(Pragmatics & Beyond New Series 20) Jan Nuyts-Aspects of A Cognitive-Pragmatic Theory of Language - On Cognition, Functionalism and Grammar-John Benjamins (1991)

A COGNITIVE-PRAGMATIC THEORY OF LANGUAGE
Pragmatics & Beyond

New Series
Editors:
Jacob L. Mey
(Odense University)
Herman Parret
(Belgian National Science Foundation, Universities of Louvain and Antwerp)
Jef Verschueren
(Belgian National Science Foundation, University of Antwerp)
Editorial Address:
Linguistics (GER)
University of Antwerp (UIA)
Universiteitsplein 1
B-2610 Wilrijk
Belgium
Editorial Board:
Norbert Dittmar (Free University of Berlin)
Bruce Fraser (Boston University)
John Heritage (University of California at Los Angeles)
David Holdcroft (University of Leeds)
Catherine Kerbrat-Orecchioni (University of Lyon 2)
Beatriz Lavandera (University of Buenos Aires)
Marina Sbis (University of Trieste)
Emanuel A. Schegloff (University of Calif ornia at Los Angeles)
Paul O. Takahara (Kobe City University of Foreign Studies)
Sandra Thompson (University of Calif ornia at Santa Barbara)
Daniel Vanderveken (University of Quebec at Trois-Rivires)
Teun A. van Dijk (University of Amsterdam)
20
Jan Nuyts
Aspects of a Cognitive-Pragmatic Theory of Language

ASPECTS OF
A COGNITIVE-PRAGMATIC
THEORY OF LANGUAGE
ON COGNITION, FUNCTIONALISM,
AND GRAMMAR
JAN NUYTS
Belgian National Fund for Scientific Research
University of Antwerp
JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
1992
Library of Congress Cataloging-in-Publication Data
Nuyts, Jan.
Aspects of a cognitive-pragmatic theory of language : on cognition, functionalism,
and grammar / Jan Nuyts.
p. cm. -- (Pragmatics & beyond, ISSN 0922-842X ; new ser. 20)
Includes bibliographical references and indexes.
1. Psycholinguistics. 2. Cognition. 3. Functionalism (Linguistics) 4. Generative gram
mar. I. Title. II. Series.
P37.N89 1991
401'.9--dc20 91-40002
ISBN 90 272 5032 4 (Eur.)/l-55619-288-6 (US) (alk. paper) CIP
Copyright 1992 - John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or
any other means, without written permission from the publisher.
for Peace
Die Sprache ist ein Labyrinth von Wegen. Du kommst von einer Seite
und kennst dich aus; du kommst von einer andern zur selben Stelle,
und kennst dich nicht mehr aus. (Wittgenstein 1958: I 203)
One opinion that has remained with me from then until now is that
the complexities of the universe, linguistic or otherwise, are so vast
that one cannot help but be awed and humbled by them, and that
arrogance in a linguist betrays at least a lack of perspective on the
problems which confront him. (Chafe 1970: 2)
While observed facts and facts deduced from facts are the flesh and
bone of scientific inquiry, its heart and soul is creative speculation
about the facts. (Givn 1979: 311)
If you want to get ahead, get a theory. (Karmiloff-Smith and Inhelder

1974: 195)
In science and philosophy alike, an exclusive preoccupation with

logical systematicity has been destructive of both historical under
standing and rational criticism. Men demonstrate their rationality, not
by ordering their concepts and beliefs in tidy formal structures, but
by their preparedness to respond to novel situations with open minds
- acknowledging the shortcomings of their former procedures and
moving beyond them. (Toulmin 1972: vii-viii)
Preface
This study discusses a number of elementary problems concerning natural

language and the sciences investigating it with the aim to contribute to the
development of a framework for modeling the cognitive systems involved in
linguistic behavior. The ideas presented here have mainly been determined by
three basic considerations.
The first assumption is that language is an extremely complex behavioral
phenomenon which is intimately linked to human thought and action in general.
Language has a considerable share in the mysteries of our cognitive world, which
are not less fascinating than those characterizing the physical world, but which
are probably even harder to access for scientific inquiry.
Secondly, this inquiry is also grounded in the conviction that one should
nevertheless try to unravel at least part of these mysteries. Our ability to disclose
reality is no doubt restricted, since our capacity to observe and understand is
determined and delimited by specific properties inherent to mankind. Anything
beyond these limits will escape our intellectual greediness forever, and this
undoubtedly includes important dimensions of our own thinking and acting.
Nevertheless, we should accept the challenge to tackle the mysteries and try to
push the limits of our understanding as far as possible. We have to do so not only
for reasons of intellectual self-esteem and satisfaction, but also, and primarily,
because it is important to do so. Due to the supremacy of physical sciences, the
role of understanding human behavior and cognition (and language is a source
par excellence for investigating the cognitive world, since it is one of the most
important means by which the mind reveals itself) is all too often underestimated.
First of all, the 'sciences of the mind' hold the key to understanding science as
a human activity in itself. Hence, they are crucial for the interpretation and
qualification of scientific theories, also in the physical sciences, and that is why
they are more central to science than the latter, contrary to the currently
prevailing opinion. On a more general level, the task of the sciences of the nrind
is much wider than just describing and explaining the mind. For, investigating the
mind is the most direct way to providing solutions for many of the problems
facing our society. If at all, then it is along this road that we can find a key to
a qualitatively better society, in which dialogue and mutual understanding are not
empty words, and in which justice and peace stand a fair chance. (Of course,
science can only try to find a key - using it is a matter of politics, and hence of
viii A COGNITIVE-PRAGMATIC THEORY
will and power.) The physical sciences can provide powerful tools, but only the
sciences of the mind can provide the manuals for using them. Hence, the latter
have an important responsibility in the struggle for survival of mankind and, ever
since the existence of the nuclear insanity, of this planet. This responsibility will
only grow with the increasing complexity and, correspondingly, increasing
dangerousness of society. That is why they are also more central to human society
than the physical sciences, again contrary to public opinion. Man is responsible
for him/herself.
However, understanding behavior and cognition presupposes a mentality
radically different from the one characterizing current research into the mind,
and into language in particular. The current fragmentation of research will have
to give way to a more integrated approach, and individualism will have to be
supplemented by collectivism. Of course, this should not endanger pluralism in
science. From a collective and integrational approach to a uniformistic and
monolithic view is but a small step, but dogmatism is deadly to rational thought.
What is required is a perspective which allows the many scattered inquiries into
various subaspects of the complex phenomenona of mind and language to be
suited into a comprehensive frame of reference, which makes them available to
the entire scientific community, and which at the same time functions as a
selector and director for further research. This is the third basic concern of this
study.
The ideas presented here owe a lot to the contributions of many. First of
all, I have inherited much of my thinking about language from Georges De
Schutter. The model introduced in the third chapter as a culmination point of
this book is a further development of joint research (De Schutter and Nuyts
1983). Secondly, Jef Verschueren has considerably widened my views on language
and has added many new perspectives to my conception of linguistics (not least
through our cooperation on Nuyts and Verschueren 1987). Detailed comments
on earlier versions of (parts of) this text have been due, not only to De Schutter
and Verschueren, but also to (in alphabetical ordering) Johan Van der Auwera,
Simon Dik, Steven Gillis, Louis Goossens, Willy Van Langendonck, Walter De
Mulder, Herman Parret, and Frank Platteau. I hope they will not be too
disappointed if they notice how stubborn I am, but I suppose they already knew.
It is impossible to mention all those who have contributed in more indirect ways
to my understanding of the different topics discussed here.
Research for this book, which is a radically shortened and completely
revised version of my PhD-thesis (Nuyts 1988), was supported by the Belgian
State Department for National Education and Dutch Culture during stays at the
Institutes for General Linguistics of the Universities of Salzburg and Amsterdam,
PREFACE ix
from 1981 untill 1983, and by the Belgian National Fund for Scientific Research
which has employed me as a researcher since 1983. The final touch was added
during a stay at the University of California at Berkeley in spring, summer, and
fall 1989, which was made possible by grants from the Belgian National Fund for
Scientific Research, from the Commission for Educational Exchange between the
USA, Belgium and Luxembourg (Fulbright), and from NATO, which allowed me
to use its money for peaceful purposes.
Mouton de Gruyter and Elsevier Science Publishers have granted
permission to integrate an adapted version of papers which have appeared in
Semiotica 1992 ('On determining the functions of language') and in the Journal
of Pragmatics 1990 ('Negative-raising reconsidered'), as parts of sections 1.2 and
3.3 of this book.
Last but not least, this research would not have been possible without social
support from my family and friends. Invaluable.
Contents
Introduction 1
1. Elements of the theory

1.0. Introduction 3
1.1. Cognition and cognitive science 3
1.1.1. Cognition as an object in language research 3
1.1.2. What does it mean to investigate cognition? 5
1.1.3. Possible arguments against cognitivism 11
1.2. Functions and language functions 26
1.2.1. On 'functions' 26
1.2.2. On the literature on the functionality of language 32
1.2.3. Toward a typology of language functions 45
1.3. Functionalism and functional explanation 64
1.3.1. Functional explanation and the pragmatic perspective 64
1.3.2. Functionalism vs formalism in linguistics 68
1.3.3. A review of arguments against functionalism 71
Notes 83
2. Linguistic knowledge and language use: Chomsky revisited

2.0. Introduction 91
2.1. The structure of Chomskyan theory 91
2.2. Knowing and using language 99
2.2.1. TG and competence and performance 99
2.2.2. Characteristics of behavior and the nature of knowledge 105
2.2.3. TG and the processes of speaking and understanding 124
2.3. On the development of language 149
2.3.1. LAD and language acquisition 149
2.3.2. Ontogenesis from a psychological and biological perspective 56
2.3.3. On the phylogenesis of language 169
Notes 177
3. Linguistic and conceptual systems in language processing

3.0. Introduction 183
3.1. Some possibilities and limitations of FG 183
xii A COGNITIVE-PRAGMATIC THEORY
3.1.1. The basics of FG functional syntax 183

3.1.2. On the cognitive plausibility of FG 190
3.1.3. Some alleged descriptive problems for FG 197
3.2. An integrated view of the systems for language processing 219
3.2.1. Language processing and general knowledge 219
3.2.2. Production and interpretation, and knowledge and control 244
3.2.3. Knowledge systems involved in language production 256
3.3. Negative raising: A test case 285
3.3.1. Traditional' views of negative raising 285
3.3.2. Reconsidering the traditional conception and accounts of NR 292
3.3.3. Toward a cognitive-pragmatic approach 301
Notes 322
Conclusion 333
References 335
Subject index 379
Index of names 393

Introduction
This book is about a theory of natural language which is basically built on two
observations, viz.: (i) Linguistic behavior requires an extensive cognitive infra
structure in a human being, and the sciences of language have to reveal the
principles, systems and processes constituting this infrastructure (i.e. a cognitivist
point of view). (ii) Linguistic behavior is functional for a human being, and the
sciences of language have to account for this instrumental character (i.e. a
pragmatic or functionalist point of view). The main assumption is that these
observations, each of which is non-controversial within specific research traditions,
are not contradictory (as is sometimes suggested), or not even complementary,
but only involve two different dimensions of one and the same phenomenon.
Accounting for language means accounting for both, simultaneously and
coherently. Hence the concept of a 'cognitive-pragmatic' theory of language. In
this book I will try to substantiate this view by considering its implications for a
wide range of issues.
In the first chapter, the elements of the cognitive-pragmatic theory are
presented. This involves a discussion of the cognitivist and the functionalist points
of view and their interaction, and of the implications for the organization and
methodology of research on natural language. Topics of special attention will be
the need for a coordination of research efforts throughout the various sciences
of language within the framework of an integrated cognitive science on the one
hand, and the problem of determining the functions of language on the other.
As to the former, the focus, also in the rest of this book, will be on the potential
contribution of research traditionally classified as linguistic to such an integrative
approach, and on the implications of an integrated cognitive science for linguistic
research. This also means that this study will mainly be concerned with adult
linguistic behavior and the 'grown-up' cognitive systems determining it.
In the second chapter some core hypotheses concerning the basic properties
of the cognitive systems involved in linguistic behavior are formulated, mainly in
contrast with the claims put forward by Chomsky. The aim is to show that
Chomsky's formalistically (anti-functionalistically) inspired proposals concerning
the nature of linguistic knowledge and its relation to language use are not
maintainable, and to argue for an alternative knowledge-for-use conception which
fits the framework of the cognitive-pragmatic theory of language. The
argumentation is further underpinned by means of a discussion of the problems
2 A COGNITIVE-PRAGMATIC THEORY
of innateness, acquisition, and phylogenesis.

The third chapter contains a discussion of several more specific issues
related to modeling the cognitive systems involved in linguistic behavior, from
the perspective of the cognitive-pragmatic theory of language. The starting point
is a consideration of the possibilities and limitations in this respect of one specific
functional linguistic model, viz. Dik's (1978a, 1989a) Functional Grammar. The
discussion leads to the introduction of the concept of a Functional Procedural
Grammar (De Schutter and Nuyts 1983, Nuyts 1985b), which includes a modified
version of Functional Grammar in its verbal (syntactic) component, but which
also attempts to provide a more integrative picture of language production in
specific, by introducing systems which can account for the encoding of a
conceptualization into a verbal pattern (including an account of the level of
discourse). Main points of concern will also be the nature of conceptual
knowledge, and the relationship between language production and language
understanding. In order to test some of the main differences between Functional
Grammar and Functional Procedural Grammar and to demonstrate the
possibilities of the latter, the 'classical' problem of negative raising is considered.
1. Elements of the Theory
1.0. Introduction
Let's start by considering the basics of the cognitive-pragmatic theory of language.

In 1.1 I will be concerned with the cognitivist view, with special attention to its
consequences for the general conception of research into natural language. 1.2
and 1.3 will be devoted to the functionalist view and its interaction with the
cognitivist view. 1.2 will concentrate on the poorly understood but from a
functionalist perspective crucial question of what is involved in the functionality
of language. In 1.3 the principle of functional explanation as a basic research
strategy for investigating cognition will be discussed and arguments against func-
tionalism in language research will be considered.
1.1. Cognition and cognitive science
1.1.1. Cognition as an object in language research
As Toulmin (1972) once remarked, clarity and unanimity about goals is one of
the most essential preconditions for the development of a full-scale science.
Language research seems to lack this feature, however. Restricting ourselves here
to theoretical research,1 most scholars will agree about the claim that the goal
is to describe and explain (in an object-inherent' sense)2 the object, as in any
other science. But the point is that there is no unanimity about what this involves.
This is reflected in the state of the art of language research. At present, one
can hardly speak of a coherent 'science of language'. At best, there is a
'conglomerate of fields and domains of inquiry into language', each of which has
its own object of investigation (some aspect or dimension of natural language),
terminology, research methods, and/or descriptive systems. Of course, there is
nothing wrong with the fact that language is approached from many different
angles and in many different ways. On the contrary: given its extreme complexity
and many-sidedness, this will be inevitable. But the point is that there appears
to be no coherence behind the diversity. Most subdomains tend to remain
isolated, because different terminology obstructs communication and/or because
there is little interest in what is going on in other subdomains, obvious rela-
tionships notwithstanding.
Consider, for example, textlinguistics, discourse analysis, conversational
analysis, those parts of speech act theory which analyze speech act sequences or
pairs, and psycholinguistic research into speaker strategies of discourse
linearization. They are all concerned with the structure of discourse, whether
monological (textual) or conversational. Yet interactions between them are
remarkably rare. Or consider text linguistics and sentence grammar. They have
been opposing each other on the question which phenomena belong to which
discipline, and they have even been quarreling about who is right and who is
wrong (cf. e.g. Petfi (ed.) 1979). Yet language is obviously organized both on
the discourse and sentence levels without there being any opposition.
This state of affairs is counterproductive, leading to a waste of energy spent
on pointless discussions or the pursuit of insights already available in other sub-
domains, or, even worse, leading to theoretical mistakes. One example is the
tendency toward 'imperialism' of subdisciplines, i.e. the tendency to integrate phe
nomena which do not really belong within their realm. For instance, many
sentence grammars, including generative grammar (henceforth TG), 3 have tried
to handle anaphoric reference exhaustively in terms of intrasentential principles,
while much of this phenomenon is obviously discourse bound (cf. 3.1.3.3).
(Another example is the use of semantic selection restrictions in grammar - cf.
3.2.3.1.)
Except for few scholars (like Chomsky - cf. 2.2) most investigators will agree
that language is a clearly identifiable and legitimately distinguishable object of
inquiry, despite the division of the field. But if it is, there is an urgent need for
an integration of the many individual research efforts concerned with it (as has
long been recognized by many leading scholars, including Sapir 1929: 166,
Greenberg 1958: v, Reichling 1969: 22, and Toulmin 1971: 370). We should be
looking for a disciplinary framework which can serve as an umbrella for the many
related, and in many cases perfectly compatible, subdisciplines, and which can
provide for a coherent terminology, for the correlation of different research
methods, and, ultimately, for an integrated descriptive system for the many
aspects and facets of natural language. Such a framework would allow for an
appropriate selection of relevant research topics against the background of an
encompassing picture of natural language reflecting the current state of the art.
It could be the 'receptacle' for the results of the various sub-inquiries, thus
making them available to the entire community of language researchers. Finally,
it could serve as common ground and arbitral force for the discussion between
alternative opinions on specific sub-problems.
The question remains what this disciplinary framework should be. This
ELEMENTS OF THE THEORY 5
obviously depends on what is involved in describing and explaining language. An

answer more and more frequently heard these days is that language research is
basically cognitive research. I do subscribe to this view, hence 'cognitive science'
may be the appropriate umbrella for the different subdisciplines, although the
concept of cognitive science may have to differ considerably from current practice
in it (cf. below). The suggestion in itself can be legitimized on the basis of the
simple observation that language is a type of systematic behavior of human be
ings. Chomsky's (1959) criticism on behaviorism (i.c. on Skinner 1957) has
certainly made clear that the complexity of linguistic behavior forces one to
accept the existence of a (most probably very intricate) infrastructure making its
performance possible. This infrastructure is what is called 'cognition of language',
and, obviously, an explanatory description of natural language should reveal what
is involved in it and how it causes the (of course also to be described) systemati-
city in linguistic behavior. This is what I call a cognitivist view of language.
Of course, this objective covers a number of dimensions, viz. (i) description
of the 'adult' cognitive systems directly responsible for the linguistic behavior of
adult language users; (ii) description of the acquisition and development of the
cognitive systems in children responsible for the 'stages' of linguistic behavior
from the initial form to the adult form (ontogenesis); and (iii) description of the
development of the linguistic capacity in the course of the evolution of the
species homo sapiens (phylogenesis). The present study is mainly concerned with
(i) (but see 2.3 concerning (ii) and (iii)).
Many investigators will object to the above characterization of cognition as
the object of description for language research. Unjustly, I would claim. Before
considering why, however, let's first have a look at what investigating a cognitive
phenomenon implies, since this has a crucial bearing on the argumentation. (A
more elaborate discussion of the following is in Nuyts 1988.)
1.1,2, What does it mean to investigate cognition?
In trying to investigate the human cognitive world, one is faced with a problem
the behaviorists rightly considered to be crucial, viz. that it is a 'black box', not
accessible to immediate scientific observation. Of course, the behaviorists4 have
exaggerated the problem. Due to their radically positivist attitude, they have
abolished not only the unjustified introspectionism characterizing their prede
cessors' methodology, but also all kinds of cognitive awareness which can not be
equated with introspection. Thus, language users are well aware of the meaning
or the significance of (the information carried by) utterances as physically observ-
able acoustic signals. Of course, meaning is not physically observable in itself, but
it certainly is intersubjectively testable to a high degree (cf. Heringer 1983 on the
problem of meaning and intersubjectivity). Another type of awareness about
which language users can report, and which is therefore intersubjectively testable,
involves the wellformedness and appropriateness in situations of linguistic utter
ances. There is no good reason why students of language should not use these
phenomena in their research, though they certainly should be careful not to
overestimate their possibilities, since that might lead to unwarranted intro-
spectionism again (cf. 2.2.2.3). (Cf. also Seuren 1977, Schnelle 1981a.)
But human cognition is a black box insofar that its 'machinery' (as Seuren
1977 calls it) cannot be observed. We may be aware of meaning, but we are not
aware of its nature, its way of being represented in the mind, the mechanisms
applying to it, etc. We have no access to the mechanisms linking meanings with
the acoustic patterns of utterances or vice versa (linguistic attitudes do not inform
us about these matters either), and neither to the mechanisms responsible for
having linguistic attitudes.
The fact that human cognition must be related somehow to the physical
structure of the brain does not bring us much further, since the nature of the
functional relationship between them is totally unknown. Neurolinguistics and
phonetics have come to fairly detailed insights into the (neuro)physiology of
speech (articulation) and perception (hearing). But how the linguistic systems
responsible for the construction of articulatory patterns and the interpretation
of perceptual patterns is present in the neurological systems remains a mystery.
(Cf. Luria 1973, 1976, Dingwall 1975, 1979, Marin 1976, Nuyts 1982b.) Most
probably, advances in this domain will only be possible if neurolinguists obtain
detailed hypotheses concerning the cognitive systems of language, for the
cognitive functions of neurophysiological structures and mechanisms are not
apparent from their physical appearance at all.
Of course, in developing hypotheses concerning cognition one should not
neglect neurological evidence - but the point is that such evidence is only of very
indirect use, contrary to what is sometimes suggested. For example, the fact that
there appear to be two main areas for language in the brain, Broca's motpr area
and Wernicke's sensory area, linked by a large bundle of fibres (the Arcuate
Fasciculus), is not in conflict with the hypothesis that the cognitive systems
responsible for linguistic behavior consist of separate production and interpreta
tion modules. But it is wrong to consider this to be strong evidence for this hypo
thesis, as e.g. Straight (1976) does. The fact that the neurological data can easily
be interpreted from the perspective of this cognitive theory is at best no
indication against it. It does not mean that there cannot be other cognitive
theories reconcilable with them. In fact, as Cooper and Zurif (1983) indicate, the
relationship between Broca's and Wernicke's areas and the cognitive functions
of language production and interpretation might be much more complicated than
would appear at first sight. In the same vain, Schnelle's (1981a, 1981b) claim that
the microstructural organization of the brain (the level of individual neurons and
neuronal groups) provides empirical access to the organization and functioning
of the linguistic systems should be handled with care. Again, Schnelle's (e.g. 1984)
cognitive model may be compatible with our knowledge about micro-level neuro
physiology, but that is not an argument for his claims. There is no reason to
believe that the functional relationship between mind and brain is simple. (Cf.
also Morton 1981.)
These considerations lead us straight into the mind-brain discussions in
philosophy. Probably, this matter should be approached quite differently from
the traditional materialist, idealist, or dualist perspectives, and Ryle's criticism
on the Cartesian 'ghost in the machine' might be on the right track.
When two terms belong to the same category, it is proper to construct conjunctive
propositions embodying them. Thus a purchaser may say that he bought a left-hand
glove and a right-hand glove, but not that he bought a left-hand glove, a right-hand
glove and a pair of gloves. [...]. Now the dogma of the Ghost in the Machine does
just this. It maintains that there exist both bodies and minds; that there occur
physical processes and mental processes; that there are mechanical causes of cor
poreal movements and mental causes of corporeal movements. [... Neither] of the
illegitimately conjoined propositions is absurd in itself. [... But] the phrase 'there
occur mental processes' does not mean the same sort of thing as 'there occur phys
ical processes' [...]. The belief that there is a polar opposition between Mind and
Matter is the belief that they are terms of the same logical type. [...] It is perfectly
proper to say, in one logical tone of voice, that there exist minds, and to say, in
another logical tone of voice, that there exist bodies. (Ryle 1949: 23-24)
(Cf. also Searle 1983.) If this is accurate, it will always remain necessary to talk
both in terms of neurophysiology and of cognition, and to develop both
neurological and cognitive theories and models. Looking for correlations between
them is essentially a matter of finding the translation system which can mediate
between both dimensions.
If a mapping between psychological processing and neurophysiological structures

is possible, it will only come about after the key theoretical constructs are
established for each level of explanation. In the interim, there are restricted circum
stances in which neurophysiological observations can play a role in the development
of psychological models. But these circumstances require the consideration of such
evidence, not only in terms of the physical organization of the brain, but also in
terms of its functional organization. (Mehler et al. 1984: 111)
(Cf. also Putnam 1973, 1980b, Fodor 1976, Popper and Eccles 1977, Pylyshyn
1980, Morton 1981, 1984.)5
How can we get at an explanatory description of cognitive systems, then?
Let's adopt the classical terminological distinction between a taxonomy and a
process model, the former being a description of behavioral data and the syste-
maticity in them, the latter being a description which deals with the behavioral
data in terms of the cognitive processes and systems responsible for them (cf. e.g.
Des Tombe 1976). Note that a process model can still be said to be a taxonomy,
but one which aims at 'psychological' or (henceforth) 'cognitive adequacy'.
Developing a process model is bound to be a very indirect business, though
it does remain empirical. The data available include not only the utterances used
in their contexts, their meanings, and the attitudes speakers have about them,
but also all kinds of non-verbal behavioral cues accompanying verbal behavior
(gesture, facial expression, gaze, hesitation and pausing in speech, etc.),6 and
all kinds of indirect, experimentally invocable data such as reaction time or
process time, response accuracy, etc. All of these can give us indirect indications
about the cognitive machinery: they are all products or effects of its activities,
thus suggesting which potential or capacity it must have, yet without providing
us with a picture of it.
This means that investigating cognition requires a hypothetico-deductive
methodology. It has to be based on an interactive research procedure, involving
both an analysis of all kinds of behavior and logical reasoning (or 'creative
speculation') concerning the infrastructure required to cause this behavior. (Cf.
also Chomsky 1959: 27,1975c: 304.) Ultimately, this is one of the most important
insights Chomsky has introduced in post-structuralist language research.) It is an
illusion to believe that induction can bring us anywhere in science, as is long
recognized in the theory of science.7
We can do without a principle of induction. We invent our hypotheses, the bolder,

the better. And then we test them; which means that we submit them to the most
gruelling criticism of which we can think. Bold, imaginative hypotheses, controlled
by severe criticism: this is about all there is to the method of science. Science is
predictive without being inductive. (Popper 1981: 19)
This appears to apply particularly strongly to research on cognition.

Actually, one should not confuse induction and empirical methodology. As
indicated above, research on cognition has to be empirical. The data determine
what the theory has to explain, and any hypothesis has to be tested against the
data. But the point is that mere observation of data does not lead to hypotheses.
Data-observation can only serve as a source of inspiration and as a testing
ground. The question what comes first, the creative reasoning or the observation,
is equivalent to the question whether the chicken preceded the egg. Developing
a process model involves a spiral movement from one to the other, starting from
some elementary theoretical notions which originate, who knows where, and
gradually developing and expanding into an ever more detailed system, against
the background of accumulating data.
A type of description based on pure induction could be called a 'pure
taxonomy', then. Examples of this category can be found among the descriptions
by the American 'taxonomie' (Katz 1964) structuralists. Although even there the
level of 'pure taxonomy' is frequently surpassed, once they start introducing
hypothetico-deductive notions (albeit very superficial ones) like 'phoneme' or
'morpheme'. Most European structuralists, and post-structuralism anywhere,
make extensive use of such notions and principles. A systematization of behavior
by means of these we may call a 'systematic taxonomy'. Developing a process
model also requires the introduction of hypothetico-deductive notions, but this
is not sufficient in itself. It requires postulates concerning the form and contents
of the global cognitive systems, which means trying to fill in what happens in
language use, i.e. in linking utterances to meanings. Of course, the hypothetico-
deductive notions can be cognitively adequate, and in that case they do have to
figure somehow in the hypotheses concerning the global systems. Developing a
process model thus involves relating relevant notions and principles to each other
and to the observable data by means of systems of rules or procedures.
What are the criteria to decide on whether a specific description of an
aspect of (linguistic) behavior is cognitively adequate, i.e. can be characterized
as a process model? In view of the foregoing, two main sets of factors can be
pointed out, leading to two fundamental types of adequacy criteria.
(i) The effects the model predicts have to correspond to the actual behavior. This
involves two aspects. On the one hand, the model may only predict effects which
(can) actually occur (observational adequacy). On the other hand, it has to
predict all effects which (can) occur within the range of phenomena it intends
to cover (observational completeness). Both constitute the principle of 'external
descriptive adequacy'.
(ii) The lines of reasoning followed in developing the model have to be plausible.
Again, this involves two aspects. On the one hand, the main theoretical options
from which the model is built have to be reasonable from the perspective of the
overall nature of the linguistic systems (conceptual plausibility). This can be made
more specific in the requirements (a) that these options be compatible with
widely accepted theoretical preconceptions and insights concerning language
gained from research on other dimensions such as ontogenesis, phylogenesis, and
pathology, and (b) that the model in itself can be fully integrated with models
on other facets of normal adult linguistic behavior, or behavior in general. On

the other hand, the model has to be logically consistent and coherent from the
perspective of the accepted theoretical options (conceptual coherence). Together,
these constitute the principle of 'internal descriptive adequacy'.
In principle, we can call a model cognitively adequate only if it fully
conforms to these criteria, i.e. if it is a perfectly consistent system accounting for
all behavioral phenomena (including indirect or accompanying ones) within its
range, which fits completely and organically into a global model of all facets of
linguistic behavior, and behavior in general, and which in turn is perfectly
integratable in an encompassing scheme including models on pathological
behavior, ontogenesis and phylogenesis, etc. As long as not all varieties and
aspects of behavior are accounted for by the model, an attempt to have it deal
with new data can lead to extensive revisions of its internal structure, and of
accounts for types of behavior dealt with in earlier stages. And according to the
same principle, integration of the model in a larger context, or changes occurring
in a larger context in which it was integrated, can force one to thoroughly change
it. (See also Fillmore 1984: 104-105.)
This is, of course, Utopian. It is long recognized in the theory of science that
judging a hypothesis cannot be a matter of testing whether there is a correspon
dence with the ontology of the object. This would imply a verificationist point
of view, but since Popper (1963) it has become clear that the only possibility is
falsification.
One can also put it like this: all our scientific assertions, all our scientific theories,
remain for ever hypothetical. They are conjectures. They may, sometimes, be true:
they may correspond to the facts. But even if they are true, we can never be certain
of this: certainty cannot be obtained. (Popper 1981: 19)
More specifically, the problems in determining the adequacy of models on cog

nitive phenomena are twofold. First, one can never be sure that all possible facts
within the range of the model have been accounted for. The facts are not
exhaustively definable, and one can never be sure that no facts hitherto unac
counted for will crop up. This is basically the problem of the underdetermined-
ness of models, also referred to by e.g. Chomsky (1975a, 1980a). On the other
hand, one can never be sure that, for a specific set of facts, it is impossible to
develop one or more alternative models which are perfectly equivalent with the
model under consideration, from the viewpoint of both its internal and external
adequacy.8 One can therefore only speak in terms of the probability of a model
being cognitively adequate, and judging a model is a matter of rating this
probability. The more data a model accounts for, and the better it fits within an
encompassing framework, the higher the probability that it is an adequate
characterization of the object of inquiry.

Of course, judgments strongly depend upon the state of development of the
science at a certain moment: a process model is a model for which so far no
refutation has been possible. Given the current state of the art in language
research, or research on human behavior in general, in which large domains and
important aspects of the data are hardly explored, let alone accounted for in
models, one can only adopt a minimalistic attitude. Assigning this title is
necessarily a matter of belief in a model being promising for future development.
But we should not have too many illusions about the appropriateness of our
current descriptive systems.
Still, there is no reason to remain content with the current state of the field,
and the question is, then, what kinds of reasons there could be for not accepting
the cognitivist view as a basis for an integrative approach to linguistic behavior.
1.13. Possible arguments against cognitivism
1.1.3.1. The individual and the social dimension of language

An obvious argument against the claim that cognitivism can be a sufficient basis
for integrating the language sciences, and that cognitive science could be an
appropriate umbrella for them, is that cognition only covers one dimension of
language, viz. the 'psychological' or 'personal' one, and that this way its social or
group dimension would be neglected. This argument is probably based on a
mistaken (though certainly very widespread) conception of cognition. One should
not confuse this notion with individuality.
Cognition does not at all exclude the social dimension of linguistic behavior,
quite to the contrary. It stands beyond doubt that human beings are both
individuals and social beings. There is no opposition between these facets, they
are complementary and inextricably linked together like two sides of one coin.
There cannot be a social context without individuals constituting it, but indi
viduality only exists against a social background. One cannot meaningfully discuss
individuality without considering the social dimension, and vice versa. Hence,
research into the cognitive world of human beings is research into their social
behavior not less than their individuality. And attempts to eliminate the social
dimension in this research necessarily lead to restricted, and potentially even
mistaken models.
Still, some may persist that cognition only covers phenomena within one
individual (albeit personal and social aspects of the individual), while many
language phenomena are typically group phenomena (cf. e.g. Verschueren's
1987a distinction between micro and macro level analyses). But again, there is
no contradiction between these observations. It is true that most linguistic
phenomena are not specific to one individual but occur within groups. Still, the
linguistic traits inherent to a specific group are obviously carried by the individ
uals constituting that group. A speaking individual can be considered a set of
linguistic characteristics, some of which are strictly personal, and most of which
are common to larger groups to which the individual belongs. A linguistic group
is determined by a set of characteristics carried by most or all individuals
constituting it. Hence, group traits have no existence outside the cognitive systems
of the individuals constituting the group. A group is an (albeit very complex)
function of its members and their cognitive set-up, and the group dimension is
an 'emanation' of the individual dimension.
This does not mean that one should not describe groups and their charac
teristics, on the contrary. Neglecting the fact that most linguistic phenomena are
group phenomena would again lead to restricted, and potentially even mistaken
models. But still, describing a group is really nothing else than describing a
prototypical language user of that group, i.e. is describing cognitive phenomena,
both individual and social. (See also 2.2.2.)
The above point of view should not be mistaken for the claim that group
phenomena merely derive from personal traits of the individuals constituting the
group, either. Many characteristics of an individual are acquired from other
members of the group to which it belongs and are therefore due to the social
dimension of cognition. By internalizing such characteristics the individual
becomes one of those carrying the traits of the group, and, ipso facto, constituting
the group. Hence, not any arbitrary set of individuals constitutes a group - only
a set of individuals behaving socially (being socially engaged) toward each other
constitutes a group.
Let's illustrate this by means of the phenomenon of a language variety being
used as standard in a certain community. Obviously, this can never be understood
from the perspective of the speaking individual as an individual, but only from
the perspective of a principle of normativity inherent to a certain social
community, i.e. a guideline for behavior imposed upon the individual members
of the group. Yet it still remains a matter of the individual members of the group.
Each of them knows the principles and rules constituting the language variety
labeled standard, and this knowledge only exists in the individual members of the
group. The fact that this variety is normatively accepted as a standard for use
within the group is entirely due to the fact that the individuals constituting it
accept the group and want to belong to it, and thus start using the language
variety according to the norms of the group.
Of course, the individual language user can think in terms of the group
without having to think in terms of the individual members of the group. Con
ceptually, a group has properties which need not be grasped in terms of its
members. Moreover, the group can be considered to be more important than,
and is certainly considered to be larger than and superceding its individual
members. Hence, it can be very difficult for an individual to decide to no longer
be a member of a group. Still, these are matters of the social dimension of cogni
tion and behavior of the individual belonging to the group. The constitution of
a group is only possible through a conception of community in the cognition of
individual language users. A language user remains a member of a group just as
long as (s)he conceptually recognizes the group as normative for behavior, and
a group exists only as long as the conception of community remains in at least
some of its members.
In a similar vain, a 'macro phenomenon' such as diachronic language change
can only be understood as a change occurring in the knowledge of the language
in individual members of the community speaking it. It starts in one or a few
individuals, and due to all kinds of factors it gradually spreads among the other
members of the community. There is nothing else to 'carry' the language change
than individual cognition, even though it does not remain restricted to the
cognition of one individual.
Some (e.g. Verschueren p.c.) might argue that this claim sounds like a
reduction of the group to the individual, comparable to the reduction of mind
to brain which was rejected in the previous section. But this is not a valid
comparison. Mind and brain are notions of a different logical category, while
individual and group are notions of the same logical category (viz. the category
'mind'). In fact, while the notion of individual cognition has an obvious biological
counterpart in the logical domain of matter, viz. the brain, the notion of group
has no such biological counterpart. 'Society' will not do, since that is a notion
covering both the dimensions of mind and matter. Society is a set of entities
interacting both mentally and, ipso facto, physically.
In sum, the notion of cognition can be considered the all encompassing no
tion for language research, and this in no way excludes or denies the fact that
language research is concerned with a primarily social phenomenon mainly occur
ring in larger groups of language users.
1.1.3.2. Linguists, language psychologists, and cognition

An argument against the cognitivist view no doubt used by many 'traditional'
linguists is that investigating cognitive systems may be something psychologists
are concerned with but falls outside their own scope of interests. Their aims are
of a different nature, including matters such as finding the general systematicity

behind the many instantiations of linguistic behavior (individual utterances);
determining why language users, given specific circumstances, use certain
utterances and not others; determining what makes utterances fulfill their role
in specific circumstances; characterizing what constitutes communication; etc. The
point is that there is no contradiction whatsoever between these and the aim to
describe the cognitive systems determining linguistic behavior. For, what else
should the latter be but trying to answer the types of questions mentioned, in
terms of abstract and general notions, principles, and rules constituting a model?
Apparently, many are misled by the persistent belief that only professional
psychologists can be investigating cognitive reality. They seem to suppose that
penetrating the cognitive world requires some mysterious scientific act, logically
different from the acts performed by other scientists studying behavior, and that
there is a 'cognitive explanation' of behavior logically different from other types
of explanation of it. As Ryle (1949: 308) has already argued, it is not true "that
psychology is the sole empirical study of people's mental powers, propensities
and performances". As pointed out in 1.1.2, nobody has a privileged access to
the cognitive world. What language psychologists do is not logically different from
what linguists do.
Just like any other language researcher the language psychologist is bound
to work with the variety of indirect data available. Of course, the stereotypical
picture of a language psychologist is one of an experimenter working with
'special' data such as reaction times, speech pauses, response accuracy, etc., or
of someone working with deviant data such as slips of the tongue and data from
language pathology. This would distinguish the psychologist from the majority of
other language researchers, including linguists. This picture is biased:9 of course,
psychologists are working with 'normal', 'primary' linguistic data, i.e. normal
linguistic utterances and their structure, as well. But even insofar as it is appropri
ate, the difference with linguists is not at all fundamental. These 'special' data
are not less indirect than the 'normal' behavioral facts, on the contrary.
An investigation of deviant data requires knowledge of what is 'normal', and
their interpretation with respect to the cognitive systems and mechanisms
involved in 'normal' linguistic behavior requires a rather detailed hypthesis
concerning the latter (cf. Nuyts 1982a: 81-84).
Similarly, experimental data can only be interpreted on the basis of hypo
theses concerning the actual linguistic structures that were used in the ex
periment. Moreover, one has to be careful with them. Results in terms of reac
tion time, response accuracy, etc. appear to be subject to considerable variability,
no doubt partly due to habit formation and degrees of acquaintance with
experimental conditions on the part of the subjects investigated. This makes

results often hard to compare. Moreover, it is often very difficult to see what
experiments really measure (cf. Levelt 1973c: 75ff). For example, the use of
reaction time as a measure of processing complexity hinges upon the question
whether the processes involved are serial or parallel (cf. Berwick and Weinberg
1983,1984, who criticize the experimental investigations on TG in the sixties and
early seventies on this basis). Furthermore, reaction time can most probably not
be used at all as a measure for non-automatic processes (cf. Pylyshyn 1980).
Again, an appropriate use of experimental data already requires a rather detailed
theory on the cognitive systems involved in linguistic behavior. Finally, the history
of experimental investigation has proven that results can lead to quite different
interpretations, and that abuses are no exception. In sum, without questioning
their value as tests for hypotheses under appropriate conditions, experimental
data clearly can only serve as additional data, and they are not of a kind to play
a major role in discovering the cognitive reality behind linguistic behavior. (More
on the role of experimentation in theorizing in Valian 1979: 19ff, Miller et al.
1984, Poison et al. 1984, Nuyts 1986a.)
Psychologists do not arrive at their theories in ways different from linguists,
either. In fact, modern (post-behavioristic) language psychology started of as a
discipline aiming to experimentally test a linguistic model, TG. Once the results
turned out to be negative, and generative grammarians appeared unwilling to
accept the conclusions (cf. 2.2.3.2), psychologists mainly dissociated themselves
from linguistic theorizing. They started to develop their own hypotheses on the
basis of different theoretical preconceptions, yet preserving the principle of
hypothetico-deductive modeling introduced by Chomsky. Yet this same principle
has also been generally accepted in post-Chomskyan linguistics, not only in
Chomsky's own domains of investigation, phonology and sentence-syntax, but also
in semantics, discourse studies, etc. Hence, the majority of modern linguistic
models, not less than psychological models, satisfy the basic methodological
requirement for developing a process model.
Psychologists may seem to be pursuing questions with their models which
linguists would never bring up: e.g. how does a language user resolve ambiguity;
how does one interpret metaphors; is language processing top-down or bottom-
up; what is the role of syntax in understanding; etc. Yet the differences have to
do, on the one hand with the circumstance that language psychologists have so
far been concerned with language understanding far more than with language
production, while linguists mainly think in terms of language production, due to
historical factors (viz. Chomsky's concern with productive grammars - cf. ch.2,
ch.3). Hence, they partly describe different aspects of linguistic behavior. On the
other hand, they have to do with a difference in attitude toward model formation.
Prototypical linguistic models cover rather restricted subdomains of natural
language (e.g. phonology, sentence syntax, text structure), and in many cases
theory development is even restricted to concept development concerning very
limited phenomena (e.g. speech acts, implicatures, modality) only. The aim is
mainly to account for as many data as possible within their range. The attention
is typically focussed on representations needed to account for the data, while
rules or procedures involving them often remain implicit. In addition, linguistic
phenomena outside the scope of the approach are often completely dis
regarded. 10 This general attitude implies that linguistic models are very well docu
mented and empirically supported, but also vulnerable from the perspective of
their internal descriptive adequacy - lack of orientation within a wider domain
does not enhance the chances for the model to be apt for integration in an
encompassing model. The fact that they often involve representations without
procedures might be a problem in particular, since it is likely that the former can
only be judged appropriately if embedded in the latter (e.g. S.Palmer 1978).
Alternatively, prototypical language psychologists, though often concerned
with limited problems too, tend to frame these in a wider perspective. Their
models are usually much more encompassing than the linguistic ones, in that they
cover aspects of larger parts of the process of language interpretation and/or
production (although until recently major dimensions such as discourse organiza
tion frequently remained out of scope here too). Moreover, they do not only
propose representations, but also processes. Yet these models are incomplete
in different respects. They provide a gestalt of the systems determining language
behavior, yet leave open many (or even most) of the details (e.g. Schlesinger
1977, Herrmann 1983). They also cover a much more restricted set of data,
making them vulnerable from the perspective of their external descriptive
adequacy. This is amply illustrated by AI-models, which are made to run on
computers: they only accept or produce small parts and aspects of a language
(e.g. Winograd 1972, Davey 1978). Moreover, because of the simulation they have
to be fully formalized, but in order to achieve this a serious amount of 'kludging'
(Lehnert 1984a) is required, i.e. filling in details on an arbitrary basis (cf. also
Davey 1978, Steedman and Johnson-Laird 1978, Poison et al. 1984).
Hence these differences are not of a principled nature. Both linguistic and
psychological models are - naturally, given their hypothetico-deductive basis -
more or less explicit systems of rules or procedures and representations aiming
to account for all kinds of data available. Each approach has its own advantages
and disadvantages from the perspective of revealing the cognitive systems causing
linguistic behavior. In fact, there appears to be a remarkable complementarity
between the two subdisciplines, in all respects, which only goes to show that an
integration could only be advantageous for each of them and for the science of
language in general. It would certainly be inappropriate to suggest that linguists
are concerned with developing taxonomies, while psychologists are concerned
with developing process models, as is sometimes done in the psychological
literature. This does not correspond with the facts of actual research, and unjustly
legitimizes psychologists not to take notice of developments within linguistic
theory.
The argumentation developed by Katz (1981,1984) against cognitivism (or
'conceptualism', as he calls it) in linguistics is not likely to challenge the above
conclusion. Katz does not reject the psychologists' attempts to investigate the
cognitive world, he only denies, on principled (philosophical) grounds, that
linguists should be concerned with it, suggesting that they should take a Platonist
attitude.
Platonism draws a fundamental distinction between the knowledge speakers have

of their language and the languages that speakers have knowledge of. (Katz 1984:
34)
According to Katz languages are abstract objects having no ontology, hence

linguists have the task of describing abstract objects.
Of course, there is a difference between languages (as behavior) and
knowledge of languages (cognition). Yet what should it mean to say that langua
ges are abstract objects without an ontology? They are physically observable and
conceptually interpretable phenomena, and every hypothesis about their structure
is an hypothesis about something that exists, and, hence, has an ontology in that
sense of the word. In fact, knowledge is much more abstract (in any normal
meaning of the word) than languages, since there is no way to physically observe
the former. The existence of knowledge can only be derived from the existence
of languages. Languages can only be a result of the working of cognitive systems,
and the structure detected in the former has to be due to the latter. Katz
commits the same type of mistake Chomsky commits: Chomsky only claims
interest in the 'knowledge' part of the 'languages vs knowledge' differentiation
(cf. 2.2), while Katz only claims interest in the 'languages' part. But the point is
that both parts cannot be separated.
Let's take a closer look at Katz' arguments.
In requiring conformity to a concrete reality, psychological reality conditions impose

constraints of a kind different from the requirement that grammars correctly
describe the sound pattern, well-formedness, ambiguity, and other structural feat
ures of sentences. Psychological reality conditions in linguistics do not concern the
grammatical structure of sentences, but concern particulars of subjective experience

or human biology. [...] Thus, with conceptualism [...] there is a possibility of conflict
between a demand that grammars satisfy an extrinsic, ideologically inspired
constraint and the traditional demand that grammars meet intrinsic constraints
concerning the successful description and explanation of the grammatical structure.
If such conflicts can exist, then linguists cannot adopt extrinsic, psychological
constraints. Linguists, like other scientists, must always try to choose the best
available theories, and hence cannot adopt an ontological policy that would select
worse theories of natural languages over better ones. (Katz 1984: 36-37)
Katz illustrates the conflicts he refers to with the following.

(i) A conceptualist could be forced to choose between two rules differing form
ally, but making identical predictions concerning linguistic facts, i.e., (s)he could
be forced to choose between notational variants:
if one accepts conceptualism, one could be committed to claiming that, say, a

grammatical counterpart to the calculus in Polish notion [sic] is the true theory
of the language and a grammatical counterpart in Principia notation is not, even
though they are mere notation variants, since the human mind could be constructed
in such a way as to represent its grammatical knowledge in the one form rather
than in the other. (Katz 1984: 38)
(ii) A conceptualist could be forced to choose a theory inferior to other possible

theories in terms of (for example) simplicity, for it is possible that a language user
knows a grammar more complex than the simplest possible grammar of the
language.
(iii) A conceptualist could even be obliged to choose wrong theories over right
ones.
Let me make my point by way of an analogy. Major calculator companies, such as

Texas Instruments and Hewlett-Packard, construct some calculators on the basis
of principles that incorrectly determine the values of a function for a range of argu
ments that, for empirical reasons, can never be inputs to the device. Companies
do this because such 'incorrect principles' are either less complicated to build into
the device [...] or more efficient in on-line computation [...]. Since computations
that produce the incorrect values of the function will never take place, these savings
arefreeand clear. Now it is plainly absurd to suppose that God, Nature, or Evolu
tion wouldfindit impossible to do what Texas Instruments and Hewlett-Packard
can do. Hence we may imagine that, for essentially similar reasons, such heuristic
principles have been built into the human brain as internalized competence rules
for language processing, that is, as its knowledge of the language. In this best of
all possible worlds, we have been provided with a language mechanism that requires
less brain utilization and is more efficient in on-line processing. But for all such
benefits, and not withstanding the fact that these internalized rules give the correct
results for all sentences that can occur in performance, the rules falsely predict
grammatical facts about sentences that can never occur in performance (because
they are, say, too incredibly long or complex). For example, the internalized rules
might convert all strings of words above a certain very great length, n, into
word-salad, so that the best theory of the speaker's competence falsely predicts that
strings of English words exceeding length n are ungrammatical. Or, the internalized
rules might turn out to be nothing but a huge, finite list of n sentences, each of
which is paired with a structural description. If this is what turns out to be in our
heads, a psychologically real grammar must falsely predict that English contains
only finitely many sentences, and only n of them at that. Given that no acceptable
metatheory for grammars ought to allow us to be committed, even contingently,
to false theories of natural languages when they are avoidable, it follows that, in
committing us to these and indefinitely many further potentially psychologically
real but linguistically false grammars of natural languages, conceptualism is
unacceptable. (Katz 1984: 39-40)
Katz is using a strange concept of scientific truth, however. Normally, it

means not being in contradiction with whatever type of empirically based
arguments available. Hence, in as far as the notational variants in (i) have
different empirically observable implications, it is possible to test them. Imagine
(as a totally fictive example) we would have to choose between the Polish and
Principia notations for formalizing human knowledge of the world. It is not
unlikely that the choice leads to different rule systems for converting the
knowledge representations into linguistic utterances, and this difference might
imply different degrees of processing load. If so, we could use experimental
conditions to test the alternatives. Obviously, this leads to empirical evidence for
a linguist not less than for a psychologist as to the 'truth' of the alternatives. It
is also possible that no type of evidence whatsoever can be found to test the
alternatives. In that case they really are perfectly equivalent, but then the
psychologist cannot and should not choose between them either. All (s)he (or
the linguist) can say is that so far there are alternative proposals.
As for (ii), what does it mean for a grammar to be the simplest one pos
sible? Obviously, simplicity is a criterion that can be used to choose between two
perfectly equivalent hypotheses in the above sense. As long as it is possible to
adduce data showing that one grammar is better than another (even if it is more
complex), no reasonable scientist (linguist or psychologist) would hesitate to
choose the more complex one. If no data are left, the psychologist no less than
the linguist is obliged to choose on the basis of simplicity. Unless one is able to
show that simplicity is not a sound scientific criterion to help finding 'true'
hypotheses (which would, again, be valid for linguists and psychologists alike).
However, I am not aware of good arguments for this suggestion (cf. 2.2.3.1).
Similarly for (iii). Either there is evidence that grammatical rules turn word
strings exceeding length n into word salad or not (if such a model would be
simpler, accounting for the same data as other models, then both psychologists
and linguists have to choose it). Either there is evidence that humans only know
a list of utterances, or there is evidence that they really do systematize and
acquire rules (and there cannot be much doubt about the latter). But this is
evidence for all scientists, no matter whether they call themselves linguists or
psychologists. Psychologists are neither in the service of companies constructing
human brains, nor are they God, Nature, or Evolution. It sounds ridiculous that
they should choose 'wrong' theories as hypotheses concerning cognitive systems,
though that is precisely what Katz suggests. There is no scientific metatheory
which would allow this - in fact, there is no difference at all between the linguists'
and the psychologists' metatheories.
Katz also claims that there exist properties of utterances which cannot be
explained by grammars as cognitive theories (he clearly alludes to TG as a
specific case), e.g. the analyticity of (1).
(1) Nightmares are dreams
(1) is necessarily true by virtue of its linguistic configuration. Obviously, if a

model of language cannot account for this fact, and more particularly for the fact
that this necessary truth has to do with word order in (1), this model is
insufficient. However, this is the case also from the perspective of its cognitive
plausibility, for there are (semantic) data questioning its adequacy. This argument
of Katz' is not an argument against conceptualism, but against a model (cf. also
1.1.3.3). A cognitively plausible model has to be able to account for (1).
As to the origins of such necessary truths, Katz claims that conceptualism
is obliged to categorize them as
consequences of principles that human beings, by virtue of their psychological or

biological make-up, cannot take to be false. Such necessary truths come out on the
conceptualist's account as merely what human beings are psychologically or
biologically forced to conceive to be true no matter what. But this is a far cry from
what is true no matter what. [...]
If we raise the prospect of beings different from us whose psychology makes
them take (1) [...] to be false, the conceptualist must embrace relativism. The
conceptualist must say, 'We have our logic, they have theirs.' The Platonist is the
only one who can say, as Frege said in a similar connection, 'We have here a
hitherto unknown type of madness'. Only Platonism enables us to say that such
necessary truths are true no matter what - no matter even if we discover that
human cognitive apparatus is built to take (1) [...] to be false. (Katz 1984: 41)
Yet first of all, everybody knowing English perceives (1) to be necessarily true,
hence a cognitive theory claiming the opposite is patently false. Secondly, it is
evident that (1) (or its equivalent in other languages) is necessarily true due to
our perception of the world. In as far as other beings have a conception of the
world in which (1) is not true, (1) is no longer 'true no matter what'. It is a
necessary truth for individuals of the species homo sapiens, because they cannot
imagine another world. Any account of human beings' behavior and cognition
has to put it that way. That does not mean, however, that there could not exist
other worlds for other beings. Relativism is unavoidable - Katz obviously suffers
from anthropocentrism.
In short, one cannot but subscribe to Chomsky's words when he says:
what is 'psychological reality', as distinct from 'truth, in a certain domain'? [...] I
am not convinced that there is any such distinction [...] [It is generally claimed that]
the evidence available in principle falls into two epistemological categories: some
is labeled 'evidence for psychological reality', and some merely counts as evidence
for a good theory. Surely this position makes absolutely no sense [...]. What we
should say [...] is that any theory of language, grammar, or whatever carries a truth
claim if it is serious, though the supporting argument is, and must be, inconclusive.
We will always search for more evidence and for deeper understanding of given
evidence which also may lead to change of theory. What the best evidence is
depends on the state of the field. The best evidence may be provided by as yet
unexplained facts drawn from the language being studied, or from similar facts
about other languages, or from psycholinguistic experiment, or from clinical studies
of language disability, or from neurology, or from innumerable other sources. We
should always be on the outlook for new kinds of evidence, and cannot know in
advance what they will be. But there is no distinction of epistemological category.
In each case, we have evidence - good or bad, convincing or not - as to the truth
of the theories we are constructing; or, if one prefers, as to their 'psychological
reality' [...). (Chomsky 1980a: 107-109)
1.1.3.3. Cognitivism is not a theory

The foregoing argumentation does not imply that the notion 'cognitive reality'
is synonymous with the notion of scientific truth, however. Chomsky adds to the
quotation concluding the previous section that "this term ['psychological reality']
is best abandoned as seriously misleading". Yet the fact that the term can be
misleading (as the previous section has shown) is certainly not a sufficient reason
for abandoning it and replacing it by the concept of 'truth'. Not only because the
notion of 'truth' is potentially dangerous in connection with scientific hypotheses,
not unlikely to invoke the illusion of certainty about correspondence to reality
(a verificationist attitude), but also because of the fact that, though 'being
cognitively real' is something like 'truth in a certain domain', not every 'truth in
a certain domain' is equal to 'being cognitively real'. Truth can be defined over
any arbitrarily delimited set of facts, but truth in such a domain does not have
to correspond to truth within the domain of all possible relevant facts (cf. also
Pylyshyn 1973). (Though truth in the latter does imply truth in the former.) That
is, there is no doubt many more possible ('true') hypotheses concerning a domain
of facts considered from that domain in isolation than considered from the
unlimited context of related facts. This is basically what is implied in the concept
of internal descriptive adequacy. Thus, a 'true' hypothesis within the domain of,
e.g., constituent organization in itself does not have to be 'true' from the
perspective of the organization of utterances as means for communicating
information, and, hence, does not have to be 'cognitively real'.
Black and Chiat's (1981) arguments against the notion of 'psychological
(cognitive) reality' are ill-directed as well. They, too, claim that the notion should
be abandoned, because it has, according to them, caused an inappropriate
relationship between linguistics and psycholinguistics. What they really do,
however, is argue that linguistic descriptions TG-style are of no direct use to the
psycholinguists' aim to develop process models: the latter can borrow relevant
notions from the former, but any direct orientation toward TG has to lead
psycholinguists astray (as has been demonstrated in the past). Obviously, this
argumentation does not affect the notion of cognitive reality at all. It only affects
the claim by TG-linguists that this notion applies to their description: if their
argumentation is sound, Black and Chiat show that TG can only be taken as a
taxonomy which cannot contribute directly to gaining insight into cognition.
Hence, they have every reason to maintain the notion of 'cognitive reality', to
distinguish clearly between TG and the types of models they advocate as to their
explanatory status. Black and Chiat stress that their argumentation is not a
criticism of TG. In fact they maintain that TG-linguists, contrary to their own
repeated claims, have not taken account of what is involved in pursuing 'cognitive
reality', and that this notion has had no influence whatsoever on the TG-model.
If that is not a criticism, what would be one?
It is important to realize that accepting Chomsky's cognitivist attitude does
not imply that one has to accept his view of cognition as well - and rejecting
Chomsky's 'cognitive theory' does not imply rejecting cognitivism as a basic option
for language research, including linguistics. Cognitivism is not the monopoly of
one specific language theory, it is not a theory in itself. Rather, it is an orientation
for a theory. This is not a superfluous remark, for it seems that the reluctance
of many to accept a cognitivist attitude toward linguistics is directly inspired by
their rejection of the way Chomsky has actualized it, i.e. by a rejection of the
contents of Chomskyan theory (formalism and inneism).
This is certainly the case for many language psychologists who, like Black
and Chiat, are convinced that TG is not likely to contribute much to developing
a process model of linguistic behavior. Apparently, many have generalized this

conclusion to the whole of linguistics (or at least to syntax), as is reflected in their
general reluctance to consider linguistic models in developing their own models
of linguistic performance (cf. the isomorphism/non-isomorphism debate in Levelt
1973c - see also Valian 1977).11 Some psychologists seem not to be aware of
the fact that there are other theories in linguistics next to TG, and many seem
not to be aware that the competence/performance distinction is not generally
accepted in linguistics (cf. ch.2 and 3). 12
From a more linguistic/philosophical perspective, the correlation of cog-
nitivism in linguistics and Chomskyan theory is also illustrated in Parret (1979a).
According to him,
Such an ontologizing view of grammar [that the entire grammatical theoretical

apparatus exists in reality as a substantial property of a 'mind existing in reality*]
is hard to maintain, and one should rather accept the modest view that the
grammatical apparatus is the result of theory formation, and is thus being recon
structed without having to 'exist in reality*. (Parret 1979a: 70; my translation)
Being for or against mentalism is a metaphysical option, and this has nothing to
do with the logical structure of the linguistic theory and its necessarily epistemolog-
ical qualities of self-consistency, exhaustivity and simplicity. (Parret 1979a: 76; my
translation)
(see also Parret 1982). Parret does subscribe to Chomsky's criticism on beha
viorism and empiricism as empirical theories of language acquisition and language
use, but instead of Chomsky's 'mentalistic rationalism' he advocates an 'anti-
mentalistic (or, as he calls it, transcendental) rationalism' as an alternative.
(1) this is rationalism since knowledge is postulated which is not due to experience;
(2) yet it is not mentalistic rationalism since postulating this type of knowledge
does not imply a mental reality at the basis of linguistic behavior; (3) it is
transcendental rationalism in so far that the knowledge postulated is supposed by
the linguistic theory to be an epistemological condition for allowing linguistic
behavior. (Parret 1979a: 76; my translation)
Parret seems to suggest that linguistics should indeed be hypothetico-deductive

(as Chomsky claims), i.e. should start from an abstract a priori 'postulate of
knowledge' (cf. ibidem: 66-67), but that this 'postulate of knowledge' has only
theoretical value. I.e. it is only knowledge of the linguist, and it cannot be pro
jected into the mind of the language user.13
Yet, how to relate this to the claim that this 'postulate of knowledge' is a
'condition of allowance' for linguistic behavior? If linguistic behavior seems to
require the knowledge postulated, how can it be that language users behave
linguistically without having this knowledge? Also, what is left of the claim that
a grammar is a reconstruction? How could that be if there is nothing to be
reconstructed?
The source of the problem with Parret's claims is revealed once he starts
correlating mentalism and anti-mentalism with specific theoretical options. He
opposes Chomsky's paradigmatic view that language is primarily a cognitive or
psychic phenomenon independent of its social uses (traditionally called a strictly
formalist attitude) to the paradigmatic view (which he accepts) that language is
primarily an object of usage with a social role and history, i.e. an instrument of
communication determined by the ways it is used (traditionally called a function
alist attitude). In this respect, he claims that
The anti-mentalist attacks autonomous linguistic creativity and claims that thought
is not a solitary and circular process but a public form of behavior with a conflictive
history, a form of behavior which is constantly being corrected in dialogue and
conversation. (Parret 1979a: 77; my translation)
Apparently, Parret is only a rationalist as far as methodology is concerned, but

he opposes a rationalist theory concerning the object of investigation. His
empiricist view is even more apparent when he says that "rules are open norms
of behavior, and learning rules is built into social interaction" (ibidem: 90; my
translation).
This paradigmatic opposition has nothing to do with cognitivism, however.
It only involves the nature of the language user's knowledge, i.e. the nature of
cognitive reality. Parret's 'rules as open norms of behavior' are claims about the
cognitive reality of language as well (contrary to Parret's statement that
"grammatical rules are not 'psychologically real' and don't have to be" (ibidem:
90; my translation)). If Parret disagrees with Chomsky on the nature and origins
of linguistic knowledge and advocates another theory, then this must be because
he is convinced that the alternative must be more adequate to explain observable
linguistic behavior. If he succeeds in making his alternative plausible, then there
is every reason to suppose that the model resulting from his theoretical options
(if it satisfies all criteria of adequacy) is a more plausible hypothesis concerning
the linguistic knowledge making the language user's behavior possible. This has
nothing to do with metaphysics.
This means that paradigmatic oppositions such as those between functional-
ism and formalism are resolvable, contrary to what Parret claims. Of course, as
long as they are only formulated at a very general philosophical level they mainly
reflect accidental biases of scholars. Hence, they easily lead into endless black-
and-white discussions. A case in point are the rationalist and empiricist views on
the origins of knowledge. At a philosophical level these are often depicted as
radically opposed either/or positions. Yet who can be considered a rationalist and
who an empiricist is far from clear if it comes to more specific discussions, and
the sharp differentiation is soon lost in a mosaic of possible positions which can
be judged on their scientific adequacy. Similarly for the functionalism-formalism
opposition. At a philosophical level, a formalist characterizes a human being
mainly as a knowing being, a functionalist as a socially acting being. This is
obviously a matter of perspective, for a human being is both. One cannot be a
socially acting individual without having knowledge, and vice versa. Yet once
these general positions are made concrete in the form of specific theories and
models, as is the case in linguistics, a decision concerning plausibility is perfectly
possible by means of the adequacy criteria suggested in 1.1.2.14
It is precisely on the basis of the integration advocated above of sub-
disciplines of the language sciences such as linguistics and language psychology
that such decisions can be brought closer. As Katz once remarked,
since the psychologist and the mentalistic linguist are constructing theories of the
same kind, [...] it follows that the linguist's theory is subject to the requirement that
it harmonize with the psychologist's theories. (Katz 1964: 133)
(and vice versa, of course - e.g. Watt 1974).
[Hence, this] requirement enables us to refute a linguistic theory if we can find

psychological theories or facts that are inconsistent with it. (Katz 1964: 134)
If there is serious evidence available in language psychology that TG cannot be

a cognitively plausible model, this should lead linguists to abandon it in favor of
other hypotheses. More generally, an integration of linguistic and psychological
models could lead relatively easily to a serious reduction among the variety of
existing linguistic grammars, as it would lead to a serious reduction of psychologi
cal models. Also, though it is important to allow as many methodologies as
possible in language research (Charniak 1984 - or, as John Marshall (in
Verschueren (ed.) 1987: 53) put it, "defining a discipline in terms of its metho
dology is a recipe for total disaster"), not all methodologies are equally
appropriate (Poison et al. 1984). An integration of the language sciences could
help deciding on this matter as well.
In sum, there appear to be no reasons not to accept cognitivism as a basic
goal for research into natural language, and to accept cognitive science as an
umbrella for the various sub-approaches to this complex phenomenon. Of course,
developing this cognitive science will not be an easy job. It will require a serious
change of mentality in many researchers, and the creation of a scientific
infrastructure which is mainly lacking at this moment. For the time being, we
might start by accepting the domestic rules suggested by G.Mandler (1984: 313-
315) as a means to create a cooperative atmosphere.
1.2. Functions and language functions
1.2.1. On functions'
The second basic component of the cognitive-pragmatic theory of language is

functionalism. This theoretical option builds on observations concerning the
functional nature of language. There is much confusion about what functionality
of language involves, however, and I will first inquire into this matter.
In order to do so, we should start with an analysis of the notion of 'function'
in general, as it is used in scientific jargon. For it is sometimes claimed that "the
word 'function' is highly ambiguous, and an exhaustive list of its many meanings
would be very long" (Nagel 1961: 522). Yet this might be a case of not seeing
the wood for the trees. Let's consider two surveys in the literature of the uses
and meanings of this notion.
Both Nagel (1961) and Greenberg (1958) first of all distinguish the usage
of the term typical for mathematics and the physical sciences. Given two sets of
elements, A and B, if B is a function of A then there exists a functional rule
assigning to each element of A (the 'argument' or 'independent variable') an
element of B (the 'value' or 'dependent variable'). An example from mathematics
(y being a function of x) and from physics, respectively:15
(2) y = x2 + 2x + 5
(3) The pressure exerted upon the walls of a container by a mixture of

alcohol and water vapors is a function of the temperature and the
concentration of each vapor
These are precisely definable relations, predictable in each single value. Both
Greenberg and Nagel also provide an example of this usage of the term,in the
human sciences (they even - accidentally? - quote the same example).
(4) The suicide rate in a community is a function of the degree of social

cohesion in that society
Apart from this probably most basic usage, both Nagel and Greenberg also
mention a number of other usages, most of them primarily occurring in biology
but having analogous applications in the human sciences or in disciplines based
upon the latter, such as economy.

Greenberg distinguishes three main types.
(i) 'Function' is used to express the role of an activity of a unit for the main
tenance of the system of which the unit is a part. This is called 'activity function'.
In biology, this amounts to the role of an activity of an organ for the maintenance
of the organism possessing it, as in (5). An example from the human sciences is
(6).
(5) The function of seeing is to detect enemies
(6) The function of shaking hands is confirming a social relationship
(ii) 'Function' is used to refer to the activity of a unit. Greenberg calls this
'organic function', and examples from biology and the human sciences are:
(7) The function of the eyes is seeing
(8) Among certain people clans have the function of distributing lands
(iii) 'Function' signifies the role of an activity of a unit, not for the global system
of which it is part, but for another unit within that system. Greenberg calls this
'internal function', and examples from biology and economy, respectively, are:
(9) The circulatory system has the function to bring material from which
muscle fibres are maintained
(10) Road transportation has the function to allow a smooth provisioning

of sales units
Nagel mentions five usages, which do not directly correspond to

Greenberg's, however.
(i) 'Function' is used to refer to more or less inclusive processes or operations
of or in a unit, without specifying their role or effect. Nagel's examples:
(11) The functioning of the stomach
(12) The functioning of the postal system
(ii) The term is used to refer to organic processes in living beings which are
typical of them, and which are indispensable for their existence. These are called
'vital functions'. They are mostly ascribed to the organism as a whole, although
they are often related to the activity of specific organs within the organism.
Examples:
(13) Respiration is a vital function of/for humans
(14) In every type of civilization, every custom, material object, idea and
belief fulfills some vital function, has some task to accomplish,
represents an indispensable part within a working whole.
(B.Malinowski, quoted in Nagel 1961: 521)
(iii) 'Function' is used to refer to a generally recognized use or utility of an

object, or to an expected effect of an action, as in the following examples.
(15) The function of an axe is to cut wood
(16) The function of spreading manure on a field is to fertilize the soil
(iv) The term is often used to indicate a series of consequences of a unit or

activity for the global system to which it belongs, or for various other units within
that system. Contrary to the usages in (ii) and (iii), the consequences need not
be intended, neither need they be useful for the living being, nor do they have
to be positive contributions to the maintenance of the global system. Some of
Nagel's examples:
(17) One function of the liver is to store sugar in the body
(18) The publication of research findings has the function not only of
making them generally available, but also of submitting them to the
criticism of the scientific community as well as of establishing
priorities in discovery
(v) 'Function' is often used to refer to the role of an item for preserving a certain
property or condition in the global system of which it is a part. Nagel illustrates
this usage with quotes:
(19) If we consider any recurrent part of the life-process [of an organ

ism], such as respiration, digestion, etc., its function is the part it
plays in, the contribution it makes to, the life of the organism as a
whole.
(20) The function of any recurrent activity, such as the punishment of a

crime, or a funeral ceremony, is the part it plays in the social life as
a whole and therefore the contribution it makes to the maintenance
of the structural continuity, (both from Radcliffe-Brown, quoted in

Nagel 1961: 521-522)
Nagel claims that apart from these five usages of the term, several others may
be found.
These surveys show that the notion 'function' is used in several sciences to
refer to quite different phenomena. But that does not mean that it is ambiguous.
It is an inherent property of lexical items that they have flexibility in usage. The
real issue is whether there exists a regularity in what is referred to by the term.
The above lists allow for the following systematization.
First, one can distinguish two basic though at least metaphorically related
usages which might be called 'quantitative function' and 'qualitative function'. The
former corresponds to the usage typical of mathematics and the positive sciences,
as described above. Here the term expresses a dependency relationship between
sets of units, which is made explicit by functional rules such as (2). The
'qualitative function', which is typical of biology and the human sciences, also
expresses a dependency relationship between entities, but one which is not purely
quantificational. Again, two sub-types may be distinguished, corresponding to
Greenberg's distinction between organic and activity function. They can also be
found in Nagel's survey, albeit indirectly. His examples (11) and (12) (usage (i))
are directly correlated to the organic function in that they refer precisely to what
in a proposition expressing an organic function is mentioned as value of the
function. The distinction runs through all of Nagel's other categories: (15) and
(17) are cases of organic function, (13), (16) and (18) of activity function.
The activity function as described by Greenberg is just one variant of a
usage of the term for referring to a particular role of an element for another
element, in principle according to scheme (21) (E stands for 'entity' (a unit or
system), A for 'activity' or 'property').
(21)[E1A1][E1A2]
Hence, if one states that the function of A 1 and/or E 1 is to enable E 2 to do A2,
or to preserve A 2 in E 2 (or something similar), one is formulating a goal or role
of A1 and/or E 1 . The actual manner of formulating can greatly differ, probably
depending on the specific constellation of the phenomenon referred to. In most
cases, part of scheme (21) is left implicit (e.g., in (5) only A 1 and A 2 are
mentioned, in (13) E1 is not mentioned). Greenberg's term 'activity function'
seems too restricted to cover this usage, and henceforth I will call this 'role
function'.
Greenberg's organic function refers to an activity or property inherent to
some unit or system within a wider system, i.e. to a simple scheme of the type
[EA]. Still, if one states that A is a function of E, or that the function of E
is A, one clearly implies that A is the 'task' of E within the system it is part of.
That is, A has a specific role to fulfill within the larger system. In this connection,
Nagel (1979) refers to the 'goal-supporting view' of functions.16
Though the structure of the role function and organic function are different,
there is a clear relationship between them. Both are intended to refer to 'goal-
directed' activities and processes, and thus to goal-directed systems (Nagel 1979).
The role function indicates the goal of an entity or activity, and in the organic
function the goal-directedness of an entity performing some activity is suggested.
As such, the structure [E A] as formulated in the organic function can be
considered a basic building block of scheme (21). This can be illustrated nicely
by means of (5) and (7) above: what is expressed in (7) (an organic function)
clearly grounds what is expressed in (5) (a role function). Thus, the following
characterization of the meaning of a function-ascribing statement seems quite
to the point:
x functions as a <p in s (or: the function of x in s is to ) relative to an analytical

account A of s's capacity to just in case x is capable of -ing in s and A
appropriately and adequately accounts for s's capacity to by, in part, appealing
to the capacity of x to in s. (Cummins 1975: 762)
Just as in the quantitative function, the indication of a qualitative functional

relationship requires further specification, though it is not possible to do that in
the form of a simple functional rule. The specification strongly depends upon
one's goals and the actual object under consideration. Still, it always boils down
to showing how the unit or activity is able to fulfill the formulated goal, task or
role (cf. Cummins 1975).
The other dimensions mentioned in Greenberg's and Nagel's surveys are
all modalities of the phenomena expressed in role functions and/or organic func
tions. They do not influence the basic meanings of the term, though they may
have implications for an analysis of the phenomena themselves (some of them
will return in our analysis of the functionality of language).
Nagel's usages (ii) and (v) are both specific cases within the general type
of 'role functions': keeping an organism alive is just a special case of preserving
a specific condition in an organism (compare (14) and (19)/(20)). Of course, life
is the 'uttermost' condition in an organism, and as such it deserves a special
status. Yet formulating a role function for maintaining other conditions in the
organism usually also somehow implies a role for the uttermost condition of
survival, if only very indirectly. Together, Nagel's categories (ii) and (v) point at
what is involved in Greenberg's internal function. It is possible to formulate

relations between complexes of type [EA] which are located either at different
hierarchical levels (a relation between an entity/activity and the global system it
belongs to, i.e. what might be called Vertical function'), or at the same hierar
chical level (a relation between entities/activities within one global system, i.e.
what could be called 'horizontal function'). However, both are just variants of
the category 'role function'.
Note that by combining organic and role functions, and horizontal and
vertical functions, it is possible to represent a system as a functional chain or
network making explicit its functional organization. The culminating point of the
network is always the cluster [organism/system survival/functioning].
Another variable of function ascriptions is involved in Nagel's category (iii).
Notions like 'utility' or 'expected effect' suggest intentionality behind the system.
Hence, the essence of category (iii) seems to be the question whether the relation
involved in the functional statement is due to a conscious choice or action by an
intentional being or not. This question can apply both to organic and role
functions (cf. (15) and (16), respectively). It does not influence the basic usages
of the notion function as such, but it does have influence on an analysis of the
phenomena involved, since intentional goal-directedness is of a different order
than non-intentional goal-directedness (cf. Nagel 1979).
Finally, Nagel's category (iv) involves the factor of the number of functions
formulated with respect to some entity or activity. However, it is hard to see what
this has to do with the notion 'function' as such - it only involves the question
whether the phenomenon involved has more than one activity or role to fulfill
or not (and as such it does have impact on the possible analysis of this
phenomenon). Apparently, Nagel believes that in case of the specification of
more than one relationship of an entity/ activity with other entities/activities this
does not have to involve a positive contribution or utility any more. Yet this is
doubtful (in fact, (17) and (18) are not different from the other examples in this
respect, and Nagel (1979) has clearly changed his mind on this matter). 'Function'
seems only to be used in case of a positive role of an entity/activity within a
larger context. Functions in biology always involve a contribution of one
phenomenon to another, the former being important for the maintenance of the
latter. Similarly, one would never say that 'the function of an axe is to split the
heads of humans', for an axe is simply not intended for that activity. One might
say that 'this axe has had the function to split a man's skull', but only if the axe
has been used with that intention. It always implies that the axe has had a
positive contribution from the perspective of the one who used it, in making it
possible to split a man's skull. The formulation of several functions, as in (18),
does not change this.

No doubt, there are still other uses of the notion 'function' than those
mentioned above. However, they all seem to boil down to the same basic
meaning, viz. the expression of a positive dependency relationship between
phenomena, with the suggestion that it deserves further explication. Apparently,
there is nothing wrong with the term. Of course, it is entirely up to the user to
be clear about the exact composition and the properties of the relation one is
formulating. This is not always easy. While in mathematics and the positive
sciences the phenomena involved are normally strictly definable, this is much less
so in biology, and even less in the human sciences. Still, it only makes sense to
use the term 'function' if one is able to clearly indicate which components (as
in scheme (21)) are involved. Problems which sometimes surround the term, and
frequently lead to criticism or doubt of it, particularly also in linguistics, mostly
derive from vagueness on the part of the researchers concerning their formulation
of the functional relationship.
1.2.2. On the literature on the functionality of language
1.2.2.1. What is involved in the functionality of language?

The usage of the term 'function' in the language sciences largely corresponds to
the qualitative function just described. It points to the fact that entities related
to the linguistic system have properties or do things by means of which they play
a role for the functioning of other entities.
Thus, the term function is used to refer to the pragmatic (information
structural), semantic or syntactic status of constituents in an utterance which
makes them fulfill some role for each other and/or contribute to the functionality
of the utterance as a whole (as in 'the function of constituent X is subject';
corresponding to this are the notions syntactic, semantic, and pragmatic function
in Dik's Functional Grammar - cf. 3.1). More generally, the term is used to
indicate the role of a specific sub-structure within a larger utterance unit. Thus,
one can speak of the meaning distinguishing function of phonemes in mjnimal
pairs, the function of words in constituents, 'function words', the function of
utterances and paragraphs within a text, etc. This nearly always involves the speci
fication of an 'organic function' of structural units, implying that through this
function they contribute to the functioning of linguistic structures at higher levels,
and, ultimately, of the linguistic system in general. They specify elements of the
functional network organizing language and allowing it to perform its role in an
even wider context.
The term is also used to refer to the role of (hypothetical) components or

sub-structures of the cognitive system determining linguistic behavior, within this
system as a whole. Thus, one can speak of the function (the role) of the lexicon
in a grammar, of world knowledge in parsing, etc. Again, this involves aspects
of the internal organization of language (albeit along a different dimension),
suggesting that they are part of a functional network causing language as a whole
to fulfill its role in a wider context.
Finally, the term function is used to refer to this ultimate question con
cerning language as a unit, viz. the activities and roles it performs for the wider
organism using it as a behavioral system. This is the matter of the functionality
of language as such, and this is the focus of attention here.
The functional nature of linguistic behavior should be seen within the
general goal-directed nature of human behavior. A human is to a large degree
a purposefully (even intentionally) acting being. The things one does normally
make sense in the light of one's perspective on the contexts in which one figures.
This is not to say that one cannot behave in senseless ways (in pathological cases,
by mistake, or deliberately). But it only makes sense to talk about 'senseless
behavior' against the background of the observation that normal behavior is
meaningful, i.e. fulfills a role for the actor. It does not mean that every aspect
of human performance is meaningful or goal-directed either. The snoring of a
sleeping person, e.g., is no more a goal-directed act, or does no more contribute
to the performance of a goal directed act, than the flapping of laundry in the
wind or the whistling of a water-kettle on the stove. Neither is the smacking
sound produced while chewing a goal-directed act (intentional smacking aside)
or a positive contribution to the goal-directed process of food consumption.
Nearly every purposeful act performed by the body produces 'noise', side effects,
and one has to carefully distinguish between things 'a human being does' and
things 'proceeding from the body'.
In fact, the same distinction has to be made in domains other than the
behavioral. Consider biology. A heart makes a beating sound, and pumps blood.
Yet only the latter is meaningful from a functional perspective. The former does
not fulfill a role, it is not a function of the heart. It is only a side effect of its
pumping. (Cf. Cummins 1975.) Of course, the heartbeat does say something
about the way the heart performs its task. That is why a physician uses the
heartbeat as a diagnostic means to check how the heart of a patient fulfills its
role as a blood pump. Similarly, snoring can tell us something about the
functioning of the respiratory organs, and smacking about how food consumption
is performed. Thus, side effects have to be part of the functional analysis of a
system under consideration, although they must be assigned a separate theoretical
status.
Returning to linguistic behavior in particular, it is evident (if alone from its
prominent place in human behavior in general) that this is not just some side
effect or incidental happening as is snoring or smacking. Apparently, an individual
is performing actions when using language, and its purpose is to contribute to
the performance of these actions (which can themselves be directed at further
goals).17 In the light of our above analysis, then, we are facing the question of
how functional scheme (21) should be instantiated for language as an 'entity'.
More specifically, there are two sub-questions, viz. (i) what are the actions a user
performs by means of linguistic behavior (i.e. what are its role functions); and
(ii) what are the activities or properties of language which allow its user to per
form these actions (i.e. what are its organic functions)? This is indicated in
scheme (22), which is represented hierarchically because we obviously have a case
of Vertical role function' here. 18
(22) [Euser ?]
[Elanguage ?]
It is possible that language also causes side effects (in the sense mentioned
above). Though this matter has to be kept apart from the functions, since it may
be important for a functional analysis of language, we will nevertheless have to
pay attention to it.
If one takes a look at the literature dealing with the functions of language,
then, one is struck by the disparity of and the lack of clarity in the views
presented. Proposals are sometimes wildly diverging, but motivation and/or
argumentation is often lacking, and systematic comparison with alternatives is
hardly ever found. This is quite an embarrassment, given the potentially crucial
role of these matters for any functionalist approach to natural language. Let's
scrutinize some of the more prominent suggestions and see where they lead us.
(A more elaborate discussion in Nuyts 1984, 1985a, and 1988.)
The cause for the often considerable differences between proposals' seems
to lie not only in the existence of mere contradictions between analyses of
identical phenomena, but also in the fact that not all proposals appear to concern
the same phenomena. The analyses can be about different aspects of (22) (with
out there being an awareness of this: something like (22) is never made explicit
in the literature), or even about phenomena which only indirectly relate to (22).
In this respect, one can roughly distinguish three (albeit internally far from coher
ent) sets of proposals. Two of them mainly concern (at different levels of ab-
straction) the notion of communication, which is clearly relevant for the role
function aspect of (22) (1.2.2.2), and the third one mainly concerns the organic
function aspect of (22) (1.2.2.3).
1.2.2.2. Communication and sub-aspects of it

No doubt the most frequent view is that the function of language is communi
cation. As Sapir (1931: 105) already stated, "Language is the communicative
process par excellence in every known society", and this view is hardly ever
contested (in most of the more complex proposals to be discussed below the
functions are also situated within the 'language for communication' view). Of
course, this statement remains somewhat vague, since the notion of communica
tion is ambiguous. There are at least two usages in the literature, viz. (i) a general
one, referring to the process of interaction between individuals, and (ii) a specific
one, referring to the process of interaction between individuals by means of
natural language. (Let's call them communication or interaction, and verbal
communication or interaction, respectively.) This ambiguity is interesting in so
far as it indicates how the 'communicative function' view relates to (22). One gets
one specific type of communication if individuals use language for it. Language
is one among a series of instruments for communication. Thus communication
is a more global activity of the language user, and language fulfills a role for
performing it. This is an answer to the upper question mark in (22).
(23) [Euser Acommunication]
[E l a n g u a g e ?]
Though hardly anybody will dispute that communication is a function of
language (at this level of analysis), there is a clear tendency among non-func
tionalists to dispute the claim that communication is the function of language.
They usually refer to all kinds of non-interactive linguistic behavior (among them
Chomsky 1975a), and/or (more rarely) claim that language is primarily an
instrument of thought (e.g. Harman 1975). If so, this would mean that the seco'nd
part of scheme (23) is not complete. This point of view seems inaccurate, but the
discussion hinges upon a more precise analysis of communication, and therefore
I postpone it until later (see 1.2.3.3, 1.3.3).
A second set of proposals (including Ogden and Richards 1949, Jakobson
1960, Hymes 1962, Robinson 1972, Copi 1978, Dik and Kooij 1979, Dik 1986a
- also speech act classifications like Austin 1962 and Searle 1976 belong here) 19
is mainly concerned with phenomena situated at a level of abstraction different
from the one represented in (23). Most of the functions discussed in these propo
sals concern sub-aspects or instantiations of the phenomenon of communication,
in that they refer to (types of) communicative acts a language user can perform
by means of specific (sets of) linguistic utterances. For example, with utterance
(24)
(24) I feel profoundly unhappy
one can perform the communicative act of expressing an emotion. Hence, one
can say that the function of this utterance (and a function of language in general)
is to allow the expression of emotion. This function is present in.each of the
above mentioned proposals of this category. Or utterance (25)
(25) Close the damned door!
can be used to perform the communicative act of making an appeal to the hearer
to perform a further act. Hence, this is another function of language, mentioned
e.g. in the proposals by Jakobson, Hymes, Searle, Copi, and Dik. These functions
specify the role of specific (sets of) utterances for the language user, and in terms
of a scheme of type (21) they could be conceived as in (26).
(26) [Euser Acommunicative a c t y]
[Eutterancex ?]
Let's call these with Dik and Kooij (1979) 'sub-functions of communication'.
A general problem with overviews of this type is whether they have overall
validity. It is questionable whether it is possible to make an exhaustive list of sub-
functions of communication (contrary to what is often claimed in e.g. speech act
theory) since language allows the performance of an endless variety of communi
cative acts. Of course, it is possible to distinguish main categories of sub-func
tions, which can be further subdivided, in principle until one reaches the level
of the individual act. In fact, all sub-functions mentioned in the proposals'under
consideration are wider categories. Yet then the problem is that there can be
considerable variability in categorizing sub-functions (as is apparent from the
proposals; and see Searle 1976), depending on the criteria used and on the level
of abstraction chosen within the phenomenon of communication.
It is a major difficulty with the proposals that they hardly ever make explicit
which criteria are used and which level of abstraction is chosen (Searle is the only
exception). Still, though this is by no means the case for all proposals (see below),
sometimes they do show a systematicity in the classification, and such cases

clearly demonstrate how there can be an orientation to quite different dimensions
of communication. (In this way proposals sometimes implicitly relate to particular
organic functions of language.) For instance, the function specified with respect
to (24) above states a referential object of a speech act (which relates it to the
fact that language in communication refers to or provides information about
states of affairs in reality, which is an organic function - cf. 1.2.3). Thus, a clas
sification can be oriented toward types of referential objects of communicative
acts.) The function specified with respect to (25), however, indicates a type of
intentionality of a speech act (thus, it relates to the fact that in communication
the speaker pursues intentions with language, which is another organic function
- 1.2.3). Hence, a classification can also be oriented toward types of intentions
involved in communicative acts. No doubt many more criteria for classification
can be found, and communicative acts can be classified according to each of
these.
It is also questionable whether such classifications have cross-cultural
validity (again contrary to what is sometimes suggested). No doubt, there are
differences in what can be done with language depending on the cultural back
ground of the speakers. Thus, while the universality of the fact that language is
used for communication is beyond doubt, this does not imply that it is possible
to give a universal specification of which specific communicative acts can be
performed with language (Hymes 1969).
Speech act classifications like Austin's and Searle's, and also the proposals
by Dik and Kooij (1979) (with exception of the 'aesthetic function' - cf. below)
and Dik (1986a), are examples of pure surveys of sub-functions of communication
which are also explicitly intended as such. Dik and Kooij's (1979: 29ff) typology
juxtaposes sub-functions related to different dimensions (they also admit that
there are various possibilities to classify communicative acts), but speech act
classifications, and also Dik's (1986a) classification, which is closely related to
them, tend to consistently orient themselves toward one dimension of communi
cation, viz. intentionality.
Ogden and Richards' (1949) proposal, which basically distinguishes between
two usages of language, viz. to symbolize a referent (the symbolic function), and
to denote emotions, attitudes, moods, etc. of the speaker (the emotive function,
which is further subdivided, but the details do not matter for the present), can
also be considered a 'pure' taxonomy of sub-functions, but one oriented toward
the informative dimension of communication. One has to be careful with the
basic distinction they suggest in terms of types of 'referential objects' of
communicative acts, however. It relies upon a differentiation between the extra-
subjective reality and the intra-subjective emotions. One should never forget that
language users can only provide information about their intra-subjective interpre
tation of extra-subjective reality (1.2.3.1).
Robinson's (1972) proposal, which covers fourteen functions, some of which
are even further subdivided (and the author himself states that "the scheme is
by no means complete, and probably not exhaustive" (ibidem: 49)), appears to
be a wild mixture of different phenomena (the indication that "the functions are
not all at the same level of analysis" (ibidem: 49) is an understatement). Many
of his functions are sub-functions of communication taken from various
dimensions of communication or to be situated at different levels of abstraction,
but some even relate to quite different phenomena (e.g., 'performatives' are in
his list, while they obviously are a formal category, not a function). It stands to
reason that such a conflation of dimensions and levels is not likely to enhance
our understanding of the functionality of language.
Yet there is one distinction in his typology which leads us to an interesting
phenomenon not to be found in other proposals. Among Robinson's functions
are 'expression of affect' on the one hand, and 'marking of the emitter for
emotional state, personality and identity' on the other. Though they might seem
to be akin, these are quite different things. While the former is a sub-function
of communication (cf. (24) above), the latter is an unavoidable implication of the
fact that language use is always performed by a person with an emotional life,
a social and cultural background, etc., which are permanently present in behavior.
Thus, the use of language for expressing affect also provides the hearer with
information concerning the speaker's emotional state etc., and the same is true
if language is used for requesting or instructing, or whatever other communicative
act. However, marking the emitter is not an organic function either: this is not
something which serves the speaker in pursuing the goal involved in the
communicative act, it is not a property language must have in order to be fit for
use in communication. Rather, it is a result of system-internal properties of the
'mechanism' producing (linguistic) behavior. Hence, this is not a function of
linguistic behavior at all. We have come across a 'side effect' or 'noise' produced
through language behavior (similar in status to the beating sound produced by
a heart pumping blood - cf. 1.2.2.1), and a functional analysis of language has to
take this factor into account, next to the functions of language proper. (See 1.2.3.)
Of course, the above does not imply that a speaker cannot try to control
or even consciously change the influence of emotional situation, character, social
and cultural background, etc. on behavior. Maybe the degree to which this is
possible can differ between individuals. Yet any speaker will try, for example, to
moderate or change the effects of emotions on language use if they may have
a negative effect on the communication (one had better not show nervousness
when applying for a job), or if (s)he does not want to show emotional involve
ment, or if (s)he wants to feign some emotion, etc. These are no longer simply
cases of marking the emitter however. For the speaker these are cases of (albeit
indirect) expression of emotion (even though they are often intended not to be
conceived by the hearer as a purposeful expression), resorting to the sub-func
tions of communication. Anyway, in many cases one will not care about whether
one's emotion, personality, etc. are marked in behavior. To a certain degree one
is not even able to avoid it, since part of one's 'internal life', and part of the ways
it is marked in behavior, are beyond control. Moreover, one's true emotions,
personality, etc. will also be reflected in the way one tries to control or feign
emotion, personality, etc. in behavior. Marking of the emitter in behavior is as
unavoidable as the presence of an emitter to produce behavior.
The proposal by Jakobson (1960) might be a case of a function typology
which ends up stating functions at a different level from the one intended. As
a reminder: Jakobson starts from six different constitutive factors of a communi
cative situation.
The addresser sends a message to the addressee. To be operative the message

requires a context referred to [...], graspable by the addressee, and either verbal or
capable of being verbalized; a code fully, or at least partially, common to the ad
dresser and addressee [...]; and, finally, a contact, a physical channel and psycho
logical connection between the addresser and addressee [...]. (Jakobson 1960: 21)
These determine six functions: context the referential function, addresser the
emotive function, addressee the conative function, contact the phatic function,
code the metalinguistic function, and message the poetic function.
The fact that Jakobson (like Bhler, whose typology does relate to the level
of the organic functions) conceived his function typology from the perspective
of constitutive and inalienable factors in communication suggests that what he
was really looking for is an answer to the remaining question mark in (23).20
Yet, instead of wondering what these factors imply for the constitution of
utterances a speaker uses for the purpose of communication, he has switched to
a completely different level of analysis in formulating his functions. He has
started discussing communicative acts in which one explicitly orients oneself
toward one of these factors (mainly in the sense that the factor becomes a refe
rential object in the act). Consequently, he has formulated sub-functions of
communication.
Hence, his referential, emotive and conative functions, which at first sight
might be taken to correspond to Bhler's three functions (1.2.2.3), are conceived
in such a way that they perfectly correspond to functions occurring e.g. in the
typology by Dik and Kooij (1979), i.e. functions of the type stated with respect
to (24) and (25) above. The same goes for the phatic function, which is akin to
the contact function in Dik and Kooij's system. As an example, Jakobson (1960:
24) cites (27).
(27) Are you listening?
Obviously, (27) only illustrates the fact that contact can be the object of a
communicative act. Nothing is said about the potential role of the channel in
constituting language use in communication in general. The same can be said for
the metalinguistic function (which is also to be found in Robinson's typology).
Jakobson illustrates that language can be used to perform communicative acts
about language. Yet that is nothing more than stating a sub-function of
communication (cf. also Busse 1975).
A similar criticism can be made for the three functions (the informative,
expressive and directive) formulated by Copi (1978: 55ff): the similarity with
Bhler's typology indicates that he was aiming at organic functions, yet they turn
out to be no different from Jakobson's, Dik and Kooij's, etc. functions.
This confusion of levels of analysis is also illustrated by the fact that
Jakobson and Copi more or less explicitly claim exhaustiveness with respect to
their functions, just like Bhler and Halliday. In fact, it must be possible to
formulate an exhaustive list of organic functions of language. Since Jakobson and
Copi have formulated sub-functions of communication however, there is actually
no question of the exhaustiveness of their typologies.
Though Hymes' (1962) proposal strongly relies on Jakobson's, the above
criticism does not hold for him. The only difference with Jakobson is that Hymes
replaces the factor 'context' by two factors, 'topic' and 'setting', and correspond
ingly states two functions (instead of one), viz. the referential and the contextual
function respectively. The motivation for this change is not made very clear.
Anyway, the basic nature of Hymes' functions is identical to Jakobson's. The main
difference is that Hymes seems perfectly aware of their nature. He explicitly
presents them as types of communicative acts, and he clearly states that he is not
really looking for constants (or universals) in communication (unlike what is
involved in searching for the organic functions of language) but that as an
ethnomethodologist he is interested in the variables in it (cf. also Hymes 1972).
He also explicitly denies exhaustiveness. And his points of dispute with Jakobson,
e.g. as to the existence of a one-to-one correlation of factors in communication
and functions of language (Hymes 1962: 120ff), mark the profound difference
in attitude toward the function typologies in both scholars.
We still have not considered the poetic function mentioned by Jakobson
(and Hymes), which under the name of 'aesthetic function' also occurs in Dik
and Kooij's and Robinson's typologies. This is certainly not a sub-function of
communication. There is much dispute about the precise definition of poeticity,
but it seems plausible to consider it a question of the way an utterance is
conceived with respect to both content and form. Hence, 'making' poeticity is not
a communicative act, it is inherent to a communicative act. It is a characteristic
of linguistic behavior: every utterance has poeticity. Still, one cannot say that it
is an organic function of language either, for it is not an activity or property of
utterances contributing to the achievement of their role in communication.
Rather, it is a matter of the way a language user performs a communicative act,
of the way one uses language to reach one's goals. This can depend on conscious
choices (poeticity can obviously be actively pursued in verbal behavior), but can
be determined unconsciously as well. As such, poeticity is no doubt closely related
to the factor of 'marking of the emitter' discussed above.
Of course, the way an utterance is constituted can become a topic as such
for the language user, in language games, in abstract (what might be called 'non-
informative') literary productions ('l'art pour l'art'), etc. In such cases the
language user tries to express the aesthetics of the linguistic medium as such, and
one could say that the verbal means become an object of communication, just
like emotions, weather and politics can be an object of communication. Also in
more 'realistic' literature this is at least one of the objects of communication.
Hence, showing the aesthetics of the linguistic medium is a sub-function of
communication. Still, poeticity as such is rather an inherent factor to language
use in general, hence Jakobson is no doubt right in claiming that every utterance
has a poetic value. This is the case for a newspaper article, an advertisement, or
a political debate, no less than for a poem. Although normally the degree of
poeticity of the latter is considerably lower than the degree of poeticity of the
former (but there are notable exceptions, such as Cicero's speeches). Poeticity
is obviously a gradual matter. (A similar criticism on Jakobson's poetic function
in Busse 1975.)
1.2.2.3. On the organic functions of language

So far, we have only found indirect suggestions as to probable organic functions
of language in communication (that language provides information, that it serves
intentions). Only few function typologies in the literature appear to contribute
more directly to insight into this matter. The main examples are Bhler (1918),
(1934) and Halliday (1976).21 Yet there are clear differences between these
proposals, so let's take a closer look at them.
Bhler's 'Organon Modell der Sprache' starts from the basic view that "die
Sprache sei ein Organum, um einer dem anderen etwas mitzuteilen ber die
Dinge" (1934: 24). The production of an utterance is a reaction to the stimulation
by an object ('Ding') within the field of perception of 'einer'. Hearing the ut
terance is for 'der andere' a stimulus to look at the object. This leads Bhler to
postulating three functions of language, viz. (i) 'Darstellung': it represents objects
and situations; (ii) 'Ausdruck': it depends on transmitters, and expresses their
personality; and (iii) 'Appell': it is an impulse to receivers, and orients their
internal or external behavior.
Halliday (1970b: 141, 1974b: 95) has criticized Bhler's analysis (and
function typologies in general) for being much too extrinsic. Such an approach
is, according to Halliday, sociologically or psychologically oriented, but cannot
throw light on the nature of linguistic structure. As an alternative, he advocates
an intrinsic approach in which each systematic contrast in a grammar is assigned
to some network of choices (according to the principle accepted in stratificational
grammar - Lamb 1966) which is related to a particular function. He distinguishes
three relatively independent networks (cf. Halliday 1976:14), each corresponding
to one of the following functions of language: (i) the ideational function: language
expresses contents, viz. speakers' experience of reality, and of their own internal
life; (ii) the interpersonal function: language is used by speakers to interfere in
the process of interaction, to establish and maintain interpersonal relations in
communication; and (iii) the textual function: language is used to establish links
with itself and with elements of the situation in which it is used. The textual
function is considered the internal function of language, and it is said to be
instrumental to both other functions.
Apparently, the only function for which there is a fair match between
Halliday's and Bhler's systems is the ideational/Darstellung function. Halliday's
interpersonal function is a junction of Bhler's Ausdruck and Appell. Halliday's
textual function is completely absent in Bhler's system.
In line with his general criticism on Bhler's approach, Halliday motivates
his junction of the Ausdruck (or expressive) function and Appell (or conative)
function by claiming that
The difference between these two is significant psychologically, but linguistically

it is very tenuous: is an interrogative, for example, a demand to be given informa
tion (conative), or an expression of a desire for knowledge (expressive)? It is not
surprising to find that expressive and conative are not really distinct in the language
system. They are combined into a single 'personal' function - or, as I would prefer
to call it, to bring out its social nature, an 'interpersonal' function. (Halliday 1976:
27)
No doubt, Halliday is right in claiming that it is irrelevant to wonder whether an

interrogative is a request for information or an expression of a desire for

knowledge, for a question is both. Yet this is not really the point of Buhler's
distinction between Ausdruck and Appell. There is a problem with the distinction,
but this has to do with the way Bhler approaches the entire phenomenon of
communication.
Bhler clearly takes an 'objective' point of view in analyzing communication.
He tries to escape from the perspective of the acting individual and considers
language from the point of view of a neutral observer isolated from the actual
communicative processes in the language users. Reichling basically makes the
same point when stating that
The communication process in language is not a mechanically closed and homo

geneous continuum [...]. There is only continuity on the level of the physical
phenomena. The language experiences of speaker and hearer develop outside of
it and over it. Precisely insofar as there is 'language' [...], there is duality; insofar
as there is physical cohesion, there is unity. In order for the communication to be
complete, the process must be continued consciously on each of the poles, speaker
and hearer.
The philosophical problem we are confronted with here, viz. how this unity
in disunity is possible, should not cause worry to any linguist. We should just accept
the fact [...]. Communication implies two distinct processes, one in the speaker, one
in the hearer. In between and apart from these there are undoubtedly sound waves,
yet speaker and hearer do not 'communicate' in sound waves, but in language;
physically there is 'contact' in sound waves. Bhler rightly looks for the explanation
of the uniformity of these language experiences in a sign function. Yet in his
formulation he situates it where it is not to be found, viz. in the sensory perceptible
world, and not in the experience, where he should look for it and practically finds
it. (Reichling 1935: 28-29; my translation)
Thus, Bhler has not really considered the functionality of language for the
language user. He has only inventoried some essential moments in a commun
icative situation, leaving open what happens in the interaction partners involved
in it. Moreover, in doing so he has focussed on just one phenomenon from the
perspective of two participants in the communication, viz. the phenomenon that
language refers to something, thus implying only one of the organic functions,of
language, viz. what Halliday has called the ideational function.23 And even in
his formulation of this function his objective point of view is apparent, for he only
refers to the extra-subjective reality, not to the intra-subjective reality (cf. the
discussion of Ogden and Richards 1949 in 1.2.2.2). Hence, even linguistic
reference is separated from the language user. Halliday does not commit this
mistake in defining his ideational function.24
Consequently, Halliday's discussion of the interpersonal function involves
matters which are not really at stake for Bhler. Yet, in so far as the distinction
between Ausdruck and Appell does touch upon the essential importance of the
point of view of the speaker on the one hand and the hearer on the other in
communication, Halliday's motivation for conflating the two elements in one
function does not hold water. The problem signalled in his discussion of the
interrogative is not really to the point. What is important is that in a communica
tive act there are two clearly distinguishable parties, viz. the speaker, who per
forms the act with some purpose, and the hearer who is to be reached and
influenced one way or another by the speaker. Hence, even from the perspective
of the speaker, the difference between his own position and the hearer's position
(each in terms of interests, goals, rights, values, face, etc.) is crucial. This may
be a 'psychological' matter, but precisely to the degree that it is cognitively
relevant, there is every reason to consider a communicative act to have a double
functionality in this respect. More generally, from a functional perspective
Halliday's claim that relevance at a psychological or social (i.e. cognitive) level
does not mean relevance at the linguistic level is hard to come by.
Halliday's textual function can hardly be considered a phenomenon relevant
at the level of organic functions. It is even doubtful whether one can call this a
function at all, in one of the senses of the word discussed in 1.2.1. Of course, it
is true that in communication language usually takes the form of a text. Linguistic
acts are not normally performed in isolation, but occur in groups, each one
embedded in a verbal context. Yet this is a matter of the formal appearance of
language: the phenomenon of the text is nothing but one aspect of the syntax of
language, and it is in this format that language fulfills its functions for the user.
Hence, as Halliday himself puts it, the textual dimension is instrumental for the
other functions, in the same way as language in general can be said to be
instrumental for its functions. Halliday is also right in claiming that the textual
embedding of an utterance has an effect on the organization of an utterance. Yet
this is not different from the fact that the sentential embedding of a constituent
has effect on the organization of that constituent.25
Criticism on Halliday's functions (as well as on his one-to-one approach to
formal system and function) can also be found in Van der Geest (1975), yet the
alternatives he advocates are not unproblematic either. He replaces the
interpersonal function by an intentional function, which represents the speaker's
interests in communication only. In view of the discussion above this is partly
appropriate. Yet Van der Geest does not introduce the required second function
relating to the fact that there is also a hearer involved in communication.
He also accepts a textual function, yet it is defined so vaguely that ap
propriate comments are hard to give. It says that an utterance is somehow related
to context and situation. In as far as this covers the same type of phenomena as
Halliday's function, the criticism given above remains valid. In as far as it also
refers to the fact that an utterance relates to the entire communicative situation
in which it is used, this is begging the question of what this relationship involves,
and what the functionality of language involves in this respect. It is not a formula
tion of a function of language however.
Finally, Van der Geest splits up Halliday's ideational function into a
referential function (an utterance refers to a certain state of affairs) and an
attitudinal function (it provides information about the speaker's interpretation
of the state of affairs referred to). This is questionable, however. If this is
intended to mean that an utterance refers to the extra-subjective world and to
the intra-subjective interpretation of it, Van der Geest is introducing an irrelevant
and even dangerous distinction. One can only refer to one's intra-subjective
interpretation of reality, since that is the only thing one knows (cf. 1.2.3.1). Of
course, somehow language must also refer to the extra-subjective reality, since
the intra-subjective interpretation is in some way based on it. Yet from the
perspective of the functionality of language for its user that is beside the point.
Alternatively, if Van der Geest is distinguishing between the fact that language
can refer to (the speaker's interpretation of) reality and to the speaker's attitudes
toward (his interpretation of) reality, he has returned to a distinction between
sub-functions of communication of the type made by Ogden and Richards (1949).
At the level of the organic functions, the only thing that matters is that language
provides information about things (reality, attitudes, and no matter what else).
1.2.3. Toward a typology of language functions
1.2.3.1. Communication and how to analyze it

The balance of our discussion of the literature is rather meager. We have only
found a clear indication that the (or at least a) role function of language is
communication.
(23) [Euser Acommunication]
[Elanguage ?]
The determination of the organic functions of language in communication

remains a problem, however. Though the proposals in the literature have
provided some partial answers (the observation that language refers, that it serves
the speaker's purposes, but also relates to the hearer's perspective), no clear-cut
typology has emerged. Let's therefore try to provide a more systematic answer.
An obvious way to approach the problem is an analysis of the phenomenon
of communication as such (i.e. the way also followed by Bhler and Jakobson).
Since language functions in communication, if we know what a language user
does when communicating we can determine the properties language must have
or the activities it must fulfill in order to be suitable as an instrument in
communication. Of course, we still have not dealt with the question whether
communication is the only role function of language. For the time being the
answer is not really important. The observation that language is used for com
munication implies that it must have organic functions allowing this, which have
to be revealed anyway in view of a functional analysis of the system. If afterwards
language would turn out to have further role functions, we can again determine
the organic functions it ought to fulfill for that purpose, and relate them to the
organic functions noted on the basis of our analysis of communication. (Cf.
1.23.3.)
In order to avoid the types of problems noted with respect to the analyses
discussed in the previous sections, some basic remarks on the nature of commun
ication and its implications for the further discussion are in order. Let's take
Franck's (1980) characterization of communicative acts as a starting point.
Wir knnen von interaktiver Handlung sprechen, wenn ein Handelnder A eine
Handlung Ha nach einer anderen Person B bzw. nach erwarteten oder vorhandenen
Handlungsresultaten von B ausrichtet und damit rechnet, da sein Verhalten von
B wahrgenommen und als Handlung gegenber B nach erwartbaren bzw. kon
ventionellen Kriterien interpretiert wird bzw. wenn B das Verhalten von A als eine
auf ihn bezogene Handlung Ha wertet. [...]. Kommunikative Handlungen sind
interaktive Handlungen mit einem Aussagegehalt, die so angelegt sind, da der
andere die Handlung aufgrund von (Symbol-)Konventionen als eine bestimmte
Handlung erkennt, die die augenblickliche Interaktionssituation fr ihn in
bestimmter Weise verndert. [...]. Jede Kommunikation hat das Ziel der Verstndi
gung. (Franck 1980: 36-37)
This characterization requires one amendment. Franck seems to be differen

tiating between interactive and communicative acts on the basis of the absence
or presence of an 'Aussagegehalt' or 'expressive content'. Yet in principle every
interactive act has an (expressive) content, otherwise one cannot speak of
'interpretation' of the act on the part of the receiver. This goes for acts by
animals as well as by humans, and by non-verbal means as well as by means of
language. As an example of a non-communicative interaction Franck mentions
a boxing match. Of course, the 'content' of a blow of the fist in this situation is
not immediately obvious. That is however only due to the conventions of the
game, which exhaustively determine the value of the acts performed (and as such
a blow does gain a very specific content in this situation). If a person consciously
and purposefully hits someone else outside the context of this game however, this
blow has an obvious content, and will no doubt be interpreted by the victim and
generate consequences for the relationship between the individuals involved. The
same goes for acts like kissing, hand waving, etc.
What does differ for various kinds of communicative acts are the modalities
of the expression of the contents, due to the nature of the medium chosen. There
are e.g. considerable differences in this respect (a.o. in directness) between a kiss,
a red heart drawn on a piece of paper, and the words 'I love you', although the
content can be identical. Yet this is a matter for a functional analysis of the
interactive system under consideration and does not have to affect our analysis
of the phenomenon of communication as such.
Communicative acts aim to achieve mutual understanding. In order to fully
understand what this means and implies for an analysis of communication, a
digression into human epistemology is necessary. If we axiomatically accept the
existence of individuals and of a world (what I have called the extra-subjective
reality so far) independent of them (Givn 1979: 317), the question to be raised
is: To what degree are individuals able to know the outside world? Is their
knowledge a photographic reduplication of this world (objective), or is it rather
a newly created reality only indirectly related (if at all) to the outside world
(subjective)? (See also Geysen and Nuyts 1982 on this matter.)
Perception seems to be more than a passive experiencing of stimuli from
the outside world. Information is processed at different levels, from the physical
to the conceptual, and processing implies manipulating the information depending
on the codes of each intermediate system. No doubt, this has influence on the
information itself, distorting it in certain ways. Conceptually, new information
does not remain isolated, but is actively integrated in and further interpreted
from the perspective of the knowledge existing in the individual, additionally
causing values, feelings etc. Hence, conceptualization (i.e. anchoring information
in memory) fully depends on one's background, which in turn is dependent on
genetic predispositions and earlier experiences with the environment. One's
mental world of knowledge, values, feelings etc. is at the same time the result
of and the basis for one's perceiving of and acting in the world.
This view is nicely illustrated in Piaget's theory of assimilation and ac
commodation as principles of the cognitive development of individuals. According
to Piaget one first tries to assimilate new information to the categories already
present in one's cognitive structures. This is so until contradictions start to occur
in the cognitive systems, disturbing one's equilibrium. Only from that moment
on is one forced to reorganize and accommodate the cognitive categories, so as

to resolve the contradictions and regain an equilibrium. (A more substantial
discussion in Piaget 1971, 1980a, b, Piaget and Inhelder 1972, Van Bezooijen
1976, Gillis 1983.) These principles were intended to account for child
development, but they seem applicable at the level of concept acquisition in
adults as well (although other characteristics of the acquisition process may be
quite different).
An impressive amount of examples illustrating the information-distorting
nature of perception can be found in Watzlawick (1976). He concludes
that the notions of 'order' and 'chaos' do not have an objective content, contrary
to the general opinion, but - like so many things in life - derive their meaning from
the point of view of the observer. (Watzlawick 1976: 62; my translation)
In Wittgenstein's words:
Ich bin meine Welt

Alles was wir sehen, knnte auch anders sein.
Alles was wir berhaupt beschreiben knnen, knnte auch anders sein.
Es gibt keine Ordnung der Dinge a priori
(Wittgenstein 1922: 5.63 and 5.634)
What remains of the outside world in the individual's knowledge can only be
indirectly related to the 'original', and a subjectivist (or 'experientalist', as Lakoff
and Johnson 1980: 185ff call it)26 view of human knowledge seems unavoidable.
(See also Van Dijk 1981.)
How does the subjective interpretation of the outside world in two indivi
duals relate, then? Obviously, this is neither a matter of identity, nor a matter
of fundamental difference. Except for identical twins, no two humans have an
identical genetic base, yet the overall principles of genetic determination in
humans are universal. No two humans have an identical 'acquired background'
(individual, social, cultural), in that the variety of environments in which they
have been, and the interactions they have had with their environments, can never
have been the same (even genetically identical twins raised in similar environ
ments end up being somewhat different). Still, from a certain level on, the reality
in which each individual lives obeys the same general principles and rules.27
Communication has to be situated in and viewed from the perspective of
the tension between community and difference, between variability and identity
in humans' subjective experience of the world. Communication is basically poss
ible because there is a partial correspondence between the subjective reality of
two individuals, and it is necessary because there is a partial discrepancy between
their realities.28 For personal and social reasons (ultimately for reasons of
survival of the individual and the species), humans have an interest in and/or
desire for a clear definition of their understanding of the world relative to others'
understanding, and, if possible, in/for closing the gap between different subjective
interpretations of reality. In principle, their aim is to find the largest possible in
tersubjective consensus about their interpretation of reality (and this obviously
includes the social and interpersonal reality as well as the physical reality).
These observations on the nature of communication have some clear conse
quences for its analysis. In principle, one could approach the matter from two
angles, viz. (i) an objective one, i.e. the perspective of a neutral observer not
involved in the interaction, or (ii) a subjective one, i.e. the perspective of one of
the interaction partners. Obviously, the above discussion implies that communica
tion can only be understood from the second angle (cf. also Parret 1982), not only
because interaction concerns the subjective views of the interaction partners, but
also because communication itself is subject to interpretation by those involved
in it. Each individual makes a personal interpretation of the acts performed and
the social and personal relationships holding within the situation, and can only
hypothesize about the communication partner's judgment of these factors (cf. also
Dik 1977: 215). The possibility to communicate, and the actual constitution of
the acts performed, fully depends on these interpretations and hypotheses. Taking
the objective perspective could only lead one astray, as was clearly demonstrated
by Bhler's failure to find the organic functions of language.30
Another element of the perspective to be taken in analyzing communication
which deserves some attention concerns the exact status of the elements 'speaker'
and 'hearer', and 'producing' and 'understanding'. In certain branches of
pragmatics one frequently finds a tendency to take an exclusively interpretive
perspective on communication, often on principled grounds, i.e. to analyze
communication only from the perspective of understanding linguistic acts. In
terms of hypotheses, this mostly results in a one-sided concern with hearer
performance and a total disregard for speaker performance. Some notorious
examples are Parret (1979b, 1980, 1982, 1986), Franck (1980), and Wilson and
Sperber (1984), and even Levinson (1983) tends in this direction. The source of
this attitude is not always obvious, and probably is not the same in all instances.
In some cases (especially among researchers having a 'minimalistic' conception
of pragmatics - cf. 1.3.1) this might be caused by the acceptance of the
Chomskyan model of language, in which pragmatic (and, thus, communicative)
aspects of language are automatically assigned an interpretive role vis-a-vis its
syntactic aspects. I will go into the plausibility of this view later. Of greater
interest in the present context is the type of reasoning Parret invokes for this
point of view.
Parret's argument is rooted in his concept of communication as a discursive

activity which is the expression of theoretical and practical reasoning.
the centrality of discourse-bound rationality and prescriptive-inferential reasoning

transforms pragmatics into a theory of understanding. Understanding has an asym
metric priority on production once one takes seriously the specificity of reasoning
processes in discourse. (Parret 1986: 755)
Of course, the view on the discursiveness of communication and the role of

rationality and reasoning in it is in perfect agreement with the above charac
terization of the nature of communication. Similarly, it is basically sound to claim
that communication turns around subjects' understanding of reality. Yet this is
only part of the truth, for rational reasoning is no less a matter of creating a
subjective view of the world. There is no contradiction whatsoever between
understanding and creation. On the contrary: both are essential and often hardly
distinguishable dimensions of all cognitive activity.
This matter should not be confused with the question as to the status of
and relationship between productive and interpretive performance in linguistic
behavior, however. Both 'modes' have elements of understanding and creation
in the sense just mentioned, being guided by similar subjective positions in
communication. As far as the relation between these modes in communication
is concerned, Parret's suggestion that interpretation (the perspective of the
hearer) has priority over production (the perspective of the speaker) is mistaken,
for, trivially, the acts constituting communicative behavior are made by the
speaker, not by the hearer. If nothing has been produced, there is nothing to be
understood. The speaker perspective always dominates the hearer perspective,
since the latter undergoes the act by the former. (Of course, this does not say
anything at all about the relationship between the interaction partners, each of
which is both speaker and hearer, constantly changing roles in communication.)
If one really wants to use these terms, then there is an 'asymmetric priority' of
production over interpretation. This is reflected in the fact that an analysis of the
functionality of linguistic behavior automatically takes a productive orientation,
i.e. the perspective of the one constituting the act, not the one receiving the act
(as is clearly demonstrated in the discussion of proposals in foregoing sections).
Interpreting an act is reconstructing the way in which it is functional for the one
performing it.
Of course, this has nothing to do with the question whether in analyzing
language performance in communication one should pay more attention to the
processes of production or interpretation. Both are indispensable aspects of
communicative behavior, and each normal language user possesses both capaci
ties. Hence, it is inappropriate to claim that "Lis a language of a community only
in case the members of this community understand the discursive sequence of

L" (Parret 1986: 756) (see also Parret 1982), as it would be inappropriate to
claim this with respect to the production of L. L is the language of a community
only if its members can produce and understand L.
1.2.3.2. Basic factors and organic functions

Thus, the question we ought to answer is: what are the elements in a com
municative situation playing a role if a language user constitutes an utterance to
perform some communicative act? Let's start from the most simple, prototypical
structure of communication, in which two speakers/hearers (S/Hs) act toward
each other in a dyadic exchange, and consider more complex or deviant inter
action patterns later.
In order for a communicative act to occur, one S/H must have an intention
with respect to the other, concerning an aspect of (the interpreted) reality
(henceforth the 'state of affairs', SoA) which is somehow relevant in the actual
situation (i.e. to both S/Hs). Relevance must be taken broadly: the S/H can be
looking for clarification about the SoA, can be eager to learn more about it, can
want to tell something about it, can want to change it, can want to respond to
one of these desires concerning the SoA from the side of the other S/H, etc. A
communicative exchange is initiated through an act according to this scheme by
one of the S/Hs; it is settled if the SoA is recognized as relevant and the
conditions of the situation as acceptable by the other S/H; and it lasts as long as
it remains relevant and acceptable for both S/Hs (of course, the pattern can be
continued by switching to a new SoA relevant to both S/Hs). Each individual act
in this situation is performed according to the actual S/H's interpretation of the
situation.
This means that an S/H has to take into consideration at least the following
factors in deciding how to act.
(i) The frame of reference

S/H has to be concerned with the relevant SoA occurring in the interpreted
reality/The notion of 'interpreted reality' may be somewhat misleading, for an
SoA can involve everything that can be thought of, including knowledge of the
outside world, but also emotions, attitudes, abstract or fictive states and events,
etc., i.e. things which have never been experienced in the outside world (cf. also
Dik 1982). Let's call this entire domain of referential objects in S/Hs knowledge
the 'universe of interpretation' (UI).
Deciding on whether and how to explicate an SoA to another S/H (hen
ceforth H/S) is not a simple matter (cf. Grice's 1975, 1989 maxims). S/H must
also have an hypothesis about H/S's knowledge of the SoA, and take into account
how the SoA in S/H's own and in H/S's knowledge relate (cf. the role of mutual
knowledge in discourse - e.g. Clark and Marshall 1981, Kreckel 1981, Smith (ed.)
1982).
(ii) The intention

S/H has to be careful to follow his/her intentions with respect to the SoA in
making clear to H/S in which way the SoA is relevant. Intention may be an
intuitively clear notion, yet it is difficult to grasp theoretically. Though it is used
frequently in the literature, a satisfactory definition is hard to come by. Even
semantic theories which accept the notion as central (e.g. Grice 1968,1969,1989,
Strawson 1971) often do not clearly explain it (but see Searle 1983). Cognitive
psychology does not bring us much further either. Anyway, intention is a 'layered',
hierarchical phenomenon (Searle 1983): an act can have several levels of inten-
tionality, starting with the intention that the act be performed, and ending with
the intention that the act achieve its ultimate purpose. Specifically in the case
of communication, there is also always a 'communicative intention', i.e. that H/S
would recognize the act as such, and mostly what might be called a 'per-
locutionary intention' (to use a term from speech act theory), i.e. the intention
that H/S would accept the SoA communicated to be true, provide information
about it, perform some change in it, etc. (cf. the speech act classifications).
Perlocutionary intentions can in principle be embedded endlessly. Of course, from
the level of perlocutionary intentions on S/H can also have more than one in
tention with an act at one level in the intentional chain.
In conceiving an act, one again has to be concerned not only with one's own
intentions but also with what in one's hypothesis could be the intentions of H/S.
What are the chances that H/S accepts one's intention? Does H/S also have
intentions concerning the SoA? Such matters can determine whether and how
one performs an act and pursues one's objectives with it.
Note that Franck's (1980) claim that intentions are not expressed in (lin
guistic) behavior and are therefore not retrievable for an H/S, is basically illogical.
It is true that not all intentions have to be made explicit. S/H can judge some of
them to be unimportant for H/S. Sometimes S/H may want to hide the real
intentions (which is an intention again, of course), and suggest fake intentions
(i.e. lying, misleading, etc.). Anyway, as far as there are intentions directly
relevant to the communicative act to be performed, S/H has to make them ex
plicit, for if H/S does not grasp them, (s)he cannot comply with them. Even
hidden or background intentions must somehow be retrievable from an act, for
the act must have something to do with them if it is to serve them.
(iii) The social relationship

Communication is grounded in the fact that an individual is a social being related
to and dependent on other individuals. This is reflected in each intentional act
performed. A communicative situation automatically creates a specific social and
interpersonal reality which offers possibilities but at the same times creates
restrictions and obligations concerning possible roles and related behavioral
patterns for the interaction partners. These are due partly to general socio-
culturally rooted conventions concerning role patterns in society, and partly to
the strictly intersubjective relationship between the interaction partners. This
involves a complex and often very subtle interplay of rules concerning matters
such as politeness and deference, power (natural or institutionalized), solidarity,
prestige, 'face', etc. (cf. e.g. Argyle 1973, Brown and Levinson 1979, Lakoff and
Tannen 1979).
Obviously, S/H has to evaluate the relationship with H/S in order to decide
whether to perform an act, and in acting S/H has to comply with the rules as they
apply to this relationship, in general and in the specific situation (unless S/H
wants to break them, of course). Again, S/H also has to form hypotheses about
H/S's conception of the relationship with S/H, and about H/S's expectations as
to the rules of behavior related to the values of the relationship. This can
determine what S/H can expect from H/S, and how to act in order to expect
most, or at least to avoid that H/S would refuse to accept the specific role pattern
imposed upon the situation by S/H on the basis of S/H's intentions.
The three factors mentioned so far correspond to the three components

of the communicative situation indicated in Bhler's 'Organon Modell'. As noted
in 1.2.2.2, Jakobson and Hymes have suggested some further components to be
taken into account, and at least one of them does have to be considered a
relevant factor for S/H's decision what to do and how to do it.
(iv) The setting

Communicative interaction is always spatially and temporally embedded in some
physically, culturally and socially specified situation, and it is influenced by the
characteristics of this situation or setting. This is not to be taken in the sense that
this situation can belong to the frame of reference for S/H. Jakobson's factor of
'context' is characterized this way, and as such it (or rather S/H's subjective
interpretation of it) is nothing more than a potentially relevant SoA to which
S/H's acts can refer. Hymes' distinction between 'topic' and 'setting' seems to
introduce an appropriate differentiation in the status of the situation. The former
concerns the situation as part of the frame of reference for S/H, like Jakobson's
'context', but the latter concerns the situation as an always present factor
influencing S/H's acts, independently from being a relevant SoA expressed in the
acts. Thus, the setting can work as an 'interfering' factor, as in the case of
interaction in a noisy context (a pub, e.g.), or as a 'tempering' factor, e.g. in
places which radiate power or authority (e.g. a church, a courtroom). Or, most
importantly, it can be a 'defining' factor, in specific institutionally ruled
interactions (e.g. a game, an examination, a lawsuit, a medical consultation, etc.).
S/H has to judge this factor, first of all to decide whether it is feasible or
acceptable to perform an act. If it is, it influences the way S/H can do so. Respect
for a church can e.g. hold one from talking even though one may have something
to say, or, if the intention is too strong, it can make one use reduced utterances
in a whispering mode. (On physical factors influencing communicative acts, cf.
e.g. Sapir 1912.) Institutional settings to a large degree determine which inten
tions interaction partners may have, which SoAs can be relevant, and which roles
the interaction partners can take (cf. e.g. Leodolter 1975, Ehlich and Rehbein
(eds.) 1983, Fisher and Todd (eds.) 1983, Redder (ed.) 1983).
The 'channel', which Jakobson mentions as a separate factor, can be
considered a special sub-element of the setting, for it co-determines the nature
of the setting. Thus, the channel can partly determine or limit the possible means
for communication, or can rule out certain aspects of the medium. For example,
the telephone excludes the use of non-verbal, gestural and visual behavior.
Written communication also excludes intonation. Moreover, the channel co-
determines the physical and cultural setting of S/H and H/S. In case of telephone
conversation it is temporally identical but spatially different. In written
communication it is both temporally and spatially different. (Cf. Schegloff 1979,
Settekorn 1981, stman 1987, De Rycker 1987.) Evidently, these elements co-
determine whether and how S/H performs an act.
As with the factors mentioned before, S/H also has to make hypotheses
concerning the way H/S evaluates the different aspects of the setting. Does H/S
have respect for a church? Does H/S feel obliged by the institutional setting, and
to what degree? What is the setting in which the person at the other end of the
telephone line is speaking?
The two remaining factors in Jakobson's typology, 'message' and 'code', are
not relevant for determining the functions of language. Concerning the former,
I can basically refer back to the discussion of the poetic and textual functions in
1.2.2. It is precisely for linguistic behavior (the message) that we are trying to find
out which factors determine its occurrence, hence this element is neutral in our
consideration (as is accurately suggested in Bhler's discussion).
Exactly the same remark can be made for the code. The code is inherent
in the medium of communication, in that it exhaustively determines its structural
character. Morse code, e.g., is characterized by a certain syntax, just like

language, and more specifically each single language. A message is always en
coded in a code (sic). Of course, S/H and H/S have to agree on the choice of
some code which they both sufficiently master to warrant understandability (i.e.
expressibility of an SoA). The actual choice does not only have to do with
understandability, however. Often the interaction partners can choose between
several languages or language variants, registers, sociolects or idiolects, etc. which
they both master, and then the choice can be determined by intentions, social
conventions, or the pressure of the setting (cf. e.g. the political and/or social
implications of the choice of Dutch or French in Brussels, of French or English
in Quebec, of English or Welsh in Wales, etc., or the situation boundness of the
choice of formal or informal language). (Cf. e.g. Clyne 1972, Heller et al. 1982,
Woolard 1987.) Hence, just like all other aspects of language, the choice of a
code is determined precisely by the four factors outlined above.
So far we have only considered the prototypical communicative situation.
Are these four factors also sufficient to account for more complicated or deviant
communication patterns? First, often more than two interaction partners are
present. If so, there seem to be two possibilities for S/H (with different in
termediate stages): either (s)he considers all H/Ss to be an entity (a group),
which means that the dyadic pattern remains basically the same; or (s)he
differentiates between the different H/Ss present, and then at least some of the
factors in the basic interaction pattern discussed above are multiplied with the
number of H/Ss. S/H may have to determine intentions toward each H/S (unless
they are identical for all of them) and take into account the probable intentions
of each of them, and S/H has to take into consideration potential differences with
each H/S's interpretation of reality (the relevant SoA), and the social and in
terpersonal relationship with each H/S as well as each H/S's possible conception
of it. Under certain circumstances it may also be necessary to take into account
differences in the setting of each H/S (cf. a triadic telephone conversation).
Finally, S/H may have to take into consideration this entire pattern as it exists
between the different H/Ss. (Cf. e.g. Clark and Carlson 1982 on the role of
mutual knowledge of two H/Ss.) Yet all of this is not simply a matter of multi
plication of interaction patterns, for the different patterns no doubt interfere in
S/H's behavior. (Often an S/H will communicate differently with an H/S
depending on whether they are alone or whether there are other H/Ss present.)
Anyway, the same basic factors remain valid, only their realization is much more
complicated.
Second, there are types of communication in which the direction of the
action is more or less one-sided. This is the case e.g. in speeches, in which there
does remain a minimal form of direct reaction in terms of the general attitude
of the public, and of questions or remarks from individuals in the public. In radio
and television broadcasts, as in other forms of mass media communication, in
which the channel imposes one-directionality, the manifest reaction from H/Ss
is as good as non-existent. Of course, there is a reaction in the H/S, but S/H
cannot observe it (except afterwards, in calls to the studio or letters to the editor,
e.g.). Yet in these cases it does remain necessary for S/H to make clear the SoA
(s)he has in mind, also on the basis of an hypothesis about what the average H/S
might know about it; S/H has to follow intentions in acting, probably also pre
conceiving possible intentions from H/Ss (it does not make sense to announce
a radio game to which listeners can respond by calling the studio, if one can be
sure that nobody is going to respond); S/H obviously has to respect the norms
inherent in the role-relationship with the average H/S; and mass-media involve
a specific type of setting (also determined by the channels they use), which again
has to be taken into account.
Non-active but 'participating' H/Ss present in the communicative situation
next to active H/Ss might receive more or less the same status as the public of
a speech (of course, there can be different degrees of involvement of H/Ss in the
conversation - cf. Clark 1987). S/H can develop an interaction scheme for them,
next to the interaction scheme for the active H/S(s), yet according to the same
principles as in one-directional communication. Purely accidental H/Ss, which
have nothing to do with the conversation itself, might also influence the
communication, but then as a sub-element of the factor 'setting'. Again, the basic
factors all appear to remain intact. There are only differences in how they are
realized.
We may therefore conclude that these four factors are always valid for all
types of communicative situations. Communicative success depends on the way
in which S/H manages to maneuver within the possibilities they offer and restric
tions they impose, in deciding what to do and how to do it. Moreover, they are
always equally important, although various phenomena may seem to contradict
this.
The fact that communicative success means realizing one's intentions might
be taken as a suggestion that intention is the most important factor. In fact, there
are many meaning theories which almost exclusively focus on the notion of inten
tion. This is mistaken, however. Intention may be the causing or 'driving' factor
of communication, but an intention never stands alone. It always has an object
(some SoA), and it is always oriented toward an H/S within some environment.
Hence, realizing intentions in the first place crucially depends on how one
manages to make clear what one's intentions are all about. Realizing intentions
also crucially depends on H/S's willingness and ability to comply with them, i.e.
to be cooperative, hence one absolutely has to take into account the interpersonal
and social obligations and the conventions inherent in the setting. Failing to do
so would not only lead to a communicative breakdown but could also result in
sanctions from H/S. 32
Though it is certainly appropriate to react against predominantly intention-
oriented meaning theories like speech act theory, one should also be careful with
a number of attempts, especially from an ethnolinguistic perspective, to go the
other direction and reduce the role of intention in communication. Duranti (1988)
argues that in Samoan culture the effect of communicative acts is often much
more important to the community than the intention of the acting person (see
also Du Bois 1987).33 This does not change the fact that communicative (or
whatever other) acts involve an intention on the part of the actor, however. As
Searle puts it:
[It is a mistake] to suppose there is some close connection, perhaps even identity,
between intention and responsibility. [...We] hold people responsible for many
things they do not intend and we do not hold them responsible for many things
they do intend. An example of the former type is the driver who recklessly runs
over a child. He did not intend to run over the child but he is held responsible.
And an example of the latter is the man who is forced at gunpoint to sign a
contract. He intended to sign the contract but is not held responsible. (Searle 1983:
103)
It is beyond doubt that cultural and social circumstances influence intentions of

language users and the ways they try to realize them. If one knows that in one's
community one will be held responsible for the effects of one's acts irrespective
of the intentions, this will obviously influence one's intentions, and will cause
greater caution in deciding on whether and how to pursue them. But intention
does remain an indispensable concept in explaining behavior even in the Samoan
cases Duranti discusses.
Du Bois (1987) argues that there are even cases of language use in which
intention does not play a role at all. He specifically refers to different types of
ritual speech, such as the sixteen cowrie divination of the Yoruba or symbol-
spinning among the Sisala. But again, he fails to distinguish between intention
and responsibility, or in this case more generally 'personality'. In any case, there
is intention behind the utterances spoken by the medium of an oracle at least
in three respects. First, the diviner must have the intention to respect the oracle
and speak the phrases which correspond to what the oracle 'says' (e.g. the
number of cowrie shells facing mouth up). Second, the phrases spoken by the
diviner have been coined some time by some person who has had an intention
with them, probably even directly concerning the divinatory process. Third, the
fact that in many of the societies which have divination the utterances triggered
by an oracle are not taken to be produced by a person, not even some abstract
deity, still does not mean that the utterances are not perceived by the consulter
as having an intentionality. They do have to be perceived as such, for otherwise
the consulter of the oracle would not act accordingly. As such, the communication
process between the consulter and the oracle/diviner cannot be understood
without the element intention as an inherent factor. (See Nuyts 1990b.)
In judging the importance of the different factors in communication, one
should also not be misled by the impression that sometimes certain factors
become predominant over the others, due to the situational circumstances. This
involves different cases.
One type occurs if one of the central factors in communication becomes
the topic of conversation. This may seem trivial, but it is precisely here that e.g.
Jakobson and Copi have been misled in formulating their function typologies.
One might for instance get the impression that in an examination in school the
frame of reference is the main factor in communication. This is appropriate only
in the sense that communication is centered around the topic of having
knowledge of reality. The decisions of the interlocutors are still determined by
all the factors in the communicative situation, the frame of reference being only
one of them. This involves an institutional setting which largely prescribes roles,
intentions and topics. The examiner has the intention to find out things, not about
SoAs as such, but rather about the examinee's knowledge of specific SoAs. The
examinee has the intention to show that (s)he has adequate knowledge of the
SoAs involved (or, sometimes, not to show that (s)he has no adequate knowledge
of them). The examiner has absolute rights to test the examinee's knowledge, yet
also has to respect general rules concerning fairness, courtesy, etc. The social role
of the examinee is one of subjectedness to the examiner.
It is easy to provide similar examples for the other factors in the com
munication. The social relationship between S/H and H/S can become the topic
of their conversation if they want to end it, settle it, improve it, etc., in an
institutional context (e.g. in case of divorce, in a lawyer's office) or not (e.g. the
dispute between a drunken husband and his wife late at night in their bedroom).
Intention can become the topic of an interaction if one of the interlocutors starts
questioning the intention behind the previous act of the other interlocutor. The
setting can become a topic if the interaction partners start questioning the rules
inherent in some institutional setting (e.g. a student complaining to a teacher
about the unfairness of examination procedures). In all such cases, the apparent
dominancy of one factor boils down to its being the relevant SoA in the
interaction. Each act concerning this SoA has to conform to the values for all
factors mentioned above, including the values of the factor which is the current
topic of communication (there is an intention in talking about an intention, one
has to act according to social norms in talking about social norms, etc.). Of
course, if a basic factor is under discussion, its values can change with each new
turn in the interaction. But this is not exceptional either: the factors discussed
above are always subject to constant reconsideration in the course of interaction.
Basically the same situation occurs if S/H indirectly communicates to H/S
a personal attitude toward the latter by speaking e.g. in an insulting, humiliating,
or impolite way without this being an element as such of the SoA communicated.
The social relationship can still be said to be a relevant SoA to S/H, next to the
SoA which the acts are really about. S/H's intention with respect to the former
(which is simply 'showing it') is realized by manipulating the conventions related
to the social and interpersonal relationship with H/S in very specific ways. Thus
once again this does not change which factors determine the acts by S/H, but only
their specific value.
Quite a different type of interaction in which one factor seems to acquire
a predominant position over the others is 'phatic communion' (Malinowski 1923)
or 'small-talk'. In such situations, it seems that "It is not what is said that matters
so much, as that something is said" (Sapir 1933: 16), and one might be inclined
to think that the factor 'social relationship' is absolutely dominant here. Again,
this is only partly accurate. Small-talk occurs in situations in which individuals
who know each other are together without there being an immediate purpose
for it. Often this creates a psychological need and/or social pressure to establish
or maintain contact, because silence can be felt to be hostile. The purpose is to
show that the individuals present remain on good terms (small-talk is unlikely
to occur between individuals who are hostile toward each other). Still, whether
small-talk occurs is strongly determined by factors like personality (people
strongly differ in 'talkativeness': some never say a word unless they really have
to, others never stop talking) and culture (silence certainly does not mean the
same everywhere: cf. Basso 1972 and several contributions in Tannen and Saville-
Troike (eds.) 1985). Nevertheless, it is true that it is mainly a psycho-social
phenomenon.
Yet this does not change the fact that all factors mentioned above exert
their specific influence on the communicative decisions to be taken by the
interlocutors. What we have here is a type of a more or less conventionalized
setting, the nature of which determines the specific values of the other factors.
Only a limited number of SoAs can be the topic of the communication (the
weather, memories, etc., in any case SoAs which are relatively unimportant and
are more or less certain to cause mutual agreement, or at least not to contain
the possibility of invoking dispute or conflict between the interaction partners).
The intention is to stay in contact and to show a positive attitude. The inter
personal pattern is one of communicative (though not necessarily social) equality,
while respecting the norms related to the general societal status of each
individual.
In summary, it seems that the four factors presented above are universal
determinants of communication, although the discussion has shown that there
can be rather complex interactions between them. Human thinking and acting
is not a flat one-level phenomenon, but is multidimensional and hierarchical.
There are constantly things in one's interpreted world which attract attention and
keep one mentally busy. Since they are the focus of one's attention, they motivate
actions. Actions have to conform to the set of conditioning factors noted above.
Obviously, the things which keep one busy influence the values of the condition
ing factors, and it appears that this can happen in quite different ways.
Anyway, the above leads us directly to an answer to the remaining question
in (23). If language has the role function to allow its user to communicate, it has
to fulfill the following organic functions:
(i) An informative function: language allows its user to make explicit the SoA
relevant in the communicative situation.
(ii) An intentional function: language allows its user to pursue intentions with the
relevant SoA.35
(iii) A socializing function: language allows its user to conform to the rules and
norms existing with respect to the interpersonal and social relationship with the
interaction partner.
(iv) A contextualizing function: language allows its user to adapt to the specific
characteristics of the setting in which the interaction occurs.
Thus we can complete scheme (23) as follows:
(28) [E u s e r Acommunication]
[Elanguage A informative ]
intentional
socializing
contextualizing
The factors mentioned are all subject to direct control by S/H when
constituting communicative acts that are functionally appropriate. As noted in
1.2.2.2, there is one further element to be taken into account in communication,
which has to receive a different status, however, viz. the inherent properties of
S/H. Each S/H is a unique constellation of in part constantly developing
emotional and rational properties (idiosyncratic, socially or culturally determined,
and universal ones) determining personality, which is furthermore influenced by
external factors determining temporary fluctuations in S/H's psychic situation
(intoxication, sleepiness, emotional excitement or disorientation, etc.). Without
being an element in the functionality of linguistic acts this factor unavoidably
determines the way the individual controls the 'constitutive' factors in the
communication and proceeds to comply with them in acting.36 Thus, it
determines general properties of human interaction such as the tendency to
minimize efforts to reach the goal (the inertia of all behavior), but also the typical
but temporarily fluctuating (idiosyncratic, social, and cultural) 'style' in which each
individual performs acts in order to reach objectives ('marking of the emitter').
The effects of this element are intricately interwoven with the effects of the
constitutive factors in S/H's communicative acts, and they may even counteract
the effects of the latter. The tendency to minimize efforts, for example, will often
have a counterproductive effect on the functional use of language. (On individual,
social and/or cultural style, see e.g. Bernstein 1971, Labov 1972b, Van Dijk 1978,
Gumperz 1982, Tannen 1987, and several contributions in Fillmore et al. (eds.)
1979, Gumperz (ed.) 1982.)
The fact that the personality and psychic situation of S/H is not a controlled
factor does not mean that it cannot become an element of direct relevance in
communication. The simplest case is when it directly becomes the relevant SoA
constituting the frame of reference of communicative acts. A more complicated
case is the one discussed in 1.2.2.2, when one attempts to control and manipulate
the personality factors as reflected in one's behavior. This is comparable to the
indirect use of the social relationship to communicate an attitude toward H/S
discussed above: the personality becomes a second SoA of relevance to S/H, next
to the SoA involved in the frame of reference. In both cases, the fact that
personality becomes the contents of one of the constitutive factors (the frame
of reference) in communication does not make it a constitutive factor in itself,
and does not change the fact that personality also remains an uncontrollable
factor determining the way S/H communicates about his/her personality.
Of course, if personality and temporary psychic situation are reflected in
communicative behavior, this means that although acts do not fulfill a 'psycho
logical function', for the receiver they do have a 'psychological meaningfulness'.
If we consider the above analysis from the perspective of the receiver, a
communicative act can be said to be meaningful in a direct and an indirect way.
The direct meaningfulness involves S/H's intention concerning the SoA under
consideration. This is what the act toward H/S is all about, and what H/S ought
to understand in order to grasp what S/H is doing. The indirect meaningfulness
involves the way S/H applies the rules and norms inherent in the social and
interpersonal relationship holding between S/H and H/S, and the personal
characteristics of S/H which find expression in the act. S/H's act normally is not
about these things (though the former is a controlled factor), hence they are not
of primary importance for H/S's understanding the act. Yet H/S always can and
normally will derive information from these factors too, and S/H can consciously
exploit this fact, as noted above.
1.2.3.3. Are there other role functions of language?

Finally, let's turn to the question whether communication is the sole role function
of language (cf. 1.2.2.2, 1.2.3.1). There are quite some instances of language use
in which one can dispute the appropriateness of calling them communicative. This
involves the many cases in which linguistic acts are not directed toward other
individuals. Chomsky (1975a: 53ff) claims that this type of language use, which
according to him is at least as frequent as interpersonal language use, cannot be
called communicative since it has a decisively different structure. He reproaches
Searle (1969,1972) for nevertheless including these usages under communication,
"since the notion 'communication' is now deprived of its essential and interesting
character" (Chomsky 1975a: 57). Yet upon closer examination the differences
between this 'individualistic' language use and real communicative language use
do not appear to be all that fundamental. Though they receive values specific
for the particular characteristics of the situation, the same basic factors can be
pointed out as determinants for S/H's decisions in forming the linguistic acts (cf.
also Firth 1952).
Obviously, the factor of personality is as relevant in this type of behavior
as it is in any other (this factor first of all determines how much, and on which
occasions, 'individualistic' language use will occur).
As to the functional factors, 'individualistic' language use always concerns
some SoA currently relevant to S/H, just like communicative language use. And
S/H does not start talking about this SoA without reason, i.e. (s)he must have
some intention with it. Yet the directionality of this intention is often different
from its directionality in communicative language use: it is not directed at another
H/S, but at oneself. This reflexivity implies that the types of intentions involved
in talk-to-self can be quite of a different nature. For example, one can have
problems in seeing the precise internal correlations of some phenomenon and
try to get it straight by formulating it and writing it down on paper. This is
certainly one of the main intentions behind scientists' making notes while doing
research. Another intention behind research notes can be to help the memoriza
tion of the SoAs involved. Other intentions behind talk-to-self can be attempts
to convince oneself of something, stimulate oneself, etc., or even simply talk
about what one is doing or thinking about in order to avoid silence, because one
hates silence, or is afraid, or likes to hear oneself talking, etc. No doubt, the exact
intention influences the construction of the 'talk-to-self acts'.
The factor 'social relationship' receives a different value in talk-to-self, too.
Maybe it is difficult to use the term 'social' here. Nevertheless, everybody has
attitudes toward and opinions about oneself, and can have changing feelings
about oneself. Some people like oneself, others hate oneself; some feel superior,
others inferior; some like who they are, others would like to be someone else;
there are moments in which one is proud of oneself, moments in which one
detests oneself, is angry at oneself, etc. This relationship with oneself no doubt
influences the way one talks to oneself.
Finally, the setting has its influence as well. Talk-to-self will be quite
different at home in the kitchen and in the bus to work. Cursing oneself for
having made a mistake will sound differently in the office amidst colleagues at
work and at night in bed.
Actually, individualistic language use is not necessarily purely reflexive.
Individuals very often privately simulate discourse with others. In that case, S/H
can imagine a fictive alter ego, or even an existing person, and start talking to
this 'H/S' as if they were involved in a real conversation. For example, if one is
about to make some important choice, one can set up a fictive antagonist holding
a different opinion than the one one is inclined to hold, and start a dialectic
communication process discussing the pros and cons with this antagonist. One
is talking to an 'imagined real' person or 'possible' person if one is re-doing a
conversation one had before with someone else, or if one is preparing a speech
or talk or anticipating a conversation to be held with another person in the
future. Especially in the latter cases, the different factors discussed will all have
the same type of impact on the linguistic acts by S/H as in real conversation,
although the fact that the 'addressee' is not really present obviously gives S/H
much more freedom.
In short, the differences between individualistic and communicative language
use do not appear to be fundamental at all as far as the basic principles of the
process are concerned. Hence it can only be a matter of terminology if one wants
to criticize Searle's use of the term 'communication' to refer to both 'types' of
linguistic behavior.37 In any case, the four organic functions which language
fulfills in communication remain valid for this 'deviant' use of language as well.
They seem to be good candidates for the status of 'universals of language use'
or 'pragmatic universals'.
The above discussion also shows that there is nothing wrong with the claim
that language is an instrument for thought, but that it is rather strange that this
claim is considered to be in conflict with the claim that language is an instrument
for communication (e.g. Harman 1975). Language certainly does support thinking,
in individualistic language use as represented above, but also in real communica
tion, to the same degree as communication in itself is an inalienable aspect of
human thinking. There is no opposition whatsoever here.
1.3. Functionalism and functional explanation
1.3.1. Functional explanation and the pragmatic perspective
If linguistic behavior is functional for its user, explaining linguistic behavior (in
the sense discussed in 1.1) obviously has to include an account of how the
organization of this behavior correlates with its functionality.38 This is a matter
of observational adequacy for a process model. A process model has to uncover
the systematicity in the relationship between the structural organization of
utterances and their meaningfulness in actual situations from the perspective of
the uses to which they are put. Moreover, it has to explicate the way the cognitive
systems responsible for linguistic behavior go about producing and interpreting
utterances in ways appropriate to the functional conditions inherent in actual
situations.
In view of the discussion in 1.1.2 concerning the 'black box' problem of
cognition, this fact implies that considerations concerning the functional di
mension of linguistic behavior such as those made in 1.2 should play an essential
role in directing the development of a process model. If it is clear that linguistic
utterances do not just have a structure and a meaning in isolation, but that
structures are meaningful (in different ways) in the light of what they are used
for in actual situations, then taking this functional dimension into account ob
viously provides a firmer basis for developing hypotheses about the black box
(since encompassing more facts about language) than just considering the
structure and the meaning of utterances in isolation, independent from their func
tional dimension. This means that functional explanations39 have to occupy a
prominent place as methodological guidelines for theory formation concerning
language in cognitive science. Since at least part of the organizational principles
of linguistic behavior and the cognitive systems responsible for it must be
motivated or relevant from the perspective of the functionality of the behavior,
it is crucial in the development of hypotheses about these organizational prin

ciples to reason in terms of how they could make sense from the perspective of
their functionality.
This is what I call a functionalist point of view in language research. More
specifically, functionalism can be characterized as the methodology making use
of functional explanation as a basic research strategy, which is the methodology
inherent to a functionalist or pragmatic language theory. In this specification the
term 'pragmatic' is used in a very wide sense, and in view of the 'classical', partly
terminological, partly theoretical dispute concerning the delimitation of the notion
of pragmatics, especially relative to the notions of 'semantics' and 'syntax', this
deserves some further explanation.
The use of the term pragmatics in present-day language research certainly
reflects the rather chaotic situation of the language sciences sketched in 1.1.
There is no uniform or generally accepted definition available, and though the
term is used throughout the different subdisciplines of the language sciences,
'pragmaticians' in different subdisciplines often appear to be unaware of each
other's research, let alone they would be trying to find a common ground. Symp
tomatic of the 'state of the art' is that the enormous amount of literature using
the term notwithstanding, there are hardly any textbooks aiming to provide a co
herent picture of what is covered by the notion. One of the few attempts to
provide a characterization of pragmatics, and no doubt the most prominent one,
is Levinson (1983). But one can seriously doubt its value as a genuine textbook.
(Concerning the following, see also Verschueren 1985, 1987a.)
One main problem is that Levinson - deliberately - only covers a relatively
restricted part of research calling itself pragmatic, viz. what he terms the 'Anglo-
American tradition'. This would be acceptable if the limitation would be based
on a clearly characterized conception of pragmatics, for then we would have a
textbook for at least one coherent domain of research calling itself pragmatic,
although we would still be left with the question how to relate this concept of
pragmatics to others, and whether it would not be possible to find a common
ground for all conceptions. Levinson does not provide the clear characterization
required, however.
Though he spends most of his first chapter considering possible definitions
of the notion pragmatics, Levinson's conclusion is that none of them is really
satisfactory. His discussion even reflects disbelief in the possibility of ever finding
a coherent unifying conception of the notion. He only retains some vague
specification of pragmatics as the study of language in context. This is obviously
insufficient. 'Context' certainly is a relevant notion in language research intimately
linked to the concept of pragmatics (see below), yet as such it is much too inde-
terminate and general to serve as a defining concept for the latter. In fact,
Levinson (1983: 22ff) himself admits that a characterization sufficiently clear for
his purpose is very hard to give. In the end, his characterization of the 'Anglo-
American' concept of pragmatics ends up being nothing but a survey of a number
of topics which happen to have received much attention in this research tradition
(deixis, implicatures, presuppositions, speech acts, and conversational organiza
tion). Levinson tries to defend the absence of clear criteria characterizing this
Anglo-American tradition by claiming that it is always difficult to define a
discipline or paradigm, and that in order to know what it represents one has to
take a look at what scientists working in it are doing. Yet one would be inclined
to think that researchers working in a discipline know what they are doing and
thus are able to provide at least some kind of criteria to delimit their discipline
or tradition from others.
In fact, there is little reason to believe that this limited Anglo-American
tradition of pragmatics or any other topic oriented conception (a 'component
view') of it can be seriously motivated as an independent field of inquiry. On the
contrary, there are good reasons to accept a 'perspective view' (Verschueren
1987b) of pragmatics as a basically functionalist approach to language in all its
aspects. Levinson argues that such a wide conception of pragmatics (which he
terms the 'continental tradition') is much too encompassing, since
such a definition, or scope, for pragmatics would fail to distinguish linguistic

pragmatics from many other disciplines interested in functional approaches to
language, including psycholinguistics and sociolinguistics (Levinson 1983: 7)
Yet why would it be necessary to avoid a definition of pragmatics which could

encompass not only research in linguistics, but also in psycholinguistics, socio
linguistics, etc.? In the light of the discussion in 1.1 this is rather to be taken as
a merit of the 'continental tradition', since it could enhance the possibilities of
finding a common ground for research throughout the many subbranches of the
language sciences. Levinson may not have faith in the possibility to reach an
integrative science of language, but that is not an argument to preclude attempts
in this direction by a priori ruling out a definition of pragmatics which could lead
to links between research in different disciplines (including research in the Anglo-
American tradition of pragmatics).
More important is the fact that, contrary to the component view, the pers
pective view of pragmatics fits naturally into the pattern of language research.
To make this point, it is useful to return to Morris' (1938) and (1946) discussions
of pragmatics, which already show the conflict between a limited and a wide con
ception of it (see also Parret 1979a, 1986). Morris (1938) defines 'syntactics',
semantics and pragmatics as three equal components of semiotics, their objects

being respectively
"the study of the syntactical relations of signs to one another in abstraction from
the relations of signs to objects or to interpreters", "the relation of signs to their
designata and so to the objects which they may or do denote", "the science of the
relation of signs to their interpreters" (Morris 1938/1971: 28, 35, 43 resp.)
Although all three are said to be on the same level, Morris' further discussion
shows that his concept of pragmatics is actually much broader than what could
be expected from a component of semiotics equal to syntax and semantics. In
fact, in Morris (1946) syntax, semantics, and pragmatics still form a triplet, yet
(following Carnap 1942) a hierarchical one with syntax being the least inclusive
and pragmatics being the most inclusive approach (Morris 1946/1971: 301ff). The
step from here to the view that pragmatics is a concept on a higher level than
syntax and semantics is not very big.
In fact, in Morris' definition of the notions, pragmatics as the study of signs
in relationship to their users has to include the study of what is involved in
semantics and syntax as well (as Morris 1938/1971: 64 himself recognized). For
the former is necessarily the study of what users of signs do with them, and how
they do it, and this precisely covers the study of how signs are put into structural
relationships and how these structurally organized strings of signs carry their
meanings in view of what the user does with them. Hence, it is perfectly natural
to define syntax as the study of the structural dimension of linguistic utterances
at the different levels of organization, and to define semantics as the study of the
meaningfulness of linguistic utterances at the different levels of organization. But
it is hard to define pragmatics as being involved with some third natural domain
of phenomena concerning language. As Levinson himself admits, it is difficult to
make a non-arbitrary distinction between matters which could belong to semantics
and pragmatics respectively within the domain of the meaningfulness of language;
and phenomena such as deixis, presupposition, implicature or conversational
structure all involve matters of structure of language (syntax) and meaningfulness
of language (semantics) alike.
Levinson tries to argue for a separate pragmatic component on the basis
of the already mentioned notion of context. If a complete description of the lex
icon has to contain an indication of the meaning of all words of a language, then
one necessarily has to refer to contextual information, since rather many ('prag
matic') words are only contextually definable. Semantics even has to be
dependent on pragmatics, so he claims: e.g., the exact meaning of a connective
in an utterance is context-dependent, which means that the truth values of the
utterance have to be made dependent on its contextual properties.
Yet these illustrations only show that an adequate description of the

structure and meaning of utterances is not possible without taking into account
contextual factors, or more generally factors concerning the usage of these
utterances. In fact, the context is important in considering nearly every aspect
of language, if only because one cannot know in advance whether it can have
influence or not. Though it is true that most of the phenomena discussed by
Levinson are among the most strongly context-sensitive ones, many other
phenomena not contained in Levinson's concept of pragmatics are context-sensi
tive, too. This goes e.g. for the choice of 'normal', denotative lexical entries, for
much of the word order of utterances, and even in phonology and phonetics
much is context dependent. Thus, context certainly does not lead to a domain
of linguistic phenomena clearly distinguishable from the domain of phenomena
belonging to syntax and semantics. To the contrary, context adds to the picture
one gets in considering aspects of language belonging to semantics or syntax (as
Levinson himself recognizes in considering why pragmatics has recently generated
so much interest), which only underscores the adequacy of the perspective view
of pragmatics.40
1.3.2. Functionalism vs formalism in linguistics
Functionalism is not generally accepted in language research, at least not within

traditional linguistics, where there is a rather sharp distinction between what is
frequently called a 'formalist paradigm' and a 'functionalist paradigm'. This
paradigmatic opposition is much less articulated or even non-existent in other
disciplines in language research. In language sociology or sociolinguistics, for
instance, there appears to be nothing of the kind, and also the large majority of
language psychologists can be said to be basically functionalist.41
Within linguistics, the opposition is rather old. One of the first clear
statements of it, which at once reveals what this is really all about, was made by
Karl Brugmann.
Die bis jetzt angestellten Errterungen ber die beste Art der Behandlung des
Satzes gehen im wesentlichen von dem formalen und dem logischen Verhltnis der
Teile des Gesprochenen aus, die Darstellung der Syntax [...soll...] die Glieder der
Rede an der Arbeit zeigen, den Zusammenhang aufweisen, in welchen sie sich
einfgen. Das betrifft alles die Form der Stze. Nun verkennt aber niemand, dass
die Form des Satzes sehr hufig nicht dem entspricht [...] was beim Sprechenden
der Grundtrieb fr seine sprachliche usserung gewesen ist. Ich will z.B., dass einer
mit mir gehe, kleide meine Aufforderung an ihn aber in die Form des Fragens: du
kommst doch mit? oder kommst du nicht mit? Es ergibt sich somit die Frage an die
wissenschaftliche Grammatik, speziell an die mit der Psychologie Hand in Hand

gehende Satzlehre: welche Mittel der Satzgestaltung stehen und standen einer
Sprache zu Gebote, um die mannigfaltigen seelischen Grundfunktionen, die zum
Sprechen drngen, zum Ausdruck zu bringen?
Das Seelenleben des Menschen in seinem schwer bersehbaren Reichtum
an Empfindungen und Vorstellungen strebt ja in sehr verschiedenen, die Unter
schiede der seelischen Antriebe so gut als mglich ausdrckenden Satzgestaltungen
sich verstndlich darzustellen. Durch die Sprache ussert der Mensch Lust- und
Unlustempfindungen, Wnsche, Bitten, Befehle, Beschwerden, erweist Vorstellung
en und Ansinnen ab, warnt, droht, verwnscht, bekundet Zweifel, tut Fragen und
beantwortet Fragen, stellt Behauptungen auf, urteilt ber Vorgnge, Gegenstnde
und Personen u.dgl. mehr, und das alles, jedes in seiner seelischen Eigenart, ringt
nach mglichst unmissverstndlicher Darstellung durch die Form des Satzes. Wie
verhlt sich da [...] die jeweilige Satzform zu der psychischen Grundstimmung, aus
der sie hervorgegangen ist? (Brugmann 1918: 1-2)
This characterization clearly shows that functionalists are taking more facts about
language into account, and are doing this in a more integrative way, than
formalists do. As far as syntax is concerned, both study the structural organization
of utterances, 42 yet formalists do this in isolation without considering how the
structure relates to what the S/H is doing with it (which in the first place implies
that they are not taking semantic matters into account), while functionalists
integrate considerations of structure and use (which at least implies that seman
tics has to be taken into account immediately, even if it is not described as such).
As far as semantics is concerned, both paradigms study the meaningfulness of
utterances, yet formalists mainly concentrate on truth conditions for the logical
patterns behind utterances without considering meaningfulness related to what
the utterance is used for, while functionalists study meaning from the perspective
of the uses of language.
Originally, the difference between the two paradigms was purely a matter
of methodology. This was clearly manifested in the structuralist era in syntax. The
mainstream of American structuralism was strongly influenced by positivism and
inductivism, hence on methodological grounds there was a refusal to consider
elements of meaning and functionality in analyzing the structural organization
of languages (though there are notorious exceptions including Sapir and Boas).
On the other hand, positivism did not affect most European structuralists, hence
they did develop a strong functionalist tradition in the investigation of language
structure (represented in a number of 'schools' - still existing today - including
the Prague, the British, the Dutch, and the French schools). (See also Firth 1955
on structuralism in the US and Europe.)43
In the post-structuralist era the distinction between both paradigms (which
are not clearly geographically bound anymore) is no longer only methodological,
however. This is mainly due to Chomsky's writings. Continuing the tradition of

American structuralism (2.2.3.1), but now starting from a clearly hypothetico-de-
ductive methodology, Chomsky has developed a system of rules and represen
tations for explaining syntactic structures completely independent of conside
rations of their functionality. Although there is no longer a taboo on semantics,
the part of the model accounting for it only receives a secondary status as an
interpretive device. Chomsky also pays lip service to the need for a model of the
functioning of language (of 'communicative competence'), but so far there has
not been a trace of it, and if it would be developed it would only receive a third
order position without affecting the semantic or the syntactic component. Hence,
there is no question of an integrative account of utterance organization and
functionality in Chomsky's model, contrary to what is the case in models
accounting for the structural organization of utterances which stem from the
functionalist tradition (e.g. Dik's Functional Grammar or Halliday's Systemic
Grammar). By claiming that his model has cognitive reality, however, and by
explicitly rejecting a functionalist conception of the cognitive systems of language,
Chomsky has also promoted the formalist-functionalist opposition to a principled,
theoretical matter. (Chomskyan formalism as a theory will be topic in ch.2.)
Of course, this raises the question what reasons formalists in linguistics can
have to reject the (albeit very simple and straightforward) argumentation for
functionalism given above.
Before going into this, however, in order to avoid misunderstandings it is
important to point out that formalism is not opposed to functionalism with
respect to the question whether one is using formalisms in linguistic description
or not. Every attempt at modeling an object requires formalization. Hence,
functionalists are using formal notations as well as formalists, only the nature of
the formal systems is basically different. Formalists often tend to reproach
functionalists for not reaching sufficient degrees of formalization, and especially
TG scholars use the high degree of formalization of their model to suggest that
their approach is scientifically much more valuable and theoretically much more
advanced than functionalist approaches. Yet, apart from the fact that it is simply
not true that functionalist hypotheses never reach a high degree of formaliza
tion,44 this matter is a very poor criterion in the discussion.
Formalization of a model is the most specific and precise way to formulate
it. In principle, adequate formalization is only feasible if one has gained a very
detailed picture of which types of representations and rules or procedures might
be involved in the behavior under consideration. This does not mean that one
cannot already make use of formalisms in early stages of developing hypotheses,
as a means to advance one's understanding of the problem. The advantage of
doing this (and one of the reasons why AI is so attractive in cognitive research)
is that one is forced to spell out problems in every detail, making explicit many
facets which would go unnoticed otherwise (Steedman and Johnson-Laird 1980).
Still, using formalization this way also holds a number of dangers. Firstly, one is
often forced to introduce ad hoc notations for solving problems without there
being any indication as to the theoretical implications ('kludging', cf. 1.1.3.2).
Secondly, formalisms easily lead to system pressure, i.e. they easily distract
attention from the object of investigation and start living their own life in that
their possibilities and limitations exhaustively determine what can be done in the
theory. This is the point where empirical science turns into logic.
Hence, the presence of a formalism does not say anything at all about
adequacy, neither of the formalism itself, nor of the theoretical views behind it.
The fact that TG is highly formalized only means that its proponents do prefer
to use the advantages of this research alley (one can wonder whether they have
managed to avoid its dangers), even for investigating a relatively small problem
area. Yet especially in such a case the meaning and possibilities of a formalism
have to be put in the right perspective (cf. also Steedman and Johnson-Laird
1978). The TG-formalism has been developed on the basis of an exclusive
concern with sentence structure and with complete exclusion of everything else.
As follows from the principle of internal descriptive adequacy (1.1.2), however,
the more limited a problem area and the perspective from which it is ap
proached, the higher the danger that it receives idiosyncratic solutions which will
turn out to be untenable once integrated into a wider framework. If on this basis
alone the risk that the TG-formalism will turn out to be inappropriate is high.
Given our present state of knowledge about language and cognition in general,
it is perfectly understandable that some researchers prefer not to pin themselves
down to any formalism yet.
1.3.3. A review of arguments against functionalism
Recent criticism on functionalism in linguistics is mainly due to members of the

Chomskyan school. Much of the discussion appears to have little concern with
scientific eagerness to find out about language and language research,45 but
let's concentrate here on what seem to be real arguments. A first set of
arguments concerns the status of the functionality of language as such. The claims
in this respect boil down to a rejection of the view that language is a phenome
non of a specifically functional nature, or at least to a denial that there is
anything interesting to be said about or learned from functional uses of language.
Koster (1983) claims that the view of language as a socially and culturally
embedded functional system is 'common sense', and therefore ought to be
rejected. Yet it is hard to see how the fact that a claim is 'common sense' can
be used as an argument against it. The observation that language is used for
communication is so straightforward that it has to be common sense. The fact
that an apple from a tree normally falls straight down is common sense too, but
there is no physical theory which can afford not to account for it, let alone there
would be one daring to deny it. The fact that the principles behind such simple
observations can turn out to be much more complicated than might appear in
'popular experience' does not change anything at all about the appropriateness
and importance of the initial observation. Koster also commits the traditional
mistake of suggesting that the views of language as a cognitive system somehow
represented in the brain of individuals on the one hand, and of language as a
socially and culturally embedded system on the other, are basically incompatible
and irreconcilable (cf. 1.1.3.1). If this confusion is eliminated from the presenta
tion, his argumentation against functionalism on the basis of Popper's (1972)
theory of the three worlds (which is actually seriously misrepresented) does not
hold water anymore.
More specifically, formalists often deny that the main function of language
is communication and claim that it serves a multitude of functions, a fact which
would make it impossible to use the functionality of language as a basis for theory
formation (e.g. Chomsky 1975a: 53ff, 1979: 85ff, 1980a: 229ff, Leuninger 1977,
Koster 1982b, c, 1983, Newmeyer 1983: 96ff). The main problem with these
claims is that they are simply conjectured without further discussion or
motivation.
Consider one of Chomsky's statements in this respect.
While it is quite commonly argued that the purpose of language is communication

and that it is pointless to study language apart from its communicative function,
there is no formulation of this belief, to my knowledge, from which any substantive
proposals follow. The same may be said of the idea that the essential purpose of
language is to achieve certain instrumental ends, to satisfy needs, and so on. Surely
language can be used for such purposes - or for others. It is difficult to say what
'the purpose' of language is, except, perhaps, the expression of thought, a rather
empty formulation. The functions of language are various. It is unclear what might
be meant by the statement that some of them are 'central' or 'essential'. (Chomsky
1980a: 230)
Obviously, this involves nothing but a simple disavowal of a huge amount of

literature (only Searle's views on this matter are explicitly mentioned in Chomsky
1975a - cf. 1.2.3.3). There is no further attempt to show why the rejected views
are untenable. Sometimes claims of this type are illustrated by means of concrete
examples which ought to show cases of non-communicative language use in some
meaning of the word communication, but this is hardly ever accompanied by a
clear definition of what is meant by communication, let alone by a discussion of
alternative definitions which might be used to account for the examples. Let's
take a closer look at some of Chomsky's examples, which are also frequently
referred to elsewhere in the formalist literature.
As a graduate student, I spent two years writing a lengthy manuscript, assuming

throughout that it would never be published or read by anyone. I meant everything
I wrote, intending nothing as to what anyone would believe about my beliefs, in
fact taking it for granted that there would be no audience. Once a year, along with
many others, I write a letter to the Bureau of Internal Revenue explaining, with
as much eloquence as I can muster, why I am not paying part of my income tax.
I mean what I say in explaining this. I do not, however, have the intention of
communicating to the reader, or getting him to believe or do something, for the
simple reason that I know perfectly well that the 'reader' (probably some computer)
couldn't care less. What my statements in the letter mean, what I mean - in one
sense - in making these statements, is not explicable in terms of what I mean, what
I intend, in writing the letter, namely to express support for people undertaking
resistance to the criminal violence of the state in more meaningful ways. Once, I
had the curious experience of making a speech against the Vietnam war to a group
of soldiers who were advancing in full combat gear, rifles in hand, to clear the area
where I was speaking. I meant what I said - my statements had their strict and
literal meaning - but this had little to do with my intentions at that moment.
(Chomsky 1975a: 61-62)
There is quite some vagueness in the presentation of these cases. Anyway,

the manuscript not intended for publication is a clear case of talk-to-self
discussed in 1.2.3.3, in which the intention must have been to help memorize
research results, or to use formulation as a heuristic means. There is no reason
not to consider this a case of communication. (The fact that Chomsky 1975b has
later published most of this manuscript only shows that he has judged it to be
suitable as a communication toward the scientific community as well.)
Unless Chomsky writes a letter to the Bureau of Internal Revenue only, to
satisfy his own conscience (in which case we have another example of talk-to-
self), he certainly does this because he hopes it will have some effect even if the
chances are extremely low. 'Having effect' includes at least that it is registered
as existing, and can be read by those involved. Thus, Chomsky must reckon with
the possibility that there can be an audience sometime. Or maybe Chomsky
writes the letter simply because he wants to show to his political relatives that
he has written it and, hence, is on their side prepared to do something for their
cause. Again, he cannot but direct the letter to the officials involved, even if he
is sure they will not read it. That is why Chomsky writes his letters with as much
eloquence as he can muster, according to the specific values of the factors
discussed in 1.2. Otherwise he could as well send a white sheet of paper, or do
nothing at all.
Exactly the same applies to the Vietnam speech. Even if he did not expect
anyone to listen to him, he must have considered it important enough as a sym
bolic act (i.e. a case in which "someone speaks honestly, merely out of a sense
of integrity, fully aware that his audience will refuse to comprehend or even
consider what he is saying" (Chomsky 1980a: 230)). This constitutes the most
basic intentional level which results in delivering the speech as if there would
have been a real audience. If one would have analyzed Chomsky's actual speech,
this would no doubt have revealed to what degree he was influenced by the real
fact that no one was listening and to what degree he has been able to keep up
the idea that he should speak as if someone was listening.
As a last example Chomsky (1980a: 230) mentions an "informal conver
sation conducted for the sole purpose of maintaining casual friendly relations,
with no particular concern as to its content", but this is nothing more than the
case of small talk as discussed in 1.2.3.2.
These illustrations reveal an extremely simplistic view of communication
and functionality and a refusal to look beyond first impressions. As far as syntax
is concerned, Chomskyans constantly argue for high degrees of abstraction in
hypotheses, but if it comes to the matter of functionality they suddenly lack even
the slightest sense for generality or abstraction. It is hard not to see this as an
expression of a lack of willingness to consider what discussions such as those
presented in 1.2 are really about. In any case, to refute claims as to the functions
of language it can hardly be sufficient just to present some rather vague personal
impressions of individual cases. A serious argumentation is feasible only if all
potentially relevant elements are taken into consideration.
But even if the argumentation in 1.2 would turn out to be wrong and it
would appear impossible to maintain that communication is the sole role function
of language, this still would not be an argument against functionalism. In the first
place, it remains the case that language is also used as an instrument for
communication (even in the most narrow definition of this term - this observation
has never been denied, not even by formalists). Consequently, the organization
of linguistic behavior still must be such as to allow this usage, and a process
model of linguistic behavior still has to show how this is possible. Hence, a theory
which only takes into account the communicative function would albeit be limited
(it would unjustly neglect other functions), but it would still have better chances
of (partial) adequacy than a theory which considers no functions whatsoever. (See
also Foley and Van Valin 1984: 9.) In the second place, it would be necessary
to develop a precise inventory and characterization of the other role functions
and corresponding organic functions of language, and the best hypothesis would
be one providing an integral account of how language can serve all of these
functions. To reach this best hypothesis, however, functional explanations are
again indispensable.
The argumentation against functionalism also involves the principle of
functional explanation as such. In the first place, these arguments often refer to
the status of this type of explanation in science in general, and in biology in
particular. Sometimes it is claimed that they have a bad reputation in the theory
of science (e.g. Koster 1982a). As a matter of fact, functional explanation has
received some criticism. One of the most famous is Hempel (1965), but see Nagel
(1979: 305ff) for a critique on his arguments. Another frequently quoted criticist
is Nagel (1961). He is mainly concerned with the question whether functional or
teleological explanations are different from non-teleological explanations or not:
the fact that teleological explanations are usually advanced only in connection with
'goal-directed' systems affects the claim that, in respect to its asserted content, every
teleological explanation is translatable into an equivalent nonteleological one. The
answer is clearly in the negative, [...], every statement about the subject matter of
a teleological explanation can in principle be rendered in nonteleological language,
so that such explanations together with all assertions about the context of their use
are translatable into logically equivalent nonteleological formulations. [...]. [...A]
teleological explanation does connote more than does its prima facie equivalent
nonteleological translation. For the former presupposes, while the latter normally
does not, that the system under consideration in the explanation is directively
organized. Nevertheless, [...] this 'surplus meaning' of teleological statements can
always be expressed in nonteleological language. (Nagel 1961: 421)
Yet, as Nagel (1979) himself admits, this only involves the rather trivial matter
of the formulation of the explanation, i.e. it involves teleological or functional
statements, and not the essence of the explanation type. As is convincingly shown
by Cummins (1975), and agreed on by Nagel (1979: 313ff), functional or teleo
logical explanations do have their own systematicity and are therefore not
reducible to other explanation types (this is also implicitly admitted in the above
quotation). Thus,
The use of teleological explanations in the study of directively organized systems

is as congruent with the spirit of modern science as is the use of non-teleological
ones (Nagel 1961: 422)
Teleological concepts and teleological explanations do not constitute a species of

intellectual constructions that are inherently obscure and should therefore be
regarded with suspicion. (Nagel 1979: 316)
A further comparison of functional explanations in language research and

in other disciplines (mainly biology) is only likely to strengthen the case for
functionalism in the language sciences. As far as the validity of the use of such
explanations is concerned, linguistic behavior perfectly fits the 'maximal' set of
criteria mentioned by Nagel (1979: 286) to qualify as a goal-directed or
teleological proces.
(i) Linguistic behavior has plasticity, i.e. communicative goals can be achieved
following alternate paths or starting from different initial positions and in quite
different circumstances (with quite different values of the factors specified in 1.2).
This does not have to imply that linguistic behavior must always be successful:
it can and often does fail to fulfill its functions, but that is fully in agreement with
the nature of goal-directedness.
(ii) Linguistic behavior is persistent, i.e. even in case of disturbance, internal or
external, it continues pursuing its goals (at least if the disturbance is not too
severe). For example, verbal communication is still possible even in a noisy
environment, with food in one's mouth, etc. Internal (e.g. pathological)
disturbance is often compensated for within the system. In aphasia, e.g., verbal
communication is continued and improved to a certain degree by compensating
for the lost behavioral functions, partly by increasing theflexibilityand widening
the activity (quantitatively and qualitatively) of non-disturbed functions, partly
by reinstalling (parts of) the lost functions in other system components.
(iii) In addition, linguistic behavior is intentionally directed, which according to
Nagel (1979: 278ff) raises it above any suspicion as a goal-directed process.
Since according to the theory of science functional explanations are valid for goal-
directed processes in general, they are ipso facto valid for language, too.
As to the structure of functional explanations in the language sciences as
compared to other disciplines, without going into an analysis of the literature
(which mainly concentrates on biology; cf. among others Ayala 1970, Toulmin
1972, Wright 1973, Mayr 1976: 383ff, Dobzhansky et al. 1977),46 the characteri
zation in 1.2.1 and 1.3.1 appears to largely correspond to Cummins' (1975)
determination of functional explanations and analyses (which he considers to be
valid for biology as well as psychology). Cummins' analysis is also compatible with
Nagel's (1979), contrary to what the latter suggests (Nagel 1979: 293), but the
point is that one should not confuse two different sorts of functional explanation
of an object under consideration.
The analysis of goal-directed or teleological systems47 often involves the
claim that their presence in an organism is not accidental but is determined
precisely by the positive effects they have for the organism. That is, their origins
and development toward their present form can be explained in functional terms.
(Cf. e.g. Dobzhansky et al. 1977: 498, Nagel 1979. This suggestion is often present
in linguistic functionalism too - cf. e.g. Dik 1986a.) This type of explanation is
quite different from the one discussed in the present chapter, however: while this
chapter is about 'methodological explanation', the above suggestion involves an
'object-inherent explanation' (cf. note 2). It is important to keep these two
matters separated. It is possible that an object has purely accidentally grown into
a functional system, and thus does not allow an 'object-inherent' functional
explanation. But that has no effect on the fact that, if we observe that
synchronically the object has a certain functionality, it is possible to use this
observation as a methodological guideline for developing a model of this object
in its present state. (Of course, especially for biological and cognitive systems,
including language, it is very unlikely that their growth would not be influenced
by functional pressure - cf. 2.3.) In any case, Chomsky (1980a: 231) is wrong if
he claims that "functional analysis is relevant only on the evolutionary level", and
that "functional explanation applies on the evolutionary level" (cf. also Chomsky
1979: 85ff, 1982: 29).
In some respects one also has to be careful with the comparison of func
tional explanation in biology and language research, however, for the different
nature of the objects in both disciplines makes for some differences in the
application of the explanation. Chomsky likes to compare 'physical' and 'mental'
organs.
We may usefully think of the language faculty, the number faculty, and others, as
'mental organs', analogous to the heart or the visual system or the system of motor
coordination and planning. There appears to be no clear demarcation line between
physical organs, perceptual and motor systems, and cognitive faculties in the
respects in question. (Chomsky 1980a: 39)
He also uses this in methodological discussions. For example, in a comment on

Searle he states that
there is no doubt that the physiologist, studying the heart, will pay attention to the
fact that it pumps blood. But he will also study the structure of the heart and the
origin of this structure in the individual and the species, making no dogmatic
assumptions about the possibility of 'explaining' this structure in functional terms.
(Chomsky 1975a: 57)
This cannot be used as an argument against functionalism in language research

for different reasons. Just like a biologist, a functionalist language researcher
studies the structure of the object next to its functionality without making
dogmatic assumptions about the relationship between the two (cf. below). But
both in language and in biological organs, structure and function are not
independent. The result of a biological investigation, e.g. of the heart, will always
have to be a coherent and integrated description of the heart as a functional
system with a particular form. Similarly, the result of a linguistic investigation of
utterance structure, e.g., has to be an integrated and coherent description of the
utterance as a functional system with a particular form. That is exactly what
functionalists are aiming at.
Actually, not a single biological theory of the heart will try to consider this
organ as a completely autonomous and isolated entity, contrary to the way
Chomsky approaches syntax. At least the real-time input and output of the organ,
and as such also the natural embedding and position of the organ within the
entire organism, will be taken into account. Hence, even in this respect Chomsky
cannot make an appeal to biology to legitimate his own methodology.
For the rest, there is one main difference between biological organs and
'cognitive organs', viz. the former are material and therefore observable entities,
while the latter are non-material and therefore (as argued in 1.1) not directly
observable. This makes quite a difference to the possibilities of developing
hypotheses concerning these objects. A biologist can concentrate on the structure
of the tissue of the heart valves, e.g., without at the same time considering the
function of the heart valves. But a language researcher cannot discuss the
structure of the lexicon or the word order component this way, because (s)he
cannot observe it like that. One can concentrate on the formal properties of
constituent organization in utterances and afterwards or separately consider the
functionality of utterances (not a single functionalist who objects to this), but then
one still has no hypothesis about the cognitive system producing or interpreting
these utterances with their specific constituent organization. To find out about
the organization of this system, an integrated consideration of form and function
is indispensable. That is what formalists by definition refuse to do.
Finally, it will be obvious that Koster's (1982a: 361) claim that "in really
successful sciences like physics [...] functional explanations do not play an
appreciable role" (my translation) is not likely to contribute much to the dis
cussion. This observation is trivial since physics does not deal with goal-directed
systems, contrary to the sciences dealing with living organisms. What is interesting
about such a remark is that it is an expression of an apparently very deep-seated
tendency in Chomskyan linguistics toward scientism, i.e. the view that all sciences
should strive for unification and to that end should orient themselves toward the
physical sciences (cf. Greenberg 1963,1971a for an analysis of this view). Another
expression of this tendency is the appeal to the 'Galilean method' (Chomsky
1980a, Koster 1983) which would be followed in TG.
Of course, this attitude completely neglects the profound differences

between objects of investigation, which force scientists in the positive sciences,
in biology, and in the cognitive sciences to adopt fundamentally different research
methods, and which make a reduction of these different sciences to each other
impossible. It is very well possible that there are basic principles which are
universal for all things being, but they will only be found if every science deals
with its own object according to the methodology required by it. It can be useful
for a scientist to take a look at and try to learn from what is happening in other
sciences, but only insofar as the peculiarities of the own object of investigation
are kept in mind. Hence, it is not unlikely that the 'Galilean method' in language
research is going to lead us away from Galilean insights in language.
Finally, it is often suggested that functionalism has a number of charac
teristics which make it dubious from the perspective of empirical science. Thus,
it is sometimes claimed to be absolute, dogmatic (cf. the quotation from Chomsky
1975a: 57 above), reductionist, and/or simplistic. Yet first of all, it is unclear what
'absolute functionalism' should be. Using functional explanations in no way
involves the suggestions that the object under consideration is 'absolutely
functional' (cf. also Dik 1983a: 5). In fact, the notion of 'absolute functionality'
makes no sense at all. A functional system always has a form, which offers
possibilities but at the same time holds restrictions for pursuing its goals. In most
cases a particular goal can be pursued by different systems. For example,
communication is possible through language but also through other non-verbal
means in humans and animals. As Greenberg puts it:
The same function can always be performed by alternate structures, though only
a limited variety of such structures is possible. Hence the mere fact that a particular
function is performed is not sufficient to account for the specific structural
arrangement. We therefore consider it in relation to its historical antecedents. We
also investigate functional alternatives on a comparative basis, in order to discover
the range of possible structures capable of fulfilling a given function.
The consideration of functional alternatives brings with it, inevitably, the
related concept of functional efficiency. We can compare diverse structures with
the same or similar functions with regard to the effectiveness with which the
function is performed. (Greenberg 1958: 78)
Though human language is no doubt the most sophisticated natural system for
communication, it still has a certain degree of efficiency and may be better suited
in some ways than in others for its purposes (cf. 2.3.3). Anyway, in this context,
what would absolute functionality be like?
Functionalism is not dogmatic either: it holds no apriorism whatsoever
about the structures of language or the functions of language, nor about the
correlations between the two. This is a purely empirical question. In trying to

develop a model of the cognitive systems determining linguistic behavior, the
functionalist cannot invoke hypotheses predicting effects which do not correspond
to observations concerning the form and function of this behavior. Hence, it is
very well possible that one will have to conclude that certain aspects of the
organization of linguistic behavior are not functional at all (cf. e.g. Labov 1987).
One can put it this way:
Saying that a certain feature of linguistic design or change cannot be functionally

explained is tantamount to saying that we have not yet been able to find a function
al explanation for that feature. (Dik 1986a: 22)
But in as far as it remains impossible in the course of further investigation to

ascribe a certain functionality to some item or process, its status in the model
will have to reflect this. Of course, in principle this is also a functional expla
nation of this item. Hence, the following statement for evolutionary biology could
be applied to language research as well:
Teleological explanations [...] have great heuristic value. They are also occasionally
very facile, precisely because they may be difficult to test empirically. Every effort
should be made to formulate teleological explanations in a fashion that makes them
readily subject to empirical testing. When appropriate empirical tests cannot be
formulated [...the scientist...] should use teleological explanations only with the
greatest restraint (Dobzhansky et al. 1977: 504)
The above also implies that there is nothing to the claim that functionalism
would be reductionist. Newmeyer argues that
If linguistic form were in any sense a direct reflection of extragrammatical prin

ciples, then there would be no need at all for language-particular syntactic state
ments. One presumably would need do no more than derive grammatical effects
from the relevant principles. However, this is impossible. (Newmeyer 1983: 101)
Yet once again, functionalists not less than formalists take into account the
empirical fact that there exists a language-specific syntax. They do not derive
grammatical form from extragrammatical principles, whatever these would be;
they observe the grammatical form of existing languages, just like formalists do.
However, unlike formalists they do not remain content with that observation but
try to correlate the form with the uses to which it is put. There is nothing extra-
linguistic about that either, for claims about the functionality of language are
based on observations concerning linguistic behavior as well.
Finally, functionalism does not have to be based on 'simplistic' assumptions
as to the form-function correlation at all. Most probably "there is a many/many
relation between form and function", as Newmeyer (1983: 111) justly states. But
that is hardly an argument against functionalism, contrary to what he suggests.
It is not clear what he means when stating that
generativists are hardly surprised to discover that grammatical systems, in keeping

with all other known natural systems, exhibit a mismatch between form and
function (Newmeyer 1983: 111-112)
Of course, language, just like any other functional system, can be, and often is
used non-functionally, and, hence, can fail to achieve its objectives. This fits
perfectly into the functionalist view of language, and functional failure is even
a very interesting field of inquiry since it can teach us a lot about form-function
correlations. Yet it seems that what Newmeyer really wants to suggest is that
there is no form-function correlation at all. If so, he should explain how he
believes it would be possible that language is systematically used for communica
tion, and he should provide at least one example of a natural system of which
it has been suggested ever before that, if it fulfills some function, there is no
correlation whatsoever between form and function.
If there is no simple correlation between form and function, this can only
mean that we have to look for a complex correlation. In fact, in view of the many
and intricately interrelated factors involved in the functionality of language (cf.
1.2) a complex correlation is precisely what we can expect. As Dik states,
a functional explanation of grammatical phenomena will typically not be based on

an assumption of simple form-function correlations, but will instead involve a
network of interacting requirements and constraints, each of which may be
understood in functional terms itself, but which interact in complex ways and in
a certain sense 'compete' for recognition and expression in the final design of
linguistic expressions. This competition between different interacting and counter
acting principles explains why there is no single optimal solution for grammatical
organization. [...] A non-simplistic view of 'functional explanation' will hold, rather,
that the organization of a natural language is a solution to a complex problem, the
'solution space' of which is circumscribed by a variety of interacting and
counteracting functionally motivated principles. A natural language, then, can be
seen as one of a set of possible solutions to a complex problem: the achievement
of inter-human communication. (Dik 1986a: 17-18)
Dik extensively illustrates this, and I refer to his exposition as a counter-balance

for the examples Newmeyer cites in an attempt to criticize the suggestion as to
a simple form-function correlation.
Actually, this matter of the form-function relationship also affects Halliday's
hypotheses, which do involve a one-to-one relationship between formal and func
tional categories in the grammar (cf. 1.2.2.3). Of course, Malliday is right in claim-
ing that one has to formulate concrete relations between both types of categories:
that is what functionalism is all about. His criticism on the extrinsic nature of
other function typologies such as Bhler's is unjust in that the latter also do imply
that the function categories suggested should be related to the formal categories
of language (and Bhler 1934 e.g. also includes an explicit attempt to do this).
Of course, as compared to Halliday's view most other function typologies can be
said to be extrinsic in that, unlike Halliday's, they do not immediately propose
a specific type of form-function relationship. That might be a problem for
Halliday rather than for the other proposals, however.
Halliday starts from an aprioristic assumption about the form-function
relationship and even makes the formulation of function categories and formal
networks dependent on this view. Yet there is no room for apriorisms in this
matter. There are observations on the form of language and on the functionality
of language, and an attempt to match them might in principle lead one to postu
late a one-to-one relationship between them, but this does not have to be the
case: formal categories in a grammar could as well turn out to be determined
by the 'highest common factor' of the functional categories. (Cf. also Van der
Geest 1975: 23-26.) As suggested above, the latter seems to be much more prob
able, and Dik's (1986a) examples certainly support this view. But Halliday's a
priori conception of the form-function correlation not only can be held respon
sible for the problems with his function categories (cf. 1.2.2.3), but also does not
allow him to see the multitude of factors which could determine an S/H's choice
for a specific formal means of expression.
Let's illustrate this point by means of two notions from Halliday's grammar.
First consider the notion of 'theme', which according to Halliday belongs into the
textual function network. It is perfectly plausible to define a theme as something
determined by the textual deployment of information. If so, its choice is
determined by the necessity to transmit information about the relevant SoA in
such a way as to make it interpretable by H/S. Thus, theme certainly also has to
do with the informative (or ideational, in Halliday's terms) function. But since
in most contexts there are different possible choices of themes, the actual choice
is no doubt co-determined by the intentions, but maybe also by what S/H
considers to be an acceptable choice for H/S, or a possible choice in the setting.
Hence, theme can also have to do with the intentional, the socializing and/or the
contextualizing functions.
For another example, consider deixis, which is again a matter of the textual
function in Halliday's system. It is straightforward that deixis first of all has to
do with reference to the state of affairs expressed, i.e. with the informative
function. Yet most languages offer different alternatives to refer to something.
For example, in person deixis there are often various options, the choice being
determined by what one wants to express (e.g. the intention to express a positive
or a negative attitude toward a person referred to can lead to the choice of
expressions like 'the man over there' vs 'him there'), by matters of politeness and
social stratification (e.g. the choice of jij or u "you" in Dutch), and by the setting
(when speaking about the king one sometimes can refer to him as 'he', sometimes
one is obliged to talk about 'his majesty'). Hence, the deictic system can also be
sensitive to the intentional, socializing, and/or contextualizing function.
To summarize, there do not appear to be serious arguments against a func
tionalist view in language research. In developing a model of the cognitive
systems and processes responsible for linguistic behavior one obviously has to ac
cept a pragmatic perspective, and research has to be based on a functionalist
methodology.
Notes
1. Of course, next to theoretical objectives language research also has a variety of appli-
cational goals, including contributions to language therapy, first and foreign language
teaching, communication technology, problems of international and intercultural com
munication, etc. Even though science has developed into an autonomous institution, given
the needs of society from which it emerged and the obligations this creates, these goals
should be the ultimate objectives. The point is that achievements in these domains
crucially depend on a profound theoretical knowledge concerning the research object. Such
knowledge can only be acquired through fundamental, non-applied research. Hence, the
applicational goals are second order, certainly in the case of the science(s) of natural
language, in which knowledge of the object still is rudimentary. (Which is not meant to
imply that one should not pay attention to matters of application, even in the present state
of the art!)
2. The notion of 'explaining' is used in different ways in scientific literature, also in this
study. In the present context, two uses have to be distinguished.
(i) On the one hand, 'explaining' means revealing connections in the object of investi
gation: it is showing how it is constituted and how it correlates with other objects in the
'surrounding reality'. A description (a model) can be said to be explanatory in this sense
if it 'accurately' (see 1.1.2) represents the internal and external organization of the object
of investigation. (This is to be understood in a wide sense: the 'object' can also be a
process, like the process of language acquisition.) Let's call this 'object-inherent
explanation'.
(ii) On the other hand, 'explaining' means arguing for a certain hypothesis concerning an
object of investigation, as a scientific act, based on all kinds of scientific criteria. This
concerns the strategies and principles used in trying to reach an adequate description of
the object. This use of the term is found in notions like 'functional explanation', 'simplicity
as an explanatory principle', etc. Let's call this 'methodological explanation'.
A more elaborate treatment of this distinction in Nuyts (1987, 1988). There, these types
of explanation were called 'ontological' and 'epistemological', respectively. But due to the
heavy philosophical burden upon these terms, I will avoid them here.
3. In the following this abbreviation covers the entire paradigm of Chomskyan Transfor
mational Generative Grammar, from the early versions to Government and Binding.
4. This only refers to the traditional, radically positivist behaviorists, like Skinner (1957).
Osgood's (1980) 'three-stage integrational, mediational' neo-behaviorism, e.g., is far away
from original behaviorism and can hardly be distinguished from other trends in recent
language psychology, except for the terminology used. In fact, one can wonder why Osgood
wants to stick to this label of behaviorism and to the (very misleading) terminology related
to it.
5. Chomsky clearly takes a different point of view: he is rather a materialist, considering the
current indirectness in investigating cognition (a problem which he accepts, at least in
principle - cf. below) to be accidental, due to underdeveloped insights. Cf. also Chomsky
(1980a: 107-109). A similar point of view is defended by Katz (1964).
A survey of different opinions on this mind-body problem in Block (ed. 1980). An
interesting view on the origins of the dichotomy in humans thinking about themselves in
Minsky (1968).
6. This source of information has hardly been used in language research so far, and it will
not get much attention in this book either. But see, e.g., Butterworth and Beattie (1978),
Argyle and Trower (1979), Beattie (1980), Bolinger (1983), Goodwin (1983), Heath (1984).
7. This principle seems valid for human knowledge acquisition in general, as Karmiloff-Smith
and Inhelder (1974) demonstrate.
8. This leads to the paradoxical insight that science can never reach its theoretical goal to
fully explain its object. At least, we cannot know whether we can reach it, since, even if
we do reach it, we will not be aware of it. Scientists' daily attempts to work towards 'truth'
can only be understood in terms of belief, for there is no absolute truth. Still, this does
not mean that no progress is possible in science: it is possible to enhance the probability
that descriptions are an adequate characterization of their objects of inquiry. See below.
9. Though it is kept upright in many psychological textbooks: Clark and Clark (1977), e.g.,
nearly exclusively survey experimental research.
10. This does not mean that the basic attitude toward them is identical in all models. For
example, TG advocates a limitation in principle to syntax, phonology, etc., on the basis
of the competence-performance dichotomy (i.e. the restriction is theoretical, cf. ch.2),
while in Dik's Functional Grammar (as in most other functionalist models) it is ac
knowledged that ultimately the grammar will have to be embedded in an encompassing
model of human communicative capacities (i.e. the restriction is purely practical - in fact,
in Functional Grammar it is gradually being resolved, cf. ch.3).
11. Among the notorious exceptions are Winograd's (1972) and Davey's (1978) attempts to
integrate the principles of Halliday's Systemic Grammar into their processing models. Also
speech act theory, Gricean maxims, and related semantically and pragmatically oriented
linguistic research is more frequently taken notice of in psychological or AI approaches.
12. It is remarkable how often psychologists present TG as representative for syntactic

research in linguistics, and, consequently, tend to reject the latter as a whole together with
TG. Cf. e.g. Clark and Clark (1977:3ff), Clark and Haviland (1977:16), Schlesinger (1977:
2ff), McNeill (1979: 281ff).
13. This does not seem to imply that Parret denies the existence of mind. In Parret (1979a)
this remains open, but Parret (1983) does provide clear indications when he says:
The psychological context as it is relevant to pragmatics, is not the whole 'life of the mind'
- to use a dangerous metaphor - but only that part of mental activity that realizes itself
in grammatically determined procedures of production and understanding of linguistic
sequences. The problem of the acquisition of language, the innateness discussion, and the
plausibility of the neurophysiological or biological foundation of mental activity in langua
ge use [...] are of no direct relevance to pragmatics. (Parret 1983: 97-98)
(see also Parret 1986). Note that this quotation suggests once more the mistaken sup
position that explaining in pragmatics and psychology are different things.
14. Parret (1979a) claims that his position concerning the irresolvability of paradigmatic
oppositions corresponds to current views in the philosophy of science. Yet in the
mainstream of the philosophy of science, the principle of paradigm shift is generally
acknowledged indeed. While Kuhn (1962) still mainly ascribed this to psychological or
social factors, the important role of genuine scientific argumentation has since been
recognized, too (e.g. Lakatos 1970, Toulmin 1970, 1972).
15. Some of the statements in the examples in this section may be obsolete from the per
spective of current theories in the sciences from which they derive. This is of no con
sequence for the argumentation, however, since the only thing that matters here is what
types of phenomena the notion function can in principle be used for.
16. Though Nagel (1979) is only referring to biological functions, his expositions may be
generalized to functions in the human sciences in many respects.
17. Thus, one can have a chain of goals, ultimately ending at the maintenance of the indi
vidual, although the distance from the actual linguistic act to this ultimate goal can be
very long. In any case, one should not believe that goal-directedness is a simple or
inarticulate matter - cf. 1.2.3.2.
18. One might be tempted to remark that (22) does not specify the highest level at which it
is relevant to talk about language functions: not the individual language user, but the
linguistic community is the most encompassing organism for which language has a role.
I can refer back to 1.1.3.1 for an answer. Though there is no doubt about language having
a function for communities, one can never avoid to consider this from the point of view
of individual language users. It is always an individual speaking for a community, and
language only fulfills functions for a community through the filter of the cognitive
interpretation of this community and its interests by the individual language user. The
functions of language for a community are the functions of language for a socially
behaving individual.
19. Though speech acts are not normally called 'functions', their relationship with them is
apparent if one considers Searle's (1976:1) initial question, viz. "How many ways of using
language are there?" They concern things an individual can do with language. Other
proposals for classifying speech acts include Bach and Harnish (1979), Ballmer (1979),
Zaefferer (1979), and Ballmer and Brennenstuhl (1981). Also Dik's (1986a) proposal is
akin to speech act classifications. A criticism on the principles of most attempts to classify
speech acts in Verschueren (1983).
20. Of course, Jakobson distinguishes more factors in the communicative situation than
Bhler. On the relevance of these additional factors for the analysis of the organic
functions, see 1.2.3.2.
21. Halliday has presented his view on the functionality of language in numerous places, also
including Halliday (1970a, b, 1973, 1974a, b, 1979).
22. Halliday (1974a) has tried to show that the theory of 'functional sentence perspective'
(FSP) of the Prague school starts from the same set of functions. In fact, FSP distinguishes
three 'strata' in the linguistic system, viz. those of the grammatical structure of the
sentence, the semantic structure of the sentence, and the organization of the utterance
(Dane 1964, and several contributions in Dane (ed.) 1974). Yet FSP itself has never
related these strata to functions, and the correspondence might not be as evident as
Halliday would like it to be. Within the Prague school, there does not appear to be one
generally accepted view on the functionality of language. Bhler's typology appears to have
been very influential (Horlek 1964, Vachek 1966, Dane 1987), but there are also many
diverging views. (Cf. Horlek 1964 for a short overview.) Most of them are difficult to ac
cess for linguistic reasons, but Horlek's (1964) exposition remains purely programmatic
and impressionistic, and focusses exclusively on sub-functions of the poetic function. It
is not unlikely that Horlek's (1964: 45) conclusion typifies the situation in the Prague
school in general: "Il rsulte sans doute assez clairement de cet expos, qu'une thorie
systmatique des fonctions de la langue devra encore tre faite." (See also Dane 1987.)
23. This brings Bhler close to a strong (but biasing) tradition (which persists even today in
many semantic theories) to consider reference the most important or even only basic
'activity' of language. Cf. e.g. Dempe (1930, 1935).
24. Busse (1975) provides a survey of other criticisms (often in the same vain) on Bhler's
typology.
25. The core of the problem with Halliday's discussion of functions is precisely the extremely
intrinsic approach he advocates. His focus on the linguistic system as such has no doubt
distracted his attention from the real question, viz. what language does in communication.
And it has led him into a much too simplistic view on the form-function correlation. See
1.3.3.
26. Of course, Lakoff and Johnson (1980: 226ff) are right in stressing that one should not
exaggerate subjectivism: there still exists a relationship between reality and knowledge,
even though its exact shape is not and most probably cannot be known.
27. This is not to say that reality does not change. I subscribe to Toulmin's (1972) stressing
the fundamental historicity and variability of all existing things (human and non-human).
Yet in itself this is a principle valid for all humans, too, and the elements of the changes
are identical for all environments. There is always a certain degree of variability and a
certain degree of identity, in whatever you consider.
28. Of course, the more and the larger the differences, the greater the need for communica
tion, but also the harder it becomes. This is basically the problem of social stratification,
of intercultural and international communication, etc.
29. Of course, this characterization has to be interpreted broadly. Orders, e.g., can also be
seen as a matter of consensus over an interpretation of reality. Depending on the social
relationship between the interaction partners, the one receiving the order will only
perform according to it if (s)he agrees with the sender about the appropriateness of the
intended change in reality. Even in case the social situation does not allow a refusal to
obey an order, e.g. in a military context, communication is a matter of consensus: the one
receiving the order is supposed to accept that the view of reality of the one in command
is appropriate, and to accept it unconditionally.
30. This seems to me to be one of the main elements of difference and continuity at the same
time, between Wittgenstein (1922) and (1958). Wittgenstein (1922) has been trying to
analyze language as a mirror of reality. But which reality? His conclusion that "Wovon
man nicht sprechen kann, darber mu man schweigen" (ibidem: 7) seems to signal that
he has become aware of the fact that it makes no sense to try and relate language to an
objective world: we cannot talk about this world because we cannot know it.
Meine Stze erlutern dadurch, da sie der, welcher mich versteht, am Ende als unsinnig
erkennt, wenn er durch sie - auf ihnen - ber sie hinausgestiegen ist. (Er mu sozusagen
die Leiter wegwerfen, nachdem er auf ihr hinaufgestiegen ist.)
Er mu diese Stze berwinden, dann sieht er die Welt richtig, (ibidem: 6.54)
Consequently, Wittgenstein (1958) consistently starts from a subjectivist (but not a
solipsist!) perspective in analyzing language. He has thrown his ladder away.
31. Thus, a complete model of the linguistic capacity of language users must grasp both
dimensions, in equal rights. Although it is partially right (but incomplete) to claim that
"the production of discourse should be seen as a procedure of understanding" (Parret 1986:
756), it is mistaken to claim "that the concept of discourse production should be shaped
on the basis of the model of the concept of understanding" (ibidem), just like the opposite
claim would be mistaken. Cf. 3.2.2.
32. On notions such as 'obligation', 'convention', 'cooperation', and 'sanction', see e.g.
Wunderlich (1974), Franck (1980).
33. Actually, this is often true in Western culture as well. An employee can be fired because
of a mistake, even though (s)he did not commit it intentionally. A minister can be forced
to resign because of mistakes made by the administration for which (s)he is responsible,
without the minister's intentions having any influence on this.
34. I prefer to use the term 'informational function' rather than the more often used term
'referential function' because the latter is much too reminiscent of reference to the extra-
subjective reality. As discussed above, language always refers to the subject's interpretation
of reality in the widest sense of the word (this is true even for deictic elements).
35. I prefer to use the term 'intentional function' rather than the term 'psychological function'
(which I have used before - Nuyts 1983a), because the latter might unjustly suggest that
elements of personality also belong to this dimension. See the discussion of Robinson's
'marking of the emitter' in 1.2.2.2.
36. Note that the characterizations of style and of poeticity (cf. 1.2.2.2) show a very clear
resemblance. Both are matters having to do with or resulting from the way linguistic
means are used to achieve goals. This corresponds perfectly to the intuition that both
phenomena are closely related, and to the practice in poetics to handle them in a mutual
relationship.
37. Actually, Searle (1983,1986) has changed his view on this matter, in the wrong direction
it seems to me. See Nuyts (1990b) for a discussion.
38. For the sake of simplicity of the exposition, the notion of 'functionality' will henceforth
be intended to cover the element of style as well, even though this is strictly speaking not
an element of the functional dimension of language.
39. 'Explanation' is used here in the sense of 'methodological explanation' - cf. note 2.
40. The above discussion of the notions syntax, semantics and pragmatics obviously concerns
the meta-level of research domains or paradigms in the language sciences, yet these
notions are traditionally also used to refer to phenomena within the object-domain of
language as such. Thus, syntax also refers to the structural properties of language (mainly
at the sentence level), semantics is often used as a synonym of meaning, and pragmatics
occurs e.g. in 'pragmatic implicatures', 'pragmatic particles' (Levinson 1983), 'pragmatic
functions' (Dik 1978a), 'pragmatic information or knowledge' (Schank et al. 1985), etc.
Since syntax and semantics as fields of inquiry have a more or less clearly definable object-
domain, it is no problem to use these terms both at the meta-level and within the object-
domain. Yet if pragmatics is not the study of a specific domain of linguistic phenomena,
one can wonder whether it is profitable to use this term in the object-domain as well. At
least, one has to be careful in doing this, the more since the term has all too often been
used as a cover for phenomena which were hard to grasp (the 'pragmatic wastebasket').
41. This does not mean that all language psychologists take functionalism in all its conse
quences (cf. also Ninio and Snow 1988). Very often language is only seen as a device for
transmitting information, i.e. only the 'informative function' is taken into account. Some
language psychologists even take seriously a number of claims made by formalist linguists
(e.g. the competence/performance dichotomy), even though upon closer examination these
claims appear to be basically incompatible with a functionalist view (cf. ch.2). Still they
can be said to be basically functionalist since they are trying to model (aspects of)
language use (performance).
42. It is a basic misunderstanding to suggest, as e.g. Miller (1986: 195-196) does, that
functionalists do not study the form of utterances. Cf. Nuyts (1986b: 226).
43. Note that one cannot compare, let alone equate functionalism and behaviorism, contrary
to what is sometimes suggested. Behaviorists can be called functionalists only in the sense
that they do not just consider the structure of behavioral products, but try to account for
this structure on the basis of their use. This is the end of the comparison, however, since
for them behavior is only a matter of conditioning. Their positivism makes it impossible
for them to talk in terms of functions and meaningfulness beyond purely superficial
correlation.
44. For example, Halliday's Systemic Grammar reaches a considerable degree of formalization,
and it has even been used as a basis for AI models for language processing (Winograd
1972, Davey 1978, among others). Moreover, since most of AI research in general is
basically functionalist in orientation (cf. Dresner and Hornstein 1976, Schank and
Wilensky 1977, Winograd 1977a) functionalism certainly has reached the highest degree
of sophistication in matters formalization conceivable at present.
45. The manipulative and biasing character of the statements of some formalists in this respect
(a good example is Koster 1982a, b, c, 1983) raises the suspicion that they are unwilling
to enter a fair debate. Often functionalist points of view are flagrantly misrepresented or
ignored, and criticisms all too often contain nothing but a pure (and aggressive) denial
of alternative views, without argumentation or motivation. For some the only aim appears
to be to defend the monopoly of a specific model at any price, even the price of scientific
integrity. Maybe this is an illustration of what Kuhn (1962) had in mind when discussing
the way endangered scientific paradigms defend themselves in times of scientific revolution
and paradigm change. See also Nuyts (1983b).
46. It goes without argumentation that Koster's (1982b) suggestion that functional explana
tions in biology are on their way back is patently false.
47. As Mayr (1976: 395ff) remarks, the term 'teleological system' can be misleading, for from
a synchronical perspective a system cannot be called 'goal-directed'. This only makes sense
from a historical perspective, if the system does develop in the direction of an
improvement of the facilities it offers in the framework of the functions it fulfills. From
a synchronical perspective the term can only be intended to refer to a system which
produces goal-directed behavior. Unfortunately, as Mayr indicates there is no good
alternative term which could eliminate this unwanted ambiguity.
2. Linguistic Knowledge and Language Use:
Chomsky Revisited
2.0. Introduction
In order to get a general picture of the main properties of the cognitive systems
determining linguistic behavior, let's start by considering the principles of the
Chomskyan formalist conception of them and confront these with some basic
views fitting into the cognitive-pragmatic theory. In 2.11 will consider the general
structure of Chomsky's hypotheses. 2.2 discusses some major principles of the
adult language user's cognitive infrastructure, and in 2.3 the matter is further
underpinned by considering the properties of ontogenesis and phylogenesis of
the human linguistic capacity.
2.1. The structure of Chomskyan theory
Of all major linguistic theories developed so far, Chomsky's has no doubt made
the strongest claim as to cognitive plausibility, and has had the widest influence
in language research in general (and even beyond that). Chomsky has strongly
contributed to language psychologists' final break with hard core behaviorism,
and has invoked an entirely new orientation of their research, which was
henceforth rather called 'psycholinguistic' (cf. Hrmann 1978).
Yet, while in the early sixties Chomskyan theory rapidly gained supporters
both in linguistics and language psychology, from the late sixties on it has caused
a strong reaction, not only in linguistics and language psychology, but also in
language philosophy and socio- and ethnolinguistics, very often because of its
lack of attention for language use and functionality. The criticism and the argu
mentation for alternative views has hardly had any influence on the Chomskyan
paradigm, however. Criticism has consistently been rejected as irrelevant or inap
propriate (cf. e.g. Chomsky 1975a, 1980a). Alternative views are either com
pletely ignored or classified as mistaken or theoretically uninteresting (not always
on the basis of clear arguments and often in particularly unkind terms).
Examples are Dresher and Hornstein (1976) on AI; Lightfoot (1982: 172ff) on
theories of language acquisition; Koster (1982a) on an alternative syntactic
model; Chomsky's repeated denial that there exist alternatives for his innateness
theory; or Chomsky's (1979: 148ff) discussion of generative semantics and
related tendencies, which is basically limited to statements like:
Lakoff [...] is working on 'cognitive grammar', which integrates language with

nonlinguistic systems. I don't see any theory in prospect there. (Chomsky 1979:
150)
(See also 1.3.3 on Chomskyan reactions to functionalism.)

It is not astonishing, then, that many researchers have started reacting in
similar ways to the generative paradigm, and this has even had the very un
fortunate effect that after a short period of cooperation between linguists and
language psychologists, the gulf between them has considerably grown again.
Unnecessarily, however, since reactions to the generative paradigm in linguistics
and language psychology are often based on the same principle, viz. a tendency
toward a more functionalist view.
Anyway, given the historical importance of Chomskyan theory, and given
that any new attempt to contribute to cognitive science from a theoretical
linguistic perspective runs the risk of being affiliated with (some of) its views,
and, hence, of being misunderstood and/or received with skepticism in other
disciplines, it is important to pay due attention to its tenets and explicate the
points of conflict and contrast in modeling linguistic behavior in the framework
of our cognitive-pragmatic theory. What will interest me are not the details of
the syntactic descriptions in TG and UG, but rather the general conception of
these models and the way they are situated in a more encompassing view of the
cognitive infrastructure of language.1
Chomsky considers the task of linguistic theory to reside in three questions,
viz.
(1) What is the precise nature of linguistic ability?
(2) How is it put to use?
(3) How does it arise in the individual?
(cf. Chomsky and Miller 1963: 271, Chomsky 1986a, b). 2 This requires
the precise specification of two kinds of abstract device, the first serving as a
perceptual model and the second, as a model for the acquisition of language.
(4a) utteranceAstructural description
(4b) primary linguistic dataBgenerative grammar
(Chomsky 1964a: 922; sentence numbers adapted)
LINGUISTIC KNOWLEDGE AND LANGUAGE USE 93
(see also Chomsky 1963: 329). Chomsky frames the development of these two
devices in terms of a meta-theoretical system of levels of adequacy. This system
provides an important key to the overall organization of and the internal cor
relations existing within Chomskyan theory, so we should pay some closer atten
tion to it. This is necessary also because the relationship of Chomsky's notions
of adequacy with those stated in 1.1.2 should be made explicit.
The system consists of three levels which are defined as follows:
The lowest level of success is achieved if the grammar presents the observed
primary data correctly. A second and higher level of success is achieved when the
grammar gives a correct account of the linguistic intuition of the native speaker,
and specifies the observed data (in particular) in terms of significant generaliza
tions that express underlying regularities in the language. A third and still higher
level of success is achieved when the associated linguistic theory provides a general
basis for selecting a grammar that achieves the second level of success over other
grammars consistent with the relevant observed data that do not achieve this level
of success. In this case, we can say that the linguistic theory in question suggests
an explanation for the linguistic intuition of the native speaker. It can be inter
preted as asserting that data of the observed kind will enable a speaker whose
intrinsic capacities are as represented in this general theory to construct for
himself a grammar that characterizes exactly this linguistic intuition.
For later reference, let us refer to these roughly delimited levels of success
as the levels of observational adequacy, descriptive adequacy, and explanatory ade
quacy, respectively. (Chomsky 1964a: 923-24)
(see also Chomsky 1964b, 1965). Descriptive adequacy would involve developing
a grammar for the individual language (i.e. developing language-specific TGs),
modeling the competence of adult S/Hs of the language to do what is involved
in scheme (4a) above. Explanatory adequacy involves a specification of device
B in (4b) above, i.e. a universal Language Acquisition Device (LAD) which
allows S/Hs of individual languages to acquire their competence grammar.
Yet Chomsky is not always consistent in using this terminology. Sometimes
he states that a language-specific TG can already be said to be an explanatory
theory of the competence of the language user. Yet, he wants to maintain the
distinction between descriptive and explanatory adequacy, because
Universal Grammar [...] constitutes an explanatory theory of a much deeper sort

than particular grammar, although the particular grammar of a language can also
be regarded as an explanatory theory. (Chomsky 1970: 428; reprint 1972: 27)
On the other hand, there also appears to be something like 'principled' de

scriptive adequacy, which is closely related to explanatory adequacy, and which
is introduced in relation to a distinction between external and internal criteria
for evaluating a specific grammar.
On one level (that of descriptive adequacy), the grammar is justified to the extent
that it correctly describes its object, namely the linguistic intuition - the tacit
competence - of the native speaker. In this sense, the grammar is justified on
external grounds, on grounds of correspondence to linguistic fact. On a much
deeper and hence much more rarely attainable level (that of explanatory ade
quacy), a grammar is justified to the extent that it is a principled descriptively
adequate system, in that the linguistic theory with which it is associated selects this
grammar over others, given primary linguistic data with which all are compatible.
In this sense, the grammar is justified on internal grounds, on grounds of its
relation to a linguistic theory that constitutes an explanatory hypothesis about the
form of language as such. (Chomsky 1965: 27)
In this respect, there is also some confusion about explanatory adequacy and the
status of the language universals. Thus, while Chomsky (1965: 34) states that an
explanatory theory is a theory which provides an evaluation criterion for
descriptive theories and that linguistic universals are an element of descriptive
theories, only two pages later he states that "the attempt to achieve explanatory
adequacy [is] the attempt to discover linguistic universals" (ibidem: 36). Related
to this is the fact that Chomsky also uses the notion of descriptive adequacy for
two things, viz. with respect to a grammar and with respect to a theory:
A grammar can be regarded as a theory of a language; it is descriptively adequate

to the extent that it correctly describes the intrinsic competence of the idealized
native speaker. [...] A linguistic theory must contain a definition of 'grammar', that
is, a specification of the class of potential grammars. We may, correspondingly, say
that a linguistic theory is descriptively adequate if it makes a descriptively adequate
grammar available for each natural language. (Chomsky 1965: 24)
According to this statement the first of the above characterizations of what is

involved in explanatory adequacy would be appropriate. Yet, since the "defini
tion of 'grammar'" is considered to be a part of LAD, and since LAD is what an
explanatorily adequate theory aims to specify, the second characterization of
explanatory adequacy is somehow adequate as well.
This rather confusing use of the adequacy notions is no doubt due to the
fact that this system involves two different matters: on the one hand a charac
terization of a hierarchy of levels of adequacy for descriptions of a certain
object, each of which implies a certain type of adequacy requirements in the
sense discussed in 1.1.2 (i.e. a meta-theoretical framework for the judgment of
hypotheses as to the degree of adequacy they reach), on the other hand a cha
racterization of types of descriptions which can be distinguished, in correlation
with types of goals which can be pursued in language research according to
Chomsky (i.e. a meta-theoretical framework stating goals and listing correlated

types of descriptions). The names 'observational', 'descriptive', and 'explanatory
adequacy' are therefore used for denoting types of descriptions and degrees of
adequacy alike.
As a system of degrees of adequacy, this framework specifically concerns
the evaluation of a description of adult linguistic behavior, i.e. device A in (4a).
The three levels are inclusive, in that adequacy at one level always implies
adequacy at the lower level(s). According to Chomsky, a description of adult
linguistic behavior is an appropriate cognitive theory only if it also satisfies the
conditions which are specified at the level of explanatory adequacy. It should be
noted that the conditions which Chomsky has specified at each level often relate
to matters to which specific attention has been paid in the development of TG.
Hence, these conditions are not formulated theory-independently (unlike those
specified in 1.1.2), and apart from the question whether TG actually does satisfy
them, one does not have to agree with the claim that they are sufficient to
warrant that a certain model is explanatory. See 2.2 and 2.3.
It is possible to trace the adequacy criteria specified in 1.1.2 in this system.
Chomsky's criterion to achieve descriptive adequacy, that a description must
make significant generalizations over the observed data, obviously refers to the
requirement that a process model be developed on a hypothetico-deductive
basis. Observational adequacy thus covers descriptions which are not hypotheti
co-deductive (cf. Chomsky 1964a: 914-918), and Chomsky's situation of the de
scriptions of his taxonomic predecessors at this level is in agreement with this.
The criteria of external and internal descriptive adequacy specified in 1.1.2 can
clearly be correlated with the levels of descriptive and explanatory adequacy in
Chomsky's system. As we saw above, Chomsky (1965: 27) uses the notions of
'internal' and 'external' criteria of evaluation himself in this respect. Thus, in
Chomsky's system descriptive adequacy of a description is a matter of satisfying
the condition that it be in accordance with the observable facts, i.e. that it be
'observationally adequate' in the sense indicated in 1.1.2. Chomsky specifically
refers to the linguistic intuitions of native S/Hs as the main kind of data a
descriptively adequate description ought to account for, but that is an idiosyn
cratic element of Chomsky's theory, to which I will return in 2.2.2.3. Explanatory
adequacy furthermore requires that the description also satisfies the criterion of
conceptual plausibility and coherence, as it was called in 1.1.2. More specifically,
according to Chomsky it ought to be preferable over other descriptively
adequate descriptions on the basis of some evaluation criterion. Since grammars
are always based only on a limited part of all potentially relevant data in a
language (i.e. the problem of underdetermination of grammars - cf. Chomsky
1964a: 937-938, 1975a: passim; see also 1.1.2), there will often be more than one
grammar compatible with the observed data. The evaluation criterion has to
provide a means to pick from the alternatives the one which really is descriptive
ly adequate. The evaluation criterion that is (or was) used in TG is the simplicity
criterion (Chomsky 1965: 37ff). Again the status of this simplicity criterion in TG
is idiosyncratic, and I will return to this in 2.2.3.1.
Next to these elements, Chomsky also assigns a very important role in his
model to language universals. Yet as we have seen he is not always clear about
whether he would consider them a matter of descriptive or rather of explanatory
adequacy of a description. In a certain sense this lack of clarity is a fair
expression of the ambivalent status of universals in developing a hypothesis on
the cognitive systems determining adult linguistic behavior. Universals are
basically theoretical concepts which are part of a hypothesis concerning
individual languages and natural language in general. On the one hand, they
relate to the question of external adequacy in that their presence in individual
grammars of languages has to be in agreement with the observed data from all
possible individual languages. On the other hand, they also play a role in the
internal evaluation of grammars of individual languages. If descriptions of a
great many languages recurrently show the presence of a certain phenomenon,
one can take this phenomenon to be a good candidate for the status of
universality. If the grammar of a specific language does not include this potential
universal but could include it if it would be rewritten in a certain way (while
remaining externally adequate, of course), one can say that the original grammar
is to be rejected on internal grounds and has to be replaced by its rewritten
version.3
Chomsky frequently switches from the usage of the adequacy notions just
mentioned to their usage to denote different types of descriptions with different
purposes, however. In the latter, the level of observational adequacy covers
descriptions which do not concern the cognitive dimension of language. It covers
all descriptions which in the former usage are observationally adequate and
descriptively adequate (i.e. it also covers all systematic taxonomies which for
internal reasons cannot be considered cognitively plausible). The descriptive
system which is explanatorily adequate in the former usage (i.e. is adequa'te on
both external and internal grounds) figures at the level of descriptive adequacy
in the latter usage of the term. That is why, in Chomsky's terminology, a
description can be 'principled descriptively adequate' next to simply 'descriptively
adequate'. Chomsky needs this formulation if he uses the term 'descriptive
adequacy' to name a certain type of description (a possible TG) while at the
same time indicating that this description not only meets the external criteria for
descriptive adequacy but also the internal criteria for explanatory adequacy (the
terms now used in the first usage indicated above). Alternatively, he sometimes
states that when a description is not both internally (explanatorily) and externally
(descriptively) adequate at the same time, it is not descriptively adequate (i.e.
it is not a type of description at that level). Compare the quotes from Chomsky
(1964a: 923-924) and (1965: 27) above. The different formulations occur
depending on whether Chomsky is trying to stick to one of the two usages of the
adequacy notions.
In its usage to name a description type, explanatory adequacy covers a
description of the innate device for acquiring the adult competence grammar.
This includes (at least in the original view - cf. 2.3.1) the internal evaluation
measure which in the other usage of the adequacy notions serves as a meta-
theoretical concept for judging the explanatory adequacy of the competence
grammar, and the set of universals (UG) extracted from individual grammars of
languages. Thus, this level covers exactly those elements which Chomsky invokes
as criteria to warrant that a certain TG is not only descriptively (externally)
adequate but also explanatorily (internally) adequate. In the present usage of the
notions, their status as meta-theoretical elements for judging the adequacy of the
description located at the level of descriptive adequacy (as a type of description)
is transformed into a status as elements of an actual hypothesis of a cognitive
system playing a role in language acquisition. Thus, from elements of 'method
ological explanation' of a description of adult language, they are switched into
elements of an 'object-inherent explanation' of adult language (cf. 1.1.1, note 2
on these types of explanation). (See 2.3 on these matters.)
In itself, this double usage of the adequacy criteria, though somewhat
confusing, might seem harmless. Yet it does mean that one should be careful in
interpreting Chomsky's statements on specific descriptions in terms of these
notions. Thus, as is illustrated in the above quotations, he constantly speaks of
'descriptions reaching some level of adequacy' in an absolute, unmodified way.
That is all right if the system is used to characterize conditions under which one
can say that a description has reached a certain degree of adequacy, i.e. if it is
used as a prescriptive system indicating what descriptions ought to be able to do
so as to be adequate at the different levels. A meta-theoretical framework for
evaluating models normally depicts the ideal situation, thus prescribing what the
scientist should accomplish in developing models. But Chomsky continues to
speak in these absolute terms when using the system to classify descriptions
according to the goals they pursue, i.e. if he uses the system descriptively to
characterize the status of actual models somehow figuring in his theory
development. That is where the terminology becomes dangerous. Given the basic
impossibility of verification (cf. 1.1.2), one can never know, and thus should
never claim, that a model has reached a certain 'level of adequacy' or goal (cf.
also Seuren 1975, 1977). Adequacy is always relative to evidence known and
applicable at a certain moment in time. It is always adequacy in absence of
counterarguments, and it never involves certainty or positive confirmation.
The above exposition at once shows how intimately Chomsky's theories of
adult competence and LAD are linked. In fact, they fulfill a rather ambiguous
role relative to each other. On the one hand, the competence theory is
presented as a means to achieve the main goal of finding out about LAD:
the study of acquisition and universal grammar does not arise as an ancillary
problem in the course of constructing grammars, but is rather the focus of our
interest - grammars interest us insofar as they contribute to these further investi
gations (Chomsky and Katz 1974: 350)
(see also Lightfoot 1982). On the other hand, since the acquisition theory
involves all elements warranting the internal adequacy of the competence theory,
the former also serves to support the plausibility of the latter. There can be no
doubt about the need for a relationship between adult and child language
theories, yet one can question the type of mutual dependence proposed by
Chomsky.
It is certainly true that a description of adult linguistic capacities constitutes
an element in developing an acquisition theory. However, it is not appropriate
to claim that an acquisition theory should be the main objective in language
research. The main objective is to explain language, and this has different
aspects of which an account of its acquisition and genetic basis is one, and an
account of adult linguistic knowledge and processing is another and equally
important one. A description of the cognitive phenomena involved in language
acquisition and its genetic basis may be explanatory at a 'deeper' level, since it
is also a further explanation of the cognitive phenomena involved in adult
language behavior. But it is at least dangerous to say that it is also explanatory
in a more 'principled' way. Such a claim can lead to reductionism as it can be
used as an alibi to neglect the problem of developing a full-scale account of the
adult cognitive systems. The tendency in Chomskyan linguistics to concentrate
on UG and to neglect grammars of individual languages might be an illustration
of this. This attitude endangers the very enterprise of developing an acquisition
model as well, however, since a full account of the acquisition process (including
its genetic basis) does require a full account of its 'final' result, i.e. the adult lan
guage system. On the other hand, it also has to be clear that a language
acquisition theory or a theory about its genetic basis cannot be derived solely
from a theory of adult linguistic knowledge, as Chomsky tries to do. An
acquisition theory has to cover data which go far beyond the data relevant to an
adult language theory. It has its own criteria of scientific adequacy, independent
ly of a theory of adult linguistic knowledge. (See also 2.3.1.)
Conversely, it certainly is a strong argument for the explanatory value of
a theory of adult linguistic knowledge if it is backed up by an acquisition theory
supporting its internal descriptive adequacy, in the sense that it would fit into the
acquisition theory as a model of the 'final stage' of the developmental process.
But this does not mean that the internal adequacy of an adult language theory
entirely depends upon there being such an acquisition theory, contrary to what
Chomsky seems to suggest. Rather, since developing an acquisition theory in
part also depends upon an adult language theory, it is important that this adult
language theory be strongly supported independently of the acquisition theory,
e.g. by being as complete as possible (covering as many aspects of linguistic
behavior as possible). This is necessary in order to make sure that what the adult
language theory contributes to the acquisition theory is plausible.
Briefly, it is best to assume that developing an adult and a child language
theory, although related, are two different things (cf. also Stich 1972). They need
much more separate consideration than is granted in the Chomskyan framework.
Therefore, although this is somewhat against the spirit of Chomskyan theory, in
the following I will treat both aspects in Chomsky's system separately. Since the
present study is mainly concerned with adult linguistic behavior, I will first
consider this aspect.
2.2. Knowing and using language
2.2.1. TG and competence and performance
As mentioned in the foregoing section, Chomsky (1964a) characterizes the

problem of developing a theory of adult linguistic behavior basically as being a
matter of providing a specification of device A in scheme (4a):
(4) a. utteranceAstructural description
In itself, this does not seem to be an ideal initial characterization of the

basic problem, though. As argued in chapter 1, what we have to start from is
that there are two observable dimensions directly related to linguistic behavior,
viz. the actual linguistic utterances as structured phenomena occurring in specific
contexts on the one hand, and the meanings we can attach to those linguistic
utterances (their functionality) on the other hand. This automatically leads to a
general concept of linguistic behavior as a bi-directional phenomenon covering

both the expression of meanings through acoustic patterns (i.e. language
production) and the abstraction of meanings from acoustic patterns (i.e.
language understanding). This overall picture seems unavoidable, trivial as it may
be. Whatever the details of the cognitive organization of language may turn out
to be (they will no doubt be rather sophisticated), the systems and mechanisms
constituting it necessarily have to be such that they allow information to be
encoded in syntactically organized acoustic patterns, and conversely that they
allow the percepts of those acoustic patterns to be decoded into information
(and all of this according to the criteria specified in 1.2). A model of the
cognitive organization of the linguistic capacity somehow has to give a mean
ingful substantiation to this global picture if it is to be adequate.
There is no reason to believe that Chomsky would deny this global concep
tion, despite the fact that (4a) does not show it. (4a) only indicates the need for
a perception theory, while a complete characterization of linguistic behavior
obviously also has to include a production theory. Yet in this respect there is a
discrepancy between (4a) and Chomsky's discourse in other places, where he
does speak about production and perception. Secondly, the characterization of
perception in (4a) only shows part of the problem: while Chomsky indicates that
the output of the system should be a structural description, a more complete and
more neutral specification should indicate that the output is a meaning pattern
of the input-utterance. Thus, a completely neutral and pre-theoretical characteri
zation of the initial problem would be (4a'):
(4) a'. utteranceAmeaning
Chomsky's formulation in (4a) is clearly influenced by the aspect of language

which interests him the most, viz. syntax. But elsewhere Chomsky does state that
a grammar is a device for 'pairing meanings and phonetic patterns' (cf. below).
Let's then consider how Chomsky conceives of the specification of device
A. As is well known, one of the main observations which Chomsky has always
said to be crucial in determining his views on language is that language users ap
pear to be able to produce and understand an unlimited number of utterances,
most of which they have never heard or produced before (i.e. the creativity of
language users - cf. e.g. Chomsky 1964a). To account for this, he claims, one has
to assume that S/Hs have the competence to perform the activities involved in
actually doing this. The existence of this competence is also reflected in the fact
that S/Hs have intuitions concerning what is and what is not possible in their
language: whether utterances are 'grammatical' (syntactically well-formed),
ambiguous, synonymous, etc. Chomsky claims that these intuitions must directly
rely upon an internal mental structure also lying at the heart of creative
language use.
According to Chomsky (e.g. 1963) the only type of formal mechanism
which can account for this ability to judge and use an infinite number of
structures, including ambiguities and synonyms, is a transformational generative
grammar of the type first proposed in Chomsky (1957), viz. a phrase structure
grammar consisting of rewrite rules which generate 'deep structures', augmented
with a system of transformation rules (gradually developed into one 'move a
rule with a series of constraints on its application) forming the 'surface
structures' of the grammatical utterances of the language.4 This mechanism is
purely syntactic. From the very beginning Chomsky has maintained that
Despite the undeniable interest and importance of semantic [...] studies of langua
ge, they appear to have no direct relevance to the problem of determining or
characterizing the set of grammatical utterances. I think that we are forced to
conclude that grammar is autonomous and independent of meaning. (Chomsky
1957: 17)
It is, of course, impossible to prove that semantic notions are of no use in
grammar [...]. Investigation of such proposals, however, invariably seems to lead
to the conclusion that only a purely formal basis can provide a firm and productive
foundation for the construction of grammatical theory, (ibidem: 100)
Of course, being able to use language also involves being able to assign meaning
to the well-formed structures in a language, hence Chomsky later did introduce
a semantics, yet without changing the formal syntactic core. It has been assigned
to an interpretive component next to the interpretive phonological component.
Thus, Chomsky claims that each native S/H must have "developed and internally
represented a generative grammar" (Chomsky 1965: 25).
Yet, he is quick to point out that this hypothesis concerning the form of
linguistic competence of the S/H should not be mistaken for a model of how the
native S/H uses language. The theory of competence is only an answer to the
question formulated in (1) in 2.1, viz. 'What is the precise nature of linguistic
ability?', not to the question in (2), viz. 'How is it put to use?'. It is not a fll
account of device A in scheme (4a) above: that would involve developing a
performance theory. The competence theory is only one component of it, albeit
one without which the device would not be able to operate.
The use of language undoubtedly involves many factors beyond the grammar [...but
it] is natural to suppose that models of the speaker and hearer will incorporate the
'competence grammar' as a basic element; that is, such models will incorporate the
system of rules and principles that determines the language that the language user
employs and its structure. In fact, virtually all serious work on language use makes
some such assumption, either implicitly or explicitly. (Chomsky 1975b: 9)
The linguist interested in competence does not have to bother with performance
- research on competence can be independent of research on performance, and
should even be prior to it. In fact, Chomsky maintains that
There seems to be little reason to question the traditional view that investigation
of performance will proceed only so far as understanding of underlying competence
permits. (Chomsky 1965: 10)
This competence-performance distinction is a core notion in Chomskyan

theory, but it has also become one of the most widely discussed and most con
troversial notions in language research in general (witness the huge number of
criticisms, including Fodor and Garrett 1966, Hymes 1971, Schlesinger 1971,
Stemmer 1971, Pylyshyn 1972,1973, Derwing 1973, Lakoff 1973,1974, Clark and
Haviland 1974, Kintsch 1974, Parret 1974, Whitaker 1974, Cooper 1975, Des
Tombe 1976, Neumaier 1979, Valian 1979, Longuet-Higgins 1981). Chomsky
(e.g. 1969a, 1975a, c, 1980a) has rejected most of the reactions as being
misunderstandings of the dichotomy. Yet he cannot be held blameless if it comes
to pointing out the causes of the misunderstandings, for his specification of the
dichotomy is so vague that it can be interpreted in many different ways, and on
different occasions he tends to suggest different readings himself, depending on
the argumentation he is pursuing.
Chomsky has rephrased the dichotomy in terms of the distinction generally
accepted in psychology between what an organism can and what it does
(Chomsky 1963: 326), or more specifically in terms of knowledge and use of
language. Yet this does not clarify much, for this is only restating the problem
in different terms: now one has to specify what is meant by 'knowledge' and how
it relates to behavior. As to the former, Chomsky (e.g. 1975c) often complains
that he is not to blame for the lack of clarity and the ambiguity of the notion
'knowledge' in the literature in general. Although this problem is real, the way
he uses the terminology himself is not likely to decrease confusion. Character
izing linguistic knowledge as 'tacit knowledge' on the one hand, but on the other
hand also as a 'theory' which the language user has about language, albeit an
'unconscious theory', (cf. e.g. Chomsky 1975c: 315) does not clarify much (cf.
2.2.3.3). Further specifying the notions of competence and linguistic knowledge
as 'being in the mental state of having internalized a generative grammar' (e.g.
1975c) does not help much either, for this only leads us back to where we
started from. There is a strong tendency in Chomsky's writings to use the notions
competence, linguistic knowledge, and TG as mutually defining, answering
questions about one of them simply by referring to one of the others, and thus
leaving us with circular argumentation.
The specification of the relationship between the notions of competence
and performance does not bring one much further either. One thing is clear: the
competence-performance dichotomy involves an attempt to break with a simple
conception of how to model the cognitive phenomena involved in scheme (4a)
(or (4a'), for that matter). This can be taken in at least two 'gradations'.
(i) In its 'weak' reading, the dichotomy involves a differentiation in the relevance
of certain observations on linguistic behavior for developing hypotheses on the
contents of the syntactic knowledge of an S/H. This means referring phenomena
to other components (modules) involved in performance, or excluding
phenomena as irrelevant for understanding knowledge in general. This includes
the concept of the ideal S/H. Chomsky claims that there is a discrepancy
between the intuitions and the actual linguistic behavior of S/Hs, in that the
latter is crowded with utterances which according to the former are ungrammati-
cal. The deviance of actual behavior is due to all kinds of factors which are
independent of S/H's pure knowledge of language, but which interfere with this
knowledge when it is put to use in actual behavior. Thus,
Linguistic theory is concerned primarily with an ideal speaker-listener, in a

completely homogeneous speech-community, who knows its language perfectly and
is unaffected by such grammatically irrelevant conditions as memory limitations,
distractions, shifts of attention and interest, and errors (random or characteristic)
in applying his knowledge of the language in actual performance. [...] We thus
make a fundamental distinction between competence (the speaker-hearer's
knowledge of his language) and performance (the actual use of language in
concrete situations). Only under the idealization set forth in the preceding
paragraph is performance a direct reflection of competence. (Chomsky 1965: 3-4)
(ii) In the 'strong' reading of the dichotomy there is also at least a partial
departure from the bi-directionality of production and perception in theorizing
about knowledge.
A generative grammar is not a model for a speaker or a hearer. It attempts to

characterize in the most neutral possible terms the knowledge of the language that
provides the basis for actual use of language by a speaker-hearer. When we speak
of a grammar as generating a sentence with a certain structural description, we
mean simply that the grammar assigns this structural description to the sentence.
When we say that a sentence has a certain derivation with respect to a particular
generative grammar, we say nothing about how the speaker or hearer might
proceed, in some practical or efficient way, to construct such a derivation. These
questions belong to the theory of language use - the theory of performance.
(Chomsky 1965: 9)
However, so far Chomsky has not provided an answer to the question what
performance should involve and how it should be organized in any more detail,
let alone an indication beyond the vague suggestions made above of how the
grammatical competence (and other competences, such as the 'communicative
competence', which Chomsky (e.g. 1975a, 1980a) postulates without giving details
about them) could figure in it.5 In any case performance concerns more than
just the matter of 'putting knowledge to use', since it also covers other cognitive
systems, and thus presumably in part other knowledge involved in actual
speaking and hearing. But as mentioned, Chomsky does not consider a more de
tailed specification of this to be the linguist's problem. Yet this way the neutral
observer is also left with all obscurities and ambiguities involved in the notion
of competence. If one notion in a dichotomy lacks specificity, the other notion
suffers from it as well.
In fact, since the mid-seventies Chomsky has motivated his silence
concerning the concept of performance with a skeptical note.
Roughly, where we deal with cognitive structures [such as the generative grammar
of the language user - JN], either in a mature state of knowledge and belief or in
the initial state, we face problems, but not mysteries. When we ask how humans
make use of these cognitive structures [i.e. how they enter into the performance
- JN], how and why they make choices and behave as they do, although there is
much that we can say as human beings with intuition and insight, there is little, I
believe, that we can say as scientists. What I have called elsewhere 'the creative
aspect of language use' remains [...] a mystery to us. (Chomsky 1975a: 138)
(see also Chomsky 1986a, b). 6 In other words, while research on competence
is doing fine, all attempts to say something about performance (and this refers
to a gigantic amount of literature) have been in vain. Chomsky (e.g. 1980a,
1986a, b) suggests that this may be due to limitations in our ability to understand
things scientifically: maybe linguistic performance lies beyond the borders of
what humans can grasp with their 'scientific competence'.
This point of view is hard to come by, though. It is not improbable that the
human capacity to understand things is indeed limited.7 The field of human
cognition in particular is so complex and difficult to access that it might be
beyond our abilities of rational thinking (though we do not know this for sure,
and though we should not stop trying to find out about it). Yet if this is the case
indeed, this argument applies to the study of cognition in general. That is, if it
applies to our understanding of how S/Hs go about in speaking and understand
ing, it ipso facto also applies to our understanding of S/Hs' knowledge underlying
these processes. Both are parts of the same black box (1.1.2), and the conditions
for their inquiry are identical.
So, maybe the causes for the mystical nature of performance for Chomsky
should be sought elsewhere. As Lakoff says,
The only consistent way in which I can understand his use of the term 'performan
ce' is that he takes it to be a wastebasket for all the phenomena that cannot be
accomodated [sic] by whatever theory he happens to be maintaining at a given
time. (Lakoff 1974: 153)
In one sense the question of what one calls 'competence' and what one calls
'performance' is a piddling issue of terminology - the facts are the same, call them
what you will. In another sense, though, it is a matter of critical importance, if like
Chomsky, one uses the terms 'performance' and 'competence' to characterize what
kinds of facts you feel a linguistic theory should be responsible for. In the latter
case, such a decision can have an effect on whether a given fact is a crucial coun
terexample, or 'merely a matter of performance' which can be brushed under the
rug. Chomsky's shifting definitions of performance provide him with a rug big
enough to cover the Himalayas, (ibidem: 155)
Let's consider the dichotomy in somewhat more detail. I will start with the
'weak' reading.
2.2.2 Characteristics of behavior and the nature of knowledge
2.2.2.1. The impact of data on the development of hypotheses

One of the most characteristic features of TG is that it is based on a strict
limitation of the set of data that is taken into account, in different respects: the
functionality of syntactic structures, the variability and changeability of behavior,
deviant structures, etc. are all excluded (cf. 2.2.2.2). Chomsky motivates his
position on the grounds that linguistic behavior is a very complex phenomenon
in which the effect of the individual 'module' of syntactic competence is troubled
by the interference of many other elements. Thus, to inquire into this compe
tence one has to eliminate everything that is not relevant to it. This is related to
his view that "the fundamental concepts are grammar and knowing a grammar,
and language and knowing a language are derivative" (Chomsky 1980a: 126; cf.
also Chomsky 1981b), or, in a more recent terminology, that the linguist should
not be concerned with the 'E(xternal)-language' but with the 'I(nternal)-
language' (Chomsky 1986a: 7ff; see also 1986b). As such, the limitation of the
facts taken to be relevant is not just a methodological matter. I receives a
theoretical meaning in that it is presented as being cognitively rooted. It is sug
gested that the limited set of facts is also precisely the set which relates to the
syntactic knowledge of S/H. However, this position is questionable on principled
grounds.
Of course, most probably Chomsky is right that actual linguistic behavior
is the result of the interaction of a wide variety of cognitive factors, and that it
gives a troubled picture of the knowledge involved due to factors such as absent-
mindedness or fatigue which are irrelevant to it (cf. below). It is true that in
developing a process model the researcher's interests are with the internal
systems, the effects of which are not a priori obvious in the 'superficial' behavior.
Yet this does not imply that the 'E-language', no matter how derivative and
complicated, is of no importance anymore. In statements like those above,
Chomsky seems to assume that competence is a directly accessible system (cf.
also his e.g. 1980a characterization of competence systems as being 'mental
organs' comparable to biological organs - 1.3.3), and that it is immediately
obvious which facts are relevant to it and which are not. But such an assumption
is inappropriate since there is no way to observe cognitive systems directly (cf.
1.1). The I-language can only be accessed through observing the E-language.
Hypotheses on the I-language hinge upon what can be observed in the E-
language, and a theory of the former is automatically also a theory of the latter.
Thus, in investigating a complex phenomenon such as linguistic behavior
it is important to systematically try and isolate parts and aspects of the entire
range of data which can be studied separately, before tackling larger sets of data.
This has to involve all kinds of intermediate idealizations and simplifications of
the object under consideration. Such a research strategy is a matter of pure
necessity: without it one would never be able to grasp the phenomenon. One
cannot draw any direct theoretical consequences from this methodological
organization of research, however. Research on one specific subfield of data
cannot lead to final conclusions about one specific sub-structure in cognition. An
at face value separable domain of facts neither has to be due to one cognitive
sub-system, nor has to contain all facts relevant to that cognitive sub-system.
Getting a picture of the organization of cognition (and of each sub-system in it)
is like putting together a jigsaw puzzle from the bits and pieces derived from
many different individual studies on many different separated aspects and parts
of the entire field of behavioral data.
In other words, though it is right to call the cognitive world modlar in
some way (cf. 2.2.3.1), this does not mean that one can investigate it module for
module. (As argued in 1.3.3, this is precisely where the cognitive sciences and
biology differ.) As Chomsky justly observes, the modules of cognition do not
show themselves as discrete elements in behavior. But unlike Chomsky's
assumptions this unavoidably leads to the conclusion that there is no way to
determine in advance how aspects of behavior are related to specific modules,
or which aspects of behavior say what about which modules. As argued in 1.1,
the correlations will only be evident if a complete map of the cognitive world has
been drawn, and if all registered aspects of the data have found their place
therein in the sense that it follows from the map which modules are responsible
for which facts. Chomsky's concept of linguistic research runs radically counter
to this principle, and this obviously also questions his notion of competence.
Chomsky's views in this respect also lead to an unjust attitude toward data
which might be taken to challenge his model (this is not the first time this
criticism is formulated). The initial idealization and limitation of the data
considered relevant to develop the model is used afterwards to reject data which
do not fall within the initially determined domain as being irrelevant. There
appears to be some awareness of this in the Chomskyan literature, and this is
reflected in a number of explicit attempts to defend the TG-attitude toward
data-observation and counter-evidence by questioning Popper's principle of
falsification (cf. e.g. Lightfoot 1982, Koster 1983). Yet, these arguments are
based on a reading of the falsification principle which is certainly not in the spirit
of Popper's initial formulation of it, or of the way it is actually applied in science.
Obviously, "refutation is not a trivial matter, to be effected by simply citing an
unanalyzed fact about some language" (Lightfoot 1982: 97). Theories are
constantly being bombarded with counter-examples by their opponents, but
insofar as this concerns individual cases on a limited scale, this does not cause
anybody to abandon them since they can normally be supposed to be able to
solve the problems (cf. also Lakoff 1974).
However, if an hypothesis is constantly confronted with types of data for
which it structurally has no, or only ad hoc solutions, the matter is different.8
Of course, one can then raise the question whether the hypothesis should really
provide an account for these facts. Obviously, if it only concerns the form and
content of a particular cognitive sub-system, it can only be held responsible for
the parts of the data which are due to this sub-system. Here the above argument
against the TG-strategies may seem to backfire, for proponents of the hypothesis
can hide behind the fact that one has no simple way to show the relevance of
the crucial data for the model. Yet the point is that rejecting the force of
falsification of data for a model on these grounds cannot be done just by
pointing at self-imposed a priori restrictions, contrary to what is often the case
in TG. This can only be done by showing that the restrictions in the model are
theoretically plausible in terms of a larger framework which covers a wider range
of facts 'surrounding' the facts covered by the model, and by showing which sub
system(s) in this larger framework should then be responsible for the rejected
data. This is not what is done in TG though, for the data rejected are referred
to the theory of performance which, however, is completely left unspecified.

Finally, the above also implies that the division of labor which Chomsky
imposes on the research on competence and performance (cf. 2.2.1) is inappro
priate. It is all right if a researcher wants to restrict attention to a limited field,
but one cannot turn such personal choices into directions for an entire scientific
discipline. Whatever the contents of the theories of competence and per
formance in Chomsky's views are, since the paradigms investigating them both
would be full-scale scientific subdisciplines handling phenomena which are
intimately related, they should be mutually relevant and each should be able to
support or undermine the other. (Cf. also Schlesinger 1967, 1971, Derwing 1973.)
Thus, it is basically unsound to postulate a one-directional relationship between
them. Even if one of them somehow could be said to be involved with a more
'fundamental' aspect of the object of investigation, this still does not mean that
the other subdiscipline could not provide insights which are crucial for the
research on this 'more fundamental' aspect (this point is similar to the one on
the relationship between adult and child language theories made in 2.1). One
sub-discipline may become dominant over the other (may become a 'pilot
science') because it seems to have made a larger progress in the analysis of the
object under investigation. But this practical circumstance has nothing to do with,
and thus cannot be baked into, the principles of the theories developed in the
subdisciplines, and it remains the case that insights from the 'following discipline'
are relevant for the theories in the pilot discipline. In its early days TG certainly
had the right to call itself a 'pilot paradigm', but with the insights accumulated
since then in other subdisciplines and in alternative theories, there is not much
left to warrant this position.
Let's try to substantiate this, first by taking a closer look at some of the
specific limitations built into the TG-concept of linguistic competence.
2.2.2.2. Some idealizations and limitations in TG reconsidered

From the perspective of this study, the most severe limitation is no doubt the
exclusion (already indicated in 1.3) of the functional dimension in modeling
syntax. This is one of the cases (others will be pointed out below) in which
Chomsky seems to deliberately put on 'eye-flaps'. What I am referring to is an
attitude to refuse to inquire into potential correlations or the mutual relevance
of two phenomena. If a researcher is faced with the possibility of there being
such a relationship, one would expect him to try to find out about this. It may
turn out that the relationship is non-existent, but this conclusion is only possible
after having tried to find one. Yet in such circumstances Chomsky often
postulates on an a priori basis that there are no correlations and dismisses
attempts to nevertheless look for them as irrelevant. The attitude toward

functionality is one instance of this more general attitude. But, as argued at
length in 1.3, particularly in this case it is totally unjustified. Since there obviously
is a relationship between form and function in linguistic behavior, an adequate
model of the cognitive systems and mechanisms involved in it has to provide an
integrated picture for the two dimensions. This is the case for a model of
language in general, but also of any sub-aspect of it, such as syntax.
The exclusion of the functional dimension already appears from the total
absence of semantically relevant notions in the syntactic core of the grammar.
Chomsky repeatedly claims that there is no reason to assume that semantic
notions are relevant for determining the structural properties of language (cf.
e.g. the quotations from Chomsky 1957 in 2.2.1), but this is either trivial or
wrong. Obviously, in determining the form of constituents in language (in terms
of notions such as V, NP, PP) no reference to semantic concepts is required, and
these elements should receive a full-fledged status as formal notions in any
model. Still, that does not mean that 'grammar is autonomous and independent
of meaning'. There are also aspects of syntactic organization of utterances which
can only be understood in semantic terms (relational concepts such as
theme/rheme, topic/focus or predicate, semantic concepts such as agent,
beneficiary, etc.) related to the conceptual status of the different structural
entities relative to each other. Since accounting for language involves showing
how utterances can carry meanings - this is the case no matter whether one
accepts the view on functionality expressed in 1.2 or whether one takes
Chomsky's very vague view that language is a means to express thoughts - it
seems obvious that the correlation of these formal and semantic elements of
syntax should be one of the core elements of a grammar which aims to say
something about the cognitive organization of language. But in TG this is
referred to an interpretive semantic appendage.
Of course, Chomsky's rejection of matters functional does not end here.
In his view the semantic component is still detached from the problem of the
functionality of language use. The determination of the latter is relegated to a
totally separate mental module, viz. that of 'communicative' or 'pragmatic
competence' (e.g. Chomsky 1980a), which is also somehow involved in
performance. Yet many syntactic patterns have a very specific communicative
significance which is specifically bound to their structural configuration. One can
expect a cognitively relevant grammar to account for this in the core of its
syntax.
In this connection, let's more specifically consider the element of creativity,
which figures so importantly in Chomsky's conception of what a linguistic theory
should be able to explain. Or, rather, which originally figured so importantly in

it, for more recently the creative use of language is also referred to the mysteries
of performance (cf. e.g. the quotation from Chomsky 1975a: 138 in 2.2.1). Maybe
this may seem a logical move: the creative use of language should be none of the
business for a mechanisms which does not say how language is used. Yet if
competence is something that is crucial for the behavioral processes of language
production and interpretation to take place, and if creativity really is such an
important feature of linguistic behavior (and Chomsky certainly was right that
it is), one may expect this competence also to contain the elements which allow
this creativity. (The paradox in the 'strong' reading of the competence-
performance dichotomy is already apparent here - cf. 2.2.3.)
Now, Chomsky's original concept of creativity certainly was a very narrow
one. What he meant by it was mainly that a language user is able to produce
and understand an infinite number of sentences, many more than the number
of seconds in a lifetime, and of which he only encounters a fraction. Thus,
the grammar of a language must, for empirical adequacy, allow for infinite use of
finite means, and we assigned this recursive property to the syntactic component.
(Chomsky 1972: 155)
Hence, it is mainly the principle of recursivity, which is a core element in the

phrase structure component of TG, which is considered crucial to account for
creativity (cf. also Chomsky 1965: 8). 9
Yet this can hardly be said to be an account for a non-trivial concept of
creativity (cf. also McCawley 1974). Recursivity concerns the possibility to
expand structures in length by repeating a structural element in its own internal
pattern. Thus, this can only account for a purely quantitative concept of
creativity: on the basis of a limited number of elements one can build an endless
number of structures by allowing the elements to self-embed, and by putting no
restrictions on the number of such self-embeddings. Real creativity in linguistic
behavior does not directly appear to be only a matter of unlimited length and
quantitative complexity of structures, though. On the contrary, linguistic behavior
is certainly not characterized by those features. S/Hs do not and cannot produce
and understand endlessly long and complex constructions, as has been repeatedly
remarked in the literature.
This is not an argument against recursivity in natural language syntax, as
Chomsky has justly replied. It is a fact that utterances have self-embeddings of
different kinds, hence a recursive principle is unavoidable. The length-limitations
in actually occurring structures must be due to limitations in processing and
memory capacities of humans.10 This explains why there is some variability in
the length and complexity of constructions produced and understood, even by

one S/H: depending on the circumstances (time and means available) one can
be more or less able to avoid memory limitations and distractions.
Of course, all of this does not change the fact that creativity involves a
(quasi) 'infinite use of finite means': a checklist view of cognition, in which all
occurring structures are somehow stored in a memory, is obviously unfit to
account for it.11 The point is that much more is involved than recursivity.
From a structural perspective, it involves the potential to combine in different
ways all single elements involved in the construction of an utterance: the dif
ferent lexical elements available in the (in itself open-ended) lexicon combining
with each other in structural patterns at different hierarchical levels, including
intonational patterns. Even within the limits of a relatively narrow length
variation, the sum of all possible combinations of basic elements available is
gigantic. Maybe it is not truly endless: a length-limited combination of a finite
number of elements produces a finite number of forms. Still, given the huge
number of basic elements and combinatorial principles, and the open-endedness
of the set of basic elements, the limit to infinity seems academic.
Yet all of the above is still not sufficient to really account for the S/H's
creativity. S/Hs do not combine basic elements into structures just like that, they
do this with an eye to certain goals. Hence, what creativity really is about is the
optimal exploitation of the structural possibilities available to reach all kinds of
goals in as efficient a way as possible. It resides in the way in which one adapts
oneself through linguistic means to the requirements of the different factors
involved in the communicative situation in which one is placed (cf. 1.2). Thus,
accounting for the potentiality for creative language use unavoidably leads to the
matter of the correlation of form and function again: what is the functional
dimension of the different possible combinations of basic elements in (a)
language?
An account of creative language use also requires that one goes beyond the
notion of 'rule-governed creativity' (Chomsky 1964a) and takes into account
'rule-changing creativity', contrary to Chomsky's argumentation (cf. also Chomsky
1974). For creativity certainly also resides in the capacity to use the means
available in unexpected and innovative ways, thus constantly renewing language.
This is, for instance, the essence of much of literary creativity: sorting special
effects to reach some communicative goal through a 'forced use' of linguistic
means available. But also 'everyday' language use is crowded with 'rule-changing'
acts: think of the principle of metaphor which is pervasive throughout everyday
interaction and which allows one to constantly represent things in new ways. Of
course, Chomsky (1974: 29) is right that "'creativity' [...] presupposes a
framework of rule", and that "if all constraints are abandoned, there can be no
creative acts".12 Creativity has to follow a certain order - otherwise it would
not be comprehensible. But this can be a non-linguistic order which allows one
to 'break' a specifically linguistic order. Of course, the ultimate borderline
remains the requirement to reach one's goal. The 'constraints' on creativity can
only be formulated if the functional dimension is taken into account.
The concept of rule-changing creativity at once brings us to a core problem
any cognitive theory unavoidably has to face, viz. the distinction between
knowledge and use of language and the role of all kinds of properties of lin
guistic behavior in determining these notions and their mutual relationship.
Chomsky uses this distinction pervasively to explain away all kinds of elements.
Of course, the value of the theoretical distinction as such between knowledge as
a potential and the use of it as an actualization of this potential is beyond doubt.
The principle of recursivity is one clear indication for this. It potentially allows
for infinitely long structures, yet it is never actually used to form them. Another
element which forces one to accept the distinction is the existence of errors in
behavior.13
By 'error' I mean that an S/H at a certain moment in a certain context
forms a structure or develops an interpretation of one which deviates from the
structure or interpretation (s)he would form in the 'ideal' case, without there
being an intention to do so. 14 In more functional terms: an error occurs when
one does not manage to say what one wants to say in a way which is accurate
according to one's linguistic knowledge, or does not succeed in understanding
what one should understand from what has been said according to an optimal
application of one's knowledge. Errors can occur at all levels of linguistic
organization, from very superficial ones (phonetic/phonological, syntactic, lexical
- errors investigated so far are mainly of this type) to very deep ones (illogical
meaning patterns, inappropriate illocutionary decisions, wrong presuppositions
or inferences, etc.). The communicative impact of an error also seems to differ
with the 'depth' at which it occurs: as a general rule, it seems that the deeper
its roots, the harder it is to recognize it as an error, and, hence, the more
profound its impact on communication can be. Anyway, errors obviously are not
part of an S/H's linguistic knowledge. They must be due to external factors such
as distractions causing knowledge to be used in wrong ways. On the basis of the
same reasoning as used for the principle of recursivity, a hypothesis of an S/H's
linguistic knowledge has to represent this knowledge in a 'pure' form. These
observations have led Chomsky to the conclusion that phenomena such as errors
and actual length of structures produced and understood should entirely remain
out of consideration in modeling knowledge. Yet the situation might be more
complicated than he suggests.

One certainly cannot keep ignoring these phenomena ad infinitum in
theorizing about linguistic behavior, since limitations in processability and liability
to err are basic characteristics rather than peripheral phenomena in behavior.
A plausible suggestion is that these matters have to do with the existence of a
control system on the one hand and a certain memory space on the other.
'Control' is obviously not meant in the TG sense, but in the AI sense. Although
this is a very common concept in this branch of cognitive science (cf. e.g.
Winston 1984), albeit one about which there is a wide variety of opinions, it is
almost completely absent in other sub-disciplines. Unjustly though, for many
phenomena in linguistic behavior are inexplicable without it, no matter how
unclear its precise implications are in our present state of the art. The notion of
memory is of course much more widespread (cf. the cognitive-psychological
notions of short-term memory (STM) and long-term memory (LTM)).
If we would characterize the linguistic knowledge of an S/H as the 'wheel-
work' on which the 'machinery' of the mind runs, the control mechanism (or
rather, the control mechanisms at different hierarchical levels and of different
types - cf. 3.2.2) could be characterized as the sub-part of the machinery which
puts the wheel-work of knowledge to work to actually produce and interpret
utterances. Metaphorically, we could say that the control system transforms the
potential energy of linguistic knowledge into kinetic energy. Memory (or rather,
again, several memories of different types) can in this context be characterized
as the work and storage space within which the control system operates and
keeps interim or final products of the operation of the wheel-work (of course,
linguistic knowledge is kept itself in some kind of - in any case long-term -
memory).
One can further assume that the control system has a certain 'capacity' to
operate the knowledge, and this capacity is in part stable. That is, the system has
an average capacity, which can no doubt differ from individual to individual, and
which can therefore figure as one factor in explaining differences in intelligence
between individuals. But the capacity in one individual is in part also determined
by hic et nunc factors such as degree of fatigue, influence of alcohol and other
products, climatological and environmental influences, etc. which influence the
average capacity and make the actual capacity differ from day to day. Thus,
while knowledge is a relatively stable phenomenon (but see the next section) the
capacity of the control phenomenon is relatively variable. The memory space
available probably also differs from individual to individual, but each individual
can be supposed to have a norm capacity which cannot be surpassed, and which
is more or less stable too. Use of knowledge is restricted by the memory space
available.
In this framework, errors can be a consequence of factors disturbing the
optimal functioning of the control system so that it makes mistakes in using
knowledge, or of the fact that part of the crucial memory capacity is filled with
interfering information not required for the task at hand (i.e. distraction), either
quantitatively (by leaving too little space for the task) or qualitatively (by mixing
up with information involved in the task). Limitations in applying the recursive
principle are due to inherent memory limitations, and fluctuations in it to the
varying degree to which memory capacity is available in specific circumstances,
or to the varying fitness of the control system.
In the present stage of our knowledge about cognitive phenomena, this is
obviously rather speculative. Yet one should theorize about such matters if one
is ever to provide a serious explanation of linguistic behavior. The principles of
fallibility and limitation should follow from any plausible cognitive model. More
specifically, it has to show the potential for precisely those types of failures and
limits which occur in behavior, and no others (Levelt 1973c). Note that this is
still a matter of potentiality: both the control system and memory are part of the
infrastructure of cognition which makes real action possible. In fact, Chomsky's
claim that theorizing on cognition is about the potentiality which can explain the
factuality of behavior is rather trivial. Insofar as Chomsky's concept of
'performance' stands for the reality of behavior (the factuality of acting), and
'competence' for the potentiality of behavior, there is no need for a 'performan
ce theory' next to a competence theory. Every model about some aspect of
cognition is then a competence model. 'Performance' follows from what is
specified in the model. To take up Chomsky's favored topic of comparison with
biology again (cf. 1.3.3), a model of the heart is a model of a system having the
potential to pump blood. There is no need for a separate model of how the
pumping goes about, for this follows automatically from the first model.
Anyway, the above implies that although a model of knowledge as such
should not bother with accounting for failures of the control system or with
memory overload, in developing such a model one cannot ignore the wider
context of a model in which such things are indeed taken into account. That is,
a researcher can decide only to develop a model of linguistic knowledge and to
leave an account of matters such as control and memory to separate models, or
to a wider-ranging 'performance model' (to use Chomsky's term, although it is
very misleading in this context), which also has to include a model of knowledge.
However, then this researcher has to accept that these other models or the
wider-ranging 'performance model' will have a lot to say about the way linguistic
knowledge can be represented in his/her model, for an answer to the questions
of how a control system functions and uses knowledge and how a memory is
structured and allows input of information obviously has implications for the
question how knowledge could be organized.
Actually, the above suggestions about the role of errors in theorizing
should not be misconstrued. No model of the cognitive systems and mechanisms
determining linguistic behavior will involve an explicit indication of which specific
mistakes can occur in actual language use - it only has to contain the potential
for failures which can actually occur. But this does not change the fact that an
investigation of occurring errors can contribute a great deal to our understanding
of cognition, including knowledge. So, even if one wants only to concentrate on
a model of knowledge, as Chomsky does, although one can start by idealizing
away errors to simplify the data one should not ignore them in the long run. So,
claiming that one should leave them out of consideration altogether is a mistake
in different respects.15
Another characteristic of linguistic behavior which is (nearly) completely
ignored in a Chomskyan conception of knowledge is its variability and changea
bility. Chomsky's concept of knowledge is extremely static and universalistic. Yet
unlike errors, and more in line with matters related to the functional dimension
of behavior, these phenomena do seem to touch upon knowledge directly.
Deviance from 'normal' linguistic behavior certainly does not have to involve
only the inappropriate use of knowledge. In fact, it is often very hard to decide
whether something can be characterized as an error or not, as it is a serious
problem to determine what can count as 'normal' in behavior.
First of all, one has to keep in mind that one can deliberately make
'mistakes' by consciously deviating from one's knowledge in producing an
utterance, e.g. in order to induce some effect in an H/S by being funny, by
mimicking something, etc. In principle one can deviate in a similar way from
one's knowledge in interpreting an utterance, but this is much rarer because it
is not particularly useful (unless one wants to be funny to oneself). However,
there is another way in which one can purposefully deviate from one's
knowledge in understanding an utterance, viz. in correcting errors which the S/H
has made, or in adapting to deviant language use by the S/H. If the input ut
terance does not deviate too heavily from what according to one's knowledge
would be 'correct', one is able to unmask the deviance and further act upon a
corrected version of it (while also interpreting the deviation as such, if there is
a chance that there is an intent behind it). On the other hand, it appears that
at least part of the errors unconsciously made by S/Hs mark a tendency to make
an utterance which is functionally more adapted than utterances which would be
possible according to their knowledge (cf. e.g. Brown 1980).
The borderline between these types of 'erroneous' uses of knowledge and

the 'rule-changing' behavior involved in creative language use is obviously very
small, if there is one at all. At least in the conscious cases, but maybe also in the
unconscious 'functionalizing' cases mentioned, one can assume that the S/H
modifies or adds rules or procedures in knowledge, sometimes just for single
usage, but perhaps sometimes also permanently. Probably the control system has
to be invoked to account for how this happens, but nevertheless it involves a
change in knowledge. Of course, such changes will probably never concern the
most fundamental principles of the cognitive organization of knowledge, and
neither will they involve a massive reshuffling in the system (unless pathology is
involved). This is necessary on purely functional grounds. There has to be
sufficient similarity in the codes used by interlocutors to warrant mutual
understanding. Hence, S/Hs have to take care that their code for interaction in
their 'standard' environment remains close enough to the codes of the other
S/Hs involved. On the other hand, small changes occurring in individual S/Hs'
knowledge are no doubt potential causes for large scale changes in the code
used in a community, and thus lie at the heart of diachronic language change.
Hence, contrary to Chomsky's views there is good reason to believe that
knowledge is changeable, and does change permanently in one individual, even
after the period of language acquisition proper. Change is not a peripheral
phenomenon, and though one may initially want to idealize it away, an adequate
theory of knowledge will nevertheless have to include the principle of change
ability at its very heart, and will have to be able to account for how it operates,
according to which principles, and what the differences in status and/or format
are between 'old' vs 'newly or temporarily created' elements of knowledge. In
fact, this is no doubt one of the domains in which research on adult language
can contribute significantly to research on language acquisition, for understand
ing change of knowledge is one of the core concerns in the latter.
The phenomenon of changeability unavoidably leads us to the related
phenomenon of variability, i.e. the differences in individual, social, and cultural
styles (cf. 1.2) (including languages proper). This involves not only the
interpersonal difference in styles used by different S/Hs, but also the intra
personal phenomenon that one S/H often uses different styles at different mo
ments. Chomsky's attitude toward variability seems illogical. Differences between
languages do have to be described in individual TGs, but other forms of stylistic
variation are completely ignored in characterizing knowledge, as follows from the
notion of a 'homogeneous language community' in his concept of the ideal S/H.
Chomsky (e.g. 1979: 53ff) claims that research such as Labov's (1972b) on so-
ciolects (and ipso facto all research on styles) may have societal, educative, or
political value, but has no theoretical value whatsoever for understanding

knowledge.16 Yet this is hard to come by since the difference between
sociolects and languages seems only gradual. Language-variation is mainly
geographically determined, sociolects are mainly socially determined, 'culturo-
lects' are mainly culturally determined, and these three dimensions interact in
many ways. What should be the criteria making the geographical dimension
theoretically more interesting than the others? It is also hard to reconcile
Chomsky's (1979: 54) statement that individual variability obviously will have to
be accounted for sometime and his (1979: 55) statement that research on
individual variability "on the linguistic level [...] is evident and banal".
There is no way around the fact that all these matters of variability directly
involve knowledge. Again, it may be valid as an interim strategy to focus on the
description of the linguistic knowledge of S/Hs in one specific, idealized (and
thus probably totally imaginary) community, assuming that it is homogeneous.
But one cannot avoid soon coming to grips with the fact that in real communities
there are differences between individuals and groups of individuals, and that
individuals and groups within one community can also belong to other communi-
ties at the same time.17 Of course, this does not amount to specifying in
models of linguistic knowledge each single variable form and each possible
intrapersonal combination of them one comes across. That would indeed be
banal. The point is to grasp the main types of variability, according to all dimen
sions (geographical, social, cultural, individual) in terms of a limited number of
representative prototypes of S/Hs. In fact, to return to our comparison with
biology, research on the physical world is not different from research on the
cognitive world in this respect either. Each instance of a biological organ
somehow differs from all others, yet it does not make sense to develop a model
of each instance, and biologists only develop one or a few prototypes of an
organ. Still, this does not mean that they ignore variability in organs, and the
principles of variability are taken into account to come to an adequate model of
a prototypical organ.18
It is also better not to talk about an 'ideal S/H' in this connection. One can
speak of an ideal S/H as far as the elimination of mistakes in representing
knowledge is concerned: one can assume that each S/H incorporates an ideal
(each S/H has perfect knowledge of a language - idiosyncratically) even though
it never shows up on the surface of his/her behavior. But the principle of
variability involves that each S/H contains another 'ideal'. The prototype being
described is basically a model of one possible S/H, incorporating the ideal
version of this one S/H in the knowledge component. There is only room for the
notion of an ideal within the range of a single S/H. Thus, it is weird that
Chomsky puts the matters of errors and of variability on a par in his notion of
the 'ideal S/H' for they are fundamentally different. The latter gives the
idealization involved a completely different and in fact unacceptable theoretical
status. It introduces a notion of normativity for which there is no room in a
scientific description of cognition. An S/H has a concept of norm which is
somehow linked to his/her linguistic knowledge and perception of the communi
ties (s)he is part of. A scientist cannot claim to be describing a 'norm S/H',
however, because this involves reference to an objective norm-point which does
not exist unless one wants to get involved in dangerous evaluations of right and
wrong. At this level, as an 'objective' observer one can only determine types.
Anyway, as far as the intrapersonal domain is concerned, an adequate ac
count of an S/H's linguistic knowledge has to answer questions such as: how are
the different idiolects, dialects, sociolects, cultural styles, and languages known
by an S/H represented and organized, individually and relative to each other
(this obviously has to involve the distinction between native and second, third,
etc. style or language); how are they related; how does the S/H differentiate
between them in terms of values and possible contexts of usage (the conditions
for talking formally or informally, for speaking a dialect or the standard
language, etc.); how do they interfere; etc.? Such matters can have far-reaching
implications for the formalization of linguistic knowledge. Given the existence
of many different styles and language variants in one's immediate environment,
one has to develop strategies to understand all of these, even though one may
only be able to produce a limited set of them oneself. Thus, an S/H has a much
wider interpretive than productive capacity, and this clearly reflects upon the
question whether linguistic knowledge might be mode-specific or rather mode-
independent, i.e. whether it might be correlated to the directionality of language
production and interpretation or not (cf. 3.2.2).
The importance of understanding the differences between languages and
styles in general is obvious: Chomsky is right that universals have a crucial role
to play in characterizing knowledge, but to discover what is universal in language
one also has to discover what is variable in language. There is no opposition
involved here: universality and variability are complementary dimensions of
language. In fact, variability in itself is a universal phenomenon, the principles
of which have to be understood. In any case, there is no a priori answer to the
question of how much in language is universal and how much is variable, and
Chomsky's universalistic view of linguistic knowledge is unwarranted because it
is aprioristic and not based on an investigation of variability. This is another
example of the self-imposed 'eye-flaps' mentioned above, and we end up with
another circular argument: on the one hand he explicitly idealizes away nearly
all forms of variability, and on the other hand he claims that variability in
linguistic knowledge is minimal.
In fact, even Chomsky's consideration of variability between languages
proper, and the way he abstracts universals from them, can be questioned. Espe
cially in the early days of TG and UG, the assumption was that one can detect
universals on the basis of the analysis of one language. Thus UG originally was
mainly based on English, and attempts to get really involved with typological
research are rather recent. However, logically speaking it is impossible to
postulate universals on the basis of one language (cf. also Dik 1982). One might
react that this may be true for 'substantial' universals (concerning the concepts
and principles directly determining the properties of languages), but since
Chomsky's universals are mainly of the 'formal' type (involving the formats of
rules and rule systems constituting the cognitive systems) it is possible to derive
them from the grammar of a single language. For there is little reason to believe
that in this respect there is much variability between S/Hs of different languages
and styles. Yet how one structures a grammar of a language and which rules and
rule-systems one accepts obviously has to be based on what one encounters in
the language as such. What one postulates about the universal properties of
grammars (and thus about which rules and rule systems have to be present in
each single grammar of a language - cf. 2.1) has to be based on what one finds
to be universal in languages. In other words, the difference between formal and
substantial universals is highly deceptive, in the same way as the difference
between E-language and I-language discussed in 2.2.2.1 is.
Thus, it is not astonishing that the recent involvement with typologically
more diverse languages did bring about a change in the view on UG, and led to
the introduction of the parameter approach to universals (cf. Chomsky 1986a,
b). This concept also introduces a new paradox, however, for it boils down to
juxtaposing forms one finds in different groups of languages rather than trying
to find common (and truly universal) principles underlying all languages. Of
course, it provides an easy way to avoid that 'universals' postulated in an earlier
phase of research would be endangered by new evidence, but the question is
whether this is more than another form of playing fast and loose with data.
Finally, postulating a structure for a grammar in terms of rules and systems
of rules, and thus postulating universals in this respect, is obviously also
impossible if one does not take into account the functional dimension of
behavior. This brings us back to where we started from in this section: the
grammar has to show how syntactic patterns can carry meanings and serve func
tions. Chomsky's refusal to take these matters into consideration in postulating
hypotheses about knowledge also endangers his concept of universality in
language. (I return to the universals problem in 2.3.)
2.2.2.3. Intuitions and linguistic knowledge

To support his concept of knowledge Chomsky frequently refers to the intuitions
of native S/Hs. As indicated in 2.1, he considers judgments concerning grammati-
cality, ambiguity and synonymy of utterances to be of special importance in
supporting the adequacy of TG, and this often involves the suggestion that they
are even more important than the primary data (the utterances) themselves. The
reason for this seems to be that he believes them to somehow provide a direct
access to the competence of an S/H. At least, that is what is suggested if he
characterizes them as "the speaker-hearer's judgments that the facts are as
determined by the grammar" (Chomsky 1975b: 36). This adds to the impression
of competence being an 'obvious' structure (cf. e.g. also the E-language vs I-
language discussion in 2.2.2.1), but again this is illusory.
Obviously, linguistic intuitions are significant phenomena which somehow
have something to do with linguistic behavior as such (speaking and understand
ing), and which do say something about it. The fact that we know that something
is an error, or that a particular utterance is ambiguous or synonymous, is mainly
based on an intuitive judgment of the relevant cases. Hence, these are data
which do have to be taken into account in developing a process model. In fact,
it is simply impossible for a linguist, even for a fervent corpus analyst, not to use
at least his/her own intuitions in research (cf. Pylyshyn 1973). But a more
systematic use of them as advocated by Chomsky certainly is necessary as well.
However, in order to be able to use them in appropriate ways in theorizing it is
important to be well aware of the precise nature of intuitive judgments, and thus
of the precise nature of the information they provide. This requires at least a
rudimentary theory about what is involved in the cognitive process of forming
and expressing an intuitive judgment. What are the characteristics of the
mechanisms generating them, how do these mechanisms (and thus the contents
of the judgments) relate to the systems and mechanisms determining linguistic
behavior proper, and which factors determine or influence the process? Yet the
central role intuitions are assigned in the theory notwithstanding, a a active
consideration of such questions is entirely lacking in TG. In fact, though the
answers to these questions are all but evident, and though at present very little
is known about them (which in itself should be a warning to be more careful
with them than Chomskyan linguists are), what is known is sufficient to raise a
number of objections against the way in which TG uses intuitions.
Linguistic judgments are just one type of reflective consideration about
objects, and are thus only one instance of the much more general capacity of
humans to reflect upon things. What reflection involves is largely unknown, even
within cognitive psychology. No doubt the control mechanism mentioned in the
previous section has something to do with it, but it is by no means self-evident
how or what. It is clear, though, that the objects of reflection in the case of
linguistic intuitions are not the cognitive systems and mechanisms underlying
linguistic behavior, but the utterances produced and understood. The reflection
probably does rely somehow upon the S/H's linguistic knowledge (cf. 2.2.3.2),
and thus does say something about this knowledge. But this does remain indirect
information, like all other types of data.
Hence, intuitive judgments do not provide privileged access to the black
box of cognition. Therefore, Chomsky's claim that an S/H can judge whether the
facts are as determined by the grammar, and thus that these judgments warrant
the adequacy of the grammar, is based on a circular argumentation. Grammati-
cality (conforming to the rules of the scientific grammar) is not an observational
notion, but is and remains a theoretical notion, with or without the intuitions (cf.
Seuren 1977). An S/H says something about utterances being (un)well-formed,
or being (un)acceptable, or being similar or ambiguous, and many more
things. But this does not mean that the S/H says that an utterance has or has not
a structural description which corresponds to what is specified in the rules of the
grammar, or that two utterances have the same underlying structural description,
or that one utterance has two different structural descriptions, etc. These are
interpretations of these judgments by the linguist, which find their reflection in
the concept of grammaticality. (See also Parret 1974, 1979a.) The grammar is
also meant as an explanation of these facts, and in this case not less than in the
case of any other data the question remains whether the interpretations and the
resulting explanation are adequate.
The above also implies that intuitive judgments are certainly not more
important than the primary data of linguistic behavior. On the contrary, since
intuitive judgments of utterances are data of a quite different type than the
direct linguistic behavior, they provide an albeit interesting and even important,
but nevertheless additional perspective. They have a secondary status in that as
reflections upon primary data they can never be considered independently of the
primary data themselves (see also Clark and Haviland 1974, Snow and Meijer
1977, Nuyts 1983a). This appears also to be in accordance with what is known
from the ontogenesis of intuitive judgments (cf. De Viiliers and De Villiers 1972,
Gleitman et al. 1972). Thus, once again, the fact that one considers intuitions
does not at all mean that one can bypass the E-language.
One of the most essential objections to the current use of intuitions in TG,
but certainly also in language research in general, resides in the lack of attention
that is paid to the variability and uncertainty that characterizes them (cf. e.g.
Schlesinger 1971, Lakoff 1973, Gleitman and Gleitman 1979, Ross 1979).
Chomsky (e.g. 1965: 18ff) is certainly aware of this problem, but the point is that
he does not search for the reasons and does not take the problem into account.
Maybe one might suggest that this variability is just a practical problem which
one has to live with: intuitions are just simply uncertain. Obviously, like all other
behavior intuitive judgments are influenced by distractions, tiredness, etc., which
disturb the 'pure' judgments that might come out if it were only for the
knowledge of the S/H. This problem exists for all kinds of behavioral data,
including the primary linguistic utterances. But there are reasons to assume that
at least in part there is more to it. Partial explanations for the variability might
be found, first in the fact that intuitive judgments are task-oriented, i.e. sensitive
to the nature of the questioning and its purpose, and second in the fact that they
are (or at least can be) context-dependent, i.e. involve considerations of
suitability of utterances in particular contexts.
As Lakoff (1974: 156) states, one can characterize a judgment of well-
formedness of an utterance as being a judgment of whether the utterance can
be part of some possible triplet of the form '[sentence, meaning structure, con-
text]'.20 This can involve unambiguous and more or less absolute judgments.
E.g., utterance (5), or the well-known Chomskyan example (6)
(5) John have given from Mary two pair of socks
(6) Colorless green ideas sleep furiously
are respectively on purely syntactic and purely semantic grounds ill-formed, and
probably no one will ever have doubts about this (unless one has such a rich
phantasy that one can create a possible world for (6)). Such utterances simply
cannot be part of a possible triplet of the form indicated above. In many cases
judgments cannot be made in such a clear-cut fashion, however, and it is in these
cases that variability tends to occur. This involves the (of course scalar) category
of 'marked' utterances, i.e. utterances which can only occur under certain limited
contextual conditions. If one assumes that judging utterances means judging
them in a possible context of usage,21 then the S/H's judgment of a marked
utterance depends on (i) whether it is accompanied by an indication of a
context, and (ii) if not, whether the S/H manages to find a possible context. In
the latter case, the more strongly marked the utterance, the harder the
judgment. If the conditions are bad (e.g. if the S/H is short of fantasy, or if the
external circumstances disfavor the contexts in which the utterance would be
appropriate and thus make it more unlikely that the S/H will find them) the
judgment can be quite different from what it would have been in optimal
conditions. (Cf. also Levelt et al. 1977.) If in eliciting judgments a context is
suggested, the judge will normally interpret the task to be to judge the utterance
in this context. If the context is not kept constant, however, and if in one case
a possible, in another case an impossible context for the utterance is provided,
one will get an 'OK' in one case and an 'out' in the other.
These observations lead to two kinds of remarks, the first with respect to
the methodology of using intuitive judgments. Judging utterances appears to be
a process subject to many kinds of influences. The context and the way in which
an utterance is presented for judgment are crucial, as is the precise formulation
of what should be evaluated. In fact, the situation is even more complicated than
what is suggested above, for the influencing context appears to go much further
than just the natural context of an utterance (its being embedded in a discourse).
Even a purely artificial context can influence intuitive judgments. It has been
shown that judgments of the well-formedness of an isolated utterance presented
in a list of other isolated utterances can be quite different depending on whether
the other utterances are clearly well-formed or are rather dubious (cf. Bever
1970a, 1971, 1974, Levelt 1973c). These facts are too easily ignored in TG, but
also in language research in general. (See also Levelt 1972, Labov 1972a, Snow
1975b, Snow and Meijer 1977, a.o.) Moreover, intuitive judgments of utterances
appear to be strongly subject to habit formation: each linguist is no doubt aware
of the fact that one loses a sense for well-formedness of a construction if one is
involved with it for too long. This is devastating especially for the widespread
practice among linguists (also in TG) not to use intuitions systematically (i.e. on
the basis of elicitation from a representative sample of S/Hs) but to rely solely
on their own personal judgments.
A second and theoretically more important remark is that TG does not ap
pear to be in accordance with what intuitions really seem to tell us. The fact that
they involve the question of whether an utterance fits into some context is not
directly an argument for the concept of an autonomous syntactic system of the
type represented in the generative core of TG. Judging an utterance appears to
involve the question of whether it can be functional or not, and this only
supports the view that one cannot avoid taking the functional dimension into
account in theorizing about linguistic knowledge. The existence of absolute
judgments such as the one relating to (5) above does not affect this position. As
argued, there is no dispute over the fact that an S/H knows syntactic principles
but only over how (s)he knows them, and how they are present in the cognitive
systems. In (5) there are serious violations of the syntactic possibilities of English
(contrary to (6), which is syntactically perfect), and they clearly make the
utterance unsuitable for normal functional usage. (See also McCawley 1974.)
Actually, one can also wonder how Chomsky reconciles the claim that TG
should be able to account for intuitions about ambiguity and synonymy with the
nature of this model. In their traditional definition, synonymy and ambiguity have
to do with meanings related to utterances, and one might expect that an account
would require that the relationship of form and meaning be at the core of one's
model.
More generally, it is unclear how TG will ever be able to provide a
satisfactory account for the gradual character of intuitive judgments related to
marked utterances. Of course, one can refer to the distinction between core and
periphery which has been introduced in TG. While in standard theory the aim
was to have the grammar generate all grammatical utterances of a language, the
more recent core grammar versions (Chomsky 1981a) only aim to generate the
'certain' cases of grammatical utterances. Uncertain (marked) cases are referred
to the periphery, and they are accounted for in terms of degrees of deviance
from the core. Yet markedness does not involve 'degrees of (un)well-formedness'
of utterances, but conditions under which they are or are not well-formed. In the
appropriate contexts, marked utterances are perfectly well-formed in the same
way as unmarked utterances are. Again, a serious concept of markedness cannot
avoid taking into account the functional dimension: characterizing the degree of
markedness of an utterance somehow has to be based on a specification of the
type and the degree of specificity of the contexts in which it can be used. It is
not astonishing that his repeated reference to a markedness-theory with respect
to the core-periphery distinction notwithstanding, Chomsky has not yet been able
to come up with such a theory (cf. e.g. Chomsky 1982: 107ff). The absence of
such a theory also casts doubt on how decisions are made about which
utterances should be handled in the core and which in the periphery, however.
If one reads that "We are [...] compelled to rely heavily on grammar-internal
considerations and comparative evidence, that is, on the possibilities of
constructing a reasonable theory of UG" (Chomsky 1981a: 9), it is hard not to
understand this as a matter of taking into account the facts which suit the
existing framework and relegating to the periphery the facts which are likely to
be problematic for the theory. Another illustration of the discussion in 2.2.2.1.
2.2.3. TG and the processes of speaking and understanding
2.2.3.1. On the problem of interpreting TG in a cognitive framework

So far, we have mainly been concerned with the role of aspects of linguistic
behavior in theorizing on linguistic knowledge (the weak reading of the

competence-performance dichotomy). It is time now to turn to the more specific
problem of the relationship of the concept of competence as embodied in TG
to a wider concept of a model of the systems involved in speaking and
understanding, and hence to the strong reading of the competence-performance
dichotomy (cf. 2.2.1). As mentioned, Chomsky himself has not much to say about
this. In fact, even apart from the problem of the total absence of the functional
dimension in the model, it is hard to see how a system like a TG could naturally
fit into a model which tries to come to terms with the unavoidable, bi-directional
picture of the cognitive systems involved in language use (cf. 2.2.1). Formally, a
transformational generative grammar is a system which blindly produces constitu
ent patterns by first expanding an initial symbol S according to rewrite rules and
by then reshuffling them according to transformational principles, and which next
phonetically interprets these patterns as if they were to be produced, but also
semantically interprets them as if they were to be understood. How could such
a system ever contribute to the process of mapping a meaning pattern onto a
phonetic pattern and/or vise versa?
Actually, one should be very careful in interpreting Chomsky's own state
ments about what TG is or does. He often says that TG assigns structural de
scriptions to utterances (using the term 'generate' at the cognitive level in a
completely different way from its usage in the theory of formal grammars, where
it simply means 'produce' - cf. the quotation from Chomsky 1965: 7 in 2.2.1) and
that it is not a productive device but a system which relates meanings and
phonetic patterns in a neutral way. This is very misleading, though. One cannot
give a sentence to TG and ask it to provide a structural description of it. Thus,
it cannot assign structural descriptions to an utterance, it can only produce
structural descriptions which afterwards could somehow be assigned to
utterances. The mechanisms for the latter part of the job are not specified in the
Chomskyan theory, however. Moreover, TG can form phonetic patterns and
logical forms, in pairs based on one structural description, but it cannot relate
sounds and meanings 'in a neutral way' (which can only be interpreted as mean
ing 'in both directions' - cf. 2.2.3.3). Formally, TG is a strictly ordered and
directional system, and if it is said to be a hypothesis about a cognitive system,
then this can only be understood as a claim that this cognitive system is and does
precisely what TG is and does.
Similarly, at a more detailed level, Seuren (1975: 9) claims that one should
not consider transformations to be productive rules, but only neutral mappings
between patterns. The seeming directionality is, according to him, purely a
matter of 'logical priority': if one would not specify transformations from deep
structure to surface structure, one would lose a lot of essential generalizations

and relapse to taxonomism. Yet this is precisely the point of theorizing about
cognitive systems. If a proper description of the phenomena requires that one
assumes logical priorities in the model, and if one aims at cognitive validity in
this model, then this can only mean that one suggests that these logical priorities
also exist in the cognitive reality. Seuren (1975: 11; my translation) states that
it is "remarkable how many linguists have failed to entirely liberate themselves
from the view that transformations would describe psycholinguistic processes",
but what else would be the point of Chomsky's making mentalistic claims about
his grammar? If, formally speaking, a transformation takes in one structure and
changes a number of things to form another structure, or more generally if the
model first generates deep structures, then transforms them into surface
structures, and finally interprets these phonetically and logically; and if this is
presented as a claim about cognition; then one can only interpret this as a claim
that this is precisely what happens cognitively. All other interpretations would
mean a mystification of the concept of a model of a cognitive object. In
presenting a model as cognitively real, one cannot arbitrarily change the meaning
of the formal characteristics of the model.
One of the main problems in fitting TG into a more encompassing
cognitive model, then, is its 'closed', autonomous nature. One does not expect
cognitive systems to produce structures at random from a single symbol, out of
the blue. At the macro level they are input-output systems which take in strings
of one type and transpose them into strings of another type (in case of language,
the transposition is between sound patterns and meaning patterns), and one
would expect to find the same type of system at the micro level of the sub
systems constituting the overall system. Again, our comparison with the physical
world is useful here: all physical organs, without exception, are input-output sys
tems, and as is so often claimed by Chomsky, there is no reason to assume that
in this type of matters the cognitive world is different from the physical world.
In fact, Chomsky does assume an input-output format for the semantic and
phonological components. What could motivate a different format for the syntac
tic component?
Chomsky likes to use the concept of modularity to try and make sense of
his autonomous syntactic system. Yet rather than being modular, his concept of
cognition (or at least of the competence part of it) is partialistic. In his
conception, 'modules' are entirely isolated systems which are not integrated in
the overall system of which they are supposed to be a part. A more plausible
conception of modularity seems to be one which involves sub-systems of
knowledge organized at hierarchical levels but which together form an integrated
system. This is the way in which the physical world is modular as well: the heart
is a separable organ, which is nevertheless fully integrated in the circulatory
system at one level, and in the human body at the highest level, and which
cannot be understood outside this context.
All of this unavoidably leads one to the question of whether Chomsky has
really thought about matters cognitive in first developing his model. As is well
known, Chomsky's initial concern was to find a solution to problems of
immediate constituent analysis which confronted his structuralist predecessors
(cf. Chomsky 1957, 1963, Chomsky and Miller 1963). That is what the core of
TG really is: a formalized system for immediate constituent analysis. This also
implies that Chomsky's sharply distancing himself from his taxonomic predeces
sors (which he keeps repeating even today - cf. Chomsky 1986a, b) is misleading,
for both the object of his research and his attitude toward it (the purely formal
approach) is firmly rooted in the American structuralist tradition (cf. also Chafe
1974, Lakoff 1974, McCawley 1974, Seuren 1975, Derwing and Baker 1978,
Kaldewaij 1986). Of course, his approach to the problem (the hypothetico-
deductive methodology) is entirely different, but even in this respect links with
his predecessors are obvious. The concept of transformation, e.g., had already
been used by Chomsky's teacher Harris (e.g. 1952, 1957). Chomsky's and Harris'
concept of transformation may not be identical, the main difference being that
the latter is non-directional while the former is directional (cf. Chomsky 1974:
36, 1979: 119ff). But the principle is identical: one structure is related to another
through some reordering pattern. Actually, it is ironic that in his cognitive
statements about TG Chomsky would precisely interpret transformation as being
a non-directional phenomenon.
Of course, since constituent structure and order are essential elements of
the organization of utterances, we do have to assume that the linguistic knowl
edge of an S/H contains systems which allow for their production and interpre
tation. Yet it is far from evident that a mechanism which can be used by the
linguist to analyze linguistic patterns detected in languages would be an accurate
hypothesis of these systems. The linguist can be said to know this mechanism as
part of his/her scientific knowledge (of course, in this trivial sense TG does
represent knowledge), but that is something completely different from saying
that the normal S/H knows this mechanism as part of the cognitive systems
determining linguistic behavior. A linguist's theoretical knowledge and an S/H's
linguistic knowledge have to be clearly kept apart (cf. 2.2.3.3).
Hence, it is not unlikely that the cognitive claims concerning TG are a post
hoc interpretation of a system which has really been developed on purely formal
(mathematical) grounds (cf. Steinberg 1975, 1982, McCawley 1974). In any case,
Chomsky's mentalistic views only appeared for the first time in his (1959) review
of Skinner, and they had not been fully articulated before Chomsky (1968) (cf.
also Derwing and Baker 1978). Chomsky (1974, 1975b, 1979) himself claims that
his mentalistic views were already present at a much earlier time, without having
surfaced in his writings, however. If so, there is not much evidence that they
have had a decisive influence on the conception of his model.
Finally, also consider the fact that TG takes a simplicity criterion to be the
main evaluation measure in deciding on the plausibility of TGs. There is little
reason to question the validity of simplicity as a concern in developing a model.
As Seuren (1975) states, and contrary to what McCawley (1968, 1977, 1980)
suggests, though there are no direct arguments for or against it, it has proven its
value in the long history of the sciences in general. In fact, positive evidence for
it might even be found in the fact that human behavior tends to be characterized
by a principle of inertia (cf. 1.2.3.2). According to the same principle, there is no
reason to believe that a child acquires a more complicated system of knowledge
about language than necessary, and thus one can expect adult linguistic
knowledge to obey to some principle of simplicity. Yet McCawley certainly is
right in having doubts concerning the simplicity criterion as used in TG. The
above argument also implies that simplicity is mainly a principle which
supplements (or maybe sometimes counteracts) the more substantial principle
that linguistic knowledge should be sufficiently sophisticated and organized in a
sufficiently efficient way so as to allow the system to be optimally operational.
Thus, in developing (or testing) a model the principle of simplicity cannot be but
a 'rest criterion', to be used next to a more substantial principle which can guide
one in organizing (or judging) it, such as the principle of functionality. The point
is that in TG there does not seem to be such a 'constructive' principle available.
Chomsky rejects functionality as a guideline in theorizing, but he does not come
up with an alternative. (Cf. also Nuyts 1983a.)
Consequently, there is no guarantee that the application of the simplicity
criterion in TG is well-directed. As Watt (1974) argues, simplicity in TG means:
how can we formulate the grammar and the rules in it as generally and concisely
as possible. But simplicity for an S/H is more likely a matter of being able to
formulate or interpret an utterance as easily and/or as quickly as possible, i.e.
having as short a derivation as possible to get in an efficient way to an utterance
or an interpretation of it. This can lead to a quite different type of model.
Chomsky would reply that this is a matter of a performance model, not of a
competence model which is not supposed to say anything about speaking and
hearing. This brings us back to the question whether Chomsky's concept of
competence as represented in TG has anything to say about cognition.
2.2.3.2. Psycholinguists on the plausibility of TG

The prima facie improbability of the cognitive claims of TG is substantiated by
much of the psycholinguistic research which Chomsky's theory has generated
since the early sixties. On the one hand, there have been attempts to test TG as
a mechanism on an experimental basis; on the other hand, there have been
attempts to build a performance model including TG as a central component.
There are surveys of this research in abundance (cf. e.g. Fodor and Garrett
1966, Greene 1972, Levelt 1973c, Fodor et al. 1974, Johnson-Laird 1974, 1983,
Cairns and Cairns 1976, Des Tombe 1976), and a short summary of the results
will be sufficient here.
Experimental research has essentially tested two hypotheses. One is the
'derivational theory of complexity' (DTC), which says that if TG is part of the
performance mechanisms, there should be a measurable increase in processing
load corresponding to an increase in the number of transformations which apply
to the derivation of an utterance. The 'coding hypothesis' says that if TG is part
of the performance mechanisms, we can expect that an H/S memorizes a
representation of the deep structure of an utterance together with a marking of
the transformations which link the deep structure to the surface structure of the
utterance. Despite the initially positive results (which were probably due to the
simplicity of the experiments, and which could often also be interpreted in many
different ways - the eternal problem with this type of data, cf. 1.1.3.2) soon
evidence against the DTC began to accumulate, which led to the conclusion that
the problem may be not that our procedures fail to measure perceptual complexity,
but rather that it is a mistake to claim psychological reality for the operations
whereby grammars generate structural descriptions. (Fodor and Garrett 1966:152)
The conclusions concerning the representations produced by TG were less clear-

cut. Some have taken the experimental results to suggest that they are plausible
(cf. Fodor et al. 1974), but others have argued that the results can be explained
much better on the basis of semantic factors (cf. e.g. Greene 1972). In any case,
the naive version of the coding hypothesis, as e.g. proposed by Mehler (1963),
is certainly wrong. Even after short time intervals, a language user normally does
not remember the syntactic structure of utterances heard but only an enriched
semantic representation of it (cf. e.g. Johnson-Laird 1970, Bransford and Franks
1972, Bransford et al. 1972). Although more recent research has shown some
forms of memory for surface structure features (cf. 3.2.1.2), there is no evidence
whatsoever that a TG deep structure plays any role in memory, or even in
language processing in general (Johnson-Laird 1974, 1983).
In the more theoretically oriented attempts to develop performance models
with a TG in them, the attention has mainly focused on the interpretive proces
ses, i.e. on sentence parsing. One type of attempt was an analysis-by-synthesis
system which was already suggested in Miller and Chomsky (1963). However, it
soon became clear that if one includes TG in a simple and straightforward way
in such a model, one gets unacceptable results since it is a matter of pure
accident whether the grammar is going to produce a structural description which
fits the input sentence, and it can take a gigantic number of trials before the
right description occurs. Solutions to this problem require the inclusion of an
additional set of procedures which actually reduces the role of TG to a mini
mum, up to a point where it could better be eliminated altogether. (Cf. Fodor
et al. 1974, Des Tombe 1976.) Fodor et al. (1974) also discuss the feasibility of
an analysis-by-synthesis system for language production, and conclude that it is
very dubious (cf. also R.Wiese 1983). Another attempt has been to build an
'inverted' TG into a parser, but the problems with this model were hardly less
severe (cf. Fodor et al. 1974, Des Tombe 1976), and moreover this did not really
involve TG anymore, but rather a different model construed on the basis of TG
as a taxonomic source (Levelt 1973c). (On TG and parsers, see also Johnson-
Laird 1983.)
The overwhelming reaction to all of this has been skepticism about TG,
and Chomsky's views in general. The general opinion has been well voiced by
Greene:
when Chomsky makes the stronger claim that he is describing the structure of
cognitive processes, can he ignore the extreme complexity of the relation between
speakers' linguistic knowledge and how it is actually put to use in the real world?
It would be a pity if the previous blindness of psychologists were to be matched
by an equal reluctance on the part of transformational linguists to face up to facts
about language use that do not square with their account of the organization of
linguistic rules. When one looks at all that remains to be discovered about the
learning and use of man's most important faculty, one hopes that, rather than
succumbing to disenchantment or taking up embattled positions from which they
point out only the weaknesses of their opponents' theories, psychologists and
linguists will meet the challenge of trying to find a theory that will account for all
aspects of language. (Greene 1972: 196)
For lack of a positive reaction from Chomskyan linguistics, most language

psychologists have rejected TG and competence-performance altogether as rele
vant notions for developing theories on the cognitive systems and mechanisms
determining linguistic behavior (cf. Marslen-Wilson 1976, Johnson-Laird 1977b,
1983, Flores d'Arcais and Schreuder 1983). Some have put it in clear terms,
stating that "the truth is that the 'mentalism' of most of us generativists is as
irrelevant to our linguistic practice as our taste in movies." (Watt 1974: 32), and
that "generative grammar is unique as a taxonomy having the pretention to be

a process model" (Des Tombe 1976: 134; my translation). Some (in particular
those belonging to the research tradition headed by Fodor, Bever and Garrett)
still accept that TG can contribute something by providing insights for the
development of parsers (as a taxonomy, so to speak). Only very few have
opposed these general conclusions and have further defended the mentalistic
claims in TG, on questionable grounds though.
In Berwick and Weinberg's (1983, 1984) theory of parsing, e.g., TG is still
assigned a very prominent role. In the first place they claim that there is a non
sequitur between the experimental results and the rejection of TG-based parsers.
They state that if one drops one of the implicit suppositions inherent in the
DTC, viz. that all procedures are serial, and if one assumes instead that there
is a partial parallel processing, one can easily develop a parser on the basis of
TG which is in perfect agreement with the experimental data. Of course, since
TG in itself assumes a strict ordering of the rules, allowing for parallel
processing would already mean modifying the model. On the other hand, it is
true that the experimental data as such are not the strongest arguments against
TG: as Valian (1979) notes, it depends on the version of the model one
considers, whether the results apply or not.
Ultimately, these considerations do not matter much for the question of
the plausibility of TG, however. Berwick and Weinberg are talking about parsers
which are (partially) based on TG, or are 'compatible' with it, but there is no
question of TG as such being an integrated part of the parser. They talk about
a 'grammar-parser' pair and argue for a very abstract relation between the two
(they call it a weakening of the type-transparency hypothesis). What this
amounts to is saying that the construction of a parser cannot happen without
taking into account things one can learn from TG, i.e. using TG as a taxonomy
for developing a real process model (which is basically what is also claimed by
Fodor et al. 1974, although they more or less reject the conclusions of the
latter). That this is what is really at stake becomes very clear in Berwick and
Weinberg (1984: 145ff). This does not contain any argument whatsoever for the
cognitive plausibility of TG, however. On the contrary, by proposing this
relationship between grammar and parser they suggest that it is impossible to
integrate TG directly into a performance model, although that would be
necessary for the grammar to be cognitively plausible (cf. the principle of
internal descriptive adequacy, 1.1.2). Even in their framework, the only model
which concerns the cognition of language is the parser. (The same type of
argumentation applies to Marshall 1980 and Bierwisch 1983. See also the
discussion of Black and Chiat 1981 in 1.1.3.3.)22
In this discussion one should keep a clear eye on the role of the formalism
of a model in determining its cognitive plausibility. Cognitive plausibility is not
only a matter of the representations which are postulated to account for the
structure of the data, but also of the mechanisms (the rules or procedures) which
are specified and which handle the representations. As Pylyshyn states, for any
domain of behavior
there are an infinite number of machines that compute any such function and
therefore there are an infinite number of competence theories which will do the
required job. All of these are equivalent with respect to the computation which
they perform, but differ in various other respects - such as in the number of states
they have, the number of cycles they take to perform each computation, the
maximum amount of memory they require, and the relation between the above
parameters and the behavior sequence being computed [...]. (Pylyshyn 1973: 44)
One example of computational equivalence involves transformational generative

grammars and augmented transition network grammars (ATN - Woods 1970).
For each TG an equivalent ATN (which is really an automaton, i.e. an accepting
system) can be developed (cf. Levelt 1973a, Johnson-Laird 1983).
In view of this, one can try to develop all kinds of hierarchies of types of
models or theories about a certain cognitive object (e.g. Marr 1977, 1982, Stabler
1983), but ultimately, what matters is that for each cognitive domain there can
basically be only one (type of) model which accurately characterizes the
processes which take place in the cognitive system of the S/H. It is only this one
(type of) model which, among all computationally equivalent models, can really
be called cognitively real. The other models all belong to the category of
taxonomies. This is not to say that it is always obvious which among a series of
equivalent models can be called cognitively real (cf. 1.1.2). What this means is
that one has to assume that there is just one model which has 'strong equivalen
ce' (Pylyshyn 1980) to the cognitive systems, and that in trying to develop a
cognitively real model one has to try to find this one model. If we observe that
one type of model, i.c. TG, can certainly not have 'strong equivalence' with the
cognitive systems under consideration, we have to conclude that this model is not
cognitively real. (Cf. also Pylyshyn 1972.)
Sometimes it is suggested that maybe Chomsky's aim with TG was only to
provide a cognitively plausible model of representations. His claim that compe
tence does not say anything about how an S/H does things leaves this reading
open. (Cf. Pylyshyn 1973, Catlin 1978.) If that is the case, his model can
obviously not be called a 'competence theory', and his claim that it is a model
of a cognitive 'organ' is then plainly misleading. But his statements in general
leave little doubt about the fact that he means more than this.
Quite a different type of attempt to secure the cognitive relevance of TG

has been to suggest that it may represent a cognitive system, but one that is not
involved in the processes of speaking and understanding, is basically independent
of them, and is only used in forming intuitive judgments about the syntactic well-
formedness of utterances. Thus, TG would represent knowledge, which is
however not used in performance. This view is expressed in Bever (1970a, b,
1971, 1974), who thus rejects the suggestion by Fodor and Garrett (1966) and
others that one should assume an abstract relationship between TG and a
performance model. Bever (1975) continues to use this concept of an autono
mous 'psycho-grammar' underlying linguistic intuitions, without mentioning TG
itself though. (See also Des Tombe 1976 on this matter.)
As an argument for this view Bever (1975) suggests that one cannot avoid
differentiating between the linguistic knowledge as expressed in the intuitive
judgments, and the actual language use, for there is an obvious discrepancy
between the contents of the judgments and the syntax of actual utterances. He
(1975: 69) illustrates this with the classical argument that on the one hand an
embedded construction such as (7)
(7) The dog the cat the cricket chirped at miowed at barked at [sic]
is perfect according to the intuitions on well-formedness, but is unusable, while

on the other hand there are numerous intuitively ill-formed utterances showing
up in behavior. This is basically the argument which has also been used by
Chomsky in relation to his competence-performance distinction, though Bever's
position clearly deviates from the Chomskyan view. In fact, Chomsky would
presumably reject Bever's position since it runs counter to his claim that
competence is essential for, and lies at the basis of performance (but see
Lightfoot 1982: 192ff).
Of course, the claim that (7) is unusable can be contested. (7) (albeit
without the final at) is very well usable under conditions in which the S/H has
time to eliminate memory limitations, to puzzle, etc. - the pure fact that (7) is
occurring on this page and can be understood unambiguously proves this. If (7)
is brought up for someone to judge, the answer will depend on very similar
factors. If it is read quickly, the answer will no doubt be negative. But if the
judge gets time to puzzle (which is normally what is the case in discussions of
this type of utterance: in written format one can read it over and again) (s)he
will have to admit that it is formally okay, although no unbiased judge will
hesitate to add that it is highly unnatural and marked.
Bever (1974) thinks in terms of a rather radical distinction between judg
ments of well-formedness and of acceptability, the former purportedly being due
to the psycho-grammar, the latter to the actual performance mechanisms. He

claims that there are many utterances which are well-formed, yet unacceptable
(like (7)), and vice versa. This distinction is dubious. Even professionally trained
judges are obviously not able to distinguish systematically between the categories
of well-formedness and acceptability. Bever's (1974) own examples of acceptable
but ill-formed utterances all involve cases which many would not determine as
ill-formed but at most as not being in agreement with what according to
normative school grammars would be described as 'correct' in the language.
Moreover, the fact that to a certain degree people are able to understand ill-
formed utterances (cf. 2.2.2.2) does not imply that they would judge these
utterances to be acceptable, contrary to what Bever suggests. If a differentiation
between the two categories is possible at all, this will only be in terms of the
degree of flexibility one applies in judging utterances.
However, even if Bever would be right in differentiating between well-
formedness and acceptability, the type of observations he adduces does not
provide an argument for his distinction between a psycho-grammar and the
performance systems. An argument against Bever's view is that it implies the
rather weird view that there are two types of linguistic knowledge, viz. the
knowledge involved in the psycho-grammar and the knowledge involved in
speaking and understanding (see also Marslen-Wilson 1976). The latter type of
knowledge cannot be avoided, since to be able to use a language one has to
know the language, and while using the language one must somehow be using
the knowledge of the language. The two types of knowledge would obviously
involve a considerable overlap in information. The presumed distinction between
acceptability and well-formedness can, on the other hand, be very well accounted
for on the basis of one system of knowledge, by taking into account the
difference between the knowledge proper and the use of it as discussed in
2.2.2.2. Supposing that intuitions involve reflections upon utterances on the basis
of the knowledge underlying performance (i.e. on the basis of some kind of off
line usage of the mechanisms which are responsible for the on-line production
and/or interpretation of utterances), we can assume that the well-formedness
intuitions are based on the question of whether an utterance is possible in terms
of the knowledge proper (which is not to say that if one comes across a
discrepancy between knowledge and utterance one also knows where or why
there is a discrepancy), while the acceptability judgments are based on the
question of whether interpreting or producing the utterance would put a heavy
burden on the control system and/or the memory.
Bever's argument does not support TG anyway, for the mechanisms
involved in forming intuitive judgments, whatever their exact format, must be
input-output mechanisms, or at least input-mechanisms, since they have to take

a full utterance as a starting point for forming a judgment. Moreover, we are still
facing the problems discussed in 2.2.2.3 concerning the treatment of intuitions
in TG. Maybe this is why Bever (1975) is silent about whether he considers TG
to be a model of his psycho-grammar.
Finally, Bever tries to make a point for his theory on the basis of a
characteristic of language acquisition, viz. that the capacity for speaking always
develops later than the capacity for understanding. He claims that the psycho-
grammar in the adult is a superfluous relic of a system which in the acquisition
of language plays the essential role of mediating between the mechanisms of
production and understanding. According to Bever, without such a mechanisms
a child would never be able to learn to speak.
However, first of all, the disproportion between the productive and the
interpretive capacity, which certainly does exist in the child, to a certain degree
persists in the adult, in that one is always able to understand more than one can
speak (cf. 2.2.2.2), and in that learning, which continues in the adult (e.g. in
acquiring a foreign language), shows the same basic relationship between the
two modes. So, if one assumes a special mechanism in the child, there is no
reason to suppose that this mechanism would no longer operate in the adult.
Second, one can wonder why such a mechanism should exist at all. It is plausible
to assume that at least in part the same knowledge is constitutive for the
processes of production and understanding (cf. 3.2.2), and that the former
operate on an integral subset of the knowledge on which the latter operate. Yet
then the question is how knowledge in general is acquired, and how the
interpretive as well as the productive mechanisms develop. It is hard to see why
the nature of these sub-processes would be so different that for the productive
systems a separate acquisition mechanism would have to be postulated which
would not be required for the interpretive systems. But even if some such
mechanism would appear to be plausible, it is even more difficult to see why this
very same mechanism should also be the source for intuitive judgments. It is by
no means clear what one could have to do with the other.
Anyway, no matter how improbable all of these suggestions are, they do
not change anything about the matter of the plausibility of TG. For obviously
TG does not have anything to tell about how the conversion processes between
the mechanisms of production and understanding could take place either.
2.2.3.3. Knowledge and the capacity for behavior

Chomsky himself has never accepted the results of psycholinguistic research, and
he still stands by his original views on competence and performance and the
cognitivist reading of TG. In fact, it is striking that he has never reacted

systematically to the psycholinguistic research concerning his theory. Moreover,
what has happened in the post-generative era in language psychology is painfully
absent in all of his recent writings. Probably, Chomsky did react in a more
indirect way, by inconspicuously sharpening the competence-performance
dichotomy toward the strong reading of it (the change in attitude toward
creativity - cf. 2.2.2.2 - points in the same direction). Obviously, if one presents
competence as being independent of performance, i.e. as a system which does
not present the way an S/H goes about in speaking and understanding, the
psycholinguistic results can easily be discarded as irrelevant since they are based
on an interpretation of TG as being directly involved in performance. They took
TG to be a working part of a performance model, and thus a model which does
say something about how an S/H goes about in language use. Similarly, the
criticism that TG does not bring together form and function can be rejected
since that can be said to be a matter for the performance systems as well.
This strong reading of the competence-performance distinction adds a new
dimension to Chomsky's static view of linguistic knowledge, which was already
noted in 2.2.2.2. Yet upon closer examination it is as untenable as the aspects
discussed earlier. In fact, the strong reading involves a paradox. Chomsky sticks
to his claim that competence is essential for performance and should be an
inherent and elementary component of the performance systems. It involves the
'knowledge store' required to be able to speak and understand. Now, the only
way to make sense of this claim is to assume that the competence really is used
in performance, for if performance were possible without competence the latter
would not represent knowledge required for performance. But if competence is
used in performance, how can it be that it does not say anything about how an
S/H goes about in speaking and understanding? In a nutshell, the strong reading
of the dichotomy really says that competence is a system which is not used in
performance, although it is used in performance.
One may try to argue that this does not take the difference between
knowledge and the use of it into account. Yet the fact that this distinction is
legitimate (cf. 2.2.2.2) does not mean that knowledge is in any sense independent
of its use. The only way to make sense of the relationship is by accepting that
it is a direct one. There is nothing mysterious about it. Linguistic knowledge
holds the potentiality for behavior, and thus also per definition the latency of the
dynamics of the behavior as activity.
In fact, it is hard to see what would be the point of calling the TG-hypo-
thesis of linguistic knowledge non-active if the parts of the system are all
instructions for doing something (rewriting or transforming), and if the system
in general is a strictly ordered set of such instructions for action. True, through
the years TG has developed in the direction of a model in which the active
character has more and more been backgrounded. Especially in the Government
and Binding theory the base-component is rather implicit, and the trans
formational component is reduced to one single instruction ('do something' - cf.
Chomsky 1986a, b) with the burden nearly entirely being placed on the
formulation of constraints on this act. Yet, apart from the fact that such a
'negativistic' concept of cognition is not very plausible either (one does not act
by considering what one cannot do - that would leave one immobilized - but by
considering what one can do), it does not change the fact that the grammar
basically consists of instructions for (not) acting.
It seems unavoidable that the formulation of knowledge takes the format
of a combination of representations (patterns or schemes) and mechanisms
operating upon them (rules, or procedures, or whatever their name). TG is no
exception. But what should performance be other than activating the mecha
nisms constituting the knowledge which underlies it? More specifically, if a TG
as a hypothesis of knowledge contains the statement 'move something', what
should use of this knowledge in performance be other than doing this, viz.
moving something in an input structure according to the possibilities shown in
the model?
One could try to understand the statement that TG is neutral with respect
to speaking and understanding as the claim that it is a mechanism which does
what it does both in speaking and understanding, and hence that it says some
thing about both modes of processing. The principle of such a mechanism for
double usage is not per se absurd: since production and understanding are
supposedly in large part based on the same knowledge, there is no a priori
reason to exclude that this is realized in the form of shared mechanisms (cf.
3.2.2). This has basically been the interpretation of TG used by the psycholin
guists, with the conclusions discussed above. However, the claim that it does not
say anything about either production or interpretation means introducing a
complete mystification of the relationship between knowledge and use of it.
The above may seem to bring us close to a general type of view on cogni
tion which has repeatedly been criticized by Chomsky as being fundamentally
wrong, viz. what he calls an empiricist conception of cognition as a set of
dispositions or capacities for acting. Chomsky's discussions in this respect always
draw a black-and-white picture of the matter.
To know a language, I am assuming, is to be in a certain mental state, which per

sists as a relatively steady component of transitory mental states. What kind of
mental state? I assume further that to be in such a mental state is to have a cer-
tain mental structure consisting of a system of rules and principles that generate
and relate mental representations of various types. Alternatively, one might
attempt to characterize knowledge of language - perhaps knowledge more generally
- as a capacity or ability to do something, as a system of dispositions of some kind,
in which case one might be led (misled, I think) to conclude that behavior provides
a criterion for the possession of knowledge. (Chomsky 1980a: 48)
By the latter view he appears to mean "the concept of mental capacities as

lacking structured vehicles" (Chomsky 1980a: 49), and he claims (not surprisingly
given his general belief that there are no serious alternatives to his view) that
this concept is "merely a promissory note" (Chomsky 1980a: 48), and that
the attraction will dim when it is recognized how little can be said in terms of 'sets
of dispositions'. Generally the dispositional analysis is put forth on the grounds
that the facts do not compel us to adopt the alternative, which is true, but hardly
relevant. Surely it suffices that the alternative stands as the 'best explanation', if
that much is correct. In the absence of a coherent alternative, and with at least
partial successes to show from the study of theories of rules and representations,
I will continue to assume that it is correct to analyze knowledge of language [...]
in terms of mental structures of rules and representations. (Chomsky 1980a: 49)
Chomsky is obviously fighting ghosts here. I suspect that today nobody

doubts the fact that what Chomsky refers to as 'the alternative view' is unten
able. This fact does not contribute to the plausibility of the TG-hypothesis of
knowledge, however, contrary to what Chomsky suggests. Currently, all models
of cognition assume that knowledge should take the format of some kind of
system of representations and mechanisms (rules, procedures). TG is just one
instance of this concept, next to the wealth of other linguistic, AI, and language
psychological models currently under development.
The fact that the 'alternative view' is wrong does not contribute much to
the plausibility of the strong reading of the competence-performance dichotomy,
or to the claim that it is wrong to characterize knowledge as a capacity or ability
to act, either. In fact, the quotations above are very misleading on this point
(due to the vague usage of terms such as capacity, ability, or disposition), and
they pervert the real points at issue in modeling cognition. There is no a' priori
opposition between the raw positions drawn by Chomsky. On the contrary, the
matter of the S/H's capacity to speak and understand is really the starting point
for our considerations. It is the existence of this capacity which forces us to
account for it. As said in 2.2.1, this is also Chomsky's starting point - of course,
he speaks about the S/H's competence rather than the S/H's capacity, but what's
in a name? (Cf. also Kenny 1981.) Our systems of mechanisms and representa
tions are devised to account for this capacity or competence, hence what they
have to do is show how speaking and understanding are possible or go about,

nothing else. The fact that Chomsky has to develop the opposition sketched
above in order to make TG acceptable is really an argument against TG. Since
the opposition basically involves turning away from his own starting point, this
only shows that his model has not succeeded in contributing to an explanation
of that for which it was meant to be an explanation. He has developed a system
of rules and representations, but apparently in the wrong way.
Chomsky tries to add further considerations against the 'alternative view',
suggesting that they are positive indications for his views. For example, he says:
If, as I am assuming, to know a language is to be in a certain mental state compris

ed of a structure of rules and principles (comparably, for certain other aspects of
cognition), then in theory one could know a language without having the capacity
to use it, a possibility that would be denied on the contrary view just mentioned.
(Chomsky 1980a: 51)
Of course, the extreme position to which he is reacting cannot handle this fact.
But a more sensible view in which the linguistic systems involve knowledge to
speak and to understand does not have difficulties with this. In terms of the
basic concept of cognition sketched in 2.2.2.2, in which the capacity for linguistic
behavior does not only involve knowledge, but also a control system and opera
tional space (memory), it is perfectly possible to have a situation in which the
knowledge is intact but in which the control system or the memory are disabled,
thus making the use of knowledge impossible. In this view it is also perfectly
sound to claim, as Chomsky does, that it should be possible in principle that two
S/Hs have identical linguistic knowledge yet differ in their capacity for language
use: this is the case if they differ in the capacity of their control system or their
memory space.
Actually, one should again be careful with Chomsky's argumentation in this
respect concerning the (methodological) question of what behavior tells us about
knowledge. He argues that it is wrong to use behavior as a criterion for the
presence of knowledge (cf. the quotation from Chomsky 1980a: 48 above). This
is another attempt (cf. 2.2.2.1) to distance himself from behavior as a criterion
to judge a theory of cognition. One fictive but possible example of intact knowl
edge without the capacity to use it that Chomsky brings up is the case of an S/H
who as the result of brain damage is no longer able to speak or understand
language, but who after some time completely recovers and regains full capacity
for behavior. Chomsky is absolutely right that it is logical to assume that during
the aphasia this S/H knew language perfectly, even while being unable to use it.
But then he brings up a second fictive example of an aphasic person who is
similar to the first one, but who due to an additional neurological disturbance
never regains the capacity to speak or understand. Chomsky claims that it would
be perverse to assume that this second aphasic did lose the knowledge of
language, and this is considered proof that linguistic behavior is no criterion for
assuming knowledge. This argument is fallible in several respects, however.
First, granted that in principle the aphasic can have kept his/her knowl
edge, the assumption that (s)he has knowledge, and which knowledge (s)he has,
can only be based on observations of the behavior (s)he showed before the
neurological damage took place. Though Chomsky is right in claiming that there
is no simplistic correlation between the assumption of knowledge and the
observation of behavior, there nevertheless always must be some indication from
behavior in order to be able to postulate knowledge. In fact, because our second
aphasic never regains the capacity for behavior, we actually have no means to
decide on whether after the brain damage the knowledge is still intact, or
whether it has been partially or completely disturbed by the lesion, or whether
it is simply gradually dying away because it cannot be used by the control system
anymore.
Chomsky argues that we could use neurological evidence to decide on this,
and in his discussion of our examples he suggests that this can be done without
any problems, by presupposing (and thus presenting as known) that the addition
al disturbance in the second patient did not affect the neurological locations of
knowledge. But the point is that such facts are not evident at all, especially in
our present state of knowledge about the relations between the cognitive and the
neurological dimensions (cf. 1.1.2). Chomsky has no means of knowing whether
the additional neurological disturbance of the second aphasic is indeed irrelevant
for linguistic knowledge. In fact, if the constellation of factors which Chomsky
describes would really occur, every sensible neurolinguist would start speculating
that the area of the brain additionally affected in the second patient might very
well have something to do with linguistic knowledge. Yet it is very hard to decide
on this, for it is very well possible that a very similar disturbance in one aphasic
is, and in another is not relevant to the linguistic knowledge. Clinical research
has repeatedly shown that very similar brain lesions can lead to strongly
diverging behavioral disturbances.
If my argumentation in 1.1.2 is correct, it will actually never be possible to
systematically use neurological evidence in matters such as the one under
discussion here. As argued, finding correlations between neurology and cognition
will most probably only be possible on the basis of full-fledged cognitive and
neurological theories. To put it in terms of a caricature: it will always remain
impossible on the basis of neurological analyses to determine what it means that
a language user knows English or Chinese, or more in general what it means to
know something, and what the format and the contents of this knowledge are (as
it is impossible to derive from a cognitive theory what the neurological correlates
of cognitive systems are). Of course, once it has been established on the basis
of a neurological and a cognitive theory what the neurological correlates of
knowing a language are, one could check whether our second aphasic still has
this neurological infrastructure. In any case, one will first need to develop a
cognitive model of what it means to know (a) language, and that can only be
done by considering behavioral data.
Chomsky continues his argumentation by postulating that there is no
reason not to assume that a child in some phase of its development can have a
complete knowledge of its language without actually being able to use it. Here
the problem is that on the basis of its behavior there has never been any
indication that the child has complete linguistic knowledge. Again, Chomsky
argues that all kinds of evidence could be adduced. Yet what he puts forward
as possible evidence is either neurological, for which the above remarks remain
valid, or are totally fictive behavioral facts, yet no facts which would even
remotely show that the position suggested is plausible.
Let's also consider Chomsky's discussion of whether knowledge should be
characterized as 'knowledge how to do something' or as 'knowledge that
something is the case' (Ryle 1949). Since he often characterizes linguistic
knowledge as 'S/H's theory of language', many have understood this as involving
the latter. (E.g. Harman 1967, 1969, Wells 1969, Schwartz 1969, Cooper 1975,
Neumaier 1979.) This runs counter to the common assumption that linguistic
knowledge is knowledge how, an assumption which is also implicit in the
discussion above (though one should be careful with this - cf. below). In fact,
Chomsky (e.g. 1969a, b, 1975c) explicitly rejects the characterization in terms of
knowledge how, and he even questions the relevance of this qualification for
discussing 'tacit knowledge' of an S/H at all. To support his argumentation, he
puts forward a number of comparisons of language with other behavioral
phenomena involving knowledge, and it is useful to consider these at length.
the discussion of 'knowing how' often takes a [...] misleading course [...]. It is said
that to know how, to be able to act intelligently, is to have a skill, a disposition or
complex of dispositions to act in a certain way. Consider the standard example,
knowing how to play chess. This is sometimes described as a skill acquired by
training, a complex of dispositions to make certain appropriate moves, a skill that
improves with practice. A person is said to know how to play chess if he normally
makes the permitted moves, avoids the wrong moves, and so on (Ryle). But there
is a possible source of confusion here. Consider the difference between the two
questions: 'Does John know how to play chess?'; 'How well does John (know how
to) play chess?' It is the latter, not the former, that asks about a skill and how well
it is exercised. The former asks about possession of knowledge, which can be

partial or improved only in the way in which knowledge of facts can be partial or
improved.
Knowing how to ride a bicycle may be a matter of having certain habits and
reflexes, and knowing how to drive a car may be a matter of having certain skills.
Similarly, being able to play chess well may be a matter of having certain skills,
acting thoughtfully, and so on. But knowing how to play chess seems to me more
like knowing how to get from New York to Chicago than like how to ride a bicycle
or drive a car. If a person does not know some specific rule of chess, say the rule
of castling, he may still play chess very well; he may (in principle) be world
champion, although it would still be correct to say that he does not know how to
play chess in the way the rankest amateur who loses every game may very well
know how to play chess. We may say that this chess champion has partial
knowledge, but his defect of knowledge is very different from the defects in
knowing how to play chess on the part of the poor player who knows all the rules
(perhaps unconsciously, without ability to formulate them). Discussions of skills,
dispositions, thoughtful action, and so on, seems to me appropriate in relation to
ability to play well, but not in relation to knowing how to play; the latter concept
has an irreducible intellectual component.
It is true that we say that a person knows how to play if he normally makes
the right moves, etc. (as Ryle puts it); that is, his actions provide the evidence that
leads us to attribute to him the knowledge how to play chess - perhaps erroneous
ly. But to attribute to him this knowledge is not the same thing at all as attributing
to him the ability to play well, perhaps on the same evidence. The concepts are
entirely different, though the evidence for applying them may overlap.
Knowledge of language, knowledge of classical mathematics, knowledge of
chess, seem to me to be instances of 'knowing how', if at all, only in the sense in
which knowing to get from New York to Chicago is an instance of 'knowing how*.
The latter implies no particular ability to get from New York to Chicago. That
requires other talents apart from knowing how to get from New York to Chicago.
And, as in the case of playing chess, the person most successful in getting from
New York to Chicago, repeatedly and characteristically, may not know how to get
from New York to Chicago as completely, as fully, as perfectly, as someone who
repeatedly fails. Similarly, two people may know English in exactly the same way,
they may have exactly the same knowledge of language, and yet differ in their
dispositions, in how well they speak and understand the language that both know;
as they might in principle match perfectly in these dispositions, and yet differ in
their knowledge of language. Ability to speak and understand involves not only
knowledge of language, in the strict sense, but much else as well. The concepts
should be kept quite separate. (Chomsky 1975c: 316-318)
First, it should be pointed out that the difference between 'capacities' such
as biking or car driving on one hand, and playing chess or going from NY to
Chicago on the other, may be qualitatively smaller than Chomsky suggests. This
becomes clear, at least in part, if one considers that with respect to the former
as well, one can differentiate between the questions of whether someone can do
them and whether someone can do them well. Thus, it would be a mistake to
underestimate the cognitive component in being able to bike or drive. Of course,
there are serious gradual differences between these various phenomena: the
cognitive component is obviously larger in the ability to play chess than in the
ability to bike or drive. The importance of the cognitive component in the latter
depends on how one defines these capacities. The actions as such only imply
knowledge of how a bike or a car functions and hence how one can move them.
For the rest this capacity is only based on motor control.23 However, if one
also includes knowledge of traffic regulations, the cognitive component obviously
increases considerably. Playing chess, on the other hand, involves a considerable
knowledge component but hardly a motor component. Being able to get from
NY to Chicago strongly involves both.
Secondly, one should be well aware of one main difference between all
behavioral phenomena mentioned so far and the capacity for language, and this
involves the notion of consciousness. This difference is likely to be qualitative,
and therefore it questions Chomsky's comparisons at least in some respects. This
requires some elaboration.
In his characterization of knowledge Chomsky is very easy-going with the
factor of consciousness. This already follows from his specification of it as an
'unconscious theory'. This expression has something of a contradictio in terminis.
In its every-day usage the term 'theory' involves a person's concept or model of
an object. This person is normally a scientist, but one could generalize the term
to include everyone involved in analyzing an object. Thus, in a wide sense it
could include a mechanic who has a theory about how cars function and about
why specific cars under consideration do not function properly, or John Q.Public
who has a theory about why one should or should not believe in a god. In any
case, this always involves conscious knowledge of facts or hypotheses, about
which the person having it normally can report. Combining it with the notion of
'tacitness' or unconsciousness is weird, to say the least.
Actually, an average S/H certainly does have a 'theory' about language (i.e.
conscious knowledge and belief about it) next to 'tacit knowledge' of language.
However, one can be sure that the cognitive status of these two 'types' of
linguistic knowledge is quite different. Each normal S/H acquires the unconscious
knowledge required for linguistic behavior in childhood, without there being
anyone who says what this knowledge (the representations and mechanisms
constituting it) should look like, and without ever having to become conscious
of the fact that (s)he is acquiring this knowledge, or being able to become
conscious of which mechanisms and representations (s)he is acquiring. This
knowledge arises simply by being confronted with linguistic behavior in the

environment (albeit for a long time, and without the behavior being deficient,
contrary to what Chomsky claims - cf. 2.3.1). Of course, there are moments of
consciousness in the child acquiring language, e.g. when it starts playing with
language, when it starts asking questions about the correctness of forms, etc. But
these concern the language as such, not the knowledge of it.
The amount and the contents of conscious knowledge (henceforth
'knowing') about language most probably differ wildly from individual to
individual, depending on the way in which one is consciously confronted with
language. This knowing is acquired by consciously reflecting upon one's own and
others' linguistic behavior, and someone who is constantly working with language
(a writer, a translator, a linguist, or even a politician) will obviously acquire more
of it than someone who is never really involved with language as such; although
everyone sometimes gets involved in language games, irrespective of educational,
social, or cultural backgrounds. In Western society nearly everybody also
acquires 'knowing' about language through school education, which involves
teaching of rules which are supposed to apply if one wants to use a (native or
foreign) language correctly. What one learns in this respect obviously depends
on the school grammar used, the views of the teacher, etc. Anthropological
research has shown that also in less- or non-schooled societies conscious
knowledge about language can play a very important role in the culture. Langua
ge can e.g. be an object of taboo, or it can be ascribed magical values, etc. The
individuals in these societies also acquire a very specific theory about language.
Obviously, the tacit knowledge of mechanisms and representations for
speaking and understanding is independent of, and especially as far as the native
language is concerned is primary relative to, one's knowing about language. One
does not need to know anything about language in order to be able to use
language. However, that does not mean that knowing about language cannot
influence one's tacit knowledge for behavior, or at least the use of this
knowledge in behavior. A writer, for example, no doubt arrives at a certain type
of language use through conscious reflection on what language is and can do,
and by adjusting performance to what (s)he believes in this respect. Also,
learning a foreign language is often (at least in part) accomplished by means of
consciously learning the rules of the language described in some school
grammar. One then starts using the language by consciously applying these rules.
It is known, though, that this is a very laborious way to learn a foreign language
(if alone because the rules of the school grammar are never complete and
entirely accurate). Direct exposure to the behavior of native S/Hs in a natural
environment is often a much better way. If one does use the path of consciously
learning rules, one will become fluent in the foreign language only if one no
longer has to use one's conscious knowledge of it to analyze or construe
utterances, and if one does not have to think about it anymore.24
How one should conceive of the cognitive status of conscious vs unconsci
ous knowledge and what this means in terms of the format of the cognitive
systems involved is an open question. No doubt the control mechanism (cf.
2.2.2.2) has to play a role in an account. Probably one needs a distinction
between knowledge systems which are directly accessible by the main or central
control mechanism, which would be the carrier of our 'self-consciousness', and
systems which are not accessible by the central processor and can only be con
trolled by what in AI are often called 'demons' or 'experts'. Maybe one has to
assume that some of these demons can still be directed by the central processor
(cf. 3.2.2), but they certainly never allow it to enter their domain and they never
provide information to it about how they go about performing their task. Such
a system would also allow one to account for the fact that 'conscious' knowledge
is not really always conscious but is rather knowledge which can be made con
scious. It is not conscious when it is not accessed by the central processor, which
means either that it is not used, or that it is used by a type of demon or expert
which can do the job for the central processor, but which allows the central
processor to take over if it wants. I return to this in 3.2.2.
Anyway, when Chomsky is talking about linguistic knowledge he obviously
has in mind the unconscious type (and he would do better not to use the notion
of a theory). However, returning to his comparisons, the cognitive component
in being able to play chess or to go from NY to Chicago, or even in being able
to bike or drive, each time involves conscious knowledge. Chomsky's suggestion
in the above quotation that playing chess involves unconscious knowledge is
inappropriate. Every chess player, good or bad, is able to explain the rules of the
game. Playing chess, going from NY to Chicago, biking, and driving are all
matters which are learned consciously, and which remain conscious, even though
some of them may become automatized (which in the above framework means
that they become controlled by an expert which at any time can give over
control to the central processor).25
Even if we suppose that, despite this crucial difference between language
and the other capacities, the matter of the relationship between capacity and
knowledge is comparable in both, the comparison only leads to further argu
ments against Chomsky's concept of knowledge. Although in the above quotation
he again tries to argue for a sharp distinction between knowledge and capacity,
he only manages to adduce points which imply that a capacity has to involve
more than just knowledge. Let's scrutinize his characterization of playing chess.
Knowing chess is obviously a flexible notion, but let's assume that one is in this
'state of mind' if one knows all and just all basic rules of the game: the starting
positions and movement and attacking possibilities of the different pieces, the
principle of turns, and the final goal and types of possible final results and when
they apply. These are the building blocks of any game of chess, and a player is
not allowed to deviate from them. Chomsky is obviously right in claiming that
to know chess in this sense by no means implies that one is able to play it well.
Though it is as evident that it does imply that one is able to play it, even if very
poorly. Thus, the capacity to play chess, good or bad, unavoidably starts with
knowledge of the rules, and it is entirely dependent upon this knowledge.
To be able to play chess well, next to the knowledge of the basic rules one
also needs a general capacity for thinking and problem solving which allows one
to combine the single rules of the game into complex strategies, to analyze a
game at any given moment in terms of the possibilities available to each player,
and to plan ahead and adapt one's strategies to the new circumstances which
occur after each turn. In part this is a matter of developing and memorizing
general strategies which can be applied time and again in playing chess (cf. the
different types of 'classical' openings), and which consist of fixed combinations
of the basic rules of the game (this is also clearly a matter of knowledge), and
in part it is a matter of the hic et nunc combination of the basic rules into
strategies, or the adaptation of general strategies to the particular circumstances
of the current game, which is always different from all other games. The
difference between playing well and playing badly is a matter of being or not
being experienced and/or inventive in how to combine the rules into strategies
appropriate to the circumstances. This shows that being able to play chess well
remains largely a matter of knowing the elementary rules. All that is additionally
required is sufficient memory space to store information about the current game,
and a sufficiently sophisticated central processor which can keep track of the
game and apply the appropriate rules at the right moments.
In Chomsky's example of a player who does not know a rule like 'castling',
we are faced with someone who actually knows and plays a game which is
somewhat different from normal chess (the rule of castling is one of the basic
rules of chess and therefore co-defining for the game). Maybe this player might
be able to beat someone who does know the game in its entirety, if (s)he has
much more experience or is more inventive in combining the other rules. But
apart from the fact that one can be sure that a chess player who reaches the
level of the world championship does know the rule of castling, everyone know
ing chess knows that someone who never applies this rule in games has little
chance to win against someone who does apply it and who is otherwise as good
in playing the game. Castling is too important in too many circumstances.

Anyway, Chomsky's exposition does not bring him any further in the matter
of the relationship of knowledge and capacity. As stated earlier, to know chess
is to know the starting position of the pieces and to know how one can move
them (the types and sizes of the moves they can make) and how they can attack
the opponent's pieces. Thus, if one knows chess (if one can play chess, well or
poorly), one knows nothing but a series of minimal directions for action, which
depending on the circumstances one can actually apply in playing the game, and
which one should not violate (although one is obviously able to do so). Of
course, the relationship between knowing and actually playing the game is again
not simplistic. A player can know the rule of castling but can fail to use it
because (s)he is clumsy or fails to recognize its advantages in the game. In this
case one does have a means to check whether the player knows the rule: since
this knowledge is conscious, one can simply ask. If one would want to investigate
this knowledge, this would put one in a situation completely different from the
one a scholar of language is facing. One does not even have to observe the
player performing the game to be able to ascribe knowledge to him/her.
Nevertheless, as in the case of language, the observation that a player normally
makes the right moves and avoids the wrong ones would again be a sufficient
criterion to claim (s)he has knowledge of the game. The degree to which (s)he
manages to make accurate moves even in very complex situations is a criterion
to decide on whether (s)he is a good or a bad player. There is nothing special
going on here, contrary to what Chomsky suggests in the above quotation.
It seems, then, that the matters of knowledge and capacity in chess and
language are comparable to a certain degree, but what the comparison shows
is a picture of the cognition of language of the type sketched earlier on in this
section, rather than the picture of it sketched by Chomsky. Basically the same
picture would evolve if one would further analyze the other examples in the
above quotation from Chomsky. I will not elaborate much on them, but it is
useful to take a quick look at the case of going from NY to Chicago.
First, it is much harder to determine what it means to know how to get
from NY to Chicago, and it is unclear what it would mean to say that one knows
'completely' how to do this, as Chomsky does. Not only are there tens or even
hundreds of ways to do this (and everybody can know one or more different
ways), it is also unclear how to delimit knowledge of one possibility. Consider the
possibility of traveling by train. In that case knowledge of how to get from NY
to Chicago also involves knowledge about how to travel by train, where train
stations are situated and what one has to do in them, how one can buy tickets
and which money and banking systems exist and can be used in this case, etc.
Even apart from this, what would knowledge about how to get from NY to
Chicago be but knowledge about possible actions to take so as to bridge the
distance involved? What would (consciously) doing it be but actually taking these
actions? Again, the knowledge is an inherent component of the capacity to get
from one place to the other, and a component which is directly and straightfor
wardly involved in actually making the trip. All kinds of things can go wrong
which have nothing to do with knowledge, but this does not affect the basic
relationship of knowledge and capacity.
Whether these cases should be called 'knowledge how' or 'knowledge that'
is not immediately obvious. They are a bit of both, depending on how one consi
ders the matter. Knowledge that implies conscious knowledge of facts, and that
is what is involved in the phenomena under consideration. Knowledge of chess
is knowledge of the fact that the rules of the game are the way they are. Yet at
the same time it is knowledge how, for the rules are directions for action, saying
what you should do to play the game. In the case of linguistic knowledge
underlying speaking and understanding, however, one can certainly not speak of
it in terms of knowledge that, since there is no consciousness involved. Maybe
it would be better in this case not to talk about knowledge at all, because of the
potential confusion with the phenomenon of conscious knowledge, but if one
does want to use the term, the only characterization which is more or less
accurate for what is involved is 'knowledge how'.
To conclude, let's summarize the main point. The assumption that there
is linguistic knowledge derives from the observation that there is linguistic
behavior. Thus, a model of knowledge can only be such that it accounts for the
behavior, and nothing more than the behavior, in a straightforward way.
Accounting for the behavior means uncovering the processes of assigning a
sound pattern to a meaning and vice versa. A model which is not able to
contribute to this, such as is the case with TG, cannot count as a plausible model
of knowledge. The argument that this knowledge is not involved in speaking and
understanding is fallible, for only of knowledge which does underlie speaking and
understanding, and which therefore does find expression in behavior, can we
know that it exists, and can we develop theories. Even if there would be knowl
edge independent of behavior, we would have no way to discover it since we
would have no access whatsoever to it. Hence, the only sensible way to under
stand the strong reading of the competence-performance distinction is as an
attempt to escape from the pressure of data so as to safeguard the cognitivist
claims of the theory.
23. On the development of language
2.3.1. LAD and language acquisition
The second part of Chomsky's theory, his concept of an LAD, has not been less
of an object of discussion. (Cf. e.g. Derwing 1973, Levelt 1973c, 1975, McCawley
1977, 1980, Neumaier 1979, Schlesinger 1982.) Let's see whether the matter of
the development of language can tell us anything more concerning the
plausibility of Chomsky's adult language theory, as he claims it does.
As mentioned in 2.1, for Chomsky the matter of developing a language
acquisition theory is a matter of providing a 'deeper' explanation for the
competence of the native S/H. This boils down to providing a model of B in
scheme (4b).
(4) b. primary linguistic dataBgenerative grammar
Through the years there have been some changes in how Chomsky conceives of
B, although the main principles have remained the same. He basically believes
that the problem a child faces is identical to the problem a linguist faces in try
ing to select an adequate TG for a particular language (cf. 1965: 33). Language
acquisition takes place in a very short period of time, on the basis of a limited
sample of primary linguistic data, which according to Chomsky are moreover
very deficient. Still, the child appears to have remarkably few problems in
acquiring its TG. (Cf. also Chomsky 1986a, b.) To explain this, Chomsky claims
one has to assume that
A child who is capable of language learning must have [...]

(i) a technique for representing input signals
(ii) a way of representing structural information about these signals
(iii) some initial delimitation of a class of possible hypotheses about language
structure
(iv) a method for determining what each such hypothesis implies with respect
to each sentence
(v) a method for selecting one of the (presumably, infinitely many) hypotheses
that are allowed by (iii) and are compatible with the given primary linguistic
data
(Chomsky 1965: 30)
A language-acquisition device that meets conditions (i) - (iv) is capable of utilizing

such primary linguistic data as the empirical basis for language learning. This
device must search through the set of possible hypotheses G1 G2, ..., which are
available to it by virtue of condition (iii), and must select grammars that are
compatible with the primary linguistic data, represented in terms of (i) and (ii). It
is possible to test compatibility by virtue of the fact that the device meets
condition (iv). The device would then select one of these potential grammars by
the evaluation measure guaranteed by (v). (Chomsky 1965: 32)
This might suggest that the evaluation measure, i.e. the simplicity criterion, is the
most important element, yet Chomsky makes sure to point out that
the real problem is almost always to restrict the range of possible hypotheses by
adding additional structure to the notion 'generative grammar'. For the construc
tion of a reasonable acquisition model, it is necessary to reduce the class of attain
able grammars compatible with given primary linguistic data to the point where
selection among them can be made by a formal evaluation measure. This requires
a precise and narrow delimitation of the notion 'generative grammar' - a restrictive
and rich hypothesis concerning the universal properties that determine the form
of language. (Chomsky 1965: 35)
This hypothesis is, of course, UG. Chomsky grants that this system must leave
sufficient room for language variation, since a child, no matter where it comes
from, is able to learn any language spoken in its environment. But since he is
convinced that variability between languages is relatively minor, and since the
speed of the acquisition process is according to him incredibly high, U G must
be a very rich and specific hypothesis which must be assumed to be innate.
The principles of general linguistics regarding the nature of rules, their organiza
tion, the principles by which they function, the kinds of representations to which
they apply and which they form, all constitute part of the innate condition that
puts a limit on admissible hypotheses. (Chomsky 1972: 171; quotation marks
omitted)
There is every reason to suppose that [the] mental organ, human language,
develops in accordance with its genetically determined characteristics, with some
minor modifications that give one language or another, depending on experience.
(Chomsky 1976: 57)
The more recent version of LAD seems to be even stricter in the assump
tions made in this respect. At least since Chomsky (1980a, 1981a) the simplicity
criterion has disappeared from it, 26 and the UG, now in its parameter format,
takes on the entire burden of explaining the child's acquisition of its grammar.
We no longer consider UG to provide a format for rule systems and an evaluation

metric. Rather, UG consists of various subsystems of principles (it has the modular
structure that we regularly discover in investigation of cognitive systems), many of
which are associated with parameters that have to be fixed by experience. The
parameters must have the property that they can be fixed by quite simple data,
since this is what is available to the child [...]. Once the values of the parameters
are set, the whole system is operative. [...We] may think of LIG as an intricate
system associated with a finite set of switches, each of which has a finite number
of positions (perhaps two). Experience is required to set the switches. When they
are set, the system functions. (Chomsky 1986a: 30)
(cf. also Lightfoot 1982: 32ff, Chomsky 1986b). In any of the versions, "learning
is primarily a matter of filling in detail within [this] structure that is innate"
(Chomsky 1975a: 39).
Obviously, Chomsky is not offering a complete theory of language acquisi
tion. He only aims to provide the innate 'initial stage' with which an S/H in
development can start acquiring language.27 Yet, as with the competence
model relative to performance, his claim is that this LAD must be the core of
any acquisition theory. One can seriously doubt whether this theory says anything
at all about language in the child, though.
First, consider the way Chomsky conceives the problem. To say that a child
is in basically the same position as a linguist trying to develop a grammar for a
language sounds strange, to say the least (cf. also Schank and Wilensky 1977,
Putnam 1980a). There is no comparability whatsoever between the process of
developing a hypothesis of a cognitive object O by an adult researcher and the
process of acquisition of O by a child, i.e. between the mechanisms of conscious
scientific reasoning to arrive at knowledge about what S/Hs must know to use
language, and the unconscious development of the capacity to use language. The
utterances playing a role in each may be of the same type (though even that is
questionable - cf. 2.3.2), but there is an enormous difference between the
conditions under which each 'task' is performed, both internally (concerning the
organism's - the scientist's vs the child's - internal make-up) and externally
(concerning the environmental conditions for the process). In fact, it is weird to
talk in terms of a child 'having a theory of language' and 'choosing between
alternative hypotheses' which it 'has formulated on the basis of a structural
interpretation of the data', for reasons similar to those discussed in 2.2.3.3 with
respect to the adult's 'theory' of language.
Second, consider the way Chomsky methodologically handles the problem.
He treats it in purely logical terms, according to the following parameters: given
the adult competence as a formalized mechanism, and given a certain type of
input for the system, what should we assume as being initially given for the adult
competence to be learnable or acquirable?28 Of course, a logical analysis of
the acquisition problem can be a valid contribution to understanding it, provided
the logical terms sufficiently resemble the real conditions of the process. But that
does not quite seem to be the case with Chomsky's analysis. Firstly, his
conception of the problem is entirely based upon the adequacy of TG as a
model of adult competence. Our discussion in 2.2 gives reason to doubt the
validity of this claim. Secondly, he starts from assumptions about the input data
of the acquisition process without having done any investigation into what they
really are. In fact, his conception of the problem idealizes away the entire
process of development of the behavior and the input in each phase of it.
Chomsky (e.g. 1965: 32nl9, 35n22, 1981b: 10) argues that his approach is a
treatment in principle, in which for the sake of the argument the acquisition is
turned into an 'all at once' and instantaneously accomplished process (see also
Pinker 1981: 60ff, Hyams 1986: 168ff). He admits that 'real-time' acquisition is
different, but he has never paid closer attention to it. In fact, he has never paid
any attention to empirical research on the acquisition process at all (obviously
because he does not consider it to contain any theoretical perspectives - cf.
Chomsky 1975a, 1980a, Lightfoot 1982).29
Yet consider what the formulation of a real hypothesis of the genetic basis
of language acquisition would require. The indirectness problem discussed in
1.1.2 is especially strong here. The genetic basis does not express itself in any
simple or direct way in observable behavior. Rather, it leads to a very complicat
ed pattern of development of also non-observable cognitive systems, which in
their different stages of development do cause observable behavioral phenome
na. Thus, determining what is innate is only possible if one has a serious picture
of the entire developmental pattern of the different cognitive systems involved
in language use, including the adult systems. This obviously requires a meticulous
investigation of the behavior in each phase of the development, and of the exact
quantity and quality of the input and the responses of the systems in terms of
possible influences on their further development (cf. Lieven 1989).
Thus, Chomsky's logical problem holds little resemblance to the real prob
lem of language acquisition, and under the conditions he specifies one simply
cannot claim to be able to provide a theory of the genetic systems underlying the
acquisition process. A syntactician concerned with the constituent structure of
adult linguistic utterances has no means of saying anything about it. (Cf. also
Cohen 1981.) In other words, Chomsky's LAD is no hypothesis about some
innate mechanism, and his claim to the contrary is dogmatic and ad hoc.
Of course, nobody can deny that language universals do have a very special
relevance for questions about the genetic determination of language. However,
first, a set of linguistic universals is and remains a hypothetical construct of
which the plausibility is tied up with the plausibility of the theory from which it
derives. This is especially the case with the type of universals specified in UG,
which are strongly determined by the TG formalism (as opposed to universals
of the type specified in the Greenberg 1966 tradition, which are much more
theory independent) which has itself proven to be suspect, and which are
moreover derived in a dubious way on the bases of a limited number of
languages and without taking into account variability (cf. 2.2.2.2). Second, even
apart from that, it is impossible to tell straight away what the role of universals
in the acquisition process is, how they came into existence and how they
developed. Just claiming that they are innate is without any foundation. (Cf. also
Catlin 1978.) They could have emerged as well through an interaction, mutually
and with environmental factors (which probably also obey to universal principles
- cf. 1.2), of more profound innate principles. There is no reason whatsoever to
believe that the relationship between the genetic basis and the adult systems is
a simple one. Again, one should not expect the cognitive world to be different
from the physical world in this respect. (Cf. 2.3.2.)
Of course, the latter seems to be a suggestion in the direction of a view
which Chomsky explicitly rejects, viz. what he would term an 'empiricist' lan-
guage theory.30
The suggestion that we seek less and less specific theories of innate endowment
that subsume linguistic universals merely expresses [...] the conventional belief that
the language faculty has no special properties, that there is simply a generalized
learning capacity. (Chomsky 1975a: 208)
Consider again the question whether cognitive functions are both diverse, and
determined in considerable detail by a rich innate endowment. If the answer is
positive, for some organism, that organism is fortunate indeed. It can then live in
a rich and complex world of understanding shared with others similarly endowed,
extending far beyond limited and varying experience. Were it not for this
endowment, individuals would grow into mental amoeboids, unlike one another,
each merely reflecting the limited and impoverished environment in which he or
she develops, lacking entirely the finely articulated, diverse and refined cognitive
organs that make possible the rich and creative mental life that is characteristic of
all individuals not seriously impaired by individual or social pathology. (Chomsky
1980a: 45-46)
(Cf. also Chomsky 1965: 54, 1980b, c, d.) However, this is once again turning
matters upside down. There is no doubt about the fact that a cognitive faculty
like language does have specific properties. That is an empirical matter. But this
does not imply, in any a priori way, a specific conception of innateness. That is
again an empirical matter, to be settled by research on the acquisition process.
In fact, Chomsky's aprioristic claim about the innateness of universals is
methodologically paralyzing, for it bars any real attempt to look for a develop
mental explanation of adult language and the universals in it (cf. Clark and Clark
1977, Cellrier 1980a, b, Schlesinger 1982). Such research could lead to the
conclusion that Chomsky is right in assuming a very strict innateness, but it could
also lead to any theory between this strong innatist view and a very weak innatist
view. Chomsky (1978) reproaches Piaget (1980a - originally published in 1975)
and Luria (1974) that they are dogmatic in rejecting his strong innatist view of
language.31 But their claims are at least based on serious empirical research
into the problem. It is rather dogmatic to reject this fact. This is another illustra
tion of the 'eye-flaps' mentioned in 2.2.2.2.
Thus, the type of argument often used by Chomsky, that a linguistic prin
ciple such as the 'structure-dependence' principle e.g. (assuming that it is
adequate, for the sake of the argument) could never be accounted for on the
basis of learning, and therefore makes any type of general learning theory a
priori invalid (e.g. Piattelli-Palmarini (ed.) 1980), does not hold water. Chomsky
has never investigated whether a child is indeed (un)able to learn such a
principle, hence his claim is ad hoc. If a linguist is able to find out about such
a principle, why should one a priori assume that a child is not? (See also Papert
1980, Putnam 1980a.) In fact, 'learnability' is a criterion which seems to play a
rather important role in deciding whether elements of a TG of an individual
language can be included in UG (cf. e.g. Chomsky 1981a, Lasnik 1981, Wexler
1981, Williams 1981), but since this criterion has no roots whatsoever in any type
of investigation of what is learnable for a child, this only adds to the suspicious
nature of UG already noted in 2.2.2.
Of course, Chomsky again claims that so far no serious alternatives have
been developed for his views. Maybe he is right: nobody has been daring enough
to propose a full-fledged model of the innate principles of language the way he
did this (though there is a massive body of research which does point in certain
directions - cf. 2.3.2). But that is no argument whatsoever for his views, and in
fact it is not blaming research on language development either. Given our
current knowledge and the complexity of the problem, we have no doubt still a
long way to go before we reach a point where a serious theory of the innate
principles of human cognition becomes feasible at all. As Putnam says:
Chomsky feels that the 'prospects' of 'general learning theory' are bad. I tend to
agree. I see no reason to think that the detailed functioning of the human mind,
will ever be transparent to the human mind. But the existence of general
intelligence is one question; the prospect for a revealing description of it is
another.
Incidentally, if the innateness hypothesis is right, I am also not optimistic
about the prospects for a revealing description of the innate template of language.
The examples Chomsky has given us of how to go about inferring the structure of
the template [...] are such bad arguments that they cast serious doubt on the
feasibility of the whole program, at least at this point in history (especially if there
exist both general intelligence and an innate template).

On the other hand, we may well be able to discover interesting facts and
laws about general intelligence without being able to describe it completely, or to
model it by, say, a computer program. There may be progress in studying general
intelligence without its being the case that we ever succeed in writing down a
'general learning theory' in the sense of a mathematical model of multipurpose
learning. (Putnam 1980a: 297)
The future will show whether we have to accept Putnam's pessimism.

All of this means that LAD cannot be used as an argument for the internal
descriptive adequacy of TG, since it is not independently motivated (cf. 2.1). The
relationship between TG and LAD is inherently circular. It is interesting to see
how Chomsky has tried to escape this circularity with respect to the simplicity
criterion. It is obviously strange to suppose that a child possesses a simplicity
criterion with which it can choose among alternative representations of the
language it is learning (cf. also Parret 1974, McCawley 1977, 1980), but, as
indicated in 2.2.3.1, it is reasonable to assume that simplicity does play some role
in the acquisition of language. Of course, empirical research again has to provide
evidence for this. Now, Chomsky has claimed that choice of an evaluation
measure is an empirical matter in his theory too (cf. also Lasnik 1981).
Given primarily linguistic data D, different choices of an evaluation measure will

assign quite different ranks to alternative hypotheses (alternative grammars) as to
the language of which D is a sample, and will therefore lead to entirely different
predictions as to how a person who learns a language on the basis of D will inter
pret new sentences not in D. Consequently, choice of an evaluation measure is an
empirical matter, and particular proposals are correct or incorrect. (Chomsky 1965:
37)
In other words, given that we know what UG looks like and, hence, which is the
variety of possible TGs for a given set of input data (sample D of language L),
we can investigate which simplicity criterion selects the appropriate TG from the
alternatives by checking whether the grammar chosen makes correct predictions
about data from L which were not present in the initial set D. Thus, the
'empirical' choice of the criterion depends on whether we have found the correct
TG on empirical grounds. This brings us in the peculiar position that the linguist
needs a simplicity criterion to choose between different grammars which remain
possible after having investigated an unavoidably limited sample of data (cf. the
problem of underdeterminedness of grammars - 1.1.2), while the simplicity
criterion has to be chosen on the basis of knowing which TG is appropriate by
taking into account more data.
Of course, Chomsky would reply that it is not the linguist but the child who
postulates TGs on the basis of sample D, and selects between them on the basis
of a simplicity criterion. The linguist's task is to detect the child's criterion, and
this can be done in the way described. It is probably in this vain that one should
see Chomsky and Katz's (1974: 351-352) claim that the simplicity criterion in
LAD should not be confused with the simplicity criterion used by linguists for
the selection of an explanatorily adequate competence description. They say that
the latter is a methodological criterion while the former is a scientific hypothesis
which has to be judged according to the normal methodological criteria which
apply to any scientific hypothesis (including a simplicity criterion?!). This seems
to be a departure from the original assumption, though: Chomsky's claim that
the linguist's and the child's problem in getting to know the grammar of a
language is basically the same does seem to imply that their simplicity criteria
are the same as well. Anyway, this does not help, for this claim about the
empirical nature of finding the simplicity criterion in LAD assumes that it is
evident to the linguist which TG the child ultimately develops, and that it is
evident that the LAD-view of acquisition is appropriate and that one has a
plausible model of UG. Neither of the two is the case, though. The fact that the
simplicity criterion has disappeared from recent versions of LAD might be an
indication that Chomsky has finally accepted the problematic nature of at least
this part of his theory.
2.3.2. Ontogenesis from a psychological and biological perspective
Let's see what empirical research on language acquisition has to tell us about
Chomsky's claims. Originally, Chomsky's views had quite some impact on
developmental psychology (McNeill 1966 having been one of their main
proponents), and even today there are still attempts to apply them (especially
the parameter version of LAD - e.g. Hyams 1986). However, the bulk of
acquisition research (including McNeill 1979) has soon abandoned the
assumptions of Chomskyan LAD on the grounds that they are very implausible
from an empirical point of view. The overall tendency is to take a clearly
functionalist perspective: it appears impossible to understand the systematicity
at the different levels of language development, and in the pattern of develop
ment from one level into another, without taking into account the embedding of
child language in a general picture of the child as an acting being. (Cf. e.g. Van
der Geest 1974, 1975, Bruner 1974, Clark and Clark 1977, Bates 1978, 1979,
Bates and MacWhinney 1979, Schlesinger 1982, Ninio and Snow 1988.)
The process of acquiring language is deeply affected by the process of becoming

a competent member of a society. [...At the same time the] process of becoming
a competent member of society is realized to a large extent through language,
through acquiring knowledge of its functions, its social distribution and interpreta
tions in and across socially defined situations i.e., through exchanges of language
in particular social situations. (Ochs and Schieffelin 1984: 277)
More specifically with respect to Chomsky's claim that the acquisition

process is incredibly fast and is accomplished easily despite bad conditions (the
deficient input data), acquisition research has led to different assumptions. The
speed of the process is relative if one considers that it takes at least 5 years for
a child to become fluent in language, and that it does not master its subtleties
before the ages of 10 to 14. If one moreover takes into account that these 10 to
14 years are a permanent exercise (mainly unconscious, but in part also
conscious, e.g. at school) in language use, Clark and Clark (1977: 373) seem
right when claiming that "acquiring a language is a long and complicated
process". (Cf. also Fraser's discussion in McNeill 1966, Putnam 1971, Morton
1971, and many others.) If this process does proceed successfully in most cases,
this is no doubt due to the fact that the input data are normally not deficient or
accidental at all. That is what appears from the massive amount of research on
adult-child interaction (e.g. Levelt 1975, Snow 1975a, Clark and Clark 1977,
Gelman and Shatz 1977, Snow and Ferguson (eds.) 1977, Waterson and Snow
(eds.) 1978, Wells and Robinson 1982, Allen and Shatz 1983, Shatz 1984, Bruner
1985).
First of all, there is the phenomenon called 'baby-talk' or 'motherese'.
Maybe this can be called deficient input from the perspective of adult language,
since it does not involve full-fledged utterances as they occur in adult-adult
conversation. It is syntactically and lexically highly selective and restricted, but
it does appear to be internally consistent and fully adapted to the level of the
linguistic capacity the adult detects in the child's performance. Thus, it does
change and becomes more and more complex depending on the growing
capacities of the child. This suggests that the child's language input consists of
a whole series of increasingly complex sub-languages, which are adapted to and
stimulate the development of progressively more complex cognitive language
systems in the child, until the adult system has been reached.
In interpreting the significance of this phenomenon one has to be careful.
Recent research (e.g. Ochs 1983) has shown that baby-talk as we know it may
not be a universal phenomenon, but may be culture-specific and depending on
the Western view of the child (psychologically) and the Western family form
(sociologically). In some cultures mother-child interaction appears to be very
indirect up to a certain age of the child, in that older children take over the role
of the mother and mediate between mother and child. Of course, one question
is whether the caretaker does not use baby-talk as well. Even four-year-old
children appear to be able to use it (cf. Gelman and Shatz 1977). Another
question is whether this different situation does not affect the language
development of the child. This requires further investigation. But it is beyond
doubt that the nature of the adult behavior toward the child does affect its
acquisition process and the 'final' product, not only in the trivial sense that it
acquires the language spoken in its environment, but also in the sense that styles
of language use toward the child seem to affect the course it takes in its devel
opment (e.g. Lieven 1989), and appear to have lasting results on the (former)
child's adult behavior. The latter is also apparent from sociolinguistic research,
which has shown that the totally different nature of linguistic input in higher vs
lower social classes has a clear influence on the course of the child's develop
ment and on the language variant it ultimately acquires (e.g. Bernstein 1971).
The input language (baby-talk, and presumably also other types) is accura
te also from a completely different perspective.
We too readily overlook the fact [...] that speech makes its ontogenetic progress
in highly familiar contexts that have already been well conventionalized by the
infant and his mother. (Bruner 1974: 261)
(cf. also Bruner 1978). Hence the child's linguistic input
provides him with instances which are from his point of view not nearly as
fragmentary and disordered as we are sometimes invited to think. Most speech is
functionally highly coherent, in the sense that its structure reflects what it is being
made to do, and this is what [...] makes it accessible to the learner. (Halliday
1970a: 323)
In adult-child conversation, adults normally make optimal use of the communica

tive situation by heavily relying upon the direct non-verbal environment (e.g.
through deixis) and by simultaneously using non-verbal means of interaction,
which the child appears to master before language. (Cf. e.g. Campbell and
Wales 1970, Bates et al. 1977, Bates 1979.)
Of course, Chomsky rejects the relevance of these observations (though he
does not question their correctness!) for developing a theory on language ac
quisition, in line with his concept of autonomous syntactic competence in adult
cognition. With respect to the functionality of language, he claims that the
tendency to act purposefully may initiate the acquisition process, but that this
does not mean that it also influences the process of acquiring syntax in particular
(e.g. Chomsky 1962, 1965: 33). One can only reply to this with the remark that
this is another case of an aprioristic rejection of a potentially relevant correlation
(the 'eye-flaps' - 2.2.2.2) which is very likely to exist. In fact, Levelt (1973c, 1975)
has observed that, formal-logically, the problem of acquiring syntax on a purely
formal basis appears unresolvable. According to him, research into formal
inferencing mechanisms for TGs has shown that if one takes into account the
real input-characteristics of language acquisition, a purely syntactic system of
language is not learnable. Hence, Chomsky's LAD simply cannot be formulated.
By real input-characteristics he means that acquisition does not happen through
'informant presentation' but through 'text presentation' of the language, i.e. the
input is not qualified as being grammatical or ungrammatical. In fact, component
(ii) of Chomsky's (1965: 30) scheme of LAD (see the quotation in 2.3.1) implies
that a child does have a means of knowing whether the utterances it encounters
are grammatical or not in the language it is acquiring. Since it does not know the
language yet, this must be due to the input proper.
Let us, in any event, assume tentatively that the primary linguistic data consist of
signals classified as sentences and nonsentences, and a partial and tentative pairing
of signals with structural descriptions. (Chomsky 1965: 32)
This is obviously not a 'real input-characteristic'. According to Levelt, under real

conditions, one cannot but accept that semantic and pragmatic factors do play
a decisive role in acquiring language, including the principles of its syntactic
organization.
Concerning the role of non-verbal interaction systems in language acquisi
tion, Chomsky claims that only
If it were possible to show that 'prelinguistic symbolic systems' share certain

significant properties with natural language, we could then argue that these
properties of natural language are somehow acquired by 'analogy', though we
would now face the problem of explaining how the 'prelinguistic symbolic systems'
developed these properties and how the analogies are established. (Chomsky
1969a: 69)
However, since for Chomsky these properties can only involve syntax, and since
natural language syntax seems unique, it is obvious what this answer means. Yet
as Goodman put it,
Chomsky argues that if the symbol systems we acquire prior to the acquisition of
a full-fledged language have the characteristics he finds to be essential for a langu
age, then acquisition of the prior systems would be equally remarkable and in need
of explanation, while if the prior systems do not have these features, they can be
of little help in acquiring a language. The latter part of this argument seems to me
quite specious. If a man has made a clock by hand, that is a remarkable accomp
lishment. I understand it better if he tells me how he first fashioned crude tools,
used these to make more refined ones, and eventually made the clock. I do not
protest that he could not have made the clock without first having made a
timekeeping instrument. The tools used in making the clock are not timekeepers,
and from the clock alone we cannot infer back to any specific characterization of
the tools made and used at any stage along the way. (Goodman 1969: 139ff)
Another, closely related element which has strongly been brought to the
fore in developmental psychology is the intimate relationship between language
acquisition and the general cognitive development of the child. No matter whe
ther one follows Piaget's (1923; Piaget and Inhelder 1972) or Wygotsky's (1974)
views, in either tradition language development is understood as a process which
is strongly interwoven with, and in many respects depends upon, other aspects
of the child's cognitive capacities, including its capacity to conceptualize the
world. (Cf. e.g. Bates 1979, Clark 1983, Gillis 1983, 1984, Shatz 1984.) These
observations cast serious doubt on Chomsky's claim that language acquisition is
an autonomous process in which there is no structural contribution from other
developments, and that it is only based on specifically linguistic innate principles.
(Attempts to relate syntactic principles to more general cognitive principles have
been made in Lyons and Wales (eds.) 1966, Hayes (ed.) 1970, Flores d'Arcais
and Levelt (eds.) 1970, Morton (ed.) 1971, and specifically in Bever 1970a, b,
1971, 1976, 1982.)
We are again facing Chomsky's deliberately accepted eye-flaps in refusing
to be open-minded toward the possibility of finding correlations in this respect.
There is no a priori way of excluding such correlations, and the fact that one
looks for them certainly does not imply that one denies the specific properties
of linguistic knowledge. If one finds correlations, this does not mean that one
can reduce the linguistic principles to the more general cognitive principles.
Again, comparison with the physical world is useful here. As Chomsky claims,
human cognitive systems, when seriously investigated, prove to be no less

marvelous and intricate than the physical structures that develop in the life of the
organism. Why, then, should we not study the acquisition of a cognitive structure
such as language more or less as we study some complex bodily organ? (Chomsky
1975a: 10)
In fact, no biologist hesitates to try to find principles of organization and

development which are common to the physical constitution of living beings in
general, even though there can be no doubt about the uniqueness of each organ
in an organism.
Let's consider the problem of innateness in language from the perspective
of some of the views inherent in biology, ethology, and neurophysiology. Though
at our current state of knowledge this is not likely to provide us with firm
arguments for one or another concrete model, it does clearly point to a certain
direction for further thinking.
At first sight Chomsky's views might seem to find approval, e.g. by Monod.
On sait que, selon Chomsky et son cole, sous l'extrme diversit des langues
humaines, l'analyse linguistique en profondeur rvle une 'forme' commune
toutes ces langues. Cette forme doit donc, d'aprs Chomsky, tre considre
comme inne et caractristique de l'espce. Cette conception a scandalis certains
philosophes ou anthropologistes qui y voient un retour la mtaphysique
cartsienne. A condition d'en accepter le contenu biologique implicite, cette
conception ne me choque nullement. (Monod 1971: 150)
Yet, the question is whether he was fully aware of the nature of Chomsky's
innateness theory when writing this, for the views he expresses elsewhere in the
very same book are hardly compatible with Chomsky's view (the fact that the
title of Lightfoot's 1982 book alludes to a statement by Monod 1971 notwith
standing).
The concept of innateness with relation to cognitive systems is uncontro-
versial. Everybody has to accept some form of innateness, and even the beha-
viorists did (cf. Quine 1969, Putnam 1971, Piattelli-Palmarini (ed.) 1980). What
has shocked many people is the extreme degree of innateness postulated by
Chomsky. This was clearly a reaction to the behaviorist tabula rasa view of
acquisition. The failure of behaviorism certainly has shown that it is necessary
to assume much richer internal mechanisms behind language behavior and
acquisition, and that a view of acquisition as the passive obedience to con
ditioning factors from the environment is untenable. Thus, it is reasonable to
assume that the genetic basis of language contains more than just some 'general
learning capacity': the latter may be required for the acquisition of the capacity
to play chess, e.g., but the acquisition of language is qualitatively different (cf.
2.2.3.3). But there still are a wide variety of positions possible before one
reaches Chomsky's (contrary to the black-and-white situation which Chomsky
always suggests). The fact that conditioning does not provide a viable explana
tion does not mean that one should develop allergic reactions to any notion of
'learning' (cf. Lehrman 1970, Whitaker 1973). This has to be kept clearly in mind
with respect to the genetic determination of cognitive or behavioral systems:
This point is very well taken. It seems unlikely that genes actually transmit
behavior as we observe it in the living animal because the course that an individual
takes in its peregrinations through life must necessarily depend on environmental
contingencies which could not have been 'programmed and prepared for' in
advance. Inheritance must confine itself to propensities, to dormant potentialities
that await actualization by extra-organic stimuli, but it is possible that innate facili-
tatory or inhibitory factors are genetically transmitted which heighten the likeli
hood of one course of events over another. When put into these terms, it becomes
quite clear that nature-nurture cannot be a dichotomy of factors but only an inter-
action of factors. To think of these terms as incompatible opposites only obscures
the interesting aspects of the origin of behavior. (Lenneberg 1967: 22)
(Cf. also Lehrman 1970, Whitaker 1973, Dingwall 1975.) Thus, "there is no need
to assume 'genes for language"' (Lenneberg 1967: 265; cf. also Lenneberg
1969.)32
A main problem in the matter of the genetic basis of cognitive systems
concerns the question of how they relate to physical (biological) structures. The
latter are obviously crucial in the analysis of the genetic determination of the
former, but still, one should clearly differentiate between the two. To see this
point, bear in mind the argumentation in 1.1.2 concerning the relatedness yet
irreducibility of the mental and the physical world in humans. In the words of
H.Jackson, one of the pioneers of neurolinguistics:
speech and words are psychical terms; words have, of course, anatomical substrata
or bases as other psychical states have. We must as carefully distinguish betwixt
words and their physical bases as we do betwixt colour and its physical basis; a
psychical state is always accompanied by a physical state, but nevertheless the two
things have distinct natures. (Jackson 1878, quoted in Gurney 1973: 58)
Thus, we must assume that the cognitive systems involved in (linguistic) behavior
are somehow coded in the neurological structures of the brain, and that
cognitive activities involved in behavior are correlated to neurophysiological
activities, although we do not know how. But if cognitive systems are 'coded' in
neurological systems, this obviously implies that they are not identical. A similar
argument is provided by Kenny (1981), who also compares the matter with the
software-hardware distinction in computers: they are not mutually reducible
either, although they are inextricably linked.33 The point is further supported by
the observation that in case of brain damage, after an initial fall-out of certain
linguistic capacities a strong rehabilitation of the lost functions often occurs,34
and by the observation that very similar brain lesions can lead to quite different
patterns of behavioral disturbance. Apparently, there is quite some variability in
the correlation of specific cognitive functions and specific physiological locations.
(Cf. e.g. Smith 1978, Zangwill 1978.) If neurological and cognitive systems really
were identical, such facts could not possibly occur. Note also that it is apparently
possible to simulate to some degree limited cognitive functions in a comput-
er.35 Although some kind of 'hardware' is indispensable, cognitive systems
apparently do not even need neurological structures to exist (cf. also Miller
1978a). (From a completely different perspective, a similar point is made by
Maturana 1978.)
In view of this, even if one assumes that the biological correlates of
behavioral systems, like all biological systems, are 'strongly' genetically
determined (a matter I will not discuss here), this does not imply that a similar
degree of genetic determination has to be assumed for the cognitive systems
related to them. Chomsky's argument that
there seems little reason to insist that the brain is unique in the biological world
in that it is unstructured and undifferentiated, developing on the basis of uniform
principles of growth or learning - say those of some learning theory, or of some
yet-to-be conceived general multipurpose learning strategy - that are common to
all domains. (Chomsky 1980a: 39)
does not hold water. The question of the genetic determination of the brain has
no direct bearing on the question of the genetic determination of the cognitive
systems coded in it. In fact, the deficiency of Chomsky's argument already
surfaces in the fact that it is strange to talk about a brain's development in terms
of 'learning'.
Maybe the genetic determination of cognitive systems could be conceived
as follows. The capacity to develop cognitive systems is based on the possibilities
offered by the physical system 'carrying' them, i.e. the brain. This system and its
possibilities are genetically determined in the way all other biological systems
are. These possibilities do not say anything directly about the cognitive systems
which are actually acquired, however. To use the computer metaphor again: the
hardware limits the possible software of a machine, but it still leaves room for
differences in the software (unless one develops a completely deterministic
machine) and therefore does not say anything directly about the software. The
cognitive systems themselves are determined by, or construed according to, the
'experiences' of the organism in the environments in which it occurs during its
development, i.e. they are due to 'nurture'.
One might reply that this is not typical for cognitive systems. The pheno-
type of a physical system is determined by the physical environment in which it
develops as well, and therefore the genotype does not say anything directly about
the phenotype of it either (cf. Lenneberg 1967: 239ff). This environmental
determination seems to be of a completely different nature, however. It does not
involve filling in open possibilities or encoding 'information' into the system (cf.
also Changeux 1980). That is why it is absurd to talk about a heart, or kidneys,
or a brain 'learning' from environmental influences.
How should we conceive of the process of developing cognitive systems in
a neurological structure in more specific terms, then? In ethology, a distinction
is made between two main types of genetic programs for behavioral systems in
animals.
We can ask what differences exist between genetic programs responsible for
behaviors formerly called innate and those considered as experientally acquired.
A genetic program which does not allow appreciable modifications during the
process of translation into the phenotype I call a closed program - closed because
nothing can be inserted in it through experience. Such closed programs are
widespread among the so called lower animals. A genetic program which allows
for additional input during lifespan of its owner I call an open program. (Mayr
1974: 651)
(See also Mayr 1976.) This should not be misunderstood: a closed program does
not have to be genetically more specific than an open program. Rather, this
involves the matter of the specificity of the resulting neurological systems which
'accept' cognitive systems.36 As Mayr states,
Even this improvement in terminological precision does not remove all our
difficulties. A particular instinctive behavior act is, of course, never controlled
directly by the genotype but rather by a behavior program in the nervous system
which resulted from the translation of the original genetic program. It is
particularly important to make this distinction for the open program. The new
information acquired through experience is inserted into the translated program
in the nervous system rather than into the genetic program because, as we know,
there is no inheritance of acquired characters. (Mayr 1974: 651)
In view of this, genetic specification involves at least the global develop

mental pattern of the central nervous system, and in each phase of the
development a certain degree of 'plasticity' in its different areas, each of which
has a certain specialization determining the types of mental systems that can be
encoded in them (the latter is necessarily vague for lack of better knowledge, but
Lamendella's 1977b types of neurofunctional systems may somewhat illuminate
what is involved). The open-closed distinction should be seen as a scale for that
matter. Programs at the closed end of the scale determine neurological systems
which have nearly no plasticity. The neurological systems developing from them
allow nearly no environmental influence, and which behavioral systems can occur
in them is entirely a priori fixed. Whether the latter are 'implemented' and
become active depends on whether the external circumstances which can trigger
them actually occur. Thus, also systems based on closed programs require
experience of stimuli, but only very specific types of experience have an effect,
which is entirely preconceived. 37 Normally this involves very simple and
stereotypical behavioral patterns such as reflexes (illustrations in Mayr 1974,
1976, Dingwall 1975).
Programs at the open end of the scale cause neurological systems which,
at least during specific, 'critical' periods, are plastic, and therefore sensitive to
large-scale experiencing (or learning, though not in the sense of conditioning or
associative learning - cf. Lehrman 1970, Whitaker 1973: 99ff). In these
neurological systems specific information is introduced basically through what
ethologists call imprinting or Trgung'. A very simple case often used to
illustrate this principle involves birds such as geese, which accept as a parent or
partner the first moving object producing noise they encounter after birth. This
can be another goose, but also a human, or a mechanical goose, etc. Another,
somewhat more sophisticated case is the acquisition of types of songs and
dialects in them by birds. (For illustrations, see Lehrman 1970, Whitaker 1973,
Mayr 1974, 1976.) As soon as the critical period ends the plasticity is lost,
though, and filling in further information through experience is no longer
possible. If birds which normally 'learn' their song are separated from their
environment from birth until the end of their developmental period, they never
acquire their species-typical song, and are therefore unrecognizable to their
congeners. Of course, in many types of organisms, even after the critical periods
for all neurological systems in them have ended a certain degree of associative
learning remains possible, on the basis of specific neurological systems assuming
the role of 'general purpose memory' (see Whitaker 1973, Lamendella 1977b).
What plasticity involves at the neurological level, and what the neurological
correlate of experiencing is, remains an open question. Some suggestions can be
found in Lenneberg (1967: 158ff), Lindsey and Norman (1972), Gurney (1973),
Whitaker (1973), Jacobson (1975), Changeux (1980).38 In any case, it is gene
rally assumed that open programs offer strong advantages to an organism from
the viewpoint of natural selection, because they allow a much wider and deeper
adaptation of the organism to the life conditions existing in its direct envi
ronment. Chomsky's (1980a: 45-46; see quotation in 2.3.1) claim that if there
would not be rich innate mechanisms determining cognitive systems 'individuals
would grow into mental amoeboids' has only rhetorical value. The issue is not
whether there are rich innate structures, but what is the relationship between
nature and nurture in terms of their contribution to behavior. Nature prohibits
development into an amoeboid, but nurture prohibits an organism developing
into an inflexible piece of rock.
Turning to the matter of the genetic determination of language proper,
there are many reasons to believe that it is based on an open program (or rather
programs) and that the role of experience in its development is considerable.
Relevant to this matter is the distinction between 'precocial' and 'altricial'
animals.
Precocial animals are those which reach maturity in specified motor and sensory
systems shortly after birth; altricial animals have more or less extended periods of
immaturity after birth, during which these motor and sensory systems develop to
their adult values. (Whitaker 1973: 98)
It seems logical to correlate the degree of altriciality of animals to the degree of

plasticity of the neurological systems, and hence to the sensitivity to experience
during critical periods, both in degree per system and in number of neurological
domains.
Now, humans appear to be the most altricial animals to date.
The human neonate is considerably more immature at birth than our closest of
kin, with a concomitant prolongation of differentiation periods and increased
susceptibility for various factors to impinge upon the direction of further
development. The acquisition of language plays a definite part in this developmen
tal history, its emergence occupying a fixed position within the array of develop
mental milestones, and there are definite indications that its development is
contingent upon a certain aspect of what might be called (for lack of more
accurate knowledge) cerebral plasticity. (Lenneberg 1967: 244-245)
Between the ages of 1;6 and 14 (roughly) humans appear to have a critical
period for the acquisition of language, which involves a considerable flexibility
in what is acquired. This phenomenon of a critical period for language
acquisition has especially been noticed in neurolinguistics. One should be careful
with it, because its precise impact and its nature in neurophysiological terms is
not yet clear. It seems to be related to the phenomenon of lateralization, but it
is still unknown in which way. Nevertheless, there are sufficient observations
showing its existence and impact on cognitive functioning (cf. also Lenneberg
1967, Lamendella 1977b), including the following.
(i) After the critical period a human seems no longer able to acquire a language
completely and perfectly. This is not only the case for a 'foreign' language, but
also for the native language, as has, among others, been shown by a number of
cases of 'wolf children', or also by Genie (Curtiss 1977).
(ii) In general, in case of child aphasia the disturbed functions are largely or
often even entirely recovered, for example by hemispherical 'division of labor'.
In adults however, recovery is normally much more partial (although mild
disturbances can be largely recovered). The transition from one to the other
situation seems to coincide with the fading away of the critical period.
(iii) Retarded children run through a more or less normal acquisition process but
at a much slower pace than normal children. Yet what retarded children have
not acquired by the end of the critical period they never acquire.
In view of these observations, there is much reason to believe that language
acquisition is based on a very extensive process which holds some relationship

to 'imprinting' in lower species (though one has to remain careful with this
comparison - cf. e.g. Lenneberg 1967: 175ff, Whitaker 1973).
One can also wonder whether the critical period is uniquely linguistic. This
relates to the question whether there are physiological systems which are
sensitive to specifically linguistic input, and hence whether there is room left for
specifically linguistic innateness, albeit in an indirect way. Lenneberg (1967) does
postulate a uniquely linguistic critical period. However, some facts point in the
opposite direction. In the same period a number of other (e.g. conceptual)
capacities develop, and there is no indication that language develops indepen
dently of them. On the contrary, the phenomenon of lateralization, related to the
critical period for language, also appears to involve a series of other cognitive
functions, including manual performance, i.e. performing goal-directed actions
through the hands (cf. the dominance of the left or right hand in this). (Cf. e.g.
Luria 1973, Clark and Clark 1977, Lamendella 1977b.) In fact, Lenneberg (1971)
did change his mind, accepting that linguistic capacity must be grounded in a
more general cognitive capacity characteristic for Homo Sapiens. Even though
the evidence is only suggestive, as Monod states,
De l penser que si, chez l'enfant, l'acquisition du langage parat aussi miracu
leusement spontane c'est qu'il s'inscrit dans la trame mme d'un dveloppement
pigntique dont l'une des fonctions est de l'accueillir, il n'y a qu'un pas que, pour
ma part, je n'hsite pas franchir. Pour tenter d'tre un peu plus prcis: de cette
croissance post-natale du cortex dpend sans aucun doute le dveloppement de la
fonction cognitive elle mme. C'est l'acquisition du langage au cours mme de
cette pignse qui permettrait de l'associer la fonction cognitive et cela de faon
si intime qu'il est trs difficile pour nous de dissocier, par l'introspection, la
performance linguistique de la connaissance qu'elle explicite.
On admet en gnral que le langage ne constitue qu'une 'superstructure',
ce qu'il parat, bien entendu, par l'extrme diversit des langues humaines, produits
de la deuxime volution, celle de la culture. Cependant l'ampleur et le raffine
ment chez Homo sapiens des fonctions cognitives ne trouvent, de toute vidence,
leur raison d'tre que dans et par le langage. Prives de cet instrument, elles sont,
pour la plus grande part, inutilisables, paralyses. En ce sens, la capacit
linguistique ne peut plus tre considre comme une superstructure. Il faut
admettre qu'entre les fonctions cognitives et le langage symbolique qu'elles
appellent et par quoi elles s'explicitent, il y a chez l'homme moderne une troite
symbiose, qui ne peut tre le produit que d'une longue volution commune.
(Monod 1971: 149-150)
Finally, it seems possible to postulate the following principles for the

ontogenesis of language. What is genetically programmed is, at least, the
neurological infrastructure offering the possibility to acquire the linguistic system,
and moreover the succession of phases of neurological development and in each

phase a certain degree of neurological plasticity in some or all of the subdomains
involved. At each level in this development cognitive systems are constructed on
the basis of experience, depending on the specific possibilities offered by the
neurological systems and on their degree of plasticity. It is very likely that
different cognitive, including linguistic, sub-functions are due to neurological
systems with different degrees of plasticity, which means that their genetic
determination is variable. Both in its global course and in each phase, cognitive
development is an integrated development in which the linguistic system as it
appears in the adult gradually takes shape in integration with other cognitive
systems, such as the conceptual system and other non-verbal communication
systems. These principles also basically underlie the model of Lamendella
(1977a, b).
Universals in child and adult language can then be handled in the following
terms. As Greenberg says,
functional requirements provide bounds [...] outside which language risks a break
down of communication [...]. As different as languages are, [...] linguistic structure
realizes but an infinitesimal portion of the logical possibilities of [...] sign systems.
(Greenberg 1958: 85)
[But] such similarities are a reflection of the basic biological unity of man as a
behavioral animal and the fundamental similarity of the functional task that a
communicative system must perform in any society. (Greenberg 1963: 162)
Toulmin (1971, 1972) illustrates this principle by means of the form of honey
combs of bees. Normally, the cells of the combs have a hexagonal form. Does
this mean that this form is directly represented in the genetic material of bees?
No way. What certainly is genetically determined is the capacity to build combs.
But their hexagonal form can be explained on functional principles. Physically,
this form is very economic: it allows the largest possible number of cells within
a given space. In a cloud of soap-bubbles each separate bubble also takes this
form (three-dimensionally) under the purely physical conditions of air-pressure.
Hence, if bees have limited space the form of the cells is hexagonal on the same
grounds. (Further illustrations of this principle can be found in Thompson 1917;
see also Bates 1978.) The same principle can be applied to linguistic universals:
we [...] no longer need to explain the universal forms of linguistic behaviour as

reflecting corresponding innate forms in their underlying capacities and endow
ments. For those universal forms might simply reflect the universality of the
'objective functions' which language is everywhere used to perform. As with
honey-bees, so with men: the universal forms in actual behaviour would be
intelligible, only by considering how generalized 'innate propensities' find

expression as applied to the specific problems of their actual lives. (Toulmin 1972:
465)
(See also Whitaker 1973.)

A typically Chomskyan counterargument for the role of functional conside
rations in the acquisition process, such as the observation that "No one supposes
that children learn to have an eye capable of sight because it would be useful for
this function to be fulfilled." (Chomsky 1980a: 229) obviously does not hold
water, for at least two reasons. First, because it misrepresents the role of
functional considerations with respect to biological organs: it is fully accepted
that functionality influences the development of organs, yet of course not in the
way suggested by Chomsky. Second, because this comparison is not valid given
the entirely different nature of the way cognitive development is determined.
Actually, there is no conflict between the supposition that functionality is
an important determining factor for cognitive systems and the supposition that
cognitive systems can be genetically determined (of course always in the indirect
way indicated above). As for other physical organs, it is perfectly sound to
assume that the physiological correlates of originally acquired cognitive systems
gradually get incorporated in the genetic basis, in the sense that the possibilities
of the neurological systems accepting them are genetically narrowed down to
allow them only to accept those specific cognitive functions. (Cf. also Lamendella
1976, Bates 1978.)
2.3.3. On the phylogenesis of language
Finally, let's briefly consider the matter of the phylogenesis of language. Our
knowledge in this area is very poor, but again it contains sufficient indications
to add to our argumentation against the Chomskyan concept of language.
As Toulmin (1971, 1972) states, one can roughly distinguish three possible
theories about the origins of language.
(i) Language could have resulted from a sudden genetic change caused by a very
drastic mutation. The consequences were so positive that the mutation soon
became dominant in the population, and eliminated all competing populations.
(ii) One could distinguish between the physiological prerequisites for language
and language itself. The former could have developed gradually in hominids, yet
without something language-like occurring in behavior. Until language was
invented, all of a sudden, by 'cultural mutation'.
(iii) Linguistic capacity could have developed through a gradual accumulation of
physiological and behavioral properties which were advantageous for pre- and
non-linguistic reasons. The progressive physiological changes which occurred this
way were accompanied by a gradual expansion of the sub-functions of language,
ending in the system we have at present.
(i) and (ii) do not, while (iii) does imply a role for precursors for language.
As Toulmin states, in view of the generally accepted (neo-)Darwinian theo-
ry of evolution only (iii) is really plausible as a theory. 39 In fact,
it strains credulity to pretend that language as we know it suddenly sprang up

intact as a cultural invention in the absence of extensive cognitive and communica
tive preadaptations. (Lamendella 1976: 398)
L'hypothse qui me parat la plus vraisemblable est que, apparue trs tt dans
notre ligne, la communication symbolique la plus rudimentaire, par les possibi
lits radicalement neuves qu'elle offrait, a constitu l'un de ces 'choix' initiaux qui
engagent l'avenir de l'espce en crant une pression de slection nouvelle; cette
slection devait favoriser le dveloppement de la performance linguistique
elle-mme et par consquent celle de l'organe qui la sert, le cerveau. Il y a, je
crois, en faveur de cette hypothse, d'assez puissants arguments. (Monod 1971:
145-146)
Actually, theory (iii) does not imply that mutation did not play any role in
the development of language.
Mutation was conceived as a thoroughly drastic phenomenon by De Vries and

other mutationists at the beginning of the century. Mutations, for them, had the
capacity in a single stroke to create entirely new characters, indeed entirely new
species. This concept of the drastic nature of mutation has been thoroughly
invalidated by subsequent genetic research. In the course of the last 50 years it has
been confirmed again and again that most mutations have hardly visible or entirely
invisible effects. This has made it necessary to reformulate the concept of
mutation: there is no genetic difference between drastic mutations and those that
are components of so-called continuous variation. The genetics of both types of
variation is the same. The modern evolutionist no longer sees in mutation a deus
ex machina suddenly engendering new species or other drastic evolutionary
inventions; mutation is simply the source of inexhaustible genetic variability. (Mayr
1976: 44-45)
(See also Stebbins 1982: 79, passim.) Rejecting mutationism also does not imply
that one cannot assume evolution to be a process having variable speed, with
periods of stability and periods of radical change, rather than being gradual
throughout. Contrary to what is sometimes claimed, this is perfectly compatible
with Darwinian evolutionary theory. Mutation and adaptation are not contradic
tory elements in any way. (Cf. Stebbins 1982.)
It is not immediately obvious which of the above theories would be

adopted by Chomsky. He is normally content to state that at present we can only
speculate about the phylogenesis of language, and that these speculations are
necessarily ad hoc for lack of serious evidence. Thus, he follows the 'discontinui
ty' hypothesis of Lenneberg (1967) (cf. also Marshall 1970). Yet, this seems
more of a way to avoid the problems which would appear in comparing the
Chomskyan views on language with the view on language emerging from the
most reasonable theory about its origins.
As Toulmin (1971, 1972) shows, if one takes seriously Chomsky's autonom
ist view of language relative to other cognitive (and thus also communicative)
systems, and his emphasis on the species-specificity of language, he seems to
automatically end up at the position of the early Mendelians, the 'mutationists'
as Mayr calls them. Thus, it is not astonishing to notice Chomsky-supporter
Koster (1982b) claiming that one should no longer stick to the Darwinian theory
of evolution, and that this theory is even bypassed in biology. For rectification,
he is referred to the literature on evolutionary theory, including Monod (1971),
Mayr (1976), Dobzhansky et al. (1977), Stebbins (1982), Plotkin (ed., 1982). (It
seems he should also revise his interpretation of the - albeit popularizing - work
by Gould (e.g. 1977).) Apparently, Mayr is still right when he says:
When H.J.Muller exclaimed in 1959, "One hundred years without Darwin is

enough!" he reflected the disappointment of the evolutionist, frustrated by the
seeming inability of nonbiologists to accept Darwinian evolutionism. In spite of the
worldwide Darwin jubilees in that year and an unprecedented level of evolutionary
research in the past 25 years, the situation has not greatly improved since Muller's
outcry. Perhaps it would have been too much to expect fundamentalists to abandon
their dogmas and accept the scientific evidence, but what is rather discouraging is
that in considerable areas of the European continent there is still very active
support for a number of non-Darwinian interpretations that are not supported by
any evidence. The situation is made worse when physical scientists and mathemati
cians attempt to 'prove' by calculations and computer simulations that the
Darwinian model will not work. Many philosophers, perhaps the majority,
including such a distinguished person as Karl Popper, are still holding out for an
alternative to Darwinism. Well-read, well-educated lay people like the jurist
Macbeth show in their writings how little they understand the Darwinian theory.
(Mayr 1976: 7; references omitted)
Apparently, linguists often do not constitute an exception. The reason is spelled

out by Toulmin (1972), or also by Mayr himself:
the delay is due to the nature of the Darwinian explanation, which was, and in
many ways still is, alien to the Western tradition. Shaped by Plato's essentialism,
the Western mind has tended to think in terms of unchanging types and essences
and to regard variation asfleetingand unimportant. The thinking of the dominant

schools of philosophy has throughout been incompatible with an adequate
consideration of the importance of variation. (Mayr 1976: 2)
In any case, as Monod states,
Evnement unique: les linguistes modernes ont insist sur le fait que le langage
symbolique de l'homme est absolument irrductible aux moyens de communication
trs divers (auditifs, tactiles, visuels ou autres) employs par les animaux. Attitude
justifie sans aucun doute. Mais de l affirmer que la discontinuit dans l'evolu
tion a t absolue, que le langage humain ds l'origine ne devait strictement rien,
par exemple, un systme d'appels et avertissements varis tels qu'en changent
les grands singes, cela me parat un pas difficile franchir, en tout cas une
hypothse inutile. (Monod 1971: 144)
Lenneberg (1967) maintained that a consideration of evidence and hypotheses

on the matter did not contain indications strong enough to develop one or
another continuity hypothesis of language.
The biological history of language is 'covert'; its evolution is hidden in the series
of transformations, structural and functional, that took place in the course of the
formation of modern man. It is tied to the history of physiological adaptations,
cognitive specializations, and sensory specifications. (Lenneberg 1967: 264-265)
In fact, Lenneberg's doubts concerning a great deal of the research, although

sometimes exaggerated, often was justified (cf. Dingwall 1979). There have been
a number of daring attempts to sketch the evolution of language on the basis of
far-fetched interpretations of very limited materials and these efforts can hardly
stand up to criticism (e.g. Hockett and Asher 1964). Moreover, attempts to
compare language with symbolic systems of animals totally unrelated to humans,
such as bees or birds, are unlikely to teach us anything about the phylogenesis
of language. Givn's (1979) attempt to ground his conclusions on the basis of the
communicative capacities of his dog can hardly be taken seriously. Finally, it is
true that to date, evidence which is methodologically sound is insufficient to
formulate a firm model of the evolution of language.
Nevertheless, the evidence which is available does not give any reason to
doubt the adequacy of the general theory sketched in (iii) above. The fact that
this theory is in accord with current theories in evolutionary biology is a strong
argument for it. Lenneberg's discontinuity hypothesis can only be understood as
a methodological claim (the evidence is insufficient to show on empirical grounds
what the origins of language are), and not as a theoretical claim (language has
no roots in systems lower on the evolutionary ladder). (Cf. Dingwall 1975, 1979,
Harnad et al. (eds.) 1976, Clark and Clark 1977.) The future will have to show
whether we will ever be able to develop a full-fledged model of the phylogenesis

of language. Types of research which can play an important role in this include:
investigations into the capacity of apes to learn non-verbal symbolic systems with
a syntax which tends in the direction of the multi-articulated natural language
syntax (systems which these apes have however not developed in their natural
environment), such as the famous experiments by the Hayes, the Gardners,
Premack, Fouts, etc. (cf. Linden 1976); investigations into the interaction systems
used by apes living in their natural environment; and both in combination with
investigations into the comparative physiology of apes and man (including fossils
of their precursors). (Cf. Tanner and Zihlman 1976, Menzel and Johnson 1976,
Dingwall 1979, Premack 1980.)
Of course, Chomsky once again rejects the relevance for understanding
language of research comparing symbolic systems in lower primates with
language.
Efforts to induce symbolic behavior in other species might illuminate the specific
properties of human language, just as the study of how birds fly might be
advanced, in principle, by an investigation of how people jump or fish swim. Some
might argue that more is to be expected in the latter case: after all, flying and
jumping are both forms of locomotion; both involve going up and coming down;
with diligent effort and special training people can jump higher and farther.
Perhaps some hopelessly confused observer might argue, on these grounds, that
the distinction between jumping and flying is arbitrary, a matter of degree; people
can really fly, just like birds, only less well. Analogous proposals in the case of
language seem to me to have no greater force or significance. (Chomsky 1975a: 41)
Once again this is a purely aprioristic position (the 'eye-flaps'). Nobody can
seriously doubt the fact that the symbolic system called natural language, with
its complex syntactic organization at different levels, is unique in the animal
world. As Greenberg says, thanks to this system
It becomes possible to state lies, hypotheses, and past and future states of affairs.
In the phraseology of semanticists, it is grammar which makes man the time-
binding animal. (Greenberg 1958: 62)
(cf. also Greenberg 1968). Hence, there is no discussion possible about the fact
that language is species-specific in this respect (cf. also Campbell 1971).
However, does that mean that the way this unique phenomenon came about is
irrelevant for understanding its nature? One should, of course, not expect this
to contribute directly to, say, the syntactic analysis of embedded constructions.
But it can very well contribute to our understanding of the general principles of
how language works and is situated in human cognition in general. Hence it can
also indirectly contribute to the syntactic analysis of language if it would provide

evidence for one concept of grammar rather than another.
I guess that nobody will present an analysis of how fish swim and humans
jump as a direct contribution to an understanding of how birds fly. But that is
beside the point. A comparative investigation of how birds fly, fish swim, and
humans jump can be a sensible part of an analysis of how animals solve the
problem of locomotion. As such it can contribute to our understanding of the
principle behind humans having legs, birds having wings and feathers, and fish
having fins. The relevance of that matter for the problem of the analysis of the
genetic determination of phenotypes is obvious. If that is not a topic which does
(or should) interest Chomsky?
Of course, if Chomsky would want to recognize the relevance of these
matters he would also have to recognize the centrality of the concept of
functionality in language. Only if he would get rid of his unwillingness to consider
language an instrument of communication, he could see the relevance of
comparing natural language and primate symbolic systems in understanding what
makes language unique, how it is possible that "language evolves by begetting
that which is not language but transcends it, even while it is dependent upon it"
(Greenberg 1958: 65), and why only humans are able to acquire this system for
communication.
It is hoped that the [...] analysis of the basic essentials of symbolic behavior, while
it renders full justice to the status of language as an evolutionary emergent, may
also serve to make this emergence appear understandable so that, even as Darwin
believed, no conceptually unbridgeable gap separates man from his nonhuman
ancestors. (Greenberg 1959: 123)
As to the latter question, what comparative research has already made

clear is that human natural language could never have emerged without a consi
derable change in the physiology of the vocal apparatus (cf. e.g. Negus 1949,
Lieberman 1975, Dingwall 1975). It has also been shown that the relative volume
of the brain has increased considerably from the higher primates to Homo
Sapiens, although the structural changes in brain-organization have remained
relatively small. Nevertheless, the expansion of the brain suggests an enormous
increase in cognitive capacities in general (cf. Altman 1978), which may have
paved the way to the development of a cognitively more complex system of
communication. Thus, what is species-specific to humans is not so much the
system of natural language as such, but rather the high degree of cognitive
sophistication in general (cf. Lenneberg 1971). Again, this leaves little room for
Chomsky's concept of language as an autonomous system.
Another source of insight into the evolution of language could be the
ontogenesis of language. As already indicated in 2.3.2, language is not the only

system for communication in humans, and some of these other systems appear
to play an important role in the child's acquisition of language.
it seems important to repeat that language does not replace the primate call
system, but is superimposed upon it. While the primate call system still holds pride
of place in the most profound aspects of human relationships, words have evolved
to capitalize environmental processes and the abstract concepts arising from new
complexities of social structure. Language has ushered in the consciousness of self,
of I and thou. But the older system of communication profoundly influences our
behaviour, including our use of language. The primate call system is a substrate
in which language is rooted and from which it has grown. Language has conferred
humanity upon a primate species; myth and technology are its fruits. (Campbell
1971: 19-20)
(See also Lieberman 1973, Lamendella 1977a, Dingwall 1979.) This observation
has often raised the idea that the ontogenesis of language could be an, albeit
very deformed, reflection of its phylogenesis (i.e. what is called 'recapitula-
tionism'). This view has received quite some criticism, but this criticism was
mainly concerned with simplistic views of the correlation. A more sophisticated
concept of the theory is provided by Lamendella (1976, 1977a). This leads him
to the following basic view of how language might have evolved.
The stages of communication development do not merely represent a gradual

accretion of bits and pieces of language that suddenly fall into place as a linguistic
system. At each stage the child is operating with a nonlanguage system of a
definite type [...] adequate to satisfy the child's needs at that time in that
environment. These systems exist for a while and then give way to new levels of
communication systems. (Lamendella 1976: 401)
[Hence] progressively higher levels of communication could be achieved in a

precursor-descendant series by: a. the development of a communication system
within a particular group of hominids; b. the incorporation of this system into the
genetic material in the form of neural systems organized to facilitate its acquisi
tion; and c. the creation of a population base with the new genotype, allowing the
development of a higher level communication system. It is this sort of bootstrap
evolution that probably lies behind the development of language systems by our
species. The result of such a process would be precisely such a series of prematura-
tional stages of communication as is observable in the ontogeny of modern Homo
Sapiens. (Lamendella 1976: 406)
This is spelled out in terms of a model presented in Lamendella (1976: 404-405).

All of the preceding shows that both ontogenetically and phylogenetically
it seems very well possible to provide, at least to a certain degree, a functional
explanation (in an object-inherent sense - cf. note 2 in chapter 1) of language.
Especially in these times of high-tech communication forms, nobody can deny

that communicative efficiency is a factor of utmost importance for the proper
functioning of a community, and hence has a special value in the battle for the
'survival of the fittest'. More sophisticated communication allows for a more
coherent and efficient form of community and thus strengthens the individuals
in the group, and the group in general, in its capacity for survival. Hence, it is
only reasonable to assume that it is pressure to reach higher efficiency and
sophistication in the functional domain of communication which caused natural
language to evolve in Homo Sapiens. Of course, this was only possible due to a
number of physiological developments which created the biological infrastructure
allowing this more sophisticated system to evolve (and which may of course also
have been favored by functional demands). The modalities of any communica
tion system are determined by the physical structure in which the system is
implemented, and by the medium which the system uses to transmit information
and which, as long as telepathy is impossible, is part of the body. Hence, the
structure and organization of human language are indisputably co-determined
by the quantity, flexibility, and specificity of the neurological systems available,
and by the possibilities of the vocal and auditory tract.
Thus, functionality has been and is the motor and orientation for the
development of natural language, but the carrier was and is the physiology of the
human body. The general increase in the cognitive abilities of mankind has
created the possibility of developing functionally more sophisticated systems, but
at the same time these abilities limit the type of functional systems that can be
developed and the degree of functional efficiency they can reach. Thus,
functional systems are simultaneously positively and negatively determined by
cognitive capacities. This confirms the claim made in 1.3.3 that there is no room
for a concept of functional absoluteness or absolute functionality.
In fact, though human language is the most efficient and certainly the most
sophisticated (natural) communication system to date, the simple fact that even
in humans 'more primitive' alternatives remain operative confirms the functional
limitations of language. Everyone knows from experience how hard it can be to
verbally express thoughts, how often verbal communication can break down even
without the interlocutors making obvious 'mistakes', and how often (e.g. in
emotional situations, or in emergency cases) people retreat to alternative non
verbal means of interaction to express themselves because they are more direct
and easily manageable than language. Specifically, emotionally loaded situations
offer a nice illustration of one way in which the medium plays a role in the
efficiency of a communication system. Because language uses acoustics as a
means of transmission, it is a very indirect communicative device which therefore
also creates a relative distance. There is always air between the interlocutors.
For expressing all kinds of emotions (love, aggression, etc.) there is functionally
a strong need for directness and closeness in transmitting the message. A touch
of the hand, or a kiss, or a blow of the fist are therefore much more efficient for
that purpose. Apparently, the increase in functional efficiency of language has
also brought about an increase in functional specialization, at the cost of a
decrease in the variety of usages to which it can be put. But the maintenance of
functional alternatives in the organism compensates for this loss. Maybe humans
are heading for a diversification of the functional domain of communication in
their further evolution.
Notes
1. It should be noted in advance that it is not always easy to get a precise picture of
Chomsky's views. This is due in part to developments over the years in Chomsky's
specification of and argumentation for certain views and notions, often without this being
made explicit. It is also due in part to the fact that some notions remain rather vague (as
will become apparent below). There are some general overviews of Chomsky's cognitive
theory (e.g. Lightfoot 1982, Newmeyer 1983), yet they do not clarify or add much to
Chomsky's own writings. Hence I will mainly limit my discussion to the latter. To avoid
accusations of misrepresenting the theory (a very frequent reaction to criticisms, e.g. in
Chomsky 1975a, 1980a), I will make extensive use of quotations.
2. In Chomsky (1986a, b) (2) and (3) have switched places. Of course, this corresponds with
Chomsky's attitude towards matters of language use. See 2.2.
3. This shows that the borderline between internal and external evaluation of a description
is largely theoretical, and is not very sharp in practice.
4. On the formal characteristics of TG, and their origins, see e.g. Chomsky (1957, 1963,
1975b), Chomsky and Miller (1963), Levelt (1973a, b), Lakoff (1987: 219ff). The
formalism has changed considerably since those early days (cf. Newmeyer 1980 for an
overview), but I will not be concerned with the alterations, although they might be
indicative for Chomsky's own view on the problems his theory faces. Chomsky did npt
change the basic tenets of the theory, and that is what interests me. Note also that the
notion of 'grammar' is used both in a narrow sense to refer to precisely the type of formal
mechanism used in Chomsky's TG, and in a wider sense to refer to any type of model
accounting for natural language syntax. Though this ambiguity is unfortunate, given its
widespread usage of the term in the latter sense, and given the absence of an alternative
term for the narrow sense, we will have to live with this.
5. The only relatively concrete proposal Chomsky has ever made about how the competence
model could be embedded in a performance model is in Miller and Chomsky (1963), in
which an analysis-through-synthesis model of perception is suggested. Cf. 2.2.3.2 for a

discussion.
6. Note that language creativity, which was one of Chomsky's initial points of inspiration
in developing his competence model, is now explicitly relegated to the domain of
performance research. I will return to this in 2.2.2.2.
7. This goes for human understanding in general, not just for scientific understanding.
Chomsky suggests that 'scientific competence' is another, separate module of the mind,
just like linguistic competence. But this use of the concept of modularity seems
particularly implausible. It is hard to see why scientific thinking should be fundamentally
(qualitatively) distinct from thinking in general. Scientific thinking seems to be no more
than an extreme form of thinking as a general human ability.
8. It is obviously unacceptable to turn the fact that hypotheses are necessarily under-
determined by the data (cf. 1.1.2) into an excuse for not handling certain problems. Yet,
this strategy is sometimes suggested - cf. e.g. Chomsky 1964a: 937-938.
9. More recently Chomsky (e.g. 1974) has denied ever to have had the intention to relate
creativity and recursivity. Yet, in a quote like the one just cited, the suggestion for this
correlation seems rather strong. See also Neumaier (1979).
10. On the other hand, this may very well be taken as an indication that recursivity should
figure much less prominently in the organization of the syntactic component than is the
case in TG. For a way to do this without using a phrase structure grammar, see 3.1.
11. This does not mean that there are no structures which must be available as gestalts in
some memory: stereotypes, formulaic expressions, idioms, etc. must somehow be present
as 'prefabricated' entities. Cf. 3.2.3.1.
12. I will not bother with the question whether the word 'rule' is appropriate here. I would
not make a distinction between the notions of rule and procedure - but to avoid
interference with the (rather confusing) philosophical literature on the former notion, I
will use the term procedure as much as possible.
13. Chomsky likes to talk about the deficiency of linguistic behavior, but this does not seem
to be an accurate picture of linguistic reality, despite the existence of mistakes.
14. Note that this explicitly includes 'interpretation errors'. Research on errors so far has
mainly concentrated on the production side, for obvious reasons: they are much more
directly observable. Yet, interpretation errors are no less important in causing
communicative disturbances.
15. One might remark that so far the contribution of research on errors in our understanding
of cognition has been rather limited. But, first of all, although there is some body of work
(e.g. Maclay and Osgood 1959, Fromkin (ed.) 1973, 1975, (ed.) 1980, Garrett 1975,
Dechert and Raupach (eds.) 1980, R.Wiese 1983), this is still very modest. Moreover,
much of this research has mainly focussed on describing the types of errors occurring,
rather than on how to interpret them in terms of a theoretical model (which is only
natural given how little we know about errors).
16. It will not surprise anybody that according to Chomsky (1979) Labov does not really have
a theory at all.
17. In fact, the notion of 'community' as such already involves an idealization, since it is an
extremely fuzzy object: groups can involve different subgroups, they can be determined
according to quite different criteria, and determined according to different criteria they
can have endless overlaps.
18. Of course, the role of variability in biology is smaller than in cognitive science, because
there is less variability in the physical world than in the cognitive world. There are far
fewer significantly different types of human hearts than there are significantly different
languages and styles. This has to do with the difference between the material vs abstract
nature of each. Cf. 2.3.
19. In principle, TG makes a distinction between acceptability, which is not relevant to

competence, and grammaticality, yet it has never been very prominent except that the
former notion is often used as a means to get 'unwanted' intuitions out of the domain
of competence and under the rug of performance. I return to this distinction in 2.2.3.2.
20. Lakoff used the term 'logical structure' rather than 'meaning structure', yet the latter term
is more neutral and is also more in line with his more recent views on language and logic.
21. This view is supported by the fact that judgments of marked utterances are often accom
panied by modifying statements such as: 'can, but only under conditions xyz'.
22. A parallel discussion has been going on with respect to the concept of a 'story-grammar'.
Thus, for Mandler (1982a, b, c) a story grammar is only a means to grasp the building
blocks of a discourse, but the real explanatory concept is that of a story scheme. See also
the reaction of Hjelmquist (1979) to the proposal for a discourse grammar by Hurtig
(1977).
23. It is obviously ridiculous to assume that in order to be able to bike one has to know the
principles of gravity. But the fact that one does not need this knowledge does not mean
that one needs no 'knowledge' at all to bike, or more generally that the capacity to act
does not have to be based on knowledge, contrary to what is sometimes suggested - cf.
Hook (ed. 1969).
24. Lamendella's (1977b) theory of the problem of first versus second language acquisition
nicely corresponds to this view on types of knowledge and their interaction.
25. In support of the distinction between linguistic knowledge and knowledge for these other
capacities, I suppose that Chomsky would not want to claim that being able to play chess
or to go from NY to Chicago are based on specifically innate principles in the same way
as he claims linguistic knowledge is.
26. However, in studies on the logical problem of language acquisition which mainly accept
the Chomskyan framework (e.g. Baker and McCarthy (eds.) 1981), the evaluation metric,
and specifically the simplicity criterion, does remain a central topic of consideration in
basically the same way it used to be in Chomsky's writings (e.g. Lasnik 1981).
27. Of course, 'innate' is a very vague term which can be interpreted in many different ways.
It is not always very clear which meaning Chomsky has in mind (cf. Neumaier 1979), but
it is most likely that he means 'genetically specified'. Cf. the quotation above from
Chomsky (1976: 57); see also Marshall (1980).
28. This way of formulating the problem obviously has its roots in the theory of formal
languages. Cf. e.g. Levelt (1973a), Baker and McCarthy (eds. 1981).
29. Caplan and Chomsky (1980) is an excellent example of how little language acquisition
theory is involved in Chomsky's 'language acquisition theory'.
30. Chomsky has often tried to link the innateness debate to the empiricism-rationalism
controversy in traditional philosophy, suggesting that his position is nothing more than
a continuation of Cartesian views (e.g. Chomsky 1966). Given the vagueness of this
debate (cf. 1.1.3.3) this is not likely to contribute much to solving this problem as an
empirical question. Chomsky (1969c, 1971, 1975a) has even tried to make arguments of
an ideological nature by linking these positions to specific political views, but the
irrelevance of such matters to empirical research need not be proven. Chomsky's
interpretation of the philosophical positions in this debate has often been criticized, on
the grounds that he would unjustly equate empiricism and behaviorism (cf. e.g. Danto
1969, Wells 1969, Bense 1973), and more generally on the grounds that the original
positions were logical theses on the origins of knowledge, not biological or cognitive ones
(cf. Abelson 1969, Black 1970, Cooper 1972, Parret 1979a: 78ff - a reply to Cooper in
Chomsky and Katz 1975). Yet leaving aside the question whether Chomsky has correctly
interpreted the original positions, the nature of his approach may be less different from
the traditional ones than suggested in these criticisms. First, in principle, if a problem
exists with respect to the origins of human knowledge, this is a cognitive problem no less
than a logical problem. Thus, criticism on Chomsky's 'psychological' interpretation of the
debate is unwarranted on these grounds. Second, in practice, since the treatment of the
problem in the traditional philosophical debate was not based on empirical research into
the acquisition of knowledge, it is hardly valuable as a contribution to cognitive
theorizing on the matter. But, as argued, since Chomsky's claim is also basically non-
empirical, the same goes for his claims in this respect.
31. Chomsky also reproaches these authors that they are using different measures for cogni
tive and physical systems. On this problem, see 2.3.2.
32. Despite this clear observation, Lenneberg (1967) still tends to put most stress on the
strong genetic determination of behavior, and he seems to have a variable position on the
effect of the factor of 'experience' from the environment (cf. e.g. ibidem: 220ff). This fact
is no doubt influenced by Chomsky's very strong formulations in this respect, but it does
not mean that Lenneberg really means the same thing as Chomsky does when he talks
about 'strong innateness'. 'Strong' is a relative concept. Though Chomsky likes to refer
to Lenneberg to give his position more strength from a biological perspective, closer
examination, especially also of Lenneberg (1969, 1971), reveals that there are serious
differences between the two points of view. (Cf. also Catlin 1978.)
33. The comparison with the software-hardware distinction is somewhat dangerous, though.
A better comparison would be with the machine-language vs machine distinction.
Software is linked to the machine-language by an interface, and it is only the machine-
language which is directly linked to the machine network. This type of correlation might
be undesirable as a metaphor for the link between cognitive and neurological systems.
34. This is certainly the case during childhood. It is less the case during adulthood, although
then the degree and the nature of rehabilitation seems to depend strongly on the nature
of the lesion.
35. I believe that it will only be possible to make a fully valid implementation of human
cognition on a computer if one manages to build a computer which mimics the human
brain in all its details. But that is a discussion I will not pursue here.
36. Note that those who do not accept the argument concerning the different status of
cognitive and physical systems above do have to assume that the difference between
closed and open programs is a matter of more or less precise genetic specification of the
neurological structures. Hence, they would have to assume that not all physical organs
are genetically determined in the same way.
37. This clearly shows the relevance of the distinction made above in the role of the
environment on the development of physical and cognitive systems. All neurological
systems are equally subject to the normal influence of the physical environment which
affects all physical systems, irrespective of their degree of plasticity.
38. Changeux (1980) still seems to be undecided about whether Chomsky's innateness
hypothesis is compatible with neurological data about plasticity, but Piaget (1980c: 278)
claims that not long afterwards he has changed his opinion, resulting in a view which is
basically in agreement with the above one.
39. Although, strangely enough, Darwin (1872) himself adopted a rather non-Darwinian posi
tion in his considerations of the evolution of communication systems. See also Lieberman
(1975).
3. Linguistic and Conceptual Systems
in Language Processing
3.0. Introduction
To further develop the basic views on the cognitive systems determining

language introduced so far, let's turn to the functionalist paradigm in linguistics.
As a basis for the discussion I will consider one specific framework there, viz.
Dik's Functional Grammar (FG) and the concept of cognition surrounding it. In
3.1 I discuss the plausibility of FG as a model of (aspects of) linguistic
knowledge. 3.2 is an attempt to sketch a more encompassing concept of
language processing, and language production in particular, called Functional
Procedural Grammar (FPG), which endorses most of the main principles of FG
but deviates from the general concept of cognition proposed by Dik. 3.3 contains
an analysis of 'negative raising', as a case study in the debate.
3.1. Some possibilities and limitations of FG
3.1.1. The basics of FG functional syntax
The observation that there is not much plausibility in an autonomous syntactic

competence, la Chomsky, obviously does not change the fact that syntactic
complexity is one of the most essential characteristics of human natural language
which makes it unique among the natural sign systems. Hence, syntax does have
to figure prominently in any plausible model of the cognitive systems determin
ing linguistic behavior. The point is that it has to be a proper subpart of these
systems, directly involved in the overall processes of producing and interpreting
behavior in the sense discussed in the previous chapter. If one still wants to use
the term competence (which I will not, to avoid confusion with the Chomskyan
notion), syntactic competence is not something which exists next to a communi
cative competence. The former is an element of the latter. That is precisely the
way functional approaches to grammar conceive of it. Therefore, rather than
abandoning linguistics altogether, as many language psychologists have done, let's
turn to the functional paradigm in it as a starting point for expanding the basic
views on the cognitive systems determining language use introduced so far.

Rather than approaching the issue by means of a comparison of alternative
models within this paradigm (a necessary enterprise, but too elaborate for the
present occasion), I will start from one particular model within it, viz. Dik's
(1978a, 1989a) FG. There are several reasons for this choice. FG is, next to
Halliday's Systemic Grammar, one of the few relatively established functionalist
grammar models (in terms of permanency of the framework and quantity of
research it generates). But while Systemic Grammar is based on the (rather
dubious) principle of direct linkage between functions and structural elements
(cf. 1.3.3), FG at least has the advantage that it explicitly accepts the principle
of multidimensionality of the form-function correlation argued for in ch.1.
Also, FG is not strictly formalized. There is a clear formalism for
representing the underlying structure of utterances, but many parts of the model
are open-ended in that there is only a general indication of the types of
phenomena that should be handled and the types of mechanisms (rules or
procedures) this might require (in terms of principles they should obey), without
there being a fully developed model of (all or part of) the mechanisms
potentially involved.1 Rather than being an argument against FG (cf. 1.3.2), this
has the advantage of not imposing limits on one's way of thinking about
phenomena without there being good evidence for doing so, and that it leaves
many roads open for integrating FG into a wider framework.
Another reason for considering FG here is that it has the interesting
property of, in recent years, being surrounded with the rudiments of a more
general theory of human cognition. Like most linguistic models, FG started out
as an attempt to deal with natural language syntax without much concern for
farther-reaching issues, such as the way it could fit into a more encompassing
model of the cognitive systems determining linguistic behavior. As such it did not
really do justice to its professed aims. In fact, Dik (1978a: 6ff) has formulated
three criteria which FG should satisfy in order to achieve 'explanatory adequacy':
(i) pragmatic adequacy: the model should be conceived as part of a wider
pragmatic theory and should be compatible with what is known about verbal
interaction in general;
(ii) psychological adequacy:
Describing the language system, i.e. the system of rules underlying the construction
of linguistic expressions, is not the same thing as describing the psychological
structures, principles, and strategies which determine the way in which linguistic
expressions are perceived, interpreted, processed, stored, retrieved, and produced.
A grammar should nonetheless be psychologically adequate in that it should not
be incompatible with strongly validated psychological hypotheses about language
processing. (Dik 1978a: 7)
SYSTEMS IN LANGUAGE PROCESSING 185
(iii) typological adequacy: the linguistic theory should provide grammars for
typologically widely diverging languages and has to account for the similarities
and differences between these languages.
In terms of the adequacy criteria specified in 1.1.2, 'typological adequacy' is
partly an element of the principle of 'external descriptive adequacy' and partly
of the principle of 'internal descriptive adequacy' (cf. 2.1), and 'pragmatic
adequacy' entirely corresponds to the principle of 'internal descriptive adequacy'.
However, from the argumentation in ch.1 it is clear that 'psychological adequacy'
is a notion of a different order: a model is cognitively adequate if it is internally
and externally descriptively adequate; and once it has reached that level the
grammar is not merely compatible with a theory about the cognitive systems of
language processing, but is such a theory.
In actual practice FG so far has only paid due attention to typological
adequacy, and there has been no special concern with elements related to
pragmatic adequacy (and 'psychological adequacy' in as far as it implies taking
into account research in language psychology). Even discourse structure is not
considered: despite Dik's (1978a: 15) initial statement that FG "is meant to cover
any type of linguistic expression, to the extent that the internal structure of that
expression is governed by grammatical rules", and thus that it is "not restricted
to the internal structure of sentences, inasmuch as there are combinations of
sentences related by syntactic and semantic rules", it is a pure sentence-grammar.
There is a growing interest in the influence of discourse structure on intra-clausal
elements (cf. e.g. Bolkestein 1983, 1985, 1987, Hannay 1983, 1985a, b, Hannay
and Vester 1987, Siewierska 1987), but the organization of discourse is still
completely absent in the model.2
Nevertheless, from Dik (1983a: 6ff) on there is a gradual reformulation of
the principle of 'psychological adequacy', in the sense that FG is now said to
have to aim at directly being part of a model of the cognitive mechanisms
involved in language use, which would of course also be an incorporation of the
'pragmatic theory' mentioned in the principle of pragmatic adequacy (cf. also
Dik 1989a: 13, 51-53). Simultaneously Dik (e.g. 1986b, 1987b, c, d, 1988, 1989b)
has made suggestions about what such a larger cognitive framework surrounding
FG could look like. Although these suggestions have mainly been programmatic
so far (FG as most recently presented in Dik 1989a still stands on its own, and
apart from one paper by Dik 1990 himself the treatment of linguistic phenomena
in FG still does not involve the wider framework), the proposals are sufficient
to serve as a useful point of departure for the discussion later on of how a
general model of the cognitive mechanisms determining linguistic behavior could
be conceived.
However, let's first concentrate on FG as such and take a look at its

general properties.3 FG is built on a number of principles which reflect Dik's
concern to avoid too much abstractness in linguistic description, viz. (Dik 1978a:
10ff, 1989a: 17ff):
(i) Avoid transformations in the sense of structure changing operations. This
involves more specifically the rejection of deletions, substitutions, or permuta
tions of elements which have already been specified earlier in the grammatical
derivation.
(ii) Avoid filtering devices such as are accepted in TG (e.g. Chomsky and Lasnik
1977) which eliminate structures which have been formed earlier on in the
grammar.
(iii) Avoid abstract semantic predicates. This involves the rejection of semantic
decomposition and the assumption that there is no level of analysis in the
grammar in terms of language-independent concepts (cf. also Dik 1978b). This
applies even to the meaning definitions of lexical entries.
On the other hand, the functionalist attitude reflects itself in the fact that the
model makes central use of relational (or functional, as they are somewhat
misleadingly called in FG) notions at different levels rather than only of
categorial notions, and in the fact that most phenomena are considered gradual
(in terms of hierarchies, priorities, and markedness) rather than discrete or
binary (cf. Dik 1989a: ch.2).
The representation of the underlying structure of utterances in FG has
been considerably expanded in recent years, as compared to the Dik (1978a)
version. It now involves layering', i.e. a system of hierarchically ordered levels
in the representation (cf. Hengeveld 1989, Dik 1989a). This system has been
introduced to account for the scope and behavior of operators such as tense,
aspect, and mood, and of different sorts of modifiers (adverbials, complements,
etc.) in utterances. The first part of the FG machinery is concerned with the
construction of this layered structure.
The building blocks are the set of predicates and terms of a language,
which are together contained in the 'fund' (Dik 1980a) of the model. In FG all
'content words', including nouns and adjectives, are considered predicates. There
is a distinction between 'basic predicates' and 'derived predicates' on the basis
of whether or not they can be derived from other elements in the language on
the basis of a productive process. Hence, idioms are also treated as basic
predicates. The basic predicates are contained in the lexicon, which is the core
component of FG. Each basic predicate is accompanied by the following
information:
(i) An indication of its category (V(erbal), N(ominal), A(djectival)).
(ii) A 'predicate frame' containing an indication of the argument pattern which

normally accompanies the predicate, and for each argument slot an indication
of the 'semantic function' (Ag(ent), Rec(ipient), Go(al), etc.) and the selection
restrictions for filling in the slot, as shown in (l)-(3).
(1) eat v (x1: animate ( X 1 )) Ag (x2: food (x 2 )) Go
(2) girlN (X1)
(3) sympathetic A (x1)
(iii) Meaning postulates, which have the form of predications themselves, and
which together constitute the meaning definition of the predicate.
(iv) An indication of non-productive allomorphs of the stem (Dik 1981c, 1989a:
294ff).
Predicates (including their frames) can be input to 'predicate formation
rules' (Dik 1980a: ch.2, De Groot 1987) which are responsible for all kinds of
productive derivational processes of a morphological nature (e.g. the formation
of diminutives, comparatives, superlatives, adjectives from nouns, etc.) or of a
syntactic nature (e.g. valency reduction, causative formation, nominalization, the
'modalization' of predicates through modals, etc.). Predicate formation can
change the meaning of a predicate, and how it does is indicated in each
individual rule.
Next to predicates there are also terms in the fund, i.e. the referring
elements of a language. Very few terms are basic in the sense that they can only
be used as terms and cannot be derived from other elements in the language.
Examples would be personal pronouns and proper names. Basic terms are
specified in the lexicon. The vast majority of terms is derived by means of 'term
formation rules' however. (4) shows the structure of terms.
(4) (xi: 1 (xi): 2 (xi): ... : n (xi))
xi symbolizes the referent, n stands for one or more term operators (definite-
ness, number, etc.), and each1(xi) is an open predication, i.e. a predicate frame
in which all positions except that for xi are again to be filled by term structures.
(This is how recursivity is realized in FG.) Each open predication introduced by
a colon Y acts as a restrictor on the value of xi, and the ordering of these
restrictors is relevant in that each following restrictor further limits the earlier
ones. Thus, term (5a) is represented as (5b) (Dik 1989a: 55).
(5) a. the big snake that lives in the zoo

b. (dlx i : snake N (x i ) : bigA (x i ) : live v (x i ) Pos

(dlxj: zoo N (x j ) ) Loc )
Basic and derived predicate frames together constitute the 'nuclear

predicate frames', and if their argument slots are filled with terms we have
arrived at 'nuclear predications'. They form the basis for the layered structure
of utterances. To arrive at an actual speech act the predication is surrounded by
a hierarchical series of 'shells', and it may be completed with operators
(grammatically expressed modifiers) and/or satellites (non-obligatory modifying
constituents, simply put - a more precise definition in Dik et al. 1990) at one or
more of these levels, as shown in (6) (from Hengeveld 1990: 12).
(6) (E1: [ 4 ILL: 4 (S) (A) (3X1: [ ] (X1): 3 (X 1 )] (E1): 5 (E1))
2e1: [1Pred: l (x1: PredN (x1)) ... (en)l (e 1 ): 2 (e 1 )

A predication designates a SoA. On top of the predicate and arguments, it may
sometimes be necessary to specify further the internal structure of this SoA by
adding predicate operators ( 1 ) e.g. specifying certain types of aspect and
predicate negation, and/or predicate satellites (1) such as adverbials of manner,
speed, and instrument. The variable (e1) symbolizes the SoA expressed by the
resulting structure, which is a complete predication. This predication can further
be provided with predication operators ( 2 ) such as tense, certain types of
aspect, and 'objective modality', and/or predication satellites (2) including
adverbials of time and place. They locate the SoA expressed in the predication
in space and time (real or imaginary), thus turning it into a 'potential fact' which
is symbolized by the variable X1. This entire structure, called 'proposition', can
again be modified by proposition operators ( ) including evidentially and
'subjective modality', and/or proposition satellites (3) such as adverbials
expressing 'propositional attitudes'. This pattern becomes a speech act by
embedding it in an 'illocutionary frame' which contains an abstract 'illocutionary
predicate' (ILL, i.e. DECL(arative), INT(erogative) or IMP(erative)) and an
abstract representation of the speaker (S) and the addressee (A). ILL can again
be modified by illocutionary operators ( 4 ) mitigating or reinforcing the basic
illocution, and/or illocutionary satellites (4) specifying properties of the speech
act. The E 1 variable symbolizes the speech act, and finally clause satellites ( 5 )
locating the speech act in a discourse context may be added.
This underlying representation of utterances is considered unordered from
a syntactic perspective. That is, it has no left-to-right linear order (it is so
presented in print only because it is the easiest notation), but only obeys to
semantic principles of organization. With a view to determining how the surface

structure of the utterance should be organized, two further types of relational
elements are introduced, viz. 'syntactic functions' and 'pragmatic functions'.
These functions are added to the semantic function specifications of constituents.
Hence a constituent can sometimes end up with three function markings.
FG distinguishes two syntactic functions, subject and object. They are
defined in semantic rather than syntactic terms, viz. as the (respectively) primary
and secondary vantage point from which the SoA expressed in the predication
is presented. Thus, they indicate the perspective from which the S/H presents
the SoA. Their assignment-possibilities are determined by a number of factors
(cf. Dik 1989a: ch.10, 11), the most important of which is the semantic function
of the arguments and satellites. The hierarchy of (declining) preference of
assignment is more or less: agent or other first argument, goal, recipient,
beneficiary, instrument, location, and time. Thus, in principle they can only be
assigned to arguments and predicate satellites, very occasionally also to some
predication satellites.
There are two clause-internal pragmatic functions in FG, viz. topic and
focus (with a number of subtypes within each category - cf. Dik 1989a: ch.13).
They concern the informational status of constituents. In general, a topic is 'what
the clause is about', and a focus is what is communicatively most salient in terms
of the information of the addressee. Topic can only be assigned to a term, focus
to any part of the underlying clause pattern.
With these function specifications added, we have a fully specified
underlying representation of the clause, which is input to the second main part
of the FG-machinery, viz. the expression component responsible for developing
the surface syntactic pattern of the utterance. There are three main types of
expression rules (which sometimes interact - Dik 1989a: 290ff): those determin
ing the form of constituents, those determining word order, and those specifying
the prosodic contours.
The first type is responsible for 'translating' the effect of operators and
function markings into the form of the terms (by introducing case markings,
adpositions, etc.) and the predicate (by introducing auxiliaries, agreement
markers, voice and tense markings, etc.), i.e. mainly matters related to
inflectional morphology (cf. Dik 1989a: ch.15).
Word order is managed by means of placement rules. They assign a
position to the elements of the unordered underlying representation relative to
each other on the basis of specific ordering templates valid in specific languages.
(7) indicates the general format of an ordering template at the clause level.
(7) P1 (V) S (V) O (V)

P1 is a special sentence-initial position normally reserved for constituents such

as question words, relative pronouns, or subordinators, or for constituents with
a pragmatic function marking (languages may have more such special positions).
V, S, and O are the positions for verb, subject, and object. An ordering template
is the result of a complex set of ordering principles (cf. Dik 1989a: ch.16) which
interact and sometimes even compete, and each language draws on them
differently when organizing utterances (i.e. languages realize (7) - or similar
patterns within the domain of the term - differently, and sometimes even have
several patterns, e.g. for different illocutions, or for main and subordinate clause,
etc.). Thus, these principles are not absolute laws, but rather tendencies, each
of which can be overridden by one of the others.
Finally, prosodic rules not only take care of the realization of lexical tone
and accent, and of modifications due to morphological adjustments, but also
express effects of pragmatic function assignment and illocution (See Dik 1989a:
ch.18).
3.1.2. On the cognitive plausibility of FG
It is certainly too early to make definite claims about the adequacy of FG as a

model of aspects of the cognitive systems determining language behavior (in
terms of amounts of interdisciplinary research done to test the model). Yet there
are reasons to believe it has some initial plausibility, at least a lot more than TG,
despite what some seem to believe. It has become a tradition among TG-
researchers to classify alternative grammar theories as notational variants of TG:
Chomsky has done this e.g. with respect to Case Grammar and Generative
Semantics, and so have a number of Chomsky followers and sympathizers with
respect to FG (Hoekstra 1978, Koster 1982a, Miller 1986, Muysken 1988).7
(See also Van Riemsdijk 1982.) Miller (1986) has even tried to give this
argument (more specifically the claim that FG is a 'subtheory' of TG) the shape
of a formal proof to the effect that for every FG producing a certain set of
representations it is possible to develop a TG which produces the same set of
representations, but not vice versa.
However, apart from the fact that even purely formally Miller's 'proof is
non-conclusive (Nuyts 1986b, 1988: 513ff), what he and his fellow-critics fail to
understand is that a formalism is not (or ought not to be) something arbitrary.
As argued in 2.2.3.2, even if two models are 'computationally equivalent' this
does not imply that they are equally plausible from a cognitive viewpoint. But
FG and TG are not even computationally equivalent. A formalism is linked to,
and is thus meaningful in terms of, the theory out of which it has developed, and
the theories behind FG and TG are basically incompatible. To develop his proof,
Miller has to devise a type of TG which is simply unacceptable from the
perspective of Chomskyan theory, among others because it introduces semantic
notions in the syntactic core of the grammar. At the same time, his model
remains unacceptable in the FG framework because it maintains a number of
formal characteristics which are unacceptable from the theoretical perspective
taken there. It may be true that one could introduce an unordered deep
structure in TG (Koster 1982a), but the fact that it would be quite arbitrary to
do so and that therefore it is not done, whereas in FG it is a basic option, is
significant. It may be true that concern with predicate structure and predicate
formation is not particular for FG (Muysken 1988), but the fact that FG handles
them in its own way which makes types of problems occurring in a TG-
framework irrelevant (cf. Mackenzie 1988) is again significant. Most of these
critics doubt the links between the FG model and its (functionalist) theoretical
options, so let's briefly go into this.
First, consider the relational ('functional') notions. Their role from the
perspective of a functional approach to language is obvious: as noted in 2.2.2.2,
they are elements which allow to link the structural and the functional
dimensions of language.8 Their relevance in a grammar (or a language
processor) is by now generally accepted, not only in linguistics (even in TG), but
also in language psychology (e.g. Clark and Clark 1977, Bates and MacWhinney
1982, Jarvella and Engelkamp 1983). A typical property of FG is that it includes
three different types, rather than one or two as is done in many other frame
works. Relational Grammar only uses syntactic functions, Case Grammars
normally have semantic and syntactic functions, Prague School linguistics mainly
focusses on pragmatic functions, TG uses semantic functions (case roles) and
pragmatic functions (thematic roles).9 Similarly, in language psychology Kempen
(e.g. 1978) only uses syntactic functions, Schlesinger (1977) only semantic and
pragmatic functions.10 This has caused some to criticize FG for being 'eclectic'
(e.g. Hoekstra 1978, Perlmutter 1981), and for aiming only "to have sufficient
stipulations available to be able to 'note down' what can be observed in
utterances" (Koster 1982a: 364; my translation) without explaining much. Yet
these relational categories have all proven their value for explaining linguistic
structure, each in its own way, and the FG literature contains numerous
examples to show the value of combining these elements in order to account for
various structural phenomena (see also 3.1.3). In fact, each of them is related to
a specific dimension of phenomena, all of which have to be integrated in an
account of linguistic behavior anyway: semantic functions to the conceptual
organization of SoAs in the S/H's knowledge of the world, pragmatic functions

to the informational status of conceptual elements in terms of the knowledge of
the S/H and H/S in an actual situation, and syntactic functions to the way
components of the conceptual SoA are linearly organized in the structure of
utterances. Thus, while one can have quarrels with the way FG characterizes
these functional levels and their mutual relationships (see below), the idea to use
all of them seems perfectly sound from the perspective of the assumption that
linguistic utterances are determined by several functional factors.11
The way in which the function categories are introduced in the FG-model
is also explicable in terms of functionalist considerations. The choice between
treatment in an interpretive module (as in TG) or in the core of the grammar
(as in FG) is not arbitrary (or, in Miller's 1986 terms, the fact that FG adopts
a 'radical version of the Katz-Postal hypothesis' is not non-empirical). In line
with the argumentation in ch.2, FG is not a competence-model in the
Chomskyan sense. It represents linguistic knowledge as directly constituting the
capacity to behave linguistically (cf. 2.2.3.3). Hence, it involves what can be
called a 'procedural' concept of linguistic knowledge, in correspondence with the
notion of 'procedural semantics' as used in AI for a similar concept of general
knowledge systems (cf. e.g. Winograd 1975, 1976, Miller and Johnson-Laird 1976,
Johnson-Laird 1977a, 1983, Wilks 1982). The fact that FG involves 'declarative'
knowledge representations such as predicate frames and word ordering
templates is not in conflict with this claim. As Winograd (1975) states,
representation types such as scripts and frames (the FG predicate frames and
ordering templates are instances of these in AI and language psychology very
common structures - cf. Schank and Abelson 1977, Schank 1980, Schank and
Birnbaum 1984, J.Mandler 1984) are unavoidable anyway (cf. also Johnson-Laird
1983). The point is that if declarative knowledge is postulated, it is homoge
neously integrated in the procedural system and is handled directly by the
procedures. In this respect, FG fully corresponds to similar types of models in
AI and language psychology.
Thus, FG is not independent of speaking and understanding: as currently
formulated, it is obviously conceived in a productive mode, and therefore' has to
be seen as (part of) a language production system (cf. Dik 1983a: 6f, 1987c,
1989a: 13, 51ff).12 Hence, since speaking is a matter of expressing meanings
(that is an observational statement - cf. 1.1), FG cannot but adopt 'an extreme
version of the Katz-Postal hypothesis': in a production model semantics has to
precede syntax, and an interpretive semantics is out of the question. Upon the
same argument, relational categories of the type under consideration have to
figure in the core of the grammar (cf. 2.2.2.2). The task of a production
grammar is to map a meaning into a syntactic structure, and since relational

notions are elements of correlation between meaning and structure, they have
to be available where syntactic decisions are to be taken. (For similar reasons,
these relational notions also have to be central in an interpretive module.) From
this perspective their treatment in TG is entirely ad hoc.
Actually, these considerations also have repercussions for some elements
of current FG. First, consider the position of pragmatic functions in the model,
which are assigned after the syntactic functions. Since they directly relate to the
informational status of conceptual information, one would expect them to be
very fundamental in the grammar, much more fundamental than the syntactic
functions which are more specifically related to the structural organization of
utterances. Of course, the FG definition of syntactic functions as 'perspectivizers'
in the SoA is semantic, too. But perspectivization can best be understood as
resulting from the status of conceptual information along several dimensions,
including the informational value of elements in the SoA (cf. the fact that topic
and subject often coincide), and their roles in the SoA (cf. the dependence of
syntactic function assignment in FG on semantic functions). As such, perspectivi
zation is complex and derivative, whereas the informational status of elements
is basic. This issue will repeatedly reoccur in this chapter.
Second, the above discussion also implies that the (albeit sporadic)
attempts to treat phenomena (e.g. the violation of selection restrictions - Dik
1978a: 44ff, 1980c, 1989a: 79ff) on an interpretive basis in FG are dangerous.
Interpretation strategies are required in an interpretive grammar, but in as far
as current FG is oriented at the production side of language processes, it should
rather include an account of the speaker's decision processes with respect to
these phenomena. The discussion in terms of interpretation strategies may give
the impression that the phenomena in question are sufficiently dealt with, but
that is not the case, and the result is a production grammar which probably lacks
important features to account for these phenomena.
Several aspects of the FG conception of the lexicon, the structure of its
entries, and its position in a grammar appear motivated from a functionalist
perspective as well, and again they are similar to proposals in AI and language
psychology. The basic distinction between the lexicon and lexically related
processes and the expression component realizing surface structures, and the
claim that the former is central for the latter, is comparable to the distinction
and organization of the system determining the lexical hierarchy of utterances
and the system for word ordering in the model proposed by Kempen and
Hoenkamp (1982, 1987). The distinction as such is well supported by analyses
of speech errors (cf. Garrett 1975, 1980, Bock 1987). Also, it is functionally quite
logical to suppose that syntax depends on the lexicon, rather than lexical
selection being syntactically motivated (as in TG), since normally an S/H first
makes lexical choices and only then enters the syntactic processes. (The situation
is probably similar in language understanding.) S/H first has to determine which
verbal elements are best suited to express the relevant aspects of the relevant
SoA before being able to start organizing these verbal elements into a syntactic
surface pattern. The structuring obviously depends on what there is to be
structured, and not the other way around.
In this respect, the FG view that predicates in the lexicon carry a predicate
frame is also attractive and intuitively plausible: in this way, lexical choice
immediately leads to the introduction of syntactically important information.
While other elements of the FG lexicon and the discussions surrounding them
(e.g. concerning the treatment of productive regularities and special forms - cf.
e.g. Dik 1979b, 1981c, 1989a, Samuelsdorff 1982, Watters 1985) are quite
comparable to debates about the lexicon in language psychology (cf. Butterworth
1983 - some of these matters will be discussed in 3.2.3.1), the direct inclusion of
predicate frames is all but traditional, neither in production nor in interpretation
models (yet, again, see Kempen 1978 and Kempen and Hoenkamp 1987 for a
similar proposal). Still, information about the valency of predicates is required
anyway, and lexical predicate frames are a very elegant solution to problems
which would otherwise have to be dealt with in terms of the procedures for
language production (and understanding) directly. In fact, proposals which do
not include lexical predicate frames often face considerable difficulties. For
example, Schlesinger (1977: 39ff) is obliged to introduce 'syntactic selection
restrictions' in the lexicon, but after all this boils down to the equivalent of
predicate frames. Goldman (1975) has to introduce a separate store of syntactic
frames (the 'concexicon') next to the lexicon, which only makes for an
unnecessary complication of the model. Moreover, Kempen (1976) provides
experimental evidence which supports the principle of predicate frames.
Actually, FG predications are based on the assumption that the structure
of utterances is essentially 'predicate + arguments', rather than 'subject +
predicate' (the option taken in TG, even though the notion 'subject' is not used).
This is, of course, quite widespread in linguistics and has a long tradition within
this discipline, but it is also frequently found in language psychology (cf. e.g.
Schlesinger 1977, Kempen 1978, Garrett 1980) (probably in correspondence with
the use of valency notations for conceptual representations). One linguistic
reason for this option is the fact that one can easily do without the 'NP + VP'
notation to describe linguistic structure (as FG constantly demonstrates) whereas
one cannot do without the valency notation (even TG has it, albeit in the
interpretive semantic component) - and introducing both seems an unnecessary

complication. Another linguistic reason is the existence of languages with a basic
VSO word order. A valency representation has no difficulties with such a
pattern, but in the 'subject + predicate' system one has to assume that the
subject systematically penetrates the predicate, thus masking the underlying
organization, which is obviously quite unattractive. Experimental evidence for the
valency representation is provided by Levelt (1973c: 25ff). He has tested
language users' intuitions concerning the structural cohesion of constituents in
utterances, and they do not seem to correspond to an 'NP + VP' system la
TG. The hierarchical constituent structure in TG appears too strong, and Levelt
(albeit very carefully) concludes that a dependency or valency system much
better fits the data.
The choice for syntactically unordered underlying predications and for an
expression component operating in terms of placement rather than transforma
tion perfectly suits the principles of a cognitive-pragmatic theory as well. FG
does not stand alone with these options. In language psychology this is e.g. also
proposed by Schlesinger (1977); and it corresponds with the way Garrett (1975,
1980) characterizes his two levels in language production, viz. the functional level
(comparable to the level of predication in FG) and the positional level
(comparable to the ordered surface structure in FG). (See also Bock 1987.) In
general, though there still are models in language psychology using transforma
tions (Goldman 1975, Kempen 1978, Kempen and Hoenkamp 1982),13 there
is a general tendency to reject this type of mechanism (cf. 2.2.3.2), for good
reasons. Since one should not assume that the conceptual representation of an
SoA is (natural language) syntactically organized (cf. Levelt 1973c: 166ff), and
since predications are (in some sense, depending on one's conception of the
entire process - cf. 3.2) basic elements for mapping a conceptual SoA into a
syntactically organized surface structure, it is very reasonable to assume that
these predications do not yet have a syntactic organization but still obey
semantic organizational principles derived from the conceptual SoA. (The claim
that predications are unordered obviously does not mean that they are not
organized, but only that their organization is not syntactic.) If this is appropriate,
the organizational principles of the underlying and the surface structures are
entirely different in nature, hence it is impossible to state the mapping between
them in terms of transformations, since a transformation implies that the
ordering principles before and after the operation are similar. Of course, one
could still assume a placement system followed by a replacement system. Yet the
idea that an S/H would first order and then undo the ordering (by deletion,
movement, or whatever) is intuitively implausible and inelegant, and as long as
there is no indication that a combination of these two systems is required to

account for data it should be avoided in a process model.
FG may not seem consistent in banning transformations, since the
predicate formation rules do involve similar processes. Yet the functionality of
predicate formation and expression is obviously quite different. The latter only
develops and organizes a structure chosen by S/H to express a conceptual SoA.
The former involves systematic processes available to S/H in choosing an
appropriate verbal representation for a conceptual SoA. That is, it is a system
which, if necessary, allows S/H to adapt structures available in the lexicon to
make them more suitable as a verbal representation of what (s)he wants to say.
As such, this system is functionally well motivated.
Finally, the introduction of a layered underlying representation in FG in
order to account for all kinds of SoA modifying elements in language fits into
a more widespread tendency toward such types of representation in linguistics
(cf. Lyons 1977, or Role and Reference Grammar - Foley and Van Valin 1984,
Van Valin 1990), but also in language psychology (e.g. Johnson-Laird 1983,
Clark 1987). There is no question about the validity of the principle of layering
as such, but there are some caveats with the way it is motivated in FG. In
Hengeveld's (1989a: 128ff) characterization the structure in (6) above consists
of two times two layers: the x and e variables constitute the 'representational
level', the X and E variables the 'interpersonal level'. The former is said to
concern the 'narrated event', the latter the 'speech event', and a predication is
claimed to have two different functions: designating a SoA at the representation
al level, and representing a potential fact as the content of a speech act at the
interpersonal level.
In view of the discussion about the functionality of language in 1.2, these
specifications are not without problems. It is a bit unclear how to understand the
notion 'interpersonal' here. On the one hand, the correlation with the notion of
a speech event suggests that it is used in the 'communicative' sense, which would
mean that it refers to a role function. In that case, since 'representing SoAs' is
an organic function of language, Hengeveld is using functional notions from two
different and non-conjoinable dimensions to characterize two parts of one single
structure. One could reply that they relate to two different levels in that
structure, too. That fact does not help, however, since there is no relationship
whatsoever between layering in clause structure and distinctions in the way one
can speak of language functions as discussed in 1.2.
On the other hand, Hengeveld refers to Halliday for the term 'inter
personal', which suggests that it is a specification of an organic function of
language. But, as argued in 1.2.2.3 this is a disputable functional category.
Moreover, even if one distinguishes between an intentional and a socializing

function, there is no reason to assign a 'higher level status' to structural elements
related to these elements as compared to elements related to the informative
function: all of these function categories have an equal status in determining
structural categories of language. In fact, the suggestion that specific strata in the
structure of an utterance hold a one-to-one relationship to specific functional
dimensions is akin to the Hallidayan concept of grammar, but is disputable in
itself (cf. 1.3.3). In addition, it would be inconsistent to introduce such a
principle at this level while the expression component of FG explicitly accepts
the principle of a multifunctional determination of structural elements (Dik
1986a, 1989a: 336ff).
In line with these observations, it is rather difficult to conceive of what it
means that in an utterance one can distinguish between a speech event and a
narrated event. An utterance contains elements of narration, in the sense that
it is about a 'narrated event'. But it is a speech event, it does not contain one.
These are non-conjoinable notions. Similarly, it is hard to get the point of the
claim that a predication has two functions, viz. designating a SoA and represent
ing a potential fact which is the content of a speech act. One would assume that
a potential fact is a SoA, and vice versa. So why have two different levels and
variables for this? Hence, the least one can say is that the motivation for the FG
layering system is not straightforward, and this raises the question whether it is
entirely satisfactory. A discussion of this requires the consideration of some other
issue, and I will return to this later (cf. 3.2.1.3, 3.2.3).
Some have tried to substantiate their negative attitude toward FG by
arguing that the model cannot account for a number of linguistic facts unless it
is changed in ways which go against its basic principles. A closer look at some
of these cases backs a number of the theoretical arguments in the previous
chapters, reveals much about the real possibilities of FG, but also highlights
some of the limitations of its present conception.
3.1.3. Some alleged descriptive problems for FG
3.1.3.1. How functional considerations make the difference

Standard TG analyses of causatives in Dutch, such as (8)
(8) Piet liet Jan een boterham eten

Pete let John a sandwich eat
"Pete made John eat a sandwich"
assume a bisentential deep structure containing a matrix sentence Piet liet and
a complement Jan een boterham eten. Dik (1980a: ch.3) has argued against such
an analysis on the basis of the fact that the surface properties of this 'comple
ment' differ from those of a normal complement in Dutch in several respects.
(i) Normal complements can be finite, the causative 'complement' cannot.
(9) Ik zag dat Jan wegging

I saw that John went-away
"I saw that John went away"
(10) *Piet liet dat Jan een boterham at
(ii) Normal complements can be put in the sentence-initial P1 position (cf. (7))
of the main clause, but the causative complement cannot.
(11) Dat Jan dit gedaan heeft geloof ik niet

that John this done has believe I not
'That John has done this, I don't believe"
(12) Van een boterham eten liet Piet niet
(iii) Normal complements can be referred to anaphorically, but causative

complements cannot.
(13) Piet denkt dat Jan weggaat en Marie denkt

Pete thinks that John goes-away and Mary thinks
dat ook
that too
"Pete thinks that John goes away and Mary thinks so too"
(14) *Piet liet Jan een boterham eten en Marie liet dat ook
(iv) Normal complements can be passivized, but the causative complement

cannot.
(15) Piet denkt dat Jan door Marie afgehaald wordt

Pete thinks that John by Mary fetched is
"Pete thinks that John is being fetched by Mary"
(16) *Piet liet door Jan een boterham gegeten worden
(v) Normal complements can be questioned, the causative complement cannot.

(17) Wat denkt Piet?

what thinks Pete
"What does Pete think?"
(18) *Wat laat Piet?
Hence, if the 'complement' of laten has complement status in 'deep structure',

this would always be neutralized toward the surface. According to FG principles,
and from a functional perspective in general, that is implausible. Since causative
formation is a perfectly productive process in Dutch, the FG solution for the
problem is a predicate formation rule which derives a causative predicate frame
from a normal predicate frame by adding the causative predicate and one
(agentive) argument, while leaving the rest of the original predicate frame
untouched.
(19) input: some arbitrary predicate-frame

(x 1 ) ... (x n )
output: laten v (x 0 ) Ag (x1) ... (xn)
Koster (1982a) has argued that this analysis neglects two essential facts, viz.
(i) there are two main verbs in the causative construction, laat and eten, and (ii)
there are two agents, Piet and Jan, each related to one predicate. On the
assumption that a simple sentence can only contain one agent (Fillmore 1968)
and one predicate, the FG analysis is very implausible. In fact, according to
Koster criteria (i)-(v) above are not waterproof either. He rejects (i) because
there are real complements which can only occur in the infinitive as well, as in
(20).
(20) Piet probeert het boek te lezen

Pete tries the book to read
"Pete tries to read the book"
And he rejects (ii) because there are real complements which cannot be
preposed into P1 either, as in (21).
(21) a. Het schijnt dat hij komt

it seems that he comes
"It seems that he comes"
b. *Dat hij komt schijnt het
Finally, he rejects (iv) since there are real complements which cannot occur in
the passive either.
(22) Van beloofde Piet geslagen te worden

John promised Pete beaten to be
The question is whether one can really say that an utterance such as (8)
contains two main predicates. It depends on what one calls a 'main predicate':
in a purely formal sense Koster is right since there are two independent
morphemes. However, from a functional perspective he is not: semantically laat
eten is one unit, the meaning of which is not reducible to the meaning of the
component predicates; and both morphemes function together to constitute the
message transmitted by an utterance like (8). Hence there is only one main
predicate in (8). This is not unique: there are many cases where one semantic
unit is expressed in two (or even more) formal elements. Consider the
parallelism with combinations of modal auxiliaries and main predicates. We will
meet other examples in 3.1.3.3 and 3.3. This phenomenon of formally distinct
elements constituting a functional unit without forming a real constituent (real
constituents are 'closed, impenetrable domains', but these functional units are
not in that their formal components can be separated) is poorly understood, yet
potentially interesting to understand word ordering principles in language.
The case of the two agents in (8) is more difficult. A predicate frame with
an agent as a first argument would come out of rule (19) with two agents indeed.
However, that does not mean that the 'one-sentence' analysis should be rejected,
but rather that there is something wrong with rule (19). In fact, intuitively (8)
does not contain two agents at all. Piet causes an action, although he does not
perform it. Jan, on the other hand, does perform the action, yet he is not in
command of the situation. To grasp this, one may need an approach to semantic
functions which is considerably more sophisticated than the FG view. I cannot
go deeply into this extremely complex matter here. But, as will be further
explained in 3.2.3, what may be needed is a differentiation between two sets of
semantic notions: on the one hand, a set of conceptual roles which categorize
the subtle position entities can have in an S/H's knowledge of the world; on the
other hand, a much more limited set of case categories which function in
predicate frames and which canalize the more subtle conceptual roles. (The
semantic functions in FG are floating somewhere in between.) In the framework
of such a proposal, De Schutter (1989) argues that in (8) Piet can be understood
to be an agent and Jan a causative (in the same way as the wind in the wind
opened the window is a causative). Hence, if the predicate formation rule in (19)
is changed accordingly, the problem of the occurrence of two identical roles is
solved.
Koster's objections to Dik's criteria for denying complement status to the

causative complement are not too convincing either. He is right in stating that
there are also real complements which can only occur in the infinitive, and that
Dik's criterion (i) cannot be upheld. However, he does not refute Dik's criteria
(iii) and (v). His objection against Dik's criterion (ii) is misguided. The fact that
(21b) is out has nothing to do with the issue. Consider (23):
(23) a. Het schijnt vast te staan dat hij komt

It seems firm to stand that he comes
"It seems to be certain that he comes"
b. Dat hij komt schijnt vast te staan
(23b) is perfectly all right. Het in (21a) is a dummy P1 filler which has to occur
if no other constituent is placed in P1 (P1 always has to be filled in Dutch
declaratives). In (21b) het simply does not fit. In fact, (24) is all right,
(24) Dat hij komt schijnt
albeit marginally, viz. with a strong contrastive focus reading on the complement.
The fact that (24) is somewhat difficult while (23b) is perfect is most probably
explicable in terms of another FG word order principle, viz. LIPOC ('language
independent preferred order of constituents'), which says that there is a tendency
to order constituents from left to right according to an increase in internal
complexity. In (24) the complement is structurally too heavy relative to the
verbal complex, and only strong focality can override LIPOC. In (23b) the verbal
complex is much heavier and therefore LIPOC does not work anymore.
Koster's argumentation against Dik's principle (iv) is fallacious too.
Consider the utterances in (25).
(25) a. Jan beloofde Piet dat hij geslagen zal worden

John promised Pete that he beaten will be
"John promised Pete that he will be beaten"
b. Jan beloofde Piet geslagen te zullen worden >
John promised Pete beaten to will be
"John promised Pete that he will be beaten"
c. Van beloofde Piet dat hij geslagen wordt
John promised Pete that he beaten is
A comparison of (22)/(25c) and (25a-b) shows that the reason why the former
are impossible may have to do with the tense of the complement (in the finite
as well as the infinite form). A promise obviously involves something in the
future. Hence (22) and (25c) are semantically anomalous because the object of
the promise is expressed with a present tense, suggesting that it is hic et nunc.
(On the complex relation between tense of the main predicate, tense of the
embedded predicate, and the moment of speaking in constructions like these,
see De Schutter 1981.) Once this anomaly is removed, as in (25a-b), the
utterances are fine. Hence, the problem in (22) has nothing to do with the
passive. In the causative in (16) it does, however.
3.1.3.2. One or more levels of syntactic functions?

Davies (1981) and Perlmutter (1981) have suggested that the Relational
Grammar (RG) multi-level concept of syntactic functions is superior to the
single-level approach in FG (cf. also Comrie 1979). Let's take a look at the most
important of their 'critical' cases.
One easy case is the matter of verb agreement in Achnese (an Austrone-
sian language) mentioned in Perlmutter (1981). His examples show that verb
agreement in that language is controlled by the same constituent in active and
corresponding passive constructions. Hence the crucial notion in this process is
not 'final 1' (the surface subject), but 'initial 1' (a subject at the first level of
syntactic representation in RG). Perlmutter claims this poses a problem for FG
since it cannot use its notion of subject (which corresponds to the final 1 in RG),
but cannot state the rule in terms of semantic functions either, unless at the cost
of generalization, for the 'initial 1' can have several different semantic functions,
including (in FG-terminology) agent, force, zero marking, and experiencer or
recipient. Yet, what Perlmutter seems to overlook is that the generalization can
be perfectly grasped in terms of the notion of 'first argument' (or of a case
covering all semantic functions occurring as first arguments - cf. the discussion
above concerning the need for deep semantic roles and more shallow case
categories; see also 3.2.3.3) which would then have to be better validated in FG
(but recently the notions of first and second argument do get a more central
status - cf. Dik 1989a: 232ff).
More difficult to handle is the behavior of 'inversion clauses' in Choctaw,
discussed in Davies (1981). Davies argues that these constructions can only be
accounted for if one assumes that the constituents have different syntactic
functions at different levels. Consider (26).
(26) Holisso-t am-ihaksi-tok

book-NOM 1DAT-forget-PST
"I forgot the book"
The 'subject' I determines dative agreement on the verb, 14 rather than

nominative agreement as would be expected from a normal subject. Hence

according to Davies it must be a 3 (an indirect object in RG) at some level of
analysis. Yet the same nominal gets nominative case marking when it is overtly
expressed.15 Cf. (27).
(27) Hattak-at i-mokoffa-li-tok

man-NOM 3DAT-let.loose-lNOM-PST
"The man dropped me"
Since normally constituents with 1hood are marked nominative (other constitu
ents are (optionally) marked objective), this is an indication that the first
nominal must be 1 at some level (the rule of nominative marking in RG is then
formulated in terms of 'heading a 1-arc', i.e. being 1 at some level, but not
necessarily at all levels). Similarly, the first nominal can antecede reflexives.
(28) Ili-chim-ihaksi-tok
REFL-2DAT-forget-PST
"You forgot yourself'
Otherwise it is obvious 1s which antecede reflexives in Choctaw, hence this is

another indication for the lhood of the first nominal (again, reflexivization is
formulated in terms of the process being controlled by a nominal heading a 1-
arc). This nominal cannot be a final 1, however, since it does not control a
number of processes which are typically related to final 1s. Thus, even if it is
coreferential with the subject of the main clause, it cannot be subject to equi-
NP-deletion in a subordinate clause. On the other hand, the other nominal in
the construction can be equi-victim, as is shown in (29).
(29) Is-s-mokoffa-ka-t ish-ikhana

2NOM-lDAT-let.loose-COMP-SS 2NOM-know
"You know I dropped you"
This other nominal is also marked nominative, as can be seen in (26). And it can
also control reflexivization. Thus, next to (28) there is also (30).
(30) Ish-ilim-ihaksi-tok
2NOM-REFL.DAT-forget-PST
"You forgot yourself'
These are reasons to assume that the other nominal in these constructions is also
a 1 at some level, and given the data from equi, more specifically the final 1.
Hence, in RG 'inversion clauses' get the representation in (31).
inversion nominal other nominal
According to Davies, FG could never provide an elegant account for these

cases, since it would have to refer to both syntactic and semantic functions, and
within the latter to several quite different functions. Yet it seems there is
another possibility for an account.16
This construction type, henceforth called 'experiencer construction' (EC),
is rather widespread in languages of the world,17 and it is systematically related
to a specific semantic domain (cf. McCawley 1976, Klaiman 1981): it expresses
situations in which an 'experiencer' is undergoing feelings, bodily or mental
perceptions, etc. It is a general characteristic that the experiencer constituent has
part of the properties normally ascribed to subjects (the others being inherent
to the other nominal), but the degree to which this is the case differs from
language to language (cf. Dik et al. 1983). Very often the EC has a counterpart
in which the constituent representing the individual behaves as a full-scale
subject, and in that case the individual is generally felt to be more actively
involved in having the feelings, perceptions, etc. as compared to the EC where
the individual is passively and non-volitionally undergoing them. ((28) and (30)
above seem to form such a pair. This might explain the different behavior, but
Davies does not provide enough information to be able to say much about this.)
There is a general tendency to directly link the notion of subject to (and
some even define it in terms of - cf. Keenan 1976) a whole series of behavioral
and coding properties (including the ones named above). In that view, if more
than one constituent has some of these properties, one faces the problem of
deciding which of them is the real subject. RG solves the problem simply by
assuming that each of them is subject somewhere in the derivation. An
alternative is to reconsider the link between subjecthood and these various
properties. One could assume that many of them can be determined, not by the
notion of subject directly, but rather by a variety of elements, including topicality,
definiteness, animacy, etc., which determine the 'prominence' of a constituent
(see Nichols et al. 1980) in the discourse domain. Probably one should
distinguish between types of properties in this respect. One would assume that
coding properties such as case marking and verb agreement are most directly
linked to subjecthood. But the behavioral properties normally mentioned with
respect to the EC, such as reflexivization, equi-NP-deletion, or switch-refer
ence,18 all involve matters of coreferentiality, and since this is an issue of
'retrievability' (see also 3.1.3.3) it is not unreasonable to suppose that they can
be controlled by a more general element of prominence. Of course, normally the
subject is most prominent in the discourse domain (since typically topical,
definite and animate), hence it is perfectly understandable that these properties
are normally inherent to the subject and can even become firmly attached to it
in the grammar of a language. Yet precisely in a construction such as the EC,
the arguments to deviate from this 'habit' are very strong.
We can assume that whereas in the normal 'active' counterpart of the EC
the individual is assigned first argument position and the feeling etc. is assigned
second argument position, in the EC the argument positions are opposite. This
reflects the semantic difference between the variants. In the EC the feeling,
perception, etc. is presented as a 'causative' (in De Schutter's 1989 terms - see
3.1.3.1), which is a dominant role relative to the experiencer, whereas in the
counterpart it is presented as an object (or a goal in FG terms), which is a
subordinate role relative to the experiencer. We also have reasons to assume
that in the EC the first argument (the constituent expressing the feeling etc.) is
subject. In fact, there is no language attested in the literature I have seen in
which the experiencer controls verb agreement. Only for Japanese and Choctaw
the experiencer is said to have the same case marking as a subject, although in
Choctaw the case marking on the verb is still different for subject and
experiencer. On the other hand, the other nominal nearly always has subject
marking and agreement. Moreover, assigning subject to the first argument is
always the 'unmarked' case according to the FG-theory of syntactic function
assignment, and since the feeling, perception, etc. is the thing which 'does it' in
the SoA it is also functionally appropriate to take it as the primary perspective.
However, the experiencer in the EC is normally animate (even human),
definite (in the overwhelming majority of examples of the EC in the literature
the experiencer constituent is a definite nominal, a proper noun, and - in most
cases - even a pronoun), and topical (in nearly 50 percent of the examples it is
first person singular). Hence, it is normally highly prominent, therefore highly
retrievable, and thus an excellent candidate to control processes which depend
on this factor. Moreover, it is perfectly natural that it very often occupies the P1
position (another property which appears from the literature). The degree to
which it attracts behavioral properties normally inherent to the subject then

depends on the degree to which a language has grammaticalized the link
between a certain property and the subject. In Choctaw, equi seems too strongly
associated to the subject proper to be taken over by the experiencer, but in
other languages this is different. Reflexivization, on the other hand, appears
generally more flexible, as there is no language attested in the literature
considered here in which the experiencer cannot control this phenomenon. (An
account similar to the above one can be given for reflexivization in Russian
discussed in Perlmutter 1981.)
Obviously, this is only a sketch of a solution. I will not go deeper into it,
also because the literature often provides too little information to be able to say
definite things (this is a rather general problem with typological literature). Still,
the lines of the argument seem clear enough to show that an FG type of
grammar, with one level of syntactic functions but with semantic and pragmatic
functions in addition, can handle the issue as well as RG, and moreover in a way
which does much more justice to the semantics of the construction than the RG
account.
3.1.3.3. Unordered predications and word ordering

Miller (1986) has argued that the FG option for unordered underlying
representations and for placement rather than transformation rules cannot be
upheld in the face of an analysis of a number of English constructions, unless at
a for FG unacceptable cost. His suggestions about FG are rather caricatural, but
he does bring up a number of interesting phenomena.
(i) Word order patterns and 'traffic rules'

Miller claims that the FG word ordering system, if systematically applied, leads
to an absurd multiplication of ordering templates (cf. (7) above) in a language.
Dik (1978a: 184ff) suggests that there are two basic ordering patterns for
English, viz. (32) for declaratives and (33) for interrogatives.
(32) P1 S Vf Vi O
(33) P1 Vf S Vi O
Miller claims this is insufficient. Consider 'particle movement' in (34).
(34) a. John gave back the book

b. John gave the book back
c. *John gave back it
d. John gave it back
According to him, to account for (34b) and (34d) we need (35), and for (34a)
and (34c) we need (36).
(35) P1 S Vf Vi O Part X
(36) P1 S Vf Vi Part 0[-Pro] X
(37) shows cases of 'particle movement' combined with 'dative shift'.
(37) a. John read off the figures to Mary

b. *John read off Mary the figures
c. John read Mary off the figures
d. John read Mary off the figures quickly
e. *John read Mary off quickly the figures
To account for (37b) (36) is insufficient. We would rather need (38). (35) is
insufficient for (37c-e): in order to account for the ungrammaticality of (37e), we
rather need something like (39).
(38) P1 S Vf Vi Part 0[-Pro/-Rec] X
(39) P1 S Vf Vi 0[+Rec] Part Go X
But then there are cases such as (40).
(40) a. Mary was given back the book by John

b. *Mary was given back it by John
c. Mary was given the book back by John
d. Mary was given it back by John
e. Mary was given back the book quickly by John
f. *Mary was given back quickly the book by John
g. Mary was given back the book she loved quickly
h. Mary was given back quickly the book she loved
In FG the book has no syntactic function in (40), hence Miller claims these
utterances cannot be handled in terms of (38) and (39) since the book does need
a more specific guidance for placement than just X. Thus, we need (41) and
(42).
(41) P1 S Vf Vi Part Go[-Pro] X
(42) P1 S Vf Vi Go Part X
At this point, Miller detects a 'traffic problem'. How can one determine
which pattern has to be used for a certain structure? For example, (39) should
not be used for (34d). FG will have to specify 'traffic rules' for this. Further
examples are provided by lexically determined differences in word order, as in
(43) and (44).
(43) a. John gave the book quickly back to Mary

b. John gave Mary quickly back the book
c. John read the figures quickly off to Mary
d. *John read Mary quickly off the figures
(44) a. John read the figures right off to Mary

b. *John read right off the figures to Mary
c. *John read Mary right off the figures
Apart from several new ordering templates, cases like these require 'traffic
rules', e.g. to avoid that (43a-b) would be treated like (43c-d) and vice versa.
Miller continues along these lines. If one adds cases with 'quantifier floating', e.g.
(replace John by all the boys in (43), and consider the alternative placement in
The boys all gave...), the number of templates doubles. It is doubled again if one
restarts the argument for interrogative clauses. Etc.
Of course, if one applies ordering templates this way there is an exponen
tial increase in their number. However, this is not what these templates are
intended for in FG. They should not be used to specify all possible positions for
all possible surface ordering patterns in a language, but only to specify its 'basic'
(unmarked) ordering pattern. It is a generalization over possible surface ordering
patterns. 1 9This template is only one factor among many in determining
surface ordering, and the ultimate ordering can even deviate from the template
if other factors dictate this. In fact, even though many of Miller's data are very
marginal, they are all easily explicable in terms of additional factors. In
particular, they are excellent to demonstrate once again the importance of
having pragmatic functions in a grammar (an element which Miller completely
neglects). The only change required in (32) is the specification that the template
can be interrupted between Vf and Vi, or just before Vf if there is no Vi (the
position has to be defined in terms of the main verb, and the definition has to
include allowance for an auxiliary to precede it). This is the position where, for
example, the quantifier goes in case of quantifier floating. Placement in this
position seems subject to conditions (see below), hence there may be a need to
make this a special position in the English template (something comparable to
P1).
One may be inclined to think that there is also a possibility to interrupt the
pattern between Vf/Vi and O, on the basis of cases such as (40a). Not so,
however: the examples in (45) (Miller 1986: 198, 205) are unacceptable precisely
because V and O cannot be interrupted in English.
(45) a. *John gave it the girl

b. *John read off the figures Mary
c. *John read the figures off Mary
That (40a) is possible is due to the fact that V and the verbal particle related
to it can be considered one functional unit. Hence, in neutral circumstances they
occupy the V position together. This is another illustration of the phenomenon
discussed with respect to the causative construction above, and it is an additional
factor to be taken into account in the placement rules.
Yet, the particle can sometimes be separated from the verb, as is shown
in (34b), (34d), (37c-d), (40c), etc. This may partly be due to the degree of
semantic cohesion between verb and particle. Intuitively, the semantic cohesion
in read off is stronger than in give back. The factors determining such differences
require further investigation. But it seems logical that the stronger the cohesion,
the harder it is to separate verb and particle. Anyway, they can only be
separated by O, or if there is no O (as in the passives in (40)) by the second
argument.20 For the rest, the relative placement of O (or argument) and
particle is determined by their topicality or focality (this factor may override
semantic cohesion). The topical element tends to be placed before the focal
element. That is why (34c) is out: a pronominalized element is always strongly
topical, hence it has to be placed as closely as possible to V. (34a), not (34b),
typically occurs in a context in which the book is strongly focal. Of course, in
(34b) the book can also be focal, but in that case it is an 'echo utterance' in
which the book contrasts to another entity which the previous S/H had inserted
in the same position in the same construction. These elements also explain (40b)
and (44a-b) (concerning (44c), see below). Right off is obviously one functional
entity, which is strongly focal in nature and is therefore placed after O.
Another element on which pragmatic functions have their effect is the
assignment of the object function to the second (in FG often goal) or third (in
FG often recipient) argument in a predication. The 'default' assignment is to the
second argument, yet 'dative shifting' is necessary if the third argument is too
topical relative to the goal (cf. Bock 1977, Creider 1979, Erteschik-Shir 1979,
Givn 1979). This fits into the general tendency noted above to order elements
in the post-verbal domain of the core-predication from the relatively topical to
the relatively focal. Thus, (46) (from Miller 1986: 198) is ungrammatical.
(46) *John gave the girl it
Similarly, (37b) is ungrammatical because Mary is topical, even too topical to

appear to the right of the verbal particle. Note that the above is a clear
empirical argument to support the suggestion that, contrary to the FG approach,
pragmatic functions should precede syntactic functions in a grammar.
Except in the special position before the main verb (see above), a satellite
cannot interfere between V and its arguments. Yet, deviance from this principle
is possible on the basis of LIPOC (cf. 3.1.3.1). Thus, (37e) and (40f) are
impossible, but (40h) is possible because of the complexity of the second
argument relative to the adverbial. The fact that (40e) is possible only supports
the assumption that the agent in a passive construction is no longer an argument
(cf. note 20). We are still facing (43). I have submitted these sentences to two
native and three near-native speakers, and all agreed that (43a-c) are certainly
not better than (43d) which, according to Miller, is ungrammatical. They agreed
that the utterances may be acceptable if quickly is interpreted as a modifier of
the verbal particle, thus constituting a functional unity with it similar to right off,
although in these examples this is, semantically at least, strange. But in that case
they do fit into the factors mentioned so far.
Finally, (37c-e), (43d), and (44c) require a completely different account.
My informants all agreed that these utterances can only get the somewhat
strange interpretation that 'Mary is something which is read off the figures by
John'. Hence, Mary is rather a goal argument (like a book in John read a book),
the figures is rather a satellite, and off is not a verbal particle but the preposition
of the satellite. If so, (37e) is impossible because the satellite quickly is placed
within the domain of the other satellite off the figures, which is obviously not
allowed. If one does find the suggested interpretation acceptable, (43d) and
(44c) are correct according to my informants. (44c) cannot be used to express
the meaning of (37a) because the other interpretation is too manifest.
There is much more to be said about these cases, yet the above is sufficient
to show that there is no need for a multiplication of ordering templates. The
'traffic problem' hardly exists either. If one accepts Dik's assumption 'that a
language can have a limited number of templates, one does need an indication
of which pattern should be applied in which case. But it is technically facile to
make the English patterns (32) and (33) sensitive to the illocutionary force
indicators declarative and interrogative respectively.
(ii) Deixis and 'ordering constraints'

According to Miller (1986: 207), FG, like TG, needs constraints on ordering
rules. Otherwise it will not be able to account for coreferentiality between nouns
and pronouns within one utterance, as shown in (47).
(47) a. When Johni wanted the book, hei told Mary

b. *Hei told Mary when John wanted the book
c. When hei wanted the book, Johni told Mary
d. Johni told Mary when he wanted the book
However, one can easily account for these cases on primarily semantic grounds,
through a principle of 'semantic recoverability' of the referent of a pronoun, and
without invoking a single syntactic constraint (cf. also B.Wiese 1983). Obviously,
in the default case the referent of a pronoun is the last mentioned entity in the
previous discourse which can logically be correlated with the pronoun, for
example on the basis of information provided by the pronoun on person, gender,
and number. 21 (Cf. Stenning 1978, Garrod and Sanford 1982, B.Wiese 1983.)
This immediately accounts for (47a) and (47d).
(47c) can be a problem for H/S if in the previous discourse there is a
potential referent for he. It is likely that in that case S/H will avoid this
construction for expressing the indicated coreferentiality. Otherwise, (47c) is no
problem (though somewhat marked) due to the fact that the pronoun occurs in
an embedded structure. On the basis of the initial when H/S knows that (s)he
is processing a temporal, embedded predication, and that the main predication
is still to follow. Hence, H/S knows that there may still be crucial information to
come for a final interpretation of the utterance, and the lack of an immediate
referent for the pronoun postpones its interpretation until the main clause has
been processed. There H/S will find a possible referent, and the problem is
solved.
In (47b) the situation is totally different. H/S has no indication whatsoever
that the referent may still appear later in the utterance, and therefore starts
looking for a referent in the foregoing discourse, before continuing to process
the utterance. Either H/S finds one, and misunderstands S/H (note that (47b)
is perfectly all right if S/H does refer to an entity in the foregoing discourse). Or
H/S does not find one and may start wondering whether (s)he has overlooked
something while, at the same time, the rest of the utterance and the following
discourse from S/H, which have not been processed yet, are overburdening
working memory (cf. G.Miller's 1956 hypothesis concerning the limited capacity
of STM). Hence, (47b) is not suited to communicate the coreferentiality relation
indicated. (The above nicely illustrates how the factor memory can be important
also in specific linguistic analyses - cf. 2.2.2.2.)
From the speaker perspective this means that cataphoric coreferentiality
can only be used if one can be certain that H/S will not start looking for a
referent in the previous discourse. That is normally not a problem discourse-

initially. Hence it is not abnormal to find a series of text-initial cataphoric
relations as in (48).
(48) He was born in A. He lived and worked in B. He played a crucial

role in C. And, mainly, he got famous by D. Yes, you have guessed
it right. We are talking about E.
This can obviously not go too far, for as long as H/S cannot determine a referent
(s)he cannot anchor new information in knowledge, and hence risks losing it. Of
course, a discourse like (48) precisely aims at making H/S find a referent in
knowledge. The information provided is assumed not to be new.
Within a running discourse the possibilities for an S/H to make cataphoric
relations are much more limited, since the risk that unwanted potential referents
are already present is much higher. Anyway, as with anaphoric coreferentiality
S/H has to respect a principle of unique identifiability. If that is impossible,
cataphoric reference is impossible. If it is possible, S/H can introduce a
cataphoric relation within the domain of a complex utterance (discourse-internal
cataphors across utterances seem quite unlikely), provided there are sufficient
indications that the referent of the cataphor is to follow. What 'sufficient' means
should be further investigated. In view of (47), it does include the presence of
the cataphor in an explicitly marked embedded clause. For a discussion of the
reasons an S/H may have to choose a cataphoric rather than the less marked
anaphoric reference, see Maes (1987).
Cases like (49) are also due to coreferentiality properties.
(49) a. John saw himself

b. *Himself was seen by John
c. John was seen by himself
Miller (1986: 198) claims that reflexive constructions cannot be passivized, but
(49c) shows he is not right. This construction is somewhat strange, but that is
due to the functionality of passivization and the nature of the SoA expressed in
(49). Subject being the primary vantage point from which an SoA is presented,
the only candidate for subject marking in the SoA underlying (49) is John (he is
the only entity). Hence, very specific conditions are required to make it useful
to assign subject function to this entity in the second argument position rather
than in the first argument position. Yet (49c) is grammatical, but (49b) is not,
and the explanation is simply that reflexives can only refer anaphorically, not
cataphorically. What could be the functional relevance of cataphorically referring
within a simple predication?

Since coreferentiality appears to be a semantic phenomenon, the syntactic
behavior of the elements involved cannot be understood without considering that
fact. Syntactic processes appear to be very 'intelligent' processes which take
information from many conceptual (semantic) sources into consideration. Of
course, this means that a full description of the use of pronouns in a grammar
is not possible without having a representation of these semantic-conceptual
factors and, since their impact appears to stretch out over several utterances,
also of the level of discourse organization available. Even a functionally oriented
grammar-in-vacuo like FG will not do for that purpose. Moreover, if cataphoric
and anaphoric relations are co-conditioned by the informational status of
elements (and hence by pragmatic functions - Maes 1987), then this is another
indication that pragmatic functions have to be present much deeper in a
grammar than is granted in FG. We then have to assume that the latter co-
determine the choice between nouns and pronouns as term slot fillers. Since
transformational replacement of such elements later in the grammar is
impossible in FG, this implies that the pragmatic functions have to operate not
only before the syntactic function assignment, but even at a pre-predicational
level. That is, at a level which is completely lacking in current FG. Hence, an
integrative conception of modeling the cognitive systems for language use is not
only theoretically necessary, as argued in 1.1, but also appears indispensable for
the praxis of linguistic analysis even in the syntactic domain.
(iii) 'Floating' elements and 'holds'

Another mechanism which FG will have to introduce according to Miller (1986:
208ff) is that of 'holds'. He assumes that FG needs a principle of cyclical
ordering which involves a strict succession of ordering procedures, starting from
terms, via embedded predications and ending at the level of the entire utterance.
If one strictly applies the FG principle that once an element has been placed it
cannot be replaced anymore, cyclical application of ordering principles requires
a system which holds constituents placed outside their own cyclical domain.
There should even be different types of holds. For example, consider quantifier
floating (Miller 1986: 209).
(50) a. John has said that all the boys have come
b. John has said that the boys have all come
c. *John has all said that the boys have come
Miller suggests that we need an 'all-hold' for this, which can only work when the
term which is being ordered has subject function, and which has to be emptied
at the next higher cyclical level. Or consider wh-questions.
(51) Who did John ask Peter to tell Mary to see?
Here we need a 'Q-hold', which can only be applied once in one complex
utterance, and which can only be emptied at the highest cyclical level.
According to Miller, raising involves a very complex hold-problem. He
starts from the analysis of raising in Dik (1979a). Dik proposes to allow subject
and object markings in main clauses containing certain predicates (such as
believe) to 'penetrate' the embedded predication and to be assigned to the terms
there (which can already have a syntactic function marking from their own level),
according to a number of principles. A term with two syntactic functions can
then trigger processes at both levels, at each level through the specific marking
it received from this level.
Miller criticizes this approach by claiming that it can lead to an absurd
multiplication of function assignments to one term. Consider (52) (Miller 1986:
210).
(52) a. The boy was believed by John to believe Bill to have killed the
farmer
b. Bill was believed by John to be believed by the boy to have
killed the farmer
c. The farmer was believed by John to be believed by the boy to
have been killed by Bill
These structures do indeed lead to a very complex pattern of syntactic function

assignment, sometimes even leading to three markings on one term (the details
do not matter here). Obviously, the complexity increases with each further
embedding with a similar predicate (the possible number of function markings
to one term being equal to the number of embeddings). Miller claims that to be
able to order these utterances FG needs an 'R-hold' which carries the term to
be raised until it reaches the appropriate level, and which has a very intricate
internal structure (Miller 1986: 211ff). Moreover, if raising and quantifier
floating are combined, as in (53) (Miller 1986: 213),
(53) a. John believed the boy to believe the ducklings to have all been
killed by Bill
b. John believed the ducklings to be all believed by the boy to
have been killed by Bill
c. John believed the ducklings to be believed by the boy to have
all been killed by Bill
d. The ducklings were believed by John to be all believed by the

boy to have been killed by Bill
it appears that the 'all-hold' and the 'R-hold' will have to enter into a complex
interaction, for all is left behind at a lower level than the rest of the term. For
all the above cases, Miller states that it becomes impossible to grasp the
meaning in terms of perspectivization by the S/H of the 'baroque' function
assignments which are required. Hence, they cannot be more than an ad hoc
way to code raising and passivization.
Miller is correct in stating that FG needs a specific type of mechanism in
the expression component to handle the cases mentioned. However, contrary to
what he suggests it is not very complex, and it perfectly fits the main tenets of
FG syntax. First, Miller's suggestion concerning the cyclical application of
ordering rules in FG has to be modified. It is true that rules have to be defined
in terms of the type of domain to which they apply, and that sets of rules should
be specialized for different domains at different hierarchical levels. But there is
no reason to assume that the application of the 'experts' has to be strictly
cyclically ordered. On the contrary, it is generally acknowledged that human
cognitive activity strongly relies on parallel processing (cf. Johnson-Laird 1983:
452ff, Levelt 1989), and it seems perfectly logical to assume the same for
syntactic procedures in particular. In fact, such an assumption would allow a
much better account of the mutual influence of ordering decisions at different
levels or within different domains at the same level.
For the treatment of the above cases, then, we not only need 'intra-level
procedures' (ILP) responsible for the ordering within a specific domain, but also
'super-level procedures' (SLP) responsible for the interaction between domains
and levels. The operational principles for the SLPs are simple: elements can only
be taken out of their domain if there is a clear functional reason (e.g. to allow
a better expression of strong focality of the element), if their placement in a
higher domain can satisfy this special functional status, and if the operation does
not obscure their semantic position in the underlying predication.
Let's take a look at how this operates in quantifier floating. At least one
of the aims of this phenomenon is to allow the S/H to more clearly express a
difference in topicality or focality of the quantifier, relative to the rest of the
term, than would be possible through intonation alone. (Cf. Bouma 1987 for a
more detailed discussion of the functionality of the phenomenon.) In the default
case the quantifier is much more focal than the rest of the term. The pre-main-
verb position in English declaratives mentioned above allows a clear expression
of this value. P1 might be suitable too, yet in case of a placement there the
difference with the non-floated quantifier would hardly be noticeable. Formally,
the SLP could operate as follows. The control system supervising the procedures
handling the term in which the quantifier occurs notices the strong focal value
of the quantifier relative to the rest of the term, and asks the control system
one-higher whether it has facilities to place the quantifier within that domain. If
so, an SLP takes the quantifier out of its domain and puts it into the relevant
slot in the higher domain. This SLP can take the quantifier upward no more
than one domain because otherwise its semantic position in the predication
would not be transparent anymore. Probably the same factor is responsible for
the fact that the SLP can only take a quantifier out of a subject term, at least
in English. If the object would also allow quantifier floating, this might cause
semantic ambiguity in case both subject and object are countable terms. In fact,
that is what happens in Dutch, which does allow quantifiers to float out of
objects.
(54) De meisjes hebben de jongens allemaal gezien

The girls have the boys all seen
"(All?) the girls have seen (all?) the boys"
A Dutch speaker will most likely avoid such an utterance if the context does not
provide sufficient clues to disambiguate it. This is a nice illustration of how
languages can differ in the way functional factors shape the grammatical system.
'Wh-placement' is similar. A wh-word is per definition focal, and the
control system for an embedded predication containing one automatically
appeals to higher control systems whether or not they can place it in a
pragmatically salient position, viz. P1. The request is automatically passed on to
the highest level, because all intermediate levels already have P1 fillers (all
intermediate levels are embedded clauses, and they have subordinators which
have to occupy P1). Semantic transparency is never in danger, since it is always
clear which embedded predication is missing a term slot. If the highest control
system agrees, an SLP does the job. Yet, the highest control system can also
refuse the wh-word. It may do so e.g. if the utterance echoes a previous one
(uttered by someone else) in which the S/H has not well understood a term. This
refusal is then compensated for by assigning main stress to the wh-word, as in
(55).
(55) John asked Peter to see WHO?
The highest control system may also refuse if there are several wh-words eligible
for placement in P1. In such cases it can accept one if there is a very salient one,
or it can refuse all of them. It can never accept two since there is only one P1
position, and no other main clause position can take a wh-word (for some reason
the pre-main-verb position cannot).
Finally, let's consider the raising cases. Next to Dik's (1979a) analysis, other
treatments of raising cases in FG are to be found in Bolkestein (1980, 1981a, b)
and De Groot (1981b). They differ from Dik's in the type of factor assumed to
cause the process. Rather than syntactic functions, they involve constellations of
pragmatic functions.22 This is not necessarily incompatible with the essence of
Dik's treatment, however. There must be a reason why main clause syntactic
functions are sometimes assigned to terms of the embedded clause and
sometimes not. This may have to do with the relative degree of topicality and/or
focality of the terms in the embedded clause. For instance, if one term is topical
whereas the other is focal, the topical term may receive subject assignment and
be lifted out of the embedded domain to increase the distance from the focal
part of the embedded clause. Or if the second argument in the embedded clause
has a strong contrastive focus, whereas the first argument is topical, the latter
may be demoted by passivization, which then allows the former to be placed in
the object-position of the main clause, where its pragmatic status finds better
expression.23 The procedure may then even be that first the embedded
element is 'raised' under influence of the pragmatic function marking, and that
it receives syntactic function assignment in the main clause only afterwards.
In any case, Miller's examples are not likely to cast doubt upon the validity
of the original proposal by Dik. First of all, one does not have to be astonished
that the structures discussed by Miller require 'baroque' function assignments in
which the notion of perspectivization entirely loses its meaning. Such assignments
perfectly explain why these equally 'baroque' utterances are extremely hard to
understand, and will hardly ever (if ever) be produced by anyone but (TG)-
linguists. In fact, a small test with the informants mentioned above showed that
all these structures (and those in (52) even more than those in (51)) are very
dubious, if not downright unacceptable. Each of the structures was characterized
as impossible at least once, and the informants were often unable to determine
whether they were okay or not, especially for (52). Hence, a model which would
provide an elegant solution for processing these cases would be very suspicious.
In more normal cases of subject and object raising, however, there is no
problem in understanding this in terms of perspectivization dependent on the
status of parts of the information in the embedded and the main clause. The
procedure can again be performed through an SLP. If we stick to the scenario
proposed by Dik in which the syntactic function assignment penetrates the
embedded clause, it could run as follows. The individual control systems at the
different levels know whether there has been subject and object marking at their
level, and which markings derive from their level and which not. If the main
clause control system notices that it lacks one of its markings, it will inquire
whether a lower control system has it. Alternatively, if an embedded control
system notices that one of its terms carries an 'alien' marking, it will inform the
higher control system(s) about this. The embedded control system is obliged to
hand over the term with the higher marking to the SLP, but before letting go it
registers all relevant facts about it so as to be able to normally organize its own
level. If a term has markings from several levels, as in Miller's examples, each
level from which a marking derives gets the term for a moment, so as to allow
it to register the data necessary for organizing that level. Ultimately, the SLP has
to bring the term to the highest control system from which a marking derives,
which will fully integrate it in the organization of its level.
I will not even try to sketch the interaction of raising and quantifier floating
as given in (52). It is not clear at all which of these structures are really possible.
The fact that after some puzzling one does manage to interpret them and to
semantically locate the quantifier is only due to the fact that there is only one
plural term in the constructions. Hence, these are cases where one can reach an
understanding of an utterance on the basis of purely semantic reasoning (cf.
3.2.2). Just make the boy plural, and one is at a total loss.
Hence, the 'holds' (if one wants to call them like that) are not a problem
for the FG expression system, if it is only conceived as a sufficiently 'intelligent'
system. In fact, the above is a nice illustration of the importance, also for the
practice of linguistic description, of the notion of control introduced in 2.2.2.2.
The presentation of the mechanisms was, of course, only sketchy, but it was
sufficient to demonstrate that the type of control system required in modeling
the human syntactic processor is probably similar to what in AI is known as
'blackboard control' (cf. Hayes-Roth 1985 - see also 3.2.2).
To summarize, the above case studies nicely illustrate the argumentation
in ch.2 that Chomsky's claim that semantics cannot play any role in syntax is
aprioristic and unmaintainable. It makes it impossible for formal linguists to look
for types of functionally motivated explanations such as the above, yet such
explanations not only prove to be perfectly possible (thus also showing that
semantics does play a role in syntax) but also indispensable for an appropriate
treatment of data. They also show that an FG-type of syntax, including the
different function categories, is rather promising, although adaptations and a
serious in-depth expansion of the model as it now stands seem unavoidable.
3.2. An integrated view of the systems for language processing
3.2.1. Language processing and general knowledge
3.2.1.1. Types of knowledge underlying language processing

An FG-type of grammar does seem to have quite some cognitive plausibility.
Still, it only involves a lexicon and a syntax, and both principled concerns (cf.
1.1) and factual needs (cf. 3.1.3) lead to the question of how it might fit into a
more encompassing model of the cognitive systems involved in language
behavior. One crucial aspect of this concerns the problem of which general
knowledge systems can be assumed to underlie linguistic behavior and what their
nature and organization might be. As argued in 1.2, since language is a medium
to achieve communicative goals, an S/H must control several factors in order to
be able to construe a functionally accurate utterance or to adequately interpret
an utterance. Hence, apart from a representation of the systems for linguistic
processing per se, an encompassing model of language behavior also has to
include a representation of the conceptual knowledge systems and mechanisms
involved in these factors, and has to show how the language systems are rooted
in, and how their use is oriented and directed by, these general systems. In fact,
a full-fledged hypothesis of the systems for language processing proper crucially
depends on what one assumes to be the nature and organization of these
general systems, since that determines specifically which types of mechanisms
have to be postulated for linking conceptual and linguistic representations. Alas,
although in the last decades there has been considerable research into the
structure and organization of general knowledge and the mechanisms operating
it (reasoning, problem solving, inferencing, etc.), especially in cognitive
psychology and AI, there are still very few firmly established hypotheses about
it. That is no doubt due in part to the fact that this is a relatively young topic of
investigation, and in part to the vastness and complexity of the problems
involved, as will become clear below.24 Let's see what our requirements are,
first of all with respect to the sorts of knowledge systems and their relationships,
particularly in view of the analysis in 1.2.
The factors distinguished in 1.2 seem to require at least three general
domains of knowledge, viz.
(i) direct and indirect knowledge concerning the SoA the interaction is about (cf.
the factor 'frame of reference'),
(ii) knowledge about the relationship between S/H and the interaction
partner(s), and about the social conventions and personal involvements related
to this (cf. the factor 'social relationship'), and
(iii) knowledge about the specific properties of the setting of the communication
and the conventions which relate to it (cf. the factor 'setting').
Yet it would be too simplistic to assume that these simply constitute three
separate 'components' of general knowledge. As mentioned in 1.2, elements of
(ii) and (iii) can also become part of (i), which suggests that all three types of
knowledge are intimately related and somehow form a part of a large knowledge
store from which information can be drawn for different purposes.
This assumption obviously requires further modification. The conceptual
knowledge of an S/H in a communicative situation should be further divided into
situation-dependent and situation-independent knowledge. The latter involves
S/H's more or less permanent knowledge of the world (or rather of all
imaginable worlds) acquired during his/her lifetime, which is gathered in what
we may call the 'encyclopedia' (a long-term memory). Thus this includes a vast
variety of things: knowledge about the physical world (e.g. knowledge of the
biological taxonomy of animals, or of cars) and the social world (e.g. knowledge
about the role of a prime minister in a democracy, or about the conventions for
social behavior within or among social groups); from general (as in the above
examples) to specific (e.g. knowledge about individuals: who am I, what are my
social and personal traits, and who is my neighbor, what are her habits and what
does she do on Sunday mornings in her garden); experiential knowledge (what
did I do on the day of my promotion) and 'directive' knowledge (what do you
do when you go to a restaurant); etc. As mentioned in 2.2.3.3, this can involve
both 'knowledge that' (factual knowledge: that a sparrow is a bird, that xyz is my
boss, and that the rules of chess are such and such) and 'knowledge how'
(directions for action: what to do in a restaurant, how to behave toward the king
of Belgium, how to play chess), whatever the status of this difference may be.
How this huge knowledge store is structured and organized and how the chunks
of information are accessed is largely an open matter. Theoretical constructs
such as scripts, frames, scenes, etc. (cf. e.g. Minsky 1975, Wilks 1976, Schank and
Abelson 1977, Schank 1980, J.Mandler 1984) are very likely to figure prominent
ly in any account of this, but they are still a far cry from a full theory of
knowledge representation.
Situation-dependent knowledge can also be any of the above types, but it
concerns matters which the S/H perceives or derives (linguistically or otherwise)
from the actual communicative environment. This knowledge is stored in some
temporary memory buffer. Part of it is retained only as long as it is relevant for
the current situation, but part of it can get included in the long-term situation-
independent knowledge store, according to certain principles of selection.
Probably there is flexibility in the length of retainment of information in the
temporary buffer, which should then not be too short-term. One can assume that
situation-dependent knowledge is always compared to situation-independent
knowledge to check whether it is compatible (or probably already present)
and/or relevant etc. and that this determines whether information is retained or
not, for how long, and whether it is transferred to the encyclopedia or not.
Knowledge relevant for S/H in the communicative situation normally
involves both the situation-dependent knowledge25 and those aspects of the
encyclopedia which are related to the factors discussed in 1.2 as applied to the
actual conversation. That is, it involves chunks of permanent knowledge related
to the relevant SoA, the social and personal relationship with H/S and the social
conventions applying to it, and the general properties related to the current
setting and its consequences for the participants. One other important
knowledge component involved here is S/H's hypothesis about H/S's conception
of the different factors (including H/S's intentions, and the non-functional
element of personality and momentary condition) involved in the communica
tion, including a hypothesis about H/S's hypothesis about S/H's conception of the
factors.26 This constellation may be called the 'partner model'.27 This knowl
edge is partly a matter of the encyclopedia (general knowledge S/H has about
H/S, if any), but mainly a matter of situation-dependent knowledge. It is likely
that the partner model is formulated, at least in part, in terms of conformity and
conflict with the values of the factors in S/H's own knowledge.
The entire stock of situation-dependent and -independent knowledge which
becomes relevant for S/H in a particular situation will be called S/H's 'universe
of interpretation' (UI). The UI is constantly changing, in that new bits of
situation-dependent knowledge are permanently entering, which can also activate
new elements (elements which were previously not relevant, or were over-
looked)28 or deactivate elements in the encyclopedia. Moreover, knowledge
in general (and thus also the UI) can constantly change through the operation
of reasoning procedures. The exact status of the UI as a structure is not a priori
evident. It could be a separate memory structure in which the relevant situation-
dependent and -independent knowledge is copied. Alternatively, it could be a
matter of activation of the relevant knowledge in the stores for situation-
dependent and -independent knowledge. The latter would seem more economi
cal, but probably there are other arguments to be considered. This requires
further investigation.
So far we have been talking about systems of general knowledge underlying
language use, but we have not yet said anything about the factor of intention and
the non-functional element of personality and momentary condition of S/H. We
did say something about the intention and personality elements of H/S as part
of the partner model, which are elements of S/H's general knowledge. S/H can
also have knowledge about his/her own intentions, personality, momentary
condition, etc. In this way these elements can also become part of the frame of
reference of the communication (cf. 1.2). But that involves an interpretation by
S/H of S/H's real intentions, personality etc. One should clearly distinguish
between having a certain intention, having a personality, being in an emotional
state, etc, and knowing about having an intention, a personality, an emotional
state, etc. (as is apparent from the fact that people can be seriously mistaken in
their knowledge about themselves). (Cf. Nuyts 1990a.) It is only in the former
'state' that these elements play their separate role in the communicative situation
as discussed in 1.2, and as such they cannot be said to be elements of knowl
edge, at least not in a simple way.
An account of these phenomena is likely to require recourse to the notion
of a control system (cf. 2.2.2.2). Intention may be a matter of the central control
system (which may also be taken to be the center of consciousness) being
attracted by specific chunks of knowledge (general or with respect to the current
situation) which for some reason do not conform with their informational
environment (e.g. in that they contradict other chunks of information, in that
they are vague or incomplete, in that they are not in line with S/H's picture of
an 'ideal' SoA, etc. - cf. also 3.2.1.3) and therefore stimulate the control system
to take action so as to remedy the discrepancy. (See also Johnson-Laird 1983:
ch.16 on the relationship between intentionality, consciousness, and control.)
It is much harder to understand personality, emotional state, momentary
condition, etc. in terms of cognitive systems. Momentary condition (being tired,
inattentive, etc.) may be a matter of fluctuations in the efficiency of the control
system in coordinating actions at several levels of processing. Emotional states
are probably due to strong meta-attitudes an S/H has with respect to knowledge
about a certain SoA. This SoA can be so salient to S/H (probably because the
meta-attitude is extremely strong), and can thus be so persistently present in
consciousness, that the meta-attitude exerts influence on any action taken by
S/H, whether it is related to the SoA it belongs to or not. The expression of an
emotional state is obviously a matter of this state influencing the way the control
system operates. Personality is likely to involve very many things, including ways
the control system is used to react to information (knowledge about SoAs) and
to direct actions to be taken, but also types of knowledge the S/H has gathered
in life, types of emotional states which predominate, etc. There is obviously a
very long road to travel toward seriously understanding these phenomena.
3.2.1.2. The nature of conceptual knowledge

An issue we have not touched upon thus far is the question of the nature of this
general, conceptual knowledge. In which format is it represented? Contrary to
the prevailing view in language psychology, AI, and linguistic semantics,
according to which it has a non-verbal format29 (which does not mean that
there is consensus on further issues concerning its format), Dik (1987b, c, d,
1988, 1989b, 1990) has argued that it is not only verbally represented, but even
has the format of the underlying predications (i.e. predicate frames with the
terms inserted in them) postulated in FG. This proposal is obviously inspired
by Dik's general inclination toward avoiding abstractness in linguistic description
(cf. 3.1.1). His argumentation runs as follows.
The language user's linguistic capacity, epistemological capacity (i.e.
general knowledge as discussed above), and logical capacity (the ability to reason
with this knowledge) each require some kind of symbolic representation, a
'language': linguistic utterances are generally assumed to be based on more
abstract underlying representations (L ur ), conceptual knowledge must be coded
in some kind of symbolic system (L ck ), and the reasoning processes depend on
a logical calculus which requires some logical syntax (Llog)- Of course, one might
assume that these three 'languages' are different. However, it is much more
attractive to suppose (56).
(56) There is one general cognitive representation language L , such

that L cog = L ur = L ck = L log
(Dik 1989b). Let's call this the 'cognitive simplicity postulate'. If FG-predications
are taken to be plausible hypotheses concerning L ur , then, one can conclude that
(57) Lcog = predications
Thus, the representation of the language user's knowledge of the SoA involved
in (58a) would be identical to the FG-predication underlying this utterance,
(58b).
(58) a. Mary gives an apple to John

b. give v (dlx 1 : Mary pN (x 1 )) Ag (ilx2: apple N (x 2 )) Go
(dlx 3 : John pN (x 3 )) Rec
According to Dik this assumption is further supported by the following

considerations.
(i) By accepting (57) there is no need for a very intricate interface which in the
alternative view would be required to relate the levels of linguistic and non-
linguistic representation (Dik 1989b).

(ii) Since humans already possess a representational system with endless pos
sibilities, viz. natural language, why should they not make use of this system for
conceptualization? Why should they develop 'shadow-expressions' in some other
representational language? (Dik 1987d, 1989b.)
(iii) Those claiming that conceptualization is non-linguistic seem not to be able
to avoid using lexical labels for identifying concepts when illustrating their
models in schemas etc. They often put the words in capitals and explicitly
mention that they should not be confused with real natural language words. But
where are the non-linguistic representations, then? (Dik 1987d, 1989b.)
(iv) If conceptualization is linguistic, knowledge differs with the particular
language the individual knows, while the claim as to the non-linguistic nature of
conceptualization usually implies that conceptualization is much more universal.
Yet, if the latter were the case, why is it that languages are so different? Why
should they cover rather than reveal the conceptual structures? (Dik 1987d.)
(v) If conceptualization is language-dependent (an obvious consequence of Dik's
view, since predications, though caught in a universally valid syntax, are filled
with and structured depending on language-specific lexical items), S/Hs of dif
ferent languages can hardly ever be said to have the same knowledge. But this
corresponds to the fact that knowledge is strongly culture-bound; and it explains
why translatability between languages is hardly ever complete. (Dik 1987d,
1989b.)
Next to conceptual representations, Dik (1987c, 1988, 1990) also postulates
the existence of 'perceptual representations' or 'images', which are the result of
perceptual processes. They are introduced in the framework of the 'semantic
component' of the Functional Logic (FL) which is put forward as embodying the
reasoning capacity. Expressions input to FL (i.e. knowledge in the form of
predications) are to be interpreted in terms of a system of cognitive representa
tions, called 'pictures', which consist of a mixture of conceptual representations
(having predicational format again) and perceptual representations (having a
non-linguistic format). The exact format of these pictures, and ipso facto of the
images, remains open, however.
There are various reasons to doubt the appropriateness of hypothesis (57),
though. Let's start by considering some linguistic facts from the perspective of
FG. Consider the phenomenon of layering. As mentioned in 3.1.1, each layer can
bear operators and satellites which somehow qualify the SoA or its expression.
Crucial for the present argument is the fact that at each level, operators and
satellites can represent identical qualifications (e.g. temporal, modal, or
aspectual qualifications), the difference being that operators occur if the
qualification gets expression by grammatical means, and satellites if the

qualification gets lexical expression. A possible hypothesis is that this difference
is determined by the degree of specificity of the qualification and/or of its
communicative saliency. As a general principle, one could assume that for each
qualification, all values are expressible by lexical means, while only some rough
distinctions are grammaticalized in a language (although the degree of gram-
maticalization is language-specific, of course).
Consider time deixis. The tense operator only allows expression of some
rough temporal categories (in English, e.g., past, present, future, relative to the
moment of speaking or relative to some moment in the past or in the future).
If S/H wants to give any detail about the temporal situation of an event, (s)he
has to use lexical means. It is very unlikely that there is any language allowing
a grammatical coding of, say, 'in 1585', or 'on the Monday before we first
met'.31 The categories involved in grammatical tense marking can also be
expressed by means of a satellite, and the decision to do so seems determined
by communicative considerations (focalizing the temporal element is much
easier, or even only possible if it is expressed lexically).
Now, one of the consequences of the 'no transformations principle' (cf. 3.1)
is that FG also has to adhere to what we may call a 'lexical constancy principle',
formulated in (59).
(59) after determining the predication in the fund of the grammar no

changes are allowed in the lexical setup of an utterance
Hence, it is appropriate to have the division between satellites and operators at

the level of the predication. Still, one might wish to have some deeper level in
the grammar at which each qualification can be treated as the coherent concep
tual phenomenon it appears to be, irrespective of its varying expression forms.
This level should be linked to the level of predications by means of decision
procedures sensitive to the specificity and informational saliency of the qualifica
tion. (See also 3.3.)
Not only the level of the layers, but also the 'basic' level of the 'core'
predication expressing SoAs provides arguments against (57). A consequence of
the claim as to the predicational nature of conceptual knowledge (one which Dik
1987b explicitly accepts) is that, even among S/Hs of a single language, what
appears to be one SoA can be represented in different ways. For example,
knowledge of the fact that a person named Johnson has 35 people working for
him can be represented at least in terms of the predications (which are given
here in a strongly simplified form) presented in (60) (examples taken from Dik
1987b).
(60) a. Johnson has 35 employees

have (Johnson) (35 employees)
b. Johnson employs 35 people
employ (Johnson) (35 employees)
c. There are 35 people working for Johnson
(there) (work (35 people) (for Johnson))
d. Johnson gives work to 35 people
give (Johnson) (work) (to 35 people)
Four individuals can each hold a different one of these. One can say that they
do have the same knowledge indirectly, since the language user can derive each
of these predications from each of the other ones by means of the FL. This
proposal is strongly counterintuitive, however, and even from a strict FG-
perspective it is theoretically questionable.
Intuitively, (60a-d) do seem to represent one and the same SoA, the dif
ferences between the variants being a matter of alternative presentation of this
SoA due to contextual factors. In fact, the relationship between (60a-d) is akin
to the relationship between (60b) and (61).
(61) 35 people are employed by Johnson
FG does acknowledge that (60b) and (61) have one underlying predication and
only differ in terms of pragmatic function assignment. Why should this not be
the case for (60a-d) too?
This point is more clearly seen in (62) ((62a) = (58)).
(62) a. Mary gives an apple to John

give v (dlx1: MarypN (x1))Ag (ilx2: apple N (x 2 )) Go
(dlx 3 : John pN (x 3 )) Rec
b. ?(To) John is given an apple by Mary
(?)To John an apple is given by Mary
(?)An apple is given to John by Mary
c. John gets an apple from Mary
get v (dlx 1 : John pN (x 1 )) Ben (ilx2: apple N (x 2 )) Go
(dlx 3 : Mary pN (x 3 )) So(?)
d. ??From Mary an apple is gotten by John
??An apple is gotten from Mary by John
??An apple is gotten by John from Mary
(62a) and (62c) have different predicational representations in FG, and, hence,
would be different pieces of knowledge, while (62a) and (62b), or (62c) and
(62d) are distinguished in the expression component, and thus would represent
the same piece of knowledge. But the fact that passive variants (62b) (certainly
the one closest in word order to (62c)) of (62a) sound somewhat unusual, and
that passive variants (62d) of (62c) are practically excluded, can be explained by
the fact that (62a) and (62c) already show a relationship very similar to active-
passive variation. Both (62b) and (62c) are ways to avoid giving the agent/source
in the SoA (viz. 'Mary') primary perspective (if necessary it can be left implicit
in both cases), and apparently, S/Hs prefer the lexically coded possibility (62c)
over the grammatical operation in (62b). The fact that (62b) remains acceptable
while (62d) is nearly impossible is no doubt due to the fact that it still makes
sense to passivize a normal active construction like (62a), while it would be
rather paradoxical to further passivize an already passive-like construction like
(62c). What would it mean, functionally, to have double passivization?
Another, similar case is (63) (examples from Foley and Van Valin 1984:
57).
(63) a. Peter stole the valuable book from Sam

b. Peter stole Sam the valuable book
c. Peter robbed Sam of the valuable book
d. *Peter robbed the valuable book from Sam
(63b) is not synonymous with (63a). Hence, (63a) cannot undergo normal 'dative
shift', which, as mentioned in 3.1.3.3, is a generally acknowledged means to de-
focalize or de-perspectivize the recipient in the SoA. Yet, (63c) precisely gives
the effect which dative shift of (63a) might have given. (63d) is impossible
because it would involve another dative shift, or rather an 'undoing of the shift
ing' involved in using (63c) instead of (63a). If no dative-shift is required, the
S/H will choose the less marked variant (63a). Hence, the same functional effect
frequently achieved by means of alternative function assignment and expression
rules in one predication is achieved by using different predications in (63a) and
(63c). (Similar examples from Latin are given in Bolkestein and Risselada 1987.)
Basically the same observation can be made with special constructions such
as clefts, pseudo-clefts, 'right-dislocations', etc. These are all generally accepted
to be marked variants of more neutral utterances, expressing the same basic SoA
in different ways due to marked informational conditions. But they have different
predicational representations in FG (cf. e.g. Dik 1980a: 210ff), and hence,
according to (57), would represent different conceptualizations.
This leads to the following objections:
(i) If four different individuals each can know one of the predicational
alternatives in (60), it seems unavoidable to accept that one individual can know
all four of them, too. The fact that these representations might be derivable
from each other by means of reasoning rules is no argument in se, since results
of reasoning processes can be stored in themselves. Thus, Dik's proposal implies
an enormous redundancy in human information storage, which from a cognitive
viewpoint is a very unattractive claim.
(ii) It is undesirable to accept the high degree of situation specificity of
conceptual knowledge representation introduced by this proposal. Certainly,
people do acquire knowledge in specific situations, but in order to make it avail
able for use in as many different situations as possible, it is necessary to store
it in a situation-unbound way. Otherwise, transfer of the information to
situations different from those in which it was acquired would be difficult. Dik
(1987d: 149) himself mentions among the "good reasons [...] for assuming that
knowledge representation structures are more abstract than the actual linguistic
expressions of any one language", the fact that "some linguistic expressions are
synonymous in the sense of embodying the same knowledge, presented in
different ways". He refers to active-passive variants, in which "the differences of
presentation may be contextually motivated, but they are not essential to the
knowledge of what happened". Apparently, precisely the same is the case in the
examples above.
(iii) It is unattractive to handle the relationship between variants like those in
(60) in a logic, on a par with real syllogistic reasoning in which different SoAs
are combined in order to draw conclusions about other SoAs. By doing this one
overburdens a system with two entirely different matters. At the same time, it
is illogical to treat distinctions such as the active-passive alternation in the
grammar properly, and to treat basically similar distinctions such as those
between the utterances in (60) or between (62/63a) and (62/63c) as a matter of
variability baked into the knowledge representations.
What seems to be required is a representational level more basic than the
predicational, allowing the specification of one common structure for the
alternative expressions of modifiers, for the variants in (60), (62), (63), etc. Also,
in view of the argumentation in 3.1.3.3 with respect to the position of pragmatic
functions in the grammar (which is obviously strengthened by the above cases;
and see also Bolkestein and Risselada 1987), pragmatic functions already have
to operate at this level: this level has to be linked to the predicational and
surface levels by contextualization procedures sensitive to pragmatic functions,
not only for determining word order and intonation, but also for the choice of
lexical items and construction types. The structures at this more basic level
constitute the L cog in (57), and FL would apply only to them.
One might suggest that this deeper level is still predicational, but then
containing only pragmatically neutral predications. However, it is hard to see

which of the variants in (60), e.g., could be chosen as the neutral one on a non-
arbitrary basis. Moreover, this proposal would mean allowing, at a deeper level
in the grammar, what is forbidden at more superficial levels according to the
lexical constancy principle in (59). For changing from the 'neutral' predication
at the deepest level to pragmatically marked types of predications frequently
involves considerable changes in the lexical material.
In fact, various other arguments also put into question the very idea of
conceptualizations being lexical in nature. First of all, various behavioral
phenomena indicate that, cognitively, information and lexical expression of
information are different things. Consider the simple everyday phenomenon of
having word-finding problems, or, in the domain of language pathology, amnesia.
Apparently, S/Hs can know what to say, but fail to find appropriate lexi-
calizations. Something similar has been observed in much of the experimental
research into memory for the verbal form of utterances, which was originally
intended to test the cognitive plausibility of TG deep structure (Sachs 1967,
Johnson-Laird 1970, 1974, Bransford and Franks 1972, Bransford et al. 1972,
Hesp 1990). As said in 2.2.3.2, language users appear to soon forget the exact
verbal structure of utterances and only remember the meaning enriched with
inferential information based on general knowledge of the world. It is not clear
why these things should happen if conceptualizations are lexical.
More recent experiments have shown better recognition or remembrance
of the verbal form of utterances produced or heard in experimental conditions
(e.g. Bates et al. 1978, Hjelmquist 1983, 1984, 1987). But these results do not
challenge the earlier findings, on the contrary. First of all, recognition of
meaning still proved to be much better than recognition of the verbal form in
which it was caught. In fact, there is much variability in the results, as can be
seen if one compares Hjelmquist (1984) and Hjelmquist and Gidlund (1985).
The latter are forced to relativize the conclusions of the former. This suggests
that the experimental conditions have quite some influence on the results.
Secondly, the recognition of verbal aspects of utterances can, contrary to
Hjelmquist's suggestion, most likely be explained on the basis of a principle of
'reconstructability' (cf. Brewer and Hay 1984). An S/H's utterances in context
are obviously not accidental expressions of their meaning: the situational
conditions (cf. 1.2) lead one to very specific choices in order to get a functionally
appropriate formulation. Hence, most utterances are not exchangeable in their
context (cf. Bolinger 1977). Thus, when tested on the recognition or re
membrance of verbal patterns uttered or heard before, it is not unlikely that an
S/H is able to rather accurately reconstruct which utterances have been used and
which have not, because (s)he has information stored concerning the meaning
and the specific situational conditions.
This explains the differences between these recent findings and the older
results in which no memory for verbal form was found. The latter were due to
experimental conditions in which utterances were used without a natural
conversational context, which therefore excluded reconstructability, while the
results showing some gist for verbal form stem from experiments in which there
was a realistic conversational context. Another observation which fits nicely with
the reconstructability hypothesis is that an individual's memory for verbal form
was much better for his/her own utterances than for utterances made by the
conversation partner (cf. Jarvella and Collas 1974, Hjelmquist and Gidlund
1985). Each individual has a personal linguistic style, and one can judge better
what one would say oneself than what someone else would say in a particular
context.
Another case in point is translation. Translating is transposing an
expression caught in the syntactic code of one language into an expression
caught in the syntactic code of another language while preserving the meaning
as precisely as possible. How should this be accounted for without accepting a
to-be-preserved and hence language-independent meaning next to the lexical
expressions in the languages involved? Moreover, accepting a more abstract
meaning representation as the mediator for translation between verbal structures
in languages seems hardly less attractive than the procedure which would have
to be followed according to Dik's proposal, contrary to what he (1987b, d)
suggests. In Dik's proposal translation would only be possible through a
'translation dictionary'. For example, translating the lexical item pinball machine
into Dutch would only be possible through a dictionary entry saying 'pinball
machine = flipperkast'. But suppose some language does not have a lexical item
for this object. Translation would only be possible by means of the 'meaning
definition' linked to the English word in the lexicon (cf. 3.1.1), which is again
caught in a predicational, and thus language-specific format (e.g. something in
the style of Webster's definition: "an amusement device in which a ball propelled
by a plunger scores points as it rolls down a slanting surface among pins and
targets", but then in the format of a predication). Each individual word of the
definition would have to be translated into the target language, supposing, of
course, that there is a translation entry for each of these words (see Van der
Korst 1987 for a more formal account of how this would work). It seems no less
efficient to suppose that the translator grasps an abstract concept of the object
'pinball machine', and has to look for a related concept for which the target
language does have a lexical item, or, alternatively, has to circumscribe the
concept through specifying its embedding in a larger conceptual context in terms

of the target language.
The fact that translation may be difficult (cf. Dik's argument (v) above)
does not affect this proposal. It is no doubt true that knowledge is culture-
bound. This is not at all in conflict with the claim that conceptualization is non
verbal, contrary to what Dik suggests in his argument (iv). Non-verbal does not
mean universal. If the object 'pinball machine' is not known in a culture, people
do not have a concept for it, and conceptualizations can vary endlessly in terms
of which concepts they contain and how the concepts are embedded in the
overall system. As such, conceptualization is not only culture-specific but even
to a certain degree idiosyncratic (cf. 1.2.3.1). It is also no doubt true that lan
guage plays a crucial role in transmitting culture-boundness. The way language,
being an important means for acquiring and transmitting knowledge, allows us
to structure information undoubtedly influences the way we conceptualize the
world (as is claimed in the theory of linguistic relativity - cf. Whorf 1963, Rosch
1974), just as the way we conceptualize the world no doubt influences linguistic
structure (cf. e.g. Miller and Johnson-Laird 1976 for illustrations of how general
perceptual categories are reflected in linguistic structure). Translation problems
may be due, then, to differences in the way languages allow the structuring of
the same basic information, but also, mainly, to the difficulties facing a translator
in communicating the way S/Hs of a specific language conceptualize the world
to S/Hs of another language.
The fact that languages seem to cover or 'hide' conceptualizations and do
structure information differently is not an argument either. Dik's argument (iv)
could be used against any proposal accepting representations more abstract than
surface structure. If predications were basic, why do languages not reveal them
directly? Obviously, in line with the above discussion, the answer is that
languages must allow one piece of information to be expressed in pragmatically
different ways. If predications were basic, why do different languages use such
different grammatical devices for expressing them? Here the answer is that every
language makes its own choice from the means of information structuring made
available by the transmission channel it uses (the articulatory, acoustic and
auditive), and this choice is due to a historically developed convention within the
linguistic community. Now, obviously, these answers are equally valid for non-
lexical basic representations. Of course, it is misleading in itself to claim that
linguistic utterances 'cover' non-verbal conceptualizations. Their purpose is to
allow an S/H to communicate the conceptual information as exactly as possible
(which is not an easy thing to do, as any scientist knows - again, it is not clear
why that would be if conceptualization were verbal). Only, this happens (or has
to happen) according to structural principles differing from those of the concep

tual representations.
Actually, if conceptualization is language-specific, this would seem to imply
that bilinguals or multilinguals conceptualize the world two or more times. This
is cognitively not a very attractive claim.
The main point of the above observations can also be stated theoretically.
According to traditional (Saussurean) definitions, language is a system of forms
and meanings which are conventionally linked together, the former being (more
or less) arbitrary symbolizations of the latter. From a functional perspective, the
formal symbols are used by humans to communicate, i.e. to transmit information
(meanings) in order to act upon other humans. Formulating meanings and recov
ering meanings is what communication is all about. This seems trivial, but it is
hard to see how the claim that conceptualizations are linguistic can be reconciled
with it. If lexical items are formal and arbitrary labels referring to meanings (a
view which is recognized in FG, since terms and predications do contain referen
tial variables), where are these meanings (the referents of the terms and
predications)? Unless one wants to return to an objectivist semantics, i.e. one in
which meaning is caught in terms of reference to the external world (a semantics
which Dik 1987a, 1988 clearly rejects, too) it seems hard to avoid accepting non
verbal conceptual systems representing the meaning of verbal expressions.
In fact, Dik (1987d: 164) himself introduces a notion of concepts to which
lexical labels are attached.
(64) a. x 320 = (d1x320: John Parker pN (x320))

b. x 203 = (d1 next x 203 : monthN (x 203 ): AugustpN (x203))
Moreover, as mentioned above Dik (1988, 1990) postulates that predications

have to be interpreted in terms of pictures, which (at least partially) appear to
be non-verbal in nature (we can suppose that concepts such as X 320 and X203
belong to this non-verbal part of the pictures, the 'images'). It is strange, how
ever, that he considers this interpretation process to be a matter of reasoning
with knowledge, and not of genuine utterance interpretation. Dik's proposal im
plies that conceptual knowledge is stored as non-interpreted (non-understood,
so to speak) form, which can be interpreted if required. The suggestion that the
pictures, against which representations input to FL are interpreted, include
conceptual representations also leads to the rather paradoxical suggestion that
'non-interpreted' conceptual representations can play a role in interpreting
conceptual representations. All of this brings to mind some kind of formal
(model-theoretic) semantics, which does not involve any semantics in the real
sense, however, and which is very implausible from a cognitive point of view (cf.
Lakoff 1987: ch.l4).

Correlated with this, speaking of conceptual representation in terms of a
'language', as is done in (56), is also misleading. A language is a formal system
used for communication of meanings, and hence has a semantics. But conceptu
alization is obviously not a communication system. It is improper to consider it
to be a syntax having a semantics. It must rather be understood as a 'pure
semantics', whatever that will turn out to be (cf. Johnson-Laird 1983: 419).
Otherwise we end up with a (potentially infinite) regression, in which the
conceptual language has to be interpreted in terms of another 'language' which
again has to be interpreted ... (etc.). The latter may be common practice in
many (especially formal semantic) theories (it is also proposed in Fodor 1976,
e.g.), but it is cognitively very implausible (cf. also Lakoff 1987: 227).
Let's view the matter from the wider perspective of human cognition and
action in general. Obviously, language is only one of the sources for human
perception, next to other auditory, but also visual, tactile, olfactory, etc. sources,
which all lead to knowledge of the world. Also, language is only one of the
means humans have for communication, and more generally for interaction with
the environment, and the many other behaviors are based on conceptualizations
all the same. Should we suppose, then, that for each type of perceptual and
behavioral 'channel' there is a different type of conceptual representation
specific to the channel's characteristics? Or, alternatively, that there is one
central type of conceptual representation integrating information from all
perceptual sources, and serving as the common basis for all types of behaviors?
The former view is unattractive: it implies that a lot of information might have
to be stored many times in terms of the different conceptualization types, it
requires an enormous amount of cross-interfacing between them if human
knowledge is still to be seen as an integrated system (considering a human being
as an internally segregated system would be rather counterintuitive), and it leads
to an overly complex reasoning system which has to be able to work in terms of
the different representation types. It is much more attractive to accept the latter
view, which is basically the same as the cognitive simplicity postulate given in
(56) and also used by Dik in his argumentation.
Yet, note that Dik, his own argumentation notwithstanding, abandons this
postulate by proposing that predicational representations and 'images', whatever
their nature, exist side by side, since this implies that there are two cognitive
representation systems anyway.34 Also, Dik's characterization of this distinction
in terms of a distinction between conceptual and perceptual knowledge is
somewhat puzzling. Obviously, conceptualization is based on perception (cf.
Miller and Johnson-Laird 1976, Block (ed.) 1981: 9ff), and one could only speak
of perceptual representations as intermediate representations formed in the

course of processing all kinds of inputs and leading to conceptual representa
tions. This observation makes Dik's assumption, discussed above, that conceptual
representations are to be interpreted against perceptual representations even
more difficult to follow, since that would imply turning the world upside down.
Anyway, if we do stick to the cognitive simplicity postulate and accept only
one type of core representation, the claim that it is verbal raises a number of
general questions. A child starts perceiving and behaving (also communicatively)
before it starts developing language, and as indicated in 2.3, language acquisition
strongly depends upon the development of general conceptual capacities, and
upon more primitive, non-verbal communication systems existing prior to
language. How are we to understand the nature of pre-linguistic conceptualiza
tion in a child, then? And why should a representation type due to a perceptual
and behavioral system, which develops relatively late, acquire the status of a
central representational system for conceptualization, while representations due
to earlier perceptual and behavioral systems do not?35 One could as well, or
maybe even better, claim that visually based representations acquire this status.
After all, they are crucial from the beginning of human development, and
remain so throughout human life. Yet such a suggestion would have its
problems, too: it is hard to see how concept X 2 0 3 in (64b) above, or a conceptu
alization of whatever other non-physical element, could be visual. Finally, how
are we to grapple with the knowledge that people born deaf and/or mute, who
hardly ever acquire natural language, still have cognitive capacities which seem
qualitatively undistinguishable from the cognitive capacities of normal individuals
(e.g. Lenneberg 1967)?
It seems more reasonable to assume that conceptualization is a matter of
developing a representational system which is not specifically determined by, and
thus not biased toward, one specific mode of perception and behavior such as
the linguistic or visual one, but which steers a path between the different
channels (i.e. allows to store information from the different channels in a
uniform way). (These considerations at once provide an answer to Dik's
argument (ii) above.)
Of course, between this system and the purely physical level of perception
and behavior there have to be levels of representation more specific to the
different channels. One can assume that the principles of these representations
are constantly accessible by the central system. One can also assume a perfect
parallelism between language and the other channels in this respect. For each
channel one could assume a 'lexicon' of basic 'labels' which correspond to
concepts or conceptual conglomerates of the central system, and which can be
used in perceiving and behaving through the specific channel according to the
'rules' or 'procedures' for combining them, or for decoding their combination.
Thus, the concept of garlic might have linked to it a verbal lexical, but also a
Visual-lexical', an 'olfactory-lexical', etc. label. If one is asked how garlic smells,
one can mentally 'imagine' this smell by invoking its 'olfactory-lexical' label.
Humans have not developed the olfactory channel into a means for actively
interacting or communicating with their environment (though production of smell
does play an important role in human interaction at an unconscious level, e.g.
in mating behavior - cf. Argyle and Trower 1979). That is, they do have a limited
'receptive grammar' for smell, but hardly a 'productive grammar'. Thus, in
absence of a piece of garlic, the one asked might try to imagine an object with
a comparable smell (onion for example) and provide the information requested
by verbally referring to this object.
The situation is quantitatively, but not qualitatively different for vision.
Humans not only have an extensive capacity for visual perception, but also for
visual behavior. They can mentally create quite complicated visual images, and
they can also exteriorize them in drawings, paintings, etc. Consider also the fact
that some written communication systems, e.g. the Egyptian, strongly rely on
visual representation. Hence, not only the receptive, but also the productive
'grammar' of the visual channel is rather well-developed. (An interesting
parallelism between the different channels (including the linguistic) is that the
receptive capacity always seems to be stronger than the productive capacity.)
This proposal allows us to account for the empirical evidence 'pictorialists'
have always used in their claim that imagery is pictorial in nature (cf. Block (ed.)
1981), without having to claim that conceptualization is pictorial indeed. It also
allows us to account for the difference between individuals' intuitions, frequently
mentioned in the imagery debate: some people suggest they frequently take
recourse to 'pictorial' representations in cognitive activity, while others claim
they often make use of language-like representations (cf. also Chafe 1977b). One
can suppose that the conceptual system of rather visually-oriented individuals is
inclined to draw upon the visual channel if necessary, while the conceptual
system of the more verbally-oriented individuals rather draws upon the linguistic
channel.
Note that the interface problem mentioned in Dik's argument (i) above
exists in whatever theory one adopts, since there have to be connections between
types of representations occurring in cognitive processing anyway. If concep
tualizations were verbal, there would still have to be interfaces with other levels
of representation (visual, olfactory, etc.). In Dik's 'pictures', there has to be an
interface between the predications and the images. Hence, Dik's proposal only
moves the interface-problem from grammar to other domains of inquiry. For the
linguist this may be a practical situation, but within a larger cognitive framework
it is not likely to reduce the problems at all.
The foregoing argumentation obviously does not exclude humans storing
more complex channel-specific representations, if necessary. In an experiment
Johnson-Laird (1983: 162ff) found that people better remember a verbal
representation in case of indeterminate discourse, i.e. discourse for which they
cannot form a clear interpretation (a 'mental model'). Yet, nothing in the ex
periment shows that this representation is definitively stored as a conceptual
ization (the experiment only tested memory during a short time span). Rather
(and remaining within the spirit of Johnson-Laird's proposals), one could assume
that the language user keeps this representation in memory for a while, still
trying to interpret it, and ultimately either forcing an interpretation in terms of
a mental model (which frequently happens, as Johnson-Laird states - cf. also
Bredart 1987), or, if this is not possible, gradually forgetting it. The same
mechanism might be postulated for the other channels as well.
Another case of storing channel-specific representations occurs when
memorizing a text, a picture, etc. Yet, it is not a priori clear whether in such a
case a person stores the text itself, or rather the meaning of the text plus a path
through lexicon and grammar, allowing a reconstruction of the verbal form of
the text. However, in the former case predicational representations cannot do
the job either: the language user has to memorize the surface structure of the
utterances. Moreover, it will be clear that this type of storing is rather
exceptional. This is reflected in the fact that it is much more difficult than simply
understanding and storing the meaning of a text.
Hence, it seems impossible to avoid a higher degree of abstractness in
representing conceptualization than Dik would be willing to grant.36 Obviously,
functionalism and concreteness are not as intimately linked as he suggests. On
the contrary, the above analysis shows that functional considerations automatical
ly force one into rather abstract arguments and assumptions, as was already
indicated by Wittgenstein (1958).
3.2.1.3. On the structure of conceptualization

If representations more abstract than predications (something which in Dik's
model would constitute the 'images')37 have to be introduced as conceptual
representations, and, hence, as basic representations in the grammar, what
should they look like? Again, at present no full-fledged answer is possible. There
are plenty of proposals available in the literature:38 different types of logical
representations (e.g. Winograd 1972), of semantic networks (e.g. Quillian 1967,
1968, 1969, Schank 1972a, b, 1973, 1979, Schank et al. 1975), Johnson-Laird's
(1983) mental models, etc. But they are either undeveloped, as Johnson-Laird's
mental models, or they are disputable. For example, many proposals put forward
'primitive' relations (e.g. Schank) or concepts (e.g. Wierzbicka 1972, 1980), but
one can often question whether these are primitive indeed (cf. e.g. Johnson-
Laird 1977b on Schank's model).
One difficulty is that proposals are mainly based on intuitive choices,
without clear arguments for the plausibility of the representations introduced.
This is due to the fact that at present it is hard to pin down which characteristics
the system should have in order to account for the many complex aspects of
human behavior and perception. As indicated above, one needs insight into the
requirements posed by the different channels of perception and behavior and
their interactions. But of these, language is no doubt the best explored, and even
then our knowledge is far from stable. In fact, most proposals have been con
ceived from the perspective of natural language only, which is of course also
potentially biasing (cf. Chafe 1977a for a similar critique). In any case, develop
ing a serious hypothesis about the format of conceptual representation is going
to be a very long-term project.
Hence, the fact that current discussions of conceptualizations tend to use
lexical labels to refer to concepts (cf. Dik's argument (iii) above) can hardly be
used against them. As long as it is unclear how they should be conceived, some
arbitrary notational convention is needed for the purpose of scientific com
munication. One can use numbers, as Dik does in (63), but since scientific
communication is normally linguistic, one can use linguistic means as well. The
fact that we are still naive about what conceptualizations might look like is not
an argument either. (Dik's model is faced with the same problem as far as the
'images' are concerned.) More and more physicists accept that there must be a
fifth dimension (and some claim even more) next to the four dimensions of
space and time on which modern physics is built, even though nobody has been
able to indicate how this fifth dimension is realized physically thus far. If the
argumentation for non-verbal conceptualizations is sound, we simply have to
accept the idea, and develop strategies to find out about them.
Let's see which direction we could take. Most probably conceptualization
should be conceived as a complex system of representations of 'basic' infor
mation about the world which can have different kinds of hierarchically organi
zed 'meta-level qualifications', more or less comparable to the system of layering
introduced in the FG predications.
The 'basic' level, which represents the SoAs about which the language user
has knowledge, can be assumed to consist of networks of mutually related
concepts (units like frames, scripts, etc. - cf. 3.2.1.1 - should somehow be seen
as higher order elements of organization of clusters of related concepts). Most
probably we have to assume the existence of quite complex clusters or chunks
of interrelated concepts - in any case more complex than the information which
is normally represented in one natural language utterance. The experimental
investigations into memory for wording cited above have shown that under
standing utterances involves embedding the information in a larger context of
inferentially related information; and forming an utterance can be seen as
abstracting a piece of information from a larger 'chunk' (Chafe 1977a, b) (a
discourse can be seen as a systematic deployment of all the relevant information
from a coherent chunk of conceptual knowledge - cf. 3.2.3). This might be taken
as a suggestion that a semantic network really is the most appropriate way to
represent conceptual knowledge (despite its disadvantages, cf. Johnson-Laird
1983: 211ff). If so, one can assume that the organizational principles of this
network (the types of relations occurring) are universal, while which concepts are
present, and how they are interrelated, is variable, and, hence, specific to cul
tures, social groups, education, etc. As already argued above, no two individuals
have identical conceptual networks. Thus, it is no problem to account for the
fact that houses are quite different things in different cultures. Although an
'average' western European, an 'average' Nepalese and an 'average' Eskimo all
have a concept of a house, the conceptual information linked to it and defining
it will be quite different, irrespective of the languages spoken.
With respect to the concepts, there is no need for 'semantic decomposi
tion'. One can perfectly follow Dik's (1978b) argumentation against such models
as the semantic feature theory and the generative semantics concept of semantic
decomposition (cf. also Johnson-Laird 1983: 206ff). In fact, these arguments are
not evidence against the claim that conceptualization is non-verbal, contrary to
what Dik sometimes suggests. Concepts do not have to be primitive, unanalyz-
able elements. People have concepts of whatever is a discrete phenomenon to
them, while in reality these phenomena can be complex, composed of more
fundamental elements (which obviously does not mean that there are no
primitive concepts - cf. Johnson-Laird 1983: 410ff). For example, many people
have a concept of '(biological) evolution', although this is a very complex
phenomenon. In the conceptual world of some (e.g. biologists), this may be
reflected in the fact that 'evolution' is a 'very high generalization' concept which,
at lower levels of generalization, is linked to far more detailed chunks of infor
mation. In most laymen's conceptualizations, however, evolution is a rather 'low
generalization' concept having hardly any more specific information linked to it.
In this way one can account for the intuition that some entities are complex, and
hence decomposable in much the same way as Dik does in terms of meaning
definitions, but then only at a deeper level of analysis (cf. Dik 1987d).
Since the conceptual relations are supposedly universal, however, one
might look for a set of primitive relations. The problem is to determine what
relations are 'primitive'. One could start with a minimal assumption. There can
be little doubt about the primitiveness of the relations of 'being' (B), 'having'
(H) and 'doing' (D). Probably this set is too limited, and there is no reason not
to expand it if further analyses force one to. However, it suffices to represent a
number of other relations which in the literature have been claimed to be primi
tive, at least if one allows for the existence of relations between concepts and
relations. For example, causation might be represented as in (65).
(65) x
('x' represents concepts, >D>, >H>, and >B> indicate the type of (oriented)
relationship holding between them.) 'Cause' is then a 'higher generalization'
concept over cluster (65).
By means of this set, one can also provide basic representations for the
sets of utterances discussed in the previous section. (62), e.g., involves the
transfer of an object. Schank et al. (1975) consider this a primitive relation, yet
(66) seems to provide an even more basic representation (in which 'T' stands for
the temporal situation of a conceptual relationship or cluster).
(66)
Verbal lexicon:
x1 = (dlx 1 : John pN (x1))
x2 = (dlx 2 : Mary pN (x2))
x3 = (ilx3: apple N (x3))
Of course, the D-relation in this network can have different properties. If there
is no further conceptual specification, the 'neutral' predicates give/get can be
chosen. But the D-arrow could also have a B-relation with a concept of 'throw',
e.g., which would be a 'higher generalization' concept having linked to it a 'lower
generalization' conceptual cluster specifying what is involved in throwing. In that
case, the S/H might choose throw as a predicate. Or the conceptualization in

(66) might be embedded in a wider conceptualization also involving the
knowledge that this happened in an official context, which might lead the S/H
to the predicate hand over/receive. This clearly shows how a relatively simple
utterance can require a rather complex conceptual representation.
The same can be said for the examples in (60). A conceptual representa
tion would also include indirect information provided in these utterances: that
Johnson is the boss of a business, that he pays people to do work for him, etc.
In this respect, there can (but does not have to) be a difference between (60a-b)
and (60c-d). Though they all provide the same direct information, the former
only seem possible if the further information just mentioned does apply, while
the latter are less committed with respect to this indirect information. Hence, an
S/H can choose the latter as 'synonymous' with the former in a context in which
this indirect information is or will become clear to the H/S. S/H can choose the
former if (s)he wants to provide this indirect information at once.
The above analysis suggests that there is no direct correspondence between
the distinctions concept/relation at the conceptual level and noun/verb at the
linguistic level: the relationship is complex. The relations mentioned obviously
find the most direct expression in the verbs do, have, and be, and many concepts
are most naturally expressed by nouns. That is only a matter of prototypicality,
however. Neither does there have to be a one-to-one relationship between
concepts and lexical items (as Dik's example (64b) already suggests). Though
there certainly will be a tendency to introduce a verbal lexical label for every
concept that is sufficiently salient and general, people can have concepts for
which they have no single lexical label, and which they can only express verbally
by means of circumscriptions. Thus, you will certainly not have a name for every
single chair at home, and in their immediate neighborhood you can refer to
them by means of deictic elements like that chair (combined with pointing), but
otherwise you can only use circumscriptions like the chair between the table and
the window. All of this indicates once again that conceptualization is certainly not
a 'shadow-representation' (cf. Dik's argument (ii) in the previous section) of lin
guistic representation.
Next to this 'basic' level of representation of SoAs, conceptualization also
has to involve various sorts of meta-qualifications. In fact, part of the elements
introduced in the layered structure of the clause in FG (cf. 3.1.1) can be
assumed to be relevant already at the level of conceptual knowledge. To be
excluded from this are, obviously, all elements related to the illocutionary
potential of an utterance (ILL and its mitigating and reinforcing operators) in
(6). Since illocutions are markers of an action potential (i.e. a communicative
intention), they are only introduced at the moment when the process of
constituting an utterance is being started. They are not aspects of the conceptual
knowledge representations, although the sources for their assignment (including
mitigated or reinforced variants) obviously do have to be present in conceptua
lization. Communicative intentions are determined by higher order intentions
which (as mentioned in 3.2.1.1) are 'triggered' by specific constellations in S/H's
knowledge of the SoA (for instance, conflict between pieces of information or
lacking information, a divergence of an SoA in reality from the S/H's ideal
conception of that SoA, etc. can lead to an intention to solve this by communica
tive means). Communicative intentions are also more directly determined by the
conceptual SoA (if knowledge of an SoA is to be completed this can lead to a
question, if an SoA is to be changed this can lead to a request or an order, etc.).
And they are determined by S/H's social relationship with H/S, and by the
setting (an example is the choice between an order and a request, which is kind
of a mitigated variant of an order).39
All other qualifications in the layering proposal in FG can be assumed to
be relevant in conceptualization, since they all concern the information contained
in the conceptual SoA. This is also true for what, in the system, could be called
'proposition qualifiers' (operators and satellites expressing evidentiality and
subjective modality): in line with the argumentation in 3.1.2, it is unclear why
these should be a matter of the speech event, more or less like illocutionary
matters, as opposed to lower level qualifications which would be a matter of the
SoA involved. Evidentiality concerns the source of the S/H's information, and
there is no need to relate this to the utterance per se rather than to the SoA.
Similarly for subjective modality. The distinction between evaluating an SoA in
terms of general knowledge (objective modality) and expressing a degree of
commitment to the truth of the content of a proposition (subjective modality)
(Hengeveld 1988: 236) seems counterintuitive (one would say that the latter is
no more than the result of the former), and as will be shown in 3.3.3.1, one can
easily do without it.40
As to the representation of these different qualifications, again, there is no
reason to believe that the conceptual level should be a simple 'shadow' of the
linguistic level. We may expect that all major distinctions at the conceptual level
are somehow reflected at the linguistic level, and that the latter respects the
(hierarchical) relationships holding between qualifications at the conceptual
level. Yet the structuring principles at each level may be quite different (as said
in 3.2.1.2, the distinction between operators and satellites at the predicational
level can better be assumed not to have an equivalent at the conceptual level).
As to the types of conceptual qualifications, one can largely adopt the
individual categories occurring in the FG proposal developed by Hengeveld

(1989) (cf. 3.1.1), although specific alterations (as in the case of subjective vs
objective modality) and additions may be required. One may have to add a
qualification indicating whether the SoA belongs to the real world or to a
deviant possible world (realis/irrealis), and one indicating the S/H's attitude
toward the SoA (emotional involvement in terms of liking or disliking - see
3.2.1.1 - required to account for satellites such as 'damned', or for the choice
between ameliorative vs pejorative terms or predicates). A difficult question is
whether locations in space ought to be considered part of the SoA proper or
should also be dealt with in terms of meta-level qualification.
Concerning the hierarchical ordering of the qualifications, we may have to
drop the division in 'macro-layers' proposed by Hengeveld. As argued in 3.1.2,
it seems hard to find a motivation for Hengeveld's main division; and upon
closer examination one has to accept hierarchical relations between qualifications
which in Hengeveld's system are situated at one 'macro-layer' as well (see
below). Hence, there seem to be reasons to introduce (at least at the conceptual
level - see below) a much more gradual system in which each qualification
occupies its own 'layer' or level (as in Foley and Van Valin 1984).41 The suc
cession may be more or less according to the hierarchy of individual qualifica
tions in Hengeveld's proposal, and the rationale behind it seems to be that a
qualification more inherent to the SoA is on a lower level than a qualification
imposed upon the SoA by the S/H.42 For example, a temporal location is a
more inherent property of an SoA than a qualification in terms of epistemic
modality. In view of this, the 'attitude' qualifier constitutes one of the outermost
layers. The 'reality/irreality' qualification can be expected to be somewhere be
tween temporal qualification and epistemic modality. If space is a meta-
qualification, it certainly occupies the lowest level.
In some respects Hengeveld's hierarchy seems to require changes. Deontic
modality is certainly not at the same level as epistemic modality. While epistemic
modality can only have temporal qualification within its domain, and not vice
versa, deontic modality and temporal qualification can extend over each other,
as appears in (67).
(67) a. John was probably going yesterday

b. John was obliged to go yesterday
c. John was probably obliged to go yesterday
(67a) can only be read with the temporal qualification within the domain of the
epistemic modality. (67b/c), however, are ambiguous as to the relative domain
of the temporal and deontic qualifications. In (67c), the deontic modality can
only be understood within the domain of the epistemic modality, but not vice
versa. See also (68).
(68) *John was obliged probably to go yesterday
We may also have to allow a greater flexibility in the hierarchical position

and scope of (some of the) qualifiers than is accorded in Hengeveld's system. As
to hierarchical position, a first example was given in (67b). Apparently, it is not
possible to postulate a fixed position of temporal qualification and deontic
modality relative to each other. Another example is negation, which in
Hengeveld's proposal is already introduced at two levels (predicate negation, and
polarity as a predication operator). Rather than being two different qualifiers,
this can be considered one qualification operating at two different levels.
Consider also (69).
(69) John didn't go to the movies yesterday
(69) can be read as 'not <John went to the movies yesterday>' or as 'yesterday
<John did not go to the movies>' (next to other readings, e.g. with the negative
only covering the temporal qualification - see below).
As to the extension of scope, each qualification may cover chunks of
conceptual knowledge of variable size: one relator, a set of two related concepts,
or some larger complex of concepts. (They can certainly extend over the
information provided in a series of subsequent natural language utterances; but
they can also extend over information provided in a single term only. Hence, the
different qualifications which can occur in an utterance can also occur in a term,
albeit in a different formal expression. Cf. Rijkhoff's 1990 attempt to relate
qualifications at the term level and at the predication level. See 3.2.3.) They
sometimes also seem to cover each other. Thus, a temporal location of a SoA
can be qualified in terms of its probability, without necessarily implying that the
SoA itself is also qualified in terms of this probability. For example, (67c) is
ambiguous as to whether the epistemic modality only covers the temporal quali
fication, or the SoA and the temporal qualification, or even the SoA and the
temporal and deontic qualifications. It can also be read with the epistemic
modality only having the deontic modality in its scope. Similar problems exist in
(69). Probably, part of the account for this problem is that, next to levels of
qualification, it is also possible to have levels of embedding of conceptualization
(i.e. having 'mental models' within 'mental models'). For example, in (70)
(70) Probably John thinks that I am crazy

John thinks that I am crazy is a SoA to the S/H, in itself involving a qualification
by John of a SoA, which is qualified by the S/H for epistemic modality. Anyway,
the precise relationship between extension of scope and hierarchical position is
a matter which deserves further investigation, as does the entire issue of the
structure of conceptualization.
3.2.2. Production and interpretation, and knowledge and control
Another important issue for modeling the cognitive systems involved in language
processing, which was already touched upon in 2.2.3, concerns the question how
language production and interpretation relate to each other. As mentioned, FG
(like most linguistic grammars - cf. 1.1.3.2) is molded in a productive mode while
there is no concern with the issue of language understanding. In language
psychology and AI, on the other hand, the main focus has been on language
interpretation, very often with a complete lack of attention for the issue of
language production. One can wonder whether such a one-sided approach to
one 'mode' of language processing can lead to adequate hypotheses about the
cognitive systems involved in language behavior. Of course, a full account of the
production processes is not possible without also considering the interpretation
systems in the sense that it is very likely that the feedback systems, which are
very important in production, are based on the interpretation systems (cf. Levelt
1989). (Probably something similar is true for the interpretation processes.)
However, a much more interesting, but also much more difficult aspect of this
problem involves the theoretical question to which degree and in which way one
can assume that the cognitive systems involved in production and interpretation
as such are related.
Although explicit considerations of this issue are remarkably rare, the
range of (implicit) views on the matter appears endless. Most of those explicitly
concerned with the issue suggest a link between the systems involved in the two
modes in the form of a correspondence to a smaller or larger degree of the
knowledge present in them. Yet, this can still lead to quite different process
models, depending on how one conceives of the way knowledge is systematized
in them. One of the strongest claims of a very intimate relationship between the
two modes of processing is made by Garrett (e.g. 1980, 1982a, b). Yet, except
for the issues of the similarity of types and levels of representation formed
during both processes, and of the processes of lexical access in both (Garrett
1982b), the specifics of the claim remain open (even the issues of the format of
the representations and of the processes of lexical access remain vague).
A more specific suggestion is to be found in Johnson-Laird (1977b), who

adopts a strongly procedural approach. Though ATN grammars were originally
devised for the purpose of parsing, they also appear to be applicable for
generating utterances (Simmons and Slocum 1972). Johnson-Laird suggests that
it is possible to use the same 'semantic transition network' (a variant of the
ATN) in production and interpretation, albeit with different procedures related
to the different stages encountered in running through the network. An
equivalent approach in terms of FG would be to accept that the different types
of frame-like structures (e.g. predicate-frames in the lexicon and syntactic
templates in the expression component - i.e. the declarative knowledge) are used
in both modes, but that the procedures applying to these frames are different
for both. Dik (1989a: 51ff) himself seems to want to make a much stronger claim
however: he suggests that the same system as involved in the productive FG, but
then applied in the opposite direction, could be used for modeling language
understanding.
Another type of approach is proposed in Frazier (1982), who suggests that
production and understanding involve the same set of (syntactic) rules, which in
itself is independent of either mode of processing. (This is obviously inherited
from the Chomskyan era in psycholinguistics.) Yet, since rules can only be
understood to be directions for action (cf. 2.2.3), Frazier's claim implies that
both modes make use of the same set of specifically syntactic procedures while
alongside it there have to be 'general-purpose' strategies which (paradoxically
enough) have to be specific for each mode.
The problem with these and other proposals is that they have not been
developed in sufficient detail (and to a large degree cannot be, given our current
state of knowledge about the processes of production and interpretation) to
allow any conclusions as to their practical feasibility in the framework of an
encompassing process model. Significantly, in AI, where it is necessary to have
fully operational and thus fully implemented models (cf. 1.1.3.2), in as far as
extensive generative systems have been developed next to the more common
understanding systems, the two have mostly been juxtaposed without any clear
links (this is the case e.g. in Schank et al. 1975). In fact, one should not exclude
the possibility that this is also the case cognitively: the idea that the generator
and parser are fully independent components of the linguistic systems is as
conceivable as the other suggestions above. They may have developed on the
basis of the same general principles, and operate with the same types of
conceptual knowledge and involve the same general types of control and
procedures, but still be realized separately, each in its own specific way.
The general observations pertaining to the issue are rather confusing.
Intuitively it would seem unlikely that there is nothing in common between the
two modes given that an S/H can speak and understand the same language: this
suggests that somehow the same knowledge is used in both modes. Yet, as
already mentioned in 2.2.2.2, an S/H is normally able to produce only part of
what (s)he is able to understand (and the differences can be considerable if one
considers that one can have serious 'passive' but very poor 'active' knowledge of
a foreign language). This suggests that part of the knowledge available for
interpretation is not available for production. (The opposite situation, that an
S/H would be able to produce utterances which (s)he cannot understand, seems
very unlikely.) It is hard to see what would be the best way to account for this.
Should one assume that the knowledge used in both modes is represented in a
way which makes it accessible for the two modes, and that the knowledge used
in understanding only is represented in a way which makes it accessible for the
interpretation system only (which somehow implies that interpretation can
involve two separate and complementary systems of knowledge)? Or, alternative
ly, should one assume two separate systems, each involving internally integrated
knowledge specified in a mode-specific way (which would imply that identical
linguistic principles, e.g. of word ordering, are present twice, once in each mode,
but differently implemented)? (Consider also Ruder and Finch 1987 on these
issues.)
Considerations with respect to the rather different functionality of the two
modes would then again suggest considerable differences in the organization of
the systems involved in either. As argued in 1.2.3.1, there is priority of
production over interpretation in the sense that speaking is performing an act,
while understanding is undergoing this act. This makes for a big difference in the
nature of what has to be done, if only in that there is a totally different input
and output, and a totally different relationship between the processes and the
conceptual basis against which they take place in both modes. Production starts
from conceptualization (of the relevant SoA) and has to construct a phonetic
pattern against the background of conceptualization (of the personal and social
dimensions and the properties of the setting), driven by an intention. Interpreta
tion starts from a phonetic pattern and has to (re)discover a (probably
unfamiliar) conceptualization and intention against the background of conceptu
alization.
A complicating element in evaluating the meaning of such observations,
and in understanding the entire matter of the relationship between both modes,
is the issue (already introduced in 2.2) of the relationship between linguistic
knowledge and its use by the control system. Let's go somewhat deeper into this.
It is useful first to take a closer look at the discussion in language
psychology concerning the nature of language interpretation in particular (see

also Flores d'Arcais and Schreuder 1983 for a summary survey). Though this
strongly simplifies matters, it seems that one can roughly distinguish between two
types of views, viz. an 'autonomist' and an 'interactive' view. In the 'autonomy
hypothesis' interpretation is taken to involve a strictly ordered series of analyses
(although parallel processing is possible within each level of analysis). First there
is a syntactic analysis involving the construction of a syntactic pattern, then there
is a semantic analysis involving the construction of the literal meaning, and then
there is a conceptual analysis involving the construction of the real meaning in
context. Each analysis is independent (i.e. it does not consider information from
other levels of analysis), and has to be completely finished before the next level
of analysis can be entered. This view, which again reminds us of Chomsky's
influence on language psychology, is taken, for example, in Forster (1979),
Marcus (1980, 1982, 1984), or Garrett (1982a, b) (the latter in analogy with his
views on language production).
According to the 'interactive' view, the parser constantly makes use of as
many sources of information as possible (syntactic, semantic, and conceptual),
while entering as soon as possible into a level of analysis as deep as possible.
This view is held by a majority of language psychologists, but is also most
variable in terms of the exact conception of the entire process: see e.g. Clark
and Clark (1977), Schlesinger (1977), Johnson-Laird (1977b, 1983), Hrmann
(1978), Marslen-Wilson (1980), Marslen-Wilson and Tyler (1980, 1981). It is also
common in AI, e.g. in the 'Hwim' parser (based on an ATN - cf. Woods 1982,
Arbib 1982) or in the series of 'Hearsay' processors (cf. Reddy et al. 1973,
Lesser and Erman 1977, Arbib 1982). This view is often accompanied by an
inclination to reduce the role of syntax in a parser to a minimum. For instance,
in the parsers developed by Schank and his colleagues, syntax is only used to test
or complement the semantic analysis if there is need for it (cf. Schank et al.
1975, Riesbeck and Schank 1978, Schank and Birnbaum 1984). In any case,
attention to matters syntactic is minimal in many of the interpretation models
of this type.
One should be careful in deciding on the appropriateness of these views.
First of all, one should be very wary of the tendency in the latter view to reduce
or neglect syntax, at least as far as the presence of knowledge of syntactic
principles in the parser is concerned. The syntactic characteristics of utterances
must somehow be created by the generator and must also be analyzable by the
parser. Hence, one cannot avoid assuming that an S/H possesses a complete
system of syntactic knowledge for language involving all aspects of syntax
observable in utterances, also for language understanding. In this respect one
can certainly agree with Marcus' (1984) criticism on Schank and Birnbaum
(1984). Moreover, it seems that one should allow for the possibility that the
syntactic systems are used separately in parsing, i.e. without making use of non-
syntactic and non-lexical knowledge. An S/H appears to be able to assign a
syntactic structure to syntactically well-formed but semantically anomalous
utterances. In fact, one is perfectly able to see the syntactic pattern of (71)
(taken from Marslen-Wilson and Tyler 1980: 8), which consists of randomly
chosen lexical items, or even of (72), which consists of totally meaningless items,
even though one can certainly not make use of semantic or conceptual
information to aid the syntactic parsing.
(71) The power was located into great water. No buns puzzle some in
the lead off the text
(72) The lif strook the flurk
Hence, there is no a priori reason why an H/S could not also make a syntactic
analysis prior to a semantic one in interpreting normal utterances, if there is
need for it.
On the other hand, it is difficult to see why one would want to assume that
in normal communicative circumstances an H/S has to complete a full and
isolated syntactic analysis before entering a semantic level of analysis, as is done
in the autonomist view. That would be quite inefficient: understanding means
recovering S/H's meaning, and it seems evident that a H/S will try to reach that
goal as soon as possible, pushing the analysis to deeper levels as soon as it is
feasible to do so, and making use of as much information from different sources
as possible. Hence, it seems strange to assume that a parser should make a
complete syntactic and a complete semantic analysis, probably in parallel, but
in any case each in isolation, as Berwick and Weinberg (1983) do in an attempt
to reconcile the autonomy hypothesis and the views of Marslen-Wilson and
Tyler. If there are sufficient elements for an H/S to make a serious semantic
analysis, what would be the point of continuing a syntactic analysis? It is even
reasonable to assume that if the circumstances allow, an H/S can interpret
utterances mainly on a semantic basis without seriously invoking syntactic
procedures, except in case of uncertainty about the result or in case the semantic
analysis fails, when syntactic procedures may help to resolve the problem. (The
discussion in Schank and Birnbaum 1984 makes this claim perfectly plausible.)
A discussion of the views on language understanding developed by
Marslen-Wilson and Tyler (Marslen-Wilson 1976, 1980, Marslen-Wilson and
Tyler 1980, 1981) will further illustrate what is at issue here. They suggest that
understanding (and in particular lexical recognition, which is the main topic of

their experimental investigations) is an interactive process in which at each level
all possible sources of information (lexical, syntactic, contextual, and conceptual)
are used, yet they also maintain that it is a strongly obligatory or automatic
process which runs primarily bottom-up. One may have some difficulties
reconciling these suggestions, however. By 'primarily bottom-up' Marslen-Wilson
and Tyler mean that the determination of which elements are present at each
level of analysis is accomplished primarily on the basis of an analysis of the
input, whereas the top-down role of higher order information is only restrictive
and secondary. They exclude the possibility of a 'positive' top-down effect in the
form of a pre-selection of elements potentially occurring at a certain level under
the influence of information from higher levels. More specifically with respect
to their topic of investigation, they exclude the possibility that lexical choice
could be made on the basis of contextual expectations.
By defining the set of perceptual alternatives on the basis of bottom-up informa

tion alone, a processing system not only restricts the set of alternatives at least as
much as top-down pre-selection normally would, but it also ensures (within limits)
that the correct word-candidate will be contained within this initial set.
(Marslen-Wilson and Tyler 1980: 62)
In this context the meaning of 'interactive processing' becomes very narrow.

The rationale behind this claim is not immediately apparent, though. The
experimental data they offer do render non-interactive views improbable, but
they do not seem to exclude a more strongly interactive concept of understand
ing. Thus, it is not clear what there would be against the suggestion that lexical
selection is a matter of phonological analysis of the input (bottom-up) and of
expectations derived from the context (top-down) as equivalent and mutually
supportive strategies, with the possibility that one takes the lead over the other
depending on the circumstances and requirements. Thus, after a partial
structural and semantic analysis of an utterance it is no doubt often possible
(especially with respect to topical or given elements, which are obviously more
predictable than focal or new elements) to form a hypothesis about what is still
to come, on the basis of knowledge of its context in combination with general
knowledge (Schank 1978), or even on the basis of linguistic knowledge (assuming
an FG type of lexicon, recognition of a predicate automatically implies the
identification of a potential argument pattern, which obviously creates syntactic,
but also lexical expectations). Thus, one may have predictions about lexical items
and corresponding meaning units still to come even before the processor has
registered the remainder of the utterance, and the phonological analysis then
only serves to make sure that the prediction was all right (which may of course
lead to correction if it was wrong), or to make a final selection if there were a

couple of a priori postulated options left. It is not even clear what there would
be against the assumption that sometimes an H/S makes a lexical selection on
an exclusively top-down basis (maybe this then only involves selection of the
lexical concept). If the prediction is considered precise and probable enough,
further bottom-up analysis may very well be omitted.44
Marslen-Wilson and Tyler (1980: 62) are no doubt mistaken when they
claim that "successive words in an utterance are only rarely fully predictable
from their prior context". This is contradicted by many facts. Normal conversa
tion is crowded with elliptical and incomplete utterances without this causing
interpretation problems. It is also common to observe that an H/S helps an S/H
by 'offering' a word if the latter happens to have word-finding problems (the
word may be rejected, but is also often accepted by the S/H). Particularly
frequent in lively debates is the phenomenon that an H/S does not allow the S/H
to finish the turn but starts to reply because (s)he thinks (s)he already knows
what S/H wants to say and therefore does not need the rest of the turn (this
assumption may turn out to be wrong, but is often right). Also consider the type
of communicative misunderstanding in which an H/S replies to an S/H on the
basis of an interpretation of the latter's previous turn which then turns out to
have been wrong (i.e. it turns out that the S/H has said something different).
This is nicely illustrated in the following riddle quoted in Johnson-Laird (1977b:
94).
(73) If a plane crashes exactly on the border between the USA and
Canada, where do they bury the survivors?
If one presents this to unwarned individuals, they do start wondering whether the
answer is Canada or the US, while survivors are obviously not buried any
where. 45
Marslen-Wilson and Tyler's claim that primarily bottom-up processing
would generally be safer and more efficient can also be questioned. It is certainly
untrue as compared to really mixed top-down and bottom-up strategies.
Consider lexical selection again. Even if there has been top-down pre-selction
of lexical elements, the phonological analysis of the input signal always remains
as a test. If the prediction appears correct, the entire procedure has not been
more complex or slower. On the contrary, it might be more circumstantial if
selection is made only after an analysis of the incoming signal, and if implications
for higher levels of analysis can be drawn only then. If the prediction appears
wrong, one is in the same position as if there had not been one. Pure top-down
processing (if one can still call it that if the H/S moves to the conceptual level
at once) may be more risky (it certainly is not less efficient in terms of speed of
processing), but as the above illustrations suggest H/Ss may be willing to take the
risk in certain circumstances.
Moreover, primarily bottom-up analysis also has its pitfalls. Consider the
problems such an approach would have with syntactically ambiguous utterances
in which evidence for the actual structure is available only late in the utterance,
as in Bever's (1970a: 316) example (74).
(74) The horse raced past the barn fell
If one has to process this primarily bottom-up there are two possibilities. In case
one is aware of a potential ambiguity from the very beginning (which is actually
already a top-down effect) one has to form a double analysis until one is able
to decide which structure is actually being formed. In case there is no awareness
of possible structural ambiguity (which is no doubt most frequent) one is likely
to start constructing the wrong pattern (especially if intonation does not give any
hints as to the nature of the construction, which can easily happen in (74)), viz.
that for the horse raced past the barn, which then has to be discarded once the
last word has been perceived (i.e. one is 'led down the garden path'; Clark and
Clark 1977: 81). It is much more efficient and certainly not more risky if, on the
basis of contextual information, one is able to pre-conceive what the S/H would
want to say, and thus to see from the beginning which syntactic pattern might be
involved.46
The point of these considerations is that perhaps language interpretation
is a much more flexible process, much more adaptable to external and internal
conditions, and thus also much more variable than many language psychologists
tend to assume (cf. also De Bruin and Wielinga 1982). Thus, the above
discussion is not meant to suggest that an H/S cannot proceed the way Marslen-
Wilson and Tyler propose, but only that an H/S does not always and necessarily
proceed that way. One can safely assume that an H/S always tries to be as
efficient as possible in processing an utterance. But efficiency is relative to the
circumstances. It is different in experimental conditions vs natural conversational
settings, or when the issue is to listen carefully (as in case one is questioned
during an examination, which stimulates bottom-up processing) vs to react
quickly (e.g. in a heated debate, which stimulates top-down processing), or when
one has little advance knowledge of the interaction partner vs long-term
knowledge about his/her views and ways of speaking (the latter makes top-down
predictions more likely). Also, it seems logical to assume that an S/H only uses
those elements of the processing system which are required for the task at hand.
For example, a typist who is simply re-typing a manuscript is likely not to make
a real interpretation of the text, but to limit the depth of processing to the level
of lexical recognition. There may be differences in depth of analysis in proof
reading a text, depending on the task, viz. whether one only checks typos, or also
syntax and style, or even the consistency of the text.
The above also makes it hard to see the point of the very common
hypothesis, also put forward by Marslen-Wilson and Tyler, that language
processing is an automatic and mainly obligatory process. Marslen-Wilson and
Tyler's (1980: 63) argument, specifically with respect to lexical recognition, is
that an H/S, when processing an acoustic-phonetic input, can obviously not
refrain from lexically interpreting it. Even in experimental conditions focusing on
phonetic properties of utterances this seems unavoidable. Yet, this is nothing
more than an expression of the principle which Hrmann (1978) has called
'Sinnkonstanz', viz. that a human in default circumstances spontaneously aims
at making sense of things (not only language). This certainly also applies under
experimental conditions, especially if they involve more than only phonetic
matters, as was the case with the Marslen-Wilson and Tyler experiments.
However, this says little about the nature of the processes involved in an H/S's
search for the meaning of an incoming utterance. Marslen-Wilson and Tyler
themselves postulate a difference in the process of lexical search in processing
'syntactic prose' (as in (71)) vs 'normal prose'. In the former case the H/S can
only rely on the phonological analysis of the incoming signal to select a lexical
item. In the latter case, however, they suggest that expectations with respect to
the meaning, combined with the phonological analysis, allow for a much faster
identification of a lexical item. Even such an observation already questions the
claim as to automaticity and obligatoriness of processing.
There is no question that 'low-level' processing (e.g. the phonological
analysis of an input signal) can be mainly automatic. But the notions of
interactive and flexible processing seem to imply that at higher levels of
processing a considerable goal- and situation-bound variability must be possible.
This is also implied by the fact that an H/S appears able, to a certain degree, to
detect and correct mistakes in an input signal (cf. 2.2.2.2 - see also De Bruin and
Wielinga 1982). In a purely automatic system the processing would probably
block if a mistake is encountered. Thus, it appears that in language use, and
particularly also in language interpretation, the factor control is very essential,
contrary to what Marslen-Wilson (1980) claims.
Returning to the issue of the relationship between production and
interpretation, then, the process of utterance production may be like that of
interpretation in that it can also be assumed to be strongly interactive, since it
constantly has to make use of several sources of information (viz. the sources
discussed in 3.2.1.1). In fact, the same sources should be relevant in both modes,
although they relate differently to the processes in each: in production they are
the source of decisions, in interpretation they have to be reconstructed. Also,
like interpretation, production can be assumed to be rather flexible. There may
be quite some difference in what the flexibility involves, however. For example,
while it is perfectly conceivable that an H/S skips the procedure for interpreting
subject-verb agreement if there are sufficient other indications for interpreting
the utterance, it is inconceivable that an S/H would skip the procedure for
introducing agreement in constructing the utterance, for that would lead to an
ungrammatical utterance which may endanger understanding. In production it
is unavoidable to run through all procedures which apply for realizing the
utterance. But production involves considerable processing flexibility in different
ways, not only in that there is 'rule-changing' creativity (2.2.2.2) in which an S/H
can alter procedures or even create new ones, but also in that an S/H can run
through the production procedures in several ways, depending on the circum
stances.
As argued by Kempen and Hoenkamp (cf. Kempen 1978, 1979, Kempen
and Hoenkamp 1982, 1987), an S/H can no doubt 'incrementally' feed bits and
pieces of the conceptual representation to be encoded in a sentence to the
formulator. (S)he can do so as soon as they are made available by the
'conceptualizer' and even while the latter is continuing to conceive the rest of
the conceptualization to be communicated (see also Levelt 1989). Yet this
procedures holds the risk of causing rather long sentence-internal pauses,
hesitations, and syntactically deficient utterances (cf. also De Smedt and Kempen
1987). Once one has formulated and uttered parts of the message, one is no
longer free while planning the rest since it has to be compatible with the parts
already uttered. Sometimes this may prove difficult if not downright impossible,
for example because the parts already uttered force one to continue a predicate
frame which does not appear suited for the expression of further elements of the
conceptualization (e.g. because it does not provide the appropriate argument
slots). Consequently, one either has to restart the utterance, or has to form a
structurally deficient utterance, or has to form an utterance which inaccurately
expresses one's thoughts (if syntactic wellformedness is more important than
functional accuracy). Hence, it is very unlikely that an S/H always proceeds in
an incremental way. One may do so when sloppiness is not a problem, or under
time pressure. However, when the issue is to be precise and/or careful, for
example in a public interview, one may expect an S/H to plan much more
systematically, probably even fully level by level.
Hence, one has to assume that the production processes are not purely
automatic either, but may involve quite some active control. In this respect,
Kempen (1977) suggests a double-layered control system: processing at 'low'
levels (e.g. the phonological level) would be automatic, performed by demons
which do not demand attention of the central processor and therefore can work
simultaneously (i.e. modular parallel processing), while higher level processing
would involve active decision-making by the central processor or 'monitor' itself,
and, since central attention cannot be divided, would therefore rather be serial.
It is not clear whether such a double-layered control system is sufficient,
however. If one also considers the phenomenon of consciousness, things appear
more complex. As argued in 2.2.3.3, it seems natural to relate the phenomenon
of consciousness to the central processor. Yet, although many linguistic processes
must be assumed to be 'centrally' controlled and not directed by demons,
linguistic knowledge never appears to be conscious. Either we have to assume
that there are intermediate processors between the main processor and the
demons, which do not involve consciousness (at least not to the same degree as
the main processor - introspection suggests that there can be degrees of
awareness or consciousness), and which are responsible for controlled language
processing. Or we have to assume that the demons involved in linguistic
knowledge do not allow the central processor to enter their domain (which
means that it does not really get access to the knowledge), but that some
demons can be influenced by the central processor in whether they perform their
job, while others cannot and decide autonomously on whether they will do
something.47
Also, there is no need to assume that the proportion of automatic and
controlled processing is always constant (both in production and interpretation).
Since the potential influence of sources of information other than the input
structure is likely to diminish the more the production process approaches the
surface structure, it is likely that there is always more active control at deeper
levels and more automatic processing at superficial levels. Procedures leading up
to verbalization, and even lexicalization and syntactic procedures, must often be
strongly sensitive to a variety of factors alongside the input-structure. Many
morphological and phonological processes can be purely input-oriented (i.e. they
can work automatically), however, since they only have to develop the implica
tions of decisions taken earlier on in the process (for instance transposing
function markings of terms into morphological markers expressing it). This
corresponds to the different effects that errors in the choice or performance of
procedures on the different levels can have. Errors at 'deep' levels do not
directly lead to formal mistakes, but rather to functionally maladapted utterances
which have the potential of causing communicative misunderstandings. Errors
at superficial levels, on the other hand, are more likely to lead to formal
mistakes, and are therefore much more easily recognized by the communication
partners.
Nevertheless, there can be deviances from this general tendency in the
distribution of types of control. In acting in general, one can often fall back upon
routines developed through habit formation in well-familiarized situations (i.e.
mainly automatic processing), whereas in unfamiliar situations one can and even
has to return to active, even conscious control of the same types of acts.
Similarly, one can assume that in language processing one can often fall back
upon standard routines at all levels, whereas in other circumstances the role of
active control can be quite important. For example, consider the S/H's capacity
(already discussed in 2.2.2.2) to deliberately make mistakes. This implies that the
control system can interfere even at very specific levels (including the morpho
logical, phonological, and phonetic), and keep the demons from doing their
normal jobs.48 Even a very simple and everyday phenomenon such as adjusting
one's dialectical pronunciation of the native language to a more general variant
involves active interference at very low levels.
In any case, as to the overall architecture of the control system, general
considerations lead Kempen to suggest something like 'blackboard control', i.e.
roughly the same type of control at which we arrived in our analysis of 'floating
elements' in natural language syntax (3.1.3.3). If one also takes into account the
strongly interactive nature of processing and the fact that, as has been shown in
the Hearsay- and Hwim-parsers (see above), blackboard control is well suited
for such processing, we have several indications here for the cognitive plausibility
of this control type. We can also assume that the same control types and systems
are valid in both production and interpretation, although it remains to be seen
how we can account for the above noted differences in processing flexibility in
the two modes.
All of this still leaves us with the question of the relationship between the
linguistic knowledge involved in production and in interpretation. Beyond the
rather safe suggestion that the lexicon is shared by both modes, an answer does
not seem feasible at present. In fact, an answer requires a rather precise view
of what is going on in each mode: only on the basis of rather detailed hypothe
ses about the knowledge, control, and memory structures involved in each can
we start to see what correlations may exist.49 Our current knowledge is
probably too insufficient for this. At least as far as production is concerned, this
seems to be the case indeed. Since FG as discussed so far is molded in the
productive mode, I will, for the rest of this chapter, concentrate on the issue of
the knowledge systems involved in language production and will not bother with
the issue of language interpretation anymore.
3.2.3. Knowledge systems involved in language production
3.2.3.1. Grammar and lexicon

How, then, should we conceive of the knowledge systems involved in a more
encompassing model of language generation? More specifically, if we accept an
FG-type of linguistic grammar as a plausible concept of the 'post-lexical'
procedures, how should we conceive of the pre-lexical systems which lead to the
construction of a predication starting from conceptual representations of the
type discussed in 3.2.1 and appropriately encoding50 them, and how should the
linguistic system be adapted so as to provide for a smooth continuation of the
pre-lexical systems? The following proposal, called Functional Procedural
Grammar (FPG),51 is a tentative suggestion for an answer, which is entirely
open for debate. I will only attempt to inventory the types of knowledge systems
which can be assumed to be involved. I will not try to formalize anything: the
proposal is still much too vague and general for that (cf. 1.3.2). A general
overview of the proposal is given in (75) below. As to the presentation of the
sets of procedures below, it is important to bear in mind that, as argued in the
previous section, there is not just one single way to run through them. Of course,
processing does have to respect dependency relations between procedures (for
example, it is impossible to assign a grammatical subject marking to a term as
long as it has not been assigned the syntactic function of subject). The general
ordering of sets of procedures below intends to roughly reflect such dependency
relations: procedures at lower levels are always dependent upon procedures at
higher levels. The dependency relations between sets of procedures at the same
level are not so obvious, as will appear from the discussion.
I start the present discussion at the point in cognitive processing where the
S/H has come to terms with the general factors assumed relevant for communi
cation (a general intention, a relevant SoA and the area of conceptual
information related to it, the conceptual information related to the social and
interpersonal relationship with the H/S, the conceptual information pertaining
to the properties of the setting, and the 'partner model' in each of these areas;
cf. 1.2, 3.2.1.1), and is ready to initiate communicative action. By this I do not
wish to suggest that there is a strict line to be drawn between general cognitive
processing (reasoning) and action oriented processing, or that the latter can
occur without the former. That would go against the views expressed in 1.2
about the hierarchical nature of thinking and acting and the complex interaction
of the various factors involved. However, I do wish to suggest that the knowledge
systems for the production of an action are different from (though at some
points interrelated with - cf. 3.2.3.2) the knowledge systems involved in reasoning
and general decision making, which may lead up to planning an action. Since I
am only concerned with the sets of procedures involved in verbal action here,
I draw this line for the sake of the exposition.52
As to the general organization of the production systems, one does not
have to agree with the tendency existing in many models of language production
to sharply distinguish between S/H's determination of what to say and how to say
it (cf. Dale 1988, Levelt 1989). As will appear below, there is probably a gradual
movement from the former to the latter, and it is often difficult to say which of
the two elements is involved.
In relation to this point, the discussions so far may have raised the
suggestion that the language production systems can be roughly subdivided in
pre-lexical or pre-verbal procedures resulting in the choice of lexical patterns
(predications in FG-terms), and 'post-lexical' or verbal procedures shaping these
lexical patterns into surface structures in the format of temporally ordered com
mands for the articulatory organs. Yet this distinction cannot be taken strictly.
That is reflected in the fact that it is questionable whether accessing the lexicon
(i.e. the stock of verbal means available in a language) can be situated at one
particular place in the grammar. As will appear below, at least as far as the
stock of function words and morphemes is concerned, it is likely that the
grammar should be able to get access to it at different levels: already before the
predication is constituted (e.g. for certain sentence connectors), but also still at
very superficial levels (e.g. for flexion morphemes or prepositions). Even for the
stock of content lexical items it is likely that it has to be accessible from very
early in the grammar onward, for even with respect to processes which in them
selves should be characterized as pre-verbal an S/H should sometimes be able
to take into account the lexical means available. (This obviously still leaves open
the question whether S/H already selects lexical elements at that point, or only
consults the possibilities.) Finally, language psychologists have found that the
phonological patterns of lexical items become available only very late in the
production process (cf. Garrett 1980, Kempen and Huijbers 1983, Harley 1984).
This might require access to the lexicon once again in a late stage of the
production process (but see Levelt and Schriefers 1987). In summary, the
process of verbalization appears to be continuous. There must be a certain point
in the grammar at which S/H must finally decide on the verbal pattern to be
used (i.e. during predication - see (75) and 3.2.3.3). But the lexicon has to be ac
cessible at different places in the grammar, and thus must be regarded as a
component parallel to the gramman.
(75)
UI
intention SoA social relation setting
turn type textualizing

SN
illocution sentencing

SSA
predication
lexicon
predication
expression
surface str.
Since the lexicon has principles of internal organization separate from the
grammar, let's first consider some issues pertaining to the former. While one can
subscribe to the FG view on the structure of individual entries (the predicate
frames), as well as on the systems for predicate and term formation, there are
reasons for disagreement at other points, and some additions seem in order as
well.
(i) Selection restrictions

There is no clear reason for having selection restrictions in the lexicon. The
notion of semantic selection restrictions has been heavily criticized, mainly by
language psychologists, on the basis of the consideration that the problems for
which it is invoked are of a conceptual, and not of a linguistic nature. Hence,
they should be dealt with at the level of the general conceptual knowledge, not
in the grammar. Dik (1978a: 45), does not subscribe to this view, but his (1989a:
76ff) arguments do not really counter the criticism. For example, he argues that
since the predicate blond can only be used to refer to hair, characterizing it
cannot be done without somehow mentioning the latter lexical item in the
predicate frame of blond. However, the fact that blond can only be used for hair
is obviously an element of its meaning. The concept 'blond', to which the lexical
item refers or 'points' (see below), is only related to the conceptual domain of
hair and does not belong to the normal conceptual domain of colors (in fact, no
one will think of blond if one is thinking of colors). One can argue similarly for
the other examples and arguments provided by Dik: all of them can be
explained in terms of concepts and of (changing) relationships between lexical
items and conceptual domains. Arguments in support of the view that selection
restrictions are superfluous can be found, e.g., in Levelt (1973c), Miller and
Johnson-Laird (1976), Johnson-Laird (1977b, 1983), Schlesinger (1977), Miller
(1978b). Moreover, Steen (1985) argues convincingly for the claim that selection
restrictions are not well suited to account for metaphor in language53 (see also
Lakoff and Johnson 1980 and Lakoff 1987, who treat metaphors as conceptual
phenomena).
Selection restrictions seem to be a typical example of principles which have
been introduced in linguistic grammars due to a limitation of attention to syntax.
Such a limitation unavoidably leads to 'imperialism', in the sense that attempts
are made to handle phenomena in the grammar which really do not belong in
its realm (cf. 1.1.1). In any case, as long as there are no good arguments to show
that an S/H's general knowledge of the world is not sufficient to account for the
occurrence of semantic restrictions on the insertion of terms into predicate
frames, preference should be given to the claim that no selection restrictions are
required in the linguistic lexicon. This is a matter of the simplicity of the overall
model: specifying them in the lexicon would be redundant, since the matter has
to be handled at the conceptual level anyway. (Cf. also Comrie 1980 and Dik
1980c.)
(ii) Meaning definitions

Meaning definitions appear superfluous in the lexicon as well. Again, it seems
more appropriate to account for the definition of a lexical item in terms of the
characterization in S/H's knowledge of the world of the concept or conceptual
cluster to which the lexical item is linked. One observation in favor of this view
is that people are able to provide different definitions of one lexical item
depending on the context and requirements. For example, depending on the cir
cumstances, one individual can provide quite different definitions of a govern-
ment, starting from something like 'a board of ministers who rule a state' to
something like 'a bunch of political friends earning a lot of money'. Definition
is extremely flexible, both in quantity and quality (content) of the information

provided. It would not only be hard to account for this in the lexicon, it would
also be highly redundant because all of this information has to be present in the
S/H's knowledge of the world (his 'encyclopedia') anyway. Therefore, it seems
better to handle the determination of the meaning of an item in the lexicon in
terms of a reference (a 'pointer', as it is called in AI - cf. e.g. Schank et al. 1975,
Kempen 1978) to the concept or conceptual domain in the encyclopedia of
which it is the label. Definition is then a matter of describing the situation of this
concept or conceptual domain in a wider network in terms of the aspects
relevant to the communicative situation and hence present in the UI.
(iii) Fixed expressions

The lexicon also has to contain proverbs and other formulaic expressions, idioms,
etc., which most probably are stored as fully specified predications. In fact, it is
one of the major claims of Construction Grammar (Fillmore 1988, 1989,
Fillmore et al. 1988) and of Cognitive Grammar (Langacker 1987) that there are
many more linguistic structures which have to be handled in this way than is
generally assumed. There may also be degrees of fixation of constructions: for
example, kick the bucket is certainly more fixed than the more ..., the more ....
Some language psychologists have even argued that frequently used utterances
(even non-idiomatic ones) may simply be stored as more or less fully specified
structures, rather than being constructed over and again each time they are used
by an S/H. If that is true, presumably they are stored in the lexicon as well.
(iv) Lexeme and phonological form

In view of the observation that the phonological form of a lexical item becomes
available only late in the process of language production, the distinction between
a lexical item as a structural phenomenon and its phonological realization (or
realizations, in the case of morphologically irregular variants) must have some
manifestation in the lexicon. How this should be realized in terms of the
organization of the lexicon is an open matter (cf. Harley 1984 for a discussion).
(v) Function words and content words

Another matter of organization is the relationship between the stock of function
morphemes and the stock of content morphemes in a language. Observations in
theoretical linguistics about the typical characteristics of each of these categories
(open vs closed class elements, primarily semantic vs primarily syntactic
motivation, etc.) suggest a clear differentiation between them. The observation
that the function words, e.g. in agrammatism in Broca aphasia, appear to be a
differentially disturbed category (cf. e.g. Saffran et al. 1980) points in the same
direction, as does the observation that function words appear to behave

differently from content words in non-pathological speech errors (e.g. Garrett
1980). Maybe these are arguments for postulating a separate lexicon for the
function words, but how to realize this is an open question.
(vi) The organization of content lexemes

Another open question is how the content morphemes should be organized in
their lexicon. Some speech errors (word substitutions holding a clear phonologi
cal relationship to the target word) suggest a phonological ordering principle,
others suggest a semantic ordering principle (cf. Garrett 1980). Of course, this
reminds one of the observation in (iv) above that the lexical items and their
phonological shapes become available at different places in the grammar. Does
this mean that these data are stored in different places or in different ways? Or
should these speech errors be explained in terms different from the organization
of the lexicon (e.g. in terms of processes in the grammar)? (Cf. e.g. Harley
1984.)
Let's now concentrate on the question of which types of procedures should

be assumed to be involved in the grammar itself.
3.2.3.2. Textualizing
The first step to be taken in producing language is to construct a structure which
contains the relevant information about the SoA which the S/H wants to
communicate to the H/S. Let's call this conceptual structure the 'situational
network' (SN). The SN is a representation of variable size, but one which can
certainly be larger than what is normally expressed in one single linguistic utter
ance. In fact, as argued above (e.g. 3.1.3) an integrated system of language
production must be able to handle the level of the composition and organization
of the discourse (text or conversation), alongside the level of the constitution of
the utterance. While in the organization of an utterance the verbal means of a
language have a crucial role to play, this is much less the case for the construc
tion of a discourse. This suggests that discourse organization mainly has to be
handled at a pre-verbal level in the grammar. Thus, the processes leading to the
composition of the predication are responsible, not only for the encoding of a
conceptualization into a verbal pattern, but also for the textual cohesion of a
series of related utterances, and for the adaptedness of such a textual complex
in the wider context of the running discourse.
Therefore, we can assume that the SN covers an integrated representation
of the information which the S/H would like to express during one 'turn' in the
discourse. It is also the structure the S/H will constantly be drawing upon while
producing individual utterances as parts of the turn (see 3.2.3.3). By 'turn' I

mean one period in a discourse during which the S/H has the floor without being
interrupted by the H/S (probably except for back channel cues) and without
changing the theme. This can be considered the basic unit at the discourse level
(Franck 1980).
The information in the SN is based upon the information about the SoA
in the UI of the S/H (cf. 3.2.1.1), but is basically adapted to the characteristics
of the specific situation. We can assume that the procedures involved in
constructing the SN, which are termed the 'textualizing' procedures in (75), are
valid for discourse in general, i.e. for both conversation and monologue/text.
There are no doubt differences in the actual ways these procedures are applied
in these discourse types due to the typical characteristics of each. Yet, the
principles of the system, the basic procedures, can be taken to be identical (i.e.
both types are handled in terms of one and the same cognitive system). At least
the following sets of procedures can be distinguished.
(i) Composing, monitoring, and filtering the SN

The S/H has to decide which information in the UI is going to be included in the
turn (s)he is about to produce. The notion of turn is not easy to pin down,
however, and correspondingly, the exact range of procedures involved here is not
a priori evident. A first matter is the potential size of a turn (and thus of the
SN). In normal conversation turns tend to be relatively short (sometimes they
involve only one utterance). But in principle a full-fledged monologue (a 'text')
can be considered a turn as well: during a one-hour speech one S/H has the
floor without interruption and without changing the central theme. Yet, it seems
unlikely that all information transmitted during this speech will be present in the
SN at once. Most probably there has to be an upper boundary to the size of the
SN (though it is not a priori clear where - probably research on types of memo
ries and their capacities might help here).
This would imply that there can be a lively interaction between SN and UI.
In case of long monologues, some core conceptualization of the central theme
may remain within the SN, as a reference point for different, more limited,
conceptualizations of sub-themes. Such sub-conceptualizations might be intro
duced into the SN one at a time, each one remaining until it has been formu
lated and can be replaced by another sub-conceptualization from the UI
(certainly, this can be a gradual and continuous process). In selecting sub-
conceptualizations from the UI the S/H must introduce a certain order: the
natural order of consecutive events is one possibility, but there are no doubt
several other possibilities. In formulating the processes involved here, one could
learn from the principles proposed by Van Dijk and Kintsch concerning the
structuring and treatment of information in discourse understanding (cf. e.g. Van

Dijk 1978, 1981, Kintsch and Van Dijk 1978, Van Dijk and Kintsch 1983), or in
the literature on the macro-structural organization of texts (especially narratives:
e.g. Rumelhart 1975, De Beaugrande 1980, 1984, Wilensky 1983, J.Mandler
1984). Of course, there has to be some 'bookkeeping' of what has already been
communicated and probably also how it has been communicated, and of what
is still to come.
The situation may be comparable in the case of a conversation on one cen
tral theme. Here again, some core conceptualization concerning the theme of
the conversation can be assumed to remain in the SN, while for each turn new
sub-conceptualizations are introduced from the UI. There is a wealth of litera
ture (from conversational and discourse analysis, but also from psychological and
AI research into dialogue models) which can be brought to bear on a detailed
development of these aspects with respect to conversation (procedures of turn-
taking, the relations between successive turns in terms of content and structure,
and also the way turns are organized internally - see 3.2.3.3), including Sacks et
al. (1974), Weiner and Goodenough (1977), Power (1979), Hobbs and Evans
(1980), Fortescue (1980), Franck (1980), Weijdema et al. (1982), Levinson
(1983), Lehnert (1984b).
Another problem in determining the notion of turn is the matter of
internal continuity. Within one set of utterances concerning a specific theme an
S/H can make excursions which are only indirectly related to the main theme.
However, this does not have to involve a new turn. The information for this
excursion may be present in the SN along with the information more closely
related to the theme, without requiring an exchange with the UI. Of course, if
S/H at a certain moment concludes a theme and switches to a completely new
one, there is a switch in turns, even if there has not been an interruption by H/S
in the meantime. Cognitively, one can imagine this to be a procedure in which
S/H clears the SN completely and reinstalls it for the theme of the new turn.
Another situation occurs if H/S interrupts S/H, for example because (s)he did
not understand a specific word. Again it is doubtful whether this involves a
change in turns. In any case, there is not much of a change in the S/H's S,
though there do have to be procedures regulating such 'side-sequences'.
Probably one should also allow for the contents of the SN to be changed
in the middle of the process of verbalizing what basically constitutes one turn
(changeability and flexibility of the SN might help to explain the fact that in real
discourse it is often very hard to delimit turns clearly). Such changes may even
be made necessary by the process of verbalization itself. The communicative
situation slightly changes with every single utterance, and non-verbal signals from
the H/S, e.g., which do not really interrupt H/S's turn, may still force the latter
to revise what is informatively relevant, and how it is relevant.
The information in the SN is a selection of what is available in the UI:
normally S/H does not want to provide H/S with all information concerning the
SoA available in S/H's UI, but only with the most relevant information.
Moreover, in many cases information in the SN is directly derived from the UI,
but it can also deviate from the information in the UI: it can be simplified, or
somewhat adapted, or even be radically different (cf. e.g. the phenomenon of
lying). The S/H's choices in this respect are traditionally explained in terms of
adequacy, relevance, etc., from the perspective of the mutual knowledge about
the SoA of the interlocutors, i.e. matters of the type of Grice's (1975, 1989)
maxims and developments of it (e.g. Levinson 1987). Yet, this explanation may
be too narrow: the decisions may also be determined by the other factors
involved in communication (cf. 1.2). Thus, if according to S/H's partner model
the H/S knows nothing at all about a certain complex SoA, it may be better to
simplify matters and make them understandable by omitting several aspects of
the problem, or even by presenting it in a slightly deviant way, even if this means
that the SoA expressed is not absolutely accurate relative to the real representa
tion in the UI anymore. However, equally feasible reasons for eliminating or
adapting information from the UI are: that some pieces of information may be
unacceptable to the H/S, and might therefore better be omitted if the social
relationship is not to be disturbed; or that the setting may disallow certain
information which might otherwise have been relevant; or that certain types of
information may be unfavorable in the light of the intentions the S/H is pursuing,
or that the S/H may simply intend to mislead the H/S by providing wrong in
formation.
The S/H will not verbalize all the selected information either. Given
general knowledge of the world and a normal reasoning capacity, an H/S is able
to form a much more complete picture of the SoA communicated on the basis
of a limited set of explicitly mentioned data (this is a point where reasoning and
language processing interfere - cf. 3.2.3.1). Cf. the phenomenon of 'bridging' -
Clark (1977), Clark and Haviland (1977). The huge literature on implications,
implicatures, presuppositions, background-information, inferencing, etc. can also
be brought to bear on this matter. The only difficulty is that the majority of this
research has been conceived from the perspective of language understanding,
and our knowledge of the S/H's strategies is very limited. Anyway, profiting from
this capacity, in composing and developing the SN the S/H can filter out those
elements of the information which can be inferred without having been
verbalized explicitly. This process no doubt interacts with the procedures determ-
ining the internal informational structure (theme/rheme, given/ new, etc. - see
below) of the SN (cf. Clark and Haviland 1977). Thematic and given information
can be eliminated more readily than rhematic and new information. It is also
likely that these filtering procedures can be applied in different stages of the
process of verbalization and production of the turn. For example, they are cer
tainly useful when selecting small chunks of information from the SN to be
expressed in singular utterances (see below).
Filtering interferes in different ways with the question of correspondence
of the information in the SN and the UI mentioned above. Consider lying. Direct
lies occur when there is an apparent and deliberate mismatch between the
information in SN and UI without the S/H indicating this. Indirect lies occur
when S/H suggests information which does not correspond to the information in
the UI by specifically combining facts to be mentioned explicitly which in
themselves do correspond to the information in the UI, or when S/H leaves out
information which in itself is strongly marked in terms of its informational value
(givenness/newness, saliency, etc.), i.e. which is of crucial importance in the given
communicative situation. Or consider metaphor. While 'dead' metaphors (i.e.
ones that are no longer consciously handled as metaphors) are no doubt integra
ted in the encyclopedia, S/H can also creatively form metaphors. In that case,
it involves a specific type of reconception of information from the UI in the SN.
Such metaphorical representations have a special status for the filtering
procedures (as also for the procedures for assigning pragmatic functions),
however, since H/S cannot be assumed to be able to infer such creative repre
sentations on the basis of general knowledge and reasoning.
(ii) Organizing the SN information-structurally

There is a growing awareness in the literature (even in FG - cf. 3.1) that
pragmatic functions even at the sentence level are strongly discourse-bound.
Hence, the sentence-level pragmatic functions have to be anchored at the level
of discourse organization. This leads us into a very tricky domain, about which
there is much discussion (witness the huge amount of literature) and little
consensus. At least the following distinctions seem necessary.
Discourse is based on the exchange of information on some 'theme' (also
called 'discourse topic'), i.e. some SoA which is mutually known by or to be
introduced between S/H and H/S, and which is recognized as central to the
current communicative situation. In the SN this central piece of information is
marked as theme, and the different pieces of information related to it by the
S/H as being relevant to the H/S are marked as rhematic (a survey of literature
on the notions theme/rheme is to be found in Lutz 1981). The theme/rheme dis
tinction does not have to be a simple one. Alongside the theme there can be
sub-themes, which at a certain moment are rhematic relative to the main theme,
but which, after having been introduced, can become thematic relative to the
other information in the SN (cf. the discussion above concerning the position of
the SN in larger texts). This also suggests that thematicity vs rhematicity is not
an absolute but a gradual value, and accounting for it may require the
introduction of some kind of (measurable?) scale.
The theme can behave in different ways in discourse. It can remain
implicit; it can be introduced once or several times in individual utterances; the
S/H can also mention it explicitly in a separate construction at the beginning of
the turn (which may result in what is called a 'theme' in FG: as for Paris ...).
S/H's selection of one of these options will be influenced by whether (s)he
assumes that the H/S knows about the theme or that it is sufficiently prominent
in the running discourse, and these observations suggest the introduction of at
least one more type of pragmatic function at this level, viz. given/new (e.g.
Prince 1981). The given/new status of information is a function of its degree of
introduction in the running discourse, and hence in the UI of the interaction
partners (presence in the encyclopedic knowledge is not sufficient: the
knowledge has to be activated54 - cf. Chafe 1976). The exact relationship
between the values of the functions theme/rheme and given/new is not a priori
evident. Given/new may again be a gradual rather than a binary categorization
(Hannay 1985a).
Especially the givenness/newness of information is constantly changing in
the course of verbalization of the turn. Once a piece of information from the SN
has been communicated, it acquires a much higher degree of givenness. If it is
not mentioned again, its givenness declines again in the course of the discourse.
Probably still other information-structural or related elements have to be
introduced at this level. One phenomenon which immediately comes to mind is
some notion of psychological 'saliency' for an S/H of parts of the information,
in general or specifically with respect to an H/S, and independently of its
givenness/newness or thematicity/rhematicity. For example, the fate of someone
one loves very much is simply much more salient than the fate of some person
one hardly knows, no matter what. (Some elements which might play a role here
are discussed in Osgood 1980.)
(iii) Determining the status of the information in the SN

The information in the SN may need qualification along several dimensions,
corresponding to the elements discussed in 3.2.1.3. Sometimes characterizations
in this respect can be adopted from the UI, but this is not necessarily the case:
S/H can want to deviate from the qualifications in the UI for reasons similar to
those which can make him/her change the SoA to be expressed (see above).
As in the UI, these characterizations are not always valid for the entire SN:
they can have scope over chunks of information of variable size within the SN.
Hence, these characterizations may have to be organized not only in terms of
their hierarchization relative to each other (cf. 3.1.1, 3.2.1.3), but also in terms
of various scope levels. Which characterizations are made at which level depends
on the degree of homogeneity of some chunk of information for the different
dimensions involved. Sometimes all the basic units of information require a
separate characterization, sometimes one characterization can be given for a
large chunk, with a contrastive marking at lower levels for those subsections of
the information which deviate from the characterization at the higher level.
For the sake of clarity, here is an overview of the qualifications involved
(in approximate hierarchical order, from low to high, and disregarding potential
fluctuations in the relative ordering - cf. 3.2.1.3).
(a) Space (if one should consider this a meta-qualification). Characterizations in
space can be relative to the location of the S/H or relative to some location
elsewhere in space, in terms of this location itself or of some point relative to
this location.
(b) Aspect, i.e. qualification of the present occurrence of a SoA relative to some
projected final state. This can involve imperfectivity or perfectivity, but also
inchoativity, continuativity, resultativity, terminativity, etc. Hengeveld (1989)
distinguishes two types of aspect at different hierarchical levels, and his
arguments make clear that it is likely indeed that there is need for an hierarchi
zation among types of aspectivity markings.
(c) Time. Characterizations of time can be relative to the here and now of the
communication, or relative to some deviant point on the time axis, and it can be
in terms of 'on', 'before', or 'after'. They can be precise (e.g. in 1585) or vague
(e.g. long before you were born).
(d) Deontic modality, i.e. qualification of the SoA in terms of its desirability or
(moral) acceptability.
(e) Realis/irrealis, i.e. qualification of a SoA for being in the actual world or in
some non-real possible world. This has to be made explicit, especially in the case
of fictional information (fairy tales, invented stories, but also suppositions, etc.),
if the S/H wants to be honest, of course. Characterization as fiction is also pos
sible through special characterizations in the dimensions of time and space (this
strategy is often used in e.g. fairy tales).
(f) Polarity: if the information in the SN is in conflict with the information in the
UI, the problem of negation arises (cf. e.g. Horn 1978b, 1989, Givn 1979).
(g) Epistemic modality, i.e. qualification in terms of the likelihood of the SoA
being the case or not in the possible world under consideration.55
(h) Evidentiality: characterization of the information in terms of the source (cf.

e.g. Chafe and Nichols (eds.) 1986). This applies among others if the S/H
mentions knowledge, beliefs, opinions, etc. of some other person (cf. the 'layered
dimension' discussed in Clark 1987). Correlated with this is the matter of direct
or indirect quotation, 'erlebte Rede', etc.
(i) Attitude: characterization in terms of the strictly personal feelings of the S/H
(for example, emotional involvement in terms of degrees of positive or negative
experience).
Note that several of these qualifications are closely related. For example, there
is an intuitive link between attitude and deontic modality, and between
realis/irrealis, epistemic modality, and polarity, and between time and certain
types of aspect. The meaning of these intuitions is unclear.
Probably, these qualifications also have to be marked in terms of pragmatic
function categories. For example, negation and modality are often very rhematic.
See also 3.3 on this.
The above procedures are oriented towards manipulating the conceptual

representation constituting the SoA which is talked about in one turn. In view
of S/H's general intention, and also of the properties of the setting and the social
relationship, and in interrelation with the specific information included in the
SN, the SN as a whole also has to be characterized for what (for lack of a better
term) we may call the 'turn type' S/H wants to produce: is the purpose to inform
H/S, to persuade H/S, to elicit information from H/S, to narrate, to play
language games, etc.? This is the equivalent at the discourse level of the
characterization for illocutionary force at the utterance level, i.e. a specification
for the communicative intention. If the information in the turn covers the entire
relevant SoA in the UI, the relationship between general intention and 'turn
type' may be rather straightforward. If the SN only covers a section of the entire
relevant SoA in the UI, however, S/H has to follow strategies in pursuing the
general intention, and each turn (and each utterance in it - see below) will have
a separate purpose, which is not necessarily simply related to the overall
purpose. For example, if the overall goal is to persuade someone of the
appropriateness of a point of view, one can use a number of consecutive turns
to raise problems and ask H/S for an answer, to end in a final turn in which the
central claim, which answers all the questions, is stated. Of course, the intention
with a turn can also deviate from the general intention if S/H wants to hide the
real intention or fake an intention.
Probably something similar should be assumed in the areas of social
relationship and setting, which may need specific markings per SN, and which
may deviate from the values for these factors in the UI, if S/H wants to feign
something. Some elements involved here have been discussed in 1.2, but all of
this is much less understood than matters related to the intentional component.
3.2.3.3. From conceptualization to utterance

The SN constitutes the conceptual structure on which the further production
process builds. The information involved now has to be adapted and restructured
into the format of an actual utterance. In this process, one can roughly
distinguish three successive sets of procedures, viz. (i) a set of procedures
responsible for 'sentencing', i.e. the rough conception of the chunks of
information to be included in separate utterances (the elements involved here
are comparable to the aspects of the process of verbalization discussed in Chafe
1977a, b); (ii) a set responsible for 'predication', i.e. the determination of a ver
bal pattern (the predication) which can represent the conceptualization; and (iii)
a set responsible for 'expression', i.e. the realization of the predication in a
surface structure. Next to these information-oriented procedures, there are again
developments in the areas of the intention, social relationship, and setting. Let's
consider these first.
In the area of intention, each utterance (or rather each chunk of
information which will ultimately result in an utterance) has to be assigned an
illocutionary force, of course in function of the turn type and the information
present (but also, as always, of the social relationship and the setting). This rela
tionship need not be simple, however. For instance, in a turn aiming to acquire
information from the H/S (e.g. as a result of some 'gap' in the S/H's conceptual
knowledge about a SoA), the S/H may have to form a series of statements to
introduce the 'conceptual context' of the gap, which can finally lead to the
formulation of a question indicating the gap itself. Hence, the determination of
the illocutionary force of a 'singular SoA' (SSA - see (i) below) is also dependent
on the exact nature of the information contained in the SSA.
There is reason to believe that even an SSA which is going to be expressed
as a subordinate clause receives an illocutionary force. The difference between
(76a-b),
(76) a. It is not clear whether he has done it

b. It is not clear that he has done it
could be due to questioning illocution in the embedded clause in (76a) as

opposed to declarative illocution in (76b). However, while the illocutionary force
of an SSA to be realized as a main clause can be expressed in many ways, this
is much less the case for embedded clauses.56
The matter of illocution (a major topic of literature, including the entirety
of speech act theory) is difficult to detach from matters of its expression: though
the latter involves decisions to be made at lower levels in the grammar, it cer
tainly also has impact on the exact way of characterizing the illocutionary force
itself. There are at least two types of (nevertheless clearly related) problems, viz.
the difference between primary and secondary forms of expression, and between
direct and indirect forms of expression.
Primary or secondary expressions of illocution involve the question whether
the set of procedures for expressing some 'prototypical' illocutionary force is
used consistently or not. Consider the difference between (77a-b).
(77) a. Are you going to Vienna this year?

b. You are going to Vienna this year?
(77a) is the prototypical expression for the illocution 'question' (i.e. a primary
expression of a question), while (77b) combines the expressive devices of a
question (the intonation) and a statement (the sentence pattern) in order to
express that which is nevertheless also a question (i.e. a secondary expression of
a question). Thus, we can distinguish a number of 'prototypical' illocutions which
have grammaticalized ways of expression, and which probably differ from lan
guage to language. For Dutch, English and German, e.g., there are only three:
statement, question (involving two types, yes/no and question word question),
and order. Apparently, statements can only occur with primary expressions.
Questions and orders can also occur with secondary expressions. All other
illocutionary forces may occur only with 'secondary' expressions (although this
sounds a bit strange for they do not have a 'primary' expression form at all).
The exact range of possibilities for mapping expression forms and
illocutions is unclear. Often various secondary expressions can be used for what
might be considered one illocution, often a single secondary expression form can
be used to express various illocutions. In some languages, the possibilities for
creating nuance in expression are considerable. In Dutch and German for
instance, the use of particles such as maar, wel, and eens, or aber, wohl and mal
allows the S/H to express an enormous range of illocutionary variants. How the
S/H decides in these matters is an open question, but among the influencing
factors, there appears to be not only the exact value of the illocution, and its
intensity (obviously, illocutionary forces can be strong or weak), but also the
characterization of the information in the SSA in terms of modality and probably
other qualifications. The difference between (77a) and (77b) above, for instance,
is clearly due to the degree to which the S/H has an opinion about the chances
that the person questioned will go to Vienna. (77a) does not entail a priori
assumptions, while (77b) suggests that the S/H estimates that chances of
receiving a positive answer are high.

The distinction between direct and indirect expressions of an illocutionary
force involves the question whether the S/H uses procedures inherent to the
illocution to be expressed at all, or rather applies a set of procedures which are
'normally' used for the expression of a different illocution. The classical example
of a direct/indirect pair is (78).
(78) a. Give me the sugar

b. Could you give me the sugar (please)?
Here too, the procedures determining the relationship between the illocution
and the expression which is chosen are unknown. It is certain, however, that
particularly S/H's knowledge of the social context and the setting is decisive in
motivating the choice: the difference between (78a) and (78b) is obviously
mainly a matter of politeness.
Performatives are a separate problem. Performative expression of an
illocution seems rather exceptional (cf. Weijdema et al. 1982), and the conditions
for its occurrence are not always obvious. Its selection is likely to be an
indication of the importance of the SSA and its realization in terms of the
illocutionary force, and thus of the intensity of the illocution. (79), for example,
(79) I ask you (earnestly) to close the door
is usually a repeated request stressing that the S/H considers it important that
the act requested be performed. The relationship between the performative verb
chosen and the illocution intended is not necessarily simple. Though (80)
suggests a simple statement, it can be used as a warning or an order.
(80) I tell you not to go
Predicates occurring as performatives can obviously also be used non-perform-

atively, e.g. in meta-communicative or meta-linguistic utterances. In this case
they involve some aspect of the information contained in the SSA (including
qualifications), but not the matter of its illocutionary force. It is not always obvi
ous whether a predicate is being used performatively or not, as is shown in (81).
(81a) is rather performative, whereas (81b) (which involves modality) is not.
(81) a. May I ask you to close the door?

b. Let's say he is sick (though this is disputable)
Also in the areas of the social dimension and the dimension of the setting,
each utterance needs a specification of values, which then have a direct effect
on the selection of expressive devices. Issues to be dealt with here are: which
register or language variety is to be chosen in verbalizing, which politeness value
should be applied, which level of formality is required, etc. Such elements have
an important effect, for example, on selecting lexical items and syntactic patterns
(cf. the difference between hey pal, gimme a smoke, will you? and I am sorry sir,
but could you lend me a cigarette, please?; or the wide variety of forms of address
in languages such as Japanese, all with different politeness values). As
mentioned in 1.2, there is a great deal of literature on the effect of these issues
on linguistic structure, yet there has been hardly any consideration of these
issues in terms of their role and position in a process model.
Let's now consider the procedures for transforming the conceptualization
in the SN into an actual utterance.
(i) Sentencing
Sentencing involves the construction of units of conceptual information of the
size conceivable within one utterance. Let's call such a unit a 'singular SoA'
(SSA).57 Taking the step from the complex conceptualization in the SN to
singular utterances, the procedures at this level have the task, not only to
conceive each individual SSA, but also to characterize its proper status within
the turn and to keep it in agreement with the other SSAs in the turn. They are
responsible for appropriately integrating information in singular SSAs in view of
the information in the SN to be expressed in related SSAs.
The impact of these procedures can seriously differ depending on the
complexity of the SN, although the correlation does not have to be simple. As
mentioned, the SN can be small, hence it is sometimes possible to integrate all
information in one SSA. Whether this is feasible and desirable is a decision to
be taken at this level (in composing the SN the question of how something can
be expressed verbally is of no concern at all). Even for a simple chunk of
information which could normally be expressed in one utterance, S/H can
nevertheless decide to spread it over a number of utterances. Similarly, S/H can
decide to utter in one single utterance a more complex chunk of information
which would normally be split up. These decisions can be situation-specific (e.g.
determined by the supposed capabilities of the H/S), but are often also a matter
of style and habit (individual, or cultural - cf. the difference in complexity of an
average German utterance as compared to an average English utterance).
Anyway, it seems intuitively plausible to assume that as a general rule, the
simpler the information in the SN, the simpler the application of the sentencing
procedures.
We assume that the SSA is still a non-verbal, conceptual representation,
but in conceiving it S/H can already start taking the lexical possibilities of lan
guage into account (including possibilities for extending predicate frames with
satellites and for expressing meta-markers such as qualifications by grammatical
means). This can influence decisions as to which conceptual elements may be
combined in one SSA, and how this may be done. The insight that some chunk
of information will be hard to express verbally may lead S/H to adapt the
conceptual structure in the SSA, in order to facilitate verbalization. This could
(e.g.) be in the form of introducing a comparison (there was an object there which
was something like a chair), or a metaphorical relation (this is really a laughing
painting, I can't call it anything else). Note that this implies that even at this level
of processing a S/H can still reinterpret or modify information (cf. also Chafe
1977b).
One can wonder whether one should assume that an S/H can already select
lexical items at this early stage, or rather has to postpone this until the moment
the predication is definitively shaped. Maybe both possibilities are available
depending upon the circumstances. The latter is probably the default procedure,
but the former might occur if the S/H absolutely insists on using some particular
word and makes the organization and verbalization of the rest of the information
dependent on the possibility of including it.
Anyway, from this level on procedures and representations gradually
become language-specific. Since the inventory of expressive devices is idiosyn
cratic for each language, we have to assume that the mechanisms leading to the
selection of these expressive devices are language-specific too (though they must
be based on universal principles, of course, in the same way as expressive
devices are based on universal principles).
At least the following sets of procedures have to be postulated at this level.
(a) Determining the ordering of the SSAs

In conceiving SSAs, S/H has to decide on how to introduce a consecutive order
in the complex of information in the SN. The possibilities in this respect are
(alongside the procedures under (b) below) crucial in determining which sub-
chunks of information can be separated for inclusion in singular SSAs. The
decisions will result in a linear ordering of SSAs for consecutive verbalization
and expression.
The two central factors determining the ordering procedures are the con
tent of the information in the SN and the information-structural organization of
the SN. There is a large volume of literature which can be brought to bear on
this matter (especially the former component - the latter is much less thoroughly
investigated): for example the literature on Rhetorical Structure Theory (Mann
1984, Mann and Thompson 1986, Thompson and Mann 1987), and the literature
investigating the way in which S/Hs provide 'spatial descriptions' (e.g. of apart
ments, routes, etc.) (including Linde and Labov 1975, Klein 1979, 1980,
Ullmer-Ehrich 1979, 1982, Levelt 1981, 1982a, b). (Issues at this level are
actually closely related to the issue of selecting chunks of information from the
UI for inclusion in the SN. The difference is that at this level the lexical
possibilities start to play a role, while they do not in compiling the SN.)
One tendency is the deployment of information from thematic to rhematic
and from given to new (though strong saliency of information can lead the S/H
to deviate from this default procedure). Another tendency is to let the order cor
respond to the natural grouping and ordering of the information in the SN (i.e.
a principle of iconicity - cf. Haiman (ed.) 1985). For example, a conceptual
representation of a story in the UI is no doubt organized in terms of a sequence
of distinguishable and logically (causally, temporally, spatially) successive phases
in the event, and in the default situation S/H will respect this organization in
grouping and linearizing SSAs. In describing an apartment, one normally starts
from the door and 'walks' through the apartment in some logical way. Even on
a small scale, the unmarked expression of temporally ordered information
follows the time sequence, of causally related information the cause-effect chain,
etc. (See also Osgood 1980, 1984.)
Of course, both factors mentioned can interfere, and sometimes the infor
mation-structural organization can force the S/H to deviate from logical order.
In general, if the natural ordering of the information in UI and SN is not
sufficiently self-evident (i.e. easily inferable by the H/S), the S/H has to make it
clear by introducing explicit markings in the individual SSAs. More marked
orderings require higher degrees of explicitness. Thus, in case of completely
natural ordering, the S/H can often omit any kind of explicit marking of logical
relationship, while this is normally impossible in case of non-natural ordering.
Marking logical ordering entails making decisions concerning the con
nectives to be used, and hence, at this level, first decisions are made concerning
the expression of related chunks of information in main/subordinate or
coordinate clause combinations.
(b) Relational organization of the information in the SSA

The information in each SSA has to be autonomously organizable, and the
possibilities in this respect co-determine the way in which information in the SN
is subdivided. We can assume that the SSA is organized in terms of a relational
structure involving a number of distinct roles (which will be realized as argu
ments and satellites in the utterance) and one central relator or relator-
combination (which will lead to the choice of a predicate). (This is comparable
to what Chafe 1977a, b calls 'propositionalizing' a conceptualization.) The S/H's
choices in this respect depend on the inherent relation specifications in the SN

(i.e. in terms of D-, H- and B-relators holding between concepts), and to a cer
tain degree probably also on the pragmatic functions assigned to parts of the
information in the SN (cf. below).
For example, consider the conceptual network in (82), indicating the
transfer of a book (x3) from John (x2) to Mary (x1).
This relational network can be realized in quite different ways depending on how
it is embedded in some wider conceptual network. Let's suppose the S/H
chooses the combination <D -> H> as central. Unless there is an indication of
who owns the book (i.e. if it is left open how the H-relationship should be inter
preted beyond purely physical possession), John is characterized as agent, the
book as object, and Mary as recipient. This could lead to the utterances John
gives a book to Mary or Mary gets a book from John. If it were evident to the S/H
that John is and remains the owner of the book, however, he might rather be
characterized as the source, and Mary as the beneficiary. In addition, John could
also be characterized as possessor, and the book as possessed, but this marking
is only required if the possessive relationship has not yet been explicated in the
context (which indicates that role assignment at this level cannot be performed
for SSAs in isolation, but has to be interactive with the information in the SN
and the constitution of previous and following SSAs). If this relationship is made
explicit, this could lead to the utterance John lends a book of his to Mary. If it is
not, the result will probably be John lends a book to Mary. Still other situations
occur when John donates the book to Mary, when Mary was the owner of the
book and John only temporarily possessed it, when there is no physical
transmission of the book from John to Mary, etc. Conceptual variation can be
considerable and may involve rather subtle nuances.
Which roles ought to be distinguished and how they ought to be represent
ed is a question open to further analysis. One possibility is the introduction of
strict labels such as 'agent', 'object', etc. (i.e. case roles of the type proposed by
Fillmore 1968). For a theory in this respect, see De Schutter (1989). One could
also conceive of these roles as a selective set of specified features (control
ler/controlled, causer/caused, etc.; cf. Nilsen 1972) to be assigned to entities on
the basis of their status in a wider conceptual network. This matter cannot be
separated from the question how a detailed representation of conceptualizations
in SN, UI and encyclopedia will turn out, and which semantic functions have to
be assumed in the lexicon. Roles at this level can be considered 'canalizations'
of a multitude of very nuanced relational positions which entities can have in
some wider conceptual framework, in the direction of a restricted number of
more or less prototypical relational categories. In their turn, these categories can
be mapped onto a strict set of semantic functions which play a role in the verbal
patterns of natural languages.
It is likely that the set of semantic functions at the level of the predication
(the 'semantic function hierarchy') postulated in current FG can be considerably
simplified within the present framework (cf. 3.1.3.1). Consider for example the
different types of 'first arguments' which have to be postulated, such as 'agent',
'force', etc. (cf. Dik 1978a: 36ff), and which appear to behave similarly in the
verbal pattern. (Cf. 3.1.3.2.) A similar observation can be made for 'second
arguments' (Dik 1989a). Detailed relational characterizations are only required
at the level of conceptual representations in the SSAs, while at the predicational
level it suffices to postulate a few functions which 'canalize' these different roles.
Foley and Van Valin's (1984: ch.2) 'actor' and 'undergoer' are something in that
direction.
One further element which must be accounted for in the internal organiza
tion of the SSA is the coreferentiality of entities which have already been
introduced in earlier SSAs (anaphora), or will be introduced again in later SSAs
(cataphora). This element is important in determining the definiteness of terms
and for choosing full NPs or anaphoric or cataphoric pronouns (and the
communicative value of the information involved will play a role here, too - cf.
3.1.3.3). (Relevant literature includes Chafe 1976, Bock and Engelkamp 1978,
Hawkins 1978, Clark and Marshall 1981, Maes 1987.)
(c) Organizing the SSA information-structurally

As mentioned above, an important element in deciding on the organization of
SSAs is the communicative value of the information in the SN. Information is
normally deployed according to the principle of the gradual development of
theme/rheme structure, with a more or less clear point of anchorage in given
information and an extension in the direction of new information. This is true
for the entire turn as well as for each singular SSA. Once a chunk of information
has been introduced in some SSA, its status on the givenness/newness scale in
the SN changes, and with it the possibilities for further development of SSAs.
Information included in an SSA has to be marked for informational value
again, on the basis of its informational status in the SN and with respect to the
SSA-internal situation. It is here that the values of topicality and focality enter
the scene (to be seen as a scalar division, with the topic and the focus of the
utterance-in-spe located at its extremities). Normally, topical information is given

and/or thematic information in the SN, and focal information is new and/or
rhematic and/or salient information in the SN. The situation can be rather
complicated, however. Informational values at this level can be seen as a first
canalization of the communicative status of information in the SN, in view of
processes to be applied later in the grammar. (Further canalizations, which can
conform to the topic/focus distribution or can highlight different aspects of the
information structure in the SN, will occur during lexicalization and syntactic
function assignment.) Qualifications of information in the SN can also contribute
to the topic/focus marking. Contrastive focus, for instance, is no doubt deter
mined by characterization of information in the SN as being in conflict with
information in the UI, i.e. it will often involve negation and/or modality (this
information will also have strong rhematicity or newness in the SN itself).
Related to this domain is literature (in FG and elsewhere) concerning
types of topic and focus, the discourse binding of pragmatic functions at the sen
tence level, and problems of topic continuity, topic shift, etc. (cf. e.g. Givn (ed.)
1983).
(d) Determining the status of the information in the SSA

An SSA must sometimes be qualified in terms of its status, along the same
dimensions as those discussed above with respect to the SN. This can be a rather
complex matter. As mentioned above, qualifications in the SN can have different
scope, and therefore can be specified relative to qualifications with a wider
scope in the SN. Obviously, in conceiving individual SSAs, a qualification has to
be introduced which corresponds to the qualification of the information in the
SN but which is also relative to qualifications introduced earlier. This matter
interferes with issues of expression.
Though qualifications are most often expressed in an utterance together
with information about the SoA they concern, most of them may also be
expressed separately in the turn, in an utterance entirely devoted to them. This
strategy is likely to be chosen if the qualification covers a considerable
conceptual network, which has to be expressed in several SSAs. For example,
this happened time and again or what this had to lead to is the following express
aspect, such things are simply intolerable expresses deontic modality, imagine,
once upon a time in a country far away expresses irrealis, if you ask me, this is
what I believe happened expresses epistemic modality, etc. The information in
following or preceding SSAs covered by this kind of utterance, if in accordance
with the qualification expressed in it, can remain unqualified itself. However, if
there is a SSA deviating from the overall qualification, it has to be marked
explicitly (this seems only possible if the 'qualification utterance' preceded, and
no doubt this leads to informational prominence). Actually, the existence of

these qualification utterances demonstrates that the S/H has considerable
flexibility in representing the internal correlations of information in the SN at
this level. Apparently, qualifications can be represented independently of the
information in their scope, although the relationship does have to be made
explicit, of course, so the H/S can recover it. This is done by means of pronouns
(e.g. it or this) 'dummy nouns' (e.g. such things), etc. which refer to the SoA
covered by the qualification and to be expressed in successive SSAs.
Qualifications of information in the SN are most often introduced in one
or several single SSAs, however. This can be done even if a qualification has
been introduced in a separate utterance earlier on: after saying it was in 1898,
on a cold winter night, e.g., S/H can recall the temporal situation of the SoA by
using past tense forms in the expression of each following SSA (though this is
not obligatory: the present tense can be chosen as well). The range of possibili
ties and how they are realized for the different types of qualifications mentioned,
in themselves and in mutual interaction, is unexplored territory for research.
Actually, even if qualifications are expressed in one SSA together with
information on the SoA, this still does not mean that their scope may not extend
over information in following utterances, even if they are not repeated there. For
example, after the utterance it was summer 1968 when xyz happened an S/H can
utter several utterances further describing xyz, even in the present tense, but all
of them are nevertheless in the scope of the initial temporal qualification.
Possibilities for utterance-internal realization of meta-qualifications of SSAs
are varied. Obviously this is language-specific, but in languages such as Dutch,
German or English, at least the following possibilities (to be chosen at later
levels, in the predication and expression components) are available. Time and
space can be expressed in a satellite of the utterance. For space, this involves a
very complex system of spatial adverbs (e.g. Von Stechow 1982) or terms with
spatial prepositions (e.g. Cuyckens 1985, 1988). (See also Fillmore 1982, Klein
1982, Ullmer-Ehrich 1982.) For tense, expression involves temporal adverbs or
particles (e.g. Knig 1979), or temporal prepositions (e.g. Knig 1974). (There
also appears to be a great overlap in expression forms for time and space - see
Traugott 1978.) If the temporal qualification is not too specific, it can also be
expressed in the tense forms of the predicate, i.e. by a temporal operator (e.g.
Wunderlich 1970, Kamp 1979, De Schutter 1981, Vet 1986). Aspect can be
expressed in the aspect forms of the predicate (which are in complex interaction
with the tense forms) (e.g. Comrie 1976), in adverbial satellites {repeatedly,
finally, etc.), and occasionally in the choice of lexical alternatives (e.g. Dutch eten
and opeten, resp. eat and eat all). Realis/irrealis is expressed inter alia in tense
forms of the predicate (e.g. the use of the preterit for irrealis - Wunderlich 1970,
De Schutter 1981). Negation is typically expressed with particles such as not, no,
never etc., but can also be expressed through the use of negative predicates (re-
ject, be impossible, etc.). Modality has a variety of expressions (Palmer 1986),
which, however, appear to be very similar for the different types of modality to
be distinguished. They include the mood operator, modal auxiliaries, adverbs and
particles, but also complex sentences with a 'propositional attitude predicate' (in
the first person present tense or an impersonal form) in the main clause,
expressing modality over the SoA expressed in the embedded clause (I believe
that, it is possible that, etc.). Evidentiality is also often indicated by means of
complex constructions with a mental state or communication predicate in the
third person in the main clause (John believes that, John claims that, etc.), but
other possibilities include literal quotation, the use of moods of the predicate
(e.g. the subjunctive in German), etc. Finally, the S/H's personal attitude is again
frequently expressed by means of propositional attitude predicates (I hate it that,
etc.), or through adverbs {damned etc.), the choice of lexical entries with specific
connotative values, etc.
Actually, most of these qualifications can also be expressed term-internally,
which is the case if they concern a concept or a conceptual network which is not
the central one in the SSA but only a 'secondary' one to be expressed in some
term attached to the main predicate. For example, absolute space is involved in
the man from Jersey, and relative space in the farm next to the river or the biggest
chair but one. Aspect is evidenced in the repeatedly blowing horn or the objective
aimed at. Absolute time is expressed in the one-o-clock news, relative time in the
man who had arrived too late or a house as old as the street. The fictional house
expresses irreality. Term-negation occurs in a non-specialist, (epistemic) modality
in a possible failure. Indication of the source in a very brave dog, according to
John. Personal attitude is involved in the damned pain.
Finally, it is important to be aware of the complex relationship between
qualifications of information in SN and SSA and the communicative value of
information in each of these domains. As mentioned, modality and negation
certainly have influence on the informational value of aspects of the conceptual
SoA to be expressed. Hence, qualification in the SN no doubt influences
pragmatic function assignment in the SSA. But it is also likely that meta-
qualifications in the SSA are marked for pragmatic functions themselves (as was
the case in the SN). This seems required to account for the differences in
expression of the meta-qualifications in the SSA. The different expressions of
modality and negation, for example, appear to be the result of their prominence
in discourse (cf. 3.3).
(ii) Predication
With the sentencing procedures we have passed the gradation from the
utterance-independent to the utterance-specific representation of information,
and the next step to be taken is to decide how to represent it verbally, i.e. how
to lexicalize the conceptual information. We now enter the realm of FG, and in
the following I will mainly restrict my discussion to elements which are respon
sible for transposing SSAs into predications, and to elements in which FPG dif
fers from FG.
From this level on, processes basically concentrate on the further develop
ment of individual utterances. This does not mean that influence from the way
earlier SSAs have been formulated is totally impossible. It may be stylistically
important to limit the reoccurrence of lexical forms in successive utterances, and
this necessitates access to lexical choices made in earlier utterances. The
previous utterance of the interaction partner in short turn-taking sequences can
also play a role here: in question-answer sequences, for example, there is often
syntactic correspondence (cf. e.g. Collier 1983). Apart from such matters,
however, the utterance can be considered a separately processable entity from
here on: the unit of processing is the sentence. This may be a complex sentence,
though not all complex sentences require an integrated treatment. As mentioned
in the discussion of sentencing, the selection of a coordinate or subordinate
relationship between sentences can already be made at the level of the
linearization of 'raw' SSAs. (83) is certainly an example of the latter,
(83) Before going to bed, he always takes a walk in the garden
and the SSAs involved may be handled in a relatively independent way (though
the procedures must obviously take into account the status as main or
subordinate sentence). Complex utterances such as (81), however, or combina
tions of main clause and relative clause (certain types at least), do constitute a
more integrated complex, and must therefore be considered one unit at this level
(i.e. they should be considered as rooted in one single SSA). (Observations on
different degrees of cohesion between the components of complex sentences, e.g.
by Foley and Van Valin 1984, or within FG by Hannay and Vester 1987, fit
nicely within this discussion.)
At least the following sets of procedures, leading to the introduction of the
different elements in the FG-predication, have to be accepted at this level. (Note
that, in line with the argumentation earlier in this chapter, pragmatic function
assignment has disappeared from this level.)
(a) Choosing a predicate and terms

In lexicalizing an SSA, the S/H must first choose a predicate. This choice is
determined in the first place by the need to appropriately bind all the roles
occurring in the SSA, including the possibility of having certain roles within, and
other roles outside the central relationship (i.e. as arguments and satellites, re
spectively). Further determining factors are: the informational value of the roles
and the possibilities offered by certain predicates to translate this into structural
prominence (cf. e.g. the difference between give and get, the former giving more
prominence to the agent, the latter to the recipient - cf. 3.2.1.2); the meta-
qualifications of the SSA (e.g. the choice between alternatives with different
connotative values depending on the personal attitude of the S/H, or between
a positive or a negative predicate due to the polarity of the SSA); and elements
of the conception of the communicative situation in the UI (matters of register,
politeness, etc. as indicated above). Elements of the qualification of the SSA or
its illocutionary force can also, in function of their prominence, lead to the con
struction of complex sentence types with performatives or propositional attitude
predicates.
The choice of a predicate can be a direct mapping of some conceptual
relator, as in the utterance a chair has four legs. But in many cases it implies that
a complex of conceptual relators is 'clustered', as in the mapping from (82) to
the predicates give or get. It is also possible to combine conceptual nodes and
relators in the predicate, as in walking, which could also be expressed in more
direct correspondence to the conceptual representation as take a walk. These
choices often involve predicate formation procedures causing, for example,
argument reduction, if this is necessary from the perspective of the number of
roles to be bound by the predicate. (Dik 1980a: 39ff also proposes a predicate
formation rule for a number of cases which in FPG-terms involve incorporation
of an entity and a relation.)
As mentioned, FPG follows the FG-approach to predicate formation.
Within this wider framework, however, these processes can and have to be made
sensitive to characteristics of the input structure or to external information, while
in traditional FG they operate 'out of the blue'. It is possible that individual
cases of predicate formation proposed in FG will have to be handled differently
in this wider framework, but I will not be concerned with that here.
Interactively with the choice of a predicate, terms are formed to express
the various roles in the SSA as well as the secondary relations attached to them.
This involves assigning categorial status to entities. This cannot be done in
isolation for each entity: assignment of a category to one entity can influence the
possibilities of assigning a category to other entities. An important determining
factor is which entity has been assigned a central role in the SSA: normally, this
entity occurs as head of the term, and one can assume that this serves as the
starting point for categorizing the other entities in the SSA. The processes which
must be assumed here are open to further investigation. As to the structure of
terms, one can again follow the FG-treatment. Another decision to be made in
forming terms at this level is how to express coreferentiality. Should one use full
NPs or pronouns, anaphoric or cataphoric, etc.? Furthermore, it is also possible
to select modal adverbs, modal and other particles, etc., as a (partial) expression
of different kinds of qualifications of the SSA mentioned above.
Depending on the number of terms formed, and the predicate's ability to
bind these terms in argument positions, the predicate frame can be expanded
with a number of satellites, as is assumed in FG. Term-insertion in argument
positions implies, then, that the role of the entities involved is further 'canalized'
in (or 'forced into the shackles of) the semantic function of these positions,
while the role or status (in the case of qualifications) of elements introduced in
satellite positions is retained. It is important to keep in mind the difference
suggested above between roles and semantic functions. The latter only determine
the relationships between the elements occurring in the core predication in
correlation with the predicate. The former determine the position of an entity
within the entire SSA. The step from role to semantic function therefore implies
that part of the relational value of the role is lost (thus roles like force and agent
can be narrowed to a first argument having the semantic function of actor or
something like that). In satellites no narrowing of the role occurs.
It is imaginable that while inserting the terms into the argument positions
of the predicate, one must conclude that the role of some term does not fit the
semantic function of the argument positions. In that case one has to reconceive
the SSA, or change the choice of a predicate, or produce a weird utterance
(which under circumstances may be taken as metaphorical, although unwanted).
(b) Determining predicate and term operators

In the predication chosen, the S/H must also specify operators for further
expression of qualifications of the SSA and the illocution. In as far as the
qualifications did not find expression in the lexicalization itself, they require
expression through function words and grammatical forms in the expression
component. With respect to types and ways of representing these operators, it
seems superfluous to maintain a detailed system such as Hengeveld's at this level
(cf. 3.1.1), since most of these issues are taken care of at the conceptual levels.
It may be sufficient to distinguish a small set of categories of operators here
(tense, aspect, mood, negation, sentence type), most of which are 'used' by a
number of different conceptual qualifications according to principles which ought
to be further investigated. For example, tense as an operator at the predicational

level can be used to express temporal situation, but also irreality (cf. the use of
the preterit for that purpose - on the complex factors determining tense, see De
Schutter 1981, Vet 1986). Aspect as an operator not only seems to allow
expression of aspectual qualifications, but sometimes also participates in
expressing temporal qualifications. (On the difficulties in disentangling tense and
aspect systems, see e.g. Comrie 1976, Vet 1981.) Mood can be used to express
matters related to evidentiality (cf. the use of the subjunctive in certain types of
quotatives in some languages, including German) or to modality (cf. the use of
the indicative or subjunctive in Spanish, next to matters related to illocution
(Lyons 1977, Hengeveld 1988). Instead of the illocutionary frame in Hengeveld's
proposal, or the illocutionary operator in Dik's (1989a) proposal, we only need
a 'sentence-type operator' (having the possible values of declarative, imperative,
or interrogative - in fact, this is precisely what the operator stands for in Dik's
1989a analysis).
Maybe it is also not necessary to explicitly handle matters of scope (both
of operators and satellites) at the predicational level. Consider the difficulties
that would arise if one attempted to handle the different readings of the scope
relationships in (67c) in 3.2.1.3 in terms of layers in the FG-predications (cf. also
3.3.3). It is enough to treat qualifications in these terms at the conceptual level.
At the predicational level it is sufficient to specify an operator on the predicate
or to add a satellite to the predication, and to clearly mark the focus of the
utterance (very often a qualification only appears to have the focus in its scope -
cf. Bossuyt 1982, Dik et al. 1990), or just to leave the rest to the interpretation
of H/S if there are enough contextual clues to disambiguate possible alternative
scope assignments.
(c) Syntactic function assignment

Finally, syntactic function marking must be added to the predication (at least in
those languages in which they can be assumed to exist). It is intuitively plausible
to consider this a matter of the introduction of a perspective in the SSA, as is
claimed in FG. However, as argued in 3.1.2 and 3.1.3, 'perspective' should not
be taken as a primitive notion. It is a complex and derivative concept deter
mined by a number of factors, including the informational and the relational
value of terms. The latter factor is explicitly recognized in FG, since syntactic
function assignment is directly correlated with the semantic function hierarchy.
But it is also generally recognized (also in FG-publications - e.g. Bolkestein
1985) that subject and topic often coincide, and that object-assignment to the
second or third argument is determined by the degree of topicality of each (cf.
3.1.3.3). Probably factors of thematicity and saliency in the SN also exert an
influence. Hence, syntactic function assignment can also be considered a further

means of expressing the communicative value of information in the SN.
One can also ask why only one object should be distinguished (as in FG),
rather than two (as in much of the syntactic literature, including relational
grammar - cf. Perlmutter 1981: 343ff). From a theoretical perspective it seems
reasonable to claim that all real arguments of a predication must have their
share in the perspectivization of the utterance (a line is traced throughout the
entire core predication, as is the case with all other function markings), and
therefore must be assigned a syntactic status. The disadvantages of introducing
a second object function are not obvious.
(iii) Expression
Ultimately, we have reached what is traditionally called the syntactic component
of the grammar. We can largely accept the (albeit still very vague) overall design
of the systems involved here in FG, and I can refer to the presentation in 3.1.1,
and the developments and modifications suggested in 3.1.3, for the issues
involved here. Here are some minor additional remarks.
Firstly, in current FG the role of intonation has no doubt been neglected.
Intonation is at least as important an expressive device for the communicative
value of information in the SSA as word order, and there is a very subtle
interaction between word order and intonation to reach a nuanced expression
of the informative status of different parts of the utterance. It is highly unlikely
that one will ever be able to accurately handle the relationship between
informational status and word order of terms without considering the S/H's
options in the domain of intonation. (See Bolinger 1978, 1983; or, concerning
FG, Van Buuren 1985, Dik 1989a.) Suggestions for anchoring intonation in a
language production model are to be found in Van Wijk and Kempen 1985,
although the input of information-structural factors to the process is much
stronger than they concede.)
Secondly, either within the expression component, or in the area bordering
the expression component and a still-to-be-introduced 'articulation component'
(i.e. that which is traditionally referred to as a phonological/phonetic cpmpo-
nent), there has to be a location where the phonologization of the lexical items
chosen can take place (should one want to take into account the evidence for
such a process - see above). Introducing these procedures will depend on the
further development of a number of the sub-procedures of the expression
component, viz. the morphological processes which interfere with phonological
processes. In any case, phonologization is a matter to be located as late as pos
sible in the grammar. (An extensive survey of issues related to this in Levelt
1989. See also Nuyts 1982a, 1983c for some brief considerations.)
3.3. Negative raising: A test-case
3.3.1. 'Traditional' views of negative raising
To illustrate the relevance of some of the arguments given in 3.2 for an FPG,
in particular where it deviates from FG assumptions, let's finally go somewhat
more deeply into the phenomenon traditionally called 'negative raising' (NR),
involving the relationship between utterances such as (84a) and (84b).58
(84) a. I think he is not crazy

Ik denk dat hij niet gek is
b. I don't think he is crazy
Ik denk niet dat hij gek is
The choice for NR as a test-case is not random. NR involves some of the most
difficult and least understood notions in grammar (including negation, modality,
and information structure), and its lexical and syntactic characteristics are very
complex. Hence it seems no accident that thus far nobody has been able to
adduce decisive arguments for or against one or another account. No doubt,
investigations into NR are handicapped by the very subtle semantics of the
phenomenon, which is close to the borderline of what is still accessible by means
of linguistic intuition, and which is not of a nature to be easily amenable (if at
all) to corpus-analysis. Yet, the main reason why proposals are indecisive so
far seems to be the limited nature of most linguistic models from which NR has
been approached in the past. The complexity of NR requires a theoretical
framework of considerable sophistication which is sufficiently rich to be able to
subsume and integrate all dimensions involved. As I will try to show, this is not
only a matter of the model being encompassing, but also of it allowing a
considerable degree of abstractness in representation.
Let's start by summarizing some of the main characteristics of NR noted
in the literature and the two main types of accounts proposed so far.60 The
basic observation underlying the 'traditional' conception of NR involves the
relationship between utterances such as (84a) and (84b) above. (84b) has a
reading in which the negative in the main clause seems to belong to the
subordinate clause, thus making it more or less equivalent to (84a). As for
terminology, henceforth I will call this specific reading of an utterance like (84b)
its 'NR-reading'. Utterance types (84a) and (84b) will be called 'embedded-
negative' and 'main-negative' constructions respectively. The main-negative
variant with NR-reading will also be called the 'NR-construction', and a pair of
an embedded-negative and an NR-construction will be called an 'NR-pair'.61

The use of the term NR for denoting the relationship between the
utterances in a pair such as (84a-b) is due to a transformational approach, first
introduced by Fillmore (1963: 220ff). He suggested that the main-negative
variant could be derived from the embedded-negative variant by means of a rule
(which he called 'negative transportation') raising the negative from the
embedded clause into the main clause. Though his proposal was not based on
a thorough analysis of the construction's characteristics and did not provide real
arguments, it nevertheless aroused a vivid discussion wherein various semantic,
syntactic, and lexical peculiarities of the construction were noted.
On the semantic level, it has first of all been observed that constructions
like (84a) and (84b) are not completely equivalent. (84b) is weaker in expressing
the negativity of the embedded utterance than (84a), i.e. in (84b) the S/H seems
less certain about 'him not being crazy' than in (84a). Thus, 'raising' involves an
uncertainty principle, and Horn (1978a) suggests that this might be an instance
of a more general linguistic principle that the force of negation is negatively
correlated with the distance between the negative marker and the element it
negates. In any case, (84a) and (84b) appear not to be real synonyms, and in
semantically-based syntactic approaches this has been accounted for by making
the NR-rule sensitive to this uncertainty principle (e.g., Prince 1976 characterizes
NR as a hedging strategy).
Another observation on the semantic level is that the main-negative
construction is basically ambiguous. (84b) not only holds a relationship with
(84a), but apparently also has a reading similar to (84c).
(84) c. It is not the case that I think he is crazy

Het is niet zo dat ik denk dat hij gek is
(Henceforth, this reading of a main-negative utterance will be called its non-NR-

reading, and a construction like (84c) will be called a 'cleft-negation construc
tion'.) Hence, in the syntactic approach the main-negative utterance should get
two underlying representations, one in common with the embedded-negative
utterance (and thus involving the NR-derivation), and one in common with the
cleft-negation construction.
Yet, the observation about the relationship between utterances like (84b)
and (84c) lies at the core of a completely different explanation for the 'NR-
reading' of the main-negative utterance, called the pragmatic approach (first
introduced in Bartsch 1973; see also Halpern 1976 and Epstein 1977). Note that
in truth-conditional terms (84c) is the 'privative opposite' of (84a): (84c) is true
in any situation in which (84a) is true, but not vice versa. In the pragmatic
approach, then, (84b) is claimed to have only one underlying representation, viz.
the one paraphrased by (84c). While (84b) can be derived from (84a) by
semantic implication ((84b) is always true when (84a) is true), (84a) is said to
be derivable from (84b) by means of an inferential mechanism working under
certain 'pragmatic' conditions which limit the truth-possibilities for (84b) to those
for (84a). That is, (84b) can be used to express (84a) provided the appropriate
circumstances are present, and the logic behind this is represented in (85)
(schema based on Horn and Bayer 1984: 399).
(85) a. P(x, p) V P(x, ~p)

b. ~P(x, p) (= utterance (84b))
c. P(x, ~p) (= utterance (84a))
(P stands for predicate, x for the subject of P, p for the proposition which is the
object of P). (85) states that an utterance like (84a) (i.e. (85c)) can be inferred
from an utterance like (84b) (i.e. (85b)) under the circumstances (represented
in (85a)) that it is (contextually) clear that (in casu) either T think that he is
crazy' (P(x, p)), or T think that he is not crazy' (P(x, ~p)). Thus, according to
this view there is no syntactic rule of NR.
NR-constructions also appear to have a very intricate syntactic behavior.
The most famous observations concern the behavior of strict negative polarity
items and the polarity of tag questions in NR-constructions (Lakoff 1969), but
Horn (1978a) cites a variety of similar syntactic phenomena. These all seem to
indicate that in NR-constructions the embedded clause is conditioned by the
negative in the main clause. Lakoff (1969), Prince (1976), and others have
argued that this syntactic behavior demonstrates that at some point in the
derivation of NR-constructions the negative must be in the embedded clause,
triggering the phenomena involved, and that raising must occur late in the
process. This would be a strong indication for the appropriateness of the
syntactic approach. The argumentation is faced with serious problems, however.
As is shown at length in Horn (1978a), it appears that nearly every syntactic
phenomenon involved (i) is elicited by various non-NR-constructions as well,
and/or (ii) is not elicited by some NR-constructions, and/or (iii) is troubled by
intervening factors. Moreover, as Horn and Bayer (1984: 407ff) show, the
pragmatic approach can possibly also account for the syntactic phenomena. All
in all, the exact status of these 'syntactic' peculiarities of NR-constructions
remains uncertain, and there is no clarity about how they can be handled in
either approach.62
A core problem for any account of the NR-constructions concerns the
lexical level. As Horn (1978a: 187) notes, NR-readings seem to occur with a
rather wide variety of predicates, but they all belong to categories expressing the
S/H's opinion "of the mind or of the heart" (Horn 1978a: 208), i.e. expressing
modality.63 This includes mental state predicates (think/denken, sup-
pose/veronderstellen, believe/geloven, etc.), perception predicates (seem/lijken, look
like/erop lijken, etc.), communication predicates (e.g. say/zeggen), but also
predicatively used adjectives expressing degrees of probability, as in (86).
(86) a. It is likely that he is not crazy

Het is waarschijnlijk dat hij niet gek is
b. It is not likely that he is crazy
Het is niet waarschijnlijk dat hij gek is
c. It is unlikely that he is crazy
Het is onwaarschijnlijk dat hij gek is
The relationship between (86a) and (86b/c) is very similar to the one existing
between (84a) and (84b).
Yet, not every predicate in these categories behaves as an NR-predicate.
For example, (87a-b) and (88a-b) certainly do not hold the meaning relationship
which exists in (84a-b).
(87) a. I know he is not crazy

Ik weet dat hij niet gek is
b. I don't know he is crazy
Ik weet niet dat/of hij gek is
(88) a. It is certain that he is not crazy

Het is zeker dat hij niet gek is
b. It isn't certain that he is crazy
Het is niet zeker dat hij gek is
In both the syntactic and the pragmatic approach, then, such predicates have
been excluded from the main account. Some have attempted to explain their
exceptional status in terms of their semantic nature. One consistent group of
exceptions appears to be factive and counterfactive epistemic predicates
(know/weten, be certain /zeker zijn, etc. - cf. Kiparsky and Kiparsky 1970). This has
been related to the uncertainty principle apparently involved in raising: factive
predicates express certainty about their complement and would therefore never
provide a reason for raising the negative. (How this could be accounted for in
the pragmatic approach is, of course, less obvious.)
However, there are other exceptions which do not allow this explanation,
at least at first sight: utterances with non-factive or non-counterfactive predicates

like be possible/mogelijk zijn, e.g., do not allow NR-readings either. Horn (1978a)
has tried to provide an integral explanation for all these exceptions to NR in
terms of a 'scalar model'. If the epistemic opinion predicates are plotted on a
strength scale, as in (89) (based on Horn 1978a: 194),
(89) weaker stronger
be possible believe, suppose know, realize

etc. be probable be certain
etc. etc.
it appears that only the mid-scalar category can cause NR-readings. Horn
explains this by noting that the negative counterparts of predicates on the
extreme poles of this scale take the opposite extreme position on the negative
counterpart of (89), whereas negated mid-scalar predicates still take a mid-scalar
position on the negative scale, as can be seen in (90).
(90) (weak) (strong)

4-
possible likely certain
not certain unlikely impossible
Thus, with the extreme-scalar predicates it is always the case that either the
positive or the negative variant expresses certainty or necessity (viz. the strong
one), and in Horn's view that would explain why all of them are excluded from
NR according to the uncertainty principle stated above. Negatives of mid-scalars
remain mid-scalars, however, and thus they are potential NR-predicates.
This explanation does not bring an end to our problems, for not all mid-
scalars allow NR-readings either. In fact, there is some cross-linguistic difference
in precisely which predicates do constitute exceptions in this respect. For
example, English hope seems not to be a NR-predicate, whereas Dutch hopen
does seem to be one, as can be seen in (91) and (92).
(91) a. I hope he hasn't done it

b. I don't hope he has done it
(92) a. Ik hoop dat hij het niet gedaan heeft

b. Ik hoop niet dat hij het gedaan heeft
Horn and Bayer (1984) attempt an account of these idiosyncratic cases in terms
of the concept of 'short-circuited implicatures' (SCI) (Morgan 1978). This
concept has originally been introduced to help account for indirect speech acts.
It was intended to explain why (93a), but not (93b) is standardly used to convey
a request.
(93) a. Can you close the door?

b. Are you able to close the door?
The use of (93a) for requesting is claimed to be based on a usage convention

which eliminates the need to calculate the implicature linking (93a) with the
actual request, and which makes it possible for (93a) to be used for requesting
in any context. The same implicature can exist between (93b) and the request,
yet it always has to be calculated, and an appropriate context is required. Horn
and Bayer also apply this principle to the mid-scalar NR-predicates, arguing that
those allowing a NR-reading do have an SCI-usage convention, whereas those
not allowing it do not. Since such usage conventions can vary across communi
ties, this might also explain why there is some variability across S/Hs, dialects,
and languages, as to which mid-scalars are and which are not 'neg-raisers'.
Note that, whereas Horn (1978a) still accepts a functional-syntactic
approach, albeit somewhat reluctantly, the Horn and Bayer (1984) proposal is
a clearly pragmatic one, in the line of the 'traditional' pragmatic approach la
Bartsch. The reasons for Horn's switch are not clear, however. He does not
provide decisive arguments in favor of the pragmatic approach and against the
syntactic approach (in fact, his proposals concerning the lexical aspects of NR
are compatible with both approaches). The syntactic approach has not been
abandoned in the literature either. In fact, in one of the most recent statements
on NR that I am aware of, Seuren (1985: 166ff) again advances the transforma
tional rule account. But Seuren's remarks basically reflect his earlier view
(Seuren 1974), and they do not add to the discussion either. The analysis of NR
can hardly be said to be settled, then. We are left with an unsatisfactory picture
of a construction which is volatile and strange, and the general observation that
intuitions about it are vague and uncertain only adds to this picture. Such facts
should make one suspicious, not only about the proposals made to handle this
construction, but probably also about the way the problem to be accounted for
has been conceived.
Before going into this, however, it is important to point out that the fact
that one starts out with a theoretical framework like FG or FPG does not a
priori preclude any of the traditional accounts of the phenomenon. This is not
a trivial matter, especially as far as the syntactic rule account is concerned. Until
now the NR-rule has always been formulated in transformational terms, and it
is precisely one of the main principles of the FG/FPG concept of syntax to avoid
replacement operations in the grammar (cf. 3.1). At first sight this might seem
a sufficient reason to discard the syntactic rule account immediately. Yet, in
terms of the 'intelligent' approach to the syntactic placement and expression
rules already suggested in 3.1.3.3, it is, in principle, perfectly possible to
formulate a non-transformational syntactic mechanism for NR which is
completely in accord with the main options taken in FG/FPG. In fact, the type
of mechanism suggested there for subject raising could take care of NR, too,
since the formal characteristics of NR and subject raising are very much alike.
One difference would be that the lexical element 'raised' would not be an
argument but a predicate operator; but that is not too different from what
happens in, for example, quantifier floating, where a term operator is 'raised'.
Another difference would be that the factor triggering the mechanism would not
be perspectivization but rather (according to the traditional conception of NR)
the uncertainty factor; but that would not be the first time either, since
Bolkestein (1981b) already proposed factivity as a trigger for a raising rule.
The pragmatic approach would, of course, not cause any formal problems
in FG or FPG either. The variants in the NR-pair would have to get a different
representation at the level of the underlying predication64 in FG, hence the
treatment would not be a matter of FG proper, but rather of the Functional
Logic (cf. 3.2.1.2). In FPG, this pragmatic relationship would be treated in terms
of S/H's selection of information from the UI for inclusion in the SN underlying
the entire discourse, and the SSA underlying the individual utterances in it, in
particular on the basis of knowledge of the context of conversation and the
partner model. One would have to assume that S/H's partner model involves the
supposition that H/S thinks (justly or not) that S/H fulfills condition (a) of
inference schema (85) above, and furthermore that S/H's judgment of the
communicative situation is such that it urges S/H to use a hedging strategy by
only expressing the SoA involved in (84b) (i.e. by only including this information
in SN and SSA), knowing that H/S can infer that the SoA in (84a) must hold,
too. If condition (a) did hold, and if H/S knew it to hold, H/S has discovered the
real state of the mind of S/H. If the condition did not hold, although S/H was
aware that H/S thought it would hold, S/H has misled H/S in a rather subtle way.
Of course, (85) can also cause communication failures. If the condition did hold,
but S/H was mistaken in assuming that H/S knew this, S/H has failed to
communicate what (s)he wanted to. If the condition did not hold, but S/H was
unaware of the fact that H/S did believe it to hold, H/S will understand things
S/H did not mean to say.
3.3.2. Reconsidering the traditional conception and accounts of NR
Formally, it is not a problem to state either of the traditional approaches in FG

or FPG terms, but as mentioned, the nature of the questions these traditional
accounts leave is such that one can wonder whether there is not something
structurally wrong with the way the phenomenon of NR as such has been
conceived thus far. Indeed, a major problem with the traditional conception
appears to be that it involves only part of the potentially relevant set of pairs of
alternative utterances existing within the global construction type under
consideration.
Functionally, constructions such as (84a) and (84b) (the latter in the NR-
reading - it is important to keep this reading distinguished from the one related
to (84c), cf. below) express S/H's opinion concerning a particular SoA (in FPG
terms a qualification of an SSA - cf. 3.2.3) expressed in the embedded utterance,
i.e. what is traditionally called a propositional attitude (PA - henceforth, I will
call this construction type the 'PA-construction'). More specifically, the
NR-phenomenon is restricted to the type of PA-construction in which the PA
combines two aspects, viz. polarity (positiveness/negativity) and modality. Thus,
in general, the PA-constructions considered here express the S/H's judgment
about the degree of likelihood of the occurrence of the SoA in some possible
world. This ranges from an absolutely positive qualification, over various degrees
of non-absolute qualification in a gradual order from rather positive via totally
neutral to rather negative, ending in an absolutely negative qualification of the
SoA. The variants in the PA-construction pairs apparently provide alternative
possibilities for expressing (approximately) the same PA over an SoA.
From the examples presented in 3.3.1 then, it will be obvious that the
traditional approaches to NR have only considered construction pairs in which
exactly one negative marker is present in the surface structure, either in the
main clause or in the embedded clause, and in which the main clause predicate
(let us call it the modality (M-) predicate, since it always expresses the modal
component of the PA) in the main-negative and the embedded-negative variants
is identical. This restriction is no doubt due to a tendency to use superficial
structural similarities as the main criterion for defining linguistic relationships.
This is precisely what one would expect from a purely formally-oriented
transformational approach. Yet, the pragmatic approach, which one would have
expected to have an eye for the functional dimension of the phenomenon, has
taken over this view without further thought. In any case, from a functional
perspective this restriction appears unmotivated, for these criteria give us only
a fraction of the entire range of pairs of alternatives (henceforth 'PA-pairs')

showing the same basic meaning relationship as in (84a-b).
Firstly, the same type of relationship exists in utterance pairs involving
more than one negative marker. Cf. (94a-d) and (95a-d).
(94) a. It is probable that he is not crazy

Het is waarschijnlijk dat hij niet gek is
b. It isn't probable that he is crazy
c. It is improbable that he is crazy
Het is onwaarschijnlijk dat hij gek is
d. It isn't improbable that he is not crazy
Het is niet onwaarschijnlijk dat hij niet gek is
(95) a. It is likely that he is crazy

Het is waarschijnlijk dat hij gek is
b. It isn't likely that he is not crazy
Het is niet waarschijnlijk dat hij niet gek is
c. It is unlikely that he is not crazy
Het is onwaarschijnlijk dat hij niet gek is
d. It isn't unlikely that he is crazy
Het is niet onwaarschijnlijk dat hij gek is
The nature of the meaning relationship between the variants in utterance pairs
such as (94b/c) and (94d), (95b/c) and (95d), or (95a) and (95b/c) is quite
comparable to that in a 'traditional' NR-pair like (94a) and (94b/c). Even the
difference in terms of weakness or degree of uncertainty is always present, even
though it is not always the same (see below).66
Secondly, the same type of relationship also exists in utterance pairs in
which there is no identity in the M-predicates. Consider once again the
utterances in a pair such as (88a-b), which has been excluded from the
traditional accounts precisely because of the clear difference in meaning between
the variants.

In utterance (88b) as well as in utterances (84b) or (94b), the negative can be

said to contribute to the value of the opinion (the PA) expressed about the SoA
in the embedded clause. It does indicate the possibility that according to the S/H
'he is not crazy', and in certain contexts even seriously suggests this to be the
case. Hence, (88b) does resemble (96) in meaning in the same way as, e.g., (84b)
equals (84a), including the difference in weakness of the expression.
(96) It is (very well) possible that he is not crazy

Het is (best) mogelijk dat hij niet gek is
Similarly, (88a) and (97) also communicate more or less the same message.
(97) It is impossible that he is crazy

Het is onmogelijk dat hij gek is
The main clause in (97) unmistakably indicates that in the S/H's opinion 'he is
not crazy', as in (88a), and again the negativity in (97) is somewhat weaker than
in (88a).
Thus, if one takes the functionality (and thus the meaning) to be the
primary criterion for deciding which linguistic constructions must be related in
a grammar and which not, any account put forward to handle pairs such as (84a-
b) or (94a-b) should equally well be applicable to pairs such as (94b/c-d), (95a-
b/c), (95b/c-d), (88b-96) or (88a-97).
If one accepts this encompassing picture of the phenomenon, however, this
immediately leads to a revision of some of the suppositions present in the
traditional approaches with respect to the nature of the relationship between the
utterance pairs. More specifically, revisions are required with respect to the
'uncertainty principle', with respect to the behavior of Horn's extreme scalar
predicates, and with respect to the structure of the main-negative construction.
First, utterances (94a-d) and (95a-d) indicate that the traditional picture
of the weakness factor in the relationship between the alternatives at least has
been too simplistic. Obviously, the uncertainty factor is not tied up with the
negativity of the PA-qualification of the SoA in itself: the utterances in (95)
clearly indicate that the same factor is involved in case of an overall positive
qualification of the SoA. Hence, this factor is rather a matter of the degree or
the strength of the modality component of the PA-qualification of the SoA,
irrespective of whether the PA is positive or negative. The explanation that the
weakness factor is related to the distance between the negative marker and the
thing negated is clearly void, as is particularly clear in the utterances in (95).
In fact, the hypothesis that there is a correlation between the distance of
the negative element and the strength of the negation is suspect in general.
Consider the Dutch utterances in (98).67
(98) a. Niet in deze test opgenomen zijn volgende

not in this test included are following
vragen
questions
"Not included in this test are the following questions"
b. In deze test zijn volgende vragen niet
in this test are following questions not
opgenomen
included
"This test does not include the following questions"
(98a) is more strongly negative than (98b) (due to the fact that the negative
marker is in focus in the former, but not in the latter), although the distance
between negative marker and negated element is greater in the former than in
the latter. The principle even fails in some of our cases involving a negative PA,
such as (94d) and (94b/c). The former is clearly weaker (more uncertain) than
the latter, although in the former there is a negative marker much closer to the
negated element than in the latter.
Yet, it remains a fact that the only consistent structural difference between
the utterances in the pairs involves the occurrence and placement of (a) negative
marker(s), even in a case like (95) where the overall PA expressed is positive.
Hence the differences in the degree of uncertainty, which are as consistently
present, somehow do have to be correlated with this variable. In fact, it appears
that the differences in weakness or uncertainty can be correlated with variation
in the exphcitness of a negative marker in the main clause. In both (94) and
(95), instance (d), which contains a double negative marker in the main clause,
is weakest, and (a), which contains no negative marker in the main clause at all,
is strongest. There is even a difference in strength between the incorporated and
non-incorporated variants (b) and (c) in (94) and (95): the non-incorporated one
is weaker than the incorporated one. In addition, compare (99a-c).
(99) a. I believe he is not crazy

Ik geloof dat hij niet gek is
b. I don't believe he is crazy
Ik geloof niet dat hij gek is
c. I doubt he is crazy
Ik betwijfel dat hij gek is
(99c), which lacks an overt negative marker, seems stronger than the main-
negative equivalent (99b) but is hardly weaker than the embedded-negative
variant (99a).
These observations seem to allow the generalization that there is a direct
correlation between the degree of explicitness of a negative element in the main
clause and the degree of uncertainty over the occurrence or non-occurrence of
the SoA expressed by the utterance: the more explicit the negative marker, the
greater the uncertainty involved in the qualification of the SoA. The explicitness
hierarchy of negative markers is given in (100).
(100) no negative > implicit negative > single incorporated negative >
single explicit negative > double negative
(See also 3.3.3 on this factor). Apparently, the main clause negative not only
influences the polarity component of the PA expressed, but also the modality
component involved in it.
Secondly, since utterance pairs such as (88b-96) and (88a-97) ought to be
included in an integral account of NR, we will have to find an explanation for
the meaning difference between the utterances in pairs such as (88a-b).

(96) It is (very well) possible that he is not crazy

Het is (best) mogelijk dat hij niet gek is
(97) It is impossible that he is crazy

Het is onmogelijk dat hij gek is
Horn is undoubtedly right in claiming that the apparently different behavior of

such pairs must be due to the nature of the main clause predicates (i.e. the
extreme-scalars in his scalar model). However, rather than trying to find an
explanation for why they are excluded from the mechanism under consideration,
it will be necessary to account for them in a system-internal way.
In fact, as noted above, these extreme-scalars take part in weakness
relationships as well as the mid-scalars, which shows that the factor to which
Horn (and others before him) have ascribed their seemingly exceptional status
does not hold water. A property such as factivity or counterfactivity of the M-
predicate apparently does not exclude the expression of uncertainty. On the
contrary, in many circumstances utterances such as (88a) or (97) precisely signal
a certain degree of uncertainty on the part of the S/H with respect to the status
of the SoA expressed, for otherwise S/H could as well have used (101).
(101) He is not crazy

Hij is niet gek
Nevertheless, Horn's scalar model, relating degrees of modality expressed

by the predicates and their behavior under negation, does contain the key to an
explanation. In order to see it, however, one should take the perspective, not of
the predicates involved (i.e. of superficial lexical similarity between utterances,
as Horn did), but of the complex PAs (combining modality and polarity)
expressed in the main clause of the constructions involved (i.e. of the meaning).
That is what is done in schema (102), which roughly shows some of the possible
mappings between various degrees of PA and main clause expressions.68
(102) - N +
impossible doubt uncertain possible believe certain

not-believe not-know not-impossible probable know
improbable not-improbable
Utterances such as (88a-b), (94a-b), and (96-97) express a certain degree

x of negativity about the SoA under consideration.69 This degree of negativity
can either be expressed by means of a negative marker in the embedded clause
and a main clause expression situated on the positive side of scale (102), or by
means of a positively formulated embedded clause and a main clause expression
located on the negative side of the scale. Of course, in order to have the same
value x overall, the selected main clause expression in the latter case must be the
exact opposite of that in the former case, point N on our scale being the axis.
If this condition is fulfilled, we have two semantically equivalent utterances (that
is, they express an identical conceptual structure; they are not communicatively
equivalent, though: cf. 3.3.3). Obviously, (94a-b), (88a-97), and (88b-96) more or
less fulfill this condition, and thus are cases of such equivalent variants (at least
roughly, since there does remain the small difference noted above in the strength
of the modality, which results from the fact that expressions at opposing
positions on scale (102) are not exact opposites due to the influence of the
explicitness of (a) negative marker(s) present). This is not the case at all in
(88a-b), however.
In fact, if we take a look at the position of M-predicates and their
negatives, and PAs and their opposites on scale (102), it appears that there is
only a correspondence between them in the case of the rather positive and the
rather negative attitudes. In all other cases, this relation between attitudes and
lexical items is non-existent. This is due, once again, to the fact that the main
clause negative not only determines the expression of polarity about the SoA,
but also influences the modal value expressed by the M-predicate. It turns this
modal value into its opposite, along the lines indicated by Horn in schema (90).
In this way, the negation of the M-predicate used to express the modality in
(88a) changes the modality expressed by that predicate to its opposite (i.e. from
Very strong' to Very weak'). Hence the use of that negated lexical item in (88b)
does more than only compensate for the change from the negative formulation
of the SoA in the embedded clause in (88a), to its positive formulation in (88b).
Thirdly, our reconsideration of the NR-phenomenon has implications for
the debate about the underlying structure of the main-negative construction. I
have just argued that the alternative ways of formulating the PA in the PA-
constructions are semantically equivalent (apart from the small strength
difference). Of course, this view is in conflict with the way the pragmatic
approach has determined the nature of the meaning relationship between these
alternatives, which is based on the assumption that their truth conditions are not
identical. In fact, this assumption is clearly false: as is particularly clear from
utterances such as (99a-c), in any situation in which one variant in our utterance
pairs is true, the other one is equally true.
The mistake of the pragmatic approach lies in basing its analysis, not on
the PA-construction reading of the main clause negative variants, but on an
alternative reading of the same structure. In fact, provided the main clause
negative marker is an explicit, non-incorporated one, it is possible to interpret
the role of this negative marker in at least two ways. Indicative of the distinction
are the different intonation contours accompanying the alternative interpreta
tions of the structure. An utterance such as (94b)
(94) b. It isn't probable that he is crazy

can be pronounced either with a neutral (mostly rising) intonation on isn't

probable or is niet waarschijnlijk, or with a clear intonational break between
negative and predicate. In the latter case, there still are two possible variants,
viz. with main stress on isn't (probably better expressed as is not) or niet, or on
probable/waarschijnlijk.
The PA- (or NR-)construction reading is obviously the one with the neutral
intonation. In this reading the negative does not play a distinct role relative to
the M-predicate, but rather forms a functional unit together with it (as is
reflected in the fact that it semantically interferes with it), which in its entirety
expresses the PA over the SoA.
In the other readings, however, there is a functional opposition between

negative and M-predicate, as is reflected in the intonational break between
them, and in the impossibility to incorporate the former into the latter. Here the
negative clearly does not take part in expressing the PA. Rather, it negates the
PA expressed by the M-predicate in itself, which is in fact what is done in the
cleft negation constructions of the type it is not the case that xyz, too. The
difference between the two intonational alternatives in this respect seems to be
a matter of topicality or focality of either the negation or the M-predicate. The
reading with stress on the predicate can be used if one wants to correct a
suggestion made in the conversation as to which opinion one holds with respect
to the embedded clause (normally, this utterance is followed by a statement of
which opinion can appropriately be ascribed to the S/H). In that case the PA
expressed in the main clause has strong contrastive focus. The alternative with
the main accent on the negative can be used if one wants to forcefully deny the
suggestion that one holds the opinion about the SoA expressed in the embedded
utterance (this reading is most like the cleft negation construction). Anyway, in
either case the negative in the main clause has the predicate within its scope.
The foregoing observations cast new light on the traditional accounts of the
phenomenon. First of all, particularly the observation concerning the structural
ambiguity of the main-negative variants shows that the pragmatic account of the
relationship between the alternatives in the PA-pairs in terms of a pragmatic
inference schema such as (85) is inaccurate. If the alternatives basically express
the same conceptual structure (and thus have the same truth conditions), there
is no need for an inferential procedure to relate them under specific conditions.
On the contrary, an adequate grammar has to specify the relationship between
them directly, in terms of the process of utterance production or interpretation
itself, starting from or leading to one and the same underlying representation in
any condition. The difference between them then has to be accounted for in
terms of communicative strategies of the S/H (cf. 3.3.3).
Of course, the analysis in terms of inference schema (85) does remain valid
for relating the non-PA-construction reading of (94b) with the accent on the
negative (the reading with the accent on the predicate would be inappropriate
in the context of precondition (85a)) and its semantic equivalent, the cleft
negation construction, to (94a) and the PA-construction reading of (94b). In fact,
contrary to what seems to have been assumed (Horn 1978a: 179), inference
schema (85) does apply in relating all such alternatives, including those with
Horn's extreme scalar predicates and even those with the idiosyncratic lexical
exceptions. For example, for certain, if H/S knows that either you are certain
that x is crazy or you are certain that x is not crazy, and you tell H/S that you
are 'not <certain that he is crazy>', H/S can conclude that you are certain that
x is not crazy. The same reasoning can be followed with hope. The problem is
that in normal communicative circumstances the precondition will hardly ever
be satisfied - hence this analysis seems purely academic.
Evidently, the conclusion that schema (85) does not apply to the variants
in the PA-constructions automatically implies that the entire account of the
idiosyncratic exceptions to the NR-predicates on the basis of the notion of SCI,
as suggested in Horn and Bayer (1984), is vacuous. Apart from the fact that
accounting for something in terms of accidental habit formation (after all, that
is what the SCI approach amounts to) is rather cheap, there is nothing to be
short-circuited in the relationship between the variants anyway.
The fact that the pragmatic approach is mistaken does not yet mean that
the syntactic account is accurate, however. The foregoing observations provide
serious arguments against such an approach, too. This is already apparent if one
compares the various pairs in (94) and (95). It does not require much discussion
to see that any attempt to account for these utterances in terms of a syntactic
rule would require mechanisms which are, at least from the functional-syntactic
perspective of FG or FPG, completely unacceptable. A syntactic mechanism
would imply the acceptance of a lexical correction session applying after the
operation of the word ordering rules as part of the expression component. This
session should involve not only relatively small lexical adaptations such as
negative incorporation (needed to account for cases such as (94c/d) or (95c/d))
(see Horn 1978a: 170ff for other 'small' adaptations), but also complete lexical
exchanges (needed to account for pairs such as (88a-97), (88b-96), or (99a-99c)).
This is against the FG/FPG principle that no changes in the lexical setup of an
utterance can be made after the level of the predication (cf. 3.2.1.2). In addition,
if one considers utterances like (94d), (95b) or (95d), it is clear that to get the
negative marker(s) in the appropriate position one will have to develop a series
of different scenarios for the application of the mechanism, in which double
negatives sometimes must be allowed and sometimes must be deleted. This is
likely to result in a completely ad hoc mechanism. Moreover, according to the
FG/FPG principles of syntax a deletion mechanism is out of the question
anyway.
With respect to the special non-transformational mechanism for raising
sketched in 3.1.3.3 in particular, the main clause negative in the PA-constructions
clearly does not have the characteristics that would be expected if it could be
subject to this mechanism. It would only be appropriate to formulate the
phenomenon in these terms if we could assume that the negative marker
semantically belongs to the embedded predication, but that it has some strong
informational value which can motivate the raising process, and which, thus, is
distinctive for the raised and non-raised alternatives. But none of these
conditions is fulfilled.
Of course, the main clause negative in the PA-constructions has to do with
the meaning of the embedded clause, yet, as shown above, it interferes with the
meaning of the M-predicate as well. Thus, the negative semantically belongs to
the embedded clause no more than to the main clause. It is even questionable
whether one can say that an operator such as a negative 'semantically belongs'
to an SoA. If one can, then this is at least quite different from the way the main
clause object in the subject raising constructions can be said to semantically
belong to the embedded SoA. In fact, the negation particle belongs to it in
exactly the same way as the M-predicate does. The very specific relationship
between the negative element and the M-predicate is totally unaccounted for in
the syntactic approach.
Hence, in terms of the information structure of the utterances, the negative
does not play a distinctive role either, as is apparent from the fact that it has no
functional distinctiveness in the main clause, but rather constitutes a functional
unit together with the M-predicate. One might still argue that the uncertainty
factor can be invoked as the cause of our syntactic mechanism. Yet, it is not only
suspect to use this kind of very subtle semantic factor as a substitute for the
traditional type of information-structural factors that are taken to influence
special placement of lexical elements in the functional pattern of an utterance,
such as perspectivization or topicality/focality. It is even impossible to correlate
this factor with the placement of the negative in the embedded clause or the
main clause. As mentioned above, the weakness relationships are perfectly
parallel between utterance series (94a-d) and (95a-d), but the occurrence of the
negative in the embedded clause is exactly opposed in the corresponding
utterances in (94) and (95): those utterances in (94) in which there is a negative
marker in the embedded clause do not have one in (95), and vice versa. This
means that a raising rule cannot apply in the same way in both cases, even
though the 'causing factor' would be equally distributed in (94) and (95).
3.3.3. Toward a cognitive-pragmatic approach
3.3.3.1. The basic level of representation

Now we are obviously facing the question how a more adequate account for our
phenomenon should proceed. If a syntactic rule is inappropriate as a means of
relating the alternatives in the grammar, we have to accept that the variants in
the different pairs do have a different structure at the level of the predications
in FG and FPG, viz. with the negative(s) present as (an) operator(s) located in
the structural unit in which it (they) occur(s) in the surface structure. For the
traditional NR-pairs, this means that in the predication of the embedded-
negative variant there is a negative operator attached to the predicate in the
embedded clause, whereas in the predication of the main-negative variant the
negative operator is attached to the M-predicate.
Now, since the alternatives in the PA-pairs do have the same basic
meaning (apart from the small difference in the (un)certainty of the qualifica
tion, which of course does require attention in our approach - see below), and
given the principle (argued for in 3.2.1.2) that same meaning implies same basic
conceptual representation, it would be unnatural to leave the difference between
the variants to some logical reasoning mechanism as part of a Functional Logic,
as would be proposed by Dik. The reasons are basically the same as those given
in 3.2.1.2. It seems necessary to turn to some 'deeper' level of representation in
the grammar as a common basis for our variants, a level not available in current
FG. Moreover, as the reconception of the problem in 3.3.2 has shown, in
conceiving this level we have to take distance from the lexical realization of the
utterances, at least as far as the PA itself is concerned. This is particularly so if
one wants to grasp the intuition that both the modality and the negation in our
constructions are aspects of the S/H's qualification of an SoA, and thus are
conceptually very similar phenomena, even though at the level of the predication
the former gets a lexical form (the M-predicate) and the latter is realized as a
predicate operator (which will be replaced by the negation particle in the
expression component). All of this again stresses the need for a non-verbal
conceptual representation.
In FPG, then, we can turn to the conceptual level of the SSAs as a
common basis to start from for deriving our variants. As argued in 3.2.3, this is
the most basic level of representation for individual utterances. For the examples
given so far, the representation at this level would be like (103) (where x1 stands
for a person in the examples referred to as he, arid x2 for the property of
craziness).
(103)
The SSA involves the matter of 'x1 being crazy'. In this SSA the 'having'
relationship gets a meta-qualification ( Q > H > ) for being untrue or non-existent

(N, i.e. negation) with a certain degree of certainty x (Mx, i.e. modality x). Note
that modality and negation are considered to be hierarchically equal in the cases
under consideration: even though there must be complex interactions between
them (as noted above), neither of them has the other one within its scope. Since
this level of representation is linked to the level of the predications by means of
procedures responsible for lexical choice and the marking of operators (cf. the
predication component in 3.2.3.3), FPG can account for the fact that part of the
qualification can get expression by means of lexical choice of certain predicates,
and part of it by means of a predicate operator.
Representation (103) seems required not only to account for the PA-
constructions, but also for a wider set of construction types expressing a PA
containing modality and negation. As mentioned in 3.2.3.3, there are several
ways of expressing modality in an utterance, including the insertion of an
adverbial expressing modality (probably/waarschijnlijk, possibly/mogelijkerwijze,
certainly/zeker, etc.) into the utterance expressing a positive or a negative SoA,
as in (104) (or variants of it with a different word order), or the use of modal
auxiliaries in the utterance, as in (105).70
(104) Probably he is not crazy

Misschien is hij wel niet gek
(105) He could not be crazy

Hij kon/kan (ook wel eens/best) niet gek zijn
Since these utterances roughly express the same thing as the PA-constructions
(although there are some slight differences which do need an account at this
level, as will be discussed below) there must be some level of representation
which is shared by all of them. Yet, again, this level cannot be the level of the
predication, for constructions such as (104) or (105) cannot be derived from the
predication of some PA-construction (or vice versa) by syntactic means for
reasons similar to those invoked in 3.3.2 with respect to the alternative PA-
constructions. The structural changes would be much too fundamental to be
acceptable in the functional-syntactic framework of FG/FPG.
The suggestion to assign the same underlying representation to utterances
(104), (105), and the PA-utterances runs counter to the analysis in Hengeveld
(1988). He does not even assign the same representation to the different types
of PA-utterances dealt with so far, contrary to the suggestion above. Hengeveld
considers cases such as (104), and PA-constructions with the M-predicate think
to be 'subjectively modalized', but PA-constructions with modal adjectives, but,
for example, also with the M-predicates believe and doubt (and presumably also
(105), although in his view modal auxiliaries can sometimes express subjective
modality) to be 'objectively modalized'. That is, they involve two different types
of modal qualifications, situated at two different levels in the layered structure
of the clause (cf. 3.1.1 - see also Hengeveld 1987, 1989, 1990 for this distinction).
As mentioned in 3.2.1.3, Hengeveld (1988: 236), who refers to Lyons (1977) for
the distinction, defines objective modality as a matter of S/H's evaluation of an
SoA in terms of general knowledge, and subjective modality as a matter of the
expression of the degree of S/H's commitment to the truth of the content of the
predication put forward. This differentiation is problematic, however. As Lyons
(1977: 797) admits, "its epistemological justification is, to say the least,
uncertain". In fact, in 3.1.2 it was argued that the rationale behind Hengeveld's
layering system in general is questionable, which already means that the
distinction between the two layers on which the two types of modality are
situated is uncertain. The difference in the characterizations of the two types of
modality as such is hard to come by as well: one would expect that the degree
of commitment of an S/H to the truth of the contents of an utterance is precisely
a matter of the evaluation of the SoA expressed in that utterance, in view of
general knowledge.
Of course, there is something plausible in Lyons' (1977: 797ff) distinction
between an utterance in which a modal qualification is the result of estimating
the objective and generally verifiable chances (probably even strictly quantifiab
le) that an SoA can or cannot occur, and an utterance in which the modality
results from purely subjectively guessing whether the SoA can or cannot occur.
But that is maybe a matter of a (gradual) difference in the sources S/H has for
expressing a modal qualification, and this seems to have two closely linked
components, viz. the 'quality' of the sources (in S/H's estimation), and the
question whether the source is (accessible by or known to) S/H alone, or rather
a larger social group to which S/H belongs. The former element determines the
degree of 'firmness' with which S/H puts forward the modal qualification, and
this is most probably identical to the 'uncertainty factor' discussed above. The
latter element determines the degree of 'subjectivity' or 'intersubjectivity' of the
qualification expressed by S/H, and this is a factor which belongs in the realm
of what is generally called evidentiality. Hence, modal qualifications always
involve an S/H's estimation of the chances of an SoA being true in the world or
not. The elements mentioned are only additional conceptual dimensions of or
related to the modal qualification in (103). In fact, though there is not
necessarily always a link between the (inter)subjectivity and the firmness factors,
there is a likelihood that higher subjectivity will cause lower firmness, and vice
versa. That is, if there is a link, the former will be causative to the latter. This
indicates that firmness is more inherent to the modality element than (in-
ter)subjectivity: firmness can be considered an aspect of the modal qualification
proper, whereas evidentiality is a qualification in its own right, next to modality,
although it appears that they easily interrelate.
This is reflected in the fact that the expression of the firmness element
cannot be dissociated from the expression of the modality (and polarity) element
(as appears from the discussion of the uncertainty factor above), whereas the
expression of the evidentiality element is basically independent of it. Thus, it is
true that there may be a difference between the PA-constructions with a modal
adjective in the main clause and constructions such as (104) with a modal
adverb, in terms of the degree of (inter)subjectivity of the expression. This has
little to do with a difference in what would be expressed by a modal adjective
vs a modal adverb as such, however, but rather with the different syntactic
constructions involved. Due to the impersonal nature of a PA-construction of the
type it is possible/likely/etc. S/H can use it much more easily to express a
qualification which is generally accepted or generally verifiable, even with a
reduction of the personal partaking to a bare minimum (cf. the context: okay,
I admit that it is possible/probable/etc. that xyz). A construction such as (104) is
much more neutral in this respect. This is not to say that the PA-construction
with a modal adjective can never be used to express more subjectively motivated
modal qualifications: especially those constructions which score very high in
terms of the 'uncertainty factor' (i.e. very low in terms of firmness), such as
(94d), strongly suggest subjectivity, and the S/H can no doubt always help in
suggesting a more subjective reading through additional expressive devices, such
as special intonation or non-verbal behavior.
Similarly, constructions with an M-predicate of the mental state type (I
think etc.) normally suggest a rather subjective modal qualification (and certainly
much more so than constructions with a modal adverb, such as (104)), again not
because of the predicate itself, but rather because of the first person singular
form (in fact, they cannot get a modality reading if they are not in that specific
form - cf. 3.3.3.3). Actually, it is puzzling at first sight that Hengeveld would treat
believe and doubt vs think as different categories in terms of types of modality,
while they are intuitively close together, and have in fact been treated as one
group throughout the NR literature. Probably, the point is that for believe he
does, but for think he does not distinguish between two readings, viz. what he
calls a 'de re' and a 'de dicto' reading. In its 'de dicto' reading believe means
'believe, accept the proposition xyz', in its 'de re' reading it means 'have the
impression that'. The latter is the reading at stake in the NR debate. It is only
this reading which Hengeveld considers to involve objective modality; the former
is a case of subjective modality. Now, it seems a similar distinction can be made
for think: 'de dicto' it means 'think, have conceptually present or in thought the
proposition xyz', 'de re' it means 'consider it possible, have the impression'. This
distinction (also with believe and other similar predicates) is even more or less
grammaticalized: in the former reading the predicate does, but in the latter it
does not allow that as complementizer for the embedded clause. As with believe,
1only the latter reading is at stake in the NR debate (cf. (84)).71 I assume that
when Hengeveld considers think to involve subjective modality, he means the de
dicto reading; the de re reading would then also involve objective modality. But
then, assuming that at least as far as the rough classification of categories of
constructions is concerned there is some parallelism in what Hengeveld would
call subjective vs objective and what in the present analysis would be called
subjective and intersubjective, it is odd that he would consider the 'de re'
readings of think and believe (and doubt) to be objectively modalized. In fact, at
least according to some of his own criteria to distinguish between subjective and
objective modality he would have to classify them as subjectively modalized, as
will appear below. Moreover, de dicto think and believe do belong in the realm
of evidentiality, but can hardly be said to involve (subjective) modality: the
probability as such of the SoA is not at stake. 'I believe the proposition xyz'
means 'I accept the proposition as being true', and there is no issue of the
chances of it being true.
To support his distinction between subjective and objective modality,
Hengeveld (1988: 236ff) mentions five types of behavioral differences. Let's see
what they tell us with respect to the above reconception of the issue.
(i) Hengeveld claims that objectively modalized predications can be questioned,
but subjectively modalized ones cannot. Cf. (106) and (107).
(106) Is it possible that John will come?
(107) *Possibly John will come?
Even within his own system this suggestion is problematic. Cf. (108).
(108) Do you think/believe that John will come?
One can only read (108) with the de dicto meaning of the main predicate, hence
in Hengeveld's system it would involve subjective modality. Yet, the utterance
is perfectly all right. Now, of course, one cannot really say that (106) involves a
modalization of an SoA on the part of the S/H in any way (though if expressed
with the appropriate intonation it may suggest that the S/H is quite skeptical
about the possibility of the SoA occurring). With (106) S/H only asks H/S
whether (s)he deems it possible that John will come. Similarly, in line with what
has been said above, the de dicto reading of the predicates in (108) does not
involve modality. In fact the difference is quite subtle, but unlike (106), (108)
does not even bring up the issue of the modal qualification of the SoA on the
part of the H/S: it only asks whether in H/S's conceptual world the SoA is the
case or not. This is a direct consequence of the meaning of the main predicates:
while in (106) it does not express a modality on the part of the S/H, it still
denotes a possible modal qualification. This is not the case with the de dicto
readings of the predicates in (108). Of course, it does remain a fact that (106)
can, but (107) cannot be used to ask things about the modal qualification of the
SoA, although a modal adverb also 'denotes' modality. One can think of two
different types of explanations. One explanation may be that asking whether
there is a possibility of an SoA occurring automatically puts the modal element
in an intersubjective perspective: the S/H puts up the modal qualification for
intersubjective discussion. Hence, this requires a construction which allows
expression of the intersubjective dimension. In fact, consider (109).
(109) ??Is it not impossible that John will come?
Referring to the discussion in 3.3.2 concerning the structural ambiguity of the

main clause negative variants, (109) is possible only if not is taken to negate the
modal adjective. The reading in which the negative and the modal adjective form
one functional unit expressing a modal value is impossible. As argued, a very
explicitly negative expression of a qualification (which thus scores very low in
terms of firmness) suggests subjectivity, and obviously too much so for the
purpose of putting it up for intersubjective discussion. Another explanation may
be that since questioning modality makes it informationally salient, the element
referring to the modality factor has to get a prominent place in the expression.
This condition is satisfied in (106), but not really in (107) (cf. also 3.3.3.2). Both
explanations probably have their share in the observations.
(ii) According to Hengeveld, objectively modalized predications can be hypothe
sized in a conditional utterance, but subjectively modalized ones cannot.
(110) If it is possible that John will come, I am going home
(111) *If possibly John will come, I am going home
Yet, again, consider the fact that (112), in which the predicate can only be
understood de dicto and thus presumably involves subjective modality according
to Hengeveld, is perfect.
(112) If you believe/think that John will come, I am going home
However, these observations can be explained in basically the same way as those
in (i). In (110) or (112) S/H does not really modalize the SoA: the question how
(s)he judges the chances of John's coming is not at stake. Nevertheless, while the
probability of the SoA as such is not at issue in (112), it is presented as a
condition for action on the part of the S/H in (110), and it remains a fact that
(110) can but (111) cannot be used to do that. Again this can be explained in
terms of our (inter)subjectivity factor and/or of the informational value of the
topic of modality, in the same way as the first observation. Also consider (113),
which is explicable in the same way as (109) above.
(113) ??If it is not impossible that John will come, I am going home
(iii) Hengeveld claims that subjective modality can only be formulated in positive
terms, whereas objective modality can be both positive and negative.
(114) *Impossibly John will come
He suggests that this is also true for cases with modal auxiliaries: when they are
used to express subjective modality, they cannot be negated.
(115) *John may-not be ill
(116) *John musn't be ill
But then, consider (117).
(117) a. John is, I believe/think, not ill

b. *John is, I don't believe/don't think/doubt, ill
In these utterances only a de re reading of the parenthetical predicates seems

possible, which for Hengeveld would mean that they involve objective modality.
Still, negative formulation appears impossible. Second, as is indicated by Lyons
(1977: 798), the positive variant of (116), (118),
(118) John must be ill
can express both subjective and objective modality (I guess also in Hengeveld's
view). Neither reading is possible in (116) (which only allows a deontic reading),
however. Obviously, the negation excludes both of them. Third, (115) is possible
with a reading in which the qualification is subjective if the negative is somewhat
stressed (strangely enough, even in the variant of (116) without incorporation of
the negative in the auxiliary this reading is not possible). If may is replaced by
might, the subjective reading is even very obvious. Yet, one would assume that
this is still a negative formulation of the modality (albeit a non-incorporated
one). Hence, the issue seems to have little to do with types of modality.
Admittedly, an alternative explanation is far from obvious. Maybe one should
look for one in the area of information structure. Negative polarity as involved
in (103) is rather salient (inherently so) and therefore requires some prominence
in the expression. If an S/H decides to express both the negative polarity and the
modality involved in the qualification of an SoA in one singular predication,
(s)he has to allow the negativity to clearly come to the fore. This is only possible
if the negation element directly affects the predicate, and gets some prominence.
These conditions are not satisfied in (114), (116), or (117b), but they are
satisfied in (104), in (117a), and in (105) and (115) if the negative is somewhat
stressed. In an incorporated variant of the main clause negative PA-constructions
the condition is still satisfied because there the entire qualification in terms of
modality and negativity is prominent (cf. 3.3.3.2).
(iv) Hengeveld suggests that subjective modality is bound to the moment of
speaking, whereas objective modality is not. This would explain why in the few
cases where modals can occur with a past tense form (e.g. may/might) this past
tense does not involve temporal reference but only expresses higher uncertainty,
whereas the past tense in (119) does have temporal reference.
(119) It was possible that John would come, so I went home
However, it has frequently been observed in the NR literature that constructions

of the type / don't believe/think/etc. can only get the de re reading, and
correspondingly an NR-reading, when they are first person present tense (this
has even led to a comparison with performatives - cf. 3.3.3.3). Hence, they are
bound to the here and now of the S/H as well, although according to Hengeveld
they are objectively modalized. Alternatively, in (120), a case involving subjective
modality according to Hengeveld, the modality falls within the scope of the
temporal reference.
(120) Since John was probably going to come, I went home
In fact, in neither (119) nor (120) does the modal adjective or adverb really
express a modal qualification of an embedded SoA by S/H. What we have here
are cases where the qualification is itself part of an SoA, which is reported on
by S/H (i.e. a 'mental model within a mental model' - cf. 3.2.1.3). The current
evaluation of the SoA by S/H is not at stake. It seems that direct modal
qualifications by an S/H are always bound to the hic et nunc, in the sense that
there must be a commitment from S/H to the modal qualification at the very
moment of speaking.
(v) Finally, Hengeveld claims that in reaction to objectively modalized
expressions the source of the information can be questioned, as in (121), but in
reaction to subjectively modalized ones that would be out of place, as is shown
in (122b). An adequate reply would then rather be (122c).
(121) a. It is possible that it will rain tomorrow

b. Who says so?
(122) a. Possibly it will rain tomorrow

b. *Who says so?
c. Do you think so?
It is not so certain that Hengeveld is right in excluding the sequence (122a-b),

although he is certainly right that (122a-c) is more likely to occur. On the other
hand, (122c) is possible as a reaction to (121a) too, although in this case (121b)
may be more likely. Now, also consider (123).
(123) I believe/think/doubt it will rain tomorrow
Taking (123) in its de re reading one can certainly reply with (122c), but hardly
with (121b), despite the fact that in Hengeveld's view this would be a matter of
objective modality. It seems some other explanation for these observations is
required. In fact, it would be rather strange to observe that the source of an
objectively modalized utterance can be questioned but that of a subjectively
modalized one cannot, in view of Hengeveld's assumption that subjective
modality is one type of 'epistemological modality' (which is basically what is
elsewhere called evidentiality, i.e. the issue of sources), whereas objective
modality is not. In reality, (122c) is obviously also a matter of questioning the
source, as is (121b), and the difference between (121a) and (122a) has nothing
to do with whether sources are involved or not. The reason why (122c) is easier
than (122b) as a reply to (122a), whereas (121b) is easier than (122c) as a reply
to (121a) is perfectly explicable in terms of our factor of (inter)subjectivity again.
In more intersubjective expressions such as (121a) one can ask for the group of
people holding this opinion, since the S/H can be assumed to be only one of
them (in fact, using (121b) is a hedging strategy in which one spares S/H by not
directly questioning his/her opinion). In less intersubjective expressions such as

(122a), the S/H is more likely to be responsible for the expression, and
questioning the source is more likely to involve questioning the S/H directly. In
clearly subjective expressions such as (123) one simply cannot avoid questioning
the S/H if one wants to question the source.
In addition to these criteria, Hengeveld (1988: 245) also suggests a
relationship between his subjective and objective modality and the issue of NR:
objective modality would allow raising, whereas 'epistemological modality' (and
thus subjective modality) would not. However, as appears from the above, PA-
constructions with predicates such as think or believe in their de re readings
would according to his criteria (iii), (iv), and (v) have to be classified as
subjectively modalized, yet they do allow NR-readings. It is more likely that NR
is restricted to predicates which clearly express modality (both subjective and
intersubjective), while predicates mainly expressing evidentiality cannot have NR-
readings. I return to this in 3.3.3.3.
5.5.5.2 Differentiating the alternatives

Thus, relating level (103) to the level of predications in FG/FPG is not only a
matter of differentiating between the alternatives in the PA-construction pairs,
but also of differentiating between the PA-constructions and equivalents such as
(104) and (105). As has appeared from the above discussion, part of the
explanation of S/H's choice for one of the alternatives can be found in additional
conceptual factors, to be represented as additional elements at the level of
representation in (103) itself. One of them is the element of (inter)subjectivity,
involving the source behind the modal qualification. As mentioned, PA-construc
tions with a predicatively used modal adjective more easily permit the expression
of intersubjective modality than utterances with a modal adverb, while PA-
constructions with M-predicates such as believe are best suited to suggest
subjective modal qualifications. This is only a part of the story, however. As
argued, the fact that impersonal PA-constructions are well-fitted for intersubjec
tive modality does not mean that they cannot be used for expressing more
subjective modality, too. The difference between these PA-constructions and
alternatives with modal adverbs can also be ascribed, then, to a difference in the
communicative value of the information. Obviously, PA-constructions such as
(94a-b) allow a much clearer functional distinction in terms of the relative
informational salience of the PA, or a part of it, and the SSA than adverbial
constructions such as (104). To see the exact role of the information-structural
factors involved, however, it is necessary to consider at once the differences
between the main-negative and embedded-negative variants within the PA-
constructions, for it appears that there is an interference in this respect.

Some of the distinctions between the variants in our PA-constructions can
be ascribed to a conceptual factor again. The choice between alternatives with
various degrees of negative incorporation in the main clause (e.g. probable vs not
improbable, improbable vs not probable, doubt vs not believe) is undoubtedly
determined by degrees of 'firmness' of the modal qualification, along the lines
indicated in 3.3.2. This factor seems less decisive in the choice between the main
clause negative vs embedded clause negative alternatives such as (94a-b) or
(99a-b), however. As noted, there is a difference in the 'strength' (i.e. firmness)
expressed by these variants, too. Yet, the difference in this respect is very small,
and can, in actual contexts, certainly diminish to barely relevant. In determining
the choice between these variants, then, we might rather look for differences in
the possibilities the variants offer to distinguish functionally between elements
of the information in (103), again. What seems to be involved is the information
al value (the topicality or focality) of the polarity factor and the modality factor
separately, and relative to (parts of) the SSA. As noted in 3.3.2, expressing the
negativity and the modality together in the main clause does not allow the
expression of a distinction in informational value between them, since they have
to form a lexical-functional unity there (otherwise they would indicate different
scope, and would not express the qualification in (103) anymore). A distinct
informational value of either of them can get expression if the negativity is
expressed in the embedded clause, however.
Considering this matter in the light of the global differentiation between
(94a), (94b) and (104), then, (94b) is likely to be chosen if the PA in general has
a strong focal value, while the SSA in the embedded clause is topical. (94a), on
the other hand, may be chosen if there is a strong informational distinction
between the components of the PA, and if the modality factor is strongly focal
relative to the SSA. That is not to say that the negation cannot be salient in this
case: if the argumentation with respect to Hengeveld's criterion (iii) above is
sound, negation of an SSA is always salient to a certain degree. This is reflected
in the explicitness and the intonational prominence of the negative element in
the embedded clause. Still, the negation and modality elements can have a
different type of informational prominence (e.g., they can have a different type
of focal value), or a different degree of focality. (104) is chosen if there is no
strong informational distinctiveness between the modality and the SSA.
At a more basic level, the differentiation between the various alternatives
may be carried back to the discourse functional status of the information in the
SN (in terms of thematicity/rhematicity and givenness/newness, which as said in
3.2.3.3 co-determine the assignment of topicality and focality values at the level
of the SSA), viz. the way in which the SSA under consideration has been
introduced into the discourse, and more particularly whether it has been
presented in a positive or a negative way, or whether the modality of the SSA
has been under discussion or not.
Actually, the above analysis of the differences between alternative
expressions of (103) in terms of information structure may also provide an
explanation for the phenomenon sometimes noted in the NR literature (and
called the 'embedded negation constraint' in Horn 1978a: 180), that in some
potential NR-pairs only the main-negative variant seems possible, while the
embedded-negative variant sounds awkward. This impression is likely to be the
result of the fact that in the NR literature, utterances have nearly always been
considered in isolation, out of possible usage contexts. As argued in 2.2.2.3, in
such cases one nevertheless normally unconsciously involves a context in judging
an utterance, and this is likely to be the most neutral context. If the same
context is used in judging both variants, the variant best suited for the most
neutral context will be preferred, and given the relative effect of judgments (cf.
2.2.2.3) this will increase the feeling of unacceptability with respect to the other
variant (even if it would be judged OK if inserted in an appropriate context).
Now, in the syntactic approach to NR the embedded-negative variant would
normally be considered the most neutral variant, since the other one is derived
from it. In that view the embedded negation constraint is obviously inexplicable.
In the above analysis the main-negative variant is obviously the least marked
one, since it is the most direct (or iconic) expression of (103), and since it
requires the least marked contextual conditions. This solves the mystery.
The above analysis again underscores the argumentation earlier in this
chapter, not only for levels of representation in a grammar more abstract than
the FG-predications, but also for the introduction of pragmatic functions at those
deeper levels rather than after the predications. They are simply unavoidable to
explain the process of predication formation and term selection. These claims
are further stressed if one tries to account for the structural ambiguity of the
main clause negative utterances, and for the relationship of their non-PA-
construction reading and the 'cleft negation constructions'. Cf. (124) ((124a) =
(94b)).72
(124) a. It is not probable that he is crazy

b. It is not the case that it is probable that he is crazy
Het is niet zo dat het waarschijnlijk is dat hij gek is
At which level should we account for the structural ambiguity of the main
clause negative constructions as discussed in 3.3.2? If we opted for the level of

predications and assigned the construction two different representations there,
we would be forced to handle the difference in terms of scope of the negation
at this level. However, as also argued in 3.2.3.3, this is likely to cause problems:
it seems difficult (also within Hengeveld's layering proposal at the predicational
level) to provide an account for the difference between a main clause negative
operator which has the predicate within its scope, and one which correlates with
the predicate. In fact, it is not necessary at all to handle matters of scope in the
predication if one takes into account representational level (103). Evidently, at
this level both readings must obtain a different representation, according to the
difference in conceptual meaning (cf. 3.3.2), and for the non-PA-reading the
representation would be like (125) (again, x1 = 'he' and x2 = 'craziness').
(125)
Thus, we get a qualification (QQ) (more specifically a negation (N)) of a

qualification (for positive polarity (P) and modality (Mx)) of an SSA. The
qualification of the embedded SSA and that SSA (<SSA>) together form
another SSA (< <SSA> >), and in line with other similar cases discussed above,
in these circumstances the M-predicate does not express modality for the S/H
anymore, but 'quotes' a possible qualification held by someone else (maybe S/H
him/herself, but then at a different moment, not here and now). We can assume,
then, that both conceptual structures (103) and (125) can be mapped (under the
appropriate conditions) onto one and the same lexical pattern at the level of the
predications. Still, the conceptual distinction between them does not get lost, for,
at least in situations in which the structural ambiguity would not automatically
be disambiguated by the (verbal and/or non-verbal) context, the S/H can use
expressive devices such as intonation to mark the specific status of the negative,
in terms of its scope. In this respect, (125) can lead to the two different
intonation patterns noted in 3.3.2, and as argued there this depends again on the
exact informational value of either the level <SoA> Q' or the level
< < S o A > > Q'.
As to the mapping of (125) onto the level of predications, the choice
between construction type (124a) and the cleft negation construction (124b) is
once again a matter of informational value, and it seems that this is very close
to the matter of the relationship between cleft and pseudo-cleft constructions
and their non-cleft alternatives. (124b) is clearly akin to the cleft constructions,
and one can even argue that it is a cleft construction. Functionally at least, they
are identical: (124b) is a structure intended to focus strongly upon the negativity
of the embedded clause, much more strongly than could be done in construction
(124a).73 Structurally, there seems to be some difference with a real cleft, yet
it is questionable whether this is fundamental. A real structural clefting of the
non-PA-construction reading of (124a) would result in utterance (126).
(126) It is not that I think he is crazy

Het is niet dat ik denk dat hij gek is
However, (126) is not a construction focusing upon the negative, but rather one
putting contrastive focus upon the entire embedded clause. If one really wants
to form a focusing construction for the negation in (125), this apparently
requires the introduction of some 'dummy' element such as the semantically
vague noun the casel het geval or the fuzzy particle so/zo in the 'cleft' clause,
which at once anticipates the factivity (the positive polarity) of the SSA to follow
and which is in the scope of the main clause negative (i.e. it is the factivity which
is going to be denied). Anyway, whether or not one accepts the structural
identity claim of (124b) and clefts, it will be clear that Dik's (1980a: 210ff)
argument that it is impossible to syntactically derive clefts and pseudo-clefts from
their non-cleft alternatives for reasons similar to those given in 3.3.2, is equally
valid for (124a) as compared to (124b). At the level of the predications they
must be considered to have different representations already, and the crucial
'split' in their derivation has to occur between level (125) and the level of
predication, on the basis of different pragmatic function assignments. Note also
that this approach has definite advantages over the FG-treatment in terms of
'elegance' of the grammar. In order to account for the specific information-
structural status of the clefts and pseudo-clefts Dik (1980a) has to introduce a
special type of predication with a fixed distribution of pragmatic functions, which
excommunicates the normal pragmatic function assignment procedures in FG.
The same would have to be done for (124b). This is a rather arbitrary system
within the FG machinery. In our present approach no such ad hoc mechanism
is required.
3.333. Further lexical and syntactic issues

Finally, let's see what kinds of solutions may be provided in the present context
for some of the remaining problems with the lexical and syntactic behavior of
NR-constructions already mentioned in 3.3.1. One problem as yet not considered
is that of lexical exceptions to NR, involving predicates which could be expected
to allow NR-readings yet do not, and the interlanguage variation in this respect.
Thus, as shown in (91) and (92), Dutch hopen does, but English hope does not
allow an NR-reading. On the other hand, neither fear nor its Dutch equivalent
vrezen allow the NR-reading. As noted in 3.3.2, the SCI analysis of this cannot
be appropriate, but the following observations suggest the direction for a
solution.
It has frequently been noted but never seriously been explained (Horn
1978a: 188ff) that M-predicates such as believe do not always allow the NR-
reading of a main clause negative. Upon closer examination, they appear not to
if they are used in the 'de dicto' reading (cf. 3.3.3.1), which is intuitively their
basic or 'literal' sense. Thus, (127a), in which not believe means 'not accept your
claim', cannot be used to communicate the same message as (127b), not even
if the sentence parts between brackets are not present, but if the remaining part
of the utterance is taken in the same sense.74
(127) a. I don't believe (you when you say) that he is crazy

Ik geloof (je) niet (als je zegt) dat hij gek is
b. I believe (you when you say) that he is not crazy
Ik geloof (je als je zegt) dat hij niet gek is
In line with what has been said in 3.3.3.1, what occurs here is that (127) does not
express a qualification of the SoA of the level involved in (103), in terms of
modality and polarity. In this reading, predicates such as believe and think
involve an indication of (an attitude toward) a source, hence they express
evidentiality, not modality (though they may suggest or imply a certain modal
qualification, in a similar way as evidentiality is suggested or implied in the
expressions of modality as discussed above). Hence, the observation in (127) is
perfectly understandable if one assumes that NR-readings are limited to
expressions of conceptualization (103).
On the other hand, it has been noted that many predicates not belonging
to the 'standard' categories of M-predicates nevertheless allow an NR-reading
of a main clause negative if they are occasionally used for expressing (103) (cf.
Horn 1978a: 204ff). (128a), containing a verb of communication, relates to
(128b) in the same way our traditional NR-variants do (though this is probably
somewhat less obvious in the Dutch example).
(128) a. I wouldn't say he is crazy

Ik zou niet zeggen dat hij gek is
b. I would say he is not crazy
Ik zou zeggen dat hij niet gek is
Similar observations can be made for example with respect to perception

predicates such as seem.
These observations clearly suggest that the mental state predicates, just like
the verbs of communication or perception, are not 'standard' categories of M-
predicates at all, contrary to what is generally assumed. In fact, it seems obvious
that the only M-predicates which can be called basic are those which directly
name the modality contained in the qualification in (103), i.e. the predicatively
used modal adjectives of the category of be certain, be probable, be possible, etc.
Predicates from various other categories can be used for the purpose of
expressing modality, too, but this capacity depends on the nature of these
categories. Their semantics must allow the unambiguous indication of the nature
and value of the modality, in a simple and straightforward way. It is evident that
in general the semantics of the mental state predicates, and probably also of the
perception predicates, qualifies better for this purpose than that of the
communication predicates. Hence the former are more commonly used for the
expression of (103) than the latter.
This view on the basic and non-basic predicate categories for expressing
(103) corresponds with the observations concerning the different occurrence
properties which must be fulfilled by the predicates of these categories to allow
an NR-reading of the main clause negation (again this is valid for all M-
predicates, not only the mid-scalars in Horn's model). Thus, while such a reading
is possible for the basic predicates under any condition, the mental state
predicates, perception predicates, and communication predicates must be used
in the first person singular, present tense, and without the complementizer that
linking main and subordinate clause. As Cattell (1973) notes, it is sometimes
possible to get NR-readings with third person subjects in the main clause (but
still, present tense is required), yet only so in reported speech, i.e. when the S/H
explicitly reports the subject's view live, so to speak. In addition, the communica
tion predicates must occur in the conditional form.
By the way, the characteristic of mental state predicates allowing an NR-
reading only when they are first person present has prompted Lakoff (1969) to
qualify them as performatives. As Cattell (1973) argues, such a qualification is
not appropriate because these predicates do not signify and perform an action
of S/H. Nevertheless, it has generally been acknowledged, without explanation,
that they must have something in common with the performatives. In the light
of the nature of the conceptual representation in (103), this common characteris
tic can be made comprehensible. Performatives and M-predicates both express
meta-level elements with respect to a particular SSA, for which S/H holds
responsibility in the actual communicative situation. The difference is that in a
performative this meta-element involves the specification of the kind of goal S/H
has with respect to the SSA under consideration, while in an M-predicate it is
S/H's qualification of the epistemic status of the SSA. Just like performatives, if
S/H uses the predicate with a third person subject in reported speech, (s)he
adopts the subject's conceptualization of (103), and 'performs' it for the subject,
so to speak. Also, just like performatives, if the predicate is used in the past
tense it does not qualify the SoA here and now, but it expresses some (in case
of all non-standard types of M-predicates even different) attitude which the
S/H/subject once had, irrespective of the attitude here and now. In that case, the
attitude is part of the SSA under consideration in the discourse.
Within the different non-basic categories of M-predicates, then, not each
predicate qualifies equally for the expression of (103). In fact, the same
requirements as for the categories in general seem to apply. They must allow a
straightforward expression of the M, i.e. they must allow the indication of a
personal point of view concerning the chances that the SSA is true or untrue,
without too much distortion. In lexical semantic terms, this might be reflected
in the requirement that the predicate has a relatively simple structure in which
a strong modal component is present or possible. This might explain why among
the communication predicates a relatively simple predicate such as say/zeggen
can be used for expressing the M in (103), while suggest/suggereren and
propose Ivoorstellen, e.g., which involve a much more complex attitude concerning
the SSA, cannot.
It seems that precisely the same principle can be invoked to exclude such
mental state predicates as hope and fear from the possible M-predicates. As
compared to predicates such as believe or know, they express a much more
complex mental state which does involve some element of modality but also a
strong component of emotional attitude in terms of liking or disliking the SoA.
The predominance of this meaning component makes these predicates unfit for
the purpose of expressing (103). The interpersonal, interdialect and interlan-
guage variation with respect to the possibility of using such predicates for
expressing a modal qualification should be ascribed, then, to differences in the
exact lexical meaning of these predicates. Thus in Dutch hopen, the emotional
component might be weaker than in English hope. The Dutch predicate also
seems to leave open the question whether the SoA is realized or not, more so
than the English predicate: hope might suggest that the SoA is already realized
in reality; Surely, this claim requires further investigation on the basis of a
comparative analysis of the uses of the predicates in Dutch and English.
As mentioned in 3.3.1, next to the lexical peculiarities the NR-constructions
also have an apparently mysterious syntactic behavior. I will not be able to
consider the many problems this involves in detail here, since most of them
require an extensive investigation of their own. For example, a discussion of the
famous problem with the negative polarity items (Lakoff 1969) would first
require a serious understanding of the tricky issue of negative polarity in itself,
but the state of the art is still a far cry from what would be sufficient to tackle
its interaction with the NR-problem. I will briefly discuss two phenomena,
however, since they can easily be used to demonstrate the plausibility of the
above analysis.
First, consider the possibility of adding tag questions to the embedded
clause in PA-constructions. In Lakoff's (1969) view, assuming that tags normally
have opposite polarity relative to the utterance to which they are attached (only
if they are used ironically they may have same polarity according to her), it is
remarkable that in main-negative NR-constructions a tag requires the same
polarity as the embedded sentence.
(129) a. I don't suppose the Yankees will win, will they?

Ik veronderstel niet dat de Yankees zullen winnen, wel?
b. *I don't suppose the Yankees will win, won't they?
*Ik veronderstel niet dat de Yankees zullen winnen, niet?
Cattell (1973) has already shown that Lakoffs assumption that tags normally
take opposite polarity is not true, however. Consider (130).
(130) a. He isn't crazy, is he?

Hij is niet gek, wel?
b. He isn't crazy, isn't he?
Hij is (in feite) niet gek, is het niet?
The use of tags with opposite polarity, as in (130a), indicates that one is
presenting one's own point of view, which one considers to be open for
discussion however, or about the accuracy of which one has no absolute
certainty. The use of a tag with same polarity, as in (130b) (which is not as rare
as Lakoff assumed), rather indicates that one is forwarding a point of view which
is conceivable in one's current UI, but which was not one's own until the
moment of speaking. Obviously, the use of tags is again closely linked to the
issue of expressing qualifications of SoAs in which polarity and modality are
crucial elements.
Now, since according to the above analysis PA-constructions always express
a modal and polar qualification on the part of the S/H, one would expect tags
to necessarily have a polarity opposite to the value of the qualification of the
SSA in (103). This should be the case no matter how the polarity is expressed
in the main and/or embedded clause, and this should pertain to the extreme-
scalars in Horn's model as well as to the mid-scalars. That this is indeed the case
is shown in (131), (132), and (133). (I have omitted the Dutch equivalents
because Dutch tags are more cryptic than English tags. In the given examples
this would make it hard to see whether they concern the embedded or rather the
main clause.)
(131) a. It's impossible that he has done it - has he?/*hasn't he?

b. It's certain that he hasn't done it - has he?/*hasn't he?
(132) a. I don't believe he has done it - has he?/*hasn't he?

b. I believe he hasn't done it - has he?/*hasn't he?
c. I doubt he has done it - has he?/*hasn't he?
(133) a. I don't know he has done it - has he?/*hasn't he?

b. It's possible that he hasn't done it - has he?/*hasn't he?
c. It's not impossible that he hasn't done it - has he?/*hasn't he?
As to the derivation of the TQs, then, in our FPG framework this is not a matter
of syntactic expansion (there is no room for such a procedure). They must
already be introduced at the level of the predications, more or less in the same
way as a tail-constituent, and the cause for their introduction and the determina
tion of their polarity are entirely matters for the pre-predication-levels. The
decision to construe them is a matter of the S/H's intention to receive the H/S's
opinion (probably this is a matter of the 'firmness' of the qualification again - cf.
3.3.3.1); the polarity is determined by the value of the qualification at the
sentence level conceptual representation (103).
Finally, a short remark on sentence pronominalization is due. Lindholm
(1969) (cf. also Horn 1978a: 160) has noted that in an utterance like (134)
(134) I don't think Bill paid his taxes and Mary is quite sure of it
Ik denk niet dat Bill zijn BTW betaald heeft en Marie is er zeker
van
it refers to 'Bill didn't pay his taxes', i.e. to the negative, not the positive version
of the embedded part of the first clause in the coordination. But the observa
tions in this respect are again not as simple as has been assumed. First, note that
also for this phenomenon it is easy to provide examples involving extreme-scalar
predicates in Horn's model. Thus, in (135)
(135) It is impossible that Bill paid his taxes and Mary knows it, too
Het is onmogelijk dat Bill BTW betaald heeft en Marie weet dat
ook
what Mary knows is that Bill hasn't paid his taxes. Again, this is an argument for
our restatement of the problems involved. On the other hand, it is also possible
to find examples in which this principle does not work, even in case of the
presence of mid-scalars. In (136)
(136) I don't think Bill paid his taxes and Mary doubts it, too
Ik denk niet dat Bill BTW betaald heeft en Marie betwijfelt het ook
what Mary doubts is that Bill has paid his taxes, not that he has not paid them.
Yet this does not really cause serious problems. Obviously, the different behavior
of the anaphor in (136) is the result of the fact that it is also within the scope
of an M-predicate expressing a qualification in terms of modality and negative
polarity (which is 'performed' by the S/H for Mary, so to speak). Hence, the
coreferentiality is not only a matter of correspondence between the SSAs but
also between the qualifications expressed in the two coordinate sentences, and
the way the M-predicate in the second coordinate sentence brings up the
qualification determines how the pronoun can be understood.
A crucial element in obtaining this effect is certainly also the coordinator
and/en, which suggests that the 'trend' in the second sentence continues that of
the first. If one replaces and/en by but/maar in (134), the pronominal reference
changes. With the former, the implication is that the polarity of the SSA remains
the same in the second sentence, and the S/H can repeat it in the expression of
the qualification there, which leads the anaphor to refer to the SSA as such.
Alternatively, in the expression of the qualification in the second sentence the
S/H can concentrate on the slight difference in modal qualification assigned to
the SSA by Mary and leave the polarity factor out, which leads the anaphor to
refer to the SSA and the polarity factor. Again, I cannot go into details of how
precisely to model this in FPG, but as to the principles of such an account,
according to the basic assumption that no structure changing operations are
allowed in the syntactic processing of utterances, it cannot be the result of
syntactic deletion. In fact, as argued in 3.1.3.3, anaphoric reference is, once
again, a purely conceptual and strongly discourse-bound matter which is to be
accounted for at pre-predicational levels, and it is introduced as such in the
underlying predication.
This ends our present discussion of NR and related problems. No matter
how incomplete the analysis may be, it seems to have underscored three central
issues of this volume: (i) A functionalist approach to language opens perspec

tives for the analysis of utterances which in a non-functionalist approach would
probably remain hidden. Only the functionalist perspective could lead us to the
more global picture of the construction type involved here, which unmasks NR
as a fake problem based on incomplete and biased data. Functionalism once
again appears to be an indispensable tool for the linguist. (ii) At the same time,
our analysis suggests that a functionalist approach is not necessarily a superficial
or concrete one, contrary to what is sometimes claimed (e.g. by Dik). An
argumentation on the basis of the functionality of utterances sometimes requires
distance from the superficial appearance of utterances, since this appearance all
too frequently covers what is really at stake, as the NR case has shown. (iii)
Specifically concerning FG and FPG, our analysis underscores the view defended
in FPG that in the process of language production one must assume levels of
representation and processing deeper than the level of the predications, which
systematically lead to the choice of structures at the level of the predications on
the basis of the informational value of the various elements at these deeper
levels. It seems unavoidable to assume that utterance production is an intricate
matter of lexical choice and syntactic expression in the light of a complex system
of underlying cognitive-pragmatic relationships. Hence, FG really needs
elaboration in the direction of the deeper levels of representation and processing
proposed in 3.2, if it is to remain a successful linguistic model. More generally,
as argued in 1.1, a much more integrated type of modeling than is currently in
practice in linguistics is required to be able to successfully deal with complex
issues such as NR.
Notes
1. Recent interest in computer applications of FG has led to attempts to develop more

explicit formalisms (e.g. Kwee 1979, 1981, 1987, Van der Korst 1987, Weigand 1987, and
several contributions in Connolly and Dik (eds.) 1989), but so far this is rather a 'parallel
circuit' to FG as a linguistic model, which has not led to stricter formalization of the
model in Dik (1989a), e.g.
2. Very recently (cf. e.g. Dik fc.) there is some attention for what are called 'extra-clausal
constituents'. However, this term suggests that discourse is again considered from the
perspective of the individual utterance, rather than in its own right as something with a
structure which cannot be understood from the perspective of any single utterance within
it alone (cf. also the way 'themes' (cf. note 6 below) are handled in FG - Dik 1981b, De
Groot 1981a). Thus, FG seems to inherit a basic attitude typical of traditional linguistic
research, yet one which is very biassing and potentially misleading.
3. The following summary overview is based on Dik (1978a) and especially Dik (1989a),
which differs from the former in several respects. There are competing opinions among
FG-researchers on many issues, but I will not pay much attention to them (with one
important exception). They will be mentioned in the later discussions if they are relevant
to my considerations. Also, references to the quite substantial literature are kept to a
minimum. Among the most representative publications are a series of collective volumes,
including Bolkestein et al. (1981), Hoekstra et al. (eds., 1981), Dik (ed., 1983), Bolkestein
et al. (eds., 1985a, 1985b), Van der Auwera and Goossens (eds., 1987), Nuyts and De
Schutter (eds., 1987), and Nuyts et al. (eds., 1990). Other overviews of the model can be
found in Dik (1980a: ch.1, 1980b, 1981a, 1983a, b) or Hoekstra (1981). Yet, none of these
covers the recent innovations as represented in Dik (1989a).
4. At present there are a number of competing versions of the layered structure (cf. e.g.
Hengeveld 1989, 1990, Dik 1989a, Vet 1990). But Hengeveld's version was the first one
to be proposed, and still seems most widespread among FG-ers.
5. Dik (1989a) does not accept an illocutionary frame, but rather handles the illocution as
a 4 operator.
6. In Dik (1978a) there were also two clause-external pragmatic functions, viz. theme (a
constituent like as for Paris, (the Eiffel Tower is ...)) and tail ((a nice chap), your brother).
From Dik (1989a) on these are only two of a series of extra-clausal pragmatic functions
used to characterize the status of extra-clausal constituents relative to the clause.
7. For lengthy replies to these articles, see Dik (1979c), Nuyts (1983b), Nuyts (1986b), and
Mackenzie (1988) respectively. In the following only the most relevant aspects of these
criticisms are discussed.
8. Obviously, the question whether the presence of relational notions is motivated by a

functional perspective does not depend on whether one can provide an answer to the
question whether these relational notions enhance the functionality of a language (as is
claimed by Miller 1986: 194).
9. According to Miller (1986: 196ff) syntactic functions are also present in TG, since they
can be characterized in terms of configurations of category notions in the core of the
grammar. He suggests (ibidem: 172) that it does not matter whether such matters are
expressed directly or indirectly in a grammar (hence, this would be another non-difference
between TG and FG). Yet, this is another example of mystification of the meaning of a
model. Either syntactic functions are relevant concepts for characterizing the structure
of language, in which case they have to be directly and unambiguously part of the
grammar; or they are not relevant (they might be epiphenomena of something else), in
which case they have to be omitted. 'Relevant but not present' is an 'empty category'.
10. An example of a system in language psychology using all three types of relations is
Osgood (1980, 1984). Yet, his semantic and pragmatic dimensions are quite different from
those in FG. A comparison would be very useful, but is beyond the scope of the present
study.
11. Johnson-Laird (1977b: 85ff) has suggested that semantic functions might be a superfluous
category which can entirely be accounted for in terms of conceptual correlations and
configurations. The arguments are not very clear though. Moreover, see 3.2.3 on the
matter of conceptual roles and semantic functions in the grammar.
12. Dik (1989a: 51ff) suggests that the same system, but then applied in the opposite way,
should be able to function in a language interpreter too. That is, he commits himself to
the claim that the same linguistic knowledge systems should be usable in production and
understanding. This kind of question is addressed in 3.2.2.
13. Although FG and Kempen and Hoenkamp's 'Incremental Procedural Grammar' have
many things in common, here is a point of difference. Despite their distinction between
a 'tree hierarchy structure' (predication in FG) and a word ordering system (expression
in FG) Kempen and Hoenkamp do assume a fixed structural order already in the former.
Still, their transformations have much less structural impact than those in TG.
14. Pronouns are not normally expressed as independent constituents in Choctaw, but they
do cause agreement marking on the verb.
15. Choctaw has both case marking on the verb (for nominative, accusative, or dative), and
case marking of constituents (either nominative or objective).
16. The following heavily draws on research done in collaboration with several others and
presented in Dik et al. (1983). For a much more detailed presentation of the typology,
semantics, and behavior of the construction, I refer to that paper. Yet my interpretation
of the matter here is certainly not shared by all of my fellow investigators. Dik objects
to accepting 'experiencer' as a full-fledged semantic function (for reasons discussed in the
above-mentioned paper), whereas I do consider it to be a semantic role, but then in the
framework of the more sophisticated conception of roles and cases already suggested in
3.1.3.1.
17. The extensive literature on this construction includes Kachru (1970), Sridhar (1976,
1979), Shibatani (1977), Cole and Jake (1978), Perlmutter (1979), Cole et al. (1980),
Nichols et al. (1980), Klaiman (1980, 1981), Kishna (1981), and Harris (1982). Among
the languages cited there are Kannada, Japanese, Italian, Bengali, Quechua, Hebrew,
Russian, Hindi-Urdu, Sarnami, and Georgian. Also Dutch and German have these
constructions.
18. I.e. the marking of a complement clause for whether its subject is the same or different
from the subject of the main clause. Davies does not mention this property in Choctaw,
but in some languages the experiencer can have this property too.
19. One could even argue that the specification of more than one ordering template for a
language, a possibility which Dik a priori grants, should be avoided. Dik's position in this
matter is determined by his attempt to avoid abstractness in representing actual surface
patterns. Yet, this principle can lead to a failure to see relevant generalizations. De
Schutter (1985a, b) has argued that it is perfectly possible to accept one ordering
template for Dutch and German (instead of two as proposed by Dik) if one is willing to
rely more strongly and systematically on pragmatic function categories. The result is a
somewhat more abstract template, but the generalization obtained is still perfectly
functionalist in spirit. In fact, it is very unlikely that functionalism (if consequently
applied) and concreteness really go together, contrary to what Dik assumes.
20. Officially, this is the third argument in FG. Yet, there are reasons to assume that in
passives the agent has no longer argument status - cf. the possibility to omit it, or the fact
that it is marked by a preposition. The present considerations rather support this. See
Pinkster (1985) on this problem in FG. This would of course mean that passive formation
is no longer a matter of alternative syntactic function assignment, as in current FG (in
the passive the subject is assigned to the second argument), but would have to be treated
in predicate formation. See also De Schutter (1989) for such a proposal.
21. This principle is no doubt complicated by the informational value (topicality/focality) of

entities in the discourse domain, but for the present discussion this is of no concern.
22. Bolkestein (1981b) also uses a semantic factor, viz. factivity. Yet, the question remains
whether there are not also more information-structural elements involved.
23. Note that one cannot state the motivation for raising in terms of the absolute
informational value of one or another constituent in the embedded clause, but only in
terms of a discrepancy in their informational value. This is in line with the suggestion by
De Schutter (1985a, b, 1987) that word order principles related to the informational value
of constituents may have to be specified not so much in terms of specific pragmatic
functions, but rather in terms of the relative informational value of constituents.
24. One additional problem with AI, which has the lion's share in research into the matter,
is that it is strongly application-oriented and often does not really care about the
cognitive plausibility of systems for knowledge representation or organization it develops.
25. It may sometimes be the case that only part of the situation-dependent knowledge is
relevant to the conversation itself. One can acquire knowledge about features of the
environment which have nothing to do with the factors involved in the communicative
interaction. For example, during a talk with a colleague about a piece of work one can
note that he is wearing an unusual tie (a green one instead of the red ones he normally
wears). This may have nothing to do with the conversation, but can still have retention
value and can thus be remembered for some time, probably even in the encyclopedia.
26. Maybe it is still possible that S/H can have a hypothesis about H/S's hypothesis about
S/H's hypothesis about H/S. Further embeddings seem very unlikely, however, since they
would hardly be conceivable for a normal individual (as with syntactic embeddings).
27. In De Schutter and Nuyts (1983) this was called the 'communicative universe of
interpretation'.
28. Not all of the information in the encyclopedia potentially relevant for the actual situation
is necessarily activated as part of the UI: S/H may ill-judge the relevance, or may simply
overlook relevant information. S/H may also misjudge actually irrelevant information as
relevant and include it. Hence, the notion of relevance for S/H, and of consequent
inclusion in the UI, must somehow be linked to an element of conscious decision on the
part of S/H.
29. For explicit arguments for this claim, see e.g. Schank (1972a), Chafe (1977a), Clark and
Clark (1977), Osgood (1980), Steinberg (1982), Kintsch (1984). Some of the arguments
below are similar to some of the arguments in this literature.
30. This proposal was formulated before the layered structure of the clause was introduced
in FG. At present it is not clear to me whether the layers would be included in
conceptual representations according to Dik. I return to this matter below. In any case,
for Dik the conceptual representations do not include the illocutionary operators (which
were already introduced in FG at that time - cf. De Jong 1981, Moutaouakil 1986), and
they also do not include the pragmatic and syntactic function markings.
31. This principle also works for term-modification: For example, English only allows a gram
matical marking for 'one' (singular) and 'more than one' (plural). Some languages also
have a marking for 'two', or for 'few'. But no language has a grammatical category for
every numeral it has.
32. One should obviously not confuse the fact that one has situation-specific knowledge
(knowledge about very specific situations) and the question whether this knowledge is
stored in a situation-sensitive way.
33. By reconstructability I obviously mean what Hjelmquist (1990) calls Veridical reconstruc
tion'.
34. In itself, this is not an exceptional proposal. There has been an extensive debate in the
psychological literature about the question whether conceptualization is propositional or
pictorial, and some 'pictorialists' (e.g. Kosslyn 1978) also claim that mental representation
consists of a mixture of propositional (i.e. descriptional) and pictorial representations. I
will not go into the 'imagery' debate here (an overview in Block (ed.) 1981, Johnson-
Laird 1983: 146ff).
35. These considerations are formulated from a Piagetian rather than a Wygotskian perspec
tive. However, even in Wygotski's (1974) view, in which language (inner speech) plays a
crucial role in developing general conceptual capacities (which certainly is plausible, and
is not really in conflict with Piaget's views), conceptualizations end up being non-verbal.
36. An argument such as that of Shanon (1987), that if conceptual representation would be
more abstract than the representation in terms of the medium one would lose much of
the meaningfulness of the medium itself, does not hold water. Taken to its extremes, in
case of language this would imply that people store surface structures, which is obviously
a highly implausible claim. It is true that aspects of the medium can be meaningful (e.g.
the choice of a language in bilingual contexts - cf. 1.2). However, that does not mean that
the representation in terms of the medium is stored, but only that the 'meaning' of this
aspect of the code is stored next to the direct message contained in the utterance.
Obviously, much of the medium is totally irrelevant (e.g. context-sensitive allophonic
variation) and is therefore never stored. Shanon's view leads to the position that
knowledge is a direct copy of the extra-subjective reality, but as argued in 1.2.3.1 this
claim is implausible.
37. Dik apparently does not intend his images to be visual - this is clear from his reference
to, e.g., Johnson-Laird's (1983) mental models as a possible realization for them. Actually,
Johnson-Laird's general theory corresponds to the one presented here rather than to
Dik's: in his view language understanding is a matter of forming a mental model indeed;
and reasoning is done in terms of these abstract conceptualizations, and not in terms of
some propositional logic (as Dik's FL would be).
38. Surveys of representation systems can be found in Frijda (1972) and Winograd (1977b).
A general critique of models of conceptual representation in Johnson-Laird (1983: 205ff).
39. The suggestion to draw a distinction between illocution and qualifications obviously does
not imply that there is no interaction between these categories, both conceptually (e.g.
a modal qualification of a piece of conceptual information as 'uncertain' can lead to a
request for clarification) and linguistically. See also 3.2.3 on the role of illocution. Note
also that even the marking of the illocution of an utterance is determined by the different
functional factors discussed in 1.2. This again underscores the strongly leveled and
hierarchical nature of human thinking.
40. Also many of the phenomena which are currently being ascribed to the illocutionary layer
in FG do not seem to have anything to do with illocutionary matters. Dik's (1990)
'illocutionary conditionals' (e.g. If you're hungry, there's some sandwiches in the fridge)
certainly do constitute a clearly different type as compared to his 'propositional condi
tionals' (e.g. If John stays, I will leave). Yet, the difference is that in the latter the fact
stated in the protasis specifies a condition for the occurrence of the fact stated in the
apodosis, while in the former the fact stated in the protasis is a condition for the
relevance of the fact stated in the apodosis. There is not a relationship of relevance
between a fact and a speech act, but between a fact and a fact. Similar considerations can
be made for many of the 'illocutionary satellites' specified by Dik et al. (1990).
41. Of course, this means that many of the observations made by Dik et al. (1990) in favor
of their layering system will have to be reconsidered in the present proposal. I will keep
this for another occasion. Still, for one thing, it is no problem to suppose that in the
hierarchy of qualifications there are 'break off points for the possibility to invoke certain
types of processes in the grammar. For another thing, there are potential problems with
some of the criteria used by Dik et al. Consider the argumentation on the basis of
negation. The difference in negation of utterances with predicate satellites and
predication satellites would be that in the former case only the relation between the
nucleus and the predicate satellite would be negated, while in the latter case the
predication can be negated. Yet consider an utterance such as John didn't buy cookies for
his aunt, which contains a beneficiary, i.e. a predicate satellite in the FG view. This can
be read in different ways, depending on how one pronounces it. In one of the 'natural'
readings the nuclear predication is negated, as with predication satellites. Also consider
the argument that 'representational satellites' form an informational unit with the core
predication while 'interpersonal satellites' do not, on the basis of the observation that the
former can be questioned with a yes/no question together with the core predication while
the latter cannot. An explanation might be that the 'interpersonal satellites' for which this
is illustrated all involve an element of uncertainty about whether the SoA applies (or is
relevant), while a question is precisely the result of uncertainty about whether the SoA
applies (or is relevant). The question aims to resolve the uncertainty. Hence, questions
and modal, evidential, conditional, etc. satellites are simply incompatible, independently
of how they are hierarchically organized.
42. Note that there is some similarity between this definition of the motivation for the
hierarchical order of qualifications and Hengeveld's definition of it (the lower layers have
more to do with the SoA itself, the higher layers have more to do with the relationship
between the individual and the SoA). The difference is that the present definition does
not rely on the notion 'speech event'.
43. There is only a rudimentary sketch of what a parser should be able to do in Dik (1987d),
but this has no impact whatsoever on the way FG itself is conceived.
44. Note that this suggestion leads to the introduction of an element of probability in
language processing. Cf. Levelt (1973c), Schlesinger (1977).
45. Note that this riddle would not work if the survivors would be replaced by those who
continue to live. That is due to the fact that the latter formulation would be very focal in
the given utterance, and therefore would attract the attention of the H/S.
46. Maybe the heated discussion in language psychology concerning the question how an H/S
handles ambiguity (see Kess and Hoppe 1981, Flores d'Arcais and Schreuder 1983) is
somewhat pointless, in the sense that maybe more than one procedure is possible, the
choice being due to the circumstances. That would at least explain the often rather
contradictory experimental results.
47. Actually, it is tempting to speculate on the relationship of the issue of grammaticalization

in linguistic structure to the issue of the control structure for processing: probably
something may be considered grammaticalized if there exists an automatic procedure
based entirely on demons for processing it.
48. Note that this would be a case where also in production procedures which would
normally have to be applied are not applied. Still, the difference remains that in
interpretation this is a normal thing to do, whereas in production it is exceptional.
49. Of course, looking for such correlations may again affect the hypotheses about each
mode: the simplicity criterion demands that if it is possible to bring the two modes closer
together by having them share knowledge components, one should prefer such a solution
over one in which the two modes are entirely separated.
50. Osgood (1980, 1984) presents the verbalization process as a matter of decoding, and
language understanding as a matter of encoding. This is illogical, however: it is natural
to consider a lexical pattern as a symbol representing a meaning, and not the other way
around.
51. The following is basically a further development of the (sketchy) proposal in De Schutter
and Nuyts (1983), although there are also a number of deviations from the latter. Note
that the notion 'grammar' is used here in a very wide sense, not only covering the
linguistic systems, but all systems somehow involved in the generation process. There are
some interesting parallels between the present proposal and the one by Hovy and Schank
(1984), but I will not be able to explore this here.
52. Obviously, in line with the argumentation in 1.1 for an integrative approach to modeling
language behavior it would be desirable to also include a model of reasoning in the
framework: it is necessary anyway for fully understanding knowledge representation and
organization as discussed in 3.2.1; and in view of the intricate process of decision making
in planning an action noted in 1.2 we cannot even fully understand language behavior
without it. Yet, limitations do not allow me to go any further into it here.
53. Though one does not have to agree with the overstated conclusions which he links to his
considerations, that the 'grammatical paradigm' should be abandoned altogether.
54. This shows the relevance of theoretically distinguishing between an encyclopedia (i.e. the
entire stock of knowledge possessed by a language user) and the UI (i.e. the part of the
encyclopedia which is activated in a specific situation).
55. The traditional category of modality appears not to be a uniform phenomenon. On types
of modality and their relationship, cf. e.g. Lyons (1977: 787ff), De Schutter (1983), Palmer
(1986), Hengeveld (1987, 1988).
56. This corresponds to the general picture of the syntactic stiffness of embedded utterances
in our languages. Cf. e.g. Givn (1979) for further examples concerning the expression
of informational value in the embedded clause.
57. In De Schutter and Nuyts (1983) we have simply called this 'SoA'. This term is too often
used in a more general sense (as I have done above) to warrant unequivocality, however.
58. NR in Dutch and English is very similar, and for a change I will intermingle Dutch and
English examples. I will not provide glosses for the Dutch examples since their structure
is nearly identical to the English cases, and the differences are sufficiently transparent to
be clear to the reader. Only when there are differences in the NR-behavior between
Dutch and English will I explicitly compare the two languages.
59. With an eye to this problem, I have checked my own intuitions on the relationship
between utterances with a number of test persons. Though I rarely obtained full
unanimity among the judges, the intuitions presented in the analyses below always
correspond to the opinion of the majority. Of course, given the informal nature of the
test, this is only suggestive for the adequacy of the intuitions presented.
60. I will not present in detail all aspects of the NR-problem as noted so far, partly because
this would mean only trying to repeat the excellent, very exhaustive and still accurate
survey in Horn (1978a), partly because my own discussion will mainly focus on the core
aspects of the phenomenon. The problem is so complex and many-sided that an
exhaustive account would take a separate study of some length, and except for few
remarks in 3.3.3 I will leave the discussion of all kinds of epiphenomena (some of which
will briefly be mentioned below) for another occasion.
61. The fact that I keep using the term 'negative raising' is only bound by tradition, and has
no correlation whatsoever with the type of account I would propose. On the contrary, see
3.3.3.
62. Except for some remarks in 3.3.3 - at which point some specifics about these phenomena
will be provided - I will not go into these problems here either.
63. According to Horn, NR occurs both with epistemic and deontic predicates. Thus, it is
desirable that he doesn 't come and it isn 't desirable that he comes seems to be a genuine
NR-pair. Yet, in order not to complicate the discussion too much, I will limit myself here
to epistemic cases.
64. Though the predication is only a part of the basic clause structure in current FG (cf. (6)
in 3.1.1), I will use the term for this entire underlying structure, for ease of reference.
65. Note that English has more modal adjectives than Dutch. The distinction between
probable and likely cannot be rendered in Dutch.
66. This context obviously makes it difficult to maintain the terms 'main-negative' and
'embedded-negative' utterance. Yet, since in the following I will mainly use examples of
the simpler, 'traditional' type, I will keep using this terminology for the sake of
convenience.
67. It is more difficult to find evidence for this claim in English, since it is much harder to
separate the negative marker and the predicate, and since word order in general is much
less flexible in English than in Dutch.
68. This is only a rough classification of the expressions. Even those expressions located at
approximately the same place on the scale show subtle differences in strength of the PA
expressed, among others due to the explicitness of a negative marker along the lines
indicated above.
69. For the sake of simplicity I will only discuss cases with a negative PA here. Of course, the
reasoning is parallel for cases with a positive PA.
70. In either case, Dutch also provides the option of introducing modal particles, which
allows for a much more nuanced expression of the qualification than is possible in
English.
71. This analysis provides a solution to one of the long-standing problems with the syntactic
behavior of these NR-constructions, viz. the observation that without the complementizer
the NR-reading is possible, but with the complementizer it is not possible. This has
nothing to do with the complementizer as such, but with the semantics of the verb. See
3.3.3.3.
72. I will not further use examples with mental state M-predicates such as (84b-c), because
they not only involve the problem of the structural ambiguity of the former and the
relationship to the latter, but also the problem of the de re and de dicto readings of the
M-predicate. In the non-PA-reading of (84b) and in (84c) it seems one can only get the
de dicto reading, and as argued above this implies that the predicate does not really
express modality anymore. Hence, (84a) and (84c) are not really synonymous, contrary
to what has been assumed in the NR-literature.
73. One could even argue that the PA-constructions are generally a type of cleft construction,
allowing the entire PA or the modality component in it to be put in focus.
74. Actually, completely equivalent observations can be made for Horn's extreme-scalar
predicates, such as know, which only confirms the appropriateness of the restatement of
the problem in 3.3.2.
Conclusion
This study has been mainly theoretical, but since current language research
shows a definite inclination toward data-oriented investigations (one might even
say: toward inductivism), many will feel uneasy and unhappy with it. This study
has made a virtue of speculation about many aspects of language, raising
questions rather than answering them, while in the human sciences speculation
is generally considered a dubious way of doing research. This study has taken an
interdisciplinary perspective on the topic, but the faith in such an attempt is that
nobody is satisfied because the results hardly ever lives up to the expectations
within each separate subdiscipline.
Still, it has been written because there are reasons to believe that these
properties are crucial for the healthy development of research into human
cognition in general, and language behavior in particular. They should, of course,
not replace a careful and in-depth data analysis guided by disciplinary conven
tions, but they are a necessary complement to such research. The language
sciences in general have witnessed an explosive growth of empirical investiga
tions in the last decade, but this has obviously been at the cost of theoretical and
reflective research. This study has also been written because it seems linguistics
in particular has come to important crossroads in its development and is in a
transitory phase, which may mark the final break with the 'generativist' era.
Defining problems and raising questions is always as important to a science as
answering questions, but in periods of change it is doubly important.
Still, some of the discontent many will no doubt feel with this book may be
justified. As long as theoretical speculation is not substantiated by serious
evidence, it remains vague and slippery. I have obviously not been doing much
in the present inquiry to meet that problem, except for citing massive amounts
of empirical investigations which pertain to individual issues. Also, while one may
attempt to cover several disciplines, one is not usually equally well acquainted
with each of them. My linguistic background has no doubt been pervasive also
in my concern with other disciplines. I cannot but humbly reply that there is a
limit to what an individual can accomplish: only within the context of the
collective research community can or does an individual contribution gain
validity. Obviously, much research is required to make something out of the
many loose strings dangling down from the foregoing pages. What is done here
is the result of my own judgment of how I could contribute best to the
development of a collective effort. Such a judgment is obviously relative, and

subjective, however.
References
Abelson, R.
1969 "Depth grammar and necessary truth". In S.Hook (ed.), 215-220.
Abraham, W. (ed.)
1975 Ut videam. Lisse: De Ridder.
Allen, R. and M.Shatz.

1983 "'What says meow?*". Journal of Child Language 10: 321-335.
Allerton, D.J., E.Carney and D.Holdcroft (eds.)

1979 Function and context in linguistic analysis. Cambridge: CUP.
Altman, J.
1978 "Three levels of mentation and the hierarchic organization of the human brain". In
G.Miller and E.Lenneberg (eds.), 87-109.
Arbib, M.A.
1982 "From artificial intelligence to neurolinguistics". In M.Arbib et al. (eds.), 77-94.
Arbib, M.A, D.Caplan and J.C.Marshall (eds.)

1982 Neural models of language processes. New York: Academic Press.
Arbini, R.
1973 "On explanations of linguistic competence". Philosophi 3: 59-83.
Argyle, M.
1973 Social interaction. London: Travistock.
Argyle, M. and P.Trower

1979 Person to person. London: Harper.
Aronson, L.R., E.Tobach, D.S.Lehrman and J.S.Rosenblatt (eds.)

1970 Development and evolution of behavior. San Francisco: Freeman.
Atkinson, J.M. and J.C.Heritage (eds.)

1984 Structures of social action. Cambridge: CUP.
Austin, J.L.
1962 How to do things with words. Oxford: OUP. (2nd edition, 1975)
Auwera, J.van der and L.Goossens (eds.)

1987 Ins and outs of the predication. Dordrecht: Foris.
Ayala, F.
1970 "Teleological explanations in evolutionary biology". Philosophy of Science 37: 1-15.
Bach, E. and R.T.Harms (eds.)

1968 Universals in linguistic theory. London: Holt.
Bach, K. and R.M.Harnish

1979 Linguistic communication and speech acts. Cambridge: MIT-Press.
Baker, C. and J.McCarthy (eds.)

1981 The logical problem of language acquisition. Cambridge: MIT-Press.
Ballmer, T.T.
1979 "Probleme der Klassifikation von Sprechakten". In G.Grewendorf (ed.), 247-274.
Ballmer, T.T. and W.Brennenstuhl

1981 Speech act classification. Berlin: Springer.
Bar-Hillel, Y. (ed.)
1971 Pragmatics of natural languages. Dordrecht: Reidel.
Bara, B.G. and G.Guida (eds.)

1984 Computational models of natural language processing. Amsterdam: North Holland.
Bartsch, R.
1973 "'Negative Transportation' gibt es nicht". Linguistische Berichte 27: 1-7.
Basso, K.
1972 "To give up on words': Silence in Western Apache culture". In P.Giglioni (ed.), 67-
86.
Bates, E.
1978 "Functionalism and the biology of language". Papers and Reports on Child Language
Development 15: K1-K26.
1979 The emergence of symbols. New York: Academic Press.
Bates, E. and B.MacWhinney

1979 "A functionalist approach to the acquisition of grammar". In E.Ochs and
B.Schieffelin (eds.), 167-211.
1982 "Functionalist approaches to grammar". In E.Wanner and L.Gleitman (eds.), 173-
218.
Bates, E., M.Masling and W.Kintsch

1978 "Recognition memory for aspects of dialogue". Journal of Experimental Psychology:
Human Learning and Memory 4/3: 187-197.
Bates, E., L.Benigni, I.Bretherton, L.Camaioni and V.Volterra

1977 "From gesture to the first word". In M.Lewis and L.Rosenblum (eds.), 247-307.
REFERENCES 337
Buerle, R., U.Egli and A. von Stechow (eds.)

1979 Semantics from different points of view. Berlin: Springer.
Beattie, G.W.
1980 "The role of language production processes in the organization of behaviour in face-
to-face interaction". In B.Butterworth (ed.), 69-107.
Beaugrande, R.-A.de
1980 "The pragmatics of discourse planning". Journal of Pragmatics 4: 15-42.
1984 Text production. Norwood: Ablex.
Bense, E.
1973 Mentalismus in der Sprachtheorie Noam Chomskys. Kronberg: Scriptor.
Bernstein, B.
1971 Class, codes and control, vol.1. London: Routledge.
Berwick, R. and A.Weinberg

1983 "The role of grammars in models of language use". Cognition 13: 1-61.
1984 The grammatical basis of linguistic performance. Cambridge: MIT-Press.
Beukema, F. and P.Coopmans (eds.)

1987 Linguistics in the Netherlands 1987. Dordrecht: Foris.
Bever, T.G.
1970a "The cognitive basis for linguistic structures". In J.Hayes (ed.), 279-362.
1970b "The influence of speech performance on linguistic structure". In G.Flores d'Arcais
and W.Levelt (eds.), 4-30.
1971 "The integrated study of language behaviour". In J.Morton (ed.), 158-209.
1974 "The ascent of the specious, or: There's a lot we don't know about mirrors". In
D.Cohen (ed.), 173-200.
1975 "Psychologically real grammar emerges because of its role in language acquisition".
In D.Dato (ed.), 63-75.
1976 "The influence of speech performance on linguistic structure". In T.Bever et al.
(eds.), 65-88.
1982 "Some implications of the nonspecific bases of language". In E.Wanner and
L.Gleitman (eds.), 429-449.
Bever, T.G., J.M.Carroll and L.A.Miller (eds.)

1984 Talking minds. Cambridge: MIT-Press.
Bever, T.G., J.J.Katz and D.T.Langendoen (eds.)

1976 An integrated theory of linguistic ability. Hassocks: Harvester.
Bezooijen, R.Van
1976 Piaget's ideen over de relatie tussen taal en denken. Nijmegen: Grammarij 1.
Bierwisch, M.
1983 "How on-line is language processing?" In G.Flores d'Arcais and RJarvella (eds.),
113-168.
Bierwisch, M. and K.E.Heidolph (eds.)

1970 Progress in linguistics. The Hague: Mouton.
Black, M.
1970 "Comment on Chomsky 1970". In R.Borger and F.Cioffi (eds.), 452-461.
Black, M. and S.Chiat

1981 "Psycholinguistics without 'psychological reality'". Linguistics 19: 37-61.
Block, N. (ed.)
1980 Readings in philosophy of psychology, vol.1. London: Methuen.
1981 Readings in philosophy of psychology, vol.2. London: Methuen.
Bobrow, D.G. and ACollins (eds.)

1975 Representation and understanding. New York: Academic Press.
Bock, J.K.
1977 "The effect of pragmatic presupposition on syntactic structure in question
answering". Journal of Verbal Learning and Verbal Behavior 16: 723-734.
1987 "Exploring levels of processing in sentence production". In G.Kempen (ed.), 351-363.
Bock, M. and J.Engelkamp

1978 "Textstrukturen aus sprachpsychologischer Sicht". Folia Linguistica 12: 301-318.
Bois, J.W.Du
1987 "Meaning without intention". Papers in Pragmatics 1/2: 80-122.
Bolinger, D.
1977 Meaning and form. London: Longman.
1978 "Intonation across languages". In J.Greenberg (ed.), vol.2, 471-524.
1983 "Where does intonation belong?" Journal of Semantics 2: 101-120.
Bolkestein, A.M.
1980 "The 'Ab urbe condita'-construction in Latin". In S.Daalder and M.Gerritsen (eds.),
80-92.
1981a "Embedded predications, displacement and pseudo-argument formation in Latin".
In A.M.Bolkestein et al., 63-112.
1981b "Factivity as a condition for an optional expression rule in Latin". In AM.Bolkestein
et al., 205-233.
1983 "The role of discourse in syntax". In K.Ehlich and H.Van Riemsdijk (eds.), 111-140.
1985 "Cohesiveness and syntactic variation". In AM.Bolkestein et al. (eds. 1985a), 1-14.
1987 "Discourse functions of predications". In J.Nuyts and G.De Schutter (eds.), 163-178.
Bolkestein, A.M. and R.Risselada

1987 "The pragmatic motivation of syntactic and semantic perspective". In J. Verschueren
and M.Bertuccelli-Papi (eds.), 497-512.
REFERENCES 339
Bolkestein, A.M., C.De Groot and J.L.Mackenzie (eds.)

1985a Syntax and pragmatics in functional grammar. Dordrecht: Foris.
1985b Predicates and terms in functional grammar. Dordrecht: Foris.
Bolkestein, A.M., H.A.Comb, S.C.Dik, C.De Groot, J.Gvozdanovic, ARijksbaron and C.Vet
1981 Predication and expression in functional grammar. London: Academic Press.
Borger, R. and F.Cioffi (eds.)

1970 Explanation in the behavioral sciences. Cambridge: CUP.
Bossuyt, A.
1982 Aspekten van de geschiedenis van de negatieve zin in het Nederlands. PhD-thesis, Vrije
Univ. Brussel.
Bouma, S.
1987 "The discourse function of quantifier float". In F.Beukema and P.Coopmans (eds.),
31-40.
Bower, G.H. (ed.)

1975 The psychology of learning and motivation, vol.9. New York: Academic Press.
Bransford, J.D. and J.J.Franks

1972 "The abstraction of linguistic ideas". Cognition 1: 211-249.
Bransford, J.D., R.J.Barclay and JJ.Franks

1972 "Sentence memory". Cognitive Psychology 3: 193-209.
Bredart, S.
1987 "Two ways of coping with indeterminate spatial discourse". Psychologica Belgica 27:
143-151.
Brewer, W.F. and A.E.Hay

1984 "Reconstructive recall of linguistic style". Journal of Verbal Learning and Verbal
Behavior 23: 237-249.
Brown, E.K.
1980 "Grammatical incoherence". In H.Dechert and M.Raupach (eds.), 27-37.
Brown, P. and S.C.Levinson

1979 "Social structure, groups and interaction". In K.Scherer and H.Giles (eds.), 291-335.
Brugmann, K.
1918 Verschiedenheiten der Satzgestaltung nach Massgabe der seelischen Grundfunktionen
in den indogermanischen Sprachen. Leipzig: Teubner.
Bruin, J.de and B.Wielinga

1982 "Flexibiliteit in natuurlijke taalverwerking". TTT 2/1: 16-27.
Bruner, J.S.
1974 "From communication to language". Cognition 3/3: 255-287.
1978 "On prelinguistic prerequisites of speech". In R.Campbell and P.Smith (eds.), vol.1,
199-214.
1985 "The role of interaction formats in language acquisition". In J.Forgas (ed.), 31-46.
Bhler, K.
1918 "Kritische Musterung der neueren Theorien des Satzes". Indogermanistisches
Jahrbuch 6: 1-20.
1934 Sprachtheorie. Stuttgart: Fischer. (Reprint 1982)
Busse, W.
1975 "Funktionen und Funktion der Sprache". In B.Schlieben-Lange (ed.), 207-240.
Butterworth, B.
1983 "Lexical representation". In B.Butterworth (ed.), 257-294.
Butterworth, B. (ed.)
1980 Language production, voll. London: Academic Press.
1983 Language production,vol.2. London: Academic Press.
Butterworth, B. and G.Beattie

1978 "Gesture and silence as indicators of planning in speech". In R.Campbell and
P.Smith (eds.), vol.2, 347-360.
Buuren, L.Van
1985 "Functional grammar and intonation". In AM.Bolkestein et al. (eds. 1985a), 31-47.
Cairns, H.S. and C.C.Cairns

1976 Psycholinguistics. New York: Holt.
Campbell, B.
1971 "The roots of language". In J.Morton (ed.), 10-23.
Campbell, R.N. and P.T.Smith (eds.)

1978 Recent advances in the psychology of language. New York: Plenum.
Campbell, R.N. and R.Wales

1970 "The study of language acquisition". In J.Lyons (ed.), 242-260.
Caplan, D. (ed.)
1980 Biological studies of mental processes. Cambridge: MIT-Press.
Caplan, D. and N.Chomsky

1980 "Linguistic perspectives on language development". In D.Caplan (ed.), 97-105.
Caplan, D., AR.Lecours and ASmith (eds.)

1984 Biological perspectives on language. Cambridge: MIT-Press.
Carnap, R.
1942 Introduction to semantics. Cambridge: Harvard UP.
REFERENCES 341
Catlin, J.
1978 "Discussion of the chapters by Stolzenberg and Chomsky". In G.Miller and
E.Lenneberg (eds.), 271-280.
Cattell, R.
1973 "Negative transportation and tag questions". Language 49: 612-639.
Cecco, J.P.De (ed.)

1969 The psychology of language, thought and instruction. London: Holt.
Cellrier, G.
1980a "Cognitive strategies in problem solving". In M.Piattelli-Palmarini (ed.), 67-72.
1980b "Some clarifications on innatism and constructivism". In M.Piattelli-Palmarini (ed.),
83-87.
Chafe, W.L.
1970 Meaning and the structure of language. Chicago: Univ. of Chicago Press.
1974 "Dialogue with H.Parret". In H.Parret (mod.), 1-25.
1976 "Givenness, contrastiveness, definiteness, subjects, topics, and point of view". In C.Li
(ed.), 25-55.
1977a "Creativity in verbalization and its implications for the nature of stored knowledge".
In R.Freedle (ed.), 41-55.
1977b "The recall and verbalization of past experience". In R.Cole (ed.), 215-246.
Chafe, W.L. (ed.)

1980 The pear stories. Norwood: Ablex.
Chafe, W.L. and J.Nichols (eds.)

1986 Evidentiality. Norwood: Ablex.
Changeux, J.-P.
1980 "Genetic determinism and epigenesis of the neuronal network". In M.Piattelli-
Palmarini (ed.), 185-197.
Charniak, E.
1984 "Cognitive science is methodologically fine". In W.Kintsch et al. (eds.), 263-274.
Chomsky, N.
1957 Syntactic structures. The Hague: Mouton.
1959 "Review of Skinner 1957". Language 35: 26-58.
1962 "Explanatory models in linguistics". In E.Nagel et al. (eds.), 528-550.
1963 "Formal properties of grammars". In R.Luce et al. (eds.), 323-418.
1964a "The logical basis of linguistic theory". Proc. Int. Congress of Linguists 9: 914-1008.
1964b Current issues in linguistic theory. The Hague: Mouton.
1965 Aspects of the theory of syntax. Cambridge: MIT-Press.
1966 Cartesian linguistics. New York: Harper and Row.
1968 Language and mind. New York: Harcourt.
1969a "Linguistics and philosophy". In S.Hook (ed.), 51-94.
1969b "Comments on Harman's reply". In S.Hook (ed.), 152-159.
1969c "Linguistics and politics". New Left Review 57: 21-34.

1970 "Problems of explanation in linguistics". In R.Borger and F.Cioffi (eds.), 425-451.
1971 Problems of knowledge and freedom. New York: Harcourt.
1972 Language and mind. Enlarged edition. New York: Harcourt.
1975a Reflections on language. New York: Pantheon.
1975b The logical structure of linguistic theory. New York: Plenum.
1975c "Knowledge of language". In K.Gunderson (ed.), 299-320.
1976 "On the nature of language". In S.Harnad et al. (eds.), 46-57.
1978 "On the biological basis of language capacities". In G.Miller and E.Lenneberg (eds.),
199-220.
1979 Language and responsibility. Sussex: Harvester.
1980a Rules and representations. New York: Columbia UP.
1980b "On cognitive structures and their development". In M.Piattelli-Palmarini (ed.), 35-
52.
1980c "The linguistic approach". In M.Piattelli-Palmarini (ed.), 109-117.
1980d "Discussion of Putnam's comments". In M.Piattelli-Palmarini (ed.), 310-324.
1981a Lectures on government and binding. Dordrecht: Foris. (2nd revised edition 1982)
1981b "Knowledge of language". In H.C.Longuet-Higgins et al. (eds.), 9-20.
1982 The generative enterprise. Dordrecht: Foris.
1986a "Changing perspectives on knowledge and use of language". Leuvense Bijdragen 75:
1-71.
1986b Knowledge of language. New York: Praeger.
Chomsky, N. and J.J.Katz

1974 "What the linguist is talking about". Journal of Philosophy 71: 347-367.
1975 "On innateness". Philosophical Review 84: 70-87.
Chomsky, N. and H.Lasnik

1977 "Filters and control". Linguistic Inquiry 8: 425-504.
Chomsky, N. and G.A.Miller

1963 "Introduction to the formal analysis of natural languages". In R.Luce et al. (eds.),
269-321.
Clark, E.
1983 "Meanings and concepts". In J.Flavell and E.Markman (eds.), 787-840.
Clark, H.H.
1977 "Bridging". In P.Johnson-Laird and P.Wason (eds.), 411-420.
1987 "Four dimensions of language use". In J. Verschueren and M.Bertuccelli-Papi (eds.),
9-25.
Clark, H.H. and T.B.Carlson

1982 "Speech acts and hearers' beliefs". In N.Smith (ed.), 1-59.
Clark, H.H. and E.V.Clark

1977 Psychology and language. New York: Harcourt.
REFERENCES 343
Clark, H.H. and S.E.Haviland

1974 "Psychological processes as linguistic explanation". In D.Cohen (ed.), 91-124.
1977 "Comprehension and the given-new contract". In R.Freedle (ed.), 1-40.
Clark, H.H and C.R.Marshall

1981 "Definite reference and mutual knowledge". In A.Joshi et al. (eds.), 10-63.
Clyne, M.G.
1972 Perspectives on language contact. Melbourne: Hawthorn.
Cohen, D. (ed.)
1974 Explaining linguistic phenomena. Washington: Hemisphere.
Cohen, D. and J.R.Wirth (eds.)

1975 Testing linguistic hypotheses. Washington: Hemisphere.
Cohen, L.J.
1981 "Some remarks on the nature of linguistic theory". In H.C.Longuet-Higgins et al.
(eds.), 21-29.
Cole, P. (ed.)
1978 Syntax and semantics 9: Pragmatics. New York: Academic Press.
1981 Radical pragmatics. New York: Academic Press.
Cole, P. and J.Jake

1978 "Accusative subjects in Imbabura Quechua". Studies in the Linguistic Sciences 8: 72-
96.
Cole, P. and J.L.Morgan (eds.)

1975 Syntax and semantics, vol.3. New York: Academic Press.
Cole, P., W.Harbert, G.Hermon and S.N.Sridhar

1980 "On the acquisition of subjecthood". Language 56: 719-743.
Cole, R.W. (ed.)

1977 Current issues in linguistic theory. Bloomington: Indiana UP.
Collier, R.
1983 "Syntactische correspondentie in een vraag-antwoord-context". In F.Daems and
L.Goossens (eds.), 29-42.
Comrie, B.
1976 Aspect. Cambridge: CUP.
1979 "Review of Dik 1978a". Studies in Language 3: 267-276.
1980 "Comments on Simon C.Dik, Functional Grammar". In M.Kac (ed.), 29-33.
Connolly, J. and S.C.Dik (eds.)

1989 Functional grammar and the computer. Dordrecht: Foris.
Cooper, D.E.
1972 "Innateness". Philosophical Review 81: 465-483.
1975 Knowledge of language. London: Prism.
Cooper, W.E. and E.C.T.Walker (eds.)

1979 Sentence processing. Hillsdale: Erlbaum.
Cooper, W.E. and E.B.Zurif

1983 "Aphasia: Information-processing in language production and reception". In
B.Butterworth (ed.), 225-256.
Copi, LM.
1978 Introduction to logic. 5th edition. London: Collier.
Coulmas, F. (ed.)
1981 A festschrift for native speaker. The Hague: Mouton.
Creider, C.A.
1979 "On the explanation of transformations". In T.Givn (ed.), 3-21.
Cummins, R.
1975 "Functional analysis". Journal of Philosophy 72: 741-764.
Curtiss, S.
1977 Genie. New York: Academic Press.
Cuyckens, H.
1985 "Een semantische beschrijving van plaatsaanduidende voorzetsels". In M.Dominicy
and H.Beatens Beardsmore (eds.), 18-44.
1988 "Spatial prepositions in cognitive semantics". In W.Hllen and R.Schulze (eds.), 316-
328.
Daalder, S. and M.Gerritsen (eds.)

1980 Linguistics in the Netherlands 1980. Amsterdam: North Holland.
Daems, F. and L.Goossens (eds.)

1983 Een spyeghel voor G.Jo Steenbergen. Leuven: Acco.
Dale, R.
1988 Generating referring expressions in a domain of objects and processes. PhD-thesis,
Univ. of Edinburgh.
Danes, F.
1964 "A three-level approach to syntax". Travaux Linguistiques de Prague 1: 225-240.
1987 "On Prague School functionalism in linguistics". In R.Dirven and V.Fried (eds.), 3-
38.
Danes', F. (ed.)
1974 Papers on functional sentence perspective. The Hague: Mouton.
REFERENCES 345
Danto, A.
1969 "Semantical vehicles, understanding, and innate ideas". In S.Hook (ed.), 122-137.
Darwin, C.
1872 The expression of the emotions in man and animals. London: Murray.
Dato, D.P. (ed.)

1975 Developmental psycholinguistics. Washington: Georgetown UP.
Davey, A.C.
1978 Discourse production. Edinburgh: Edinburgh UP.
Davies, W.D.
1981 "Choctaw subjects and multiple levels of syntax". In T.Hoekstra et al. (eds.), 235-271.
Dechert, H.W. and M.Raupach (eds.)

1980 Temporal variables in speech. The Hague: Mouton.
1987 Psycholinguistic models of production. Norwood: Ablex.
Dempe, H.
1930 Was ist Sprache? Weimar: Bhlau.
1935 "Die Darstellungstheorie der Sprache". Indogermanische Forschungen 53: 245-266.
Derwing, B.L.
1973 Transformational grammar as a theory of language acquisition. Cambridge: CUP.
Derwing, B.L. and W.J.Baker

1978 "On the re-integration of linguistics and psychology". In R.Campbell and P.Smith
(eds.), vol.2, 193-218.
Dietrich, R. and C.F.Graumann (eds.)

1989 Language processing in social contexts. Amsterdam: North Holland.
Dijk, T.A.Van
1978 Tekstwetenschap. Utrecht: Het Spectrum.
1981 "Subjektieve interpretatie". TTT 1/1: 56-72.
Dijk, T.A.Van and W.Kintsch

1983 Strategies of discourse comprehension. New York: Academic Press.
Dik, S.C.
1977 "Wat is pragmatiek?" In B.Tervoort (ed.), 206-221.
1978a Functional grammar. Amsterdam: North Holland.
1978b Stepwise lexical decomposition. Lisse: De Ridder.
1979a "Raising in a functional grammar". Lingua 47: 119-140.
1979b "Funktionele morfologie". In T.Hoekstra and H.Van der Hulst (eds.), 73-100.
1979c "Over funktionele grammatika". Forum der Letteren 20: 42-51.
1980a Studies in functional grammar. London: Academic Press.
1980b "Seventeen sentences". In E.Moravcsik and J.Wirth (eds.), 45-75.
1980c "Reply to Bernard Comrie's comments". In M.Kac (ed.), 35-40.
1981a "Predication and expression". In AM.Bolkestein et al., 1-17.

1981b "Embedded themes in spoken Dutch". In AM.Bolkestein et al., 113-124.
1981c "Over de behandeling van niet-produktieve regelmatigheden". Forum der Letteren 22:
39-50.
1982 "Taalbeschouwing en taaltheorie". Forum der Letteren 23/3: 184-200.
1983a "Basic principles of functional grammar". In S.Dik (ed.), 3-28.
1983b "Some basic principles of functional grammar". Proc. Int. Congress of Linguists 13:
74-88.
1986a "On the notion 'functional explanation'". Belgian Journal of Linguistics 1: 11-52.
1986b "Taalwetenschap 1986-2000". In H.Jolles et al. (eds.), 65-80.
1987a "A typology of entities". In J.Van der Auwera and L.Goossens (eds.), 1-20.
1987b "Functional grammar and its potential computer applications". In W.Meijs (ed.),
253-268.
1987c "Generating answers from a linguistically coded knowledge base". In G.Kempen
(ed.), 301-314.
1987d "Linguistically motivated knowledge representation". In M.Nagao (ed.), 145-170.
1988 "Concerning the logical component of a natural language generator". In M.Zock and
G.Sabah (eds.), 73-91.
1989a The theory of functional grammar, pt.l. Dordrecht: Foris.
1989b "Towards a unified cognitive language". In F.Heyvaert and F.Steurs (eds.), 97-110.
1990 "On the semantics of conditionals". In J.Nuyts et al. (eds.), 233-261.
fc. The theory of functional grammar, pt.2. Dordrecht: Foris.
Dik, S.C. (ed.)

1983 Advances in functional grammar. Dordrecht: Foris.
Dik, S.C. and J.G.Kooij

1979 Algemene taalwetenschap. Utrecht: Het Spectrum.
Dik, S.C., K.Hengeveld, E.Vester and C.Vet

1990 "The hierarchical structure of the clause and the typology of adverbials". In J.Nuyts
et al. (eds.), 25-70.
Dik, S.C., R.Gebruers, W.De Groot, S.Van de Kerke, R.Knaap, H.Nobach and J.Nuyts
1983 "The treatment of experiencer constructions in a functional grammar". Ms., Univ.
of Amsterdam.
Dingwall, W.O.
1975 "The species-specificity of speech". In D.Dato (ed.), 17-62.
1979 "The evolution of human communication systems". In H.Whitaker and H.Whitaker
(eds.), 1-95.
Dirven, R. and V.Fried (eds.)

1987 Functionalism in linguistics. Amsterdam: Benjamins.
Dobzhansky, T., F.J.Ayala, G.L.Stebbins and J.W.Valentine

1977 Evolution. San Francisco: Freeman.
REFERENCES 347
Dominicy, M. and H.Baetens Beardsmore (eds.)

1985 Linguistics in Belgium 7. Brussels Pre-Prints in Linguistics, Special Issue.
Dresher, B.E. and N.Hornstein

1976 "On some supposed contributions of artificial intelligence to the scientific study of
language". Cognition 4: 321-398.
Duranti, A.
1988 "Intentions, language, and social action in a Samoan context". Journal of Pragmatics
12: 13-33.
Ehlich, K. and J.Rehbein (eds.)

1983 Kommunikation in Schule und Hochschule. Tbingen: Narr.
Ehlich, K. and H.Van Riemsdijk (eds.)

1983 Connectedness in sentence, discourse and text. Tilburg: Tilburg Studies in Language
and Literature 4.
Epstein, S.
1977 Investigations in pragmatic theory. Bloomington: IULC.
Erteschik-Shir, N.
1979 "Discourse constraints on dative movement". In T.Givn (ed.), 441-467.
Fasold, R.W. and R.W.Shuy (eds.)

1977 Studies in language variation. Washington: Georgetown UP.
Fillmore, C.J.
1963 "The position of embedding transformations in a grammar". Word 19: 208-231.
1968 "The case for case". In E.Bach and R.Harms (eds.), 1-88.
1982 "Towards a descriptive framework for spatial deixis". In R.Jarvella and W.Klein
(eds.), 31-59.
1984 "Some thoughts on the boundaries and components of linguistics". In T.Bever et al.
(eds.), 73-108.
1988 "The mechanisms of 'construction grammar'". Berkeley Linguistic Society 14: 35-55.
1989 "Grammatical construction theory and the familiar dichotomies". In R.Dietrich and
C.Graumann (eds.), 17-38.
Fillmore, C.J., P.Kay and M.C.O'Connor

1988 "Regularity and idiomaticity in grammatical constructions: The case of 'let alone'".
Language 64: 501-538.
Fillmore, C.J., D.Kempler and W.S.Y.Wang (eds.)

1979 Individual differences in language ability and language behavior. New York: Academic
Press.
Firth, J.R.
1952 "Linguistic analysis as a study of meaning". In F.Palmer (ed. 1968), 12-26.
1955 "Structural linguistics". In F.Palmer (ed. 1968), 35-52.
Fisher, S. and AD.Todd (eds.)

1983 The social organization of doctor-patient communication. Washington: Center for
Applied Linguistics.
Fishman, J.A. (ed.)

1968 Readings in the sociology of language. The Hague: Mouton.
Flavell, J.H. and E.M.Markman (eds.)

1983 Cognitive development. New York: Wiley.
Flores d'Arcais, G.B. and R.J.Jarvella (eds.)

1983 The process of language understanding. Chichester: Wiley.
Flores d'Arcais, G.B. and W.J.M.Levelt (eds.)

1970 Advances in psycholinguistics. Amsterdam: North Holland.
Flores d'Arcais, G.B. and R.Schreuder

1983 "The process of language understanding". In G.Flores d'Arcais and R.Jarvella (eds.),
1-41.
Fodor, J.A.
1976 The language of thought. Hassocks: Harvester.
Fodor, J.A and M.F.Garrett

1966 "Some reflections on competence and performance". In J.Lyons and R.Wales (eds.),
135-179.
Fodor, J.A, T.G.Bever and M.F.Garrett

1974 The psychology of language. New York: McGraw-Hill.
Foley, W.A and R.D.Van Valin

1984 Functional syntax and universal grammar. Cambridge: CUP.
Forgas, J.P. (ed.)

1985 Language and social situations. New York: Springer.
Forster, K.I.
1979 "Levels of processing and the structure of the language processor". In W.Cooper and
E.Walker (eds.), 27-85.
Fortescue, M.
1980 A discourse production model f or 'twenty questions'. Amsterdam: Benjamins.
Franck, D.
1980 Grammatik und Konversation. Knigstein: Scriptor.
Fraser, C. and K.R.Scherer (eds.)

1982 Advances in the social psychology of language. Cambridge: CUP.
REFERENCES 349
Frazier, L.
1982 "Shared components of production and perception". In M.Arbib et al. (eds.), 225-
236.
Freedle, R.O. (ed.)

1977 Discourse production and comprehension. Norwood: Ablex.
Frijda, N.H.
1972 "Simulation of human long-term memory". Psychological Bulletin 11: 1-31.
Fromkin, V.A.
1975 "When does a test test a hypothesis, or: What counts as evidence?" In D.Cohen and
J.Wirth (eds.), 43-64.
Fromkin, V.A. (ed.)

1973 Speech errors as linguistic evidence. The Hague: Mouton.
1980 Errors in linguistic performance. New York: Academic Press.
Garrett, M.F.
1975 "The analysis of sentence production". In G.Bower (ed.), 133-177.
1980 "Levels of processing in sentence production". In B.Butterworth (ed.), 177-220.
1982a "A perspective on research in language production". In J.Mehler et al. (eds.), 185-
199.
1982b "Remarks on the relation between language production and language comprehen
sion systems". In M.Arbib et al. (eds.), 209-224.
Garrod, S.C and A.J.Sanford

1982 "The mental representation of discourse in a focussed memory system". Journal of
Semantics 1: 21-41.
Geest, T.Van der

1974 "Language acquisition as a hidden curriculum". Communication and Cognition 7/2:
169-190.
1975 Some aspects of communicative competence and their implications for language
acquisition. Assen: Van Gorcum.
Gelman, R. and M.Shatz

1977 "Appropriate speech adjustments". In M.Lewis and L.Rosenblum (eds.), 27-61.
George, S. (ed.)
1984 From the linguistic to the social context. Bologna: CLUEB.
Geysen, R. and J.Nuyts

1982 "Objektiviteit en interpretatie". Restant 10/2: 167-190.
Giglioni, P.P. (ed.)

1972 Language and social context. Harmondsworth: Penguin.
Gillis, S.
1983 Van pretalige naar talige interactie. Wilrijk: Antwerp Papers in Linguistics 30.
1984 De verwerving van talige referentie. PhD-thesis, Univ. of Antwerp.
Givn, T.
1979 On understanding grammar. New York: Academic Press.
Givn, T. (ed.)
1979 Syntax and semantics 12: Discourse and syntax. New York: Academic Press.
1983 Topic continuity in discourse. Amsterdam: Benjamins.
Gleitman, H. and L.Gleitman

1979 "Language use and language judgment". In C.Fillmore et al. (eds.), 103-126.
Gleitman, L.R., H.Gleitman and E.F.Shipley

1972 "The emergence of the child as grammarian". Cognition 1: 137-164.
Goldman, N.
1975 Conceptual generation. In R.Schank et al., 289-371.
Goodman, N.
1969 "The emperor's new ideas". In S.Hook (ed.), 138-142.
Goodwin, M.H.
1983 "Searching for a word as an interactive activity". Semiotics 1981: 129-138.
Goossens, L. (ed.)
1981 Bijdragen over semantiek van het 33ste Vlaams Filologencongres. Wilrijk: Antwerp
Papers in Linguistics 23.
Gould, S.J.
1977 Ever since Darwin. New York: Norton.
Grandy, R. and R.Warner (eds.)

1986 Philosophical grounds of rationality. Oxford: Claredon.
Granstrm, K. (ed.)
1983 Om kommunikation 2. Linkping: Studies in Communication 6.
Greenbaum, S. (ed.)
1977 Acceptability in language. The Hague: Mouton.
Greenberg,
J.H.
1958 Essays in linguistics. Chicago: Univ. of Chicago Press.
1959 "Language and evolution". In J.Greenberg 1971b, 106-125.
1963 "Language and linguistics". In J.Greenberg 1971b, 156-171.
1966 "Some universals of grammar with particular reference to the order of meaningful
elements". In J.Greenberg (ed.), 73-113.
1968 "The nature and definition of language". In J.Greenberg 1971b, 260-273.
1971a "Linguistics as a pilot science". In J.Greenberg 1971b, 314-329.
1971b Language, culture and communication. Edited by A.S.Dil. Stanford: Stanford UP.
REFERENCES 351
Greenberg, J.H. (ed.)

1966 Universals of language. Cambridge: MIT-Press.
1978 Universals of human language. Stanford: Stanford UP.
Greene, J.
1972 Psycholinguistics. Harmondsworth: Penguin.
Grewendorf, G. (ed.)
1979 Sprechakttheorie und Semantik. Frankfurt: Suhrkamp.
Grice, H.P.
1968 "Utterer's meaning, sentence-meaning and word-meaning". Foundations of Language
4: 225-242.
1969 "Utterer's meaning and intentions". Philosophical Review 78: 147-177.
1975 "Logic and conversation". In P.Cole and J.Morgan (eds.), 41-58.
1989 Studies in the way of words. Cambridge: Harvard UP.
Groot, C.De
1981a "On theme in functional grammar". In T.Hoekstra et al. (eds.), 75-88.
1981b "Sentence-intertwining in Hungarian". In A.M.Bolkestein et al., 41-62.
1987 Predicate formation in functional grammar. Amsterdam: Working Papers in
Functional Grammar 20.
Gumperz, J.J.
1982 Discourse strategies. Cambridge: CUP.
Gumperz, J.J. (ed.)

1982 Language and social identity. Cambridge: CUP.
Gumperz, J.J. and D.Hymes (eds.)

1972 Directions in sociolinguistics. New York: Holt.
Gunderson, K. (ed.)
1975 Language, mind and knowledge. Minneapolis: Univ. of Minnesota Press.
Gurney, R.
1973 Language, brain and interactive processes. London: Arnold.
Haiman, J. (ed.)
1985 Iconicity in syntax. Amsterdam: J.Benjamins.
Halle, M., J.Bresnan and G.AMiller (eds.)

1978 Linguistic theory and psychological reality. Cambridge: MIT-Press.
Halliday, M.A.K.
1970a "Functional diversity in language". Foundations of Language 6/3: 322-361.
1970b "Language structure and language function". In J.Lyons (ed.), 140-165.
1973 Explorations in the functions of language. London: Arnold.
1974a "The place of 'functional sentence perspective' in the system of linguistic descrip-
tion". In F.Danes' (ed.), 43-53.

1974b "Dialogue with H.Parret". In H.Parret (mod.), 81-120.
1976 System and function in language. London: Oxford UP.
1979 "Modes of meaning and modes of expression". In D.Allerton et al. (eds.), 57-79.
Halpern, R.N.
1976 "The bivalence of neg-raising predicates". Studies in the Linguistic Sciences 6: 69-81.
Hannay, M.
1983 "The focus function in functional grammar". In S.Dik (ed.), 207-223.
1985a "Inferrability, discourse-boundness, and sub-topics". In AM.Bolkestein et al. (eds.
1985a), 49-63.
1985b English existentials in functional grammar. Dordrecht: Foris.
Hannay, M. and E.Vester

1987 "Non-restrictive relatives and the representation of complex sentences". In J. Van der
Auwera and L.Goossens (eds.), 39-52.
Harley, T.A.
1984 "A critique of top-down independent levels models of speech production". Cognitive
Science 8: 191-219.
Harman, G.H.
1967 "Psychological aspects of the theory of syntax". Journal of Philosophy 64: 75-87.
1969 "Linguistic competence and empiricism". In S.Hook (ed.), 143-151.
1975 "Language, thought, and communication". In K.Gunderson (ed.), 270-298.
Harnad, S.R., H.D.Steklis and J.Lancaster (eds.)

1976 Origins and evolution of language and speech. New York: New York Academy of
Sciences.
Harris, A.
1982 "Georgian and the unaccusative hypothesis". Language 58: 290-306.
Harris, Z.S.
1952 "Discourse analysis". Language 28: 1-30.
1957 "Co-occurrence and transformation in linguistic structure". Language 33: 283-340.
Hawkins, J.A.
1978 Definiteness and indefiniteness. London: Croom Helm.
Hayes, J.R. (ed.)

1970 Cognition and the development of language. New York: Wiley.
Hayes-Roth, B.
1985 "A blackboard architecture for control". Artificial Intelligence 26: 251-321.
Heath, C.
1984 "Talk and recipiency". In J.Atkinson and J.Heritage (eds.), 247-265.
REFERENCES 353
Heller, M.S., J.P.Bartholomot, L.Levy and L.Ostiguy

1982 Le processus de francisation dans une entreprise montralaise. Quebec: L'editeur offi
ciel.
Hempel, C G .
1965 Aspects of scientific explanation. New York: Free Press.
Hengeveld, K.
1987 "Clause structure and modality in functional grammar". In J.Van der Auwera and
1988 "Illocution, mood and modality in a functional grammar of Spanish". Journal of
Semantics 6: 227-269.
1989 "Layers and operators in functional grammar". Journal of Linguistics 25: 127-157.
1990 "The hierarchical structure of utterances". In J.Nuyts et al. (eds.), 1-23.
Heringer, H.J.
1983 "Criteria of understanding and understandability". In G.Rickheit and M.Bock (eds.),
257-265.
Herrmann, T.
1983 Speech and situation. Berlin: Springer.
Hesp, C.
1990 "The functional grammar computational natural language user and psychological
adequacy". In J.Nuyts et al. (eds.), 295-312.
Heyvaert, F.J. and F.Steurs (eds.)

1989 Worlds behind words. Leuven: Leuven University Press.
Hjelmquist, E.
1979 "Psycholinguistics as scientific encounters". Folia Linguistica 13: 215-225.
1983 "Depth and surface in face-to-face communication". In K.Granstrm (ed.), 35-57.
1984 "Memory for conversations". Discourse Processes 7: 321-336.
1987 "Where is the meaning of verbal discourse?" In J.Verschueren and M.Bertuccelli-
Papi (eds.), 667-680.
1990 "Context and language". In J.Nuyts et al. (eds.), 313-325.
Hjelmquist, E. and A.Gidlund

1985 "Free recall of conversations". Text 5: 169-185.
Hobbs, J.R. and D.AEvans

1980 "Conversation as planned behavior". Cognitive Science 4: 349-377.
Hockett, C F . and R.Asher

1964 "The human revolution". Current Anthropology 5/3: 135-168.
Hoekstra, T.
1978 "Funktionele grammatika". Forum der Letteren 19/4: 293-312.
1981 "An outline of functional grammar". In T.Hoekstra et al. (eds.), 3-18.
Hoekstra, T. and H.van der Kulst (eds.)

1979 Morfologie in Nederland. Glot, special issue.
Hoekstra, T., H.Van der Hulst and M.Moortgat (eds.)

1981 Perspectives on functional grammar. Dordrecht: Foris.
Hook, S. (ed.)
1969 Language and philosophy. New York: New York UP.
Horlek, K.
1964 "Les fonctions de la langue et de la parole". Travaux Linguistiques de Prague 1: 41-
46.
Horecky, J. (ed.)
1982 Coling 82. Amsterdam: North Holland.
Hrmann, H.
1978 Meinen und Verstehen. Frankfurt: Suhrkamp.
Horn, L.R.
1978a "Remarks on neg-raising". In P.Cole (ed.), 129-220.
1978b "Some aspects of negation". In J.Greenberg (ed.), vol.4, 127-210.
1989 A natural history of negation. Chicago: Univ. of Chicago Press.
Horn, L.R. and S.Bayer

1984 "Short circuited implicature". Linguistics and Philosophy 7: 397-414.
Hovy, E.H. and R.C.Schank

1984 "Language generation by computer". In B.Bara and G.Guida (eds.), 165-195.
Hllen, W. and R.Schulze (eds.)

1988 Understanding the lexicon. Tbingen: Niemeyer.
Hurtig, R.
1977 "Toward a functional theory of discourse". In R.Freedle (ed.), 89-106.
Huxley, R. and E.Ingram (eds.)

1971 Language acquisition. London: Academic Press.
Hyams, N.M.
1986 Language acquisition and the theory of parameters. Dordrecht: Reidel.
Hymes, D.H.
1962 "The ethnography of speaking". In J.Fishman (ed. 1968), 99-138.
1969 "The functions of speech". In J.De Cecco (ed.), 78-84.
1971 "Competence and performance in linguistic theory". In R.Huxley and E.Ingram
(eds.), 3-28.
1972 "Models of the interaction of language and social life". In J.Gumperz and D.Hymes
(eds.), 35-71.
REFERENCES 355
Jacobson, M.
1975 "Brain development in relation to language". In E.Lenneberg and E.Lenneberg
(eds.), 105-119.
Jakobson, R.
1960 "Linguistics and poetics". In R.Jakobson 1981, 18-51.
1981 Selected writings, vol.3. The Hague: Mouton.
Jarvella, R.J. and J.G.Collas

1974 "Memory for intentions of sentences". Memory and Cognition 2: 185-188.
Jarvella, R.J. and J.Engelkamp

1983 "Pragmatic influences in producing and perceiving language". In G.Flores d'Arcais
and R.Jarvella (eds.), 225-270.
Jarvella, R.J. and W.H.Klein (eds.)

1982 Speech, place, and action. Chichester: Wiley.
Johnson-Laird, P.N.
1970 "The perception and memory of sentences". In J.Lyons (ed.), 261-270.
1974 "Experimental psycholinguistics". Annual Review of Psychology 25: 135-160.
1977a "Procedural semantics". Cognition 5: 189-214.
1977b "Psycholinguistics without linguistics". In N.Sutherland (ed.), 75-135.
1983 Mental models. Cambridge: CUP.
Johnson-Laird, P.N. and P.C.Wason (eds.)

1977 Thinking. Cambridge: CUP.
Jolles, H.M., A.P.Plompen and H.Weijma (eds.)

1986 Vijftien wetenschappen vijftien jaar verder. Den Haag: Staatsuitgeverij.
Jong, J.De
1981 On the treatment of focus phenomena in functional grammar. In T.Hoekstra et al.
(eds.), 89-115.
Joshi, AK., B.L.Webber and I.A.Sag (eds.)

1981 Elements of discourse understanding. Cambridge: CUP.
Kac, M. (ed.)
1980 Current syntactic theories. Bloomington: IULC.
Kachru, Y.
1970 "The syntax of ko-sentences in Hindi-Urdu". Papers in Linguistics 2: 299-314.
Kaldewaij, J.
1986 Structuralisme en transformationeel generatieve grammatika. Dordrecht: Foris.
Kamp, H.
1979 "Events, instants, and temporal reference". In R.Buerle et al. (eds.), 376-417.
Karmiloff-Smith, A. and B.Inhelder

1974 "If you want to get ahead, get a theory". Cognition 3: 195-212.
Katz, J.J.
1964 "Mentalism in linguistics". Language 40/2: 124-137.
1981 Language and other abstract objects. Oxford: Blackwell.
1984 "An outline of platonist grammar". In T.Bever et al. (eds.), 17-48.
Keenan, E.L.
1976 "Towards a universal definition of 'subject'". In C.Li (ed.), 303-333.
Kempen, G.
1976 "Syntactic constructions as retrieval plans". British Journal of Psychology 67: 149-160.
1977 "Man's sentence generator". In M.De Mey et al. (eds.), 157-164.
1978 "Sentence construction by a psychologically plausible formulator". In R.Campbell
and P.Smith (eds.), vol.2, 103-123.
1979 "A study of syntactic bookkeeping during sentence production". In H.Ueckert and
D.Rhenius (eds.), 361-368.
Kempen, G. (ed.)
1987 Natural language generation. Dordrecht: Nijhoff.
Kempen, G. and E.Hoenkamp

1982 "Incremental sentence generation". In J.Horecky (ed.), 151-156.
1987 "An incremental procedural grammar for sentence formulation". Cognitive Science
11: 201-258.
Kempen, G. and P.Huijbers

1983 "The lexicalization process in sentence production and naming". Cognition 14: 185-
209.
Kenny, A.
1981 "Language and the mind". In H.C.Longuet-Higgins et al. (eds.), 31-37.
Kess, J.F. and R.AHoppe

1981 Ambiguity in psycholinguistics. Amsterdam: Benjamins.
Kintsch, W.
1974 The representation of meaning in memory. Hillsdale: Erlbaum.
1984 "Approaches to the study of the psychology of language". In T.Bever et al. (eds.),
111-145.
Kintsch, W. and T.A.Van Dijk

1978 "Toward a model of discourse comprehension and production". Psychological Review
85: 363-384.
Kintsch, W., J.R.Miller and P.G.Polson (eds.)

1984 Method and tactics in cognitive science. Hillsdale: Erlbaum.
REFERENCES 357
Kiparsky, P. and C.Kiparsky

1970 "Fact". In M.Bierwisch and K.Heidolph (eds.), 143-173.
Kishna, S.
1981 "The recipient state construction in Sarnami". In T.Hoekstra et al. (eds.), 135-156.
Klaiman, M.H.
1980 "Bengali dative subjects". Lingua 51: 275-295.
1981 "Toward a universal semantics of indirect subject constructions". Berkeley Linguistic
Society 7: 123-135.
Klein, W.
1979 "Wegausknfte". LiLi 9: 9-57.
1980 "Verbal planning in route directions". In H.Dechert and M.Raupach (eds.), 159-168.
1982 "Local deixis in route directions". In R.Jarvella and W.Klein (eds.), 161-182.
Koefoed, G. and A.Evers (eds.)

1976 Lijnen van taaltheoretisch onderzoek. Groningen: Willink.
Knig, E.
1974 "The semantic structure of time prepositions in English". Foundations of Language
11: 551-563.
1979 "Direkte und indirekte Bewertung von Zeitintervallen durch Satzadverbien und
Gradpartikeln im Deutschen und Englischen". In H.Weydt (ed.), 175-186.
Korst, B.van der

1987 Twelve sentences: A translation procedure in terms of FG. Amsterdam: Working
Papers in Functional Grammar 19.
Kosslyn, S.
1978 "Imagery and internal representation". In E.Rosch and B.Lloyd (eds.), 217-257.
Koster, J.
1982a "Review van S.C.Dik 1980a". De Nieuwe Taalgids 75: 360-369.
1982b "Karel, Darwin en Chomsky". Hollands Maandblad 415-416: 3-21.
1982c "Leiden ontzet". Hollands Maandblad 421: 38-39.
1983 "Taalkunde: Van common sense naar taalwetenschap". TTT 3: 198-213.
Kraak, A. (ed.)
1975 Linguistics in the Netherlands 1972-1973. Assen: Van Gorcum.
Kreckel, M.
1981 Communicative acts and shared knowledge in natural discourse. New York: Academic
Press.
Kuhn, T.S.
1962 The structure of scientific revolutions. Chicago: Univ. of Chicago Press.
Kwee Tjoe-Liong
1979 A68-FG(3), Simon Dik's funktionele grammatika geschreven in algol68 versie 3.
Amsterdam: Publikaties van het Instituut voor Algemene Taalwetenschap 23.

1981 "In search of an appropriate relative clause". In T.Hoekstra et al. (eds.), 175-189.
1987 "A computer model of functional grammar". In G.Kempen (ed.), 315-331.
Labov, W.
1972a "Some principles of linguistic methodology". Language in Society 1: 97-120.
1972b Language in the inner city. Philadelphia: Univ. of Pennsylvania Press.
1987 "The overestimation of functionalism". In R.Dirven and V.Fried (eds.), 311-332.
Lakatos, I.
1970 "Falsification and the methodology of scientific research programmes". In I.Lakatos
and A.Musgrave (eds.), 91-196.
Lakatos, I. and A.Musgrave (eds.)

1970 Criticism and the growth of knowledge. Cambridge: CUP.
Lakoff, G.
1973 "Fuzzy grammar and the performance/competence terminology game". Chicago
Linguistic Society 9: 271-291.
1987 Women, fire, and dangerous things. Chicago: The Univ. of Chicago Press.
Lakoff, G. and M.Johnson

1980 Metaphors we live by. Chicago: Univ. of Chicago Press.
Lakoff, R.T.
1969 "A syntactic argument for negative transportation". Chicago Linguistic Society 5:140-
147.
Lakoff, R.T. and D.F.Tannen

1979 "Communicative strategies in conversation". Berkeley Linguistic Society 5: 581-592.
Lamb, S.M.
1966 Outline of stratificational grammar. Washington: Georgetown UP.
Lamendella, J.T.
1976 "Relations between the ontogeny and phylogeny of language". In S.Harnad et al.
(eds.), 396-412.
1977a "The limbic system in human communication". In H.Whitaker and H.Whitaker
(eds.), 157-222.
1977b "General principles of neurofunctional organization and their manifestation in
primary and nonprimary language acquisition". Language Learning 27: 155-196.
Langacker, R.W.
1987 Foundations of cognitive grammar, vol.1. Stanford: Stanford UP.
Lasnik, H.
1981 "Learnability, restrictiveness, and the evaluation metric". In C.Baker and J.McCarthy
(eds.), 1-21.
REFERENCES 359
Lehnen, W.G.
1984a "Paradigmatic issues in cognitive science". In W.Kintsch et al. (eds.), 21-49.
1984b "Problems in question answering". In L.Vaina and J.Hintikka (eds.), 137-159.
Lehnert, W.G. and M.H.Ringle (eds.)

1982 Strategies for natural language processing. Hillsdale: Erlbaum.
Lehrman, D.S.
1970 "Semantic and conceptual issues in the nature-nurture problem". In L.Aronson et
al. (eds.), 17-52.
Lenneberg, E.H.
1967 Biological foundations of language. New York: Wiley.
1969 "On explaining language". Science 164: 635-643.
1971 "Of language, knowledge, apes and brains". Journal of Psycholinguistic Research 1:
1-29.
Lenneberg, E.H. and E.Lenneberg (eds)

1975 Foundations of language development. New York: Academic Press.
Leodolter, R.
1975 Das Sprachverhalten von Angeklagten bei Gericht. Kronberg: Skriptor.
Lesser, V.R. and L.D.Erman

1977 "A retrospective view of the HEARSAY-II architecture". Int. Joint Conf on Artificial
Intelligence 5: 790-800.
Leuninger, H.
1977 "Bestimmt die kommunikative Funktion von usserungen ihre Struktur?" Papiere
zur Linguistik 13-14: 143-167.
Levelt, W.J.M.
1972 "Some psychological aspects of linguistic data". Linguistische Berichte 17: 18-30.
1973a Formele grammatica's in lingustiek en taalpsychologie, I. Deventer: Van Loghum
Slaterus.
1973b Formele grammatica's in lingustiek en taalpsychologie, II. Deventer: Van Loghum
Slaterus.
1973c Formele grammatica 's in lingustiek en taalpsychologie, III Deventer: Van Loghum
Slaterus.
1975 "What became of LAD?" In W.Abraham (ed.), 171-190.
1981 "The speaker's linearization problem". In H.C.Longuet-Higgins et al. (eds.), 91-102.
1982a "Het linearizeringsprobleem van de spreker". TTT 2: 1-15.
1982b "Linearization in describing spatial networks". In S.Peters and E.Saarinen (eds.),
199-220.
1989 Speaking. Cambridge: MIT-Press.
Levelt, W.J.M. and G.B.Flores d'Arcais (eds.)

1978 Studies in the perception of language. Chichester: Wiley.
Levelt, W.J.M. and H.Schriefers

1987 "Stages of lexical access". In G.Kempen (ed.), 395-404.
Levelt, W.J.M., J.A.W.M.Van Gent, AF.J.Haans and A.J.A.Meyers

1977 "Grammaticality, paraphrase and imagery". In S.Greenbaum (ed.), 87-101.
Levi, Y., I.M.Schlesinger and D.S.Braine (eds.)

1988 Categories and processes in language acquisition. Hillsdale: Erlbaum.
Levinson, S.
1983 Pragmatics. Cambridge: CUP.
1987 "Minimization and conversational inference". In J.Verschueren and M.Bertuccelli-
Papi (eds.), 61-129.
Lewis, M. and L.A.Rosenblum (eds.)

1977 Interaction, conversation, and the development of language. New York: Wiley.
Li, C.N. (ed.)

1976 Subject and topic. New York: Academic Press.
Lieberman, P.
1973 "On the evolution of language". Cognition 2: 59-94.
1975 On the origins of language. New York: Macmillan.
Lieven, E.
1989 "The linguistic implications of early and systematic variation in child language
development". Berkeley Linguistic Society 15: 203-214.
Lightfoot, D.
1982 The language lottery. Cambridge: MIT-Press.
Linde, C. and W.Labov

1975 "Spatial networks as a site for the study of language and thought". Language 51:924-
939.
Linden, E.
1976 Apes, men, and language. Harmondsworth: Penguin.
Lindholm, J.M.
1969 "Negative raising and sentence pronominalization". Chicago Linguistic Society 5:148-
158.
Lindsay, P.H. and D.A.Norman

1972 Human information processing. New York: Academic Press.
Longuet-Higgins, H.C.
1981 "Procedural theories of linguistic performance". In H.CLonguet-Higgins et al. (eds.),
83-90.
REFERENCES 361
Longuet-Higgins, H.C., J.Lyons and D.E.Broadbent (eds.)

1981 The psychological mechanisms of language. London: The Royal Society and The
British Academy.
Luce, R.D., R.R.Bush and E.Galanter (eds.)

1963 Handbook of mathematical psychology, II. New York: Wiley.
Luria, A.R.
1973 The working brain. Harmondsworth: Penguin.
1974 "Scientific perspectives and philosophical dead ends in modern linguistics". Cognition
3: 377-385.
1976 Basic problems of neurolinguistics. The Hague: Mouton.
Lutz, L.
1981 Zum Thema 'Thema'. Hamburg: Hamburger Buchagentur.
Lyons, J.
1977 Semantics. Cambridge: CUP.
Lyons, J. (ed.)
1970 New horizons in linguistics. Harmondsworth: Penguin.
Lyons, J. and R.J.Wales (eds.)

1966 Psycholinguistics papers. Edinburgh: Edinburgh UP.
Mackenzie, J.L.
1988 "Functionele grammatica: Een thema, geen variatie". TTT 8: 47-56.
Maclay, H. and C.E.Osgood

1959 "Hesitation phenomena in spontaneous English speech". Word 15: 19-44.
Maes, F.
1987 "The pragmatic value of cataphoric relations". In J.Nuyts and G.De Schutter (eds.),
131-146.
Malinowski, B.
1923 "The problem of meaning in primitive languages". In C.Ogden and I.Richards 1949,
296-336.
Mandier, G.
1984 "Cohabitation in the cognitive sciences". In W.Kintsch et al. (eds.), 305-315.
Mandler, J.M.
1982a "Some uses and abuses of a story grammar". Discourse Processes 5: 305-318.
1982b "Another story of grammars". Journal of Pragmatics 6: 433-440.
1982c "Recent research on story grammars". In J.-F.Le Ny and W.Kintsch (eds.), 207-218.
1984 Stories, scripts, and scenes. Hillsdale: Erlbaum.
Mann, W.C.
1984 Discourse structures for text generation. Marina del Rey: USC/ISI Research Report
84-127.
Mann, W.C. and S.A.Thompson

1986 "Relational propositions in discourse". Discourse Processes 9: 57-90.
Marcus, M.P.
1980 A theory of syntactic recognition for natural language. Cambridge: MIT-Press.
1982 "Consequences of functional deficits in a parsing model". In M.Arbib et al. (eds.),
115-133.
1984 "Some inadequate theories of human language processing". In T.Bever et al. (eds.),
253-278.
Marin, O.S.M.
1976 "Neurobiology of language". In S.Harnad et al. (eds.), 900-912.
Marr, D.
1977 "Artificial intelligence". Artificial Intelligence 9: 37-48.
1982 Vision. San Francisco: Freeman.
Marshall, J.C.
1970 "The biology of communication in man and animals". In J.Lyons (ed.), 229-241.
1980 "On the biology of language acquisition". In D.Caplan (ed.), 106-148.
Marslen-Wilson, W.D.
1976 "Linguistic descriptions and psychological assumptions in the study of sentence
perception". In R.Wales and E.Walker (eds.), 203-229.
1980 "Speech understanding as a psychological process". In J.Simon (ed.), 39-67.
Marslen-Wilson, W.D. and L.K.Tyler

1980 "The temporal structure of spoken language understanding". Cognition 8: 1-71.
1981 "Central processes in speech understanding". In H.C.Longuet-Higgins et al. (eds.),
103-118.
Maturana, H.R.
1978 "Biology of language". In G.Miller and E.Lenneberg (eds.), 27-63.
Mayr, E.
1974 "Behavior programs and evolutionary strategies". American Scientist 62: 650-659.
1976 Evolution and the diversity of life. Cambridge: Harvard UP.
McCawley, J.D.
1968 "Review of T.Sebeok (ed.), Current trends in linguistics vol.3". Language 44: 556-
593.
1977 "Acquisition models as models of acquisition". In R.Fasold and R.Shuy (eds.), 51-64.
1980 "An un-syntax". In E.Moravcsik and J.Wirth (eds.), 167-193.
McCawley, N.
1976 "From OE/ME 'impersonal' to 'personal' constructions". Chicago Linguistic Society,
REFERENCES 363
Parasession on Diachronic Syntax, 192-204.

McNeill, D.
1966 "The creation of language by children". In J.Lyons and R.Wales (eds.), 99-132.
1979 The conceptual basis of language. Hillsdale: Erlbaum.
Mehler, J.
1963 "Some effects of grammatical transformations on the recall of English sentences".
Journal of Verbal Learning and Verbal Behavior 2: 346-351.
Mehler, J., J.Morton and P.Jusczyk

1984 "On reducing language to biology". Cognitive Neuropsychology 1: 83-116.
Mehler, J., E.C.T.Walker and M.F.Garrett (eds.)

1982 Perspectives on mental representation. Hillsdale: Erlbaum.
Meijs, W. (ed.)
1987 Corpus linguistics and beyond. Amsterdam: Rodopi.
Menzel, E.W. and M.K.Johnson

1976 "Communication and cognitive organization of humans and other animals". In
S.Harnad et al. (eds.), 131-142.
Mey, J.L. (ed.)

1979 Pragmalinguistics. The Hague: Mouton.
Mey, M.De, R.Pinxten, M.Poriau and F.Vandamme (eds.)

1977 CC 77. Gent: Communication and Cognition.
Miller, G.A.
1956 "The magical number seven, plus or minus two". PsychologicalReview63: 81-96.
1978a "Introduction". In G.Miller and E.Lenneberg (eds.), 1-12.
1978b "Practical and lexical knowledge". In E.Rosch and B.Lloyd (eds.), 305-319.
Miller, G.A. and N.Chomsky

1963 "Finitary models of language users". In R.Luce et al. (eds.), 419-491.
Miller, G.A. and P.N.Johnson-Laird

1976 Language and perception. Cambridge: CUP.
Miller, G.A. and E.Lenneberg (eds.)

1978 Psychology and biology of language and thought. New York: Academic Press.
Miller, J.R., P.G.Polson and W.Kintsch

1984 "Problems of methodology in cognitive science". In W.Kintsch et al. (eds.), 1-18.
Miller, P.
1986 "On certain formal properties of Dik's functional grammar". Belgian Journal of
Linguistics 1: 171-221.
Minsky, M.L.
1968 "Matter, mind and models". In M.Minsky (ed.), 425-432.
1975 "A framework for representing knowledge". In P.Winston (ed.), 211-277.
Minsky, M.L. (ed.)

1968 Semantic information processing. Cambridge: MIT-Press.
Monod, J.
1971 Le hasard et la ncessit. Paris: Seuil.
Moravcsik, E. and J.R.Wirth (eds.)

1980 Syntax and semantics 13: Current approaches to syntax. New York: Academic Press.
Morgan, J.L.
1978 "Two types of convention in indirect speech acts". In P.Cole (ed.), 261-280.
Morris, C.
1938 Foundations of the theory of signs. Chicago: Univ. of Chicago Press.
1946 Signs, language, and behavior. New York: Prentice Hall.
1971 Writings on the general theory of signs. The Hague: Mouton.
Morscher, E., O.Neumaier and G.Zecha (eds.)

1981 Philosophie als Wissenschaft. Bad Reichenhall: Comes.
Morton, J.
1971 "What could possibly be innate?" In J.Morton (ed.), 82-97.
1981 "The status of information processing models of language". In H.C.Longuet-Higgins
et al. (eds.), 173-182.
1984 "Brain-based and non brain-based models of language". In D.Caplan et al. (eds.), 40-
64.
Morton, J. (ed.)
1971 Biological and social factors in psycholinguistics. London: Logos.
Moutaouakil, A.
1986 Towards an adequate representation of illocutionary force in functional grammar.
Amsterdam: Working Papers in Functional Grammar 10.
Mller, B. (ed.)
1985 Sprachsynthese. Hildesheim: Olms.
Muysken, P.
1988 "Taalkunde in F-groot en F-klein". TTT 8: 35-46.
Nagao, M. (ed.)
1987 Language and artificial intelligence. Amsterdam: North Holland.
Nagel, E.
1961 The structure of science. London: Routledge.
1979 Teleology revisited and other essays in the philosophy and history of science. New York:
REFERENCES 365
Columbia UP.
Nagel, E., P.Suppes and A.Tarski (eds.)

1962 Logic, methodology and philosophy of science. Stanford: Stanford UP.
Negus, V.E.
1949 The comparative anatomy and physiology of the larynx. New York: Hafner.
Neumaier, O.
1979 Biologische und soziale Voraussetzungen der Sprachkompetenz. Innsbruck: Conceptus.
Newmeyer, F.J.
1980 Linguistic theory in America. New York: Academic Press.
1983 Grammatical theory. Chicago: Univ. of Chicago Press.
Nichols, J., G.Rappoport and ATimberlake

1980 "Subject, topic, and control in Russian". Berkeley Linguistic Society 6: 372-386.
Nilsen, D.L.F.
1972 Toward a semantic specification of deep case. The Hague: Mouton.
Ninio, A. and C.Snow

1988 "Language acquisition through language use". In Y.Levi et al. (eds.), 11-30.
Nuyts, J.
1982a Een psycholinguistische benadering van de fonologische aspekten van afasie. Wilrijk:
Antwerp Papers in Linguistics 26.
1982b "Afasie en lingustiek". De Psycholoog 17: 377-383.
1983a "On the methodology of a functional language theory". In S.Dik (ed.), 369-386.
1983b "Funktionele grammatika: Een notationele variant?" De Nieuwe Taalgids 76: 19-26.
1983c "A system for phonological language description". Communication and Cognition 16:
413-438.
1984 Sprachfunktionen. Trier: LAUT A-128.
1985a Funkties, funktionalisme en psychologische realiteit. Wilrijk: Antwerp Papers in
Linguistics 41.
1985b "On formalism, functionalism and mentalism". In J.Nuyts (ed.), 113-168.
1986a "Review of V.L.Zammuner 1981 and R. Wiese 1983". Journal of Pragmatics 10: 139-
149.
1986b "What formalists seem not to understand about functionalism". Belgian Journal of
Linguistics 1: 223-237.
1987 "Pragmatics and cognition". In J. Verschueren and M.Bertuccelli-Papi (eds.), 713-730.
1988 Aspekten van een kognitief-pragmatische taaltheorie. PhD-thesis, Univ. of Antwerp.
1990a "Emotions and the functionality of language". Grazer Linguistische Studien 33, 227-
240.
1990b "Intentions and language use". Ms., Univ. of Antwerp.
Nuyts, J. (ed.)
1985 Antwerp studies in functional grammar. Wilrijk: Antwerp Papers in Linguistics 39.
Nuyts, J. and G.de Schutter (eds.)

1987 Getting one's words into line. Dordrecht: Foris.
Nuyts, J. and J.Verschueren

1987 A comprehensive bibliography of pragmatics. Amsterdam: Benjamins.
Nuyts, J., A.M.Bolkestein and C.Vet (eds.)

1990 Layers and levels of representation in language theory. Amsterdam: J.Benjamins.
Ny, J.-F. Le and W.Kintsch (eds.)

1982 Language and comprehension. Amsterdam: North Holland.
Ochs, E.
1983 "Cultural dimensions of language acquisition". In E.Ochs and B.Schieffelin (eds.),
185-191.
Ochs, E. and B.B.Schieffelin

1984 "Language acquisition and socialization". In R.Shweder and R.LeVine (eds.), 276-
320.
Ochs, E. and B.B.Schieffelin (eds.)

1979 Developmental pragmatics. New York: Academic Press.
1983 Acquiring conversational competence. London: Routledge.
Ogden, CK. and I.ARichards

1949 The meaning of meaning. London: Kegan Paul. (5th edition)
Osgood, C.E.
1980 Lectures on language performance. New York: Springer.
1984 "Towards an abstract performance grammar". In T.Bever et al. (eds.), 147-179.
stman, J.-O.
1987 "Implicit involvement in interactive writing". In J.Verschueren and M.Bertuccelli-
Papi (eds.), 155-178.
Palmer, F.R;
1986 Mood and modality. Cambridge: CUP.
Palmer, F.R. (ed.)

1968 Selected papers of J.R.Firth 1952-1959. London: Longman.
Palmer, S.E.
1978 "Fundamental aspects of cognitive representation". In E.Rosch and B.Lloyd (eds.),
259-303.
Papert, S.
1980 "The role of artificial intelligence in psychology". In M.Piattelli-Palmarini (ed.), 90-
99.
REFERENCES 367
Parret, H.
1974 "Over de waarde van teoretische constructies". Leuvense Bijdragen 63: 311-332.
1979a Filosofie en taalwetenschap. Assen: Van Gorcum.
1979b "Significance and understanding". Dialectica 33: 297-318.
1980 Contexts of understanding. Amsterdam: Benjamins.
1982 "Spreken en verstaan". TTT 2: 28-43.
1983 Semiotics and pragmatics. Amsterdam: Benjamins.
1986 "Pragmatics". In T.Sebeok (ed.), 751-760.
Parret, H. (mod.)
1974 Discussing language. The Hague: Mouton.
Perlmutter, D.M.
1979 "Working 1s and inversion in Italian, Japanese, and Quechua". Berkeley Linguistic
Society 5: 277-324.
1981 "Functional grammar and relational grammar". In T.Hoekstra et al. (eds.), 319-352.
Perlmutter, M. (ed.)
1984 Parent-child interactions and parent-child relations in child development. Hillsdale:
Erlbaum.
Peters, S. and E.Saarinen (eds.)

1982 Processes, beliefs and questions. Dordrecht: Reidel.
Petfi, J.S. (ed.)

1979 Text vs. sentence. Hamburg: Buske.
Piaget, J.
1923 Le langage et la pense chez l'enfant. Neuchtel: Delachaux.
1971 Biology and knowledge. Edinburgh: Edinburgh UP.
1980a "The psychogenesis of knowledge and its epistemological significance". In
M.Piattelli-Palmarini (ed.), 23-34.
1980b "Schemes of action and language learning". In M.Piattelli-Palmarini (ed.), 164-167.
1980c "Afterthoughts". In M.Piattelli-Palmarini (ed.), 278-284.
Piaget, J. and B.Inhelder

1972 De psychologie van het kind. Rotterdam: Lemniscaat.
Piattelli-Palmarini, M. (ed.)
1980 Language and learning. London: Routledge.
Pinker, S.
1981 "Comments on K.Wexler 1981". In C.Baker and J.McCarthy (eds.), 53-63.
Pinkster, H.
1985 "The discourse function of the passive". In A.M.Bolkestein et al. (eds. 1985a), 107-
118.
Piotkin, H.C. (ed.)

1982 Learning, development, and culture. New York: Wiley.
Poison, P.G., J.R.Miller and W.Kintsch

1984 "Methods and tactics reconsidered". In W.Kintsch et al. (eds.), 277-296.
Popper, K.R.
1963 Conjectures and refutations. London: Routledge. (4th edition, 1972)
1972 Objective knowledge. Oxford: Oxford UP.
1981 "The present significance of two arguments by Henri Poincar". In W.Morscher et
al. (eds.), 19-24.
Popper, K.R. and J.Eccles

1977 The self and its brain. Berlin: Springer.
Power, R.
1979 "The organization of purposeful dialogues". Linguistics 17: 107-152.
Premack, D.
1980 "Representational capacity and accessibility of knowledge". In M.Piattelli-Palmarini
(ed.), 205-221.
Prince, E.F.
1976 "The syntax and semantics of neg-raising, with evidence from French". Language 52:
404-426.
1981 "Toward a taxonomy of given-new information". In P.Cole (ed.), 223-255.
Psathas, G. (ed.)
1979 Everyday language. New York: Irvington.
Putnam, H.
1971 "The 'innateness hypothesis' and explanatory models in linguistics". In J.Searle (ed.),
130-139.
1973 "Reductionism and the nature of psychology". Cognition 2: 131-146.
1980a "What is innate and why". In M.Piattelli-Palmarini (ed.), 287-309.
1980b "Philosophy and our mental life". In N.Block (ed.), 134-143.
Pylyshyn, Z.W.
1972 "Competence and psychological reality". American Psychologist 27: 546-552.
1973 "The role of competence theories in cognitive psychology". Journal ofPsycholinguistic
Research 2: 21-50.
1980 "Computation and cognition". The Behavioral and Brain Sciences 3: 111-169.
Quillian, M.R.
1967 "Word concepts". Behavioral Science 12: 410-430.
1968 "Semantic memory". In M.Minsky (ed.), 227-270.
1969 "The teachable language comprehender". Communications of the ACM 12: 459-476.
REFERENCES 369
Quine, W.V.O.
1969 "Linguistics and philosophy". In S.Hook (ed.), 95-98.
Redder, A. (ed.)
1983 Kommunikation in Institutionen. Osnabrck: OBST 24.
Reddy, D.R., L.D.Erman, R.D.Fennell and R.B.Neely

1973 "The HEARSAY speech understanding system". Int. Joint Council on Artificial
Intelligence 3: 185-193.
Reichling, A.
1935 Het woord. Zwolle: Willink. (2nd printing, 1967)
1969 Verzamelde studies over hedendaagse problemen der taalwetenschap. Zwolle: Willink.
(5th edition)
Rickheit, G. and M.Bock (eds.)

1983 Psycholinguistic studies in language processing. Berlin: De Gruyter.
Riemsdijk, H.Van
1982 "Over de verscheidenheid van de menselijke taalwetenschap". Tilburg Studies in
Language and Literature 3: 63-89.
Riesbeck, C.K. and R.C.Schank

1978 "Comprehension by computer". In W.Levelt and G.Flores d'Arcais (eds.), 247-293.
Rijkhoff, J.N.M.
1990 "Toward a unified analysis of terms and predications". In J.Nuyts et al. (eds.), 165-
191.
Robinson, W.P.
1972 Language and social behaviour. Harmondsworth: Penguin.
Rosch, E.
1974 "Linguistic relativity". In A.Silverstein (ed.), 95-121.
Rosch, E. and B.B.Lloyd (eds.)

1978 Cognition and categorization. Hillsdale: Erlbaum.
Ross, J.R.
1979 "Where's English?" In C.Fillmore et al. (eds.), 127-163.
Ruder, K.F. and A.Finch

1987 Toward a cognitive-based model of language production. In H.Dechert and
M.Raupach (eds.), 109-138.
Rumelhart, D.E.
1975 "Notes on a schema for stories". In D.Bobrow and A.Collins (eds.), 211-236.
Rycker, T.A.De
1987 "Turns at writing". In J.Verschueren and MBertuccelli-Papi (eds.), 613-647.
Ryle, G.
1949 The concept of mind. Harmondsworth: Penguin. (Edition 1980)
Sachs, J.
1967 "Recognition memory for syntactic and semantic aspects of connected discourse".
Perception and Psychophysics 2: 437-442.
Sacks, H., E.A.Schegloff and G.Jefferson

1974 "A simplest systematics for the organization of turn-taking for conversation". Langu-
age 50: 696-735.
Saffran, E.M., M.F.Schwartz and O.S.M.Marin

1980 "Evidence from aphasia: Isolating the components of a production model". In
B.Butterworth (ed.), 221-241.
Samuelsdorff, P.O.
1982 "The treatment of morphology in a functional grammar with special reference to
Swahili". Folia Linguistica 16: 385-398.
Sapir, E.
1912 "Language and environment". In E.Sapir 1968, 89-103.
1929 "The status of linguistics as a science". In E.Sapir 1968, 160-166.
1931 "Communication". In E.Sapir 1968, 104-109.
1933 "Language". In E.Sapir 1968, 7-32.
1968 Selected writings of Edward Sapir in language, culture and personality. Edited by
D.G.Mandelbaum. Berkeley: Univ. of California Press.
Schank, R.C.
1972a "Conceptual dependency". Cognitive Psychology 3: 552-631.
1972b "'Semantics' in conceptual analysis". Lingua 30: 101-140.
1973 "Identification of conceptualizations underlying natural language". In R.Schank and
K.Colby (eds.), 187-247.
1978 "Predictive understanding". In R.Campbell and P.Smith (eds.), vol.2, 91-101.
1979 "Some prerequisites for a computational pragmatics". In J.Mey (ed.), 365-407.
1980 "Language and memory". Cognitive Science 4: 243-284.
Schank, R.C. and R.P.Abelson

1977 Scripts, plans, goals and understanding. Hillsdale: Erlbaum.
Schank, R.C. and L.Birnbaum

1984 "Memory, meaning, and syntax". In T.Bever et al. (eds.), 209-251.
Schank, R.C. and K.M.Colby (eds.)

1973 Computer models of thought and language. San Francisco: Freeman.
Schank, R.C. and R.Wilensky

1977 "Response to Dresher and Hornstein". Cognition 5: 133-145.
REFERENCES 371
Schank, R., L.Birnbaum and J.Mey

1985 "Integrating semantics and pragmatics". Quaderni de Semantica6: 313-325.
Schank, R.C., N.M.Goldman, C.J.Rieger III and C.K.Riesbeck

1975 Conceptual information processing. Amsterdam: North Holland.
Schegloff, E.A.
1979 "Identification and recognition in telephone conversation openings". In G.Psathas
(ed.), 23-78.
Scherer, K.R. and H.Giles (eds.)
1979 Social markers in speech. Cambridge: CUP.
Schlesinger, I.M.
1967 "A note on the relationship between psychological and linguistic theories".
Foundations of Language 3: 397-402.
1971 "On linguistic competence". In Y.Bar-Hillel (ed.), 150-172.
1977 Production and comprehension of utterances. Hillsdale: Erlbaum.
1982 Steps to language. Hillsdale: Erlbaum.
Schlieben-Lange, B. (ed.)
1975 Sprachtheorie. Hamburg: Hoffmann und Campe.
Schnelle, H.
1981a "Introspection and the description of language use". In F.Coulmas (ed.), 105-126.
1981b "Elements of theoretical net-linguistics, pt.l". Theoretical Linguistics 8: 67-100.
1984 "Concurrent parsing in programmable logic array (PLA-) nets". Proc. Intern. Conf.
on Comput. Linguistics 10: 150-153.
Schutter, G.de
1981 "Beschouwingen over het Nederlandse tempussysteem vanuit een universalistische
gezichtshoek". In L.Goossens (ed.), 21-47.
1983 "Modaliteit en andere modificaties in de Nederlandse grammatica". In F.Daems and
1985a "Pragmatic and syntactic aspects of word order in Dutch". In A.M.Bolkestein et al.
(eds. 1985a), 137-154.
1985b "Typological aspects of Dutch and German". In J.Nuyts (ed.), 73-112.
1987 "Pragmatic positions". In J.Nuyts and G.De Schutter (eds.), 103-118.
1989 Casussen, syntactische funktietoekenning en gemarkeerdheid. Wilrijk: Antwerp Papers
in Linguistics 63.
Schutter, G.de and J.Nuyts

1983 "Towards an integrated model of a functional grammar". In S.Dik (ed.), 387-404.
Schwartz, R.
1969 "On knowing a grammar". In S.Hook (ed.), 183-190.
Searle, J.R.
1969 Speech acts. Cambridge: CUP.
1972 "Chomsky's revolution in linguistics". The New York Review of Books, June 29:16-24.
1976 "A classification of illocutionary acts". Language in Society 5: 1-23.
1983 Intentionality. Cambridge: CUP.
1986 "Meaning, communication, and representation". In R.Grandy and R.Warner (eds.),
209-226.
Searle, J.R. (ed.)

1971 The philosophy of language. Oxford: Oxford UP.
Sebeok, T. (ed.)
1986 Encyclopedic dictionary of semiotics. Berlin: De Gruyter.
Settekorn, W.
1981 "Toi aussi on te tlphone come a...' - 'Oui ben je j'sais pas ce que j'ferais'".
Cahiers de Linguistique Franaise 2: 183-204.
Seuren, P.A.M.
1974 "Negative's travels". In P.Seuren (ed.), 183-208.
1975 Tussen taal en denken. Utrecht: Oosthoek.
1977 On linguistic awareness. Nijmegen: Grammarij 4.
1985 Discourse semantics. Oxford: Blackwell.
Seuren, P.A.M. (ed.)

1974 Semantic syntax. Oxford: Oxford UP.
Shanon, B.
1987 "The non-abstractness of mental representations". New Ideas in Psychology 5: 117-
126.
Shatz, M.
1984 "Contributions of mother and mind to the acquisition of communicative compe
tence". In M.Perlmutter (ed.), 33-59.
Shibatani, M.
1977 "Grammatical relations and surface cases". Language 53: 789-809.
Shuy, R.W. (ed.)

1973 Some new directions in linguistics. Washington: Georgetown UP.
Shweder, R. and R.LeVine (eds.)

1984 Culture theory. New York: CUP.
Siewierska, A.
1987 Postverbal subject pronouns in Polish in the light of topic continuity and the topic/focus
distinction. In J.Nuyts and G.De Schutter (eds.), 147-161.
Silverstein, A. (ed.)
1974 Human communication. Hillsdale: Erlbaum.
REFERENCES 373
Simmons, R. and J.Slocum

1972 "Generating English discourse from semantic networks". Communications of the
ACM 15: 891-905.
Simon, J.C. (ed.)

1980 Spoken language generation and understanding. Dordrecht: Reidel.
Skinner, B.F.
1957 Verbal behavior. New York: Appleton-Century-Crofts.
Smedt, K.De and G.Kempen

1987 "Incremental sentence production, self-correction and coordination". In G.Kempen
(ed.), 365-376.
Smith, A.
1978 "Lenneberg, Locke, Zangwill, and the neuropsychology of language and language
disorders". In G.Miller and E.Lenneberg (eds.), 133-149.
Smith, N.V. (ed.)

1982 Mutual knowledge. New York: Academic Press.
Snow, C.E.
1975a "The development of conversation between mothers and babies". Pragmatics
Microfiche 1/6: A2.
1975b "Linguists as behavioral scientists". In A.Kraak (ed.), 271-275.
Snow, C. and C.Ferguson (eds.)

1977 Talking to children. Cambridge: CUP.
Snow, C. and G.Meijer

1977 "On the secondary nature of syntactic intuitions". In S.Greenbaum (ed.), 163-177.
Sridhar, S.N.
1976 "Dative subjects". Chicago Linguistic Society 12: 582-593.
1979 "Dative subjects and the notion of subject". Lingua 49: 99-125.
Stabler, E.P.
1983 "How are grammars represented?" The Behavioral and Brain Sciences 6: 391-421.
Stebbins, G.L.
1982 Darwin to DNA, molecules to humanity. San Francisco: Freeman.
Stechow, A.Von
1982 "Three local deictics". In R.Jarvella and W.Klein (eds.), 73-99.
Steedman, M.J. and P.N.Johnson-Laird

1978 "A programmatic theory of linguistic phenomena". In R.Campbell and P.Smith
(eds.), vol.2, 171-192.
1980 "The production of sentences, utterances, and speech acts". In B.Butterworth (ed.),
111-141.
Steen, G.
1985 Grammar and metaphor. Amsterdam: Working Papers in Functional Grammar 3.
Steinberg, D.D.
1975 "Chomsky: From formalism to mentalism and psychological invalidity". Glossa 9:
218-252.
1982 Psycholinguistics. London: Longman.
Stemmer, N.
1971 "A note on competence and performance". Linguistics 65: 83-89.
Stenning, K.
1978 "Anaphora as an approach to pragmatics". In M.Halle et al. (eds.), 162-200.
Stich, S.P.
1972 "Grammar, psychology and indeterminacy". Journal of Philosophy 69: 799-818.
Straight, S.
1976 "Comprehension versus production in linguistic theory". Foundations of Language
12: 525-540.
Strawson, P.F.
1971 "Intention and convention in speech acts". In J.Searle (ed.), 23-38.
Sutherland, N.S. (ed.)

1977 Tutorial essays in psychology, voll. Hillsdale: Erlbaum.
Tannen, D.
1987 "Conversational style". In H.Dechert and M.Raupach (eds.), 251-267.
Tannen, D. and M.Saville-Troike (eds.)

1985 Perspectives on silence. Norwood: Ablex.
Tanner, N. and A.Zihlman

1976 "The evolution of human communication". In S.Harnad et al. (eds.), 467-480.
Tervoort, B.T. (ed.)

1977 Wetenschap en taal. Muiderberg: Coutinho.
Thompson, D.W.
1917 On growth and form. Cambridge: CUP.
Thompson, S.A. and W.C.Mann

1987 "Rhetorical structure theory". Papers in Pragmatics 1/1: 79-105.
Tombe, L.Des
1976 "Competence en performance". In G.Koefoed and A.Evers (eds.), 111-141.
Toulmin, S.
1970 "Does the distinction between normal and revolutionary science hold water?" In
I.Lakatos and A.Musgrave (eds.), 39-47.
REFERENCES 375
1971 "Brain and language". Synthese 22: 369-395.

1972 Human understanding I. Princeton: Princeton UP.
Traugott, E.C.
1978 "On the expression of spatio-temporal relations in language". In J.Greenberg (ed.),
vol.3, 369-400.
Ueckert, H. and D.Rhenius (eds.)

1979 Komplexe menschliche Informationsverarbeitung. Bern: Huber.
Ullmer-Ehrich, V.
1979 "Wohnraumbeschreibungen". LiLi 9: 58-83.
1982 "The structure of living space descriptions". In R.Jarvella and W.Klein (eds.), 219-
249.
Vachek, J.
1966 The linguistic school of Prague. Bloomington: Indiana UP.
Vaina, L. and J.Hintikka (eds.)

1984 Cognitive constraints on communication. Dordrecht: Reidel.
Valian, V.
1977 "Talk, talk, talk". In R.Freedle (ed.), 107-139.
1979 "The wherefores and therefores of the competence-performance distinction". In
W.Cooper and E.Walker (eds.), 1-26.
Valin, R.D.Van
1990 "Layered syntax in role and reference grammar". In J.Nuyts et al. (eds.), 193-231.
Verschueren, J.
1983 "Review article of Ballmer and Brennenstuhl 1981". Language 59: 166-175.
1985 "Review article of G.N.Leech, 'Principles of pragmatics', and S.C.Levinson,
'Pragmatics'". Journal of Linguistics 21: 459-470.
1987a Pragmatics as a theory of linguistic adaptation. Wilrijk: IPrA Working Document 1.
1987b "The pragmatic perspective". In J.Verschueren and M.Bertuccelli-Papi (eds.), 3-8.
Verschueren, J. (ed.)
1987 Concluding round table 1987 International Pragmatics Conference. Wilrijk: IPrA
Working Document 2.
Verschueren, J. and M.Bertuccelli-Papi (eds.)

1987 The pragmatic perspective. Amsterdam: Benjamins.
Vet, C.
1981 "La notion de 'monde possible' et le systme temporel et aspectuel du franais".
Langages 64: 109-124.
1986 A pragmatic approach to tense in functional grammar. Amsterdam: Working Papers
in Functional Grammar 16.
1990 "Asymmetries in the use of tense and modality". In J.Nuyts et al. (eds.), 123-137.
Villiers, P.A.De and J.G.De Villiers

1972 "Early judgments of semantic and syntactic acceptability by children". Journal of
Psycholinguistic Research 1: 299-310.
Wales, R.J. and E.Walker (eds.)

1976 New approaches to language mechanisms. Amsterdam: North Holland.
Wanner, E. and L.R.Gleitman (eds.)

1982 Language acquisition. Cambridge: CUP.
Waterson, N. and C.Snow (eds.)

1978 The development of communication. Chichester: Wiley.
Watt, W.C.
1974 "Mentalism in linguistics, II". Glossa 8: 3-40.
Watters, J.R.
1985 "The place of morphology in functional grammar". In AM.Bolkestein et al. (eds.
1985b), 85-104.
Watzlawick, P.
1976 Is 'werkelijk' waar? Deventer: Van Loghum.
Weigand, H.
1987 "Functional grammar as a formal language". In J.Van der Auwera and L.Goossens
(eds.), 179-194.
Weijdema, W., S.C.Dik, M.Oehlen, C.Dubber and A D e Blauw

1982 Strukturen in verbale interaktie. Muiderberg: Coutinho.
Weiner, S.L. and D.R.Goodenough

1977 "A move toward a psychology of conversation". In R.Freedle (ed.), 213-225.
Wells, C G . and W.P.Robinson

1982 "The role of adult speech in language development". In C.Fraser and K.Scherer
(eds.), 11-76.
Wells, R.
1969 "Innate knowledge". In S.Hook (ed.), 99-119.
Wexler, K.
1981 "Some issues in the theory of learnability". In C.Baker and J.McCarthy (eds.), 30-52.
Weydt, H. (ed.)
1979 Die Partikeln der deutschen Sprache. Berlin: De Gruyter.
Whitaker, H.A.
1973 "Comments on the innateness of language". In R.Shuy (ed.), 95-120.
1974 "Is the grammar in the brain?" In D.Cohen (ed.), 75-89.
REFERENCES 377
Whitaker, H. and H.A. Whitaker (eds.)

1977 Studies in neurolinguistics, vol. 3. New York: Academic Press.
1979 Studies in neurolinguistics, vol.4. New York: Academic Press.
Whorf, B.L.
1963 Sprache, Denken, Wirklichkeit. Hamburg: Rowohlt.
Wierzbicka, A.
1972 Semantic primitives. Frankfurt: Athenum.
1980 Lingua mentalis. Sydney: Academic Press.
Wiese, B.
1983 "Anaphora by pronouns". Linguistics 21: 373-417.
Wiese, R.
1983 Psycholinguistische Aspekte der Sprachproduktion. Hamburg: Buske.
Wijk, C.Van and G.Kempen

1985 "From sentence structure to intonation contour". In B.Mller (ed.), 157-182.
Wilensky, R.
1983 "Story grammars versus story points". The Behavioral and Brain Sciences 6: 579-620.
Wilks, Y.
1976 "Frames, scripts, stories and fantasies". Pragmatics Microfiche 2/3: B13-D14.
1982 "Some thoughts on procedural semantics". In W.Lehnert and M.Ringle (eds.), 495-
516.
Williams, E.
1981 "A readjustment in the learnability assumptions". In C.Baker and J.McCarthy (eds.),
64-78.
Wilson, D. and D.Sperber

1984 "Pragmatics". In S.George (ed.), 21-41.
Winograd, T.
1972 Understanding natural language. New York: Academic Press.
1975 "Frame representations and the declarative/procedural controversy". In D.Bobrow
and ACollins (eds.), 185-210.
1976 "Towards a procedural understanding of semantics". Revue Internationale de
Philosophie 30: 260-303.
1977a "On some contested suppositions of generative linguistics about the scientific study
of language". Cognition 5: 151-179.
1977b "Formalisms for knowledge". In P.Johnson-Laird and P.Wason (eds.), 62-71.
Winston, P.H.
1984 Artificial intelligence. Second edition. Reading: Addison-Wesley.
Winston, P.H. (ed.)

1975 The psychology of computer vision. New York: McGraw-Hill.
Wittgenstein, L.
1922 Tractatus Logico-Philosophicus. Amsterdam: Atheneum. (Edition 1976)
1958 Philosophische Untersuchungen. Frankfurt: Suhrkamp. (Edition 1980)
Woods, W.A.
1970 "Transition network grammars for natural language analysis". Communications of the
ACM 13: 591-606.
1982 "HWIM: A speech understanding system on a computer". In M.Arbib et al. (eds.),
95-113.
Woolard, K.A.
1987 "Codeswitching and comedy in Catalonia". Papers in Pragmatics 1/1: 106-122.
Wright, L.
1973 "Functions". Philosophical Review 82: 139-168.
Wunderlich, D.
1970 Tempus und Zeitreferenz im Deutschen. Mnchen: Hueber.
1974 Grundlagen der Linguistik. Opladen: Westdeutscher Verlag.
Wygotski, L.S.
1974 Denken und Sprechen. Frankfurt: Fischer.
Zaefferer, D.
1979 "Sprechakttypen in einer Montague-Grammatik". In G.Grewendorf (ed.), 386-417.
Zangwill, O.L.
1978 "Aphasia and the concept of brain centers". In G.Miller and E.Lenneberg (eds.),
119-132.
Zock, M. and G.Zabah (eds.)

1988 Advances in natural language generation, vol.1. London: Pinter.
Subject Index
abstractness 74, 186, 223, 228, 231, 236, 285, anaphora 4, 198, 212-213, 276, 282, 321
322, 324-325 animacy 204-205
address anthropology 144
forms of 272 aphasia 76, 139-141, 162, 166
adequacy 74, 99, 120-121, 128, 131, 135, 138, argument 187-189, 199, 210, 253, 274, 281-282,
148-149, 152, 299 284, 291, 325
cognitive 8-10, 20-22, 70, 91, 95-96, 109, first 202, 205, 212, 217, 276, 282
114, 126-129, 131-133, 184-185, 190, 219, 229, second 202, 205, 209-210, 212, 217, 276,
232-233, 255, 325
descriptive 93-97 third 209, 283, 325
explanatory 93-97, 99, 121, 156, 184 argument reduction 281
see also explanation argument structure 187, 194, 195, 249
external descriptive 9-10, 16, 95-97, 185 articulation component 284
see also data, empirical evidence artificial intelligence (AI) 16, 71, 89, 91, 113,
internal descriptive 9-10, 16, 22, 71, 95-99, 138, 145, 192-193, 218-219, 223, 244-245,
131, 155, 185 247, 260, 263, 325
levels of 93-98 aspect 186, 188, 224, 267-268, 277-279, 282-
observational 9, 64, 93-96 283
see also data, empirical evidence attention 254
pragmatic 184-185 attitude
psychological see speaker attitude
see adequacy (cognitive) augmented transition network (ATN) 132,
typological 185 245, 247
see also typology
adjective 186
modal 303, 305, 307, 309, 311, 317, 330 baby-talk 157-158
predicatively used 288, 303, 305, 307, 309, back channel cue 262-264
311, 317 background 38, 47-48, 57, 144, 238
adult-child interaction 157-158 cultural 37, 59, 117-118, 157, 224, 231, 238,
adverbial 188, 210, 278-279 272
modal 279, 282, 303, 305, 307, 309, 311 individual differences in 48
spatial 278 social 117
temporal 278 background information 264
agent 199-200, 210, 227, 275, 281, 325 behavior 233-235, 237
agrammatism 260 inertia of 61, 128
ambiguity 100-101, 120-121, 124, 216, 251, behaviorism 5, 23, 84, 89, 91, 161, 180
286, 299, 313-314, 328 beneficiary 275, 327
amnesia 229 bilingualism 232, 326
analysis-by-synthesis 130, 178 biological vs cognitive system 77-78, 106, 117,
126-127, 153, 160, 163, 169, 179 communication 35-64, 72-74, 76, 79, 81, 86-87,
see also neurological vs cognitive 112, 168, 174-177, 196, 221-222, 231-233,
systems 248, 297, 300, 325
biology 26-27, 29, 31, 32, 33, 75-80, 114, 117, analysis of 49-50
160, 171-172, 179, 181 constitutive factors of 39-40, 51-64, 111,
brain vs mind 6-7, 13, 84, 140-141, 162 219, 221, 256-257, 264, 327
and hardware vs software 162-163, 181 determining factor of 61
see also neurological vs cognitive dyadic 51-55
systems group 55
indirect 59
meta- 39-40, 271
case (grammatical) 202-203, 205, 324 non-verbal 8, 46-47, 158-159, 168, 170, 172-
case (semantic) 200, 202 177, 234-235, 305
see also semantic function see also phylogenesis of language (and
case grammar 190-191 non-verbal symbol systems)
cataphora 211-213, 276, 282 one-sided 55-56
causation 239, 274 sub-function of 36-41, 45
causative 200, 205 topic of
causative construction 197-202, 209 see theme
channel 40, 54, 56 universals of 37, 40, 60
character 38 written 54, 235
cleft construction 227, 314-315, 331 see also language (functions of)
cleft-negation construction 286, 299, 313-315 communicative act 36-41, 44, 46-47, 51-63,
code 54-55, 116, 230, 326 188, 196
coding hypothesis 129 communicative effect 57, 115
cognition communicative efficiency 176-177
black box of 5-6, 64, 104, 121 communicative failure 57, 81, 176, 250, 254,
limits to our understanding of 104 291
modularity of 106-107, 126-127, 178 communicative goal 111-112, 219, 318
nature of 24, 184-185 see also illocution, intention (communica
see also biological vs cognitive system, tive) brain vs mind, language (cognitive systems
of), language acquisition (and general communicative situation 39, 44, 51, 53, 55-59,
cognitive development), phylogenesis of 62, 86, 111, 158, 192, 220-222, 260, 263,
language (and general cognitive evolution) 265, 272, 281, 291, 317
communicative strategy 299
cognitive development 47-48, 77 communicative success 56
see also language acquisition (and general comparison 273
cognitive development) competence 93, 98, 100, 105-107, 110, 120,
cognitive grammar 92, 260 132, 138, 149, 151, 178-179, 183, 192
cognitive reality communicative 70, 104, 109, 183
see adequacy (cognitive), cognition see also linguistic knowledge
cognitive science 5, 11, 25, 79, 92, 113, 179 competence vs performance 23, 84, 89, 101-
cognitive simplicity postulate 223, 233-234 105, 108, 110, 114, 125, 128, 130, 133, 135-
cognitivism 5, 11, 17-25, 127-128, 130-131, 138,136,
148,148
151, 177-179
see also linguistic knowledge (and
cohesion 261 language use), performance
SUBJECT INDEX 381
complement control 113-116, 121, 134, 139-140, 145-146,

see sentence (embedded) 216-218, 222, 245-246, 252, 254-255, 328
complementizer 306, 317, 331 blackboard 218, 255
computational equivalence 132, 190 individual differences in 113, 139
see also notational variant see also demon, language processing
concept 224, 230-232, 234-235, 237-240, 243, (automaticity of)
259-260, 275, 279, 281-282 conversation 262, 299
conceptual domain 259-260 side-sequence in 263
conceptual relation 237-240, 243, 274-275, see also discourse
281 conversational analysis 4, 263
conceptual role 193, 200, 202, 274-276, 281- coordinator 321
282, 324 coreferentiality 203, 205, 210-213, 276, 282,
conceptualization 47-48, 160, 168, 191-193, 321
195-196, 213, 219, 223-244, 246-250, 253, culture
256, 258-259, 261-263, 268-269, 272-277, see background (cultural)
279-283, 297, 299, 302, 304, 307, 311-312,
314, 316-318, 320-321, 324, 326-327
embedding of 243, 310, 314 data
see also quotation deviant 14, 105, 115
language-(in)dependence of 186, 223-238, experimental 8, 14-15, 129, 131, 229-230,
302, 326 249, 328
(non-)universality of 231, 238-239 idealization of 106-107, 115-119, 152, 179
structure of 194, 236-244 indirectness of 8, 14-15, 121, 152
see also knowledge of the world intuitive
conditional construction 307-308, 327 see linguistic intuition
connotation 281 linguistic 8-10, 93-95, 120-122, 149, 151-
consciousness 143-145, 148, 151, 222, 254 152, 155, 159
see also linguistic knowledge (conscious deficiency of 149, 157-158, 178
vs unconscious) neurological 140-141
constituent 109, 189-190, 200-201, 204, 213, non-verbal 8
324 relevance of 103, 105-107
categorial status of 109, 186, 281-282, 323 see also adequacy (external descriptive,
complexity of 201 observational), empirical evidence
relational status of 109, 186, 283 dative shift 207, 209, 227
see also term deafness 234
constituent structure 78, 127, 195 declarative 206, 210, 215, 269-271, 283
constraint 137 deep structure 101, 129, 198-199, 229
construction grammar 260 definiteness 204-205, 276
content deixis 82, 210-213, 225, 240
see meaning deletion 300, 321
context 40, 53, 65-68, 122-124, 153, 158, 163, demon 145, 215, 254-255, 328
181, 216, 226, 228-231, 240, 247, 249-251, see also control, language processing
259, 261, 275, 283, 287, 290-291, 294, 299, (automaticity of)
312-314, 327 derivational theory of complexity 129, 131
see also setting description
contextualization 228 types of 8, 94-97, 132
see also formalization, model methodological 77, 83-84, 97

diachronic language change 13, 116 object-inherent 77, 83, 97
dialect 118, 255 see also adequacy (explanatory)
discourse 205, 211-213, 236, 238, 265-266, 268, expression 269, 273, 278, 284
277, 279, 291, 312-313, 318, 321-322, 325 see also rule (expression)
see also conversation, text expression component 189, 193, 195-197, 215,
discourse analysis 4, 263 218, 245, 282, 284, 300, 302, 324
discourse structure 4, 179, 185, 188, 261-263, see also rule (expression)
265
discourse topic
see theme factivity/counterfactivity 288-289, 291, 296,
divination 57-58 315, 325
falsification vs verification 10, 21, 98, 107
feedback 244
echo utterance 209, 216 fiction 267
elliptical construction 250 filter 186
emotion 36, 37, 38-39, 61, 176-177, 222, 242, flexion 257
268, 318 focality 201, 215, 225, 227, 328
empirical evidence 19-21, 98, 119 see also topicality/focality
see also data, adequacy (observational, focus 283, 295, 315, 331
external descriptive) contrastive 201, 217, 277, 299, 315
empirical method 8 see also topic/focus
empiricism 23-25, 137, 153, 180 formalism 91, 127, 218, 292
encyclopedia 220-221, 260, 265-266, 276, 325, formalism vs functionalism 24-25, 68-74, 78,
329 80, 89
see also knowledge of the world, universe see also functionalism
of interpretation formality 272
epistemology 47 formalization 16, 70-71, 89, 132, 184, 190, 256,
equi-NP-deletion 203, 205-206 322
error 112-118, 120, 178-179, 193, 252, 254-255, see also description, model
261 frame 192, 220, 238, 245
ethnolinguistics 57, 91 frame of reference 51-53, 58, 61, 219, 222
ethology 160, 163, 165 see also state of affairs
evaluation criterion 94-97, 128, 150, 155, 180 function 26-31, 85
see also simplicity in language research 32-33
evidentiality 188, 241, 268, 279, 283, 304-306, organic 29-33
310-311, 316 role 29-33
evolutionary theory 170-171, 181 see also language (functions of)
experience 163-166, 168, 180 function word 257, 260-261, 282
experiencer 204-206, 324 functional failure
experiencer construction 202-206 see communicative failure
expert functional grammar (FG) 32, 70, 84, 183-197,
see demon 199-202, 206-210, 213-215, 217-219, 223-
explanation 3, 83-84, 85, 98, 149, 191 227, 232, 237, 240-242, 244-245, 249, 255-
cognitive 14, 85 258, 265-266, 276-277, 280-285, 290-292,
functional 64-65, 75-81, 89, 175, 218 300, 302-303, 311, 313, 315, 322-328, 330
SUBJECT INDEX 383
functional logic 224, 226, 228, 232, 291, 302, hearer model
327 see partner model
functional procedural grammar 183, 256-285, hedging 291, 310
290-292, 300, 302-303, 311, 320-322 human sciences 26-27, 29, 32
functional sentence perspective 86, 191 hypothetico-deductive method 8-9, 15-16, 23,
functional unit of forms 200, 209-210, 298, 70, 95, 127
301, 307, 312
functionalism 26, 34, 65-66, 71-83, 89, 92, 156,
183-184, 186, 191-193, 196, 199-200, 213, iconicity 274, 313
232, 236, 292, 322, 325 idiolect 116-118
see also formalism vs functionalism idiom 186, 260
illocution 188, 190, 210, 240-241, 268-271,
281-283, 323, 327
gender 211 direct vs indirect expression of 270-271, 290
generative grammar (TG) 4, 15, 22-23, 25,
70-71, 84-85, 92, 95-96, 101-103, 105, 107- primary vs secondary expression of 270
109, 113, 119-120, 123-139, 148-152, 154- see also communicative goal, intention
156, 159, 177-179, 186, 190-195, 197, 210, (communicative)
217, 229, 323-324 illocutionary frame 188, 283, 323
generative semantics 92, 190, 238 imagery 235, 326
genetic program immediate constituent analysis 127
open vs closed 164-165, 181 imperative 87, 241, 270-271, 283
given/new 249, 265-266, 274, 276-277, 312 implication 264, 287
see also new implicature 264, 290
goal 205, 209-210, 275 short-circuited 290, 300, 316
goal-directedness 30-31, 33, 75-76, 78, 85, 89 incremental procedural grammar 324
government and binding 137 individual vs social group 11-13
grammar 4, 18-20, 23-24, 81-82, 93-98, 100- individualistic language use
101, 105, 109, 119, 121, 128-129, 174, 177, see talk-to-self
183-185, 190-193, 213, 216-217, 219, 225, induction 8-9, 69
228-229, 231, 236, 244, 256, 259, 285, 294, inferencing 219, 229, 238, 264-265, 287, 291,
299, 301-302, 313, 315, 323, 329 299
autonomous 133-135, 171, 174 see also information processing, reasoning
see also syntax (autonomous) information
directionality of 125-127, 137 see knowledge of the world
integration of 71, 84, 107, 114, 126, 130, information processing 47, 220-221, 238
184-185, 213, 219, 285, 322, 329 see also inferencing, reasoning
see also model (integration of) information structure 82, 189, 191, 193, 213,
linguist's vs child's development of 151, 217, 225, 227, 265-266, 273-274, 276-277,
156 279, 281, 283-285, 301, 307-309, 311-315,
semantics and 101, 109, 124, 191, 193, 213, 322, 325, 329
218 see also pragmatic function
grammaticalization 206, 225, 270, 306, 328 innateness 92, 150-155, 160-161, 165, 167, 179-
181
see also language (genetic basis of)
institution 109, 113-114, 120-121, 125-127, 130,

see setting (institutional) 132, 145, 147, 152-153, 157, 173, 183-
intelligence 113 185, 190, 213, 219, 244-284
intention 31, 33, 37, 51-52, 54-62, 73-74, 76, checklist view of 111
82-83, 88, 221-222, 241, 246, 256, 264, 268- modeling of 114-115, 117, 138, 244
269, 320 see also process model
communicative 240-241, 268 see also cognition, linguistic knowledge
see also communicative goal, illocution formal vs functional dimension of 69-70,
levels of 52, 241 77-82, 109, 111, 136, 183, 191, 197, 232
retrievability of 52 functional dimension of 64-65, 69-72, 74,
interaction 77-81, 91, 105, 108-109, 112, 115-116,
see communication 119, 123-125, 128, 158, 168-169, 174,
interpersonal relation 176-177, 212, 215-216, 229, 246, 292,
see social relation 294, 323
interrogative 42-43, 198, 206, 208, 210, 269- functions of 26, 33-46, 60-64, 72, 74-75,
270, 283, 306-307, 310-311, 328 85-86, 157, 196-197
intersubjectivity 49 extrinsic vs intrisic approach to 42, 82
organic 34-35, 37, 38, 39, 40, 41-46,
intonation 189-190, 215-216, 228, 251, 270, 284, 298-299, 305, 307, 309, 312, 314
60, 63, 196
introspection 5-6 role 34-35, 45-46, 60, 196
inversion clause see also communication, function,
see experiencer construction thought
genetic basis of 98, 150, 152-153, 161-165,
168-169, 180
knowledge 102-103, 179-180 see also innateness
and capacity to act 137-139, 141-143, 145- individual vs group aspects of 11-13, 85-86,
148, 179 116
see also linguistic knowledge (and individual vs social dimension of 11-13,
language use) 72
how 141-142, 148, 220 native vs foreign 144-145, 179
that 141, 148, 220 side effect of 34, 38
see also linguistic knowledge species-specificity of 171, 173-174
knowledge of the world 19, 48, 160, 192, 200, variability in 105, 115-119, 150, 153, 179,
212, 219-229, 231, 233, 237-238, 240-241, 224
245, 249, 259-269, 271-280, 291, 304, 325, language acquisition 5, 23, 91-92, 97-99, 116,
327 128, 135, 141, 143-144, 149-169, 175, 179-
situation-dependent vs independent 220- 180, 234
221, 325-326 and general cognitive development 160,
source of 268, 279, 304, 310-311, 316 167-168, 234
see also conceptualization, encyclopedia see also cognitive development
as a logical problem 151-152, 159, 180
critical period of 166-167
language speed of 157
changeability of 105, 115-116 language acquisition device (LAD) 93-94, 98,
cognitive systems of 5-11, 33, 64-65, 70, 149-152, 155-156, 159
72, 80, 91-92, 96-98, 100-101, 106-107, see also universal grammar
SUBJECT INDEX 385
language disturbance 76, 116, 162 goal of 3-5, 13-14, 94-98

language game 144 integration of 4-5, 11, 17, 25, 66, 108
language independent preferred order of object of 3-5, 13-14, 17, 77
constituents (LIPOC) 201, 210 slate of the art of 3-4, 11, 65
language interpretation 130-131, 193-194, 211, language sociology 66, 68, 91, 158
232, 236, 238, 247-252, 263-264, 327-328 language theory 3, 9, 15, 22, 26, 94, 99, 106,
interactive vs autonomist view of 247-248 108, 128, 130, 145, 191, 285
see also language processing, language adult vs child 98-99, 149, 155
production vs interpretation language user's 102, 141, 143-144, 151
language philosophy 91 language understanding
language processing 113, 128-129, 133-134, see language interpretation
136-139, 144, 148, 183-184, 191-192, 219, language universals 94, 96-97, 118-120, 150,
244, 254, 299, 328 152-153, 168, 273
automaticity of 249, 252, 254-255 formal vs substantial 119
see also control, demon parameter approach to 119, 150-151
bottom-up vs top-down 249-252 lateralization (of the brain) 166-167
flexibility of 251-253, 255 learnability 151, 154, 159
interactive 247-250, 252, 255 lexical access 244
obligatoriness of 249, 252 lexical choice 194, 228-229, 249-250, 252, 254,
phonological 249-250, 252, 254 257, 272-274, 278-280, 303, 313, 322
semantic 247-249, 254 lexical constancy principle 225, 229
situation bound 251-253, 255 lexical item 224-225, 230, 232, 235, 240, 248-
sources of information for 247-249, 252- 250, 257-258, 261, 280, 291, 298, 300-302
254 meaning definition of 186-187, 230, 239,
syntactic 247-249, 254 259-260
see also language interpretation, language phonological form of 257, 260-261, 284
production, performance lexical recognition
language production 183, 192-195, 238, 241, see lexical choice
253-284, 322, 329 lexical vs grammatical coding 225, 227
bookkeeping in 263 see also qualification (lexical vs grammati
incremental 253 cal expression of)
see also language processing lexicalization
language production vs interpretation 6-7, see verbalization
15, 49-51, 87, 100, 103, 115, 118, 125, 135, lexicon 67-68, 111, 186-187, 193-194, 196, 219,
137, 178, 193-194, 235, 244-247, 252-255, 230, 234, 236, 245, 249, 255, 257-261, 275,
324, 328 285-286, 288
see also language interpretation linguistic attitude
language psychology 16, 66, 68, 89, 91, 129- see linguistic intuition
130, 135-138, 156, 160, 185, 191-195, 219, linguistic awareness 5-6
223, 244-247, 251, 257-258, 260, 263, 323, see also linguistic intuition
326, 328 linguistic behavior
see also linguistics vs language psychology as a criterion for knowledge 138-141, 147-
language research 148
applications of 83 nature of 33-34, 100, 103, 105-124
as cognitive research 5, 13-21 linguistic community 116-118, 179, 231, 290
division of labor in 108 homogeneous 116-117
linguistic creativity 100-101, 104, 109-112, 116, relational organization of 109, 186, 189,
136, 178, 253 191-193, 323
linguistic intuition 6, 93-95, 100-101, 103, 120- see also semantic function, syntactic
124, 133-135, 179, 195, 285, 290, 313 function
ontogenesis of 121 structure of 64-65, 67-71, 78, 99, 109, 111-
uncertainty of 122 112, 119, 121, 127, 129, 133, 185, 189-
variability of 122 195, 197, 247-248, 251, 272
see also linguistic awareness, linguistic see also syntax
utterance (acceptability of, appropriateness wellformedness of 6, 121-124, 133-134, 248,
of, grammaticality of, wellformedness of) 253
linguistic knowledge 13, 17-18, 23-25, 98, 103- see also linguistic intuition
106, 112-125, 127-128, 133-148, 160, 179, see also sentence
183, 192, 244, 246-249, 254-257, 324, 328- linguistics 16, 91-92, 138, 183, 194, 236, 244,
329 259-260, 285, 322
and language use 102-104, 112-113, 134, linguistics vs language psychology 13-21, 22-23,
136-139, 141, 144-145, 147, 246 25, 92, 183, 191, 196
see also competence vs performance, see also language psychology
knowledge (and capacity to act) literature 111
changeability of 116 lying 264-265
conscious vs unconscious 143-145, 148,
151
see also consciousness mass media communication 56
declarative 192, 245 mathematics 26, 29, 32
deviating from 115 meaning 5-6, 46-47, 69, 101, 129, 192-193, 200,
procedural 192, 245 206, 210, 215, 229-230, 232-233, 236, 247,
see also competence, knowledge, language 252, 259, 293-294, 296-298, 301-302, 307,
(cognitive systems of) 314, 318, 326
linguistic relativity 231 direct vs indirect 61-62
linguistic utterance 261-263, 265-266, 268-269, literal 247
272, 277-278, 280, 282, 284, 291, 302, 322 see also linguistic utterance (mean
acceptability of 121, 133-134, 179, 313 ingfulness of), semantics
see also linguistic intuition meaning postulate 187
appropriateness of 6, 253 memory 113-115, 134, 139, 146, 211, 221, 255,
see also linguistic intuition 262
grammaticality of 100, 103, 120-121, 124, for verbal form 129, 229-230, 236, 238
159, 179, 253 individual differences in 113, 139
see also linguistic intuition limitations of 110-111, 114, 133, 211
layered underlying representation of 186, long-term (LTM) 113, 220
188, 196-197, 224, 237, 240-243, 283, short-term (STM) 113, 211, 220-221
304, 314, 323, 326-327 mental model 236-237, 243, 310, 327
see also qualification (hierarchy of) mentalism
markedness of 122, 124, 133, 179, 186, 227, see cognitivism
313 metaphor 111, 259, 265, 273, 282
meaningfulness of 64, 67-69, 99, 109, 119, metatheory 20, 94-95, 97
124 methodology 25
see also meaning, semantics
SUBJECT INDEX 387
mind 300-301, 313

see cognition negative raising construction 285-322
modality 224, 268, 270-271, 277, 279, 283, 285- neurolinguistics 6, 140, 162, 166
286, 288-289, 292-294, 296-298, 302-305, neurological vs cognitive systems 162-169, 174,
312-314, 316-319, 321, 329, 331 176, 181
deontic 242-243, 267-268, 277, 308 see also biological vs cognitive systems,
epistemic 242-244, 267-268, 277, 279, 318 brain vs mind
firmness of 304-305, 307, 312, 320 neurophysiology 160
(inter)subjectivity of 304-308, 310-311 new 212
subjective vs objective 188, 241-242, 303- see also given/new
311 normativity 12-13, 118
model notational variant 18-19, 190
integration of 9-10, 16, 25 see also computational equivalence
see also grammar (integration of) noun 186, 213, 240, 276, 282
meaning of 125-126, 323 number 211
scope of 16
underdetermination of 10, 95, 155, 178
see also description, formalization object (syntactic) 189, 209, 214, 216-217, 283-
modifier 284, 301
see qualification objectivity 43, 47, 49
mood 186, 279, 282-283 operator 186, 188-189, 224-225, 241, 278, 282-
morpheme 257 283, 301-303
morphology 187, 189-190, 254-255, 260, 284 illocutionary 188, 283, 323, 326
muteness 234 predicate 188, 291, 302-303
mutual knowledge 51, 192, 264-265 predication 188
proposition 188, 241
term 187, 291
necessity 289 see also qualification
negation 243, 267-268, 277, 279, 282, 285-286, order
289, 294-295, 297-299, 302-303, 307-309, see imperative
312, 314-315, 317, 327
predicate 188, 243
term 279 PI 190, 198-199, 201, 205, 208, 215-216
see also polarity parallel processing 131, 215, 247-248, 254
negative incorporation 295, 298-300, 309, 312 parenthetical 308
negative marker 285-288, 292-302, 307, 309, parser
311-316, 330 see language interpretation
explicitness of 295-298, 312, 330 particle 270, 278-279
scope of 299 modal 279, 282, 330
negative polarity item 287, 319 particle movement 206-207
negative raising 285-322, 329-331 partner model 51-56, 221-222, 256, 264, 291
and uncertainty 286, 288-289, 291-296, 301- passive construction 198-199, 202, 209-210,
302, 304-305, 309, 328 212, 215, 217, 227-228, 325
pragmatic approach to 286-288, 290-292, perception 47-48, 224, 231, 233-235, 237
298-300 see also representation (perceptual)
syntactic approach to 286-288, 290-292, performance 23, 33, 109-110, 129, 131, 133-
134, 144 303

see also competence vs performance, auxiliary 200, 208, 279
language processing modal 200, 303-304, 308-309
performative 38, 271, 281, 309, 317-318 causative 199
person 211, 309, 317-318 communication 279, 288, 316-318
personality 38-39, 57, 59, 61-62, 221-222 deontic 330
perspectivization 189, 193, 205, 212, 215, 217, embedded 202, 302
227, 283-284, 291, 301 epistemic 279, 288-289, 292-294, 296-299,
persuasion 268 301-305, 311, 314, 316-318, 321, 330
philosophy of science scalar model of 289, 294, 296-297, 299,
see theory of science 317, 320-321, 331
phonetics 6, 255, 284 see also predicate (negative raising)
phonology 254, 284 finite 208-209
interpretive 101, 125-126 infinite 208-209
phrase structure grammar 101, 178 main 199-200, 202, 208-210, 279, 292, 296,
phylogenesis of language 5, 169-177 302, 306-307
and general cognitive evolution 169-171, mental state 279, 288, 305, 317-318, 331
174, 176 de re vs de dicto 305-311, 316, 331
and non-verbal symbol systems 169-175 modality
and mutation 170-171 see predicate (epistemic)
and ontogenesis 175 negative raising 288-290, 315-318
discontinuity hypothesis of 171-172 see also predicate (epistemic)
physical sciences perception 288, 317
see positive sciences propositional attitude 279, 281
plasticity predicate formation 187, 191, 196, 258, 281,
neural 164-166, 168, 181
of linguistic behavior 76 see also rule (predicate formation)
poeticity 40-41, 88 predicate frame 187-188, 192, 194, 199-200,
polarity 243, 267-268, 281, 287, 292, 294-298, 223, 245, 253, 258-259, 273, 282
303, 305, 308-309, 312-316, 319-321 predication (process) 269, 278, 280-284, 303
see also negation predication (structure) 186-189, 194, 196-197,
politeness 53, 59, 83, 271-272, 281 209, 213, 215, 223-233, 236-237, 241, 243,
positive sciences 26, 29, 32, 72, 78-79, 237 256-257, 260-261, 269, 273, 276, 280, 282-
positivism 69, 89 284, 291, 300, 302-304, 309, 311, 313-315,
possessive 275 320-322, 324, 327-328, 330
possible world 242, 267, 292 (un)ordered nature of 188-189, 191, 195,
pragmatic function 32, 189-191, 193, 206, 208- 206
210, 213, 217, 226, 228-229, 265-266, 268, see also representation (linguistic)
274, 277, 279-280, 313, 315, 323, 325-326 preposition 210, 257, 325
see also information structure spatial 278
pragmatic universals 63 temporal 278
pragmatics 49, 65-68, 88, 184-185 presupposition 264
Prague school privative opposite 286
see functional sentence perspective procedural semantics 192
predicate 186-189, 194-195, 214, 239-240, 249, procedure 16, 116, 132, 137-138, 143-144, 178,
259, 271, 274, 278-279, 281-283, 297-298, 184, 192, 215, 228, 235, 245, 253-254, 256,
SUBJECT INDEX 389
262, 269, 272-274, 280-281, 284, 303 quotation 268, 279, 283, 314
intra-level- 215 see also conceptualization (embedding of)
super-level- 215-218
syntactic 213, 215
see also rule rationalism 23-25, 180
process model 8-11, 15, 17, 22-23, 64, 74, 106, realis/irrealis 242, 267-268, 277-279, 282
120, 131, 196, 244-245, 272 reality
see also language (cognitive systems of extra-subjective vs intra-subjective 37-38,
(modeling of)) 43, 45, 47-48, 87, 327
processing capacity interpretation of 47-51, 87
limitations of 110, 113 see also state of affairs, universe of
processing load 19, 129 interpretation
prominence 204-205, 281, 307, 309, 312 reasoning 50, 219, 221, 223-224, 228, 232-233,
promising 201-202 256-257, 264-265, 302, 327, 329
pronominalization 209 see also inferencing, information proces
sentence 320-321 sing, thought
pronoun 211, 213, 276, 278, 282, 320-321, 324 recipient 209, 227, 275, 281
proposition 188 recursivity 110-112, 114, 178, 187
propositional attitude 188, 279, 281, 292-299, reference 37, 43, 45, 86, 187, 211-212, 232,
302-303, 311-312, 314, 330 260
propositional attitude construction 292, 298- reflection 120-121, 134, 144
303, 305, 309, 311-313, 315, 319, 331 reflexive 203, 212
prosody reflexive construction 203, 205-206
see intonation register 272, 281
pseudo-cleft 227, 314-315 relational grammar 191, 202-206, 284
psycholinguistics relativism 21
see language psychology relevance 51, 221, 264, 325-327
reported speech 317-318
representation 16, 70, 129, 132, 137-139, 143-
qualification 196, 224-225, 228, 240-244, 266- 144, 150, 190, 196-197, 219-220, 223-241,
268, 270-271, 273, 277-279, 281-283, 292, 244, 273, 280, 285-287, 298-299, 302-303,
294, 296, 302-303, 314, 327 311, 313-315, 322, 325-327, 329
hierarchy of 237, 242-244, 267, 303, 327- conceptual
328 see conceptualization
see also linguistic utterance (layered formal
underlying representation of) see formalization
lexical vs grammatical expression of 225, linguistic 223-224, 229-236, 240-241
273, 302-303 see also predication
see also lexical vs grammatical coding logical 236
scope of 243-244, 267, 277-278, 283, 303, olfactory 235
312, 314-315, 321 perceptual 224, 232-235
see also operator, satellite visual 234-235, 327
quantifier 208, 215-216, 218 request 241, 271, 290
quantifier floating 208, 213-216, 218, 291 responsibility 57, 88
question retardation 166
see interrogative rhematicity 268
see also thematicity/rhematicity semantic transparancy

rhetorical structure theory 273 see semantic recoverability
right dislocation 227 semantics 65, 67-70, 88, 101, 129, 189, 192,
ritual speech 57-58 211, 223, 233, 247-248, 285-286, 288, 317,
role and reference grammar 196 331
rule 18-20, 24, 70, 111-112, 116, 119, 128, 132, formal 232-233
136-139, 143-146, 148, 150, 178, 184, 235, functional 69
245 interpretive 70, 101, 109, 125-126, 192,
expression 189, 227, 291
ordering of 131, 213-218 objectivist 232
placement 189, 195, 206, 209-210, 213-218, truth-conditional 69, 286
291, 300 see also linguistic utterance (meaningful-
predicate formation 187, 196, 199-200, 281 ness of), meaning
rewrite 101 semiotics 66-67
term formation 187 sentence 280
transformation 101, 125-127, 129, 137, 186, complex 212, 214, 279-281
195-196, 206, 213, 225, 286, 290-291- coordinate 274, 280, 320-321
292, 324 embedded 198-199, 201, 203, 211-214, 216-
see also procedure 218, 269, 274, 280, 285-288, 292-293,
295, 297-302, 306, 311-315, 317, 319-
320, 324, 329
saliency 216, 222, 225, 240, 265, 266, 274, 277, finite vs infinite 198-199, 201
283, 309, 312 main 203, 211, 214, 216-218, 269, 274, 280,
satellite 188-189, 210, 224-225, 241, 273-274, 285-287, 294-302, 305, 307, 309, 311-
278, 281-283, 328 317, 319-320, 324
clause 188 relative 280
illocutionary 188, 327 simple 213, 309
predicate 188, 327 see also linguistic utterance
predication 188, 327 sentence connector 257, 274
proposition 188, 241 sentence type 270, 282-283
see also qualification sentencing 269, 272-280
scene 220 setting 53-60, 63, 82-83, 220-221, 241, 246,
scientism 78 256, 264, 268-269, 271-272
script 192, 220, 238 institutional 54, 58
selection restriction 187, 193-194, 258-259 see also context
semantic cohesion 209 silence 59
semantic decomposition 186, 238 simplicity 18-19, 96, 128, 150, 155-156, 180,
semantic feature 238 259, 329
semantic function 32, 187, 189, 191, 193, 200, see also evaluation criterion
202, 204, 206, 275, 282, 323-324 singular state of affairs (SSA) 269-284, 291-
see also case (semantic), linguistic 292, 302, 311-315, 317-318, 320-321
utterance (relational organization of) ordering of 273-274, 280
semantic function hierarchy 276, 283 relational structure of 274-276
semantic network 236-238, 275 situational network (SN) 261-269, 272-279,
semantic recoverabiliry 211, 215, 216 283-284, 291
filtering of 264-265
SUBJECT INDEX 391
small-talk 59-60, 74 surface structure 101, 129, 189, 193, 195, 208,
smell 233, 235 228, 231, 236, 257, 269, 302, 326
see also representation (olfactory) switch-reference 205
social relation 42, 53, 55-63, 82-83, 87, 219, synonymy 100-101, 120-121, 124, 228, 286
221, 241, 246, 256, 264, 268-269, 271-272 syntactic construction 228, 272
social role 53-54, 158 syntactic correspondence 280
socialization 157 syntactic domain 213-218
sociolect 116-118 syntactic function 32, 189, 191, 193, 202, 204-
sociolinguistics 207, 210, 214-215, 217, 227, 277, 283-284,
see language sociology 323, 325-326
source 275 see also linguistic utterance (relational
space 188, 242, 267, 278-279 organization of)
description of 274 syntax 65-70, 74, 88, 91-92, 100, 103, 105, 108-
speaker vs hearer 44, 49-51 109, 123, 173-174, 178, 183-184, 187-189,
speaker/hearer 192-194, 213, 215, 219, 230-231, 233, 247-
ideal 103, 116-118 249, 254-255, 259, 284-287, 291, 300, 303,
idiosyncratic properties of 48, 113, 116-118, 319-322
139, 144, 231, 238, 272 autonomous 78, 101, 109, 123, 125-126,
momentary condition of 61, 103, 106, 111, 158-159, 183, 218
113-114, 122, 221-222 see also grammar (autonomous)
prototypical 117 see also linguistic utterance (structure of),
speaker attitude 37, 45, 59, 83, 222, 242, 268, word order
279, 281, 318 systemic grammar 70, 85, 89, 184
speech act
see communicative act
speech act theory 4, 35-37, 57, 85-86, 270 tag question 287, 319-320
standard language 12 tail 320
state of affairs (SoA) 51, 53, 55-56, 58-63, 82, talk-to-self 62-63, 73
188-189, 192-197, 205, 212, 219, 221-228, taxonomy 8-9, 17, 22, 95-96, 126, 130-132
237, 240-244, 246, 256, 261-262, 264-269, telephone conversation 54-55
272, 277-279, 291-294, 296-299, 301-304, tense 186, 188, 201-202, 225, 278-279, 282-283,
306-310, 314, 316, 318-319, 328 309, 317-318
see also frame of reference, reality term 186-190, 213-218, 223, 232, 243, 259,
statement 276, 279, 281-283, 326
see declarative see also constituent
story 274 term formation 187, 258
see also text text 42, 44, 82, 236, 262
story grammar 179 macro-structure of 263
structuralism 9, 69-70, 127 see also discourse, story
style 61, 88, 116, 118-119, 158, 230, 272 text linguistics 4
subject 189, 193-195, 202-206, 212-214, 216- textualizing 262-269
217, 256, 283, 324 thematicity 283
subject/object raising 214-215, 217-218, 291, thematicity/rhematicity 266, 274, 312
301 see also rhematicity
subjectivity 47-49, 87 theme 37, 41, 58-59, 82, 262-263
subordinator 216 theme/rheme 265-266, 276-277
theory of science 75, 85 273, 277, 280-282, 328

thought 35, 64, 109, 178 verification
hierarchical nature of 60, 256, 327 see falsification vs verification
see also reasoning vision 233-235
time 188, 224-225, 242-243, 267-268, 274, 278- see also representation (visual)
279, 282, 309-310
topic 193, 283
topic/focus 189, 276-277 warning 271
see also focus wh-question 214, 216
topicality 204-205, 210, 283 wh-word 216-217
topicality/focality 209, 215-217, 249, 276-277, what to say vs how to say it 257
299, 301, 312, 325 word order 189, 193, 195, 200-201, 206-218,
see also focality 227-228, 246, 284, 300-301, 303, 324-325,
transformation 330
see rule (transformation) see also syntax
transformational generative grammar word order template 189-190, 192, 206-210,
see generative grammar 245, 324-325
translation 224, 230-231
truth 286-287, 298-299, 304, 318
analytical 20-21
scientific 19-22, 84
turn 250, 261-266, 268, 272, 276-277
turn-taking 280
turn type 268-269
typology 119, 183, 206
see also adequacy (typological)
universal grammar (UG) 92-93, 97-98, 119,

124, 150-152, 154-156
see also language acquisition device
universals
see communication (universals of),
language universals, pragmatic universals
universe of interpretation 51, 221, 260, 262-
268, 274, 276-277, 281, 291, 319, 329
see also encyclopedia, reality
valency
see argument structure
verb
see predicate
verb agreement 202-203, 205, 253, 324
verbal particle 209-210
verbalization 196, 254, 257, 261-266, 269, 272-
Index of Names
Abelson, R. 180, 192, 220 Bolkestein, A M . 185, 217, 227-228, 283, 291,
Allen, R. 157 323, 325
Altman, J. 174 Bossuyt, A. 283
Arbib, M.A. 247 Bouma, S. 215
Argyle, M. 53, 84, 235 Bransford, J.D. 129, 229
Asher, R. 172 Bredart, S. 236
Austin, J.L. 35, 37 Brennenstuhl, W. 86
Auwera, J.Van der 323 Bretherton, I. 158
Ayala, F. 76-77, 80, 171 Brewer, W.F. 229
Brown, E.K. 115
Brown, P. 53
Bach, K. 86 Brugmann, K. 68-69
Baker, C. 180 Bruin, J.De 251-252
Baker, W. 127-128 Bruner, J.S. 156-158
Ballmer, T.T. 86 Bhler, K. 39-43, 46, 49, 53-54, 82, 86
Barclay, R J . 129, 229 Busse, W. 40-41, 86
Bartholomot, J. 55 Butterworth, B. 84, 194
Bartsch, R. 286, 290 Buuren, L.Van 284
Basso, K. 59
Bates, E. 156, 158, 160, 168-169, 191, 229
Bayer, S. 287, 290, 300 Cairns, C.C. 129
Beattie, G.W. 84 Cairns, H.S. 129
Beaugrande, R.-A.De 263 Camaioni, L. 158
Benigni, L. 58 Campbell, B. 173, 175
Bense, E. 180 Campbell, R.N. 158
Bernstein, B. 61, 158 Caplan, D. 180
Berwick, R. 15, 131, 248 Carlson, T. 55
Bever, T.G. 123, 129-131, 133-135, 160, 251 Carnap, R. 67
Bezooijen, R.Van 48 Catlin, J. 132, 153, 181
Bierwisch, M. 131 Cattell, R. 317, 319
Birnbaum, L. 88, 192, 247-248 Cellrier, G. 153
Black, M. 180 Chafe, W.L. 127, 235, 237-238, 266, 268-269,
Black, M. 22, 131 273-274, 276, 326
Blauw, A.De 263, 271 Changeux, J.-P. 163, 165, 181
Block, N. 84, 233, 235, 326 Charniak, E. 25
Boas, F. 69 Chiat, S. 22, 131
Bock, J.K. 193, 195, 209, 276 Chomsky, N. 1, 4-5, 8, 10, 15, 17, 21-24, 35,
Bois, J.Du 57 62, 70, 72-74, 77-79, 84, 91-112, 114-122,
Bolinger, D. 84, 229, 284 124-130, 132-133, 135-147, 149-161, 163,
165, 169, 171, 173-174, 177-181, 183, 186,

190-192, 218, 245, 247 Eccles, J. 8
Cicero 41 Ehlich, K. 54
Clark, E. 84-85, 153, 156-157, 160, 167, 172, Engelkamp, J. 191, 276
191, 247, 251, 326 Epstein, S. 286
Qark, H. 52, 55-56, 84-85, 102, 121, 153, 156-157, Erman, L. 247
167, 172, 191, 196, 247, 251, 264-265, 268, Erteschik-Shir, N. 209
276, 326 Evans, D. 263
Clyne, M.G. 55
Cohen, L.J. 152
Cole, P. 324 Fennell, R. 247
Collas, J. 230 Ferguson, C. 157
Collier, R. 280 Fillmore, CJ. 10, 61, 199, 260, 275, 278, 286
Comb, H. 323 Finch, A. 246
Comrie, B. 202, 259, 278, 283 Firth, J.R. 62, 69
Connolly, J. 322 Fisher, S. 54
Cooper, D.E. 102, 141, 180 Flores d'Arcais, G.B. 130, 160, 247, 328
Cooper, W.E. 7 Fodor, J.A 8, 102, 129-131, 133, 233
Copi, I.M. 35-36, 40, 58 Foley, W.A 75, 196, 227, 242, 276, 280
Creider, C.A. 209 Forster, K.I. 247
Cummins, R. 30, 33, 75-76 Fortescue, M. 263
Curtiss, S. 166 Fouts, R. 173
Cuyckens, H. 278 Franck, D. 46, 49, 52, SSy 262-263
Franks, J.J. 129, 229
Fraser, B. 157
Dale, R. 257 Frazier, L. 245
Danes, F. 86 Frijda, N.H. 327
Danto, A. 180 Fromkin, V.A 178
Darwin, C 170-171, 181
Davey, A.C. 16, 85, 89
Davies, W.D. 202-204, 324 Galilei, G. 78-79
Dechert, H.W. 178 Gardner, B. 173
Dempe, H. 86 Gardner, R. 173
Derwing, B.L. 102, 108, 127-128, 149 Garrett, M.F. 102, 129-131, 133, 178, 193-195,
Descartes, R. 7, 180 244, 247, 257, 261
Dijk, T.Van 48, 61, 262-263 Garrod, S.C 211
Dik, S.C. 2, 32, 35-37, 40-41, 49, 51, 70, 77, Gebruers, R. 204, 324
79-82, 84, 86, 88, 119, 183-190, 192-194, Geest, T.Van der 44-45, 82, 156
197-198, 201-202, 204, 206, 210, 214, 217, Gelman, R. 157-158
223-225, 227-228, 230-240, 245, 258-259, Gent, J.Van 123
263, 271, 276, 281, 283-284, 302, 315, 322-328 Geysen, R. 47
Dingwall, W.O. 6, 162, 164, 172-175 Gidlund, A. 229-230
Dobzhansky, T. 76-77, 80, 171 Gillis, S. 48, 160
Dresher, B.E. 89, 91 Givn, T. 47, 172, 209, 267, 277, 329
Dubber, C 263, 271 Gleitman, H. 121-122
Duranti, A. 57 Gleitman, L. 121-122
INDEX OF NAMES 395
Goldman, N. 194-195, 237, 239, 245, 247, 260 Hjelmquist, E. 179, 229-230, 326
Goodenough, D. 263 Hobbs, J. 263
Goodman, N. 159-160 Hockett, C F . 172
Goodwin, M.H. 84 Hoekstra, T. 190-191, 323
Goossens, L. 323 Hoenkamp, E. 193-195, 253, 324
Gould, S.J. 171 Hook, S. 179
Greenberg, J.H. 4, 26-27, 29-31, 78-79, 152, Hoppe, R.A. 328
168, 173-174 Horlek, K. 86
Greene, J. 129-130 Hrmann, H. 91, 247, 252
Grice, H.P. 51-52, 85, 264 Horn, L 267, 286-290, 294, 296-300, 313, 316-317,
Groot, C.De 187, 217, 322-323 320, 330-331
Groot, W.De 204, 324 Hornstein, N. 89, 91
Gumperz, J.J. 61 Hovy, E. 329
Gurney, R. 162, 165 Huijbers, P. 257
Gvozdanovic, J. 323 Hulst, H.Van der 323
Hurtig, R. 179
Hyams, N. 152, 156
Haans, A. 123 Hymes, D.H. 35-37, 40-41, 53, 102
Haiman, J. 274
Halliday, M.AK. 40-45, 70, 81-82, 85-86, 89,
158, 184, 196-197 Inhelder, B. 48, 84, 160
Halpern, R.N. 286
Hannay, M. 185, 266, 280
Harbert, W. 324 Jackson, H. 162
Harley, T.A. 257, 260-261 Jacobson, M. 165
Harman, G.H. 35, 64, 141 Jake, J. 324
Harnad, S.R. 172 Jakobson, R. 35-36, 39-41, 46, 53-54, 58, 86
Harnish, R. 86 Jarvella, R.J. 191, 230
Harris, A 324 Jefferson, G. 263
Harris, Z. 127 Johnson, M. 48, 87, 259
Haviland, S. 85, 102, 121, 264-265 Johnson, M.K. 173
Hawkins, J.A. 276 Johnson-Laird, P.N. 16, 71, 129-130, 132, 192,
Hay, A. 229 196, 215, 222, 229, 231, 233, 236-238, 245,
Hayes, J.R. 160 247, 250, 259, 324, 326-327
Hayes, K. 173 Jong, J.De 326
Hayes, V. 173 Jusczyk, P. 7
Hayes-Roth, B. 218
Heath, C. 84
Heller, M.S. 55 Kachru, Y. 324
Hempel, C. 75 Kaldewaij, J. 127
Hengeveld, K. 186, 188, 196, 241-243, 267, Kamp, H. 278
282-283, 303-312, 314, 323, 327-328 Karmiloff-Smith, A 84
Heringer, H.J. 6 Katz, J.J. 9, 17-21, 25, 84, 98, 156, 180, 192
Hermon, G. 324 Kay, P. 260
Herrmann, T. 16 Keenan, E.L. 204
Hesp, C. 229 Kempen, G. 191, 193-195, 253-255, 257, 260,
284, 324 Lieven, E. 152, 158

Kempier, D. 61 Lightfoot, D. 91, 98, 107, 133, 151-152, 161,
Kenny, A. 138, 162 177
Kerke, S. Van de 204, 324 Linde, C. 274
Kess, J.F. 328 Linden, E. 173
Kintsch, W. 15-16, 25, 102, 229, 262-263, 326 Lindholm, J.M. 320
Kiparsky, C. 288 Lindsay, P.H. 165
Kiparsky, P. 288 Longuet-Higgins, H.C. 102
Kishna, S. 324 Luria, A.R. 6, 154, 167
Klaiman, M.H. 204, 324 Lutz, L. 265
Klein, W. 274, 278 Lyons, J. 160, 196, 283, 304, 308, 329
Knaap, R. 204, 324
Knig, E. 278
Kooij, J. 35-37, 40-41 Mackenzie, J.L. 191, 323
Korst, B.Van der 230, 322 Maclay, H. 178
Kosslyn, S. 326 MacWhinney, B. 156, 191
Koster, J. 72, 75, 78, 89, 91, 107, 171, 190-191, Maes, F. 212-213, 276
199-201 Malinowski, B. 28, 59
Kreckel, M. 52 Mandier, G. 26
Kuhn, T. 85, 89 Mandier, J.M. 179, 192, 220, 263
Kwee 322 Mann, W.C. 273
Marcus, M.P. 247-248
Marin, O.S.M. 6, 260
Labov, W. 61, 80, 116, 123, 179, 274 Marr, D. 132
Lakatos, I. 85 Marshall, C. 52, 276
Lakoff, G. 48, 87, 92, 102, 105, 107, 122, 127, Marshall, J.C. 25, 131, 171, 180
177, 179, 233, 259 Marslen-Wilson, W.D. 130, 134, 247-252
Lakoff, R.T. 53, 287, 317, 319 Masling, M. 229
Lamb, S.M. 42 Maturana, H.R. 163
Lamendella, J.T. 164-170, 175, 179 Mayr, E. 76, 89, 164-165, 170-172
Lancaster, J. 172 McCarthy, J. 180
Langacker, R.W. 260 McCawley, J.D. 110, 124, 127-128, 149, 155
Lasnik, H. 154-155, 180, 186 McCawley, N. 204
Lehnen, W.G. 16, 263 McNeill, D. 85, 156-157
Lehrman, D.S. 161-162, 165 Mehler, J. 7, 129
Lenneberg, E.H. 162-163, 165-167, 171-172, Meijer, G. 121, 123
174, 180-181, 234 Mendel, G. 171
Leodolter, R. 54 Menzel, E.W. 173
Lesser, V.R. 247 Mey, J. SS
Leuninger, H. 72 Meyers, A. 123
Levelt, W.J.M. 15, 23, 114, 123, 129-130, 132, Miller, G.A. 92, 127, 130, 162, 177, 192, 211,
149, 157, 159-160, 177, 180, 195, 215, 244, 231, 233, 259
253, 257, 259, 274, 284, 328 Miller, J.R. 15-16, 25
Levinson, S. 49, 53, 65-68, 88, 263-264 Miller, P. 89, 190-192, 206-210, 212-215, 217-218,
Levy, L. 55 323
Lieberman, P. 174-175, 181 Minsky, M. 84, 220
INDEX OF NAMES 397
Monod, J. 161, 167, 170-172 Poison, P.G. 15-16, 25

Moortgat, M. 323 Popper, K.R. 8, 10, 72, 107, 171
Morgan, J.L. 290 Postal, P. 192
Morris, C. 66-67 Power, R. 263
Morton, J. 7, 8, 157, 160 Premack, D. 173
Moutaouakil, A. 326 Prince, E.F. 266, 286-287
Muller, H. 171 Putnam, H. 8, 151, 154-155, 157, 161
Muysken, P. 190-191 Pylyshyn, Z.W. 8, 15, 22, 102, 120, 132
Nagel, E. 26-31, 75-77, 85 Quillian, M.R. 236

Neely, R. 247 Quine, W.V.O. 161
Negus, V. 174
Neumaier, O. 102, 141, 149, 178, 180
Newmeyer, F J . 72, 80-81, 177 Radcliffe-Brown, A 29
Nichols, J. 205, 268, 324 Rappoport, G. 205, 324
Nilsen, D.L.F. 275 Raupach, M. 178
Ninio, A. 89, 156 Redder, A 54
Nobach, H. 204, 324 Reddy, D.R. 247
Norman, D. 165 Rehbein, J. 54
Nuyts, J. 2, 5-6, 14, 15, 34, 47, 58, 84, 88-89, Reichling, A 4, 43
121, 128, 190, 204, 222, 284, 323-325, 329 Richards, I.A 35, 37, 43, 45
Rieger, C 237, 239, 245, 247, 260
Riemsdijk, H.Van 190
O'Connor, M. 260 Riesbeck, CK. 237, 239, 245, 247, 260
Ochs, E. 157 Rijkhoff, J. 243
Oehlen, M. 263, 271 Rijksbaron, A. 323
Ogden, CK. 35, 37, 43, 45 Risselada, R. 227-228
Osgood, C E . 84, 178, 266, 274, 323, 326, 328 Robinson, W.P. 35, 38, 40-41, 88, 157
Ostiguy, L. 55 Rosch, E. 231
stman, J.-O. 54 Ross, J.R. 122
Ruder, K F . 246
Rumelhart, D.E. 263
Palmer, F.R. 279, 329 Rycker, T.De 54
Palmer, S.E. 16 Ryle, G. 7, 14, 141-142
Papert, S. 154
Parret, H. 23-24, 49-51, 66, 85, 87, 102, 121,
155, 180 Sachs, J. 229
Perlmutter, D.M. 191, 202, 206, 284, 324 Sacks, H. 263
Petfi, J.S. 4 Saffran, E.M. 260
Piaget, J. 47-48, 154, 160, 181, 326 Samuelsdorff, P. 194
Piattelli-Palmarini, M. 154, 161 Sanford, A 211
Pinker, S. 152 Sapir, E. 4, 35, 54, 59, 69
Pinkster, H. 325 Saussure, F.De 232
Plato 17, 20, 171 Saville-Troike, M. 59
Piotkin, H.C 171 Schank, R. 88-89, 151, 192, 220, 237, 239, 245,
247-249, 260, 326, 329 Thompson, D.W. 168

Schegloff, E.A. 54, 263 Thompson, S.A. 273
Schieffelin, B. 157 Timberlake, A. 205, 324
Schlesinger, I.M. 16, 85, 102, 108, 122, 149, Todd, A. 54
153, 156, 191, 194-195, 247, 259, 328 Tombe, L.Des 8, 102, 129-131, 133
Schnelle, H. 6-7 Toulmin, S. 3-4, 76, 85, 87, 168-171
Schreuder, R. 130, 247, 328 Traugott, E.C. 278
Schriefers, H. 257 Trower, P. 84, 235
Schutter, G.De 2, 200, 202, 205, 275, 278-279, Tyler, L.K. 247-252
283, 323-325, 329
Schwartz, M. 260
Schwartz, R. 141 Ullmer-Ehrich, V. 274, 278
Searle, J.R. 7, 35-37, 52, 57, 62-63, 72, 77, 86,
88
Settekorn, W. 54 Vachek, J. 86
Seuren, P.A.M. 6, 98, 121, 125-128, 290 Valentine, J. 76-77, 80, 171
Shanon, B. 326-327 Valian, V. 15, 23, 102, 131
Shatz, M. 157-158, 160 Valin, R.D.Van 75, 196, 227, 242, 276, 280
Shibatani, M. 324 Verschueren, J. 11, 13, 25, 65-66, 86
Shipley, E. 121 Vester, E. 185, 188, 280, 283, 327
Siewierska, A. 185 Vet, C. 188, 278, 283, 323, 327
Simmons, R. 245 Villiers, J.G.De 121
Skinner, B.F. 5, 84, 128 Villiers, P.A.De 121
Slocum, J. 245 Volterra, V. 158
Smedt, K.De 253 Vries, H.De 170
Smith, A. 162
Smith, N.V. 52
Snow, C. 89, 121, 123, 156-157 Wales, R.J. 158, 160
Sperber, D. 49 Wang, W. 61
Sridhar, S.N. 324 Waterson, N. 157
Stabler, E.P. 132 Watt, W.C. 25, 128, 130
Stebbins, G.L. 76-77, 80, 170-171 Watters, J.R. 194
Stechow, A.Von 278 Watzlawick, P. 48
Steedman, M.J. 16, 71 Weigand, H. 322
Steen, G. 259 Weijdema, W. 263, 271
Steinberg, D.D. 127, 326 Weinberg, A. 15, 131, 248
Steklis, H. 172 Weiner, S.L. 263
Stemmer, N. 102 Wells, C G . 157
Stenning, K. 211 Wells, R. 141, 180
Stich, S.P. 99 Wexler, K. 154
Straight, S. 6 Whitaker, H.A. 102, 161-162, 165-167, 169
Strawson, P.F. 52 Whorf, B.L. 231
Wielinga, B. 251-252
Wierzbicka, A. 237
Tannen, D. 53, 59, 61 Wiese, B. 211
Tanner, N. 173 Wiese, R. 130, 178
INDEX OF NAMES 399
Wijk, C.Van 284

Wilensky, R. 89, 151, 263
Wilks, Y. 192, 220
Williams, E. 154
Wilson, D. 49
Winograd, T. 16, 85, 89, 192, 236, 327
Winston, P.H. 113
Wittgenstein, L. 48, 87, 236
Woods, W.A. 132, 247
Woolard, K.A. 55
Wright, L. 76
Wunderlich, D. 88, 278-279
Wygotski, L.S. 160, 326
Zaefferer, D. 86
Zangwill, O.L. 162
Zihlman, A. 173
Zurif, E. 7
Pragmatics & Beyond New Series
A complete list of titles in this series can be found on www.benjamins.com
195 Pahta, Pivi, Minna Nevala, Arja Nurmi and Minna Palander-Collin (eds.): Social Roles and
Language Practices in Late Modern English. viii,227pp.+index. Expected June 2010
194 Khnlein, Peter, Anton Benz and Candace L. Sidner (eds.): Constraints in Discourse 2.
v,178pp.+index. Expected April 2010
193 Suomela-Salmi, Eija and Fred Dervin (eds.): Cross-Linguistic and Cross-Cultural Perspectives on
Academic Discourse. 2009. vi,299pp.
192 Filipi, Anna: Toddler and Parent Interaction. The organisation of gaze, pointing and vocalisation. 2009.
xiii,268pp.
191 Ogiermann, Eva: On Apologising in Negative and Positive Politeness Cultures. 2009. x,296pp.
190 Finch, Jason, Martin Gill, Anthony Johnson, Iris Lindahl-Raittila, Inna Lindgren,
Tuija Virtanen and Brita Wrvik (eds.): Humane Readings. Essays on literary mediation and
communication in honour of Roger D. Sell. 2009. xi,160pp.
189 Peikola, Matti, Janne Skaffari and Sanna-Kaisa Tanskanen (eds.): Instructional Writing in
English. Studies in honour of Risto Hiltunen. 2009. xiii,240pp.
188 Giltrow, Janet and Dieter Stein (eds.): Genres in the Internet. Issues in the theory of genre. 2009.
ix,294pp.
187 Jucker, Andreas H. (ed.): Early Modern English News Discourse. Newspapers, pamphlets and scientific
news discourse. 2009. vii,227pp.
186 Callies, Marcus: Information Highlighting in Advanced Learner English. The syntaxpragmatics
interface in second language acquisition. 2009. xviii,293pp.
185 Mazzon, Gabriella: Interactive Dialogue Sequences in Middle English Drama. 2009. ix,228pp.
184 Stenstrm, Anna-Brita and Annette Myre Jrgensen (eds.): Youngspeak in a Multilingual
Perspective. 2009. vi,206pp.
183 Nurmi, Arja, Minna Nevala and Minna Palander-Collin (eds.): The Language of Daily Life in
England (14001800). 2009. vii,312pp.
182 Norrick, Neal R. and Delia Chiaro (eds.): Humor in Interaction. 2009. xvii,238pp.
181 Maschler, Yael: Metalanguage in Interaction. Hebrew discourse markers. 2009. xvi,258pp.
180 Jones, Kimberly and Tsuyoshi Ono (eds.): Style Shifting in Japanese. 2008. vii,335pp.
179 Simes Lucas Freitas, Elsa: Taboo in Advertising. 2008. xix,214pp.
178 Schneider, Klaus P. and Anne Barron (eds.): Variational Pragmatics. A focus on regional varieties in
pluricentric languages. 2008. vii,371pp.
177 Rue, Yong-Ju and Grace Zhang: Request Strategies. A comparative study in Mandarin Chinese and
Korean. 2008. xv,320pp.
176 Jucker, Andreas H. and Irma Taavitsainen (eds.): Speech Acts in the History of English. 2008.
viii,318pp.
175 Gmez Gonzlez, Mara de los ngeles, J. Lachlan Mackenzie and Elsa M. Gonzlez
lvarez (eds.): Languages and Cultures in Contrast and Comparison. 2008. xxii,364pp.
174 Heyd, Theresa: Email Hoaxes. Form, function, genre ecology. 2008. vii,239pp.
173 Zanotto, Mara Sophia, Lynne Cameron and Marilda C. Cavalcanti (eds.): Confronting
Metaphor in Use. An applied linguistic approach. 2008. vii,315pp.
172 Benz, Anton and Peter Khnlein (eds.): Constraints in Discourse. 2008. vii,292pp.
171 Flix-Brasdefer, J. Csar: Politeness in Mexico and the United States. A contrastive study of the
realization and perception of refusals. 2008. xiv,195pp.
170 Oakley, Todd and Anders Hougaard (eds.): Mental Spaces in Discourse and Interaction. 2008.
vi,262pp.
169 Connor, Ulla, Ed Nagelhout and William Rozycki (eds.): Contrastive Rhetoric. Reaching to
intercultural rhetoric. 2008. viii,324pp.
168 Proost, Kristel: Conceptual Structure in Lexical Items. The lexicalisation of communication concepts in
English, German and Dutch. 2007. xii,304pp.
167 Bousfield, Derek: Impoliteness in Interaction. 2008. xiii,281pp.
166 Nakane, Ikuko: Silence in Intercultural Communication. Perceptions and performance. 2007. xii,240pp.
165 Bublitz, Wolfram and Axel Hbler (eds.): Metapragmatics in Use. 2007. viii,301pp.
164 Englebretson, Robert (ed.): Stancetaking in Discourse. Subjectivity, evaluation, interaction. 2007.
viii,323pp.
163 Lytra, Vally: Play Frames and Social Identities. Contact encounters in a Greek primary school. 2007.
xii,300pp.
162 Fetzer, Anita (ed.): Context and Appropriateness. Micro meets macro. 2007. vi,265pp.
161 Celle, Agns and Ruth Huart (eds.): Connectives as Discourse Landmarks. 2007. viii,212pp.
160 Fetzer, Anita and Gerda Eva Lauerbach (eds.): Political Discourse in the Media. Cross-cultural
perspectives. 2007. viii,379pp.
159 Maynard, Senko K.: Linguistic Creativity in Japanese Discourse. Exploring the multiplicity of self,
perspective, and voice. 2007. xvi,356pp.
158 Walker, Terry: Thou and You in Early Modern English Dialogues. Trials, Depositions, and Drama
Comedy. 2007. xx,339pp.
157 Crawford Camiciottoli, Belinda: The Language of Business Studies Lectures. A corpus-assisted
analysis. 2007. xvi,236pp.
156 Vega Moreno, Rosa E.: Creativity and Convention. The pragmatics of everyday figurative speech. 2007.
xii,249pp.
155 Hedberg, Nancy and Ron Zacharski (eds.): The GrammarPragmatics Interface. Essays in honor of
Jeanette K. Gundel. 2007. viii,345pp.
154 Hbler, Axel: The Nonverbal Shift in Early Modern English Conversation. 2007. x,281pp.
153 Arnovick, Leslie K.: Written Reliquaries. The resonance of orality in medieval English texts. 2006.
xii,292pp.
152 Warren, Martin: Features of Naturalness in Conversation. 2006. x,272pp.
151 Suzuki, Satoko (ed.): Emotive Communication in Japanese. 2006. x,234pp.
150 Busse, Beatrix: Vocative Constructions in the Language of Shakespeare. 2006. xviii,525pp.
149 Locher, Miriam A.: Advice Online. Advice-giving in an American Internet health column. 2006.
xvi,277pp.
148 Flttum, Kjersti, Trine Dahl and Torodd Kinn: Academic Voices. Across languages and disciplines.
2006. x,309pp.
147 Hinrichs, Lars: Codeswitching on the Web. English and Jamaican Creole in e-mail communication.
2006. x,302pp.
146 Tanskanen, Sanna-Kaisa: Collaborating towards Coherence. Lexical cohesion in English discourse.
2006. ix,192pp.
145 Kurhila, Salla: Second Language Interaction. 2006. vii,257pp.
144 Bhrig, Kristin and Jan D. ten Thije (eds.): Beyond Misunderstanding. Linguistic analyses of
intercultural communication. 2006. vi,339pp.
143 Baker, Carolyn, Michael Emmison and Alan Firth (eds.): Calling for Help. Language and social
interaction in telephone helplines. 2005. xviii,352pp.
142 Sidnell, Jack: Talk and Practical Epistemology. The social life of knowledge in a Caribbean community.
2005. xvi,255pp.
141 Zhu, Yunxia: Written Communication across Cultures. A sociocognitive perspective on business genres.
2005. xviii,216pp.
140 Butler, Christopher S., Mara de los ngeles Gmez Gonzlez and Susana M. Doval-Surez
(eds.): The Dynamics of Language Use. Functional and contrastive perspectives. 2005. xvi,413pp.
139 Lakoff, Robin T. and Sachiko Ide (eds.): Broadening the Horizon of Linguistic Politeness. 2005.
xii,342pp.
138 Mller, Simone: Discourse Markers in Native and Non-native English Discourse. 2005. xviii,290pp.
137 Morita, Emi: Negotiation of Contingent Talk. The Japanese interactional particles ne and sa. 2005.
xvi,240pp.
136 Sassen, Claudia: Linguistic Dimensions of Crisis Talk. Formalising structures in a controlled language.
2005. ix,230pp.
135 Archer, Dawn: Questions and Answers in the English Courtroom (16401760). A sociopragmatic
analysis. 2005. xiv,374pp.
134 Skaffari, Janne, Matti Peikola, Ruth Carroll, Risto Hiltunen and Brita Wrvik (eds.):
Opening Windows on Texts and Discourses of the Past. 2005. x,418pp.
133 Marnette, Sophie: Speech and Thought Presentation in French. Concepts and strategies. 2005.
xiv,379pp.
132 Onodera, Noriko O.: Japanese Discourse Markers. Synchronic and diachronic discourse analysis. 2004.
xiv,253pp.
131 Janoschka, Anja: Web Advertising. New forms of communication on the Internet. 2004. xiv,230pp.
130 Halmari, Helena and Tuija Virtanen (eds.): Persuasion Across Genres. A linguistic approach. 2005.
x,257pp.
129 Taboada, Mara Teresa: Building Coherence and Cohesion. Task-oriented dialogue in English and
Spanish. 2004. xvii,264pp.
128 Cordella, Marisa: The Dynamic Consultation. A discourse analytical study of doctorpatient
communication. 2004. xvi,254pp.
127 Brisard, Frank, Michael Meeuwis and Bart Vandenabeele (eds.): Seduction, Community,
Speech. A Festschrift for Herman Parret. 2004. vi,202pp.
126 Wu, Yian: Spatial Demonstratives in English and Chinese. Text and Cognition. 2004. xviii,236pp.
125 Lerner, Gene H. (ed.): Conversation Analysis. Studies from the first generation. 2004. x,302pp.
124 Vine, Bernadette: Getting Things Done at Work. The discourse of power in workplace interaction. 2004.
x,278pp.
123 Mrquez Reiter, Rosina and Mara Elena Placencia (eds.): Current Trends in the Pragmatics of
Spanish. 2004. xvi,383pp.
122 Gonzlez, Montserrat: Pragmatic Markers in Oral Narrative. The case of English and Catalan. 2004.
xvi,410pp.
121 Fetzer, Anita: Recontextualizing Context. Grammaticality meets appropriateness. 2004. x,272pp.
120 Aijmer, Karin and Anna-Brita Stenstrm (eds.): Discourse Patterns in Spoken and Written Corpora.
2004. viii,279pp.
119 Hiltunen, Risto and Janne Skaffari (eds.): Discourse Perspectives on English. Medieval to modern.
2003. viii,243pp.
118 Cheng, Winnie: Intercultural Conversation. 2003. xii,279pp.
117 Wu, Ruey-Jiuan Regina: Stance in Talk. A conversation analysis of Mandarin final particles. 2004.
xvi,260pp.
116 Grant, Colin B. (ed.): Rethinking Communicative Interaction. New interdisciplinary horizons. 2003.
viii,330pp.
115 Krkkinen, Elise: Epistemic Stance in English Conversation. A description of its interactional
functions, with a focus on I think. 2003. xii,213pp.
114 Khnlein, Peter, Hannes Rieser and Henk Zeevat (eds.): Perspectives on Dialogue in the New
Millennium. 2003. xii,400pp.
113 Panther, Klaus-Uwe and Linda L. Thornburg (eds.): Metonymy and Pragmatic Inferencing. 2003.
xii,285pp.
112 Lenz, Friedrich (ed.): Deictic Conceptualisation of Space, Time and Person. 2003. xiv,279pp.
111 Ensink, Titus and Christoph Sauer (eds.): Framing and Perspectivising in Discourse. 2003. viii,227pp.
110 Androutsopoulos, Jannis K. and Alexandra Georgakopoulou (eds.): Discourse
Constructions of Youth Identities. 2003. viii,343pp.
109 Mayes, Patricia: Language, Social Structure, and Culture. A genre analysis of cooking classes in Japan and
America. 2003. xiv,228pp.
108 Barron, Anne: Acquisition in Interlanguage Pragmatics. Learning how to do things with words in a study
abroad context. 2003. xviii,403pp.
107 Taavitsainen, Irma and Andreas H. Jucker (eds.): Diachronic Perspectives on Address Term
Systems. 2003. viii,446pp.
106 Busse, Ulrich: Linguistic Variation in the Shakespeare Corpus. Morpho-syntactic variability of second
person pronouns. 2002. xiv,344pp.
105 Blackwell, Sarah E.: Implicatures in Discourse. The case of Spanish NP anaphora. 2003. xvi,303pp.
104 Beeching, Kate: Gender, Politeness and Pragmatic Particles in French. 2002. x,251pp.
103 Fetzer, Anita and Christiane Meierkord (eds.): Rethinking Sequentiality. Linguistics meets
conversational interaction. 2002. vi,300pp.
102 Leafgren, John: Degrees of Explicitness. Information structure and the packaging of Bulgarian subjects
and objects. 2002. xii,252pp.
101 Luke, K. K. and Theodossia-Soula Pavlidou (eds.): Telephone Calls. Unity and diversity in conversational
structure across languages and cultures. 2002. x,295pp.
100 Jaszczolt, Katarzyna M. and Ken Turner (eds.): Meaning Through Language Contrast. Volume 2. 2003.
viii,496pp.
99 Jaszczolt, Katarzyna M. and Ken Turner (eds.): Meaning Through Language Contrast. Volume 1. 2003.
xii,388pp.
98 Duszak, Anna (ed.): Us and Others. Social identities across languages, discourses and cultures. 2002.
viii,522pp.
97 Maynard, Senko K.: Linguistic Emotivity. Centrality of place, the topic-comment dynamic, and an ideology
of pathos in Japanese discourse. 2002. xiv,481pp.
96 Haverkate, Henk: The Syntax, Semantics and Pragmatics of Spanish Mood. 2002. vi,241pp.
95 Fitzmaurice, Susan M.: The Familiar Letter in Early Modern English. A pragmatic approach. 2002.
viii,263pp.
94 McIlvenny, Paul (ed.): Talking Gender and Sexuality. 2002. x,332pp.
93 Baron, Bettina and Helga Kotthoff (eds.): Gender in Interaction. Perspectives on femininity and
masculinity in ethnography and discourse. 2002. xxiv,357pp.
92 Gardner, Rod: When Listeners Talk. Response tokens and listener stance. 2001. xxii,281pp.
91 Gross, Joan: Speaking in Other Voices. An ethnography of Walloon puppet theaters. 2001. xxviii,341pp.
90 Kenesei, Istvn and Robert M. Harnish (eds.): Perspectives on Semantics, Pragmatics, and Discourse. A
Festschrift for Ferenc Kiefer. 2001. xxii,352pp.
89 Itakura, Hiroko: Conversational Dominance and Gender. A study of Japanese speakers in first and second
language contexts. 2001. xviii,231pp.
88 Bayraktarolu, Arn and Maria Sifianou (eds.): Linguistic Politeness Across Boundaries. The case of
Greek and Turkish. 2001. xiv,439pp.
87 Mushin, Ilana: Evidentiality and Epistemological Stance. Narrative Retelling. 2001. xviii,244pp.
86 Ifantidou, Elly: Evidentials and Relevance. 2001. xii,225pp.
85 Collins, Daniel E.: Reanimated Voices. Speech reporting in a historical-pragmatic perspective. 2001.
xx,384pp.
84 Andersen, Gisle: Pragmatic Markers and Sociolinguistic Variation. A relevance-theoretic approach to the
language of adolescents. 2001. ix,352pp.
83 Mrquez Reiter, Rosina: Linguistic Politeness in Britain and Uruguay. A contrastive study of requests and
apologies. 2000. xviii,225pp.
82 Khalil, Esam N.: Grounding in English and Arabic News Discourse. 2000. x,274pp.
81 Di Luzio, Aldo, Susanne Gnthner and Franca Orletti (eds.): Culture in Communication. Analyses of
intercultural situations. 2001. xvi,341pp.
80 Ungerer, Friedrich (ed.): English Media Texts Past and Present. Language and textual structure. 2000.
xiv,286pp.
79 Andersen, Gisle and Thorstein Fretheim (eds.): Pragmatic Markers and Propositional Attitude. 2000.
viii,273pp.
78 Sell, Roger D.: Literature as Communication. The foundations of mediating criticism. 2000. xiv,348pp.
77 Vanderveken, Daniel and Susumu Kubo (eds.): Essays in Speech Act Theory. 2002. vi,328pp.
76 Matsui, Tomoko: Bridging and Relevance. 2000. xii,251pp.
75 Pilkington, Adrian: Poetic Effects. A relevance theory perspective. 2000. xiv,214pp.
74 Trosborg, Anna (ed.): Analysing Professional Genres. 2000. xvi,256pp.
73 Hester, Stephen K. and David Francis (eds.): Local Educational Order. Ethnomethodological studies of
knowledge in action. 2000. viii,326pp.
72 Marmaridou, Sophia S.A.: Pragmatic Meaning and Cognition. 2000. xii,322pp.
71 Gmez Gonzlez, Mara de los ngeles: The ThemeTopic Interface. Evidence from English. 2001.
xxiv,438pp.
70 Sorjonen, Marja-Leena: Responding in Conversation. A study of response particles in Finnish. 2001.
x,330pp.
69 Noh, Eun-Ju: Metarepresentation. A relevance-theory approach. 2000. xii,242pp.
68 Arnovick, Leslie K.: Diachronic Pragmatics. Seven case studies in English illocutionary development. 2000.
xii,196pp.
67 Taavitsainen, Irma, Gunnel Melchers and Pivi Pahta (eds.): Writing in Nonstandard English.
2000. viii,404pp.
66 Jucker, Andreas H., Gerd Fritz and Franz Lebsanft (eds.): Historical Dialogue Analysis. 1999.
viii,478pp.
65 Cooren, Franois: The Organizing Property of Communication. 2000. xvi,272pp.
64 Svennevig, Jan: Getting Acquainted in Conversation. A study of initial interactions. 2000. x,384pp.
63 Bublitz, Wolfram, Uta Lenk and Eija Ventola (eds.): Coherence in Spoken and Written Discourse. How
to create it and how to describe it. Selected papers from the International Workshop on Coherence, Augsburg,
24-27 April 1997. 1999. xiv,300pp.
62 Tzanne, Angeliki: Talking at Cross-Purposes. The dynamics of miscommunication. 2000. xiv,263pp.
61 Mills, Margaret H. (ed.): Slavic Gender Linguistics. 1999. xviii,251pp.
60 Jacobs, Geert: Preformulating the News. An analysis of the metapragmatics of press releases. 1999.
xviii,428pp.
59 Kamio, Akio and Ken-ichi Takami (eds.): Function and Structure. In honor of Susumu Kuno. 1999.
x,398pp.
58 Rouchota, Villy and Andreas H. Jucker (eds.): Current Issues in Relevance Theory. 1998. xii,368pp.
57 Jucker, Andreas H. and Yael Ziv (eds.): Discourse Markers. Descriptions and theory. 1998. x,363pp.
56 Tanaka, Hiroko: Turn-Taking in Japanese Conversation. A Study in Grammar and Interaction. 2000.
xiv,242pp.
55 Allwood, Jens and Peter Grdenfors (eds.): Cognitive Semantics. Meaning and cognition. 1999.
x,201pp.
54 Hyland, Ken: Hedging in Scientific Research Articles. 1998. x,308pp.
53 Mosegaard Hansen, Maj-Britt: The Function of Discourse Particles. A study with special reference to
spoken standard French. 1998. xii,418pp.
52 Gillis, Steven and Annick De Houwer (eds.): The Acquisition of Dutch. With a Preface by Catherine E.
Snow. 1998. xvi,444pp.
51 Boulima, Jamila: Negotiated Interaction in Target Language Classroom Discourse. 1999. xiv,338pp.
50 Grenoble, Lenore A.: Deixis and Information Packaging in Russian Discourse. 1998. xviii,338pp.
49 Kurzon, Dennis: Discourse of Silence. 1998. vi,162pp.
48 Kamio, Akio: Territory of Information. 1997. xiv,227pp.
47 Chesterman, Andrew: Contrastive Functional Analysis. 1998. viii,230pp.
46 Georgakopoulou, Alexandra: Narrative Performances. A study of Modern Greek storytelling. 1997.
xvii,282pp.
45 Paltridge, Brian: Genre, Frames and Writing in Research Settings. 1997. x,192pp.
44 Bargiela-Chiappini, Francesca and Sandra J. Harris: Managing Language. The discourse of corporate
meetings. 1997. ix,295pp.
43 Janssen, Theo and Wim van der Wurff (eds.): Reported Speech. Forms and functions of the verb. 1996.
x,312pp.
42 Kotthoff, Helga and Ruth Wodak (eds.): Communicating Gender in Context. 1997. xxvi,424pp.
41 Ventola, Eija and Anna Mauranen (eds.): Academic Writing. Intercultural and textual issues. 1996.
xiv,258pp.
40 Diamond, Julie: Status and Power in Verbal Interaction. A study of discourse in a close-knit social network.
1996. viii,184pp.
39 Herring, Susan C. (ed.): Computer-Mediated Communication. Linguistic, social, and cross-cultural
perspectives. 1996. viii,326pp.
38 Fretheim, Thorstein and Jeanette K. Gundel (eds.): Reference and Referent Accessibility. 1996. xii,312pp.
37 Carston, Robyn and Seiji Uchida (eds.): Relevance Theory. Applications and implications. 1998. x,300pp.
36 Chilton, Paul, Mikhail V. Ilyin and Jacob L. Mey (eds.): Political Discourse in Transition in Europe
19891991. 1998. xi,272pp.
35 Jucker, Andreas H. (ed.): Historical Pragmatics. Pragmatic developments in the history of English. 1995.
xvi,624pp.
34 Barbe, Katharina: Irony in Context. 1995. x,208pp.
33 Goossens, Louis, Paul Pauwels, Brygida Rudzka-Ostyn, Anne-Marie Simon-Vandenbergen
and Johan Vanparys: By Word of Mouth. Metaphor, metonymy and linguistic action in a cognitive
perspective. 1995. xii,254pp.
32 Shibatani, Masayoshi and Sandra A. Thompson (eds.): Essays in Semantics and Pragmatics. In honor of
Charles J. Fillmore. 1996. x,322pp.
31 Wildgen, Wolfgang: Process, Image, and Meaning. A realistic model of the meaning of sentences and
narrative texts. 1994. xii,281pp.
30 Wortham, Stanton E.F.: Acting Out Participant Examples in the Classroom. 1994. xiv,178pp.
29 Barsky, Robert F.: Constructing a Productive Other. Discourse theory and the Convention refugee hearing.
1994. x,272pp.
28 Van de Walle, Lieve: Pragmatics and Classical Sanskrit. A pilot study in linguistic politeness. 1993.
xii,454pp.
27 Suter, Hans-Jrg: The Wedding Report. A prototypical approach to the study of traditional text types. 1993.
xii,314pp.
26 Stygall, Gail: Trial Language. Differential discourse processing and discursive formation. 1994. xii,226pp.
25 Couper-Kuhlen, Elizabeth: English Speech Rhythm. Form and function in everyday verbal interaction.
1993. x,346pp.
24 Maynard, Senko K.: Discourse Modality. Subjectivity, Emotion and Voice in the Japanese Language. 1993.
x,315pp.
23 Fortescue, Michael, Peter Harder and Lars Kristoffersen (eds.): Layered Structure and Reference
in a Functional Perspective. Papers from the Functional Grammar Conference, Copenhagen, 1990. 1992.
xiii,444pp.
22 Auer, Peter and Aldo Di Luzio (eds.): The Contextualization of Language. 1992. xvi,402pp.
21 Searle, John R., Herman Parret and Jef Verschueren: (On) Searle on Conversation. Compiled and
introduced by Herman Parret and Jef Verschueren. 1992. vi,154pp.
20 Nuyts, Jan: Aspects of a Cognitive-Pragmatic Theory of Language. On cognition, functionalism, and grammar.
1991. xii,399pp.
19 Baker, Carolyn and Allan Luke (eds.): Towards a Critical Sociology of Reading Pedagogy. Papers of the XII
World Congress on Reading. 1991. xxi,287pp.
18 Johnstone, Barbara: Repetition in Arabic Discourse. Paradigms, syntagms and the ecology of language. 1991.
viii,130pp.
17 Piraut-Le Bonniec, Gilberte and Marlene Dolitsky (eds.): Language Bases ... Discourse Bases. Some
aspects of contemporary French-language psycholinguistics research. 1991. vi,342pp.
16 Mann, William C. and Sandra A. Thompson (eds.): Discourse Description. Diverse linguistic analyses of a
fund-raising text. 1992. xiii,409pp.
15 Komter, Martha L.: Conflict and Cooperation in Job Interviews. A study of talks, tasks and ideas. 1991.
viii,252pp.
14 Schwartz, Ursula V.: Young Children's Dyadic Pretend Play. A communication analysis of plot structure and
plot generative strategies. 1991. vi,151pp.
13 Nuyts, Jan, A. Machtelt Bolkestein and Co Vet (eds.): Layers and Levels of Representation in Language
Theory. A functional view. 1990. xii,348pp.
12 Abraham, Werner (ed.): Discourse Particles. Descriptive and theoretical investigations on the logical,
syntactic and pragmatic properties of discourse particles in German. 1991. viii,338pp.
11 Luong, Hy V.: Discursive Practices and Linguistic Meanings. The Vietnamese system of person reference. 1990.
x,213pp.
10 Murray, Denise E.: Conversation for Action. The computer terminal as medium of communication. 1991.
xii,176pp.
9 Luke, K. K.: Utterance Particles in Cantonese Conversation. 1990. xvi,329pp.
8 Young, Lynne: Language as Behaviour, Language as Code. A study of academic English. 1991. ix,304pp.
7 Lindenfeld, Jacqueline: Speech and Sociability at French Urban Marketplaces. 1990. viii,173pp.
6:3 Blommaert, Jan and Jef Verschueren (eds.): The Pragmatics of International and Intercultural
Communication. Selected papers from the International Pragmatics Conference, Antwerp, August 1987. Volume
3: The Pragmatics of International and Intercultural Communication. 1991. viii,249pp.
6:2 Verschueren, Jef (ed.): Levels of Linguistic Adaptation. Selected papers from the International Pragmatics
Conference, Antwerp, August 1987. Volume 2: Levels of Linguistic Adaptation. 1991. viii,339pp.
6:1 Verschueren, Jef (ed.): Pragmatics at Issue. Selected papers of the International Pragmatics Conference,
Antwerp, August 1722, 1987. Volume 1: Pragmatics at Issue. 1991. viii,314pp.
5 Thelin, Nils B. (ed.): Verbal Aspect in Discourse. 1990. xvi,490pp.
4 Raffler-Engel, Walburga von (ed.): DoctorPatient Interaction. 1989. xxxviii,294pp.
3 Oleksy, Wieslaw (ed.): Contrastive Pragmatics. 1988. xiv,282pp.
2 Barton, Ellen: Nonsentential Constituents. A theory of grammatical structure and pragmatic interpretation.
1990. xviii,247pp.
1 Walter, Bettyruth: The Jury Summation as Speech Genre. An ethnographic study of what it means to those
who use it. 1988. xvii,264pp.

(Pragmatics & Beyond New Series 20) Jan Nuyts-Aspects of A Cognitive-Pragmatic Theory of Language - On Cognition, Functionalism and Grammar-John Benjamins (1991)

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

(Pragmatics & Beyond New Series 20) Jan Nuyts-Aspects of A Cognitive-Pragmatic Theory of Language - On Cognition, Functionalism and Grammar-John Benjamins (1991)

Diunggah oleh

Hak Cipta:

Format Tersedia

A COGNITIVE-PRAGMATIC THEORY OF LANGUAGE

Pragmatics & Beyond

Aspects of a Cognitive-Pragmatic Theory of Language

JOHN BENJAMINS PUBLISHING COMPANY

If you want to get ahead, get a theory. (Karmiloff-Smith and Inhelder

In science and philosophy alike, an exclusive preoccupation with

This study discusses a number of elementary problems concerning natural

1. Elements of the theory

2. Linguistic knowledge and language use: Chomsky revisited

3. Linguistic and conceptual systems in language processing

3.1.1. The basics of FG functional syntax 183

Subject index 379

Index of names 393

of innateness, acquisition, and phylogenesis.

Let's start by considering the basics of the cognitive-pragmatic theory of language.

1.1. Cognition and cognitive science

1.1.1. Cognition as an object in language research

obviously depends on what is involved in describing and explaining language. An

1.1,2, What does it mean to investigate cognition?

If a mapping between psychological processing and neurophysiological structures

We can do without a principle of induction. We invent our hypotheses, the bolder,

This appears to apply particularly strongly to research on cognition.

on other facets of normal adult linguistic behavior, or behavior in general. On

More specifically, the problems in determining the adequacy of models on cog

characterization of the object of inquiry.

1.13. Possible arguments against cognitivism

1.1.3.1. The individual and the social dimension of language

1.1.3.2. Linguists, language psychologists, and cognition

of a different nature, including matters such as finding the general systematicity

experimental conditions on the part of the subjects investigated. This makes

Platonism draws a fundamental distinction between the knowledge speakers have

According to Katz languages are abstract objects having no ontology, hence

In requiring conformity to a concrete reality, psychological reality conditions impose

grammatical structure of sentences, but concern particulars of subjective experience

Katz illustrates the conflicts he refers to with the following.

if one accepts conceptualism, one could be committed to claiming that, say, a

(ii) A conceptualist could be forced to choose a theory inferior to other possible

Let me make my point by way of an analogy. Major calculator companies, such as

Katz is using a strange concept of scientific truth, however. Normally, it

(1) Nightmares are dreams

(1) is necessarily true by virtue of its linguistic configuration. Obviously, if a

consequences of principles that human beings, by virtue of their psychological or

1.1.3.3. Cognitivism is not a theory

a process model of linguistic behavior. Apparently, many have generalized this

Such an ontologizing view of grammar [that the entire grammatical theoretical

Parret seems to suggest that linguistics should indeed be hypothetico-deductive

Apparently, Parret is only a rationalist as far as methodology is concerned, but

(and vice versa, of course - e.g. Watt 1974).

[Hence, this] requirement enables us to refute a linguistic theory if we can find

If there is serious evidence available in language psychology that TG cannot be

1.2. Functions and language functions

The second basic component of the cognitive-pragmatic theory of language is

(3) The pressure exerted upon the walls of a container by a mixture of

(4) The suicide rate in a community is a function of the degree of social

upon the latter, such as economy.

(5) The function of seeing is to detect enemies

(6) The function of shaking hands is confirming a social relationship

(7) The function of the eyes is seeing

(10) Road transportation has the function to allow a smooth provisioning

Nagel mentions five usages, which do not directly correspond to

(11) The functioning of the stomach

(12) The functioning of the postal system

(13) Respiration is a vital function of/for humans