for
Research
in
Linguistics and Language Sciences
(CRiLLS)
NEWCASTLE
WORKING PAPERS
IN LINGUISTICS
19(2), 2013
BOARD OF EDITORS
Burak Sunguralp Tekin
Caroline Burns
Jean Price
Geoffrey Poole
PUBLISHED BY
Newcastle University
Centre for Research in Linguistics and Language Sciences
FOR INQUIRIES:
CRiLLS
Room 3.01 & 3.02, Level 3,
Old Library Building
Newcastle University
Newcastle upon Tyne NE1 7RU
United Kingdom
Telephone: 0191 222 8790
Fax: 0191 222 6592
E-mail: crills@ncl.ac.uk
TABLE OF CONTENTS
20
37
59
74
Editorial Note
We would like to thank all of the reviewers for giving their time and expertise to evaluate
submissions. Finally, we would like to give our thanks to all of the contributors to this
volume.
Editors
This paper questions the assumption, widespread in linguistic theory and pragmatics, that
linguistic expressions have meaning in virtue of possessing semantic properties/content.
Problems created by this assumption are discussed and an alternative, semiotic, account of
meaning is developed that places linguistic meaning in the context of meaning in general.
1. Introduction
In a discussion of relevance theory, Dan Wedgwood comments:
Relevance theorists have tended to assume that RT can be used... as an adjunct to
fairly conventional approaches to other parts of linguistic theory. But in making the
move away from the moderate contextualism of Gricean approaches, RT has more
radical consequences, whether we like it or not. In effect, this constitutes a break from
conventional perspectives on semantics.... the nature of encoded meaning cannot be
understood without active consideration of inferential contributions to meaning. This
is nothing less than a reversal of conventional methodology, which tends to abstract
away from inferential pragmatic processes as much as possible. ... Once we reject
[conventional semantic analysis], as RT does in principle, the nature of encoded
meaning becomes an entirely open question.... A great deal of contemporary syntactic
theorising is motivated by a wish to account for the perceived semantic character of a
given sentence (hence the regular use of levels of representation of syntactic
representation like LF, LOGICAL FORM).
(2007: 679)
These are the thoughts I develop in this paper. I will argue that pragmatic theory - and
relevance theory in particular - should be seen as offering a challenge to conventional
linguistic wisdom.
What is at issue is the nature of meaning, no less. I will make a case for - or aim to
reinstate - an account of meaning at odds with conventional linguistic theorising, by which I
will assume Wedgwood means Chomskyan generative grammar (CGG). The relevant CGG
assumption is evinced by Brody (1995: 1) when he writes it is a truism that grammar relates
sound and meaning. CGG seeks to achieve this by positing two interface levels of linguistic
representation, Phonetic Form (PF) and Logical Form (LF), attributing to expressions two
sorts of property, phonological and semantic. This double-interface assumption is assumed to
be necessary to the modelling of language as sound with a meaning (Chomsky 1995: 2).
In reconstructing the idea that expressions have meaning by assigning them
semantic properties, CGG effectively equates semantics and linguistic meaning. I will argue
1
This paper is due to appear in Capone, A. et al. Perspectives on Linguistic Pragmatics (Springer Verlag). I am
grateful to Phil Carr, Jerry Fodor, Wolfram Hinzen, Magda Sztencel, Dan Wedgwood and Deirdre Wilson for
discussion. Needless to say, they are not responsible for errors or misthinkings in it, nor do/would they
necessarily agree with everything in it.
Burton-Roberts
that this insulates linguistic meaning from meaning in general and is unexplanatory.
Furthermore, it is the basis of Grices moderate contextualism. As Wedgwood observes,
relevance theory (RT) is radically contextualist in principle but shares Grices and CGGs
assumption that there is such a thing as linguistic semantics and Logical Form (LF) thought
of as a level of linguistic encoding. To that extent RTs radical contextualism is qualified.
CGGs double-interface assumption is a legacy of Saussures concept of sign (BurtonRoberts & Poole 2006). I approach the topic of this paper through a discussion of the
Saussurean sign (in Section 2) because I want to bring semiotics to bear on the subject of
meaning and to highlight some problems for CGGs (Saussurean) reconstruction of sound
with a meaning. I reject that CGG account in favour of a (roughly) Peircean concept of
sign. This, I argue, is more general in its application, more explanatory and more consistent
with the cognitive naturalism of CGG and RT.
In the light of that, Section 3 offers a general account of what meaning is, developing
ideas touched on in Burton-Roberts (2005, 2007). This account denies that linguistic
expressions have semantic properties. It distinguishes meaning and semantics. Assuming,
with RT, that only thoughts have (real) semantic content, it argues that meaning is a relation
(of something, potentially anything) to the semantic content of a thought, a cognitive relation
effected by inference. The crucial idea here is that meaning is not a (semantic) property but a
semiotic relation (to semantic properties). This goes for meaning generally; my aim is to
situate linguistic meaning within a semiotic account of meaning in general. My own view is
that the picture of meaning that emerges is consistent with RT. One might even go as far as to
say that relevance, as defined in RT, is meaning.2
2. 1. The Saussurean sign...
There are two well-known features of the Saussurean linguistic sign.
The first is that the linguistic sign is constituted by the conjunction of a concept and a
sound image. Sound image and concept combine to make a further entity, the sign itself,
which is a double entity (Saussure, 65), a two-sided psychological entity (66).3 The
relation between concept and sign and that between sound image and sign are part~whole
relations. In modern parlance, the concept is the signs semantics; the sound image is its
phonology. The relation between semantics and phonology is thus a part~part relation.
Saussure (e.g. 67) explicitly refers to the sign as the whole and to the sound image and
concept as its parts. Crucially, the same idea is evident in CGGs double-interface
assumption: there are sensorimotor systems that access one aspect of an expression and there
are conceptual-intentional systems that access another aspect of an expression, which means
that an expression has to have two kinds of symbolic objects as its parts (Chomsky 2000b:
9).
The formal study of part~part and part~whole relations is called mereology. I wont
be invoking formal mereology, but Ill borrow the term mereological in discussing the
2
I wont push this thought further but the reader may want to bear it in mind in Section 3 below. Incidentally,
Sperber & Wilson (1986/1995: Chapter 1) assume that any semiotic approach to meaning aims to reduce all
meaning to a code model of communication. That may have been true of 20th century extensions of Saussurean
semiotics (they cite Levi-Strauss and Barthes) - in connection with which they write (with justification in my
view) The recent history of semiotics has been one of simultaneous institutional success and intellectual
bankruptcy (p. 7) - but nothing could be further removed from my aim here. Whereas, in their terms, The
semiotic approach to communication... is a generalisation of the code model of verbal communication (p. 6),
my aim is precisely the opposite: to situate linguistic meaning within a more general semiotic - i.e. inferential account of meaning. My claim is that all meaning (though not all communication, see note 9 below) - and all
decoding involved in the construction of meaning - is inferential in character.
3
All references to Saussure are to Wade Baskins (1959) translation of the Cours de Linguistique Generale.
Saussurean sign. In an intuitively equivalent formulation, concept and sound image are
respectively the semantic and phonological properties of (or aspects of) the linguistic sign.
Saussure has little to say about where syntax figures in this. This is addressed in
CGG, where phonological and semantic features are treated as properties (parts, aspects) of
syntactic entities (Burton-Roberts 2011). This adds substance to the double entity idea: if
you believe in the existence of a further entity jointly constituted by phonological and
semantic features, there should indeed be something substantive to say about it e.g. that it is
syntactic - over and above what can be said about the phonological and the semantic.
The second feature of the Saussurean sign is that it involves a semiotic relation, the
signifier~signified relation. The sound image signifies the concept. Significantly, nothing in
CGG reflects this feature of Saussures thinking, a matter I address below.
Now, the mereological (part~part) relation and the semiotic (signifier~signified)
relation are at least different. Since things generally are related in more ways than one, it
might seem that sound image and concept can be related in both these ways, as Saussure
assumed (see Fig. 1).
Figure 1. The Saussurean sign
SIGN
sound image
signifier
concept
co-part, or co-property
signified
Against this, I will argue that the two relations are so different as to be incompatible;
we must choose between them.
Notice first that the part~part relation is symmetric, in the sense that they are co-parts,
co-constitutive of the Saussurean sign. Saussure (113) draws a general analogy with a sheet
of paper: in a language (a system of signs), sound image and concept each relate to the sign
as the two sides of a sheet of paper relate to the sheet. As we shall see, this symmetry was
crucial for Saussure. In sharp contrast, the semiotic relation is antisymmetric. One term is the
signifier, not the signified; the other is the signified, not the signifier. Its a relation from the
sound image (signifier) to the concept (signified). The sheet-of-paper analogy is completely
inappropriate in this connection.
Notice furthermore that on the mereological conception there is no direct relation
between sound image and concept. Their (part~part) relation to each other is entirely
derivative: they are in that relation only because each is, primarily, in the (antisymmetric)
part~whole relation to the sign. In that respect, the relation between sound image and concept
is not self-explanatory. The semiotic relation, by contrast, is direct and primary. It does not
follow from any other more direct relation. In that sense, by comparison with the
mereological relation, the semiotic relation between sound image and concept is selfexplanatory.
A further difference is that the mereological account treats as an object (the sign)
what the semiotic idea, by itself, treats purely as a relation. The mereological idea, by
definition, entails that there is an object distinct from both sound image and concept
constituted by their combination. Not so the semiotic idea. Unless mereological, a relation
Burton-Roberts
between two objects implies no further object.4 Notice that property-of talk in connection
with the mereological conception is questionable on the semiotic conception. If x stands in a
semiotic relation to y (i.e. signifies y), there is no conceptual necessity to posit an object Z
(the sign) for x and y to be properties of.
This doesnt mean that, on a purely semiotic conception, we have to abandon any idea
of sign. There is an alternative, non-mereological, conception of sign compatible with the
signifier~signified relation. On this conception it is the signifier itself that is the sign (Fig.2).
Figure 2. The purely semiotic (Peircean) sign
SIGN
(signifier)
signified
(concept)
I call it Peircean because it is consistent with the thought of C. S. Peirce (1933). But
I wont be invoking the whole panoply of Peircean semiotics. Furthermore, I will assume with Saussure, not Peirce - that the signified is always a concept (Section 3 below).
Saussure (67) explicitly rejects this (Fig. 2) concept of sign, insisting on the
mereologically constituted sign (Fig. 1). Against this I will suggest that the mereological
account is conceptually (a) insufficient and (b) unnecessary, even (c) assuming it is possible.
And there are other objections to it, notwithstanding its influence on CGGs double-interface
assumption and moderate contextualisms encoded meaning/semantics.
The mereological account is conceptually insufficient, I hold, because there is nothing
actually semiotic about it, in and of itself. Mereological (part~part and part~whole) relations
have nothing to do with meaning. What I mean is that they fail to differentiate the relation we
are concerned with from e.g. that between the barrel and the nib of a pen, the two sleeves of a
shirt, the seat and back of a chair, which are mereological. This mereological idea offers no
explanation of why a sound image (phonology) and a concept (semantics) should actually be
related. It might be objected that this ignores the fact that, for Saussure, the relation between
sound image and concept isnt only mereological, its also semiotic. But that just goes to
show the conceptual insufficiency of the mereological idea in this context. It needs to be
supplemented by the semiotic idea.
As regards conceptual necessity: having supplemented it with the semiotic relation,
what conceptual/theoretical work remains for the mereological account? It adds nothing. The
semiotic idea, we have just seen, is necessary - and it is in itself sufficient. Furthermore, it
yields a concept of sign that is at least more obvious: what is a sign if not a signifier?5 It is
also more parsimonious. It calls for just one relation between just two entities. The
mereological account multiplies beyond necessity. It posits three relations (two part~whole
relations and a part~part relation) and three entities. And even those need supplementing by
the semiotic relation.
We might even question whether in this context the mereological idea is possible, on
the grounds that phonological and semantic properties are sortally distinct. Sortally distinct
properties are such that nothing can have both sorts of property. For example: sore throats,
4
Other than an abstract set-theoretical entity. See Burton-Roberts (2011) for discussion of the issue in CGG.
Saussure explicitly denies the sign is an abstract entity (15, 102), insisting on its being a concrete entity. His
reference to parts and wholes (e.g. 67) would be inappropriate were he thinking in merely set-theoretical
terms. Sets dont have parts, they have members.
5
It is reasonable to ask what signs are signs of/for. On the Fig. 2 (purely semiotic) notion of sign, the question
receives an answer (though a general one): they are signs of/for concepts. But on the Fig.1 (Saussurean) notion
of sign, the question is simply incoherent: the mereologically constituted sign isnt a sign of of/for anything. See
below on a necessary precondition for signification (and bootstrapping).
The nearest Saussure comes to allowing this is in the idea that (mentally constituted) signs are realized in
phonetic phenomena. But even here, it is surely just the signifier (the sound image) thats realized. Since the
Saussurean sign is constituted in part by a concept, its hard to accept it could be realized in phonetic
phenomena.
Burton-Roberts
- the code (14) - of some particular language. The motivation, in short, is an extreme
version of the so-called Sapir-Whorf hypothesis.
This must be rejected, I believe. As Sperber & Wilson (1986/1995: 192) and Carston
(2002: 30-42) argue, what is thought/thinkable extends well beyond what is linguistically
encoded/encodable. This is not a merely philosophical or theoretical matter; its the stuff of
common experience. Chomsky (2000b: 76) puts it well: ...very often, I seem to be thinking
and finding it hard to articulate what I am thinking. It is a very common experience to try
to express something, to say it and to realize that is not what I meant... it is pretty hard to
make sense of that experience without assuming that you think without language. You think
and then you try to find a way to articulate what you think and sometimes you can't do it at
all;... if you are thinking, then presumably there's some kind of conceptual structure there.
The universality of such experience, the necessary precondition for signification mentioned
earlier and the related problem of how the Saussurean sign could actually evolve/arise other
than by the merest (most circular) bootstrapping (Fodor e.g. 1975), all lead me to agree there
must be some kind of conceptual structure there, a language of thought used internally
(Chomsky 2006: 9), if subconsciously, logically prior to a speakers particular language, as
argued by Fodor and assumed in RT.7
How is the set of concepts delimited? Since the relation between sound image and
concept is arbitrary, if concepts only exist as constituents of linguistic signs then the set of
concepts must be arbitrary (and, notice, semantics = linguistic semantics). Another response,
more consistent with the naturalism of CGG and RT, is that the set of concepts is the set of
humanly entertainable concepts, a set delimited by nature - human nature. This is consistent
with a Peircean conception of sign but not with what motivates the Saussurean sign.
2.2. ...CGG and pragmatics
Ive suggested we must choose between the Saussurean (mereological) sign and the
Peircean (purely semiotic) sign and have sought to show we must choose the latter. All this
would be of merely historical interest were it not that CGG makes the diametrically opposite
choice. It is Saussures mereological idea that is embodied in CGGs idea of a (syntactic)
object with phonological and semantic properties (aspectsparts). And only the
mereological idea. CGG doesnt appeal to sign. CGG eschews all reference to semiotic
concerns. In CGG we are asked to make do just with a mereological (part~part) account of
the relation between phonology and semantics.
I have argued this is at least conceptually insufficient - uninformative and
unexplanatory. Mereological relations have nothing to do with meaning. CGGs response
might be that, in attributing semantics to expressions, it assigns them their meanings. This
equates having meaning with having semantic properties in the linguistic context. The
next section presents an account of meaning that undermines that equation. In the meantime, I
here offer some observations on the mereological account in the context of CGG and RT.
In addition to the above objections to it, the mereological account poses two problems
internal to CGG (Burton-Roberts 2011). The first is this. As noted, phonological and
7
I would argue it is phylogenetically and ontogenetically prior. The passage from Chomsky, incidentally, is
consistent with a more nuanced version of the Sapir-Whorf hypothesis. I am not the first to suggest there is more
than one kind of thought. There is (a) the kind of thought explicitly referred in that quote and (b) another, also in
evidence there, which consists in gaining more conscious access to those thoughts by trying to articulate them.
See Burton-Roberts (2011).
Grice takes the Saussurean line: A plausible position is that...language is indispensable for thought
(1989:353). But (355) he is upfront - and unusually clear - on the dilemma this poses re the precondition for
signification (his intelligibility). Less clear (to me) is his proposed solution (355-6).
Burton-Roberts
3. Meaning
Meaning is not exclusively linguistic, nor is sound with a meaning. If we seek an
explanatory account of linguistic meaning, it should be laid in the nest of these (cognitive,
semiotic) truisms.
I dont claim that all of what follows is new. What it aims to show is that it is
unnecessary and unexplanatory (not to say plain wrong) to attribute any kind of semantic
property/content to linguistic expressions. To that end, section 3.1 argues for a distinction
between meaning and semantics. In the light of that, section 3.2 discusses their relation.
3.1. Distinguishing meaning and semantics
A phenomenon P is a sign if it is significant. It signifies something (X). Thereby P
means X. Consider first some well-worn examples of natural signs. (i) Those spots mean
measles (Grice 1989: 213). (ii) Pattering on the window means rain. (iii) Red litmus paper
means acid. (iv) Smoke means fire. These are examples of Peirces category of indexical
sign, involving Grices natural meaning. (ii) is a case of sound with a meaning. What
makes P a natural sign of X here is a natural (i.e. agent-less) causal relation between X and P.
Pace Peirce, whose terminology suggests that all semiotic relations are representational, these
are not representational. As a (natural, indexical) sign of fire, smoke is not a representation of
fire. By the same token, they are not ostensive.8 They are signs-of, not signs-for.
Its not nearly that simple of course. None of (i)-(iv) is objectively true, i.e. true in the
absence of a subject S noticing P on an occasion. A particular phenomenon P is a sign only if
it signifies something (X) to S. And meaning-X-to-S depends on S making inferences based
on her beliefs. Red litmus paper in a liquid will mean to S that the liquid is acid only if S
assumes it is litmus paper and has the belief B that acid causes litmus paper to turn red. Red
litmus paper means acid abstracts away from such crucial details, on the assumption that B is
(general) knowledge. Those spots may mean measles to a doctor but not to Peter. General
knowledge and even what passes for the evidence of ones eyes (ears, etc.) dont obviate
the need for inference, however subliminal (Searle 1983).
Nor is it true that P means X to S - not if X stands for the causing phenomenon.
Rather, P means that-X to S. Subject to cognitive conditions just described, smoke is
meaningful to S because it communicates to her that theres a fire where that theres a fire
identifies a thought of Ss. Meaning is by its nature communicative. I take communicates,
when it involves meaning, to be equivalent to leads S to entertain a thought.9
What follows from all this, even if obvious, is important. Although we say that P, as a
sign, is meaningful or has meaning, and although we may talk of sound with a meaning,
such talk does not identify any property of P. More to the point - and this is my point - no one
would want to say that their having meaning identifies a semantic property of those
phenomena. On an occasion of pattering-on-a-window, it is true (false) that it is raining but
no one would attribute any truth condition to the pattering. If truth conditions are what youre
after, they pertain to Ss thought that-X. What might be said to be the meaning of P is
8
However, they can be exploited ostensively in a way that either depends on their indexicality or overrides it. A
Glaswegian accent is an indexical sign of Glaswegian provenance but not if deliberately adopted (in which case,
if not intended to deceive, it will be representational). Susan putting on her coat is an indexical sign of her
intention to go out (perhaps) but she may exploit that with the semiotic intention of (ostensively) representing
her desire to do so. White smoke rising from the Vatican ostensively represents a Papal election and, if so
recognised, that will override (iv).
9
I am concerned with communication but only insofar as it can held to involve meaning (and thus thought and
inference). There are of course other kinds of communication e.g. among honey bees and (chemical) among
cells as Deirdre Wilson reminds me. I am not concerned with these.
10
Burton-Roberts
entirely extrinsic to P. Despite the predicational similarity, means X is utterly different from
rises when heated or is mercury/heavy/edible. In at least the cases discussed so far indexical signs in general - meaning and semantics are clearly distinct.
With indexical signs we are far from language but, in the respects that matter here, the
above holds quite generally. Take Peirces category of symbolic signs - Grices non-natural
meaning (meaningnn). I illustrate this first by non-linguistic examples but we are close enough
to language here, I suggest, for it to have a direct bearing on linguistic meaning. What I am
suggesting is that, as with indexical signs so with symbolic signs (whether linguistic or not):
meaning and semantics need to be distinguished. Illustrative phenomena are a red flag
(warning that artillery practice is in progress or swimming is forbidden), a gunshot (to start a
race) and a ringing bell (to stop the bus, raise the alarm, etc.), the last two being sound with a
meaning. What makes these symbolic is the fact that theres no natural causality involved
here. These are non-natural in depending on other things: (a) on a certain convention, (b) on
semiotic intention, and (c) on inferentially derived recognition by S of (a) and (b). Unlike
indexical signs, symbolic signs are representational (in virtue of (a)-(c) above); they are
ostensive. The difference is that, symbolic signs being representational, that-X is meant. (We
dont say that-theres-a-fire is meant by smoke.) Here meant = intended; and intended
here is short for intended to mean (to S). In these symbolic cases, an agent A communicates
that-X to S by means of P. In other words, A arranges for the occurrence of some
phenomenon P, intending that its occurrence will - in virtue of (c) - communicate to S (lead S
to have the thought) that-X. Since we are not telepathic, some perceptible P is necessary.
It was Grice who instated intention and its recognition at the centre of symbolic (nn)
meaning (m-intention, indeed). A convention, though important when involved, only makes
for the possibility of meaning.10 But possible meaning is not meaning (any more than a
possible queue on a motorway is a queue). A red flag actually means nothing to anyone in
its box. Even when flying and intended/meant thereby to mean/communicate e.g. that
artillery practice is in progress, it wont actually mean that if no one (S) notices it other than
F, the person who flew it. Im not sure it actually means that to F himself, crucial though its
potential meaning was to his intention in hoisting it. (And, to Fs commanding officer, it
might mean only that his order - to F, to hoist the warning flag - has been carried out). It is
the implementation of a convention on an occasion that makes for actual meaning and then
only if some (other) S assumes that it was meant/intended to mean/communicate - to her what it potentially means (rather than, in the case of the ringing bell, as a demonstration in
fire practice). As regards artillery practice is in progress or swimming is forbidden, a
flying flag might be said to be ambiguous in virtue of there being distinct conventions
governing what it potentially means. But, for that reason, such ambiguity has the same
nugatory status as possible meaning. Its not actual ambiguity. Actual ambiguity is an
occasion-specific subjective mental state of (inferential, interpretative) indecision of S
(Burton-Roberts 1994).
There are of course big differences between indexical and symbolic signs.
Nevertheless, I see no reason not to say that, when P functions for someone (S) as a symbolic
sign, P has meaning for S in the same sense as when P is an indexical sign. It is only the
cognitive mechanics that subserve the meaning that differ: (a)-(c) above in the symbolic but
not the indexical case. Indexical signs involve just one person (S) while symbolic signs
involve more than one (A and S) but that doesnt alter the general fact about meaning. It
concerns only whether the meaning involves intention (was meant) and convention. Both
communicate an idea to S, i.e. lead her to have a certain thought. It is true that, with symbolic
10
As the first two examples of note 8 above show, while all non-natural (representational, ostensive) signs
depend on intention, not all depend on convention.
11
signs, someone (by means of P) is communicating rather than merely something (P)
communicating - but its inferentially dependent communication in both cases.11
Symbolic signs also differ from indexical signs in that, when symbolic, the whole
rationale of P lies in its being (intended as) a sign. Since A arranges for it, the very
occurrence of P is motivated by that. But that doesnt (or shouldnt) tempt us into thinking
that Ps status as sign (for S) involves any intrinsic property of P. Nor would we attribute any
truth condition to a flying flag. Truth conditions lie elsewhere. My point is this: impressive
though the differences between indexical and symbolic signs are, P as a symbolic sign has
meaning (for S) but it no more has semantic properties than any indexical sign. Indeed, what
makes for meaning with symbolic signs is, if anything, doubly extrinsic to P.
Subject to sorting out issues with word (below), I see no reason not to say that words
are symbolic signs - and thus that all the above applies to them. In some quarters, thats
uncontroversial. But Grice wrote ...the distinction between natural and non-natural meaning
is, I think, what people are getting at when they display an interest in a distinction between
natural and conventional signs. But I think my formulation is better. For some things
which can meannn something are not signs (e.g. words are not) (1989: 215, my emphasis). In
the absence of any explanation or argument, I can only speculate why Grice (or anyone)
should deny that words are signs. He was interested in Peirces semiotics (Chapman 2005:
71) and opened Meaning (1989: 213-223) by discussing indexical and symbolic signs
(though Peirce, semiotics, indexical and symbolic make no appearance). So, to
speculate, could Grices denial have been in aid of driving a wedge between (semiotic)
meaning in general and word meaning? This would be consistent with the semantic corral
Grice sought to erect around word meaning. For, if the denial was motivated by the
assumption that words have meaning in virtue of having semantic properties (having
meaning in a literal, objective, property sense), that certainly would distinguish their
meaning from that of signs generally, as I have sought to show. But, widespread (well-nigh
universal) though it is, the assumption that words are unique in having meaning in virtue of
having semantic properties has actually never been defended explicitly. In the absence of
explicit supporting argument, a distinction between meaning-as-semantics and semiotic
meaning (the former linguistic but not explained beyond that) is conceptually profligate,
uninformative and not explanatory of linguistic meaning.
A more persuasive reason for denying that words are signs might go like this. At least
as treated thus far, signs are perceptible phenomena (P above). But words are not
perceptible phenomena. Hence words are not signs. Im prepared to accept that. However, the
relevant perceptible phenomena here are the phonetic phenomena we think of as uttered
words. The question is: When do such phenomena count as uttered words? And, since
uttered word suggests the existence of words independent of and prior to utterance, what is
a word anyway?
The CGG (and RT) answer is that a word is a syntactic object constituted by
phonological and semantic properties. But I doubt whether proponents of this answer really
believe that what we physically utter literally has syntactic and semantic properties. Its
agreed, surely, that what speakers utter are sounds - brute sounds. In which case, what
11
Sperber & Wilson (1986/1995: 22-23) deny this. Their discussion involves an example of the ostensive use of
an indexical sign: Suppose that Mary intends to inform Peter of the fact that she has a sore throat. All she has
to do is let him hear her hoarse voice (22). Of this they write (23) This should not be regarded as a form of
communication - a denial that must apply, a fortiori, to (non-ostensive) indexical signs as such. For myself, I
see no reason to deny that, if Peter is led to think that Mary has a sore throat from hearing her hoarse voice whether or not by speaking in that voice she intended to inform him of it - her hoarse voice (P) does
communicate to him that she has a sore throat.
12
Burton-Roberts
speakers utter can no more have semantic properties than the pattering of rain or a starters
gunshots can.
Phonetic phenomena functioning as linguistic signs involve a phonology (more
strictly, a morpho-phonology). But I dont want to say (and I dont think phonologists do)
that uttered sounds themselves have phonological properties. Actual sounds counting as
linguistic are articulatory implementations of a phonology. What a given phonological
system does is license the production of phonetic phenomena (actual sounds). That much is
uncontroversial. But in what sense does a phonology license sounds? My claim is that it
licenses them for use as symbolic (representational) signs.12 It is in virtue of this, not the
possession of semantic properties, that we might talk of linguistic sound with a meaning.
On this account, then, a word is a phonologically constituted license for the use of
sounds as symbolic signs. A word is not an object (with semantic properties) but a
phonologically constituted semiotic license. Its a rule, if you like. As such, words are indeed
not signs and dont themselves have meaning. Words (as rules) only make for the
possibility of signs and thus the possibility of meaning. P counts as an uttered word and thus
an actual sign when it is assumed to be the physical implementation of a (morpho-)
phonologically-constituted semiotic license.
This licensing is subject to the cognitive conditions (a)-(c) above and all the provisos
discussed there. Condition (a) made reference to convention. The above amounts to the
suggestion that a word is a phonologically constituted convention. By convention I simply
mean a relational locus of (non-natural, symbolic) Saussurean arbitrariness. I dont mean it is
conventional in the sense of existing in a supposedly objective public language external to
individuals (c.f. Sperber and Wilsons (1998) public words). As Chomsky argues (e.g.
1986), there are no such public languages (E-languages). Conventionality in this public
sense, insofar as it exists as a linguistic phenomenon, is rooted in the personal assumptions of
individuals (Pateman 1987). These are I-linguistic assumptions (where I stands for
individual), collectively amounting to an I-assumption about others words - effectively, the
I-assumption that others implement the same conventions as I. We notice this assumption in
inferential interpretation consciously only when faced with its fallibility. A famous fictional
example (from Sheridans The Rivals) is Mrs Malaprops Sir, you are the pineapple of
politeness! Here the need for inference regarding her I-language (her I-conventions/licenses)
and thus her I-assumptions about the public language is more apparent (intentionally so, it
being fictional) and it is generally successful (Davidson 1986).
What does pineapple mean? The semantic answer, of supposedly objective
necessity, runs: pineapple means PINEAPPLE. But this is either vacuous (pineapple means
what it means) or - insofar as it implies that pineapple always and for everyone means the
same thing - simply false. There is no literally objective/public fact here, i.e., no fact
independent of particular I-assumptions and inferences about others I-assumptions and
intentions on occasions, however well-evidenced those assumptions might seem to be. Nor is
there any call here for talk of a coming apart of speaker meaning and linguistic meaning
(Carston 2002: 18). What is the use of words - including Mrs Malaprops - if not linguistic?
Words mean what speakers mean by them on occasions of use (Recanati 1998) - and that, in
the case of Mrs Malaprops pineapple, is what I (and I believe most others) mean by
pinnacle.
More relevant here, because empirically (non-fictionally) and more widely attested,
are certain uses of disinterested, infer, refute, antisocial - uses in which they mean what I,
and I believe some others, mean by uninterested, imply, reject, unsociable. But who am I (or
12
This is the representational (as against CGGs realizational) conception of phonology proposed by the
Representational Hypothesis (e.g. BR 2000, 2011, BR & Poole 2006).
13
anyone) to damn these as misuse - let alone imply they are not linguistic - especially
when, on the evidence of my own understanding of them (given my I-appreciation of their
prevalence), they are part of my own I-language?
As work in RT has shown, even when other speakers uttered words appear to be
consistent with our own I-assumptions about standard (normal/public) meaning, they
seldom if ever actually mean exactly that. See e.g. Sperber & Wilson (1998) on tired.
Furthermore, as Carston (2002: Ch.5) vividly demonstrated in a discussion of open and
happy, it is not even clear we could ever actually grasp such standard (objectively encoded)
meanings. Meaning (actual meaning) just is personal, neither sub-personal nor supra-personal
(public).
A case in point is He was upset but not upset, offered by a witness in O. J.
Simpsons trial, cited by Carston (2002: 324 and 2004). But Carstons (2004: 839) discussion
of it illustrates what I have been questioning. She writes As far as linguistically supplied
information goes, this is a contradiction, a fact that presumably must be captured somewhere
within a semantics for natural language. I have been urging, in effect, that there is no fact
here to be captured within a semantics for English, let alone for natural language. What
there undoubtedly is here includes the articulatory/acoustic fact of the double occurrence of
upset. But only a firm commitment to the double-interface [phon+sem] assumption - and
thus to linguistic semantics - would suggest that this phonetic fact goes, of objective
necessity, hand-in-hand with some unique semantic fact. Carston herself points out that
whats intended and recognised - the actual meaning - is not contradictory, representing this
by her distinction UPSET* vs. UPSET**. It is this, as RT itself argues, thats (really)
semantic.
Convention needs more discussion than space allows. Ill mention just one issue in
this connection. With convention comes the idea of encoding. As I have discussed
elsewhere (2005, 2007), RT operates with what Ive called a constitutive notion of
encoding, according to which the encoded meaning of a word is a constitutive semantic
property of it, deterministically decoded not inferred (Carston 2002: 322-23), consistent with
CGGs Saussurean double-interface assumption. However, by encoding I mean something
different, briefly contemplated by Carston (2002, 363) but not pursued. In illustration of what
I mean, and to keep things very simple, take the Morse code, where (the convention is that)
dot-dash encodes THE LETTER A. The whole point of the Morse code is that dot-dash is
not the letter A. It has none of its properties. Dot-dash is, quite distinctly, a sign for - a
pointer to, a representation of, indeed means - THE LETTER A. I accept there is such a
thing as encoded meaning only in this latter (relational, semiotic, non-constitutive) sense,
distinguishing whats-encoded (x) from what-encodes-it - E(x) - subject always to the above
on conventions and in no sense that admits of deterministically decoded, as distinct from
inferred, meaning.
I have in effect been arguing with Fodor (1998: 9) that English has no semantics.
But Fodor immediately follows that with Learning English isnt learning a theory about what
its sentences mean, its learning how to associate its sentences with the corresponding
thoughts. There is much I would question in this but what concerns me here is the implied
equation between (not)-having-semantics and (not)-having meaning. If the claim that
English has no semantics is to be sustained with any plausibility, we need to avoid any
suggestion that what doesnt have semantics doesnt have meaning. I hope Ive shown that
this does not follow, given a distinction between meaning and semantics. On the contrary, as
I will argue in the next section, learning how to associate []s with ...thoughts precisely
amounts to learning...what []s mean. Jerry Fodor (p.c.) has dismissed this as
terminological. Nevertheless, it seems to me important if English has no semantics is to
command assent.
14
Burton-Roberts
15
This, again, involves property vs. relation. His account of conceptual content is externalist
and relational - content is constituted, exhaustively, by symbol-world relations (Fodor 1998:
14). This is not what I mean. By conceptual content I mean a (indeed the individuating)
property of a concept.13 This is an internalist constitutive notion of conceptual content - and
nativist, implying that what you acquire is not a concept but a certain kind of access to a
concept (from worldly and/or linguistic experience). However, this may be terminological,
since Fodor himself allows that concept-world relations are determined by something minddependent, which I assume has to do with the concept itself. Being a doorknob is just:
striking our kinds of minds in the way that doorknobs do (Fodor 1998: 162). This seems to
me to allow, if not demand, that the concept DOORKNOB has some kind of internal
constitutive property. This for me is its content, determining non-arbitrarily what external
phenomena it locks onto and thereby makes sense of. Concept-world relations (semantics in
an externalist, relational sense), it seems to me, arise when a concept sufficiently complex to
be entertained as a representation actually is so entertained. I think this amounts to saying that
the distinction-and-relation between internalist and externalist semantic content is the
distinction-and-relation between the language of thought and actual thoughts. End of
digression.
I earlier undertook to show that, even when something can be said both to have
semantic content and to have meaning, meaning (the relation) and semantics (one term of
that relation) must be distinguished. Whats at issue here is whether thoughts (concepts), in
and of themselves, have meaning. Ive assumed thoughts have semantic/conceptual content.
But do they, thereby (in virtue of just that fact), have meaning? My answer is No. I dont
recall this question ever having been asked but, intuitively, that seems to me the right answer.
Independently of that, my answer must be No given what Im claiming meaning is. It is
signs that have meaning but thoughts are not, in and of themselves, signs. That is, it is not
necessary, not definitional of a thought, that a thought leads you to have a thought. The only
sense in which that could be definitional would be if a given thought led you have that very
thought. Self-causing thoughts are out, I assume. Your having thought Tj does not
communicate Tj to you. In short, semantic content is not, in itself, semiotic - it is not, in itself,
meaning.
However, although I am claiming that thoughts in and of themselves dont have
meaning (are not in and of themselves signs), I am not denying that, nevertheless, thoughts
can have meaning, i.e. can function as signs in the mental life of an individual. They
generally do. (It is this that motivated my earlier move from P, for external phenomena, to
, for anything.) The point is, though, that this meaning of a given thought T, for S, is
distinct from its semantic content. The idea is simply this: having one thought T1 can lead you
to have/entertain another thought T2 whose semantic content may be entirely distinct from
that of T1. Having T1 can communicate to you (the distinct content of) T2. This in fact
strengthens the distinction between semantics and meaning. (The scare quotes around have
(meaning) and meaning of are a reminder that the meaning lies neither in T1 nor T2 but in
the relation T1T2.)
I illustrate this below but I must first address an issue Ive not attended to. Lets say
Susan (S) sees something in the kitchen which she takes to be a pile of clean washing. This
needs unpacking: is a visual stimulus and Ss first thought (T1) on seeing is THATS (or
LO!) A PILE OF CLEAN WASHING. What Im calling first thoughts are, in psychological
terminology, percepts. My previous use of P (for phenomenon) obscured the distinction
between stimulus and percept (the result of mentally processing the stimulus). Thereby it slid
13
Fodor (ibid.) comments that his atomistic theory of concepts allows him not to hold that ones inferential
dispositions determine the content of ones concepts. It seems to me that it forces him not to hold that and
forces his relational account of content. Incidentally, on atomism in RT, see Burton-Roberts (2005: 339).
16
Burton-Roberts
over the issue of whether mere stimuli are signs. Is [stimuluspercept] a semiotic relation?
In other words, is it meaning? Since Ive described the percept as a first thought (T1), you
will have guessed I want to say it is meaning. Given how S is internally constituted, the
stimulus leads S to have a percept (P for percept now), an inferentially-derived mental
representation. P is a conceptual interpretation of that stimulus. Without percepts theres no
meaning, just meaningless stimuli. The inferential move from to SMOKE() is different
from, but not different in kind from, the move from SMOKE() to CAUSED-BY-FIRE(). If
[stimuluspercept] is excluded from (on grounds of being in some sense prior to) any
semiotic move, where do we stop? Is CAUSED-BY-FIRE() a percept? Im bothering with this
because it might be felt that treating the [stimuluspercept] relation as meaning stretches
meaning too far, thereby undermining my account. I want to say it is meaning and, given
the generality of my project for meaning, I dont regard this as undermining.
So, Susans first thought (T1) is THATS (or LO!) A PILE OF CLEAN WASHING. Lets say
that, given Susans cognitive context, CS - her current mental state, her projects and
preoccupations - T1 leads her to have (communicates to her) another thought T2: JOHN HAS
DONE THE WASHING. (If T1 is true then, given CS, T2 is true; T1 by assumption is true, hence
T2 is true.) Now lets say that, given CS, T2 leads to T3: I DONT NEED TO DO ANY WASHING
RIGHT NOW and T3 in turn to T4: I CAN FINISH MY LECTURE RIGHT NOW. Susan will almost
certainly go further - subject to constraints expounded by relevance theory - but Ill stop
there. The stimulus and each of T1, T2 and T3 are signs for S, given CS. The meaning of T1
for S (and, I claim, of the stimulus itself) ultimately and indirectly lies in T4. Now, the
relations [stimulusT1] and [T1......T4] are semiotic. They are meaning relations. But they
are clearly not semantic. The stimulus has no semantics and T1, T2, T3 and T4 (more strictly,
the LoT formulae in terms of which T1T4 are couched) are semantically unrelated. The
point of this illustration has been to show that even - in fact especially - in the case of what
does have semantic/content (namely thoughts and only thoughts) and has meaning, meaning
and semantics are distinct.
As an aside, its reasonable to ask what kind of semiotic relation holds between those
thoughts of Susans. Since it involves neither convention nor semiotic intention, it is clearly
not symbolic (not nn), not representational. Although were dealing with an agent, S, the
relevant causations of thought are not agentive. So it seems it must be indexical (natural). I
reconcile myself to this conclusion as follows. We saw, with our examples of natural
(indexical) signs, that their status as signs depends on a given subjective background. The
same holds in the above illustration. The difference is that our original examples were
assessed against subjective backgrounds assumed to be shared as (supra-personal) general
knowledge and taken for granted as such, whereas in the illustration the background (CS) is
personal, with no claim to be general knowledge. Here what counts as general knowledge
is irrelevant. Whats relevant is Susans pre-existing personal projects/preoccupations.
Although we, as observers of Susan, wont take CS for granted, Susan does - necessarily, it
being her own current state of mind.
This is not to say that relevant aspects of CS cant be shared. John may have
developed some appreciation of those aspects of CS. They may be mutually manifest to Susan
and John given previous conversation between them. In which case, John may have left the
pile of washing in a prominent place in the kitchen ostensively, i.e. with the (semiotic,
communicative) intention of leading Susan to have T2 and perhaps intending her further to
derive T3 and T4. Given CS, Susan may derive those thoughts whether or not she recognises
Johns intentions. However, following an account by Susan of her preoccupations, John
might instead utter Ive done the washing. The meaning of that utterance/stimulus lies
in the semantic content of Ss thought T2 (actually it lies in UT2.) In that case, if Susan
recognises his intention, John might be said to have conversationally implicated T3 and T4.
17
18
Burton-Roberts
Noel Burton-Roberts
School of English Literature, Language and Linguistics
Percy Building, Newcastle University, Newcastle upon Tyne
NE1 7RU, United Kingdom
noel.burton-roberts@ncl.ac.uk
19
Thanks to Noel Burton-Roberts, Maria Maza, an anonymous reviewer and especially Ian Mackenzie for much
helpful discussion. Any remaining errors are of course my own.
1
Los Fueros de la Novenera son de ndole muy arcaica tanto por lo que se refiere a la lengua como al fondo
jurdico.
2
In the candle ritual, common in many cultures worldwide, the two parties gather with witnesses in a sacred
place to call upon the adjudication of a supernatural entity. Two identical candles are lit at exactly the same
moment, and the supernatural entity in question is deemed to have sided with the party whose candle stays lit for
longer.
3
I concentrate here on the NPI behaviour of ningn and put aside the question of whether it could also function
as a modal polarity item during the early Medieval period, as, in fact, it has no impact on the discussion. (The
modal contexts to be discussed in Section 1.2 below are ones in which ningn could not take on a positive,
indefinite value in other early Medieval Spanish varieties (but see footnote 6 for some additional discussion).)
21
& entraron
dentro & no hallaron ningun hombre
and entered.IND inside and NEG found.IND n-one man
And they went inside and found nobody.
(Gran Conquista de Ultramar, 13th c.)
(2)
(3)
Crucially, however, ningn is also commonly found as a subject when the verb is negated.
(4)
Cases such (4) are taken to fall together with the cases in (1) and (2) under the
assumption that sentential negation raises covertly from T to C, with the result that ningn as
a subject is within the scope of negation at LF.
This classic NPI behavior illustrated in (1)-(3) above is also seen in the Fueros de la
Novenera.
(5)
Nuylla muyllerno ha
poder defer
feyto
ninguno
While some authors (e.g., Martins (2000, 2008); Camus (2006)) have claimed that n-words such as ningn were
both negative and modal polarity items during this period, the data, at least for Old Spanish, are not entirely
clear. Keniston (1937), the source of Martins data for Old Spanish, specifically restricts himself to describing
16th century Spanish, by which time, as Poole (2011) shows, ningns change from polarity item to negative
concord item was already underway. Additionally, a search of the Corpus del Espaol finds only a handful of
examples of n-words with a positive indefinite value outside the scope of negation in unquestionably modal
contexts (particularly questions and conditionals) prior to the 15th century. (Some of the other modal contexts
mentioned by, e.g., Martins (2000: 195), such as the scope of words expressing prohibition or comparative and
before-clauses, are possibly covertly negative.) N-words prior to the 15th century do otherwise behave as
negative polarity items (as opposed to negative concord elements) in all relevant varieties. See Poole (2011) for
further discussion.
4
Unless otherwise indicated, all examples from the Fueros de la Novenera (FdN) are taken from Tilander
(1951), cited by article, while all other unattributed examples from Medieval Spanish are from the Corpus del
Espaol (Davies 2002-).
Poole
22
non deue
ser peynndrado por deuda ninguna
NEG must.IND be pledged
for debt n-one
[An ill person] may not have security taken for any debt
(FdN, 135)
(7)
The Fueros de la Novenera also contains examples parallel with (4) above, in which
ningn (or a variant n-word nuill/nuyll5) can serve as a subject of a negated verb.
(8)
(10)
Though etymologically related to a different Latin word from ningn (NULLUS none, rather than NEC NUS
not (even) one) nuill/nuyll appears to have an identical distribution to ningn in the Fueros de la Novenera.
See Tilander (1951: 26-27) for discussion and see also footnote 16.
(11)
23
In (9)-(11), the subject NP is interpreted as any, but, unlike (4) above, the verb of
which it is a subject, is crucially not negated. As mentioned above in footnote 3, although
there have been some claims that ningn in standard early medieval Spanish is a weak NPI,
and therefore can appear in some contexts with a positive indefinite value without being in
the direct scope of negation, contexts like (9)-(11) are not one of them.6 In these varieties,
negation would be required in (9)-(11) in order to license the any reading for ningn (in
which case it would be anynot, and the examples would be assimilated to (4) above).
The free choice interpretation of ningn/nuill in (9)-(11) is confirmed by a second,
incomplete copy of the Fueros de la Novenera which exists as part of manuscript 13331 in
the Biblioteca Nacional de Madrid. It contains only the first 79 articles of Fueros as an
untitled appendix to the charters of Viguera and Val de Funes, and Hergueta (1900) dates this
manuscript to the 15th century. Tilander (1951: 27) observes that, with only one exception
(which, he notes, probably reflects the original), all of the instances of ningn/nuill hombre in
contexts such as (9)-(11) have been replaced by todo hombre. Todo hombre is in fact found
in a number of places in the A manuscript of the Fueros de la Novenera and is also, in the
context of a law code or set of regulations, the expression that would be used in Modern
Spanish. Leonetti (2009) notes, this bare form of the universal quantifier is predominantly
non-specific, and in distribution resembles a Free Choice Item.
I know of only one counterexample to this claim in the literature. Camus (2006: (40a)) notes the example in
(i).
(i)
Camus (2006: 1181) refers to the context in (i) as imperative modality, and claims that this is a non-negative
context in which ningn is licensed, but I have been able to find only one other example in the Corpus del
Espaol (which from 1200-1400 contains 20 million words) outside the cases discussed here. Martins (2000:
(52)) notes the superficially similar Old Galician-Leonese example in (ii).
(ii)
Que ningun omne que en suas heridades nin en seus omnes metir manoque peyte mil
that n-one man that in his properties nor in his men puts hand that pays thousand
mors. e perda quanto
ouuer
mors. and loses as-much-as might-have
that any person who causes damange to his properties or his menpay a thousand moravedis
(unit of money) and lose everything he might have
In this example ningn is in a topic phrase (as evidenced by the reduplicated complementizer preceding the
finite verb peyte), and in at least one other variety (the later Navarro-Aragonese language of the Fuero General
de Navarra discussed in Section 3) a ningn phrase can very occasionally be a topic even after it has ceased
being licensed in the contexts in (9)-(11).
Most importantly, however, as will be discussed below, seemingly no law charters contained in the
Corpus del Espaol use ningn in this way in this context besides the Fueros de la Novenera, where it is
extremely frequent. This suggests that this free choice use of ningno was not more widely available.
Poole
24
(13)
Si algun villano
se quasare con alguna villana
el [sic] vno de eilos
if some commoner.M SE marry.IND with some commoner.F and
one of them
muere sin
creaturas el que biuo finquare
non deue
tener la heredad
dies.IND without children he who alive remain.IND NEG must.IND have the property
del muerto
of.the dead
If two commoners marry and one of them dies before there are any children, the
survivor may not inherit the property of the deceased.
(Fuero General de Navarra, Version B, 13th c.)
Although the Corpus del Espaol does not explicitly indicate anything concerning the exact dates or locations
of manuscripts which it contains, Version B, from the ADYMTE corpus, is indexed with the 13th century
materials and is replete with characteristic Aragonisms (such as the third-person possessive pronoun lur (from
Vulgar Latin *ILLRUM, which existed alongside illorum (Umphrey 1913))
8
Interestingly, qual quier who-/whichever, a very common free choice item in the 13th century, is not found
either in the Fueros de la Novenera or the Fuero General de Navarra.
25
(Note that contrary to superficial appearances, the verb ha in (15) is a variety of the
impersonal verb hay there is, and therefore the NP is not the subject.)
(15)
In (16), the NP introduced by nuill introduces a large topic, which is then resumed by
the indirect object pronoun le, suggesting very clearly that the NP is topical.
(16)
Poole
26
particular approach to polarity items and the free choice effect, and I argue that their
approach to any can account for the behaviour of ningn in early Navarro-Aragonese.
2.1. The dual behaviour of English any
As is well known, English any exhibits a cross-linguistically peculiar dual behaviour.
On the one hand, it can have a free choice interpretation in certain contexts, as evidenced by
the grammaticality of almost, a traditional diagnostic for free choice (see, e.g., Giannakidou
2001 and the references cited there).
(17) a. (Almost) anyone who wants to drive a truck must obtain a special license
b. John will read his poetry to (almost) anyone
However, any cannot be a true free choice item, as it can be also found in contexts
which are impossible for these elements; for example, within the direct scope of negation or
within the scope of other negative elements in an episodic context.
(18) a. John didnt see (*almost) any of the people.
b. Mary quit her job without (*almost) any notice.
Since almost is impossible in (18a) and (b), we know that we are dealing with
negative polarity any, as opposed to free choice any. Furthermore, as noted by Quer (1999)
(cited in Giannakidou & Quer (2012)), a true Free Choice Item such as Spanish cualquier is
ungrammatical in contexts like (17):
(19)
*Expulsaron del
partido a cualquier disidente.
Expelled.IND from.the party A FC
dissident
Intended: *They expelled FC-any dissident from the party.
(Modern Spanish; Quer 1999)
In other words, descriptively, the facts in (17) and (18) suggest that any in English is
neither a true free choice item nor a true (negative) polarity item.
2.1.1. The quantificational status of any
Contrasts such as the one between (17) and (18) have engendered a debate concerning
the quantificational status of any. 9 Since Ladusaw (1980) and Carlson (1980), polarity any
has been generally taken to be an existential quantifier. For example, in addition to the
argument from almost-modification noted in (18) above, Carlson (1980) also notes other
environments in which NPI any patterns with existential rather than universal quantifiers:
(20) a. He has a little/some/*all/*every courage.
b. He doesnt have any courage.
(21) a. We made a little/some/*all/*every headway.
b. We didnt make any headway.
9
Those who have taken the view that any is an existential quantifier include Kadmon & Landman (1993),
Giannakidou (2001) and Horn (2005); while Davison (1980), Jayez & Tovena (2005) and Dayal (2005) among
others claim that any is a universal quantifier.
27
By contrast, free choice any seems more like a universal quantifier, particularly as the
subject of a generic verb. For example, (22) seems effectively synonymous with either (23a)
or (23b).
(22)
10
11
Poole
28
Under the assumption that ningn/nuill in the Fueros de la Novenera functions like
English any, it will be defined as follows (adapting and amalgamating Giannakidou & Quers
(2012: (41) and (42)) definitions for any):
(25) a. Ningn P is an extensional indefinite of the form P(x), where x is an individual
variable.
b. The x variable is dependent: it cannot be bound by a default existential, unless there is
another nonveridical operator above the existential. If the nonveridical operator is a
Q-operator, then the Q-operator binds the x variable, as is standardly the case with
indefinites.
(26)
Furthermore, I assume the definition of must in (27), adapting von Fintel & Gillies
(2007: (6)) as per their remarks:
(27)
must (B) ()
must (B) () is true in w iff is true in all worlds that are B-accessible from w
B: the conversational background (in the sense of Kratzer), effectively a function
from worlds to sets of worlds
: the prejacent proposition
Putting together the formal definitions in (25)-(27), we see that a deontic modal
context such as that provided by deber must is one in which an element such as ningn can
be licensed. First, deontic modality provides the requisite non-veridical environment. (The
claim it is required or ought to be the case that p does not entail (or presuppose) the truth of
p.) Deontic modality also provides a set of worlds over which ningns domain
12
This is because English any, like ningn in Navarro-Aragonese and unlike true Free Choice Items, is also
licensed within the direct scope of negation, an environment in which conversational implicatures, but not
presuppositions, are cancelled.
29
exhaustification can be evaluated, and this set of worlds is universally quantified over.13 I
therefore take the Logical Form of the sentence in (11) to be (28) (putting aside the precise
question of how the relative clause is represented):
(28)
The universal quantifier provided by the deontic modal binds the dependent variable
introduced by the ningn-phrase, and (28) therefore is interpreted as meaning that, for every
world w which is a member of the set of deontically accessible worlds from our world w
(that is, worlds where the laws are the same as this world) and where x is an uncastrated ram
thief in w, x pays a fine of 5 sueldos in w. (11) further conversationally implicates that the
domain of uncastrated ram thieves is exhausted pairwise with the deontic alternative worlds
that we consider.
This approach to ningn also explains other examples in the Fueros de la Novenera
which have a free choice interpretation, but where, unlike (11) above, universal quantification
is not a plausible alternative analysis. Consider (29)-(31):
(29)
Dos uandos que se mesclen nuyt nin dia et maten ombre ninguno.
Two groups that refl mix.SBJV night nor day and kill.SBJV man n-one
saquen omiziero et peyten homizidio al rey
get.SBJV homicide and pay.SBJV homicide fine to.the king
Two groups of people who come to blows at any time and kill a man are guilty of
homicide and pay the homicide fine to the king.
(FdN, 157)
(30)
(31)
All of these statements are also true of generic contexts, and thus I assume that the analysis of (11) will carry
over straightforwardly to examples such as (10) above.
Poole
30
and pay the homicide fine. Similarly, (31) does not require that a man steal every animal in
order to fall within the scope of the statute. Instead, the statute applies across pairs of
deontically accessible worlds and stolen animals, and is relevant in every one of them.
The conversational implicature of domain exhaustification also explains the ability of
ningn/nuill to be used as a topic (discussed at the end of section 1.2.1 and also relevant for
some of the examples directly above). According to Giannakidou & Quers definition of
ningn in (25) above, in which the phrase ningn P introduces a dependent individual
variable, it seems as though it is the sort of element which should not be able to be topical.
This is made explicit by Giannakidous (2011: 1695) definition of dependent existential (cf.
the definition of dependent variable (25b) above)):
(32)
Nuylla muyllerno ha
poder defer
feyto ninguno
N-one woman NEG has.IND power of do.INF business n-one
amenos de su marido
without her husband
No woman has the authority to conduct any business without her husband
(FdN, 197)
(5) differs crucially from the previous cases discussed above because negation
provides a non-quantificational non-veridical operator. As per the definition in (25b) above,
this allows the dependent variable(s) introduced by the ningn phrase(s) to undergo
existential closure. There is no domain exhaustification, since, following Gazdar (1979) and
Horn (1991), conversational implicatures are cancelled in negative contexts. As such then,
(5) merely claims that it is not the case that there exists a woman x and a business transaction
y such that x has the power to conduct y without her husbands approval.
3. Ningn in later Navarro-Aragonese
The development of ningn in western varieties of Navarro-Aragonese can be traced
through the next centuries by examining two other early charters, the Pamplona manuscript of
the Fuero de Jaca (manuscript D of Molho 1964) and Version B of the Fuero General de
Navarra found in the Corpus del Espaol.14 Jaca was the early administrative center of the
kingdom of Aragn, and its charter dates from the end of the 11th century. Although
originally promulgated in Latin, Manuscript D of Molhos (1964) critical edition of the Fuero
de Jaca is a translation into Navarro-Aragonese done in Navarra around 1340. By contrast,
the Fuero General de Navarra was originally composed in Navarro-Aragonese, and
14
All examples in this section from the Fuero de Jaca are taken from Molho (1964), cited by article.
31
promulgated initially in 1247. As noted in footnote 6 above, Version B from the Corpus del
Espaol seems to originate from an approximately contemporaneous period and region.
Consideration of these two charters suggests that, by the early 14th century at the
latest, Navarro-Aragonese had developed into a standard old Romance variety. Ningn is,
as expected, licensed within the scope of negation (again requiring main clause negation
when in subject position)
(33)
(34)
Also, as expected from the general discussion in Section 1, ningn in later NavarroAragonese is also licensed when in the scope of other negative elements, such as the
preposition sin without
(35)
(36)
However, no examples with ningn as a subject in a deontic modal or generic nonnegative context, corresponding to (9)-(11) above, are to be found in either document.15 This
can be accounted for under the assumption that in later Navarro-Aragonese the implicature of
domain exhaustification is lost. Essentially, ningn is reanalysed as a mere NPI, a
15
While this is true, it should be noted that a handful of examples can be found in the Fuero General de
Navarra in which a ningn phrase is a topic, similar to examples (15) and (16) above. This could be because
ningn has not entirely completed the transition to a simple polarity item. An alternative possibility is that the
explanation for ningns use as a topic both in the Fueros de la Novenera and in the Fuero General de Navarra
lies in the fact that a ningn-phrase can refer to the set that the phrase picks out, in which case, following
Reinhart 1981, it can be interpreted as a topic.
Poole
32
referentially vague indefinite in Giannakidou & Quers terms, in which case it is analysed
as per the discussion of example (5) in the previous section.
Proceeding in parallel with the loss of the free choice use of ningn is the increasing
use of various alternatives for expressing free choice which existed alongside ningn. In
particular, as mentioned in Section 1.2 above, the bare quantifier todo/toda exists alongside
free choice ningn in the A manuscript of the Fueros de la Novenera, as illustrated by (37),
or by example (30) above.
(37)
I put aside here the question of how ningn/nuill itself became licensed in free choice contexts. It is a
somewhat surprising development given that both elements are etymologically related to negative elements in
Latin (NEC NUS not (even) one and NULLUS no/none, respectively). Unfortunately however, it may be that
the lack of sufficiently early Old Spanish texts will make even speculation about this question impossible.
33
process that the non-negative modal contexts are acquired. If the negative polar environment
is generally perceived by learners as most salient for elements which are licensed there, then
the direction of the change seen with ningn in Navarro-Aragonese would be expected. This
would also explain the cross-linguistic trend for indefinites to become more negative (see,
e.g, Haspelmath 1997 and Roberts & Roussou 2003 for discussion).
4. Conclusion
This paper has examined ningn in early Navarro-Aragonese, from both synchronic
and diachronic perspectives. The archaic use of ningn, as exemplified by the manuscript
version of the Fueros de la Novenera in Tilander (1951), straddles the free choice/polarity
divide, in a manner similar to English any (which is well-known for being cross-linguistically
unusual). It is particularly unusual in the context of the history of early Medieval Spanish
polarity items, as ningn does not generally exhibit this behaviour in equivalent contexts in
other early medieval Spanish varieties. I suggested that the behaviour of ningn in the
Fueros de la Novenera provides indirect support for the analysis of English any in
Giannakidou & Quer (2012), in that its properties seem amenable to a similar analysis.
Ningn is an existential, rather than a universal, quantifier and the free choice effect is the
result of a lexical semantic property of ningn by which it conversationally implicates
domain exhaustification. Diachronically, ningn in later Navarro-Aragonese appears to
develop into a standard early Medieval Spanish (negative) polarity item, losing the ability to
serve as the subject of a non-negated generic or deontic modal verb, and this change appears
to be broadly consonant with the diachronic developments of other polarity items crosslinguistically.
References
Camus Bergareche, B. (2006) La expressin de la negacin. In C. Company Company (ed.),
Sintaxis histrica de la lengua espaola. Parte I: La frase verbal, 1165-1252.
Mxico: FCE, UNAM.
Carlson, G. (1980). Polarity any is existential. Linguistic Inquiry 11, 779-804.
Davies, M. (2002-). Corpus del Espaol (100 million words, 1200s-1900s). Available online
at http://www.corpusdelespanol.org
Davison, A. (1980). Any as universal or existential? In Van Der Auwera, J. (ed.), The
Semantics of Determiners, 11-40. London: Croom Helm.
Dayal, V. (2005). The universal force of free choice any. In Rooryck, J & Van
Craenenbroeck, J. (eds), Linguistic Variation Yearbook 2004, 5-40. Amsterdam: John
Benjamins.
Endriss, C. (2009). Quantificational topics. A scopal treatment of exceptional wide scope
phenomena. New York: Springer.
von Fintel, K. & Gillies, A. (2007). An opinionated guide to epistemic modality. In Gendler,
T. & Hawthorne, J. (eds.), Oxford studies in epistemology, vol. 2, 3262. Oxford:
Oxford University Press.
Gazdar, G. (1979). Pragmatics: implicature, presupposition, and logical form. New York:
Academic Press.
Giannakidou, A. (2001). The meaning of free choice. Linguistics and Philosophy 24, 659735.
Giannakidou, A. (2011). Negative polarity and positive polarity: licensing, variation, and
compostionality. In von Heusinger, K., Maienborn, C. & Portner, P. (ed.), The
34
Poole
Kuno, S. (1972). Functional sentence perspective: a case study from Japanese and English.
Linguistic Inquiry 3, 269-320.
Kuroda, S.-Y. (1972). The categorical and the thetic judgement: evidence from Japanese
syntax. Foundations of Language 9, 153185.
Ladusaw, W. (1980). Polarity sensitivity as inherent scope relations. New York: Garland.
Leonetti, M. (2009). Remarks on focus and non-specificity. In Espinal, M. et al. (eds.),
Proceedings of the IV Nereus International Workshop Definiteness and DP Structure
in Romance Languages. Konstanz: Fachbereich Sprachwissenschaft, Universitt
Konstanz.
Martins, A. (2000). Polarity Items in Romance: Underspecification and Lexical Change. In
Pintzuk, S. et al. (eds), Diachronic Syntax: Models and Mechanisms, 191-219.
Oxford: Oxford University Press.
Martins, A. (2008). Investigating language change in a comparative setting. In Almeida, M.,
Sieberg, B. & Bernardo, A. (eds.), Questions on Language Change. Lisbon:
Colibri/Centro de Estudos Alemes e Europeus. 99-116.
Molho, M. (1964). El Fuero de Jaca. Edicin critica. Zaragoza: Escuela de Estudios
Medievales/Instituto de Estudios Pirenaicos.
[facsimile available online at
http://www.derechoaragones.es/es/catalogo_imagenes/grupo.cmd?posicion=3&path=10
1595]
Poole, G. (2011). Focus and the development of n-words in Spanish. In Berns, J. et al. (eds.),
Romance Language and Linguistic Theory 2009, 291-303. Amsterdam: John
Benjamins.
Quer, J. (1999). The quantificational force of free choice items. Paper presented at Colloque de
Syntaxe et Semanique de Paris 99.
Reinhart, T. (1981). Pragmatics and linguistics: an analysis of sentence topics. Linguistica 27, 5394.
Roberts, I. & Roussou, A. (2003). Syntactic change. A minimalist approach to
grammaticalisation. Cambridge: Cambridge University Press.
35
Rooryck, J. (1994). On two types of underspecification: towards a theory shared by syntax and
phonology. Probus 6, 207-233.
Tilander, G. (1951). Los Fueros de la Novenera. Stockholm : Leges Hispanicae Medii Aevi,
II.
[facsimilie
available
online
at
http://www.derechoaragones.es/i18n/catalogo_imagenes/grupo.cmd?path=101969]
Umphrey, G. W. (1913). The aragonese dialect. Bulletin of the University of Washington,
University Studies 5.
Seattle: Washington. [facsimile available online at
http://archive.org/details/aragonesedialect00umphuoft]
van der Wal, S. (1996). Negative Polarity Items & Negation: Tandem Acquisition. Groningen
Dissertations in Linguistics 17.
36
Geoffrey Poole
School of English Literature, Language and Linguistics
Percy Building, Newcastle University, Newcastle upon Tyne
NE1 7RU, United Kingdom
geoffrey.poole@ncl.ac.uk
Poole
LONG-DISTANCE A-MOVEMENT AND INTRODUCING EXTERNAL ARGUMENTS:
PROBLEMS POSED BY THE BEI-CONSTRUCTION IN MANDARIN*
ALISON BIGGS
(University Of Cambridge)
Abstract
The Mandarin bei-construction is traditionally identified as the Mandarin passive, although
more recently it has been described as a non-canonical passive (Liu 2012, Huang 2013).
One property of the bei-construction non-canonical of passives is the possibility of a longdistance dependency between the surface subject (the derived subject) and its gap. A second
non-canonical property is that where the agent is overt, it does not show properties consistent
with demotion.
The first part of the paper examines the availability of the long-distance dependency.
Although in many respects unlike canonical passive constructions, Section 2 shows that a
range of properties including reconstruction and sensitivity to particular verb classes
indicate that the bei-construction involves A-movement; this contrasts with most studies,
which derives the long-distance construction via operator movement. The long-distance
dependency itself is found to be possible with only (certain) ditransitives, and object control
verbs. Section 3 proposes a structure of Mandarin ditransitives and object control verbs, and
from this a derivation of the long-distance bei-construction is sketched, that obviates any
potential intervention effects. According to this analysis, the availability (or not) of longdistance movement in the long and short bei-construction does not follow from any particular
property of the bei-construction itself, but follows from independent properties of its
complement.
The second part of the paper examines the status of the agent of the bei construction in
more detail. The core proposal is that the agent of the bei-construction and external
arguments in Mandarin more generally are not introduced via Voice; this is possible in
Mandarin, but not in English, because in Mandarin the External Argument can be introduced
by a range of lexical heads. Instead, I suggest that VoiceP of the bei-construction obligatorily
selects a CausativeP as its complement, and that the agent (the external argument) is
introduced in its specifier.
1. Introduction
In the Mandarin bei-construction, the Internal Argument of the lexical verb occurs
obligatorily preposed in sentence-initial position. This NP is followed by the morpheme bei.1
* The research reported here was funded by the European Research Council Advanced Grant No. 269752
Rethinking Comparative Syntax. Many thanks to my informants (all mainland speakers of Mandarin), and for
comments and questions from Theresa Biberauer, Lisa Cheng, Anders Holmberg, Adam Ledgeway, Ian Roberts,
Michelle Sheehan, Joel Wallenberg, Norman Yeo, the audiences of the 7th Newcastle-Upon-Tyne Postgraduate
Conference in Linguistics, the LAGB 2012, SLE 45. Comments from Freddy Xuhui Hu and from an anonymous
reviewer have been particularly helpful. Any remaining errors are my own.
1
The co-verbs jiao, derived from the lexical verb call, rang derived from let/allow, and gei derived from
give, are interchangeable with bei (in the long construction) in formal styles. A co-verb is the traditional
Chinese grammar term for functional elements that semantically resemble prepositions, but which distribute
more closely with verbs (from which they derive diachronically).
38
Biggs
In a long bei-construction the agent is overt (3), but in a short bei-construction the agent is
not overt (2).2
(1)
Wo pian le ta.
I
cheat ASP he3
I cheated him.
(2)
(3)
(5)
I assume that the non-overt agent in the short bei-construction is syntactically present. The issue is tangential to
the issues under discussion here.
3
Non-Leipzig glossing: ASP aspect marker, BA object marker, BEI bei-construction, DE modification
marker
4
It is claimed that the short bei-construction (the agentless bei-construction) does not permit long-distance
dependencies (Huang 1999, Ting 1998, etc.), although Her (2009) and Bruening & Tranh (2012) argue that longdistance dependencies with the short bei-construction are possible. The analysis presented here does not force a
correlation between a long-distance dependency and an overt agent. Whether this is correct is an empirical
question.
39
relationship across intervening arguments (4)-(5), are all taken as indicative that the initial-XP
is related to its gap via A-movement.
(6)
Zhangsan bei wo tongzhi Lisi ba zanmei *(ta) de shu dou mai zou le.
Zhangsan BEI I
inform Lisi BA praise *(him) DE book all buy away ASP
Lit: Zhangsan was informed-Lisi-to-buy-up-all-the-books-that-praise-[him] by me.
(Zhangsan was such that I informed Lisi to buy up all the books that praise him.)
(7)
yi-xia.5
once
(Huang et al. 2009: 127)
Although these resumptive pronoun examples are frequently cited in the literature (Huang 1999, Huang et al.
2009 etc.) my informants found these marginal at best; however, the presence of the adverb yi-xia once
improves the example greatly (as reported in Lin 2009). The proposal developed here does not lead to an
obvious solution to this problem. I leave it to future research.
6
Pre-cursors and variants of this account include at least Feng (1995), Chiu (1995), Cheng et al. (1996) Ting
(1998), Tang (2001), Simpson & Ho (2008), Huang et al. (2009), Lin (2009), Liu (2012), Huang (2013).
40
Biggs
The object of an embedded lexical verb (the gap) is base-generated as a null category
pro with a null operator (Huang et al. 2009: 121); the operator adjoins to spec-IP, and from
this position is predicated on the subject of bei.
(8)
(9)
Predication
send proi]]
More recently it has been argued that bei has a chameleonic nature (Huang 2013),
and that its syntax can involve either control or raising, dependent on context, in both its short
and local long uses (Liu 2012, Huang 2013); yet even in this recent discussion the longdistance long bei-construction is held to be derived as in (9) (Liu 2012, Huang 2013: 100).
2.2. Problems with previous accounts
Despite this consensus there are a number of problems with an operator account. The
first set relate to whether the initial XP is base-generated in that position; the second relate to
whether the relationship between the initial XP and its gap involves operator movement.
First, according to this account, the initial XP is an Experiencer selected by bei. This is
surprising given that inanimate arguments such as nei feng xin that letter in (8) are not
typical Experiencers.
The motivation for a thematic relationship between bei and the initial XP centres on
the possibility of subject-oriented adverbs preceding the bei morpheme, where the adverb
scopes exclusively over the initial XP (Cheng et al. 1996, Ting 1998: 339, Huang 1999,
Huang et al. 2009: 115, Liu 2012, Huang 2013).
(10)
Zhangsan guyi
bei Lisi da
Zhangsan intentionally BEI Lisi hit
Zhangsan intentionally got hit by Lisi.
le.
ASP
41
(13)
Thus it is not clear that scope of reference exhibits a one-to-one relationship with
thematic roles, or that it is an arguments thematic role that exclusively determines its
compatibility with a particular class of adverb.
The behaviour of reflexives also suggests that the XP in initial position is not basegenerated in situ. The anaphor ziji self is a subject-oriented reflexive. In contrast to English,
ziji is not clause bound, but can be bound by any argument in its minimal governing category.
(14)
This pattern is also observed in the bei-construction: in the following, the initial XP
reflexive may be bound by either embedded argument:
(15)
[Zijii/j de
haizi]k
[zijii/j de
Wangwu
bei
Zhangsani jiao Wangwui piping
haizi]k le
REFL
DE children BEI Zhangsan tell
criticised
ASP
[Hisi/j own children]k were such that Zhangsani told Wangwuj to criticise [them]k.
If the initial XP were base-generated in initial position, it should behave in the same
way as base-generated topics, which cannot be bound by an embedded argument, as shown in
(16);7 a base-generated topic has the interpretation of aboutness, illustrated in (17).
(16)
(17)
*Zijii de
ben
DEM
CLF
42
Biggs
Finally, the availability of reconstruction for scope not only shows that the initial XP is
not base-generated in situ, but that the long-distance relationship is not derived via operator
movement.
In Mandarin in contrast to English the scope of an active sentence is rigid with
respect to surface ordering (>, *> in (14); >, *> in (15)):
(18)
Yi
ge nuren zhua-zou le mei ge ren.
One CLF woman arrest-go ASP every CLF person
A woman arrested everyone.
(Aoun & Li 1993: 17)
(19)
However, in the bei-construction, both narrow and wide scope readings are available
(>, >):
(20)
Mei ge ren
dou bei yi ge nuren zhua-zou le.9
Every CLF person all BEI one CLF woman arrest-go ASP
Everyone was arrested by a woman.
(Aoun & Li 1993: 17)
(21)
These examples illustrate that existentially quantified subjects not just universal
quantifiers, as reported in Aoun and Li (1993) are rigid for a wide scope reading in active
sentences in Mandarin, but that in the bei-construction, derived existentially quantified
subjects are ambiguous between wide and narrow readings.
Leaving precise analysis of quantifiers aside, the ambiguity indicates that the surface
initial XP is merged in a position that is c-commanded in the domain of the quantifier. This is
unexpected if the initial XP is base-generated in situ as an argument of bei. Furthermore, the
ambiguity indicates that the construction is not derived via operator movement, as operator
relationships do not yield reconstruction.
In short, the initial XP is derived via movement, but crucially not operator
movement.
2.3. A or A-bar movement?
If the bei-construction is derived via movement, the availability of a long-distance
dependency, as in (22), might suggest it is derived by A-movement. Indeed, Peltomaa (2006)
8
The embedded clause use of the embedded clause is intended to prevent interaction of scope with the
Definiteness Effect. For details of the Definiteness Effect in Mandarin, see Huang et al. (2009: 318-28).
9
For Huang (1999) scope is also rigid in the bei-construction. However, my informants obtained the ambiguity
reading reported in Aoun & Li (1993).
43
and Cann & Wu (2010) argue that the bei-construction is (object) topicalisation.
Topicalisation, illustrated in (23), is a canonical example of A-movement.
(22)
(23)
ASP
By contrast, in the bei-construction, most speakers can only prepose the most deeply
embedded object; those speakers that do allow a less deeply embedded argument to be
extracted require the predicate qu go to mark the gap. As shown in (24), no such marking
is required for object topicalisation.
(25)
a. Lisi bei wo jiao *(qu) qing Wangwu ji-zou le nei feng xin.
Lisi BEI I tell go
ask Wangwu send ASP that CLF letter
Lit: Lisi was told-to-ask-Wangwu-to-send-the-letter by me.
(Lisi was such that I told [him] to ask Wangwu to send that letter.)
b. Wangwu bei wo jiao Lisi qing *(qu) ji-zou le nei feng xin.
Wangwu BEI I
tell Lisi ask
go
send ASP that CLF letter
Lit: Wangwu was told-Lisi-to-ask-to-send-the-letter by me.
(Wangwu was such that I told Lisi to ask [him] to send that letter.)
44
Biggs
(30)
45
English where A-movement does not reconstruct (although see Legate 2003), in languages
such as Japanese and German the lack of A-reconstruction (in short scrambling) only
concerns binding; A-movement does allow reconstruction for scope (Wurmbrand 2010). The
reconstruction data in (17)-(20) therefore does not entail that the initial XP is related to its gap
via A-movement.
A second objection might be that the long-distance examples given above are sensitive
to island constraints, another diagnostic of A-bar movement. However, I assume that both Aand A-movement are sensitive to islands, but that (long-distance) A-movement is usually
blocked by other constraints, such as intervening arguments. A possible example would be the
following raising violation of the Complex NP Constraint for English:
(31)
*John seemed that the belief John was smart was false.
(Ian Roberts, p.c.)
This brings us to the final problem with an A-movement account: under familiar
principles of computational locality, A-movement of the NP that letter, from the complement
position of ji-zou send to sentence-initial position, should be blocked by the (many)
apparently intervening NP arguments:
(32)
This problem is examined in detail in the next Section. In brief I will argue that longdistance long bei-constructions are compatible with intervening arguments only in a restricted
set of syntactic environments: with certain ditransitive structures, and with object control
predicates. These syntactic environments are unsurprising if the bei-construction is derived
via A-movement, but unexpected if the structure is derived via operator- or A-movement.
3. Limiting intervention effects
In contrast to previous studies, I will argue that long-distance long bei-constructions can only
evade locality restrictions in a very restricted set of syntactic environments, namely with
double object constructions and with object control verbs.
The absence of A-movement intervention effects with double object constructions is
familiar cross-linguistically, so I discuss this first. Building on this, I sketch a theory of the
derivation of object control in Mandarin, accounting for the possibility of long-distance Amovement through these structures.
3.1. A-movement in ditransitives in Mandarin
As in English, in Mandarin there are many means of realising ditransitive predicates.
The first resembles an English double object construction, with the goal (or recipient)
preceding the theme linearly, with neither argument morphologically marked by a case or
preposition.
(33)
46
Biggs
The second resembles the English prepositional dative, with the theme preceding a
recipient, and the recipient obligatorily marked by the co-verb gei.10 The co-verb functions as
a dative or prepositional marker; gei licenses only recipients goals (including locations) are
not possible.
(34)
Prepositional dative
In the presence of gei, the recipient may also appear immediately post-verbally (but
still preceding the theme).11
(35)
Of interest here are the double object construction (33) and verb+gei (35) structures. In
both, the recipient (the indirect object) precedes the theme (the direct object); although the
recipient is more local to TP, the recipient cannot be preposed; however, the recipient does
not act as an intervener, permitting the theme to participate in the bei-construction.12
(36)
Mali ti le.
Mary
ASP
In the spirit of Paul and Whitman (2010), in both the DOC and the verb-gei variant, in
the active the recipient (the indirect object) is introduced in the specifier of VP; this follows
10
A co-verb is the traditional Chinese grammar term for functional elements that semantically resemble
prepositions, but which distribute more closely with verbs (and which historically derive from lexical verbs).
The co-verb gei is derived from the lexical verb give.
11
There are two additional strategies of realising ditransitives in Mandarin, in which one argument is realised
pre-verbally. A recipient marked by gei may occur pre-verbally; the recipient has a benefactive reading.
i.
Ta gei wo xie le yi feng xin.
Benefactive
gei+R+V+T
He GEI I write ASP one CLF letter
He wrote her a letter
A theme (a direct object) can also occur in pre-verbal position if marked by ba (a functional marker,
grammaticalised from a lexical verb meaning take, hold, handle; see Li 2006).
ii.
Ta ba
shu song
gei wo.
BA construction ba+T+V+R
He BA book give
GEI
I
He gave the book to me
12
The one exception seems to be wei feed, which can only be realized as a DOC, yet only permits passivisation
of a recipient. Under the account presented here, wei should be ungrammatical in a long-distance beiconstruction, as the recipient should act as an intervener.
47
from the low applicative semantics of the construction (in the sense of Pylkknen 2008) (for
full details see Paul & Whitman 2010: 266-67).
The theme (the direct object) is initially merged as a complement of V. The recipient
enters the derivation with an u[Case] feature (see also Georgala 2012), which must be valued
for convergence of the derivation. (The presence of u[Case] contrasts with English, where the
indirect object and only the indirect object in American English dialects of a double object
construction can undergo A-movement). Paul & Whitman (2010) present convincing
evidence from the placement of adverbs and the position of distributive quantifiers that the
recipient raises out of VP; however, the crucial fact here is that the recipient enters an Agree
relation with the Applicative projection.
(38)
[TP I [ASP0 sell- GEI-ASP [APPL Mary [APPL0sell-GEI [VPMary [V0 sell [one-CLF-book]]]]]]]
(39)
Mali ti le.
Mary
ASP
(41)
[TP One CLF book [VOICE0 bei [CAUSEP I [ASP0 sell-GEI [APPL Mary [APPL0sell-GEI [VP
Mary [V0 sell [one-CLF-book]]]]]]]]]
48
Biggs
(42)
The recipient (Mary), bears a u[Case]. It enters an Agree relation with the first
available Probe it encounters, the Applicative head; agreement with ApplP values its u[Case].
The inaccessibility of the recipient in the context of the bei-construction follows
straightforwardly: once the of the recipient has been valued, it is no longer active (Chomsky
2001) and cannot participate in any further Agree relations (it cannot raise again); nor is it a
potential intervener for other Agree relations (Baker 1988). The theme, lacking some u[F]
presumably u[Case] is free to raise past the indirect object to initial position.
3.2. A-movement and object control in Mandarin
As discussed in Section 2, the long-distance bei-construction requires a non-finite
complement. In addition, the long-distance construction appears to only be possible in the
context of object control verbs, such as jiao order, qing ask/invite, tuo request (from),
and pai send:
(43)
If this generalization is correct, other verbs that should be compatible with the longdistance bei-construction include quan urge, and bi (try to) force, bipo force (based on
Grano 2012: 236). If object control predicates are replace by subject control predicates, the
bei-construction is no longer possible:
(44)
49
(46)
Object control predicates, or at least those under discussion here, are triadic: in
addition to two DP arguments, they take an event as a complement.
Following much previous work, including Larson (1991), I assume that the structure
of the object control predicates is that of double object constructions, with the event as the
internal argument of the control predicate. The key difference between the two structures,
following Larson (1988, et seq.), is the introduction of a causative head in addition to the
applicative head; this anticipates discussion in Section 4.
(47)
(48)
The object control predicate jiao tell is generated with the functional heads ApplP
and CauseP, through which it undergoes head movement (to CauseP) in order to obtain the
relevant semantics.
The argument undergoing causation, Lisi, enters an Agree relation with ApplP, in the
same way described above for double object constructions. This argument is also in a
predication relation with PRO; PRO is base-generated as the external argument of the event
complement.
The derivation of the bei-construction proceeds as in the same way as for the double
object construction. The recipient (Lisi), as before bears u[Case]. It enters an Agree relation
with the most local Probe ApplP, rendering it inactive for Agree relations, and is no longer
acts a potential intervener.
50
(49)
Biggs
(50)
3.3. Summary
Long-distance A-movement in bei-constructions escapes intervention effects only in
the restricted context of (particular) ditransitive constructions, and with object control
predicates, which have the same underlying structure as double object constructions. It is
therefore not a property of the bei morpheme itself that permits a long-distance relationship;
rather, long-distance A-movement is always possible, but is ruled out in the context of active
intervening arguments. Where apparently intervening arguments are inactive for example,
because a more local (Applicative) head has already valued its Case feature intervention
effects are obviated, and a long-distance relationship is possible.
4. The introduction of the agent
We now turn to a second property of the bei-construction, unrelated to the issue of Amovement.
Although the bei-construction is generally referred to as the Mandarin passive, the
overt agent in the (long) bei-construction does not undergo demotion to an oblique (noncore argument) status, in contrast to the agent in, for example, the English be-passive. The
remainder of the paper explores the syntactic status of the agent.
(51)
(short bei-construction)
(52)
51
(long bei-construction)
4.1. The syntactic status of the agent and the taxonomy of passives
Evidence that the agent retains its subjecthood or that it is not demoted comes
from the fact that the subject-oriented anaphor ziji self is sensitive to the agent in the beiconstruction; adjuncts, such as the Prepositional Phrase gen with, cannot bind ziji self.
(53)
(54)
In addition, bei and the DP that follows it do not distribute with other adjuncts.
Adjuncts in Mandarin, such as PPs, can prepose freely to sentence-initial position, and more
generally do not exhibit a fixed ordering, as illustrated in (55). This flexibility is not true of
bei and the following DP, which must occur in a fixed position, as in (56) (Li 1990: 157-64;
Huang et al. 2009: 116-9):
(55)
(56)
Finally, coordination demonstrates that the agent forms a constituent with the
predicate that follows it. This is unexpected if the agent is an adjunct, or some other oblique.
(57)
Ta bei Lisi
He BEI Lisi
ma
le liang sheng, Wangwu ti
le san xia.13
scold ASP two CL
Wangwu kick ASP three times
13
52
Biggs
(58)
diagram in (24):
(24)
.... VP
!
#
" Causative-agentive
#
$
V'
V
%hit
!
#
#
&strike "
#
'blow #
$
VP
DP
the window
V'
V
break
!
#
#
" Inchoative-resultative
#
#
$
Optional
Incorporation
14
53
(60)
As such, the following cartography of the Mandarin vP shell seems plausible, where
Voice precedes causation.
(61)
I propose that the structure in (61) is always present in the bei-construction, and that,
regardless of the overt or non-overt status of the CAUSE head, the agent is always introduced
the specifier of CauseP.
Evidence that CauseP is always present follows from the semantic restrictions on the
types of events with which the bei-constructions (both long and short) are compatible.
For example, true (individual-level) stative predicates are incompatible with the beiconstruction.
(62)
54
(63)
Biggs
Under the lexical decomposition hypothesis, the minimal pair (57) and (58) are
distinguished by their (non-overt) decomposition: (58) but not (57) includes a (null) causative
light verb. The compatibility of (58) but not (57) with the bei-construction follows if the bei
morpheme obligatorily takes a causative complement.
Crucially, it is not the semantics inherent to individual stage-level predicates that
prevents compatibility with the bei-construction. For example, in the Mandarin resultative-de
construction, the modifier de is added to a verb, introducing a clause that describes the result
of the event denoted by V (Huang et al. 2009: 84). Cheng (2008) argues that this result-de
introduces a CauseP. Thus, under the hypothesis presented here, any predicate modified by
resultative-de should be compatible with the bei-construction. The following examples
demonstrate that once an individual-level stative predicate is placed in a resultative-de
construction, it becomes compatible with the bei-construction, as predicted.
(64)
*Shoupa
bei Zhangsan ku-shi le.
Handkerchief bei Zhangsan cry-wet ASP
The handkerchief became wet from Zhangsan crying.
(65)
Shoupa
bei Zhangsan ku-de
dou shi le.
Handkerchief bei Zhangsan cry-RESULT all wet ASP
The handkerchief was made all wet by Zhangsan from his crying.
(Based on Huang 1992: 125)
(67)
le.
55
If the introduction of the external argument is distinct from the presence of Voice in
Mandarin, then there is no reason to expect that the status of the agent should be affected by
the presence or absence of the Voice projection. It is for this reason that the bei-construction
does not yield demotion of the agent, but instead has the status of a full syntactic argument,
as observed in Section 4.1.
5. Conclusion
Having established that the long-distance dependency possible in the long beiconstruction is derived via A-movement, the first half of the paper examined the availability
of the long-distance dependency more closely. The long-distance dependency is in fact
possible with only a subset of predicates, which share the property of deactivating one of their
internal arguments. According to this analysis, the availability (or not) of long-distance
movement in the long and short bei-construction does not follow from any particular property
of the bei-construction itself, but follows from independent properties of its complement.
The second part of the paper examined the status of the agent of the bei-construction.
The core proposal made is that the agent of the bei-construction and external arguments in
Mandarin more generally are not introduced via Voice; this is possible in Mandarin, but not
in English, because in Mandarin the External Argument can be introduced by a range of
lexical heads. Instead, the VoiceP of the bei-construction obligatorily selects a CausativeP as
its complement, and the agent (the external argument) is introduced in its specifier. It follows
that there is no interaction between the Voice projection headed and the agent, and thus the
agent does not undergo demotion in the bei-construction.
References
Aoun, J. & Li, A. (1993). Syntax of scope. MIT Press: Cambridge, MA.
56
Biggs
57
Kratzer, A. (1996). Severing the external argument from its verb. In Rooryck, J. & Zaring, L.
(eds.), Phrase structure and the lexicon, 109-137. Dordrecht: Kluwer.
Larson, R. (1988). On the double object construction. Linguistic Inquiry 19, 335-391.
Larson, R. (1991). Promise and the theory of control. Linguistic Inquiry 22, 103-139.
Lasnik, H. & Fiengo, R. (1974). Complement Object Deletion. Linguistic Inquiry 5, 535-571.
Laws, J. & Yuan, B. (2010). Is the core-peripheral distinction for unaccusative verbs crosslinguistically consistent? Chinese Language and Discourse 1, 220-263.
Legate, J. A. (2003). Some interface properties of the phase. Linguistic Inquiry 34, 506515.
Li, A. (1990). Order and Constituency in Mandarin Chinese. Kluwer: Dordrecht.
Lin, J. (2001). Light Verb Syntax And The Theory Of Phrase Structure. Ph.D dissertation,
University of California, Irvine.
Lin, J. (2009). Licensing gapless bei passives. Journal of East Asian Linguistics 18, 167
177.
Liu, N. (2012). The syntactic structures of the Chinese bei passive and the English be-passive.
In Proceedings of GLOW in Asia IX.
Simpson, A. & Ho, T. (2008). The comparative syntax of passive structures in Chinese and
Vietnamese. In Chan, M. & Kang, H. (ed.), Proceedings of the 20th North American
Conference on Chinese Linguistics, 825-42. Columbus, Ohio: Ohio State University.
Paul, W. & Whitman, J. (2010). Applicative structure and Mandarin ditransitives. Duguine,
M., Huidobro, S. & Madariaga, N. (eds.), Argument structure and syntactic relations:
a cross-linguistic perspective, 261-282. Amsterdam: John Benjamins.
Peltomaa, M. (2006). Pragmatic nature of Mandarin passive-like constructions. In Werner, A.
& Leisi, L. (eds.), Passivisation and typology: form and function, 83- 114.
Amsterdam: John Benjamins.
Pylkknen, L. (2008) [2002]. Introducing arguments. Cambridge, MA: MIT Press.
Shibagaki, R. (2010). Mandarin secondary predicates. Taiwan Journal of Linguistics 8, 57-94.
Simpson, A. & Ho, T. (2008). The comparative syntax of passive structures in Chinese and
Vietnamese. In Chan, M. & Kang, H. (ed.), Proceedings of the 20th North American
Conference on Chinese Linguistics, 825-42. Columbus, Ohio: Ohio State University.
Tang, S. (2001). A complementation approach to Chinese passives and its consequences.
Linguistics 39, 257-295.
Ting, J. (1998). Deriving the bei-construction in Mandarin Chinese. Journal of East Asian
Linguistics 7, 319-354.
Wurmbrand, S. (2010). Reconstructing the A/A-distinction in reconstruction. In Stevens, J.
(ed.), Penn Working Papers in Linguistics 16.1.
58
Alison Biggs
Department of Theoretical and Applied Linguistics, University of Cambridge
886 Kings College
Cambridge
CB2 1ST
UK
arcb2@cam.ac.uk
Biggs
60
Syed
phonemes are also explained to them. They are also encouraged to speak British English (BE)
with as much native-like accuracy as possible. But despite all this effort at academic level,
practically what the students normally hear around them in the society is PE which substitutes
English dental fricatives [ ] with dental stops [t h d ]. The current study aims to test the
perception and production of advanced Pakistani learners of English who, after having
acquired PE, are taught BE. The research question is whether such students remain faithful to
the already acquired PE and produce English [ ] as stops or whether they acquire BE,
accurately producing these sounds as dental fricatives.
The remainder of this paper is divided into five sections. The next section provides a
brief introduction of the basic ideas and predictions of the speech learning model (SLM:
Flege 1995) relevant to the current study. The findings of the study will be analyzed in light
of the predictions of the SLM. Section 2 provides details of the participants and experiment
conducted for data collection. The data will be presented in section 3. Section 4 is based on a
detailed analysis and discussion of the results. The analysis will be done using classical
Optimality Theory (Prince & Smolensky 2004). Section 5 concludes this paper.
2. The speech learning model
Several models of second language acquisition have been presented to account for the
acquisition of L2 sounds. The perceptual assimilation model commonly known as PAM (Best
1994, 1995), feature model known as FM (Brown 1998, 2000) and the speech learning
model (Flege 1995) are some of the most well known models of second language acquisition.
The findings of the current study will be analyzed in light of the predictions of the speech
learning model so only the speech learning model will be discussed in detail. The reason for
this is that the SLM is based on phonetic perception of L2 learners which is the main concern
of this study. The SLM also predicts a correspondence between perception and production
whereas other models (e.g. FM, PAM, etc.) primarily account for perception of L2 learners.
Since the objective of the current experiment is to study a relationship between the perception
and the production of L2 sounds, the speech learning model suits the study.
The speech learning model divides L2 sounds into 'new', 'similar' and 'identical' out of
which the similar ones are considered most difficult to perceive for L2 learners (Flege 1995).
The SLM develops different hypotheses about different learning scenarios. According to the
model, learners perceive gradient phonetic details of L2 sounds. This implies that learners
can perceive allophonic variance of L2 sounds. If L2 learners perceive a difference between
an L2 and the closest L1 sound, they develop a separate phonetic representation for the L2
sound. The phonetic categories developed by bilinguals may be different from those of
monolinguals, either for maintaining contrast between the L2 and the corresponding L1
sound, or if the categories of bilinguals are based on features different from those of the
monolinguals of the L2. On the other hand, if L2 learners do not perceive a difference
between L2 and the corresponding L1 sounds clearly, they equate the two sounds. Flege calls
this a mechanism of equivalence classification (Flege ibid.). In case of equivalence
classification, the establishment of a new phonetic category for the new L2 sounds is blocked.
In this case, learners develop the same phonetic representation for the L2 and the closest L1
sound.
Defining, the concept of equivalence classification, Flege comments that equivalence
classification between two sounds may be of two types: weak equivalence classification and
strong equivalence classification. In case of weak equivalence classification, learners can
perceive a little difference between an L2 and the closest L1 sound but they cannot perceive it
clearly enough to develop a new phonetic representation for the L2 sound. Sometimes weak
equivalence between two sounds leads to the development of a phonetic representation which
61
is a merger of the two sounds (Flege 1987). In case of a strong equivalence classification,
establishment of a new phonetic category for L2 sounds is totally blocked. In that case,
learners perceive the L2 sound as the closest L1 sound and have the same representation in
perception and production of the two sounds.
As pointed out earlier, the speech learning model claims that learners perceive
gradient phonetic details of L2 sounds. On the other hand, it also claims a correspondence
between the perception and production of L2 sounds. This means that, according to the SLM,
there is a symmetry between phonetic perception and phonological (lexical) production
whereas it is well-known in the literature that sometimes phonetic and phonological factors
play critically different roles in L2 acquisition. The interface of phonetics and phonology and
its influence on the perception of L2 sounds is well-established in the field of second
language acquisition. See Boersma & Hamann (2009) for a detailed discussion on phoneticsphonology interface. The current study is conducted with a view to determine how well the
speech learning model can account for the perception and production of English dental
fricatives by Pakistani learners.
3. The current study
A group of thirty Masters students of English were selected for this experiment. The
participants spoke the same L1 and were studying in the same university in Pakistan. At the
MA level the students have a lot of opportunity to speak and listen to English. In Pakistan,
English is taught as a compulsory part of courses of studies from primary school regardless of
the discipline of the student. Therefore, the participants of this study were taught PE up to
Bachelor level. At the time of experiment, they were registered as regular students of the MA
English programme in their second year.1 In the Masters programme, although they still hear
PE around them, they are taught the phonology of British English (BE) in class. According to
their statements, they speak English for an average of 2 hours (standard deviation 1.20) and
listen to it spoken by non-native Pakistani speakers for an average of 2.56 hours (standard
deviation 1.65) daily. Their average age was 21.97 (standard deviation 2.63) years.
Before the main experiment, the students were asked to provide information about
their academic and linguistic background through a questionnaire. The above information
was elicited through the questionnaire. They were also requested to provide written
permission for recording of their voice and using the data for research purposes
anonymously. Permission to conduct the experiment within the campus was also obtained
from the heads of the institutions where the participants were studying.
Afterwards, the learners were given a perception and production test. The perception
test had three tasks: an identification task, a 3 alternative forced choice (hereafter 3AFC)
discrimination task and an AX discrimination task. In the identification task, the participants
were asked to identify and write, in Urdu2 and English in the relevant columns of a given
answer sheet, which consonants they heard between two low vowels in the stimuli. The
stimuli of the test carried along with other distracters, [aa] and [aa] produced by a female
native speaker of English aged 27. This task had 3 repetitions providing 90 responses (30
participants * 3 repetitions). Before being used in the test, the stimuli were played to four
native speakers of English who confirmed that the consonants were produced in correct
1
Syed
62
English pronunciation. It is important to point out an interesting fact that in Urdu, which is a
national language of the participants and Saraiki, which is their mother tongue, the letters for
dental fricative consonants [ ] exist but the sounds themselves do not exist. This is because
Pakistani languages got their script from Arabic and the sounds [ ] do exist in Arabic.
Therefore, the learners were aware of the existence of the letters for dental fricatives. Since
they had been taught the phonology of British English, they also knew the IPA symbols for
these English phonemes. Still, they were asked to point out if they felt that they were
listening to a consonant in the stimuli for which they did not have proper letter available in
Urdu3 and English. None of the participants pointed out such a difficulty. The purpose of
using Urdu letters along with English ones is that the English alphabet system lacks proper
letters for dental fricatives. Most of the participants preferred to write their answers in
English letters (not in IPA symbols). Since English lacks letters for dental fricatives, those
who wrote their answers correctly, wrote 'the' on listening to [] sound. Since Urdu has
proper letters for dental fricatives, the participants were also asked to write their answers in
Urdu. Their responses in Urdu letters further confirmed whether the participants identified
the target sounds correctly or not.
In the 3AFC task, they listened to sets of sounds including those listed below:
(1)
[aa]
[aa]
[aa]
[aa]
[ava]
[asa]
[ada]
[aga]
[aga]
[aba]
[aza]
[afa]
The learners were asked to choose by ticking one of the following on the given answer sheet
after hearing each of the sets of the above sounds:
1. The consonant in the first syllable is the same as the consonant in the second
syllable.
2. The consonant in the first syllable is the same as the consonant in the third syllable.
3. The consonant in the first syllable is different from the consonants in both the
second and the third syllable.
Some other sets of sounds as distracters were also added to the above list. There was
no repetition in this task so, in all, 30 responses were obtained in this task. The purpose of
this task was to test whether the participants could differentiate English [] from [s] and [f],
and [] from [z] and [v]. Previous studies report substitution of English dental fricatives with
either labial fricatives (Bell & Gibson 2008) or coronal fricatives (Hatten 2009, Lombardi
2003, Weinberger 1997). The third, apparently irrelevant stop given in the list of each of the
sets of the stimuli in (1) above was included with the target sounds in the stimuli to make the
task more difficult so that the participants would maintain their concentration and decide on
the basis of careful listening. All the stimuli were prepared in the voice of the same native
speaker of English.
The third task of the perception test was an AX discrimination task. All the
instructions to the participants were given by the author in their L1 which is also the L1 of the
author. The instructions were also written on the answer sheets in English. In this task, two
sounds were played and the participants were asked to determine by ticking in the relevant
column of the answer sheet if the consonants in the two sets of sounds were the same or
different. The first set of sounds in the stimuli in this task was either [aa] or [aa] spoken by
3
The L1 of the learners also uses Urdu script with some modifications.
63
the same female native speaker of English followed by [at ha] or [ad a] spoken by a female
monolingual native speaker of Saraiki which is the L1 of the participants of this study. The
L1 sounds were recorded in the voice of a female native speaker who speaks the same dialect
of Saraiki which the participants speak.4 Before the Saraiki sounds were used in the test, two
native speakers listened to them and confirmed that the sounds were produced with accurate
pronunciation. The purpose of this task was to determine whether the participants could
discriminate between English dental fricatives and Saraiki dental stops or if they confuse
them. Recall that in PE the English dental fricatives [ ] are produced as dental stops.
Aspirated and unaspirated dental stops exist in Urdu (Shackle 2007), the national language of
the country and Saraiki (Shackle 1976), the L1 of the participants. The participants were
instructed to neglect the differences in the tone, accent, pitch, etc. of the speakers and decide
only on the basis of the consonants between two as in the stimuli. This task had three
repetitions. Some distracters were also included in the list of stimuli. The stimuli were
presented in random order. A total of 90 responses (30 participants * 3 repetitions) were
obtained from this task.
In the production test, the students were asked to read in natural, normal speed the
words given in a list which contained, along with other English words, these and thieve
each three times. Previous research shows that the front vowel is neutral in its effect on the
production of L2 sounds (Syed 2011). Therefore, such words were selected for production in
which the target sounds were followed by a tense front vowel. These words start with dental
fricatives and are quite commonly used words with which the participants were already
familiar. These words carry English dental fricatives on onset of the syllable which is quite a
prominent position (Archibald 1998). Therefore, they provide suitable context for the study
of English dental fricatives. Four native speakers of BE (aged 23, 32, 40, 42) were asked to
evaluate the productions of English dental fricatives in the words these and thieve5 on a
Likert scale ranging from 5 to 1. On the scale, 5 was native-like, 4 was near native-like, 3
different from natives but understandable, 2 hardly understandable and 1 was
unintelligible. The judges were asked to evaluate the productions only on the basis of the
pronunciation of word-initial consonants in the words these and thieve without being
influenced by the (in)accuracy of productions of the other phonemes in the target words. The
scores awarded by the four judges were averaged. In the following analysis only averaged
results will be presented.6
4. Presentation of the data
The data will be presented in the following two subsections. First, the results of the
perception test are presented, followed by those of the production test. The results will be
analyzed in Section 4.
According to Shackle (1976) there are six dialects of Saraiki. All participants of this study speak central dialect
of Saraiki.
5
Only two words, namely 'these' and 'thieve' were selected as stimuli, one of which is a verb. The main reason
for this is to control or equal the effect of the adjacent vowel on the production of consonants. If a different
word starting with a voiceless dental fricative followed by another vowel e.g. [u] were selected as a stimulus, it
would be difficult to determine whether the difference in production of voiced and voiceless dental fricatives is
on account of learning or on account of the adjacent vowel. Therefore, 'thieve' was selected as a stimulus, which
like 'these' starts with a dental fricative, is followed by the same front vowel like 'these' and also consists of one
syllable. Therefore, the purpose of selecting phonologically similar words is to control the effect of different
factors like familiarity, adjacent vowel, word size, etc.
6
-fronting is common in English native speakers living in and around Eastern London. The judges were tested
for -fronting and it was confirmed that they produced the voiceless dental fricative as [], not as [f].
Syed
64
AX discrimination (N=90)
L1 [at ha]
L1[ad a]
L2 [aa]
L2[aa]
83
79
The above table shows that only 7 out of 30 students discriminated [] from [f] and 14
of them discriminated it from [s]. Similarly, 14 participants discriminated [] from [v] and 20
of them discriminated it from [z]. In this way, the performance of the participants was better
on the discrimination of voiced dental fricative than the voiceless one. They assimilated both
target sounds with the closer labio-dental fricatives in more trials than with the coronal
fricatives.
The results of the AX discrimination task show that the participants discriminated the
voiceless dental fricative from the corresponding L1 stop 83 times and the voiced dental
fricative from the corresponding L1 stop 79 out of 90 times. The excellent performance of the
participants in the AX discrimination task indicates that they can discriminate between
English dental fricatives and the L1 dental stops.
4.2. Production test results
The pariticipants productions of the target phonemes were evaluated by four nativespeaker judges. Table 2 shows the averaged scores awarded by the judges.
Table 2: Production test results
Stimuli
[]
[]
Minimum
2.00
2.00
Maximum
4.33
4.67
Mean
3.24
3.51
Std. Deviation
.50
.50
The results show that the participants performed better on the voiced [] than on voiceless [].
The variance in the means of the two scores is significant (Z= -2.81, p=.005). Overall, the
participants could not obtain an average score of 4 which means they were not rated by the
7
8
In one of the trials, the English voiced fricative was identified as an irrelevant consonant.
Recall that the participants presented their answers by writing the consonant in English and Urdu.
65
[]
0
8
19
3
0
[]
0
2
24
4
0
Category
Unintelligible
Hardly understandable
Different from natives but understandable
Near native-like
Native-like
The above table shows that none of the participants were rated as native-like in the
native evaluation and only 3 participants for the voiceless and 4 for the voiced dental fricative
were considered as near native-like. The above tests indicate that the participants were very
poor in the perception and production of English dental fricatives. They confuse English
dental fricatives mostly with labio-dental fricatives [f v] in perception. However, they can
clearly differentiate between English dental fricatives and the corresponding L1 dental stops.
Their average production is mostly far from even the near native-like category.
For a clear understanding of the pronunciation of the learners, all recordings of the
participants productions were acoustically analyzed using Praat (Boersma & Weenink 2012).
The analysis shows that all dental fricatives were produced as stops by the participants. The
following spectrogram and wave-form provide an example of the productions of the voiced
dental fricative [] of one of the participants.
Figure 1: Spectrogram of the word these
Burst
Pre-voicing
Vowel
Fricative
Syed
66
In the wave-form, pre-voicing is quite clear before the burst. It is also visible in the
shape of darkness at the bottom of the spectrogram pointed out by the arrow. During the burst
phase, the vocal folds of the speaker stopped vibrating briefly and the burst was accompanied
for a short time by a strong puff of air which is visible with the burst on the spectrogram but
it may not be considered a fricative gesture because normally a continuant maintains frication
for a considerable amount of time which is reflected in the shape of the concentration of
acoustic energy normally on the upper half of the spectrogram depending upon the place of
articulation. The final part of the spectrogram is a fricative [z] which clearly shows a
concentration of acoustic energy on the upper half of the spectrogram for a considerable time
period. There is no such acoustic correlate in the beginning of the word. This shows that the
initial consonant of the word these was produced as a stop with pre-voicing. In the L1 of the
learners, voiced stops (including [d ]) are produced with pre-voicing (Syed 2013). This means,
the participants substituted voiced English dental fricatives with the voiced dental stops of
their L1. For further confirmation of this, voice onset time for all productions of the target
phonemes were measured which are given in the following table.
Table 4: Voice onset time
Sound
[] produced as [d ]
[] produced as [t h]
N
90
90
Minimum
-173.67
14.00
Maximum
-4.33
104.33
Mean
-111.2667
67.4333
Std. Dev.
40.29
18.63
The above table shows that the voiced dental fricative [] was produced as pre-voiced
stop with negative VOT and the voiceless dental fricative [] was produced as voiceless
aspirated dental stop [t h]. These results are analyzed in the following section.
5. Analysis and Discussion
The results presented in the previous section show that Pakistani learners perceive
English dental fricatives either as labio-dental fricatives or coronal fricatives. They perceived
dental fricatives as labio-dental fricatives in more trials than as coronal fricatives. This shows
that most of the learners perceived the stimuli on the basis of acoustic cues. It has been
pointed out in previous studies (Simon 2009) that L2 learners whose L1s do not have English
dental fricatives perceive these consonants as labial on the basis of acoustic cues because the
labial [f v] are acoustically very similar to [ ] (Wester, Gilbers & Lowie 2007) and if they
perceive these sounds as [s z] this is on account of phonological reasons (Simon 2009)
because both pairs of sounds are attached to the same place node i.e. CORONAL (Clements
& Hume 1995). This means that most of the participants perceived the stimuli on the basis of
acoustic phonetic cues. Better performance of the learners in the discrimination rather than
the identification test is understandable since in a discrimination test learners simply
determine on the basis of acoustic cues the difference between two sounds and respond with
yes/no type answers, but for identification of those sounds they have to retrieve from their
L2 phonemic inventory the sound matching the stimuli which is only possible if the learners
have clearly established a separate phonetic category for the sounds in their L2 phonemic
inventory. Therefore, a better performance by L2 learners in discrimination rather than
identification tests is always predictable (Archibald 2005).
In this way, most of the participants neglected the feature [coronal] and
perceived
English dental fricatives as labials on account of acoustic similarity. Some of them, who
perceived the stimuli as coronal fricatives [s z], only neglected the feature [distributed] and
67
[strident] since the English dental fricatives [ ] are [+distributed, -strident] whereas [s z]
are [-distributed, +strident].
An interesting factor in the results is that most of the participants differentiated
English dental fricatives from the L1 dental stops accurately (see table 1). However, in
production, they substituted these English phonemes with dental stops. Since the stimuli were
nonce words, the students perceived them purely on the basis of phonetic cues. The stimuli
for the production test, however, were meaningful words of English, so the participants
produced them on the basis of their lexical (phonological) understanding. This interface of
phonetics and phonology indicates that although the learners can perceive the difference
between fricatives and stops phonetically, they produce English dental fricatives as stops
because under the influence of their already acquired Pakistani English they have developed a
specific phonemic inventory of their own which is dominant in their production. This already
acquired representation of English dental fricatives as stops only activates in a lexical or
phonological environment.
The speech learning model does not seem to account for the above situation
successfully. The study shows asymmetric results in perception and production, whereas the
model predicts a symmetry between perception and production. Therefore, the results of the
study do not accord with the SLM predictions since, i) the SLM assumes that in learners L2
phonemic inventory representation of sounds is based on only acoustic phonetic cues whereas
the above results show that the Pakistani learners perceived L2 sounds on phonetic grounds
but produced them on lexical and phonological grounds; ii) the SLM predicts a
correspondence between perception and production, however the results show an asymmetry
between the perception and production of the participants. According to the SLM, L2 learners
perceive gradient phonetic details of L2 sounds and develop a phonetic category for the
sound on the basis of their perception. Their production corresponds with the phonetic
category developed by them in their L2 phonemic inventory. In other words, according to the
SLM, phonological production is based on phonetic perception. However, the results of this
study show a different picture. This phonetics-phonology interface points out that a revision
in the speech learning model is required to incorporate the role of phonology in L2
perception.
The substitution of dental fricatives with dental stops is because of universal
markedness hierarchy since stops are the most unmarked of the consonants (Simon 2009).
There are examples of substitutions of English dental fricatives with stops (Lombardi 2003,
Yildiz 2006), but the current substitution is triggered more under the influence of PE than the
universal markedness of coronal stops. This is because the learners produced the English
voiceless dental fricative [] as the voiceless aspirated stop [t h], not as the unaspirated stop
[t ]. We already know that Pakistani speakers substitute English [] with their L1 [t h]
(Mahboob & Ahmar 2004, Rahman 1991). If it were a substitution purely on the basis of
markedness only, the participants would have substituted [] with unaspirated [t ] which also
exists in the L1 of the learners. But the mean VOT for the first consonant in the word thieve
produced as stop by the participants is 67.43 ms (see table 4). This is a VOT value for
aspirated stops not for unaspirated ones. This confirms that the substitution of the dental
fricative with the dental stop is done under the influence of PE, not because of universal
markedness.
An interesting question arises: why do Pakistani learners substitute English [] with
the L1 [t h] but not with [t ] which also exists in the L1 phonemic inventory and is more
unmarked than aspirated [t h], since [t h] along with primary articulation also involves a
secondary articulation. The answer is that it is because of the influence of English
orthography which does not have a separate letter for the voiceless dental fricative. The L1
and Urdu (the national language of Pakistan) both have two letters for voiceless dental stops,
Syed
68
one corresponding to dental stop and another 'h' which represents the aspirated dental
fricative in their L1. Since in English the letters 'th' are used to represent the voiceless dental
fricative, therefore, there is a strong probability that the participants substituted the fricative
with the stop under the influence of orthography. The influence of orthography on production
is already attested in literature (Bassette 2009, Erdener & Burnham 2005, Escudero &
Wanrooij 2010, Nicol & Barker 2010, Hayes-Harb et. al 2010, LaCharite & Paradis, 2005,
etc.). Another probability is that the learners perceive the existence of frication in the input
i.e. []; however, since they cannot produce the dental fricative, they substitute it with a stop
but to compensate for the loss of frication they add aspiration. This is because of the fact that
acoustically, frication and aspiration are similar.9 However, this is a mere hypothesis which
needs confirmation.
In summary, the participants of this study perceived English dental fricatives mostly
as labial fricatives but produced them as stops. Now the results are analyzed using classical
Optimality Theory (Prince & Smolensky 2004). The L2 phonemic inventory of Pakistani
learners does not have dental fricatives. We shall reflect this using a constraint *[ ]. This is
an extension of the constraint *C (Tesar & Smolensky, 1996). *C resists the existence of a
specific phoneme in a particular language e.g. *[p] in Arabic, etc. The substitution of dental
fricatives with labio-dental fricatives is ascribed to a constraint ACOUSTIC
CORRESPONDENCE (Wester, Gilbert & Lowie, 2007) which demands that the output
should be acoustically the same as the input. The faithfulness constraint
CORRESPONDENCE is active in the L2 grammar of the learners in different forms.
CORRESPONDENCE demands that output is identical to the input. Explaining the
CORRESPONDENCE constraint Wester, Gilbert & Lowie (2007: 485) comment:
[f] is the most similar segment to [] from an acoustic viewpoint: [f] and [] are
acoustically (and thus perceivably) surprisingly similar and are hard to differentiate
in a non-semantic context. This implies that when [] is ruled out by a
MARKEDNESS constraint, maximal acoustic CORRESPONDENCE will demand
[f] to be used as the realisation of [].
The substitution of dental fricatives with other consonants in the current context is a
result of interaction between the markedness constraint *C and the faithfulness constraint
CORRESPONDENCE. The following rankings reflect the perception of most of the
participants.
(5) *[ ], CORRESPONDENCE [continuant], ACOUSTIC CORRESPONDENCE >>
CORRESPONDENCE [COR], CORRESPONDENCE [strident] 10
/ /
*/*
[ ]
[s z]
*!
CORRESPONDENCE
[continuant]
ACOUSTIC
CORRESPONDENCE
CORRESPONDENCE
*!
[COR], [strident] 11
That is why in some phonological theories, like Element Theory (Backley 2011, Harris 1994), frication and
aspiration are represented with the same element i.e. H.
10
For those few participants who perceived English dental fricatives as coronal fricatives the
CORRESPONDENCE [COR] will be ranked higher than ACOUSTIC CORRESPONDENCE and the overall
ranking will be *[ ], CORRESPONDENCE [continuant], CORRESPONDENCE [COR] >> ACOUSTIC
CORRESPONDENCE, CORRESPONDENCE [strident]
11
Due to space problem the constraint CORRESPONDENCE [strident] is not written separately. This is a low
ranked constraint in the grammar of the learners which does not play any significant role in the current context.
69
[f v]
[t d ]
[x ]
*!
*!
*!
The most faithful candidates [ ] lose for violating *[ ]. The learners also preserve
the feature [+continuant] losing the place feature [COR] in that they perceive dental fricatives
as labio-dental continuant. This shows that the constraint CORRESPONDENCE [COR] is
low ranked in the perception ranking. The candidates [t d ] violate CORRESPONDENCE
[continuant] and the velar [x ] lose because they are not acoustically similar to the input so
they violate the constraint ACOUSTIC CORRESPONDENCE. Therefore, [f v] emerge as the
most optimal candidates.
As the results of the study show, the learners have different rankings for the
perception and production of English dental fricatives [ ]. They perceive [ ] as fricatives
but produce them as stops, which means that CORRESPONDENCE [COR] is inviolable in
their production grammar. The constraint rankings for production of voiceless and voiced
phonemes are also different from each other since the learners add aspiration to the voiceless
stop but no such thing occurs with voiced dental fricative []. As hypothesized earlier, the
learners perceive frication in [] in the input and compensate for it by adding aspiration to the
output. If this hypothesis holds, it means that the constraint CORRESPONDECE
FRICATION sits higher in the grammar of the learners. It strengthens the SLM point of view
that L2 learners perceive sounds on account of their phonetic properties. The following
ranking reflects the substitution of voiceless fricative with voiceless aspirated stop.
(6) *[], CORRESPONDENCE [COR],
CORRESPONDENCE FRICATION >>
CORRESPONDENCE [strident]
//
*[]
[]
[s]
[f]
[t ]
h
[t ]
*!
CORRESPONDENCE
CORRESPONDENCE
[distributed],
[continuant],
CORRESPONDENCE
CORRESPONDENCE
CORRESPONDENCE
[COR, distributed]
FRICATION
[continuant, strident]
*!
*!
*
*!
*
The above tableau shows that the most faithful candidate is defeated for violating a
highly ranked constraint *[] whereas [s] and [f] lose because the former violates
CORRESPONDENCE [distributed] and the latter violates CORRESPONDENCE [COR].
Candidate 4 i.e. [t ] loses because it does not obey the constraint CORRESPONDENCE
FRICATION which is also highly ranked. The input [] has frication whereas [t ] lacks it.
Therefore, the laurels go to [t h] which emerges as the most optimal candidate. However, we
have to differentiate between frication and the feature [+continuant]. The former is a phonetic
phenomenon and the latter is a phonological feature. The learners lose a phonological feature
but compensate for it by producing an output which is phonetically similar to the input. The
following ranking reflects the substitution of [] with [d ].
Syed
70
[]
[z]
[v]
[d ]
*!
CORRESPONDENCE
CORRESPONDENCE
[COR, distributed]
[continuant, strident]
*!
*!
*
*
The place feature CORONAL and its dependent [distributed] sit higher than
CORRESPONDENCE [continuant] so the dental stop [d ] emerges as the most optimal
candidate because [z] is defeated on account of a fatal violation of CORRESPONDENCE
[distributed] and [v] loses for violating CORRESPONDENCE [COR]. The successful
candidate only violates CORRESPONDENCE [continuant] which is lower ranked in the
production grammar of the learners.
6. Conclusion
This study reported on an experiment with a group of advanced Pakistani learners of
English. Perception and production test results show that Pakistani learners perceive English
dental fricatives as labial fricatives on account of acoustic cues but produce them as stops
under the influence of Pakistani English. This shows that the learners have two parallel
constraint rankings for perception and production, one of which is based on the phonetic
nature of the target sounds and the other is based on phonological features. The findings of
the study pose a challenge to the speech learning model, which claims a correspondence
between the perception and production of L2 learners. It points out a gap in the SLM, that it
is only a phonetics-based model, whereas the current study demonstrates that both phonetic
and phonological factors interact in the grammar of L2 learners. This suggests a revision in
the model by incorporating phonological factors besides phonetic factors into it.
References
Archibald, J. (1998). Second language phonology . Amsterdam/Philadelphia: John Benjamins
Publishing Company.
Archibald, J. (2005). Second language phonology as redeployment of phonological features.
Canadian Journal of Linguistics, 50 (1-4), 1000-1030.
Backley, P. (2011). An introduction to element theory. Edinburgh: Edinburgh University
Press.
Bassetti, B. (2009). Effects of adding inter-word spacing on Chinese reading: a comparison of
Chinese native readers and English readers of Chinese as a second language. Applied
Psycholinguistics, 30(4), 757-775.
Baumgardner, R. J. (1990). The indigenization of English in Pakistan. English Today, 6(1),
59-65.
Baumgardner, R. J. (1993). The English Language in Pakistan: Oxford University Press.
Bell, A. & Gibson, A. (2008). Stopping and Fronting in New Zealand Pasifika English.
University of Pennsylvania Working Papers in Linguistics, 14(2), 44-53.
Best, C. (1994). The emergence of native-language phonological influences in infants: A
perceptual assimilation model. In Goodman, C. & Nasbaum, H. (eds.), The
71
72
Syed
Rahman, T. (1991). Pakistani English: some phonological and phonetic features. World
Englishes, 10(1), 83-95.
Shackle, C. (1976). The Siraiki language of central Pakistan: A reference grammar. London:
University of London School of Oriental and African Studies.
Shackle, C. (2007). Urdu. In Cardona, G. & Jain, D. (eds.), The Indo-Aryan Languages, 315386. London: Routlege.
Simon, E. (2009). Acquiring a new second language contrast: an analysis of the English
laryngeal system of first language Dutch speakers. Second Language Research, 25(3),
377-408.
Syed, N. (2011). Perception and production of consonants of English by Pashto speakers. The
Journal of Humanities and Social Sciences, University of Peshawar, XIX(1), 119-146.
Syed, N. (2013). Voice onset time for plosives in Saraiki: Implications for the acquisition of
English aspiration contrast. EFL Annual Research Journal 2012 Shah Abdul Latif
University Khairpur 14, 1-15.
Tesar, B. & Smolensky, P. (1996). Learnability in Optimality Theory. Retrieved from
http://www.blutner.de/ot/dat/tes-smol_OTlearn-short.pdf
Weinberger, S. (1997). Minimal segments in second language phonology. In James, A. &
Leather, J. (eds.), Second Language Speech: Structure and Process, 263-311. Berlin:
Mouton de Gruyter.
Wester, F., Gilbers, D. & Lowie, W. (2007). Substitution of dental fricatives in English by
Dutch L2 speakers. Language Sciences, 29(2-3), 477-491.
Yildiz, Y. (2006). Age effects in the second language acquisition of English onset clusters by
Turkish learners: An optimality theory account. Unpublished PhD Dissertation,
University of Essex.
Nasir Abbas
Lasbela University
Uthal, Balochistan
Pakistan
nasirabbassla@gmail.com
73
PROPOSAL
FOR
PASSIVE
DOMINIC THOMPSON
(University of Glasgow)
CHRISTOPH SCHEEPERS
(University of Glasgow)
Abstract
To describe a transitive event, the English language allows a choice of two Voices. The
canonical form is the active-voice, and the alternative is passive-voice, which offers its own
semantic and syntactic functions. The passive-voice can also be divided into two further
variants: be-passives and get-passives. Though theories are numerous, literature falls short in
describing the functions and uses of these two forms.
Here we present a rethink of passive syntactic representation, simplifying its
description under a single structural unit. The proposed pvP theory is a non-derivational
account. It allows a simple explanation of the syntactic and semantic variation between the
two passive-types, while accounting for the features that are shared by all passive forms.
1. Passive Syntax
Recent descriptions of the passive construction have been shaped significantly by
Chomskys (1981) analysis in which he views passives as being syntactic derivations of the
canonical active-voice. For Chomsky, the passive is generated through a series of
transformations that are triggered by Case and thematic-role (theta-role) requirements. In an
active-voice construction (1), a transitive verb such as attack requires two arguments: a
subject and an object complement. The subject is theta-marked as the Agent (the doer) of
the action, and receives Nominative Case; while the object is theta-marked as Patient (the
undergoer of the action), and receives Accusative Case.
1)
2)
This work was funded by ESRC grant RES-062-23-2009, awarded to Christoph Scheepers and Fernanda
Ferreira.
75
Here, TP stands for Tense Phrase, equivalent to the Inflection Phrase of earlier
generative models; DP is a Determiner Phrase, usually consisting of a determiner plus a
Noun Phrase (NP); t is a trace left by a constituent that has moved; VP is a Verb Phrase; vP
is a light Verb Phrase, which is semantically impoverished and therefore is dependent on
combination with some additional phrase to produce a meaning.
The Agent of the action, who appears as the sentential subject in the active, is demoted
to appear only in an optional adjunct. This is referred to as the by-phrase, or sometimes the
agentive by-phrase to distinguish it from other by-initial adjuncts such as locatives.
This initial conception of the passive-voice is explored further by Jaeggli (1986). In
support of the idea that the subject position in the passive does not receive a theta-role,
Jaeggli notes that it is possible for expletives to appear in subject position in passive
constructions, such as in (3).
3)
Expletives are ungrammatical if they are used in positions that are necessarily
assigned a theta-role, therefore the grammaticality of (3) is taken as evidence that the subject
position does not receive one.
Jaeggli conceptualises the theta-role and Case suppressions mentioned by Chomsky as
absorption, further specifying that it is the passive bound morpheme -en that absorbs these
features; specifically defining absorption as assignment to a bound morpheme. In practice,
this passive bound morpheme, or participle suffix, is usually realised as -ed; in the literature,
for ease of distinction, it is consistently referred to as -en. Jaeggli notes that the nonassignment of Case to the object does not hold cross-linguistically, citing languages such as
Spanish and Italian as counter-examples, and concluding that Case in passive constructions
requires language-specific options.
With regard to the optional by-phrase in the passive, Jaeggli highlights that the nounphrase within is interpreted as having whatever thematic-role the passivized verb standardly
assigns to its external argument; that is, the determiner-phrase (DP) in the by-phrase is not
inevitably an Agent, but may also be a Source, Goal, Experiencer, etc. The theta-role reaches
the DP within the by-phrase via transmission: once the bound morpheme -en absorbs the
theta-role from the verb, it transmits it to the prepositional-phrase, which in turn assigns it to
the DP.
76
Collins (2005) presents a departure from this generally accepted approach initiated by
Chomsky. Rather than relying on transformations, Collins proposes an analysis that sees both
active-voice and passive-voice as having the same underlying structure with regards to
external argument structure. He asserts that the passive participle suffix (the bound morpheme
-en in Jaeggli) does not differ from the past participle suffix, citing their identical morphology
in English. Furthermore, neither suffix results in suppression or absorption of theta-roles or
Case.
To achieve a situation whereby the external argument merges in the same location in
both active and passive constructions, Collins suggests that the passive involves smuggling.
This entails the movement of a constituent x, held within a larger constituent y, across a
potential barrier to x. He introduces a Part-Phrase (PartP), containing the participle and direct
object; all of which, in the passive, raises to a second new phrase, Voice-Phrase (VoiceP),
which has by as its head.
In terms of the passive by-phrase, Collins objects to the situation in Jaegglis approach
whereby theta-role assignment to the external argument (via transmission) is entirely
different in the passive versus the active. In place of theta-transmission, Collins treats by as a
dummy-preposition. The preposition exists only to allow the inclusion of an Agentcontaining adjunct; the preposition by, when used in the passive, does not possess
interpretable features. Collins contrasts this with other uses of by, such as locative
constructions, in which the preposition does possess interpretable semantic features.
Ahn & Sailor (2010) build on Collins with a slight revision of the VoiceP. They note
that there are two previous independent instantiations of VoiceP: the first is from Kratzor
(1996), who has VoiceP as both the merging point for external arguments, and the assigner of
a theta-role to that external argument; the second is that of Collins described above, where
VoiceP is a phrase that assists in the formation of the passive, and is absent from the active.
Ahn and Sailor combine these concepts, having VoiceP present in all voices; active, passive,
and middle. They hold that passives are not derived from an underlying active-voice sentence,
concluding that there is no default voice. However, the generation of the passive still calls
for an additional phrase, FP, that is not present in the active. While they note several clear
problems with their analysis (related to the motivation for various constituent movements and
feature-checks within the VoiceP), the authors maintain their proposals central quality of
having a single phrase to motivate all grammatical voices.
Something notably absent from all of these approaches is any distinction between the
two variant passive forms; be and get. It has even been explicitly stated by some (e.g. Quirk et
al. 1972, Stein 1979, Chomsky 1981) that these two forms are structurally identical and
interchangeable. It will, however, become clear that passive get is distinct from the standard
be both syntactically and semantically, through precisely how is a matter of debate.
2. Get-Passive Syntax
Some of the first authors to include the get-passive in their discussions (e.g. Quirk et
al. 1972, Stein 1979) analysed the two passive-types as structurally identical, with a simple
choice between alternative auxiliaries; be and get. The first syntactic analysis of the passive
seeking to differentiate these two forms comes from Haegeman (1985). She begins by
refuting the earlier belief that be and get are syntactically equivalent in the passive, both
functioning as auxiliaries. She achieves this simply by demonstrating that get fails all standard
linguistic tests for determining auxiliary status. These tests involve the observable syntactic
behaviour of auxiliary and lexical verbs in specific structural contexts. Auxiliary verbs are
able to combine with the negative contraction -nt (4); can undergo subject-auxiliaryinversion (5); and they may be stranded in instances of VP-deletion (6).
4)
5)
6)
She was hired and her friend was too / *She got hired and her friend got too
77
In all of these contexts, any lexical verb will be ungrammatical; this clearly includes
passive get. To make the formation of these structures possible with any lexical verb, support
must be provided by the dummy-operator do (7); passive get clearly falls into this category.
7)
She didnt get hired / Did she get hired? / She got hired and her friend did too
Having shown the non-equivalence of the two passive-types, Haegeman proposes that
get is an ergative lexical verb that takes a passive small-clause complement, resulting in the
following surface structure:
8)
In being ergative, get is not equipped to assign Case, so when John moves in an
attempt to satisfy its Case requirement, it is unable to do so after the first move, therefore
moving again to Specifier of TP ([Spec,TP]; subject position), where it can receive Case from
T (Tense-Phrase). Ergative get also has no Agent -role to assign, ensuring the DP does not
receive two -roles. The resulting nature of the get-passive is partly passive through the
presence of passive morphology on the participle (Haegeman 1985: 70) and is also ergative,
due to get being an ergative verb. In assuming that get is not an auxiliary verb, it follows that
get does not raise to T; thus with get and be occupying different nodes in the tree structure,
their syntactic differences are easily explained.
One notable development of Haegemans approach comes from Fleisher (2008), who
introduces an additional phrase, vPpass, into the passive structure. The primary motivation for
this revision comes from the following comparison of grammaticality:
9)
10)
Sportiche (1988) gives a stranding analysis of quantifier float; that is, stranded
quantifiers are grammatical when merged in the position of an intermediate trace (a position
through which a DP has moved). This is the case in (9). The fact that (10) is judged to be
ungrammatical indicates the absence of a trace position in which to merge the quantifier all.
Furthermore, this missing intermediate position is conspicuously present in the get-causative
construction, as seen from the presence of the DP them in (11).
11)
78
functional head vpass, no position is left available between get and arrested for the object to
move through, thereby accounting for the ungrammaticality of (10). The underlying structure
as described by Fleisher is reproduced in Figure 2.
Figure 2: They got arrested
(Fleisher 2008)
The contraction of the embedded VP [arrested, ti] (i.e. not including a V level) appears to
account for the discrepancy between the examples in (9) and (10), since it eliminates the DP
landing site between get and the main verb. The proposal of a vPpass however, does not
significantly add to the analysis. Uniting the VP headed by get and the vP headed by be as
two instances of the vPpass, is primarily a notational alteration. It provides no better
explanation for the differing syntactic behaviour of get and be in passive constructions, nor
does it lend any assistance in explaining the semantic dissonance across the two forms.
Furthermore, despite assuming an analysis with get and be both merging as head of the same
phrase (vPpass), it remains the case that get does not raise to T. Fleisher notes this simply as an
idiosyncrasy.
The causative, as exemplified in (11) above, also features the passive voice. Fleisher
represents this as a null vPpass. He states that, in the causative, get is a full lexical verb heading
a VP, and taking a vPpass as a complement. The syntactic tree for this structure is reproduced
in Figure 3.
Figure 3: He got them arrested
(Fleisher 2008)
79
In Figure 3, the structure of the vPpass remains as it is in the get-passive, except that the
head of the phrase is null. The direct object of arrested moves to [Spec,vPpass], but differs
from the passive as it moves no higher. If the DP them is to remain in [Spec,vPpass] it must
receive Case in that position: Fleisher suggests an instance of Exceptional Case Marking,
either from get, or possibly from further up the structure via the vagent node.
The theories discussed so far are classed as raising approaches. They involve deriving
the passive-voice from the active via the raising of the actives thematic Patient into subject
position. Parallel to these there have been proponents of a possible control structure
underlying the get-passive. Lasnik & Fiengo (1974) seem to be the originators of this train of
thought. While they do not provide an explicit exploration of the get-passive, they do
highlight features that indicate its controlling nature. It is not until Huang (1999) that the
control approach is given deeper consideration. In Huangs estimation, the get-passive is a
subject-control verb; that is, a verb that forces the covert (unpronounced) subject of an
embedded clause to refer to the subject of the matrix clause. This unpronounced element is
referred to as PRO. Huang presents the get-passive and be-passive as very similar structurally.
The primary difference is that in the be-passive, the Patient of the action is raised from direct
object position to subject position of the matrix clause; while in the get-passive, the direct
object is the null anaphoric pronoun, PRO. This means that the subject in a get-passive has
not raised from direct object position; instead PRO necessarily refers to the subject. This
effectively places the subject simultaneously in direct object position. To exemplify, in (12)
the direct object (the Patient) is the unpronounced PRO; this has raised to become subject of
the embedded clause. John is the subject of the matrix clause, which is necessarily coreferenced with PRO (this implies an interpretation along the lines of John got himself killed).
Compare this with the be-passive, as analysed by Haegeman (1985) above, and maintained by
Huang (13).
12)
13)
Huang states that this structural difference accounts for differences between the getpassive and be-passive as given in (14) and (15), and that such differences are equal to those
recognized between general raising and control constructions, as in (16) and (17), in the case
of subject-oriented adverbs.
14)
15)
16)
17)
The sentence in (14) does sound odd. However, we do not believe that this is due to a
structural difference; it is likely to be due to the semantic interaction between cheated and
intentionally: the implicit Agent would be unable to accidentally cheat John. In terms of
syntax, it is comparable to the pair given in (18) and (19).
18)
80
19)
The sentence in (18) does not sound ill-formed as shooting can occur either
intentionally or accidentally, and therefore the adverbs presence is justified, and not
superfluous as in (14). The reason (15) is acceptable may be that a causative-reflexive
reading is possible, wherein John chose to get cheated by some unnamed party. Although for
both sentences in (18) and (19) a reading is available in which some unnamed party purposely
shot John, it is only in (19) that this causative-reflexive reading is additionally possible.
Contrary to Fleishers approach, in which distinct representations are given to the
passive and causative instantiations of get, Huang proposes a far more intimate relation
between the structures. His analysis of the causative is that it fully contains the complete getpassive structure as the complement of a causative VP shell (Huang 1999).
20)
Although this analysis seems economical specifically for the get-passive and getcausative constructions, it does not appear to generalize to other instances of the causative.
Consider sentences of the type given in (22) and (23).
22)
23)
These sentences exhibit the same structure as the get-causative construction in (21).
They differ in that the matrix verbs want and see cannot form the simple passive.
Consequently there seems to be no motivation for assuming that these verbs initially merge
lower in the structure before raising to their final positions. The embedded VP maintains its
passiveness irrespective of where the matrix verb is initially merged. This being the case, we
propose it is more likely that want and see each merge directly as head of the matrix VP. In
support of this, notice that both be and get are able to merge in the position that want or see
would have raised from.
24)
Mary wanted John to be/to get blamed for the mistake (by Bill)
25)
Mary saw John be/get blamed for the mistake (by Bill)
It is clear that want or see could not have raised from the embedded VP, as that VP is
already headed by a separate verb.
Further generalizing the idea that verbs such as want or see do not raise out of an
embedded passive VP would allow causative get to remain independent of the passive,
merging directly as the matrix verb. Such a generalization is clearly supported by the fact that
causative get can acceptably combine with passive get or passive be:
26)
81
The form in (26) that contains both instances of get may seem slightly degraded or
dispreferred, but this is likely to come from the repetition of the get lexeme. Regardless, this
suggests that causative and passive get do not have as close a relationship as Huang claims.
Butler & Tsoulas (2006) argue in favour of Huangs proposal. The authors cite
Alexiadou (2005) as noting that the reflexive reading of the passive is only possible with get.
They suggest that this can be accounted for if get realizes a control structure of the sort
developed in Huang (1999). They note further that it is frequently possible to paraphrase
control structures with a reflexive pronoun in place of PRO (Butler & Tsoulas 2006). While
the get-passive does have the ability to project a reflexive interpretation, similar to classic
subject-control verbs such as want, they fail to note that get is also able to give a nonreflexive reading equal to that of the be-passive.
Recently, Reed (2011) has argued that the structure of get-passives is in fact
ambiguous, having three variant forms. These include a verbal passive (this is the raising
passive discussed above, as found in Quirk et al. 1972, Stein 1979, Haegeman 1985, Fleisher
2008); a control structure; and an adjectival passive (from Fox & Grodzinsky 1998).
Discussions of the get-passive construction have taken several paths. They can assume
a raising or control structure, an adjectival passive, or some combination of these. However,
gets counterpart, the be-passive, is universally treated the same in these analyses. All authors
treat the syntactic structure of the be-passive in much the same way as Chomsky (1981),
providing no notable revisions. The assumption is always that get, as an intruder into the
domain of the passive, is the only variant that requires attention or reanalysis.
3. Passive Semantics and Function
Linguists have typically treated the passive as a derivative of the canonical activevoice. However passive constructions stand in parallel, serving certain purposes less available
or entirely unavailable to the active-voice. While some (e.g. Weiner & Labov 1998) have
suggested that both be-passives and get-passives are semantically equivalent in general, the
majority consider them distinct.
3.1 Passive Function
As noted by Keenan & Dryer (2006), passives allow focus to be concentrated on a
specific element, in a way very similar to other topicalizing constructions such as clefting.
What this means for the passive is that the object in a described event is elevated above other
elements in terms of its relative importance; that is, its syntactic prominence is utilised in
communicating some form of significance. Further to this, passives allow the backgrounding
of another element: the subject of the active, typically the Agent of the described action. This
reduction of focus or importance can be achieved by relegating it to the status of an oblique
NP (Keenan & Dryer 2006), that is, by consigning the agent of the action to a by-phrase, as
in Mary was hired by the manager. Focus can be further depleted by completely removing the
by-phrase, and therefore any explicit mention of the agent, as in Mary was hired.
Several other notable functional aspects of the passive are highlighted by Keenan and
Dryer. First, they can function alongside other foregrounding constructions, for example,
allowing clefting of a passivized sentence. This implies a semantic non-equivalence of the
passive and functionally similar constructions. Second, and related to this, passives are
generally better integrated into the grammar than topicalizations and dislocations, which tend
to be limited to main clauses. Finally, they point to an important absence of distinction
82
83
affected entity. Here, get-passives and be-passives are on more equal terms: both are weighted
towards having an affected subject. This suggests that the difference between these two
passive forms comes from their relative transitivity, while the degree to which the subject is
affected by the event (how Patient-like the subject is) is more or less constant.
The term affected is not used consistently in the literature, but generally, if an entity
is affected, it is conceptualised as undergoing some experience resulting from the action
described by the main verb. Contrary to Arrese (1999) discussed above, many authors, such
as Sasaki (1999), do not believe the two passive forms communicate an equal degree of
affectedness; instead they believe that the get-passive suggests greater subject affectedness.
For instance, Carter & McCarthy (1999) suggest that the get-passive has a tendency to focus
on the event itself, and how the event impacts the patient.
Cameron (1990) suggests that it is not sufficient for the Patient to simply be affected;
rather they must be materially affected. Corpus data from Downing (1996), however, gives
evidence against this, providing examples such as (27).
27)
Orfitelli (2011) goes as far as to claim that the affectedness requirement is so strong
that predicates that do not affect their internal argument are typically illicit with the getpassive, although they are allowed in the be-passive. The author exemplifies this through the
sentences in (28) and (29).
28)
29)
84
While Carter & McCarthy (1999) also suggest the get-passive communicates adversity,
they specify that the adversity is interpreted by the speaker, not necessarily the actual Patient
of the action.
For other authors, adversity is not so narrowly defined; it is understood to equally
describe both positive and negative outcomes (Chappell 1980, Sussex 1982, Siewierska 1984,
Givon 1993, Gronemeyer 1999, Sasaki 1999). The adversity may arise due to some perceived
struggle; as McIntyre (2005) describes it, the get-passive suggests that the result is hard to
attain. Sussex (1982) suggests that, although get-passives can communicate both negative
and positive events, negative get-passives have more semantic flexibility.
3.5. Patient Responsibility
Another commonly identified semantic feature of the get-passive is its ability to
communicate responsibility on the part of the subject; the Patient of the action. Numerous
authors note this implication of initiative, control, or responsibility (among many others,
Hatcher 1949, Lakoff 1971, Barber 1975, Givon & Yang 1994, Downing 1996, Sasaki 1999).
Relatedly, Vanrespaille (1991) suggests that resultativeness is a primary feature of the getpassive, and that the subject is at least partly responsible. Arrese (1997) also notes at least
partial responsibility of the Patient. Sussex (1982) claims that get-passives can imply a wide
variety of meanings, including varying degrees of purposefulness, blame, and responsibility.
The semantics of control are conceptualised by Lasnik & Fiengo (1974) in a more
complementary manner between the get-passive and be-passive. The authors claim to show
that when a passive sentence is formed with get, Patient control is assumed; when the same
sentence is formed with be, Agent control is assumed. For example, in (30) it is assumed to be
Johns intention to cause the event; while in (31), the event is Marys intention.
30)
31)
This is compatible with the suggestion of Carter & McCarthy (1999) that the getpassive holds interpersonal meaning, which they believe motivates its greater likelihood in
dialogues.
85
Complementary to the humanness of the Patient, Hatcher (1949) suggests the byphrase Agent is likely to be inanimate or non-identifiable, and when a human Agent is
present, they have a low degree of individualization. For Sussex (1982), however, Agent
humanness is another factor that increases the semantic flexibility of get-passives.
3.7. Reflexivity
For Sussex (1982), the potential reflexivity of the get-passive also increases its
semantic flexibility. It is noted by Givon & Yang (1994) that a reflexive or reciprocal sense is
only possible when the passive is formed with get. Alexiadou (2005) simply notes that the
get-passive is compatible with reflexive action.
4. The pvP Theory
In the preceding sections we have discussed existing approaches to passivization in
English. This has included considerations of the passive versus the active, and of the getpassive versus the be-passive.
All of the authors who have examined the syntax of the passive agree on the basics of
the be-passive structure, while each offering a different analysis of the get-passive. These
analyses of passive get not only differ from author to author, but are consistently
distinguished structurally from the be-passive. Despite all forms of the passive having several
semantic and syntactic features in common, no theory thus far is able to give a simple or
unified account of this collection of constructions. Each syntactic approach necessitates a
number of distinct underlying representations to describe the various passive-types, along
with assorted caveats to ensure there are no violations of established principles or conditions.
The semantics of the passive forms also raise debate. There is no suggestion in the
literature that clearly describes a central meaning of each, and no approach that appears to
account for both the differences and similarities between the get-passive and be-passive, nor
for the differences in their degree of variability.
In order to harmonize this situation, a new approach is clearly required. All the passive
forms share several attributes, from the canonical be-passive to the partial passive in causative
constructions. For example, in all forms the main verb has the same inflection; the clausal
subject is interpreted as the Patient of the action; and the inclusion of an agentive by-phrase is
optional. These can be seen in example (33).
33)
From the above it can be seen that the actual usage of the various passive forms is
more consistent than one would guess based on any given theory discussed in previous
sections. There are clearly differences between the forms too. In this paper we are considering
two of the variants; be-passives and get-passives. As we have detailed above, these
differences include syntactic ones (e.g. question formation) and semantic ones (e.g. Patient
responsibility or control). Any new approach to the passive must seek to account for these
commonalities as well as the variations between each form. This is what we aim to do in the
approach discussed here.
86
In (34), PRO is co-indexed with (necessarily refers to) the preceding DP, John. In the
case of the pvP, the PRO is also co-indexed with the preceding DP: in simple passives this is
the matrix subject, while in causatives it is the embedded DP. We will return to this shortly.
The pvP remains insular all the way up to surface-structure. In order to satisfy the
Extended Projection Principle (that all clauses must have a subject), the direct object PRO
raises to specifier of pvP (the subject position of the pvP), and moves no higher. This means
that, unlike in existing theoretical approaches (including Haegeman 1985, Huang 1999, Butler
& Tsoulas 2006, Fleisher 2008, Reed 2011), the matrix subject is always a discrete DP (i.e. it
does not raise from direct object position). The pvP exhibits structural consistency through all
passive forms, with the passivized verb, V0, being the only variable. The passivized verb is
the verb that undergoes passivization, i.e. is morphologically marked as Past Participle, and
whose thematic object is displaced. The bare syntactic structure of the pvP is detailed in
Figure 4.
Figure 4: Bare structure of the pvP
This unit, the pvP, is the core of the syntactic model we are proposing, with
implications for both the syntax and semantics of the various passive forms. Below we will
discuss the pvP as it appears within each of these passive constructions. We aim to
demonstrate both the simplicity with which each is built around this core phrase, and the
notable consistency of the structures.
87
This absence of an explicit passive auxiliary (e.g. get or be) has motivated the various
and divergent approaches (see especially Fleisher 2008). However, taking the pvP as a
constituent generates a simple state of affairs.
In a causative construction such as (35), the verb want selects three arguments: a
subject, the DP John; and two complements, the DP Mary, and the pvP PROi killed ti. The
anaphoric PRO necessarily refers to the preceding DP, Mary. The structure for (35), as
interpreted under a pvP analysis is given in Figure 5. The dashed box denotes the passive
portion of the structure; the pvP itself.
Figure 5: Representation of John wanted Mary killed under a pvP analysis
The structure in Figure 5, here formed with want as the matrix verb, is possible with
any of a small set of verbs occupying the matrix verb position, including see, have, want,
expect, and get. Taking these verbs as forming a double-complement construction in the
causative (selecting DP and pvP complements), allows a single account for the whole set of
verbs (contrary to Huang 1999). All of these verbs that can appear as the matrix verb are not
themselves passive; they all function the same as they do outside of causative or passive
sentences. The only difference is that in this instance they have been combined with a pvP
complement, rather than some other complement type. As such, we will refer to this type of
passive as a Double-Complement Passive. Presently, we simply note that there are some
88
verbs that can select a pvP complement, and other that cannot. While there may be some
underlying constraints, it would be difficult to speculate on these in the present space
limitations. However, leaving these constraints underspecified does not impact on the
theoretical framework of the pvP as such, (similarly, leaving constraints on sub-categorisation
underspecified does not speak against lexicalised theories of grammar.)
In Huangs approach (Huang 1999), the causative-passive is directly related to the getpassive, with get raising into a higher VP. This seems plausible in the case of get, however, as
noted above, the causative-passive is possible with several verbs, of which it is only get that
can form the simple passive. It is far more parsimonious to describe all of these verbs as
behaving in a common manner: none of them are passive in themselves; the passiveness of
the sentence is generated through the inclusion of a pvP argument.
4.1.3. Be and Get: Single-Complement Passives
While get (among other verbs) selects three arguments in the causative, in the socalled get-passive it selects two: a DP subject and a pvP complement, as shown in Figure 6.
Again, the dashed box contains the pvP; the passive part of the structure.
Figure 6: Representation of John got killed under a pvP analysis
As with the matrix verbs in the causative, get in Figure 6 is not passive, nor is it a
passive auxiliary of any kind; it is the same lexeme that occurs in other transitive
constructions. As above, the passiveness of the sentence comes from get selecting a pvP
complement. This differs from the majority of existing approaches to the get-passive (e.g.
Haegeman 1985, Fleisher 2008, Reed, 2011), in which get has some specific form that
appears in the passive.
Finally there is the case of the be-passive. This is the most widely agreed upon in
structural terms, and is regarded in some sense as the true passive. In practice, this situation
necessitates a multitude of additional structures to fully represent the passive, since the
nature of this accepted version of be-passive structure is not capable of accounting for the
syntax or semantics of other passive forms; the get-passive and causative constructions
require their own unique structures.
89
With the inclusion of the pvP as a constituent, the situation can once again be greatly
simplified: the be-passive can be taken as an instantiation of copula be with a pvP
complement, as shown in Figure 7. As before, the dashed box shows the structures passive
section.
Figure 7: Representation of John was killed under a pvP analysis
As with get, the lexeme be is not passive, nor is it a passive auxiliary; it simply
belongs to a group of verbs that are able to take a pvP complement. To clarify, be is not a
passive auxiliary, it is simply an auxiliary. Once again, it is the pvP that provides the
passiveness of the sentence. For these types of passive we will use the term SingleComplement Passive.
Since neither be nor get is a passive auxiliary, and only the former is an auxiliary of
any kind, referring to them in such a way is misleading. With no alternative in the existing
literature, we will use the term passive verbal-complement, or pvC. This term represents the
fact that these lexemes are needed in the construction of a simple passive sentence (i.e. are
complements), while avoiding any reference to passive auxiliary status.
Importantly, the structures in Figures (6) and (7) allow a minimal account of the
syntactic differences noted from Haegeman and Fleisher. With both get and be existing
entirely outside of a self-contained passive constituent (the pvP), the two verbs (pvCs) can
continue to act as they do in all other non-passive sentences. That is to say, get is a full lexical
verb heading a VP. On the other hand, be is an auxiliary verb, and as such can act as one in
question formation, VP-ellipsis, etc.
This approach also correctly predicts the variation in grammaticality noted above
regarding the insertion of a quantifier, repeated in (36). The issue here is that a quantifier,
such as all, can only appear grammatically in the location of a trace; a location from which a
constituent has raised. In this case the constituent that has raised is They.
36)
In the get-passive (that is, a single-complement passive as formed with the pvC get;
see Figure 6), the only trace left by the subject lies above get in the structure. This means that
90
a floated quantifier can only be placed above get (in the location of the trace), with no
locations available below. In the single-complement passive with be, the subject leaves the
same trace as in the single-complement get passive. A floating quantifier again only has one
available location, but in this case the auxiliary verb be raises higher up the structure to T0 (as
all auxiliaries do), thereby allowing the quantifier to follow be (see Figure 7).
There are some types of sentences, as in (37), that may appear to be doublecomplement passives; however these are instances of the single-complement passive.
37)
The infinitive form of be or get actually lies within an embedded clause; the TP-object
of want. The resulting structure is comparable to sentences of the type in (38):
38)
Treating the pvP as a constituent that is able to be selected as an argument means that
the sentences in (37) and (38) are structurally identical, as represented in Figure 8.
Figure 8: Representation of John wanted Mary to be [A] under a pvP analysis
91
subject position). It may be the case that the phonetically null subject of the pvP raises further
to become specifier of the TP. Alternatively a separate PRO may merge in each position; that
is, one within the pvP, and one within the TP. Either way, the empty categories form a coreferential chain, meaning that these positions all refer to the same entity, namely Mary.
4.2. A new semantic approach
The approaches in the literature to the semantics of the passive, and more specifically,
of the get-passive, have been diverse. As regards the role of the passive versus the active, we
agree with the notion that the passive is one method that interlocutors can employ or note the
salience of the Patient or to assign salience to the Patient of a described action (as noted for
example by Keenan & Dryer 2006). This is true for both passive get and passive be.
The majority of approaches to passive semantics consider what it is that get does, as
opposed to what each of the pvCs (get and be) do; generally noting a function of the getpassive, while leaving its absence from the be-passive as implied. This in itself suggests a
general consensus whereby the get-passive possesses functions supplementary to those
possessed by the be-passive, rather than each having their own distinct or complementary
functions. Furthermore, many authors, usually seeking features that are present in all
instances of the get-passive, highlight just one or two features of the get-passive, and do so
using highly specified examples to support their claims. These examples need to be so
constrained because they only apply in certain circumstances; that is not to say they are
incorrect, only that they are not universally applicable.
Under the pvP analysis we do not claim that there is a prototypical meaning of the getpassive or argue for one or other of the existing proposals. Instead, we suggest how these
various meanings are produced, and why the observed variability is possible with the getpassive.
With the passiveness of any given passive sentence contained wholly within the pvP
unit, any verb selecting the pvP as a complement is free to behave as it does in non-passive
constructions. That is, the two pvCs (get and be) are outside of the passive portion of the
sentence. We have already discussed how this applies to the syntax of the passive, and we will
now argue that it can equally be applied to passive semantics.
As proposed above, the lexeme be in the passive is taken simply as copula be; it is an
auxiliary. It functions analogous to a logical operator, roughly equivalent to an entailment
sign. John is tall describes a situation in which John entails tall. In the same way, be describes
a logical operation in which John entails a state wherein he has been killed: PROi killed ti. As
an operator, be provides no additional information beyond stating that someone came to be in
a certain situation.
On the other hand, get is a lexical verb. It has a greater depth of meaning than be,
since auxiliaries are purely functional, while lexical verbs also possess their own semantics.
Furthermore, the lexical verb get is capable of communicating several independent meanings.
Some authors have attempted to determine the core meaning of the lexeme get. These have
included result (Forchini 2008), and obtain or receive (Johansson & Oksefjell 1996).
With the pvP placing get outside of the passive portion of the syntactic structure, there
is no need to distinguish a specific passive get from the several non-passive uses of get. This
means that the semantics of the lexeme get apply equally regardless of whether it is used in
passive or non-passive constructions.
Though it is not our aim to explain how each and every interpretation of the getpassive can be derived, we will give two brief examples below.
First, under the reading of obtain, as in John got a book from the shelf, or perhaps
achieve, as in John got recognition, the reflexive interpretation of the get-passive can be
92
communicated. That is, John got killed, with a reflexive reading, could roughly paraphrase as
John obtained/achieved his own killing. The specific meaning is modulated by, among other
things, the nature of the event. Nonetheless, get is able to provide the reflexive sense via one
of its several accepted meanings.
Second, the reading of get as receive, as in John got a shock, accounts for the truly
passive interpretation of the get-passive; that is, examples where the subject is a passive
recipient of a described situation. This would include instances such as John got hit by a bus,
where John has no prior knowledge of the events occurrence. Again, the exact interpretation
can be altered, for example, by the semantics of the main verb or the context.
The main point to be made in relation to get-passive semantics is that there are
numerous meanings of get that can be utilised: in the passive, get is the same lexeme as it is in
non-passive constructions, therefore it has the same semantic range to draw upon. The bepassive, however, is limited in this regard, being an auxiliary and carrying no additional
associations. Essentially, the get-passive is more semantically flexible than the be-passive.
This is accounted for in our proposed pvP theory by virtue of the two passive verbal
complements (get and be) remaining outside of the passive structural unit.
Unlike existing approaches, which assume there are individual passive forms of be
and get, we assume that they are the same verbs that function elsewhere in the language.
There are further implications of this. For example, using the option that has a greater
semantic weight behind it allows a speaker to communicate an added degree of involvement,
or perhaps empathy. It allows a speaker to demonstrate or encode that they sense some
importance beyond a simple logical description of events.
5. Summary and Conclusions
In this paper we highlighted the variety of approaches that have been proposed to deal
with both the semantic and syntactic variations between the be-passive and other passive
forms. We have demonstrated that the syntactic approaches not only conflict with one
another, but that there is no approach that satisfactorily accounts for the several types of
passive without recourse to numerous underlying structures and caveats. Likewise, the
semantic accounts of the passive forms seek to propose some additional meaning for the getpassive, though without a conclusive answer.
We have proposed a new analysis that aims to provide a more parsimonious account,
uniting the various passive forms via a single common constituent. To achieve this we have
postulated a phrasal constituent called a passive light-verb Phrase, or pvP, which is selected
by certain verbs as an argument (in much the same way as an object is selected). The VP
within the pvP contains a phonologically null anaphor (PRO) as its object, which raises into
the pvPs subject position. This PRO necessarily refers to the preceding DP. The pvP is
headed by a null verb, . A set of verbs, both mono-transitive and di-transitive, are able to
select a pvP argument; get and be are just two among this set. Neither get nor be is in fact
passive, and neither one is a passive auxiliary, though be remains a standard auxiliary, as it
is in other constructions. This makes dedicated passive auxiliaries superfluous in our account.
We have termed them passive verbal complements, or pvCs. Under the assumption of a pvP,
be and get (the pvCs) exist outside of the passive segment of a passive sentence. This results
in a situation where the motivations underlying the syntactic and semantic differences across
these passive forms can be accounted for with the greatest degree of simplicity: the syntactic
differences between be and get in question formation, their requirement or otherwise for dosupport, and so-on, are explained by virtue of get being a lexical verb and be being an
auxiliary verb; just as they are in any other context. The structure also makes correct
93
94
Carter, R. & McCarthy, M. (1999). The English get-passive in spoken discourse: description
and implications for an interpersonal grammar. English Language and Linguistics 3,
41-58. Cambridge: Cambridge University Press.
Chafe, W. (1982). Integration and involvement in speaking, writing, and oral literature. In D.
Tannen (ed.), Spoken and written language, 1-16. Norwood, NJ: Ablex.
Chappell, H. (1980). Is the get-passive adversative? Research on Language & Social
Interaction 13(3): 411-452.
Cheshire, J. (2005). Syntactic variation and spoken language. Syntax and Variation:
Reconciling the Biological and the Social, 81-120. Amsterdam: Benjamins, 81-106.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris.
Collins, C. (2005). A Smuggling Approach to the Passive in English. Syntax 8(2).81-120.
Downing, A. (1996). The semantics of get-passives. In Hasan, R., Cloran, C. & Butt, D. (eds.),
Functional descriptions: theory in practice, 179-205. Amsterdam: John Benjamins.
Fleisher, N. (2008). Passive get, Causative get, and the Phrasehood of Passive vP.
Proceedings of CLS 41, 59-67.
Forchini, P. (2008). The get-unit in Corpora of Spontaneous and Non-spontaneous Mediated
Language: from Syntactic Versatility to Semantic and Pragmatic Similarity. In Taylor,
C. (ed.), Ecolingua. The Role of E-corpora in Translation and Language
Learning, 185-209. Trieste: EUT Edizioni Universitari.
Fox, D. & Grodzinsky, Y. (1998). Childrens passive: a view from the by-phrase. Linguistic
Inquiry 29, 311-332.
Frazer, A. & Miller, M. (2009). Double Standards in Sentence Structure. Passive Voice in
Narratives Describing Domestic Violence. Journal of Language and Social
Psychology 28(1), 62-71.
Givn, T. & Yang, L. (1994). The Rise of the English GET-passive. In Fox, B. and Hopper, P.
(eds.), Voice: Form and Function, 119-149. Amsterdam: John Benjamins.
Givn, T. (Ed.). (1993). English grammar: A function-based introduction (Vol. 2). John
Benjamins Publishing Company.
Gronemeyer, C. (1999). On deriving complex polysemy: the grammaticalization of get.
English Language and Linguistics 3(01), 1-39.
Haegeman, L. (1985). The get-passive and Burzios Generalization. Lingua 66, 53-77.
Hatcher, A. G. (1949). To get/be invited. Modern Language Notes, 433-446.
Henley, N., Miller, M., & Beazley, J. (1995). Syntax, semantics, and sexual violence agency
and the passive voice. Journal of Language and Social Psychology 14(1-2), 60-84.
Huang, C.T.J. (1999). Chinese Passives in Comparative Perspective. Tsing Hua Journal of
Chinese Studies 29, 423-509.
Jaeggli, O.A. (1986). Passive. Linguistic Inquiry 17(4):587-622.
Johansson, S. & Oksefjell, S. (1996). Towards a unified account of the syntax and semantics
of GET. In Thomas, J. & Short, M. (eds.), Using corpora for language research:
Studies in honour of Geoffrey Leech, 57-75. London/New York: Longman.
Keenan, E. & Dryer, M. (2006). Passives in the Worlds Languages. Shopen, T. (ed.).
Kratzer, A. (1996). Severing the external argument from its verb. In Rooryck, J. & Zaring, L.
(eds.) Phrase Structure and the Lexicon, 109-137. Dordrecht, Kluwer.
Lakoff, R. (1971). Passive Resistance. Papers from CLS 7, 49-61.
Lasnik, H. & Fiengo, R. (1974). Complement Object Deletion. Linguistic Inquiry 5(4), 535571.
McIntyre, A. (2005). The Semantic and Syntactic Decomposition of get: An interaction
between verb meaning and particle placement. Journal of Semantics 22(4), 401-438.
95
96
(Corresponding author)
Dominic Thompson
Institute of Neuroscience and Psychology
University of Glasgow
58 Hillhead Street
Glasgow, G12 8QB, UK
d.thompson.3@research.gla.ac.uk