(Studies in Linguistics and Philosophy. Synthese Language Library Volume 3) Avishai Margalit (Ed.) - Meaning and Use-D. Reidel Publishing Co. & The Magnes Press (1979)

TO THE MEMORY OF
YEHOSHUA BAR-HILLEL
SYNTHESE LANGUAGE LIBRARY
TEXTS AND STUDIES IN
LINGUISTICS AND PHILOSOPHY
Managing Editors:
JAAKKO HINTIKKA, Academy of Finland and Stanford University

STANLEY PETERS, The University of Texas at Austin
Editorial Board:
EMMON BACH, University of Massachusetts at Amherst

JOAN BRESNAN, M,assachusetts Institute of Technology
JOHN LYONS, University of Sussex
JULIUS M. E. MORAVCSIK, Stanford University
PATRICK SUPPES, Stanford University
DANA SCOTT, Oxford University
VOLUME 3
MEANING
AND USE
Papers Presented at the
Second Jerusalem Philosophical Encounter
April 1976
edited by
A VISHAI MARGALIT
The Hebrew University of Jerusalem, Israel
D. REIDEL PUBLISHING COMPANY

DORDRECHT : HOLLAND I BOSTON: U.S.A.
LON DON: ENGLAND
THE MAGNES PRESS, THE HEBREW UNIVERSITY

JERUSALEM
Library of Congress Cataloging in Publication Data
Jerusalem Philosophical Encounter, 2d, 1976.

Meaning and use.
(Synthese language library; v. 3)

Includes bibliographies and index.
1. Languages-Philosophy-Congresses. 2. Meaning (Philosophy)-
Congresses. 3. Pragmatics-Congresses. I. Margalit, Avishai, 1939-
II. Title. III. Series.
PI06.J46 1976 149'.94 78-16884
ISBN 14-020-3263-3
Published by D. Reidel Publishing Company.

P.O. Box 17. Dordrecht, Holland
Sold and distributed in the U.S.A., Canada, and Mexico

by D. Reidel Publishing Company, Inc.
Lincoln Building, 160 Old Derby Street, Hingham,
Mass. 02043, U.S.A.
Sold and distributed in Israel

by The Magnes Press, The Hebrew University, Jerusalem
All Rights Reserved

Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland
and copyright holders as specified on appropriate pages within.
No part of the material protected by this copyright notice may be reproduced or
utilized in any form by or any means, electronic or mechanical,
including photocopying, recording or by any informational storage and
retrieval system, without written permission from the copyright owner
Printed in Israel
TO THE MEMORY OF
YEHOSHUA BAR-HILLEL
TABLE OF CONTENTS
PREFACE IX
ADDRESS XI
W. V. QUINE / Use and Its Place in Meaning

DONALD DAVIDSON / Moods and Performances 9
Comments by W.V. Quine 21
EDDY M. ZEMACH / Awareness of Objects 23
Comments by Igal Kvart 31
ASA KASHER / What is a Theory of Use? 37
JAAKKO HINTIKKA and LAURI CARLSON / Conditionals, Generic
Quantifiers, and Other Applications ot Subgames 57
HELMUT SCHNELLE / Circumstance Sentences 93
Comments by Victor Raskin: Is There Anything Non-Circum-
stantial? 116
MICHAEL DUMMETT / What Does the Appeal to Use Do for the
Theory of Meaning? 123
Comments by Edna Ullmann-Margalit 136
A VIS HAl MARGALIT / Open Texture 141
MARCELO DASCAL / Conversational Relevance 153
Comments by Ruth Manor 175
JOHN R. SEARLE / Intentionality and the Use of Language 181
HILAR Y PUTNAM / Reference and Understanding 199
Comments by Michael Dummett 218
Hilary Putnam: Reply to Dummett's Comment 226
P. F. STRAWSON / May Bes and Might Have Beens 229
SAUL KRIPKE / A Puzzle about Belief 239
Comments by Hilary Putnam 284
INDEX 289
PREFACE
The second Jerusalem Philosophical Encounter was held in Jerusalem on

April 25-28, 1976. The symposium was originally planned to celebrate the
60th birthday of Yehoshua Bar-Hillel, philosopher and friend. But his
sudden death intervened, and turned celebration into commemoration.
The topic of the symposium was Meaning and Use. For Bar-Hillel, the
question 'meaning or use?' was of great importance, one which he took as a
question of priorities. Which approach to natural language is prior: the
formal, semantical approach, which accords a central position to the truth-
functional concept of meaning and to the theory of reference, or rather the
alternative approach which accords the central position to linguistic commu-
nication and prefers dealing with speech acts to dealing with statements? Bar-
Hillel's answer to this question, in his later years, can be summed up by our
title, meaning and use: neither approach deserves priority, each is equally
necessary, and they both complement each other. Those familiar with Bar-
Hillel's uncompromising intellectual honesty would know that this answer
does not reflect a superficial wish for domestic peace, but stems rather from
deep and informed convictions.
The issues of meaning and use dominated Bar-Hillel's intellectual life. At
the same time his day-to-day existence was guided by the idea that the
meaning of life is to be found in being useful, particularly in being useful to
the community of seekers of knowledge. This collection is meant to serve as
an expression, I hope an eloquent one, of the respect and the love with which
his memory is treasured by his friends, colleagues, and students.
lowe a special debt of gratitude to Ms. Eva Shorr, who has spared no
effort in bringing this volume to its present form. Thanks are also due to the
institutions which helped in the organization of the Encounter: the Israel
Academy of Sciences and Humanities; the S.H. Bergman Centre for Philo-
sophical Studies, which launched the Jerusalem Philosophical Encounters;
the Van Leer Jerusalem Foundation; and the Israeli Association for Logic,
Methodology, and Philosophy of Science.
The Hebrew University of Jerusalem A.M.
IX
ADDRESS
Ladies and Gentlemen, Teachers and Friends, we have gathered here this
afternoon tor the opening session ot the second Jerusalem Philosophical
Encounter on "Meaning and Use" dedicated to the memory of Yehoshua
Bar-Hillel. This is not the place to review his many contributions to the
philosophy of mathematics, language and science, nor is it the proper time to
assess his great impact as an educator and a teacher. However, on this
occasion it is fitting to say a few words in memory of a great philosopher and
a beloved friend.
There seems to be no way to define a great philosopher other than pointing
to outstanding examples. Yehoshua was one such example. If philosophy is
defined as the love of wisdom, Yehoshua was a lover of knowledge. If philo-
sophy is defined as the struggle against dogma, misconcep1ion, and confusion,
Yehoshua was a tireless fighter. And if philosophy is defined as the search
for meaning and understanding, Yehoshua was a devoted seeker.
Above all he possessed a genuine passionate intellectual curiosity that is
the trademark of great minds. His personal and intellectual style were
inseparable. All of Yehoshua's activities - personal and professional- were
characterized by the same warmth, vigor and wit. As students we admired
Yehoshua not only for his incisive mind and his personal courage and
integrity, but also for his love of man, his enthusiasm for life and his mar-
velous sense of humor.
In the history of ideas, different people are remembered for different
things: a theorem, an invention, a paradox, their impact on others, or their
contribution to the Zeitgeist. Yehoshua will be remembered, I believe, pri-
marily for what he was: a deep and penetrating critic, a warm and com-
passionate philosopher, a free spirit in search of meaning and truth.
Yehoshua lived with an intensity that his heart could not bear for long.
We have all lost something with his departure. The world was better and
more exciting when Yehoshua lived in it, and our lives are richer and deeper
because we have known him. There is probably no better way to pay tribute
XI
XII ADDRESS
to Yehoshua Bar-Hillel than this meeting of outstanding scholars who gather

here in Jerusalem to discuss the problem of Meaning and Use that was
always so close to his heart.
The Hebrew University of Jerusalem Amos Tversky

w. V. QUINE
USE AND ITS PLACE IN MEANIN G
The notion of meaning is stubborn. It does not submit readily to satisfactory

scientific formulation, and yet it is deeply rooted in everyday discourse and
not easily dispensed with. Semantics, moreover, or the theory of meaning, is
a vitally important subject, despite the disreputable character ot its ostensible
subject matter.
What, then, is the semanticist to study and analyze, if not the meanings of
words? The use of words. John Dewey was urging this point in 1925.
"Meaning," he wrote, " ... is primarily a property of behavior" (Experience
and Nature, p. 179). And just what property of behavior might meaning then
be? Well, we can take the behavior, the use, and let the meaning go.
How, then, may we set about studying the use of words? Thus take a
decidedly commonplace and unambiguous word: 'desk.' What are the cir-
cumstances of my use of this word? They include, perhaps, all the sentences
in which I ever have used or shall use the word, and all the stimulatory situa-
tions in which I uttered or shall utter those sentences. Perhaps they include
all the sentences and stimulatory situations in which I would use the word.
The sentences and stimulatory situations in which I would now use the word
might even be said to constitute the meaning of the word for me now, if we
care to rehabilitate the dubious term 'meaning.' However, the range of
sentences and stimulatory situations concerned is forbiddingly vast and ill
organized. Where is one to begin?
For a provisional solution, consider what we often actually do when asked
the meaning of a word: we define the word by equating it to some more
familiar word or phrase. Now this is itself a quick way of specifying the range
of sentences and situations in which the word is used. We are specifying that
range by identifying it with the range of sentences and situations in which
the other and more familiar word or phrase is used. Happily we can spare
ourselves the trouble of cataloguing all those sentences and situations, be-
cause our pupil has already mastered the use of the more familiar word or
phrase.
We may persist, then, in the old routine of giving meanings by citing
synonyms. The behavioral doctrine of meaning does not oppose that. Wha
1
A. Margalit (ed.), Meaning and Use, 1-8. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
2 W. V. QUINE
the behavioral doctrine of meaning contributes is theoretical. it purports to

explain this synonymy relation itself, the relation between the word whose
meaning is asked and the more familiar word or phrase that we cite in reply.
The behavioral doctrine tells us that this relation of synonymy, or sameness
of meaning, is sameness of use.
The method of giving the meaning of a word by citing a synonym is
convenient but very limited. It accounts for only a small minority of the
entries in a dictionary. Often the lexicographer will resort to what he calls a
distinction of senses: he will cite several partial synonyms, some suitable in
some kinds of context and others in others. When he does this, he has to
distinguish the kinds of context by providing a general characterization of
each, usually by reference to subject matter. And in many cases there is no
appeal even to partial synonyms; the use of a word can be taught in other
ways. In general, given any sentence all of whose words are familiar except
the word in question, what needs to be taught is how to paraphrase that
sentence into an equivalent whose words are all familiar.
General instructions for paraphrasing the sentential contexts of a word
into unproblematic sentences: such is the lexicographer's job. The citing of a
direct synonym is just one form that such instructions may take, and it is
feasible less often than not. What is more to the point than the relation of
synonymy of words to words and phrases, then, as a central concept for
semantics, is the relation of semantical equivalence of whole sentences. Given
this concept, we readily define the other: a word is synonymous to a word or
phrase if the substitution of the one for the other in a sentence always yields
an equivalent sentence.
And when do sentences count as semantically equivalent? A provisional
answer from the behavioral point of view is evident: they are equivalent if
their use is the same. Or, trying to put the matter less vaguely, we might say
that they are equivalent if their utterance would be prompted by the same
stimulatory situations.
But clearly this will not do. They cannot both be uttered at once; one must
be uttered to the exclusion of the other. On any occasion where one of the
sentences is uttered, moreover, there must have been a cause, however trivial,
for uttering it rather than the other. It may hinge merely on a phonetic
accident: the choice of a word in the one sentence may have been triggered
by a chance phonetic resemblance to a word just previously heard. Clearly
we ask too much if we ask of two equivalent sentences that they be prompted
by all the same stimulations. And anyway, if a criterion required actually
USE AND ITS PLACE IN MEANING 3
comparing the stimulatory conditions for the volunteering of sentences, it

would surely be hopeless in practice; for utterances are on the whole virtually
unpredictable. The motives for volunteering a given sentence can vary
widely, and often inscrutably: the speaker may want to instruct, or console,
or surprise, or amuse, or impress, or relieve a painful silence, or influence
someone's behavior by deception.
We can cut through all this if we limit our attention to the cognitive
equivalence of sentences; that is, to the sameness of truth conditions. We are
then spared having to speculate on the motives or circumstances for volun-
teering a sentence. Instead we can arrange the circumstances ourselves and
volunteer the sentence ourselves, in the form of a query, asking only for a
verdict of true or false. Cognitively equivalent sentences will get matching
verdicts, at least if we keep to the same speaker. He can be mistaken in his
verdicts, but no matter; he will then make the same mistake on both
sentences.
I remarked that it would be too much to require of two equivalent sentences
that their utterance be prompted by all the same stimulations. Now, however,
we are evidently in an opposite difficulty: we are requiring too little. We are
requiring only that he believe both or disbelieve both or suspend judgment
on both. This way lies little more than material equivalence; not cognitive
equivalence.
The solution to this difficulty is to be found in what John Stuart Mill called
concomitant variation. To get this effect we must limit our attention for a
while in yet another way: we must concentrate on occasion sentences. These,
as opposed to standing sentences, are sentences whose truth values change
from occasion to occasion, so that a fresh verdict has to be prompted each
time. Typically they are sentences that contain indexical words, and that
depend essentially on tenses of verbs. Examples are 'This is red' and 'There
goes a rabbit'; these might be designated more particularly as observation
sentences. Further examples are 'He is a bachelor' and 'There goes John's
old tutor'; these do not qualify as observation sentences, but still they are
occasion sentences. The truth value of 'He is a bachelor' varies with the
reference of the pronoun from occasion to occasion; similarly the truth value
of 'There goes John's old tutor' depends both on the varying reference of the
name 'John' and on who happens to be passing down the street at the time.
Now if our interrogated informant is disposed to give matching verdicts on
two such occasion sentences on every occasion on which we query the two
sentences, no matter what the attendant circumstances, then certainly the
4 W. V. QUINE
two sentences must be said to be cognitively equivalent for him. One such
pair is 'He is a bachelor' and 'He is an unmarried man.' Another such pair,
for a particular speaker, may be 'There goes John's old tutor' and 'There
goes Dr. Park.'
These two pairs of examples differ significantly from each other in that the
second pair qualifies as cognitively equivalent only for a particular speaker,
or a few speakers, while the first pair would qualify as cognitively equivalent
for each speaker of the language. It is the difference between cognitive
equivalence for an individual, or for an idiolect, and cognitive equivalence
for a language. It is the latter that we are interested in when we expound the
semantics of a language. Cognitive equivalence for the individual, however,
is the prior notion conceptually, that is, in respect of criterion. Two occasion
sentences are equivalent for him if he is disposed, on every occasion of query,
to give them matching verdicts or, on doubtful occasions, no verdict. The
summation over society comes afterward: the sentences are equivalent for the
language if equivalent for each speaker taken separately.
This unanimity requirement works all right for our core language, Basic
English so to say, which all English speakers command. However, when
recondite words are admitted, some pair of occasion sentences may fail of
cognitive equivalence for an ignorant speaker merely because of misunder-
standing. If we still want to count those sentences cognitively equivalent for
the language, we may do so by relativizing the unanimity requirement to an
elite subset of the popUlation.
Cognitive equivalence of two occasion sentences for a speaker consists in
his being disposed to give matching verdicts when queried in matching stimu-
latory circumstances. We can easily make this notion of stimulatory circum-
stances more explicit. It is a question of the external forces that impinge on
the interrogated subject at the time, and these only insofar as they affect his
nervous system by triggering his sensory receptors. Thanks to the all-or-none
law, there are no degrees or respects of triggering to distinguish. So, without
any loss of relevant information, we may simply identify the subject's external
stimulation at each moment with the set of his triggered receptors. Even this
identification is very redundant, since the triggering of some receptors will
have no effect on behavior, and the triggering of some receptors will have no
different effect from what the triggering of other neighboring receptors would
have had. However, the redundancy is harmless. Its effect is merely that two
occasion sentences that are cognitively equivalent, in the sense of comm-
anding like verdicts under identical stimulations, will also command like
verdicts under somewhat unlike stimulations.

Each overall momentary stimulation of our interrogated subject is to be
identified, I have suggested, with a subset of his receptors. The stimulation
that he undergoes at any moment is the set of receptors triggered at that
moment. This makes good sense of sameness and difference of stimulation of
that person from moment to moment. It does not make sense of sameness of
stimulation of two persons, since two persons do not share the same recep-
tors. They do not even have exactly homologous receptors, if we get down to
minutiae. But this is all very welI, for I am not having to equate stimulations
between persons. The notion of cognitive equivalence of occasion sentences
for a single person rests on sameness and difference of stimulations of that
person alone, and the subsequent summation over society appeals then to
cognitive equivalence for each separate person, with no equating of stimula-
tions between persons.
I feel that the relation of cognitive equivalence is in good shape so far as
occasion sentences are concerned. The relation is defined for the individual
and for society, and the definition can be applied by a routine of query and
verdict. There remain, of course, the other sentences - the standing
sentences.
There remain also the single words, and their relation of synonymy to
other words and phrases. We saw earlier that this relation presents no dif-
ficulty, once we have fixed the relation of equivalence of sentences. One word
is synonymous to another word or phrase if substitution of the one for the
other always yields equivalent sentences. Or, now that our equivalence re-
lation for sentences is cognitive equivalence, we should say that a word is
cognitively synonymous to a word or phrase if substitution of the one for the
other always yields cognitively equivalent sentences. Granted, the relation
even of cognitive equivalence of sentences is now under control only for
occasion sentences. However, I think this is already enough to settle cognitive
synonymy of words to words and phrases across the board. If a given word is
interchangeable with a given word or phrase in alI occasion sentences, in-
variably yielding a cognitively equivalent sentence, then I think the inter-
changeability can be depended on to hold good in alI standing sentences as
welI.
If this be granted, then a conceptual foundation for cognitive synonymy is
pretty firmly laid. The courses, as stonemasons calI them, are as folIows. First
there is the relation of sameness of overall stimulation of an individual at
different times. This is defined, theoreticalIy, by sameness of triggered
6 w. V. QUINE
receptors. Next there is the relation of cognitive equivalence of occasion

sentences for the individual. This is defined by his disposition to give match-
ing verdicts when the two sentences are queried under identical overall stimu-
lations. Next there is the relation of cognitive equivalence of occasion
sentences for the whole linguistic community. This is defined as cognitive
equivalence for each individual. Finally there is the relation of cognitive
synonymy of a word to a word or phrase. This is defined as interchangeability
in occasion sentences salva equivalentia. We could take the nominal further
step, if we liked, and define the cognitive meaning of a word as the set of its
cognitive synonyms.
Strictly speaking, this interchangeability criterion of synonymy requires
some awkward reservations regarding the positions in which the substitutions
are allowable. For instance, it would never do to require interchangeability
within direct quotations; and this reservation extends, in diminishing degrees,
to indirect quotation and other idioms of propositional attitude. I shall pass
over this difficulty, for it is a familiar and perennial one, and I have nothing
new to say about it.
Anyway we must remember that the synonymy of words and phrases,
however well defined, is not the mainstay of lexicography. What are wanted
in general, as I said earlier, are instructions for paraphrasing the sentential
contexts of a word into unproblematic sentences by whatever means; the
citing of a direct synonym is just one form that such instructions can some-
times take. The relation of equivalence of occasion sentences offers a founda-
tion equally, however, for all this. If the use of a word can be pinned down by
instructions for paraphrasing its sentential contexts at all, I expect it can be
pinned down by instructions for paraphrasing just those contexts that are
occasion sentences.
If we may measure the familiarity of words by their frequency, we may
perhaps schematize the task of the monoglot lexicographer as follows. Let
us define a gloss of a sentence s, with respect to one of its words w, as any
cognitively equivalent sentence lacking wand containing only other words
of s and words of higher frequency than w. A word may be called reducible if
all occasion sentences that contain it admit of such glosses with respect to it.
The lexicographer's task, then, is a systematic specification of glosses of
occasion sentences with respect to all reducible words. This leaves him doing
nothing about the irreducible words, which comprise the core language. I
welcome this outcome on the whole, for the monoglot lexicographer's com-
pulsive explanations of irreducible words have been a waste. But he should
still add a few supplementary cognitive equivalences for the benefit of speakers
whose frequencies diverge somewhat from the national average. For instance
he should continue to define 'gorse' as 'furze' and 'furze' as 'gorse.'
I am of course stopping short stiII of the needs of practical lexicography in
one conspicuous respect: I am attending only to the cognitive side, ignoring
emotional and poetic aspects. Regarding those further aspects I have nothing
to suggest.
My consideration of cognitive equivalence has been limited to occasion
sentences thus far, and I have urged that occasion sentences already provide a
broad enough base for lexicography. However, there is no need to limit
cognitive equivalence to occasion sentences. We can extend the relation into
standing sentences in several fragmentary but substantial ways. Standing
sentences grade off into occasion sentences, after all. Verdicts on occasion
sentences have to be prompted anew on each occasion, while verdicts on
standing sentences may stand for various periods. The shorter the periods,
the more the sentence resembles an occasion sentence. The more it resembles
an occasion sentence, the more applicable our criterion of cognitive equi-
valence: the criterion of like verdicts under like stimulation. We might even
extend this criterion to all standing sentences, provided that we take it only
as a necessary condition of cognitive equivalence and not a sufficient one.
For occasion sentences it is necessary and sufficient.
From another angle a sufficient but not necessary condition of cognitive
equivalence can be brought to bear on standing sentences. Namely, we can
exploit the relation of cognitive synonymy which I already defined on the
basis of cognitive equivalence of occasion sentences. One standing sentence is
cognitively equivalent to another if it can be transformed into the other by a
sequence of replacements of words or phrases by cognitive synonyms. This
sufficient condition can be broadened by submitting the standing sentences
not just to substitution of synonyms but also to other sorts of paraphrase:
sorts that have already been found to preserve cognitive equivalence among
occasion sentences.
These conditions do not quite add up to a definition of cognitive equi-
valence for standing sentences. If a pair of standing sentences meets the
necessary condition and not the proposed sufficient one, the question of their
cognitive equivalence has no answer. But in their incomplete way the con-
ditions ao make the notion widely applicable to standing sentences. Mean-
while it is defined for occasion sentences, and this, I have urged, is basis
enough for cognitive lexicography,
8 w. V. QUINE
I have been concerned in all these remarks with monoglot semantics, not
polyglot; not translation. Criteria are harder to come by in the polyglot
domain, particularly in the case of radical translation, where there are no
bilinguals to exploit. The most serious difference is this: cognitive equivalence
for a single individual is definable for occasion sentences generally by same-
ness of verdict under sameness of stimulation; but between two individuals
this definition carries us little beyond the observation sentences (Word
and Object, pp. 41-49).
If a bilingual is available, we can treat the two languages as his single
tandem language; and then we can indeed define cognitive equivalence of
occasion sentences generally, for him, even between the languages. But this
is still cognitive equivalence only for him and not for a linguistic community
or pair of communities. Only if we have a whole subcommunity of bilingual
can we summate over the individuals, as we did in the monoglot case, and
derive a bilingual relation of cognitive equivalence of occasion sentences at
the social level. The polyglot case thrives, it would seem, just to the extent
that it can be treated as monoglot. Thus the theory I have been developing
here has no bearing, that I can see, on the indeterminacy of translation.
Harvard University
DONALD DAVIDSON
MOODS AND PERFORMANCES
Frege held that an adequate account of language requires us to attend to

three features of sentences: reference, sense, and force. Elsewhere I have
argued that a theory of truth patterned after a Tarski-type truth definition
tells us all we need to know about sense.! Counting truth in the domain of
reference, as Frege did, the study of sense thus comes down to the study of
reference.
But how about force? In this paper I want to consider force in the only
form in which I am certain that it is a feature of sentences, that is, as it serves
to distinguish the moods. The question I am concerned with is, can a theory
of truth explain the differences between the moods?
In trying to answer this question I am responding, belatedly, alas, to a
challenge put to me by Yehoshua Bar-Hillel some years ago; he asked me
how it might be possible to represent mood within the confines of a theory of
truth.
One reason the analysis of mood is interesting is that it prods into pro-
minence certain relations between what sentences mean, and their uses. We
have on the one hand the syntactic, and presumably semantic, distinction
among moods (such as: indicative, imperative, optative, interrogative), and
on the other hand the distinction among uses of sentences (such as: to make
assertions, to give orders, to express wishes, to ask questions).
The moods classify sentences, while uses classify utterances; but the moods
indirectly classify utterances, since whatever distinguishes sentences can be
used to distinguish utterances of them. So we may ask, what is the relation
between these two ways of classifying utterances; how are assertions related
to utterances of indicative sentences, for example, or commands to utterances
of imperative sentences?
The simplest suggestion would be that the associated classes of utterances
are identical: utterances of imperatives are commands, utterances of inter-
rogatives are question-askings, etc. This idea appears to find support in
Dummett's book on Frege. Here is how Dummett explains Frege's use of the
assertion sign or judgment-stroke:
9
A. Margalit (ed.), Meaning and Use, 9-20. Dordrecht, D. Reidel. All Rights Reserved.
This Article Copyright 1979 by Donald Davidson.
10 DONALD DAVIDSON
The judgment-stroke is the sign of assertion proper, that which carries the assertive force.
It is therefore not a functional expression, or part of one: we cannot enquire of it what its
sense is, or what its reference is; it contributes to the meaning of the complex sentential
symbol in quite a different way ... it is only the sentence to which the judgment-stroke
is prefixed which may be said to express a sense or to stand for a truth-value: the whole
expression with the judgment-stroke neither expresses anything nor stands for anything -
it asserts something: it asserts, namely, that the thought expressed by what follows the
judgment-stroke is true. 2
Here Dummett says that it is sentences that make assertions, where I think it
would be more natural to say that an assertion is an utterance, and it is the
speaker who makes the assertion. However, this may be no more than a
terminological complaint; what bothers me is the implied claim that assertion
and the indicative mood can be this closely identified. For there are many
utterances of indicative sentences that are not assertions, for example indi-
cative sentences uttered in play, pretense, joke and fiction; and of course
assertions may be made by uttering sentences in other moods. (Utterances of
"Did you notice that Joan is wearing her purple hat again?" or "Notice that
Joan is wearing her purple hat again" may on occasion simply be assertions
that Joan is wearing her purple hat again.) And similarly for the other moods;
we can ask a question with an imperative or indicative ("Tell me who won
the third race," "I'd like to know your telephone number"), or issue a
command with an indicative ("In this house we remove our shoes before
entering").
Needless to say, Dummett knows all this, and if he temporarily allows
himself to overlook these cases for the sake of the larger view, it is only
because he believes that there is a clear sense in which the counter-examples
are deviant. But what is this sense? Austin made a distinction between what
he called the "normal" or "serious" uses of a sentence and the "etiolated" or
"parasitical" uses} If such a distinction could be made in a non-circular way,
and it turned out that the normal or serious use of indicatives was to make
assertions, of imperatives to issue commands, of interrogatives to ask ques-
tions, and so on, then the desired connection between the moods and uses of
sentences would be established.
There surely is some important connection between the moods and their
uses, and so we are bound to think that there is something natural, serious,
or normal, about using a sentence in a certain mood to perform a "corres-
ponding" act. The question is whether this feeling can be articulated in a way
that throws light on the nature of the moods. It is easy to see that appeal to
what is "serious" or "normal" does not go beyond an appeal to intuition. It
MOODS AND PERFORMANCES 11
is no clue to the seriousness of a command that it is uttered in the imperative

rather than the indicative; similarly, a serious question may be posed in the
imperative rather than the interrogative mood. And if "normal" means usual,
or statistically more frequent, it is dubious indeed that most indicatives are
uttered as assertions. There are too many stories, rote repetitions, illustra-
tions, suppositions, parodies, charades, chants and conspicuously unmeant
compliments. And in any case the analysis of mood cannot plausibly rest on
the results of this sort of statistical survey.
Dummett's solution is to switch from the serious or normal to the con-
ventional: an assertion is an indicative uttered under conditions specified by
convention; a command is an imperative uttered under other conventionally
given conditions; and so forth. So he writes, " ... assertion consists in the
(deliberate) utterance of a sentence which, by its form and context, is recog-
nized as being used according to a certain general convention ... " (p. 311).
And of imperatives, " ... the utterances of a sentence of a certain form, unless
special circumstances divest this act of its usual significance, in itself con-
stitutes the giving of a command" (pp. 301-302). He sums up with this advice
on how to approach the subject of the relations between the moods and their
uses:
... the correct approach is to consider utterances as conventionally demarcated into types,
by means of the form of linguistic expressions employed, and then to enquire into the
conventions governing the use of the various types of utterance (p. 302).
Dummett's view that linguistic actions like assertion and command consist in
uttering sentences in the indicative or imperative moods under conventionally
specified conditions is central to his picture of language when coupled with
the thesis that there is a further convention that assertions are made with the
intention of saying what is true. For these two ideas together would establish
a direct connection between languages as used in conventional ways and a
certain overall purpose (to say what is true).
I agree that we must find connections between how sentences are used and
what they mean if we are to give a foundational account of language. I am
doubtful, however, that either link in Dummett's chain will hold. I cannot
now discuss the second link, the supposed convention of trying to say what is
true. But it is relevant in the present context to comment on the claim that
the utterance of an indicative sentence under conventional conditions
constitutes an assertion.
One difficulty is obvious but may be superable: if there is to be a general
12 DONALD DAVIDSON
account of assertion along these lines, there will have to be conventions that
explain how assertions are made by uttering sentences not in the indicative
mood. But perhaps it is plausible that ifthere are conventions linking indica-
tives and assertions, there are additional conventions linking other moods
with assertions.
The real trouble is that the right sort of conventions do not exist. Of course
it is true that if an indicative is uttered under the right conditions, an assertion
will have been made. It may even be that we can specify conditions that are
necessary and sufficient for making an assertion; for example, I think that in
order to make an assertion a speaker must represent himself as believing what
he says. But none of this suggests that the conditions are conventional in
nature.
It must also be conceded that interpreters and speakers of a language are
generally able to tell when an assertion has been made, and that this ability
is an essential part of their linguistic competence. Furthermore, knowledge
of linguistic and other conventions plays a key role in the making and detect-
ing of assertions. Costume, stance, tone, office, role and gesture have, or may
have, conventional aspects, and all these elements can make a crucial con-
tribution to the force of an utterance. We may easily allow all this without
agreeing that merely by following a convention, indicative or imperative
utterances become assertions or commands.
There are, I think, strong reasons for rejecting the idea that making an
assertion (or issuing a command, or asking a question) is performing a purely
conventional action. One reason is, as I have been suggesting, that it is so
hard to say what the convention is. (For example, if an asserter necessarily
represents himself as believing what he says, one would have to describe the
conventions by following which one can represent oneself as believing what
one says.) A second point is this. Quite often we understand an utterance in
all relevant respects except that we do not know whether it is an assertion.
One kind of teasing consists in leaving the issue of assertion open in the mind
of the teased; historical novels, or romans aclef, deliberately leave us puzzled.
Is some conventional aspect of utterance omitted? What is it? And if we
could say, then why would not the tease or romancer include that very item
in his utterance?
Whatever is conventional about assertion can be put into words, or some-
how made an explicit part of the sentence. Let us suppose that this is not now
the case, so that Frege's assertion sign is not just the formal equivalent of the
indicative mood, but a more complete expression of the conventional element
in assertion. It is easy to see that merely speaking a sentence in the strength-

ened mood cannot be counted on to result in an assertion; every joker,
storyteller and actor will immediately take advantage of the strengthened
mood to simulate assertion. There is no point, then, in the strengthened mood;
the available indicative does as well as language can do in the service of
assertion. But since the indicative is not so strong that its mere employment
constitutes assertion, what must be added to produce assertion cannot be
merely a matter of linguistic convention.
What this argument illustrates is a basic trait of language, what may be
called the autonomy of linguistic meaning. Once a feature of language has
been given conventional expression, it can be used to serve many extra-
linguistic ends; symbolic representation necessarily breaks any close tie with
extra-linguistic purpose. Applied to the present case, this means that there
cannot be a form of speech which, solely by dint of its conventional meaning,
can be used only for a given purpose, such as making an assertion or asking
a question.
The argument has a simple form: mood is not a conventional sign of
assertion or command because nothing is, or could be, a conventional sign
of assertion or command. The reason for this, it should be stressed, is not
that the illocutionary force of a speech act is a purely mental, interior, or
intentional aspect of the act. 4 Of course assertion or command must be in-
tentional, as must meaning in the narrow sense. But it is part of the intention
that the act should be interpreted as assertive or commanding, and therefore
part of the intention that something publicly apparent should invite the
appropriate interpretation.
It would be easy to become involved in a dispute about the extent to which
a speaker's intention to perform an act that will be interpreted as being
assertive must be realized before his act is correctly called an assertion. It is
too much to insist that an assertion has been made only if it is actually inter-
preted as an assertion; it is too little to demand only that the intention be
present. We need not settle the question how far an asserter must succeed in
his intention; all that matters here is whether an asserter or commander must
intend his hearer to recognize his intention through his (the asserter's) em-
ployment of what he knows or believes to be a linguistic convention. If there
were such a convention, we should find it easy to say what it is, and easy, in
the great majority of cases, to say whether or not it has been observed. But
though we can usually determine whether or not an assertion has been made,
we cannot in general say what convention was followed. The reason we
14 DONALD DAVIDSON
cannot say is, I have urged, that there is no such convention.

It would be a mistake to conclude that there is no conventional connection
between the moods and their uses. There would, indeed, be no such connec-
tion if certain analyses of the moods were correct. David Lewis, for example,
has boldly suggested that all non-indicative sentences may be "treated as
paraphrases of the corresponding performatives, having the same base
structure, meaning, intension and truth-value."5 Thus "Fry that egg" would
have the same analysis as "I command that you fry that egg." Lewis thinks
the two sentences might have a different range of uses, but since this difference
would not, on his theory, arise from a difference in meaning, his theory
simply denies that mood has any conventional significance.
An analysis with the same consequence was proposed many years ago by
Herbert Bohnert. 6 Bohnert's proposal was that imperatives have the structure
of disjunctions of a certain kind. Thus "Fry that egg" would be rendered,
"Either you fry that egg, or X will happen," where X is something presumed
unwanted by the person addressed.
These theories draw their strength from the fact that we can, and often do,
use indicatives to do the work Dummett believes is conventionally assigned
to the other moods. And in fact if we want to move in this direction, there is
an even simpler and, I think, better, theory available, which is to assign to
imperatives the same semantic analysis as is assigned to the most directly
corresponding indicative (treat "Fry that egg" just as "You will fry that egg"
is treated).7
It is a virtue of these theories that they make evident the fact that having a
truth value is no obstacle to a sentence's being used to issue a command or
ask a question. But this merit of reductive theories also accounts for their
failure, for simply reducing imperatives or interrogatives to indicatives leaves
us with no account at all of the differences among the moods. If any of the
reductive theories is right, mood is as irrelevant to meaning as voice is often
said to be. If mood does not affect meaning, how can we hope to explain the
connection between mood and use, whatever the connection comes to?
Reductive analyses abandon rather than solve the problem with which we
began.
We are now in a position to list the characteristics a satisfactory theory of
mood should have.
(1) It must show or preserve the relations between indicatives and cor-
responding sentences in the other moods; it must, for example, articulate the
sense in which "You will take off your shoes," "Take off your shoes," and
"Will you take off your shoes?" have a common element.

(2) It must assign an element of meaning to utterances in a given mood
that is not present in utterances in other moods. And this element should
connect with the difference in force between assertions, questions and
commands in such a way as to explain our intuition ofa conventional relation
between mood and use.
(3) Finally, the theory should be semantically tractable. If the theory
conforms to the standards of a theory of truth, then I would say all is well.
And on the other hand if, as I believe Bar-Hillel held, a standard theory of
truth can be shown to be incapable of explaining mood, then truth theory is
inadequate as a general theory of language.
The difficulty in meeting the three requirements is obvious. The first two
conditions suggest that mood must be represented by operators that govern
sentences, the sentences governed being either indicative (in which case no
operator is needed for the indicative mood), or neutral (in which case an
operator is needed for every mood). The third condition, however, seems to
prohibit all but truth-functional sentential operators, and it is clear that truth-
functional operators cannot serve to give a plausible interpretation of mood.
Dummett seems to me to be right when he says that mood is not like a
functional expression, that we cannot ask what its sense or reference is, and
that (therefore) a sentence with a mood indicator "neither expresses anything
nor stands for anything." As Geach says, a mood indicator is not like any
other part of speech; " ... it is necessarily sui generis. For any other logical
sign, if not superfluous, somehow modifies the content of a proposition;
whereas this does not modify the content ... "8
Dummett and Geach make these negative claims for what seem to me
partly wrong, or confused, reasons. Dummett thinks a sentence with a mood
indicator cannot express or stand for anything because the sentence
" ... asserts something, ... namely, that the thought expressed by what
follows the judgment-stroke is true" (loc. cit.). This is something I have
urged that no expression can do; but the idea also seems wrong for another
reason. If the assertion sign asserts that the thought expressed by the rest of
the sentence is true, then the imperative sign should assert that the thought
expressed by the rest of the sentence is to be made true. But this proposal
wipes out the distinction between assertion and command. Geach says
instead that the mood indicator (understood as Frege understood it)
" ... shows that the proposition is being asserted" (loc. cit.). This proposal
preserves the needed distinction.
16 DONALD DAVIDSON
I have argued against both Geach and Dummett that no mood indicator
can show or assert or in any other way conventionally determine what force
its utterance has. But if this is so, we are left with no clear account of what
mood contributes to meaning. Indeed, we seem to have a paradox. Mood
must somehow contribute to meaning (point 2 above), since mood is clearly
a conventional feature of sentences. Yet it cannot combine with or modify
the meaning of the rest of the sentence in any known way.
Let us turn for help to what Austin called the "explicit performatives." We
have rejected the idea put forward by David Lewis that imperatives be re-
duced to explicit performatives, but it remains open to exploit analogies.
Austin drew attention to the fact that " ... we can on occasion use the
utterance 'Go' to achieve practically the same as we achieve by the utterance
'J order you to go.'''9 But how are explicit performatives to be analyzed?
Austin held that performatives have no truth value on the ground that utter-
ing a sentence like "I order you to go" is not typically to describe one's own
speech act but rather to issue an order.
This is perhaps an accurate account of how we would characterize many
speech acts that consist in uttering explicit performatives. But as a description
of what the words that are uttered mean, this view introduces an intolerable
discrepancy between the semantics of certain first-person present-tense verbs
and their other-person other-tense variants. And the problem is adventitious,
since what is special to explicit performatives is better explained as due to a
special use of words with an ordinary meaning than as due to a special
meaning.
Ifwe accept any of the usual semantics for explicit performatives, however,
the difficulty recurs in a form that is hard to avoid. According to standard
accounts of the matter, in a sentence like "Jones ordered Smith to go" the
final words ("Smith to go") serve to name or describe a sentence, or a pro-
position, or the sense of a sentence. To show the relevant embedded sentence,
we may recast the whole thus: "Jones ordered Smith to make it the case that
Smith goes." And now, on the standard accounts, the sentence "Smith goes"
cannot, in this context, have anything like its ordinary meaning. Therefore,
neither can it have anything like its ordinary range of uses. However, "I order
you to go" (or, recast, "I order you to make it the case that you go") has the
same form as "Jones ordered Smith to go," and so should have the same
analysis with appropriate changes of person and time. It follows that in
uttering "I order you to go" I cannot mean by the words "you go" anything
like what I would mean by them if they stood alone; in the present context,
I am using these words merely to refer to a sentence or the proposition it

expresses. It seems quite impossible, then, that if any standard analysis of
such sentences is correct, an utterance of "I order you to go" could be an
order to go.
Or, to make the same point with assertion: one way of establishing the
fact that I am not asserting that it is raining when I utter the words "It is
raining" is by prefixing the words "Jones asserted that." According to most
analyses of such sentences, the same effect should be expected if I prefix the
words "I assert that."
This difficulty is one among the difficulties with the usual analyses that has
prompted me to urge an entirely different approach to the semantics of in-
direct discourse, belief sentences, sentences about commands, orders, hopes,
expectations and so on: the whole unholy array of attitude-attributing
10cutions)O Leaving aside complications that arise when we quantify from
outside into the governed sentences (or their scrambled surfaces), my pro-
posal is this. Bearing in mind that it is in any case utterances, not sentences,
that have a specific truth value and semantics, we should be satisfied with an
analysis of the truth conditions of utterances of words like "Jones asserted
that it is raining." I suggest that we view such an utterance as the utterance
of two sentences; "Jones asserted that," and then, "It is raining." If I assert
that Jones asserted that it is raining, I do this by asserting "Jones asserted
that" and then uttering, usually non-assertively, the sentence that gives the
content of Jones' assertion; in this case, "It is raining." The function of the
"that" in an utterance of "Jones asserted that" is to refer to the following
utterance, which gives the content. So to put the idea in a wordy but sugges-
tive way: an utterance of "Jones asserted that it is raining" has the effect of
two utterances:
Jones made an assertion whose content is given by my next utterance.
It is raining.
This analysis accounts for the usual failure of substitutivity in attributions
of attitude without invoking any non-standard semantics, for the reference
of the "that" changes with any change in the following utterance. It also
allows the second utterance to consist, on occasion, in making an assertion,
as it will ifl say truly, "I make an assertion whose content is given by my next
utterance." Similarly, I may be giving an order in saying "You go" even if
these words follow an utterance of "I order that," or "This is an order."
I propose to treat the non-indicative moods in much the same way as
18 DONALD DAVIDSON
explicit performatives, but not by reducing the other moods to the indicative.
Here is the idea. Indicatives we may as well leave alone, since we have found
no intelligible use for an assertion sign. We will go on, as is our wont, some-
times using indicative sentences to make assertions, sometimes using them to
do other things; and we will continue to use sentences in other moods to make
assertions when we can and find it fun.
In English we mark the non-indicative moods in various, occasionally
ambiguous, ways, by changes in the verb, word order, punctuation or intona-
tion. We may think of non-indicative sentences, then, as indicative sentences
plus an expression that syntactically represents the appropriate transforma-
tion; call this expression the mood-setter. And just as a non-indicative
sentence may be decomposed into an indicative sentence and a mood-setter,
so an utterance of a non-indicative sentence may be decomposed into two
distinct speech acts, one the utterance of an indicative sentence, and the other
the utterance ofa mood-setter. It should not bother us that in fact we do not
usually perform these acts one after the other but more or less simultaneously.
Just think of someone rubbing his stomach with one hand while patting his
head with the other.
We have seen that the mood-setter cannot be treated semantically as a
sentential operator of any ordinary sort, and that it seems quite impossible
to give a plausible account of how the meaning of a non-indicative sentence
can be the result of combining the meaning of an indicative with the meaning
of the mood-setter. I suggest that we accept the semantic independence of
indicatives from their accompanying mood-setters by not trying to incor-
porate the mood-setter in a simple sentence with the indicative. There is the
indicative sentence on the one hand, and before, after, or alongside, the
mood-setter. Or, better, thinking of the utterance, there is the utterance of
the indicative elements, and there is (perhaps simultaneously) the utterance
of the mood-setter. The utterance of a non-indicative is thus always decom-
posable into the performance of two speech acts.
So far, the proposal is not clearly incompatible with the proposals of
Geach, Dummett, and perhaps others. I have, indeed, dropped the assertion
sign, but that may be considered largely a notational matter. I have also
rejected an explanation of the meaning of the mood-operator in terms of a
conventional indicator of the force with which the particular utterance is
made. So there is a vacuum at the center of my account; I have failed to say
what the mood-setter means.
Geach remarked that what I call a mood-setter cannot be regarded as any
other part of speech. This was because he thought of it as a part of a longer

sentence, and realized it did not have the semantic properties of a sentential
operator. We have removed the mood-setter from the indicative sentence it
accompanies. The only form the mood-setter can have - the only function
it can perform - is that of a sentence. It behaves like a sentence an utterance
of which refers to an utterance of an indicative sentence. If we were to re-
present in linear form the utterance of, say, the imperative sentence "Put on
your hat," it would come out as the utterance of a sentence like "My next
utterance is 'imperative," followed by an utterance of "You will put on
your hat."
This suggests the semantic situation, but syntax makes it wrong. The
mood-setter cannot be any actual sentence of English, since it represents a
certain transformation. I do not want to claim that imperative sentences
are two indicative sentences. Rather, we can give the semantics of the utter-
ance of an imperative sentence by considering two specifications of truth
conditions, the truth conditions of the utterance of an indicative sentence got
by transforming the original imperative, and the truth conditions of the
mood-setter. The mood-setter of an utterance of "Put on your hat" is true if
and only if the utterance of the indicative core is imperative in force.
Mood-setters characterize an utterance as having a certain illocutionary
force; they do not assert that it has that force, since only speakers make
assertions. But if someone wishes to give an order, he may well do it by
uttering the imperative mood-setter assertively. Then if the truth conditions
of the mood-setter hold (if what the speaker has asserted is true), his utterance
of the indicative core will constitute giving an order. There are plenty of other
ways he can give the same order; for example, by asserting "This is an
order," or "I hereby command that"; or simply by uttering "You will take
off your hat" as an order.
I believe this proposal satisfies the three requirements we listed for a
satisfactory analysis of the moods.
First, on the proposal there is an element common to the moods. Syntac-
tically, it is the indicative core, which is transformed in the non-indicative
moods. Semantically, it is the truth conditions of this indicative core.
Second, mood is systematically represented by the mood-setter (or its
absence in the case of the indicative). Mood-setters function semantically as
sentences, utterances of which are true or false according as the utterance of
the indicative core does or does not have the specified illocutionary force.
The meaning of the mood-setter is conventional in whatever sense meaning
20 DONALD DAVIDSON
in general is, but there is no suggestion that this meaning determines the
illocutionary force of an utterance of the mood-setter, of its associated in-
dicative, or of the pair. The conventional connection between mood and
force is rather this: the concept of force is part of the meaning of mood. An
utterance of an imperative sentence in effect says of itself that it has a certain
force. But this is not the "says" of "asserts" (except on occasion and in
addition). What it says, in this non-asserted sense, may as like be false as true.
This fact does not affect the conceptual connection between mood and force.
Third, a straightforward semantics, based on a theory of truth for utter-
ances, works as weII here as elsewhere. In particular, all the utterances the
theory takes as basic have a truth value in the standard sense. On the other
hand, if I am right, the utterance of a non-indicative sentence cannot be said
to have a truth value. For each utterance of a non-indicative has its mood-
setter, and so must be viewed semantically as consisting in two utterances.
Each of the two utterances has a truth value, but the combined utterance is
not the utterance of a conjunction, and so does not have a truth value.
The University of Chicago
NOTES
1 For a recent statement of my position, see "Radical Interpretation," Dialectica 27

(1973): 313-328. For its defense, see "Reply to Foster," in: Truth and Meaning: Essays
in Semantics, G. Evans and J. McDowell (eds.), Oxford, 1976.
2 Michael Dummett, Frege: Philosophy of Language, London, 1973, pp. 315, 316.
3 J.L. Austin, How to Do Things with Words, Cambridge, Mass., 1962, p. 22.
4 I am indebted to Michael Dummett for making me appreciate this point.
S David Lewis, "General Semantics," in: Semantics of Natural Language, D. Davidson
and G. Harman (eds.), Dordrecht, 1972, p. 208.
6 Herbert Bohnert, "The Semiotic Status of Commands," Philosophy of Science 12 (1945).
7 Yes-no interrogatives would then be treated, perhaps, as having the same semantics as
the corresponding affirmative indicative; or, on another option, as having the same
semantics as the alternation of the affirmative, with the negation of the affirmative,
indicative. WH-questions might be assigned the same semantics as the corresponding open
sentences in the indicative. Here as elsewhere in this article my remarks about the inter-
rogative mood are sketchy. As Jaakko Hintikka has pointed out to me, my general
program for the moods may run into trouble when a ser ious attempt is made to apply it to
interrogatives.
8 P.T. Geach, "Assertion," The Philosophical Review 74 (1965): 458.
9 J.L. Austin, How to Do Things with Words, p. 32.
10 For a fuller discussion of this proposal, see my "On Saying That," Synthese 19 (1968 -
1969): 130-146.
w. V. QUINE
COMMENTS
Some sentences are true, some are false, and some are neither. According to
what seem to be the usual standards, questions are neither true nor false;
neither are optatives; neither are imperatives. A sentence with a truth value
is a sentence in the indicative mood. But not all sentences in the indicative
mood have truth values. Occasion sentences have no truth values they can
call their own; only their utterances have truth values. Nor is it clear even
that all utterances of indicative sentences have truth values; for there is the
matter of truth-value gaps. If a sentence contains a singular term that fails to
designate, then the utterances of the sentence may lack truth value; Strawson
has plausibly urged that this account usually fits ordinary usage.
However, if an utterance does have a truth value, then according to these
standards it is an utterance of a sentence in the indicative mood. We are
therefore bound to accord the indicative mood a central position in our
semantics if, with Davidson, we are to base our semantics on truth conditions.
Two questions then arise: (I) why base our semantics on truth conditions?
and (2) if we do, how can we accommodate the non-indicative utterances?
My answer to the first question is implicit in my paper of yesterday. If we
are to explore the use of sentences systematically, we must cut through the
jungle of possible motives for the volunteering of sentences. We can do so by
volunteering the sentences ourselves, and asking only for assent and dissent;
and these verdicts are verdicts strictly of truth and falsity.
There remains the second question, Bar-Hillel's: how a semantics focused
on truth values can accommodate non-indicative utterances. Davidson has
just now offered a very original answer, by extending the very original answer
that he once gave to the corresponding question regarding indirect quotation
and other propositional attitudes. In his earlier analysis of 'Galileo said that
the earth moves' he construed the conjunction 'that' as demonstrative
pronoun and the subordinate clause as a separate utterance, thus:
Galileo said that.
The- earth moves.
Similarly, as I interpret him, he would construe the imperative 'Put on your
21
22 W. V. QUINE
hat' as in effect 'It is desirable that your hat be on,' hence:

Your hat is on.
That is desirable.
Again, both are indicative. He does not actually treat both parts as indicative,
but I think he might as well.
Giving a central role to the indicative does not mean giving a central role
to assertion, for an utterance in the indicative mood is only occasionally an
assertion. It may be the antecedent of a conditional utterance, or the negate
of a negative utterance, or an alternand of an alternative utterance. For an
indicative utterance to qualify as an assertion, it would have to stand alone
rather than as a component of a longer sentential utterance.
But grammatically independent position is still not enough to qualify an
indicative utterance as an assertion, in Davidson's view. And we easily see
why. We would like to say that Galileo said that the earth moves even if we
disagreed with him. And we certainly would like to tell you to put on your
hat without asserting that it was already on.
So assertion, for Davidson, must involve more than mere grammatically
independent utterance. In fact he makes it involve much more: he seems to
require belief, either real or seriously feigned with a view to deception. He
excludes fiction, jokes, play-acting. This narrow version of assertion conflicts,
I think, with usual attitudes; certainly with mine. I should prefer to allow
assertion and all the other performances - interrogation, command, and the
rest - to occur within fiction and jokes and play-acting, on a par with their
occurrence in serious discourse. Davidson himself remarks that no distinctive
operators could serve to exclude assertions or questions or commands from
these frivolous contexts, since the operators would straightaway be counter-
feited by the joker or play-actor or fictioneer himself. The natural course,
surely, is to allow assertions to occur in fiction and jokes and play-acting, and
then simply to withhold moral sanctions in these connections. In such
contexts it is both possible and moral to assert what one does not believe.
I am left, then, with an unsolved problem. Mere grammatically independent
utterance must not constitute assertion, if Davidson's demonstrative theory
of moods and propositional attitudes is to stand. Sincerity, on the other
hand, or studied deception, seems too much to require of assertion. I am at a
loss for an acceptable intermediate standard.
Harvard University
EDDY M. ZEMACH
AWARENESS OF OBJECTS
How does a subject become aware of some (e.g., external) objects? An

answer to this question, I think, can be given only when one recognizes that
there are two distinct language games which are essentially connected with
our concept of becoming aware of an object. I shall try to show that although
both these games are necessary in order for us to have our present concept of
being aware of something, they are quite distinct games, and can be played
independently of each other. The question "What does an awareness of X
consist in?" will thus be shown to be essentially misguided, and its apparent
meaningfulness due only to our tendency to commit a conceptual short-circuit
here and lose sight of the fact that our epistemic game is actually two games
played together in a certain way. Thus I shall attempt to show that the ques-
tion, "What does awareness of an entity consist in?" is wrong headed in
exactly the same way that the question, "What is the chemical constitution
of a trump card?" is wrong: they can only be due to a conceptual confusion
about the nature of these games.
The main language game we play is the Game of Objects (GO, for short).
There are, we say, many objects in the world; singular terms denote them;
predicates are satisfied by them; statements can be made about them; those
statements which attribute to them the predicates they satisfy are true, others
are false, or meaningless. Need I go any further? We are all accomplished GO
players and its basic notions have been meticulously clarified for us by the
work of many contemporary philosophical semanticists.
GO, however, must be played together with some other game (anyone of
whole series of games) which is to be used as the admission gamefor GO. To
explain this procedure, let me contrast it with the way we play chess. When
one sits down to play chess one begins by placing the chessmen on the boatd.
What kinds of pieces are to be used in the game, how many of each kind, and
where they are to be placed - all these questions are already determined by
the rules of the game. Suppose, however, that we wish to make this process
less arbitrary by making one's right to put one's chessmen on the board
dependent upon the outcome of another game. We may, e.g., use checkers
as the admission game of chess: the winner in a match of checkers has the
23
24 EDDY M. ZEMACH
right to place one of his chessmen on the chess board. We would start playing
chess only after having played checkers a certain number of times, thus accu-
mulating enough chessmen to play with. We may use an admission game to
determine the kind and number of pieces we are going to use in a game, their
exact locations, etc.
Now, unlike chess, GO does not come equipped with a list of all the pieces
(objects) admitted in the game. We must, therefore, have some procedure for
determining what objects are there in the world, i.e., we need a game which
could be used as an admission game, determining which objects are to be
allowed on the GO board.
The admission game we in fact use is again only too well known. I shall
call it 'PMG,' the Perception and Memory Game. This term is probably not
a very happy one, since 'Perception' and 'Memory' are used as success terms,
i.e., they already reflect the function of PMG as the admission game of GO. It
is already incorporated into the logic of 'perceive' and 'remember' that if a
perceives x then x exists, and if a remembers that p then p is the case. To
avoid this implicit connection to GO I shall write 'PMG-NI' when referring
to the non-intentional use of PMG, i.e., when PMG is played "phenomeno-
logically," without assuming that perception and memory have a role in
determining which objects are there in the world.
Players of PMG-NI would not distinguish between seeing and that which
is seen, hearing and that which is heard, feeling and that which is felt, etc., in
the same way that we do not distinguish between a smile (i.e., "that which is
being smiled") and smiling, or between a kick and "the activity of executing
the kick" (i.e., kicking).
Such a game is sketched, e.g., in Wilfrid Sellars' adverbial theory of the
mind. On his view, one senses redly, hears violinly, remembers childhoodly,
etc. Sellars is even ready to say, as would a player of PMG-NI, that one
simply reds or lauds:
The adverbial theory views such verbs as 'feels,' 'experiences,' 'senses' - and, as we
shall see, 'thinks' - as generic verbs, and the expressions formed from them by "adding
a reference to the objects felt, experienced, etc." as specific verbs. It follows from this that
in a perspicuous language, we would not use the generic verb in forming its species, but,
instead, say
Tom pains
rather than
Tom feels pain
just as we say
The book is rectangular
AWARENESS OF OBJECTS 25
rather than
The book is rectangularly shaped.
In this perspicuous language we would not say
Tom senses a red triangle
but
Tom a-red-triangles.!
We, however, train our children differently. We train them to use most
common nouns to name objects, not mental processes. We use PMG as an
admission game for GO, thus conferring upon it an intentional status, as a
final bid of Four Hearts makes Hearts the trump suit for this match. That
seeing, hearing, remembering (again I use these terms "neutrally" or "pheno-
menologically") are sources of information about objective reality is a priori
true. It is not a "happy coincidence" that these, and no others, are our ways
of finding out what things are really like; rather, this fact is a logical feature of
our version of GO - the language game of saying what things are really like.
This logical connection was already noticed by S. Shoemaker,2 who pointed
out that it is a necessary, a priori, truth that the great majority of our percep-
tions and memories would be veridical. Shoemaker rightly insists that it is
impossible inductively or empirically to establish the veracity of our sense
perceptions and memories,3 since in the process of gathering inductive evi-
dence we must assume the veracity of sense perception and memory beliefs -
i.e., presuppose exactly that which we intended to prove. Shoemaker, how-
ever, does not ask for the reasons for this "strange phenomenon." Can we
explain the a priori veracity of our perceptions? A simple explanation which
would not attribute this veracity to, say, divine grace, or to the goodness of a
Cartesian Demon, would be based upon the principle that whenever we find
a necessary a priori truth we have a rule of some game or other. In the present
case, so it seems, it is the rule of GO which decrees that seeing, remembering,
etc. are not to be conceived of as one's being in a certain state but as one's
being directly (i.e., intentionally) in touch with reality. Roughly speaking,
perceptions (etc.) are veridical since to be real is partially defined, in this
version of GO (i.e., PMG-GO), as being the object of a perception (etc.).
This point can be demonstrated by comparing PMG with some other
possible candidates for the role of admission game, e.g., dreaming, speaking,
thinking, hoping, etc. Suppose I have dreamt that the Empire State Building
was pulled down, and that many other people have had the same dream at
about the same time. Our daytime observations, however, do not bear this
dream out - we can all see the Empire State Building standing in its usual
26 EDDY M. ZEMACH
place the next morning. Which should we believe - our dreams or our day-
time seeing? Can we adjudicate between the two? Do we have to? The obvious
answer is, No. Although it is possible for one to believe that one's dreams
would come true, this only means that one's dreams will conform to one's
(or others') later sense perceptions; it presupposes, and does not challenge,
the special status of sense perception. Perception can verify dreams, dreams
cannot verify perceptions.
Another example: a preacher tells us the world will end on January I, 1970.
However, nothing significant is perceived to have happened on or after that
date. We continue to have our regular visual and acoustic sensations of trees,
houses, etc. Should we now say that we are perceiving the world, and what the
preacher said was false, or should we count these sensations as misleading
and illusory, since the world has already come to an end on January I, 1970?
The answer is of course given a priori. How can we explain that? A proposi-
tion is true if things are as it says they are. In both cases, we are quite sure
that the dream, or what the preacher had said, are not true. Therefore, it
seems that part ot the meaning of what it is for things to be in a certain way
depends upon what can be seen, or heard, or remembered.
It is evident that there is a logical connection between 'p is the case,' and
'I seem to see that p is the case,' a connection that does not exist between
'p is the case' and 'I dream that p is the case,' or 'I say that p is the case,' or
'I fancy that p is the case,' or 'I think that p is the case,' etc. But if this one
connection is not inductive or empirical, it must be a priori. If I inductively
establish that on many occasions when 'I hope that p' was true 'p' was true,
then the truth of '1 hope that p' would lend an inductive support to 'p.'
But the connection between 'I seem to see that p' and 'p' is not like that at
all. In order to make empirical verification possible we have to have an
admission game ready at hand. There must be some episternic predicates
E1 ... En such that, prima jacie, 'aEIP' would criteriologically, i.e., non-
empirically, lend support to 'p.' It is also necessary that there would also
be some other predicates, N 1 . N n , such that 'aN IP' could lend support
to 'p' only inductively. Obviously, the distinction between E and N predicates
can be made only a priori, and it determines the nature of the game one plays.
This decision cannot be given a justification, but it need not have any.
I believe that Paul Feyerabend makes a similar point, when he says that it
is not necessarily "preferable to interpret theories on the basis of an obser-
vation language rather than on the basis of a language of intuitively evident
statements ... or on the basis of a language containing short sentences."4 I
agree that this is indeed so, i.e., that we may consider 'is a very short ex-
pression' or 'is intuitively clear' as our only E predicates. It is not too difficult,
1 think, to compile a fairly long list of alternative E predicates, that is, of
games other than PMG which may be used as admission games for GO. The
following is a short sample.
Game A. One prays to God, and God makes one know what is the object
one now encounters.
Game B. Everyone is cerebrally connected to a central relay station, where
a comprehensive computer is found. The relay station continuously broad-
casts to you (directly into your brain) the exact nature of the surrounding
area. Suppose, moreover, that the said information is couched in highly
"theoretical" terms (say, of nuclear physics) and does not mention any
"observational" terms.
Game C. Each person has a sense of direction determining, in mathematical
exactitude, his location relative to the universal grid. Objects are referred to
in terms of their respective locations only, and no other consideration matters
scientifically: No observation by means of the senses is ever needed for pre-
diction, explanation, and other scientific purposes. Suppose that information
concerning the location of an entity is transmitted by telepathy. The "sensory
manifold" will thus be considered as a private, even idiosyncratic, affair, like
our moods and pains in the present framework.
It is essential to note that in any of these (and similar) games the E pre-
dicates used must be used as epistemic question-stoppers. The answer to
"How do you know that you E1P?" can only be "because I E1P." (Compare
this to our E predicates; Question: "How do you know that you are appeared-
to-redly?" Answer: "1 am appeared-to-redly.")
Simpler admission games which partially use PMG, but not in the way we
use it, can also be employed for the same purpose. Here are some examples:
Game D. Whatever anyone says, goes. It is impolite to object Of contradict
him. Thus, if Jones has said, "There is a tree over there," there is a tree over
there. The statement "There isno tree over there" is false. We all know that
there is a tree over there, and our (sufficient) ground for holding this belief is
that Jones said there is a tree over there.
Game E. If one says something in a very loud voice, what he says is true.
We accept it, that is, until somebody else says the opposite in a louder voice.
In this case we consider the first statement as refuted, and its opposite as
true.
Game F. The test of the truth of what you say is putting your arm in the
28 EDDY M. ZEMACH
fire for a second. If you cannot take the pain, what you have said is untrue.
Game G. Every object is tagged or labeled. You read the label and know
whatever there is to know about the object.
It is obvious that any GO which would be played with any of these games
as its admission game will be very different from ours; yet it is clearly con-
ceivable that, under certain conditions (in some possible worlds) it may be
more reasonable, and pragmatically expedient, to adopt any of the games
A-G, rather than PMG, as the GO admission game.
Now, I think, I can examine the mechanics of our own game, PMG-GO,
more closely. How does our language (and our attitudes in general) reflect
the "decision" to use perceiving and remembering criteriologically, i.e., as
admission agencies for our Game of Objects? The device used is simple and,
if I may say so, ingenious: Intentionality. Instead of talking about visual-
treeing and aural-rivering, as we would have done had we been playing
PMG-NI, we talk, in the same circumstances, about seeing a tree and hearing
a river. The role of PMG as object supplier for GO is thus built right into
sentences (and attitudes) using perception and memory predicates. Instead
of these three components (1) Va (where 'V' is some PMG-NI predicate)
(2) E!qJ (where 'qJ' is some GO substantive), and (3) the rule Va ~ E!qJ, we
can now employ the short, theory-laden, intentionalistic statement aVqJ
(e.g., Jack sees this tree). The object, we now say, is "immediately given" to
us. We are aware of it.
This shorthand device, intentionality, is the most fundamental feature of
PMG-GO. Using this device we can now describe our experience as that of
recognizing entities, identifying and re-identifying objects; in short, the game
is now built around the idea that it is possible for one entity to have another
entity directly present to it. However ~ and this is a crucial point - the basic
rules of this game make it impossible for there to be an entity, or a state of
an entity, which is awareness.
To talk the language of awareness, or intentionality, is to use PMG for
GO purposes, i.e., to playa game in which all PMG-NI terms are used to
denote objects or states of objects. It is therefore not an interesting psycho-
logical or ontological discovery that Hume, Kant and James could not
empirically observe consciousness, or the knowing self, but a necessary
logical feature of the PMG-GO framework. It is a constitutive rule of this
game that whatever is experienced is an object of some sort, recognized and
identified by its typical features. Thus consciousness, or experience ('ex-
perience' being the general term for all PMG-NI activities) cannot have any
character or typical features in PMG-GO. It is not the mysterious, dia-

phanous nature of consciousness which makes it so cob-web thin that it is
hard to detect. Rather, it is the rule that in playing PMG-GO one ought to
attribute every recognizable feature of his experience to an intentional object
of this experience. These features can then be taken as defining what this
object is (e.g., a tree), a certain property of that object (e.g., brown), a certain
way this object looks under certain conditions (e.g., bent when in water), etc.
Generally speaking, every determination of the PMG-NI treeing-now goes
to the tree (in Direct Realism) or to some other intentional object sanctioned
in any other version of GO, such as treeness, a tree percept, a tree-ish sensum,
etc. The consciousness of the tree is thus necessarily eliminated as a legitimate
object for inspection by an explicit rule of the game (hence, the "trans-
parency" of consciousness is not a fact to be discovered through a pheno-
menological investigation).
The epistemic-intentional language is created by adopting the following
convention: being in some positions, such as treeing or reding, is considered
as being in a privileged epistemological position vis-a-vis a certain tree, or a
red patch, such that one may just read off directly from those objects them-
selves what their (say, visual) properties are. This convention has the effect
that a certain kind of statements concerning the "properties" of "objects"
(quotation marks used here to highlight the GO use of these terms) comes to
be considered as the basic inventory of pieces to be used in the game. To
signify x's right to issue a statement of this kind, i.e., to say that he is in a
position to "read" the properties of y in the said manner, we say that x is
conscious of y. Using Sellars' terminologyS this rule may be described as
letting all PMG positions serve as entry tickets for statements (to be called
'observation statements,' or something like it) into some GO positions. This
is why the notion of a conscious state is a conceptual chimera: in PMG-NI
there are states, but no consciousness of anything. In PMG-GO there is con-
sciousness of objects, but no states of awareness.
In conclusion, I would like to suggest the following analogy which may
best summarize my account of intentional epistemic terms such as 'see,'
'remember,' etc. The semantic status ofthese terms, I would say, is similar to
that of logical (truth-functional) connectives. One of the most significant
achievements of philosophical logic in the beginning of this century has been
in showing that 'and,' 'or,' 'if ... then,' etc. are not names of "logical
objects," and do not designate relations between objects. Truth tables made
it evident to most philosophers that there are no "logical relations" between
30 EDDY M. ZEMACH
objects or states of affairs. Now I wish to say that intentional terms are in
this sense similar to logical terms: the truth conditions of statements contain-
ing them are truth-functions of the truth condition of their logical consti-
tuents. Intentional terms do not stand for anything, be it an entity, a state
or a relation. In this sense it is a mistake to symbolize 'a sees b' as 'Sea, b)'
just as it is a mistake to symbolize 'a and b' by 'R(a, b).' One cannot say
"but 'a sees the tree' is true: hence there is something which is a's seeing the
tree" any more than one can argue "but 'p and q' is true; hence there must
be something which is the conjointness ofp with q." Conscious states ought
to be analyzed away; they exist no more than logical entities do; seeing is
no more real than andness.
The Hebrew University of Jerusalem
NOTES
1 Wilfrid Sellars, "Metaphysics and the Concept of a Person," in: The Logical Way of
Doing Things, K. Lambert (ed.), New Haven, Yale, 1969, p. 235.
2 S. Shoemaker, Self Knowledge and Self Identity, Ithaca, Cornell, 1963.
3 Again I use 'perception' and 'memory' in the "phenomenological" sense, i.e., as short
for 'seeming to perceive' and 'seeming to remember.'
4 "Science without Experience," Journal of Philosophy 66 (1969): 791-794.
S "Some Reflections on Language Games," Philosophy of Science 21 (1954): 204-228;
rev. ed. in: Science, Perception and Reality, London, Routledge and Kegan Paul, 1963.
IGAL KVART
COMMENTS
In his paper Eddy Zemach introduces the language game of perception and
memory as a separate admission game to our main language game in which
we describe the world of objects. He then presents some truths involving
perception as a priori, which he explains in terms of rules of our language
games; wherefore he argues that the rules of our language game make it im-
possible for there to be an entity which is awareness, from which he concludes
that the analysis of intentional terms should not make them stand for
anything.
I shall present my comments from within the conceptual perspective in
which Zemach's paper is formulated. In my response I shall first criticize his
conception of the perception-and-memory game as an admission game, and
argue that it involves a confusion concerning the meta-language of our
object-language game, a confusion which is obstructive to the rest of his
arguments. Then I shall elaborate on the sense in which one may say that
certain truths concerning perception are a priori, and the sense in which one
may not; and then use the meta-linguistic status of perception talk to clarify
the ontological status of awareness and perception in our language game and
the analysis of intentional terms.
1. In his paper Zemach attempts to draw a distinction between two

games: the game of objects - GO, and a perception-and-memory game-
PMG, and to describe the second as an admission game to the first, in the
sense that it determines which objects there are for GO to refer to.
Zemach draws extensively on Wilfrid Sellars' paper "Some Reflections on
Language Games" (Science, Perception and Reality, London, 1963). In this
paper, the picture of a language game is provided with the following elements:
positions in the game, where "to occupy a position in a language is to think,
judge, assert that so-and-so"; moves in the game, where "to make a move in
a language is to infer from so-and-so that so-and-so" (p. 329). Though moves
in a language transpose one from one position to another, there are never-
theless positions which one can be at without having moved to, i.e. initial
positions. Observation sentences constitute such positions.
31
32 IGAL KVART
However, initial positions can be arrived at in ways other than moves in

the language. There can be transitions to (initial) positions in the language
from non-linguistic positions, which Sellars calls language-entry transitions,
in which stimuli give rise to initial positions in the game as responses.
Now the reason for which Zemach brings up his admission game is because
"unlike chess, GO does not come equipped with a list of all the pieces (objects)
admitted in the game. We must, therefore, have some procedure for deter-
minining what objects are there in the world, i.e., we need a game which could
be used as an admission game, determining which objects are to be allowed
on the GO board." However, the observation sentences in a language L, e.g.
'this table is red,' serve, among other things, precisely this function, as they
include referring expressions. So this job is already done in the language
game and the need for an admission game is at best superfluous.
Worse still, one might have thought that admission game sentences like 'I
perceive a red table' would provide a missing link between the main language
game L and objects in the world. This job is already performed in the
Sellarsian theory by the important device of language-entry transitions. But
clearly Zemach's concept of admission game does not even provide an
alternative, since the question which objects are admitted in the game would
arise again for his admission game just as well. No language can provide a
linkage with the world for another language. Thus, either he would resort
here to another admission game for his admission game, and be driven to an
infinite regress, or else would resort to Sellars' mechanism of language-entry
transitions. But then, why mediate through the admission game? The linkage
can be done - as it is done in Sellars - directly with the language L. The
admission game then is conceived to perform ajob which it does not do and
which is already done by another mechanism. Moreover, it is central to
Sellars' account, and he makes efforts to make it so, that it could fit into a
theory of how an organism might come to learn a pattern-governed behavior,
and this feature of course deserves merit. For Sellars, with this in mind, both
moves in the language as well as language-entry transitions can be represented
as stimulus-response pairs. Acceptable or not, it is an arguable model in
which Sellars' account of how a language hooks us to the world fits; but
clearly Zemach's 'I perceive a red table' admission-game positions would fit
worse in an S-R model than Sellars'; and Zemach mentions no alternative
account.
But worse still, to say 'I perceive a red table' (as in Zemach's PMG-game)
is to say: 1 occupy the position 'this is a red table' in my L-game, and I
COMMENTS 33
moved there from a (certain type of) extra-linguistic position (e.g. a perceptual
sensation, through a language-entry transition). But to realize this is to
realize that Zemach's PMG functions as a meta-language of L (or GO),
describing language-entry transitions into L, and thus is no admission
language of an independent status. This should be quite obvious, since
having a perception or a sensation is being stimulated in a certain way which
serves as an extra-linguistic position from which one can move to an initial
position in the language game (describing the content of the perception by an
observation sentence) through a language-entry transition. To use a part of
the meta-language as an admission game is not just futile, but also runs
contrary to a main purpose of Sellars in developing this account, as he says:
"I shall have achieved my present purpose if I have made plausible the idea
that an organism might come to playa language-game ... without having
to be playing a meta-language game ... " (p. 328).
Realizing this point would prove fruitful to other concerns of Zemach in
this paper. Thus, he correctly notices the phenomenon of perception reports
as question-stoppers, but provides no explanation for that. One could, how-
ever, be in a position to explain this phenomenon if one realized the meta-
linguistic character of perception-and-memory talk, instead of mistakenly
construing it as an admission game. Thus, consider:
Q: How do you know that p?
A: 1 perceived that p.
There are no further questions to be asked here. The answer, of course, does
not provide an inter-language move through some inference tickets, but
rather meta-linguistically calls on a language-entry transition, which carries
one beyond the limits ofthe language game, and thus cannot allow for further
questions-and-answers within it. A challenge within the game is a call to trace
a position through a legitimate move to it. This of course cannot be done
once one withdraws to an extra-language-game position (though through a
legitimate language-entry transition).
2. Zemach states that it is a priori true that the vast majority of our
perceptions are veridical, and that it is a priori true that we would prefer our
perceptions to our dreams in determining what is real in the world. Now it
is quite clear that, from the perspective of our present discussion, to adopt a
conceptual framework is to commit oneself to a language game, whose rules
govern the usage of the terms involved. And there is no doubt that as long as
34 IGAL KVART
we play a particular game we are bound by its rules, and thus may view
features determined by the rules as a priori, and thus as independent of our
experience; and indeed the examples mentioned above do so come out in
virtue of the nature of the language-entry transitions in the language game
we play. But although it is important to realize that we organize our ex-
perience within one conceptual frame, or language game, or another, extreme
changes in experience may give way to switching from one language game to
another. thus, imagine circumstances in which our visual, tactual and
auditory perceptions would cease to be correlated with a significant variety of
pleasureable and painful sensations, but these in turn would be highly cor-
related with the extent of our subscription to our dreams. In such circum-
stances there may occur a switch of a language game, where in the new game
the language-entry transitions would not transfer one to an initial position of
'this is red' from a position of having a red visual sensation, but would rather
be dominated by dreams results. Thus, the cases Zemach brings up as a
priori would be so only within a language game, but not in the sense of being
independent of experience in extreme cases where language games them-
selves may be so dependent. So Zemach's examples do reflect their a priori
status in the sense of being independent of experience as long as we operate
within our usual language game. But the stories he provides as examples are
under-determined as to whether circumstances could be so radically different
that a change in the language game would ensue, whereupon some of its
constitutive features which confer a prioricity would go by the board. thus
in a radical sense, the truths observed in these examples may not be inde-
pendent of experience, even though they are determined by the rules of the
game.
Again, the a priori character of true propositions involving perception
concepts is a reflection of the meta-linguistic level of perception talk, since it
is in the meta-language that the rules of the game are formulated, and it is
their special affinity to rules which gives such propositions their a priori
status.
3. To specify that a position in the language game L is an initial position

is to use the meta-language of L, as this is to express that there are language-
entry transitions to it from extra-language positions. to characterize extra-
linguistic positions in the meta-language as perception or memory positions
is to express that there are language-entry transitions from them to positions
in L. Further specifications of these positions in the meta-language would
COMMENTS 35
serve to specify which are the initial positions in L to which some language-
entry transitions from these positions lead. Consciousness characteristics
apply to these extra-language-game positions, and thus are used in the meta-
language of L as well. So the role of meta-linguistic talk involving these
concepts is the specification of the initial positions of the language game.
Thus, their role is not to refer or describe, and hence no states of conscious-
ness belong to the ontology of the meta-language, a fortiori to that of the
language game L. Therefore perception-and-memory constructions do not
have a relational form, but are rather, being meta-linguistic, analyzable as
specifying positions in the language game as initial positions (of a certain
type). However, it is a mistake on Zemach's part to assimilate them to
statements which are complex (in the sense of the propositional calculus),
since the latter are not meta-linguistic at all. The primary logical form of
perception-and-memory sentences is thus that of predication on linguistic
expressions of the language game (since it classifies them as positions). So
Zemach is right in relating the concept of 'being conscious' to an interplay
between two language games; but wrong in thinking that "they are quite
distinct games and can be played independently of each other," since they
are a language and its meta-language. But he is right in considering the
question 'What does a consciousness of x consist in' misguided, if he takes
the question to mislead to thinking that 'an awareness of x' is a referring
expression.

ASA KASHER
WHAT IS A THEORY OF USE?
I. INTRODUCTORY: BAR-HILLEL AND PRAGMATICS
Yehoshua Bar-Hillel was one of the pilgrim fathers of Pragmatics of Natural

Languages. He sailed his own positivistic ship, whose Carnapian hull he had
loaded up with an Ordinary Language freight, but instead of rebuilding his
ship on the open sea, he ventured on aiming his masts towards new colonies,
striving to join some established philosophical confederation, to unite with
some intellectual super-powers, and to indulge in some fruitful inland
navigation.
During the early 1950's, a couple of years after Strawson had launched his
pragmatic blows against Russell's semantics, and a few years before Austin
came over to Cambridge, Mass., to show how to do things with words, Bar-
Hillel published in Mind his paper "Indexical Expressions," thus adding to
the pragmatic studies of presuppositions and speech-acts a third pillar, viz.
"the investigation of indexical languages and the erection of indexica
language-systems."! As a matter of fact, none of these pillars of pragmatics
of natural languages is of a pure constitution, and telling the semantic
parts - be they in the capital of any pillar or in its base - from the pragmatic
parts of the shaft is never obvious, if possible. .
Throughout the twenty odd years that followed, Bar-Hillel spent a lot of
ink and energy in preaching pragmatics to logicians and linguistics. The
ensuing logico-linguistic papers of Richard Montague are well-known, but
the following two expressions deserve our attention too. It was Chomsky who
suggested, seven years ago, that "it may be that the next great advance in the
study oflanguage will require a forging of new intellectual tools that permit
us to bring into consideration a variety of questions that have been cast into
the waste-bin of 'pragmatics' ..."2 and it was Bar-Hillel himself who not
only dared to entitle a paper he published in 1971 "Out of the Pragmatic
"Wastebasket,"3 but also saw it fit to propose, in 1972, that "everyone who
has been engaged in Semantics or Syntax should clearly be engaged in
Pragmatics" - "yetzt und heute."4 Being Bar-Hillel's final comment during
the very last conference on language in which he took any part, these words
may indeed serve as his intellectual will.
37
Copyright (C) 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
38 ASA KASHER
II. NEW GOALS FOR THEORIES OF LANGUAGE USE
According to the traditional characterization of Pragmatics, an investigation

made in the field of language study is assigned to Pragmatics, if reference is
made in it, explicitly and essentially, to "the user of a language."5 Bar-Hillel
focussed his attention on the context of use, paving the way for Montague's
attempt to see through his brilliant formal lenses just indices - time of
utterance, place of utterance, addressee or addressees, etc. Saturated with
indexicality, intoxicated with excessive formal powers, Montague's theory
left most parts of Pragmatics of natural languages, such as the studies of
speech-acts and implicatures, out offocus, blurred. 6
In the present paper I would like to propound new goals for pragmatical
theories, present some major problems and propose some ingredients of
possible solutions. Conclusions might be drawn, somewhat indirectly, con-
cerning the long-troubled philosophical marriage of Princess Meaning and
the Earl of Use.
My starting point is almost a truism: No device is mastered and nothing is
mastered qua device without an adequate grasp of its standard uses. Thus, a
girl scout has not grasped the notion of a postage stamp, if she knows all
about perforated edges and can even tell the side that sticks from the side
that speaks, but knows nothing whatsoever about letters and postage. And a
scout-master does not have a thorough understanding of his organization if
he knows the ropes and can tell ajamboree from a merry rally, but is unaware
of the constitutive purposes of his movement.
A theory about anything which is of some standard use - be it a tool, an
artifact, an institution, or what have you - is explanatorily inadequate if it
fails to specify the constitutive elements of the use, such as function or
purpose. Accordingly, the goals of pragmatics should now become clear. It
is a truism that, in a sense, languages are used in fairly standard ways, and as
a maxim for BaHiol men has it, even a truism may be true. Hence, I propose
the goal of Pragmatics - the ultimate goal of pragmatical theories - to be
specification and explanation of the constitutive rules of the human competence
to use linguistic means for effecting literal purposes. 7
Turning to a brief elucidation of this proposal, let me first stress the point
that pragmatics is here confined to a study of competence. Keeping it off any
systematic consideration of independent factors of human behaviour, such as
psychological organization and physiological control, which take part in
linguistic as well as in non-linguistic behaviour, is on a par with keeping
WHAT IS A THEORY OF USE 39
computer engineering outside the confines of arithmetic. Pragmatics is the

arithmetic of language-use rather than the computer engineering of it.
No attempt will be made here to defend Chomsky's due distinction between
competence and performance, ~ but two observations are still in place. First,
any competence-performance delimitation is actually a theory about inter-
related systems that can be studied almost independently. Our linguistic
intuitions are data that should be explained by a competence-theory, just in
case any explanation in terms of some performance-theory is plausibly ruled
out. Indeed, complex syntactic structures, intricate meaning relations and
puzzling circumstances suggest different explanations, in terms of both com-
petence and performance theories, of usual discrepancies between com-
petence-theoretical descriptions of grammaticality and given intuitions of
acceptability.
A second observation pertains to some philosophical merits ot the
competence/performance distinction. In illustration of some light shed by it,
let me mention Ramsey's contention that logic is a normative science,9 as
extended by Wittgenstein to the realm of language. Indeed, linguistic norms
and ideal speakers are not laden with value, and here an "ought" is just a
disguised "is," awaiting beyond the unavoidable gap of abstraction and
embodiment. It seems that a philosophical defence of the concept of com-
petence, as contrasted with performance, will happily dissolve the apparent
paradox of a normative science.
Now, a blend ofWittgenstein's philosophy and Chomsky's methodology-
Summa Cantabrigiarum, if I may dub it - might not be to the taste of
everyone, but the prospects of such a blend do seem bright; parts of what
follows are in pursuit of a happy weave.
I would like to note in passing that the competence/performance distinc-
tion, when applied to linguistic use, does not coincide with Ryle's use versus
usage distinction. Usage overlaps just one facet of linguistic performance,
and use it la Ryle is unfortunately restricted to words, while we are interested
in the competence to use any linguistic means, sentences included.
For the present purposes I take it that a human competence is defined by a
finitely representable,lO flexible system of constitutive rules. It is finitely
representable because it is human; its rules are constitutive because they
define institutions; it is flexible because it tolerates some kinds of constitu-
tional amendments. In the sequel I shall take it for granted that finite
representability and flexible constitutiveness are adequacy conditions imposed
on any system of rules making up the pragmatic competence.
40 ASA KASHER
The pragmatic competence is that of using linguistic means for certain

purposes. Linguistic means are utterances of sentences. Although this is a
seemingly simple observation, it hums with complication. Here suffice it to
say that sentences are not series of inscriptions or word-types, but rather such
series under unambiguous representations on different levels, such as a level
which specifies intonational patterns and a level which assigns truth-
conditions. Since the use of any sentence is restricted to a suitable family of
contexts of utterance, an utterance of a sentence is characterizable as a pair:
a sentence and a context. There are different ways of analysing utterance, but
for the moment we are interested just in the conclusion that linguistic means
are utterances, contexts, inscriptions and sentences under their different
representations. 11
Concluding this brief introduction of general goals for pragmatical
theories, I would like to devote a few words to the concept of literal purposes
which are achieved by happy uses of linguistic means. Literal purposes are
basic in the sense that they do not induce any assumption or presupposition
about any other use of the same means under the same circumstances. Thus
using words and sentences for composing a sonnet is not a literal use of those
linguistic means, and the purpose is not a literal one, because writing a
sonnet cannot take place without the expressions of the sonnet being also
used for effecting some standard purposes of, say, asserting or requesting.
The literal use of a sentence under certain circumstances is not the most
important purpose the speaker has in his mind during the utterance of the
sentence; the central purpose of yelling "Fire!" may be to save someone's
life, but this is not the literal purpose, because it requires presupposing that
"Fire!" has another use which may contribute to a complicated process of
saving someone's life. Paraphrasing a remark by Davidson 12 I would claim
that there must be the literal use of uttering a sentence in a suitable context,
if there are other uses. According to the present proposal, pragmatics is
confined to the study of literal uses. Thus the formal patterns of sonnets or
inscriptive arabesques will not be treated of in such a theory of use. (Specific
claims about the nature of the literal uses will be made in the sequel.)
III. ADEQUACY CONDITIONS: UNITY VS. VARIETY
Approaches to a theory of language use are of two different kinds, to be

called here the variety approach and the unity approach. One might parody
both of them by describing the variety approach as that which is unable to
WHAT IS A THEORY OF USE? 41
show the wood for the trees and the unity approach as that which is unable
to show trees for the wood. The methodological point at issue is the theore-
tical role of a taxonomy of speech-acts. According to the variety approach,
advancing such a classification is a natural, if not necessary, first step,
provided that the classes are well-defined by clear and objective concepts)3
It is the latter rider which dimisses Austin's taxonomy of English illocu-
tionary verbs; I take Searle's arguments to that effect to be conclusive.1 4
However, better taxonomies of illocutionary acts or verbs might also not
stand up to criticism for a more important reason, viz. lack of what Hempel
calls "systematic import," which is the explanatory power of the system of
classes)5
According to the unity approach, which shares its methodology with the
one employed and defended by Chomsky, the significance of a taxonomy is
quite limited, even in the best case.
Mr. Unity is engaged in pursuit of universals of linguistic use. Formal
universals are abstract restrictions imposed on a class of rules for it to form a
certain competence. Substantive universals are the ultimate building blocks
of such rules)6 Universals of both kinds involve deep generalizations about
the mental equipment of any language user. It is clear that taxonomies are
unable to provide more than suggestive data for such pursuits of universals.
For example, the notion of "the direction of fit between words and the
world" should play some part in theoretical generalizations concerning
performatives, but as a mere criterion for a sheer classification it seems
worthless)7 It may be expected that current classifications of illocutions will
play in pragmatics of natural languages the same role that has been played in
syntax by phrase structure grammars. Some universals and related generali-
zations will be pointed out latter, in a nutshell.
Signs by themselves seem dead, said Wittgenstein, and what gives them
life is use. But how are things made lively for corpses of sentences and rocks
of contexts? Holding that the spirit of any linguistic use is a linguistic
institution, I would suggest the following criterion of adequacy ofpragmatical
theories:
(CA) For every context of utterance C, every sentence S of a natural

language L, and every ideal speaker Alpha of that language, the
following biconditional should be a true theorem:
Context C is linguistically appropriate for speaker Alpha to
utter in it sentence S of language L,
42 ASA KASHER
if and only if,

there is a linguistic institution of L which grants Alpha an
institutional role which enables him (or her) to achieve a
literal purpose he (or she) entertains in C, by uttering in it the
sentence s.
I shall call this adequacy condition 'Criterion A' of appropriateness.

There is a certain correspondence between this criterion and its role in
theories of use and the so-called 'Convention T,' introduced by Tarski and
defended by Davidson, and the role it plays in theories of meaning. I would
like to mention just two points of comparison. First, an adequate theory of
the linguistic appropriateness relationship among speakers, sentences and
contexts, in terms of linguistic institutions, institutional roles and literal
purposes, will constitute a major part of an adequate theory of the prag-
mati cal competence, provided that the theory fulfils some other adequacy
conditions, such as finite representability. However, such a theory may still
come short of an adequate theory of linguistic use, because a firm grasp of
an appropriateness relationship involves existential statements about institu-
tional roles and literal purposes, while the ability to use a particular language
consists of having a good grip of certain institutions, roles and purposes. A
similar argument has recently been advanced in the domain of semantics by
Dummett. 18
Secondly, there is some similarity between the form of evidence to be used
to support a theory of truth for a natural language, as suggested by Da-
vidson,19 and the form of evidence which might be used to support a theory
of appropriateness for a language.
Consider the following T-sentence and A-sentence:
(T) "Yored geshem" is true-in-Hebrew when spoken by x at time

t, if and only if, it is raining near x at t.
(A) Context C is linguistically appropriate for x to utter in it the
Hebrew sentence "Y ored geshem," if and only if, there is a
linguistic institution of Hebrew which grants x an institutional
role which enables x to achieve a literal purpose x entertains
at C, by uttering in it "Yored geshem."
Where evidence for (T) is evidence for the speaker's membership in the
Hebrew speech community, and his holding true the Hebrew sentence "Yored
geshem" under certain circumstances, evidence for (A) involves the speaker's
WHA T IS A THEORY OF USE? 43
membership in the same speech community, but also his holding appropriate
the same sentence under certain related circumstances. Whereas evidence
for (T) involves also the investigator's holding true his interpretation of the
Hebrew sentence under the same circumstances, evidence for (A) involves the
investigator's holding effective the speaker's speech-act under those cir-
cumstances.
Indeed, the attitude of holding effective is not innocent and it clearly needs
thorough analysis. On the other hand, the attitude of holding true seems less
basic than the attitude of holding appropriate, from which it seems to be
derived.
Among the key-concepts of Criterion A are: linguistic institutions,
institutional roles and literal purposes. Few words about each of them would
not be out of place.
An institution is a system of non-natural rules that govern a certain kind of
activity, by assigning roles and instituting facts, rendering acts and situations
meaningful and useful beyond their natural properties and potential.
An institutional role is a cluster of requirements a person has to fulfil in
order to operate in a certain way within a certain institution. Some institu-
tional roles, such as promisor and congratulator, seem to be determined
partly by the so-called "preparatory" and "sincerity" conditions. 20 Thus, for
one to be a congratulator he is required to believe that the event under congra-
tulation is in the hearer's interest, and he is required to be pleased at this
event. What the speaker is required to (pragmatically) presuppose at a context
of utterance in order to play a particular linguistic role is also part of the
constitutive specification of this role. 21
Another type of a linguistic institutional role is pointed out in Putnam's
theory of the division of linguistic labour, according to which necessary and
sufficient conditions for membership in the extensions of certain predicates,
for example, are known only to a subset of the community's population of
speakers - the "experts" - on whose judgments all other speakers usually
rely when employing these predicates. A similar kind of experts is suggested
by Kripke's theory of names and naming. 22
Searle's "essential" conditions determine for an iIlocutionary act of any
type what it counts as. A promise to do something, for example, counts as
placing the speaker under an obligation to do it. Indeed, this is a fact insti-
tuted by the linguistic institution of promising, through a happy activity of a
speaker who satisfies the requirements of the related institutional role.
The final ingredient of Criterion A I would like to discuss is the concept
44 ASA KASHER
of literal purpose. Earlier I stipulated that a literal purpose is one that does not
require presupposing another purpose, but here I would like to be a little
more specific.
Literal purposes of particular speech-acts seem to involve general kinds of
purposes, such as good old Communication. According to the variety
approach one should look, perhaps, for a taxonomy of such general purposes,
but let us avoid flogging a dead horse and try instead using the unity ap-
proach. To the surprise of nobody, the corps of unity are not in unity;
actually, another Homeric struggle takes place between the friends of
Thought Expression and the allies of Communication. Recently, a thorough
attempt to replace Communication by Representation at the foundations of
an intentional theory of language has been made, but it does not seem to
overcome the difficulty of Other Moods, such as questions and requests,
without overworking the concept of representation. 23
I would like to make now one natural addition to the list of general uses of
language. Institutions are commonly characterized by their coordinative
functions. Political institutions are obvious examples of social means of
coordination, and more intricate cases are indeed abundant. Now, every
linguistic institution enhances inter-subjective coordination, because every
speech-act which is performed happily within such an institution provides
information of a typical form about the speaker. The information conveyed
by one's speech-acts about oneself is always a disclosure of some preference
relations, on the part of the speaker, of some possible worlds over other,
related ones. For example: when (ideal speaker) Alpha requests Beta -
"Give me the red file, please" - he discloses his preference of any state of
affairs in which he has been given the file by Beta over another state of affairs
in which he has not been given it, ceteris paribus. If Alpha asks Beta - "Where
do you live?" - he makes it known to Beta that, everything else being equal,
he prefers a state of affairs in which he knows where Beta lives over a state of
affairs in which he does not. Similarly, upon asserting, by using the sentence
"Arabella sleeps," Alpha discloses the preference on his part of one state of
affairs, w, to another state of affairs, W', provided they differ from each other
just to the extent that in w Beta knows that Arabella sleeps while in w' Beta
does not know it.
T tried to develop this idea in much detail elsewhere,24 showing, I hope
successfully, that every speech-act carries a class of implicatures which
characterize the kind of the speech-act (assertion, request, advice, etc.) and
all these characteristic implicatures have the form of a preference relation
between two states of affairs that differ from each other in a specific way,
determined both by the type of speech-act and by the propositional content
of it. Hence, our candidate for the office of the general use of language, in
terms of which literal purposes are formulated, is the institutional disclosure
of preference relation.
All the foregoing theories, which try to establish a uniformity of literal use
of natural language - be it in terms of Communication, Coordination,
Representation, Expression of Thought, Preference-disclosure, or what have
you - are none a flash in a pan, and it seems that each of these provides
insights into use of language, but nevertheless it is clear that most of them
fail to furnish the study of language with intellectual tools that will enable us
to understand the use of natural language, as contrasted with any other type
of symbolic or institutional system. Articulated expression of thought has
been claimed to be present in painting and music,25 and therefore any attempt
to characterize natural language as an expressive vehicle of thought is
doomed to fail ifit falls short of pointing out crucial differences between, say,
Italian language and Italian metaphysical art, besides what obviously tells a
picture from a sentence. The case for foundational theories of coordination,
communication and representation is even worse, because non-linguistic
systems which manifest such uses are in profusion.
The case for institutional disclosure of preferences is more complicated.
Since there are obvious examples of non-linguistic institutions and non-
linguistic disclosures of preferences, the case should rest on the uniqueness
of institutional disclosure of preferences among all disclosures of preferences
and the uniqueness of the institution of disclosing preferences among all
institutions. The first uniqueness problem involves the distinction between
institutions and other systems, while the second uniqueness problem seems
to require for its solution intricate distinctions between different kinds of
institutions, unknown as yet.
Where the pronounced goals of pragmatical theories is to specify and
explain the human competence to use linguistic means, the theoretical dif-
ficulties posed by the second uniqueness problem (and by similar problems)
should not be overlooked. There are two major ways of tackling these pro-
blems; each of the two supports a different research program and, more
importantly, seems to endorse distinctive hypotheses about the nature of
human competence. I shall call them for short the "interactive" and the
"intra-active" views. I am going to argue that both views are wrong.
According to the interactive view, the human ability to use linguistic means
46 ASA KASHER
for effecting given purposes results from an application of the general human
powers of Use and WiII to particular systems of linguistic means. To put it
in a pseudo-formula:
(I l ) Linguistic Activity equals WiII plus Use plus Semantics and
Grammar.
It is thus viewed as a product of three interacting faculties of mind: the WiII
which determines purposes, both general and derivative ones; the indepen-
dent "tool box" of linguistic means; and the general power of Use which
consists of rationality principles, heuristics and perhaps some other kinds of
rules. Adherence to such an approach obviates the problem of characterizing
the use of language by denying it: there are no differences between the general
principles of use oflinguistic means and general principles of use of any other
means, and there is no essential distinction between the general purposes
effected by using linguistic means and those that are or might be effected
through some other tool box. Moreover, since meaning is considered under
this view to be independent of both use and purpose, any regularity to be
found in meaning and purpose relationships or in meaning and use relation-
ships is bound to play no constitutive role in any linguistic system. Semantics
and syntax are accordingly separable from the rest of our mental equipment.
According to the opposite, intra-active view, the human pragmatic com-
petence is a matchless power of our mind, which provides a general purpose
(or general purposes) of a unique kind, to be effected by putting to use, in an
unparalleled way, available and suitable linguistic means. To put this view
in a pseudo-formula:
(I 2) Linguistic Activity equals Linguistic Purposes plus Linguistic
Uses plus Semantics and Grammar.
A supporter of this approach does not deny the problem of characterizing the
use of language; on the contrary, all his attention is focussed on this very
problem. Ifhe is, for example, a linguistic institutionalist, he should hunt up
formal and substantive universals of linguistic institutions not shared by any
other kind of institution. Furthermore, under this view not only syntax and
semantics are separable from all other faculties of mind, but pragmatics is
also independent to that extent.
I hold both these views untenable, because far as they are from the truth of
the matter, they are stiII close enough to it to be able to throw each other off
its balance. Without attempting here any thorough sifting of the wheat from
the chaff, I would like to discuss two examples in some detail and point out
the emerging intermediate view.
IV. SPEECH AND ACTS: THE CASE OF RATIONALITY
It is one of the scandals of recent philosophy of language that the philoso-

phical theory of speech-acts has grown up independently of the philosophical
theory of action. The intra-active view, which sequesters language from all
the rest of the mind, is probably the implicit assumption underlying that state
of the art. My argument against it rests on the role played by principles of
rationality in linguistic activity.
When our pragmatical competence is under consideration, rather than our
actual performance, we may assume that a weak, doxastic version of the
principle of effective means is followed, namely:
(Re) Given a desired literal purpose, the ideal speaker opts for a
linguistic action which, to the best of his belief, attains that
purpose most effectively and at least cost, ceteris paribus.
The latter rider is important because it enables us to explain, for example, the
seeming lack of terseness in ordinary speech. Given the amounts of, say, time
and energy that a speaker is willing to spend on deliberation, he is supposed
to choose what he believes to be the best speech-act that he is able to perform
for attaining the literal purpose he then entertains, provided that the choice
is carried out without indulging in a process of deliberation which requires
more than the given amounts of time and energy. We are not always terse,
because usually we do not spend time on trying to find better ways of ex-
pressing what we have to say, though such ways exist probably most of the
times.
The usual version of the principle of effective means:
(R) Given a desired end - a literal purpose - the ideal speaker
chooses that linguistic action which most effectively and at
least cost attains that purpose, ceteris paribus,
is indeed too strong. The latter principle and (Re) introduce two standards
for linguistic action, marking well the distinction between constitutive rules
and strategies, that is, between playing by the rules and excellent playing.
The most interesting application of principles of rationality to linguistic
activity is Grice's theory of implicatures. 26 It is my contention that all theo-
48 ASA KASHER
retical achievements of the theory of conversational implicatures are derivable

from appropriate applications of principles of rationality to standard cases
of linguistic activity.
Grice has put forward a principle of cooperation and several derived
super-maxims and maxims. The former principle is that of making conversa-
tional contributions such as required at the stages at which they occur, by the
accepted purposes or directions of the talk exchanges in which the conversa-
tional contributors are engaged. I argued elsewhere27 that Grice's principle
is too strong and proposed replacing it by (Re) or a similar rationality
principle. Grice's super-maxims are all linguistic applications of consequences
drawn from the principle of effective means. An obvious example is the con-
sequence that means should not be used for achieving a given desired end
less or more than is required for achieving it, ceteris paribus. Grice's super-
maxim of quantity ("do not make your conversational contribution less or
more informative than is required") is evidently an application of this 'ultra-
maxim' of rationality, where information is taken to be a measure of
linguistic means.
A second ultra-maxim of rationality commends trying to achieve given
desired ends by standard employment of the available means, ceteris paribus.
Among the consequences of this ultra-maxim are Grice's super-maxim of
quality ("try to make your conversational contribution one that is true"),
Searle's maxim - "speak idiomatically unless there is some special reason
not to,"28 and also style controlling principles. For the enumeration of the
latter, a saying by Blanshard might serve as a motto: "Style is the feather in
the arrow, not the feather in the hat."29 Notice, however, that this ultra-
maxim lends no support to what has been rightly called "cobwebs of
grammarian's fetishes," such as the supposition that one should not begin
a sentence with "and" or "but," the insistence on writing "first" instead of
"firstly," or what the novelist Jean Stafford is told to put on a sign placed
over her back door: " 'Hopefully' must not be misused on these premises.
Violators will be humiliated."30
A third ultra-maxim is as follows: At every stage in your pursuit of your
desired ends, consider the means used concurrently by others, and determine
the manner of using your means accordingly; moreover, prefer using your
means in a manner which you believe is likely to help other persons in their
pursuit of their desired ends over any other way of using your means, ceteris
paribus. Without attempting a derivation of this principle here, I would offer
it as an explication of what might be found at the foundations of Grice's
suggestive but vague super-maxim of relation, commending "relevance."

Finally, Grice's super-maxim of manner ("Be perspicuous") is derivable
from a general ultra-maxim of rationality which commends giving preference
to use of available means which leads one directly to his desired ends, over
such uses of those means which lead him naturally to situations wherein
achievement of his desired ends is just a possible, non-standard result,
ceteris paribus.
The replacement of Grice's principle of cooperation and his four super-
maxims by a linguistic application of a principle of effective means and some
derived ultra-maxims, respectively, leaves the generative power of Grice's
ingenious theory of implicatures intact. Moreover, the presently proposed
theory has some advantages over Grice's one, but they will not be discussed
here)1
I take it that the indicated role of rationality principles in explaining
linguistic behaviour provides sufficient evidence against the intra-active view,
simply because such rationality principles are by no means practised in
linguistic behaviour solely.
I would like to mention in passing that some fundamental traits of natural
language can be explained on grounds of rationality in a broad sense that is
applicable to the way in which devices or institutions are organized. Putnam's
principle of the division of linguistic labour is an example of such a trait. The
hypothesis that all sentences of all colloquial kinds are indexical is another
case. 32 I do not have space here for more than providing another motto,
which is what Mr. Warnock tells us Grice once said, when they had been
looking at some parts ofthe vocabulary of perception: "How clever language
is! "33
v. USE AND MEANING: THE CASE OF IMPERFECT INFORMATION
Turning now to a refutation of the interactive view, according to which there

are only general principles of use that are applied to the independently defined
systems of syntactic and semantic rules, I would like to point out particular
rules of use, which are not derivable from any concoction of general principles
and semantic or syntactic rules.
Consider the ambiguity of sentences such as "Michael wants to meet a
spy," which are related to the philosophical referential/attributive distinction
and the linguistic specific/non-specific dichotomy. A promising way to
explain this ambiguity is to accomodate the powerful framework of Hin-
50 ASA KASHER
tikka's language-games with rules that would apply to indefinite expressions

under all their readings.3 4
I take these language-games to be played by the speaker, who utters the
sentence under consideration in an appropriate context, and a hearer. The
former tries to provide support for what he said, while the latter tries to
subject it to extreme criticism, so as to see whether there is any flaw in the
speaker's words that should stop the hearer from accepting them. Under this
interpretation, which should not be confused with the dialogical games of
Lorenzen,35 the rules of Hintikka's language-games are constitutive rules of
the linguistic institution of backing up and criticizing, which is closely related
to the linguistic institution of assertion.3 6
Now, all language-games for classical connectives, quantifiers and opera-
tors are, in game-theoretical terms, of perfect information. All the choices of
subformulae, individuals and possible worlds are done openly and every
player knows exactly what is given in any turn.
It seems that in order to treat adequately all readings of indefinite expres-
sions it is necessary to waive the condition of perfect information. Individuals
are selected and given new proper names, but these operations are not per-
formed openly, and only partial information is shared by all participants.
For example, when the reading of the sentence "Michael wants to meet a
spy" under consideration is that according to which the speaker has in mind
a certain person he knows to be a spy and of whom he asserts that Michael
wants to meet him, the speaker will move in the game by first introducing a
new proper name for that spy, say - Gamma, and then go on playing with
what corresponds to the sentence "Of Gamma it is true that Michael wants
to meet him." The hearer goes on playing with the same sentence and fOl-
mula, without having been informed about the identity of Gamma. What he
is told is just that Gamma is a spy, familiar in a way to the speaker.37 Other
readings are similarly treated.
The gist of this example is that the use of some particular expressions is
governed by special rules which involve partial descriptions of individuals.
These rules are neither applications of general principles of use, nor are they
reducible to purely semantic rules. It seems, therefore, that the case against
the interactive view can rest here.
The need to consider imperfect information arises with respect to addi-
tionallinguistic phenomena, such as nominalizations, comparatives, posses-
sives, some time indicators and perhaps also natural kind terms. The details
will be discussed elsewhere.
The present proposal carries an important observation with respect to

some meaning theories and to meaning and use theories. The truth-conditions
of the sentence "Of Gamma it is true that Michael wants to meet him"
depend on the logical status of "Gamma." For the speaker this is a fully
interpreted individual term, while for the hearer it is only partly interpreted
through a partial description. Put differently, the speaker's specified inter-
pretation includes a clause of the form 'Gamma = ... ,' where the right
hand of the equation specifies an individual in the universe of the discourse,
whereas the hearer's specified interpretation includes a clause of the form
'Gamma ESP,' where 'SP' denotes the class of spies in the universe of
discourse, known in a way to the speaker. The latter form is interesting in
two respects: it involves a semantic extension of the concept of interpreta-
tion, since it does not provide tor the constant "Gamma" an equation but a
membership statement, and it also involves a pragmatical extension of the
concept of interpretation, since it involves a speaker-dependent class of
elements of the universe of discourse. The nature of the latter extension and
the disparity of the speaker's and hearer's forms of interpretation, both cast
doubt on the possibility of carrying out any project of "pure" semantics.
VI. CONCLUDING REMARKS
The refutation of both the interactive and intra-active views results in the
emergence of an intermediate view. The pragmatic competence, i.e., the
competence to use linguistic means for effecting literal purposes consists of a
limited number of institutions, each comprising its constitutive rules of
various kinds. Some of the rules, such as those derived from certain prin-
ciples of rationality, are very general and shared by many institutions,
languages and activities. Some rules are more restricted, assigning particular
institutional roles, and some are even more limited in scope, being confined
to certain expressions of one natural language.
Notice that according to the present view it would be pointless to confine
the study of use to the vocabulary rather than to sentences and phrases, pace
Ryle and Alston.3 R Institutions govern uses of sentences and perhaps 'sub-
institutions' control the functioning of kinds of expressions, such as referen-
tial and predicative phrases.3 9
Indeed, the distinctive nature of the linguistic cluster of institutions is still
in the dark, if existent.
Nothing I have proposed here has been offered as a resting point. Are there
52 ASA KASHER
philosophical resting points at all? The going concern has been in redirecting
and unifying the study of language use. And if a theory of meaning for a
natural language has to give account of "how its speakers ... do whatever
may be done by the utterance of one or more sentences of the language,"40
then some light might be shed on additional regions of language study. It
was the poet Alexander Pope who expressed, perhaps unthinkingly, the
classical goal of linguistic theory, saying: " 'Tis not enough no harshness
gives offence; The sound must seem an echo to the sense."41 I plead for an
institutional amendment: " 'Tis not enough no harshness gives abuse; The
sound must be an echo to the use."
Tel-Aviv University
NOTES
1 Bar-Hillel (1954; 1970: 78).

2 Chomsky (1969: 81).
3 Bar-Hillel (1971).
4 Bar-Hillel (1974: 365).
5 Carnap (1959: 9).
6 In my (1975a) I tried to explain Montague's misguided linguistic elforts. The major
difference between generative and many philosophical studies oflanguage on the one hand
and Montague's and some other philosophical works on the other hand, culminates in the
contrast between Chomsky's contention that linguistics is a branch of cognitive psychology
and Montague's approach according to which "syntax, semantics, and pragmatics of
natural languages are branches of mathematics ... " (Thomason (1974: 2.
Still, the prospects for a linguistic theory which shares its goals with common theories
of human competence and its formal framework with common logical theories are not
absolutely dim.
7 In earlier formulations I used the expression "basic purpose" for what I call here
"literal purpose." I have found the former misleading to a certain extent.
8 See Kasher and Lappin (1976) for an elaborate defence. Our use of the expression
"pragmatic competence" is on a methodological par with Cresswell's use of "semantic
competence" in his (1976).
9 Quoted by Wittgenstein (1958: 38).
10 Davidson (1965) and Chomsky (1965: 222, n. 2 and references).
11 See Cresswell (1973: 111-115).
12 Davidson (1969).
13 See Hempel (1965: 137 If.).
14 Searle (1975a: 350-354).
IS Hempel (1965: 146 If.). Searle's (1975a) is an example.
16 See Chomsky (1965: 27-30), for some explanations.
WHA T IS A THEORY OF USE? 53
17 Searle (1975a: 346 f.) and Searle (1975c).

18 Dumrnett (1975).
19 Davidson (1973: 320-325).
20 Searle (1969: 60-67).
21 See Stalnaker (1973), for the related concept of presupposition.
22 Putnam (1975: 227-229) and Kripke (1972).
23 Searle (1975c).
24 Kasher (1974).
2S See Beardsley (1958: 369-378) and Hintikka (1975: 223-251).
26 Grice (1967).
27 Kasher (1976).
28 Searle (1975b: 76 f.).
29 Blanshard (1954; 1967: 50).
30 The quoted expression and the following two examples are Sir Ernest Gower's; all of
them and the latter quotation as well appear in a review by Joseph Epstein, T.L.S.
(February 13, 1976), 161-162.
31 In Kasher (1975c) I mentioned two examples. First, in case an implicature is created
by a seeming flouting of a maxim, Grice's theory does not explain why the implicit way of
conveying opinions by means of implicatures has been preferred by the speaker to an
expJi cit way of conveying his po int by means of frank assertions. Secondly, the role of the
so-called "silence-acts" in speech can be explained by using the linguistic version of the
principle of effective means, but not by the principle of cooperation.
32 I defended this point in my thesis, Hebrew University of Jerusalem, 1970. Henry Hit
pointed out to me that the idea of background information which is 'active' but not explicit
is also of the same nature.
33 Warnock (1973: 79).
34 For the general framework, see Hintikka (1973).
3S See Hintikka (1973: 80 ff.).
36 After writing this paper I read Dummett (1976) which is closely related to this idea.
37 This rider is discussed in Kasher and Gabbay (1976), following a suggestion and
arguments of Strawson. The whole conception is developed in this paper and also in
Kasher (1976), in a different context. Intuitionistic ideas are essentially involved.
38 Ryle (1953) and Alston (1967). See also Cohen (1955).
39 See Searle (1969: Chs. 4 and 5). I take reference and predication to be constituted by
sub-institutions, without committing myself to the view that there are speech-acts of
reference and predication.
40 Dummett (1975: 99).
41 In his Essay on Criticilm.
BIBLIOGRAPHY
Alston, William P., 1967, "Emotive Meaning," in: The Encyclopedia of Philosophy, Paul
Edwards (ed.), New York, Macmillan, 486-493.
Bar-Hillel, Yehoshua, 1954, "Indexical Expressions," Mind 63: 359-379; reprinted in
Bar-Hillel (1970), ch. 5.
54 ASA KASHER
Bar-H:lIel, Yehoshua, 1970, Aspects of Language, Jerusalem, Magnes.

Bar-Hillel, Yehoshua, 1971, "Out of the Pragmatic Wastebasket," Linguistic Inquiry 2:
401-407.
Bar-Hillel, Yehoshua, 1974, "General Discussion," in: Semantics and Communication,
Carl H. Heidrich (ed.), Amsterdam, North-Holland, 347-367.
Beardsley, Monroe C., 1958, Aesthetics, New York, Harcourt, Brace & World.
Blanshard, Brand, 1954, On Philosophical Style, Manchester University Press; reprinted
Bloomington and London, Indiana University Press, 1967.
Carnap, Rudolf, 1959, Introduction to Semantics, Cambridge, Mass., Harvard University
Press.
Chomsky, Noam, 1965, Aspects of the Theory of Syntax, Cambridge, Mass., MIT Press.
Chomsky, Noam, 1969, "Form and Meaning in Natural Language," in: Communication,
John D. Roslansky (ed.), Amsterdam, North-Holland, 63-85.
Cohen, L. Jonathan, 1955, "The Use of 'The Use of,'" Philosophy 30: 7-14.
Cresswell, M.J., 1973, Logics and Languages, London, Methuen.
Cresswell, M.J., 1976, "Semantic Competence," MS, Victoria University of Wellington.
Davidson, Donald, 1965, "Theories of Meaning and Learnable Languages," in: Proceedings
of the 2nd International Congress for Logic, Methodology and Philosophy of Science,
Yehoshua Bar-Hillel (ed.), Amsterdam, North-Holland, 383-394.
Davidson, Donald, 1969, "True to the Facts," Journal of Philosophy 66: 748-764.
Davidson, Donald, 1973, "Radical Interpretation," Dialectica 27: 313-328.
Dummett, M.A.E., 1975, "What is a Theory of Meaning? (I)," in: Mind and Language,
Samuel Guttenplan (ed.), Oxford, Clarendon Press, 97-138.
Dummett, M.A.E., 1976, "What is a Theory of Meaning? (II)," in: Truth and Meaning,
Gareth Evans and John McDowell (eds.), Oxford, Clarendon Press, 67-137.
Grice, H.P., 1967, "Logic and Conversation," partly published in: The Logic of Grammar,
Donald Davidson and Gilbert Harman (eds.), Encino & Belmont, Cal., Dickenson,
64-75.
Hempel, Carl G., 1965, "Fundamentals of Taxonomy," in his Aspects of Scientific Ex-
planation, New York, The Free Press, 137-154.
Hintikka, Jaakko, 1973, Logic, Language-Games and Information, Oxford, Clarendon
Press.
Hintikka, Jaakko, 1975, "Concept as Vision: On the Problem of Representation in Modern
Art and in Modern Philosophy," in his The Intentions ofIntentionality and Other Models
for Modalities, Dordrecht, Reidel, 223-251.
Kasher, Asa, 1974, "Mood Implicatures: A Logical Way of Doing Generative Pragmatics,"
Theoretical Linguistics 1: 6-38.
Kasher, Asa, 1975a, "The Proper Treatment of Montague Grammars in Natural Logic
and Linguistics," Theoretical Linguistics 2: 133-145.
Kasher, Asa, 1975b, "Pragmaticai Representations and Language Games: Beyond Ex-
tensions and Intensions," in: Rudolf Carnap, Logical Empiricist, Jaakko Hintikka (ed.),
Dordrecht, Reidel, 271-292.
Kasher, Asa, 1975c, "What are Pragmatical Representations?" in: Basic Problems in
Methodology and Linguistics. Proceedings of the 5th International Congress on Logic,
Methodology and Philosophy of Science, Part III, C. Butts and Jaakko Hintikka (eds.),
Dordrecht, Reidel, 1977, 225-242.
Kasher, Asa, 1976, "Conversational Maxims and Rationality," in: Language in Focus:
Foundations, Methods and Systems, Asa Kasher (ed.), Dordrecht, Reidel, 197-216.
Kasher, Asa and Gabbay, Dov M., 1976, "On the Semantics and the Pragmatics of
Specific and Non-Specific Indefinite Expressions, I," Theoretical Linguistics 3: 145-190.
Kasher, Asa and Lappin, Shalom, 1976, Introduction to Philosophical Linguistics, Kronberg/
Ts, Scriptor Verlag.
Kripke, Saul, 1972, "Naming and Necessity," in: Semantics of Natural Language, Donald
Davidson and Gilbert Harman (eds.), Dordrecht, Reidel, 253-355 and 763-769.
Putnam, Hilary, 1975, "The Meaning of 'Meaning,'" in his Mind, Language and Reality
(Philosophical Papers, Volume 2), Cambridge University Press, 215-271.
Ryle, Gilbert, 1953, "Use and Usage," Philosophical Review 62; reprinted in: Philosophy
and Linguistics, Colin Lyas (ed.), London, Macmillan, 45-53.
Searle, John R., 1969, Speech Acts, Cambridge University Press.
Searle, John R., 1975a, "A Taxonomy of IIIocutionary Acts," in: Language, Mind and
Knowledge, Keith Gunderson (ed.), Minneapolis, University of Minnesota Press,
344-369.
Searle, John R., 1975b, "Indirect Speech Acts," in: Speech Acts, Peter Cole and Jerry L.
Morgan (eds.), New York, Academic Press, 59-82.
Searle, John R., 1975c, "Meaning, Communication and Representation," MS, University
of California at Berkeley.
Stalnaker, Robert C., 1973, "Presuppositions," Journal of Philosophical Logic 3: 447-457.
Thomason, Richmond, 1974, "Introduction," Formal Philosophy (Selected Papers of
Richard Montague), New Haven and London, Yale University Press, 1-69.
Warnock, G.J., 1973, "Saturday Mornings," in: Essays on J.L. Austin, Sir Isaiah Berlin
et al., Oxford University Press, 31-45.
Wittgenstein, Ludwig, 1958, Philosophical Investigations, G.E.M. Anscomb (trans.), 2nd
ed., Oxford, Blackwell.
JAAKKO HINTIKKA AND LAURI CARLSON
CONDITIONALS, GENERIC QUANTIFIERS, AND OTHER

APPLICATIONS OF SUBGAMES
In examining the interrelations of use and meaning, one of the most promising
testing grounds is constituted by the theory of conditional sentences in
natural languages. On this ground the differences between different ap-
proaches to meaning and those between the several uses of "use" have
clashed dramatically, and yet left many of the principal problems un-
resolved. The truth-functIonal analysis of "if-then" sentences is as interesting
an example of an approach to meaning by means of recursive truth-
characterizations as one can hope to find. Yet it has run into a sharp criticism
from those philosophers of language whose paradigm of meaning-giving use
is usage, i.e., intralinguistic use. These philosophers are sometimes mislead-
ingly called ordinary-language philosophers. However, they have likewise
failed to solve many of the most interesting questions concerning the actual
behavior of conditionals in natural languages. The initial problems we shall
be dealing with in this work are cases in point. Hence the field is wide open
for new approaches.
In order to avoid misunderstandings, it is important to realize that there
are reasons of two different kinds why the truth-functional treatment of
natural-language conditionals is inadequate. Here we shall confine our atten-
tion to those problems that are caused by the conditional character of if-then
sentences. This is not what has primarily occupied most philosophers of
language, however, when they have been considering conditionals. What has
caught their fancy is usually the stronger logical tie that binds the antecedent
and the consequent of a natural-language conditional as compared with
purely truth-functional conditionals. This extra force is seen in problems
about counterfactuals, paradoxes of "material" implication, and so on.
This extra force of natural-language conditionals is a much less subtle
problem than the conditional character of if-then sentences in, say, English.
A suitable modal analysis of conditionals goes a long way toward solving the
problems of extra force. Furthermore, these problems are also amenable to a
treatment in terms of conversational forces. However, they will not be treated
57
58 JAAKKO HINTIKKA AND LAURI CARLSON
in the present essay, which focuses exclusively on the conditional nature of

conditionals. The phenomena caused by this nature are quite different from
the problems of extra (non-truth-functional) force of natural-language con-
ditionals. They are considerably subtler than these, and cannot be treated in
the same way.
How can they be treated, then? One promising approach - not necessarily
a completely new one, though - can be reached by taking the use that figures
in the fam,ous identification of meaning and use to be, not usage or intra-
linguistic use (use in language) but use in the sense of those rule-governed
activities ("language-games") which link a language, or a part of it, to the
world it enables us to describe or to interact with. Some such language-games
happily turn out to be games also in the strict sense of the mathematical
theory of games. Some of these have been studied by the authors of this
essay and by their associates in what they have called "game-theoretical
semantics." In this work, game-theoretical semantics is brought to bear on a
number of apparently unrelated semantical (linguistic, logical) phenomena.
They include certain "generic" (i.e., universal-quantifier) uses of words like
"a(n)" and "some," certain difficult types of pronominalization, and the
semantics of conditionals. They all turn out to be closely related to each other
and to admit of very natural explanations which all involve the same basic
idea. The most prominent one of these problems is precisely the semantics of
conditionals just mentioned.
For the fundamentals of game-theoretical semantics, the reader is referred
to Hintikka (1974), (1975) and (1976). The main idea underlying our se-
mantical games may be said to be to consider each such game as an attempted
verification of a sentence S against the schemes of a malevolent Nature who
is trying to defeat me. These games are thus games against Nature. Their two
players will be called "myself" and "Nature." The former wins if the game
ends with a true atomic sentence; the latter wins if it ends with a false one.
The rules of these games can be gathered without much difficulty from what
has been said. For instance, to verify a sentence of the form
X - some Y who Z - W
(where the "who" in "who Z" is for simplicity assumed to occupy the subject
position and to be singular) I clearly will have to choose a person, say b, such
that in the rest of the game I can verify
X - b - W, b is a(n) Y, and b Z.
CONDITIONALS, GENERIC QUANTIFIERS AND SUBGAMES 59
The game rule for "some" whose special case this recipe is will be called
(G. some).
In order to verify the sentence
X-everyY who Z-W
(with the same proviso concerning "who Z") I will have to verify
x- d - W if d is a(n) Y and d Z
for any individual d Nature might choose. This can be generalized into a
game rule (G. every) for the English quantifier word "every."
To verify "SI or S2" I will have to verify SI or verify S2, i.e., choose one
of them for the rest of the game to deal with, and to verify "SI and S2" Iwill
have to verify which ever conjunct Nature chooses. (Special attention will
have to be paid here to anaphoric relations between SI and S2')
These examples will suffice to illustrate how our games are played. The
rule for truth and falsity embodies an improved version of the old idea
that a sentece S is true if it can, in principle, be verified. This is now taken
to mean that S is true iff I have a winning strategy in the correlated game
G(S), false iff Nature has a winning strategy in G(S). If G(S) is indeterminate
(if neither player has a winning strategy), S is neither true nor false.
As a starting-point, let us recall the obvious simple-minded game-theoretic
treatment of if-conditionals. In earlier game-theoretical treatments, they
were handled by means of the following rule:
(G. if) When the game has reached a sentence of one of the forms
If X, Y
or
YifX
then I may choose either neg+[X] or Y, and the game is
continued with respect to it.
Here'neg+' refers to the process of forming the (semantical) negation
(contradictory) of a given sentence. Its analysis presents a separate problem.
This problem is not the reason why (G. if) is not wholly satisfactory. The
rules for negation will have to be discussed in game-theoretic semantics
anyway (cf. Hintikka, forthcoming). Indeed, the rule (G. if) is in many
respects a good first approximation. For instance, it enables us to discuss
the important principles that govern the order in which the game rules are
applied (cf. Hintikka, 1975).
The problem with (G. if) is connected with its purely truth-functional
character. As was already indicated, ordinary-language philosophers have
time and again claimed that a conditional like
(1) If X, Y
is not equivalent with the disjunction
(2) neg+fX] or Y.
In asserting the conditional (1) one does not assert the disjunction (2). One
somewhow makes, rather, a purely conditional assertion whose force does
not come to the play at all until its antecedent clause X is verified or other-
wise asserted. However, these ordinary-language philosophers also have
totally failed to spell out the precise logical and seman tical difference between
(1) and (2).
If anything, the game-theoretic approach encourages an emphasis on the
differences between (1) and (2). One reason for what happens in the case is
that a semantical game G(X) connected with X turns out to be indeterminate.
(It is trivially true that both players cannot have a winning strategy in one of
our semantical games. However, there is no general guarantee that either of
them must have one.) If the game is indeterminate, the associated logic is a
non-classical three-valued one. It is welI known that in such a three-valued
logic it is unnatural to define (1) as (2). Hence the possibility of indeterminacy
makes the putative game rule (G. if) unnatural, for it has precisely the force
of assimilating (1) to (2).
The purely truth-functional character of (G. if) is also seen from the fact
that this rule is virtually identical with the game rule (G.::J) for material
implication in the semantical games connected with formal first-order
languages. In fact, in the sequel we shall treat (G. if) and (G.::J) as being
essentialIy identical. The inadequacies of (G. if) as an explication of the
semantics of natural-language conditionals are to some extent paralleled by
the criticism presented by certain philosophers of mathematics and of logic
who prefer non-classical logic to the classical one.
It is also clear that rules like (G. if) do not do justice to the way in which
speakers process a conditional sentence like (1) semantically. In some sense,
we process (1) by first processing X and only then - depending on the out-
come of the first stage - processing Y. In so far as our rules of semantical
games are supposed to approximate the way in which we actually deal with
CONDITIONALS, GENERIC QUANTIFIERS AND SUB GAMES 61
ordinary-language sentences semantically - in so far as they capture the

dynamics of natural-language semantics, we might say - in so far they fail
to deal with (1) in a realistic fashion.
How can our game rule for "if" be improved on? In order to answer this
question, let us go back to the idea of(1) as a genuine conditional which only
becomes operative if and when its antecedent X is found to be true. Then,
and only then, must I be able to show that Y is true, too.
Now the different attempted ways of verifying X can be understood as the
different strategies available to me in the associated game G(X), and the
attempted ways of verifying Yare my strategies in G(Y). For, as it was
already indicated above, the basic crude but useful intuitive interpretation of
our semantical games G(Z) is that they are my attempts to verify Z against
the schemes of a malicious Nature. What I undertake to do in defending (1)
is therefore naturally understood to be to correlate to each winning strategy
of mine in G (X) a winning strategy f/J of mine in G(Y). The force of
(1) itself is to assert the existence of a functional f/J which carries out this
correlation. This simple idea captures very well the notion of conditionaliza-
tion, and we shall argue that it leads us to an eminently natural game rule for
if-sentences.
First, however, we have to develop it a little further. One trouble is that
the concept of a winning strategy is not an absolute one, but relative to an
opponent's strategy set. This makes it awkward to speak of a mapping of
my winning strategies in G(X) on my winning strategies in G(Y). The natural
thing is to consider mappings of all my strategies in G(X) into all my
strategies in G(Y).
The natural way to realize this idea through actual game rules is to have
the two players carry out a play of the game G(X) first, fought to the bitter
end. For it is part and parcel of the basic ideas of game theory that to play
a game is to choose a strategy. To playa game to the end is thus naturally
interpreted as divulging one's strategy in it. Since these strategies are (or are
represented by) functions, the strategies of the two players at the later stages
of the game will have functions as their arguments and hence be higher-order
functions (functionals). For their moves at these later stages will depend on
their already divulged strategies in a completed subgame. Thus it is the idea
of a concluded subgame that leads us to consider strategies representable by
functionals rather than (first-order) functions.
In order to force myself to consider all of my ~trategies in G(X), as re-
quired for the purpose of speaking of a function on the set of all such
strategies, the two players must exchange roles in G(X): Nature chooses one
of the strategies that would ordinarily be mine, and vice versa. If I win in
G(X), I have in effect falsified X, and no need to consider Y arises. Hence we
might declare myself a winner in this case.
However, if Nature wins, she has verified X and hence forced myself to
consider Y. In this case, the players must move on to carry out a play of
G(Y). The fact that the game is continued only if one of "my" strategies, as
chosen by Nature, wins in G(X) is the precise technical counterpart of the
earlier crude and inaccurate idea that in a conditional "If X, Y" we are
dealing with a mapping of my winning strategies in G(X) into my winning
strategies in G(Y).
My strategy in G(Y) will now depend on the way in which X was verified,
i.e., on Nature's choice of "my" strategy in G(X). Nature, in contrast,
clearly does not enjoy any comparable privilege. The outcome of this play
will decide the outcome of the overall game G(If X, Y).
Thus the game rule for G(If X, Y) can be represented by means of the
following "flow chart."
I win =I win G (If X,Y)
my ,._/
strategy/,-
Iwin=
G (X) with roles reversed I win G (If .x,Y)
Nature's~
strategy ~-
- ~
my
strategy ~(7
_/
Nature
~ G(Y) with normal roles
'wins - - - - - -..
Nature's~
strategy 1'] -- ~
Nature wins =
Nature wins G(lf X,Y)
Hence "If X, Y" is true iff there is a functional ifJ and a function ~ such
that they win against any strategy of Nature's represented by the functions
Cand'YJ.
We shall call a game rule defined by the flow chart (G. cond l ). If this game
rule strikes the reader as being rather complicated, we would like to counter
by asking whether he really feels entitled to expect a simple rule in view of
all the complicated problems ("ifficulties") about natural-language condi-
tionals. Moreover, we doubt that (G.cond l ) is felt to be very complicated
when its precise import is appreciated.
But why are not both players asked to divulge their strategies in G(X)? In
other words, why does not Nature's strategy 'YJ in G(Y) depend on my
strategy ~ in G(X)? Why is ~ as it were forgotten in G(Y)? The answer is
implicit in the intuitive motivation given above for the game rule (G.cond l ).
It was intimated there that Y comes into play only when and after X has
been verified, and its role will hence naturally depend on the way in which X
turned out to be true. Now this way of turning out to be true is what Ccodi-
fies. In contrast, ~ represents merely a hypothetical attempt to falsify X.
Intuitively, we must therefore require that G(Y) should be played so as to
disregard ~. It may be recalled here that initially we tried to establish only a
mapping of my winning strategies in G(X) into my winning strategies in
G(Y).
Thus the point of the game rule (G.cond l ) is not really to add much to the
intuitive ideas it is based on. Rather, what (G.cond l ) does is to show how
the precise dependencies such as the roles of 'YJ and ~ in G(Y) serve as
objective counterparts to our intuitive ideas of conditionality. An even more
explicit way of spelling out the same basic idea would be to say that G(Y) is
played with full knowledge of Cbut in ignorance of ~.
The formulation of (G.cond l ) in terms of subgames implies that in an
important respect the new rule does not change the character of our seman-
tical games. Before replacing (G.::: by (G.cond l ), our game-theoretical
semantics could have been said to effect a translation of each first-order
sentence into a second-order sentence of the form
(3) (afl) (a12) (aIm) (Xl) (X2)'" (xn)F(fr, f2,"" fm, Xh
X2,' .. , Xn)
where flo 12, ... , fm are such Skolem functions as serve to define my
strategies in so far as quantifier rules are concerned, and (Xl), ... , (xn) are
all the universal quantifiers of the original sentence (assuming that all
negation-signs were first driven in so as to precede immediately atomic

formulas). Furthermore, in (3) F is the original sentence with quantifiers
omitted and with each existentially quantified variable y replaced by a
different term of the form I(x" xl' ... , x k ) where (x,), (xl)' ... , (x k ) are
all the universal quantifiers within the scope of which (ay) occurs in the
original sentence. The universal quantifiers (Xl), (X2) , ... , (xn) in effect
embody Nature's strategies as far as her quantificational moves are con-
cerned. The import of (3) thus comes very close to saying just that I have a
winning strategy in the game correlated with the original sentence.
The replacement of (G. = by (G.cond l ) has the effect of replacing some of
the function and individual variables 110 /Z, ... , Xb X2, ... by variables
for functionals, i.e., higher-type functions (or for functions in the case of
universal quantifiers), plus adding to their number. Since nested occurrences
of "=>" serve to push the types of these functionals higher and higher,
we must in principle be prepared to use functionals of any finite type.
We can in fact obtain a kind of formalized expression of (G.cond l ) by
hanging on each subordinate clause X and Y two argument places, indicating
respectively my strategy and Nature's in the correlated games. Then the game
rule (G.cond l ) corresponds to a translation rule which takes us from "If
X, Y" to
Our new game rule (G.cond l ) calls for a few comments. First, the idea it
incorporates is obviously related very closely to the ideas of the intuitionists.
According to them, a conditional asserts that there is a way of obtaining a
"proof" (verification) of the consequent from any given "proof" (verification)
of the antecedent. This is very closely related to what (4) says. For basically
what it asserts is just the existence of a functional rf> which takes us from a
successful strategy in verifying X to a successful strategy in verifying Y.
Secondly, it is worth noting how the subgame idea which led us to
(G.cond l ) helps us to capture some of the dynamics of one's natural se-
manti cal processing of a conditional sentence which was mentioned above.
Intuitively speaking, we first process the antecedent. This corresponds to the
complete playing off of the game G(X) correlated with the antecedent. (This
is what brings in subgames or, as we really ought to call them, completed or
closed subgames.) Only after we have made clear to ourselves what the world
would be like if the antecedent is true do we move on to consider what the
consequent says on this assumption. This second stage corresponds to play-
ing the game G(y), and its conditionality is reflected by the dependence of
my strategy (1)(C) in G(Y) on Nature's strategy C in G(X), played with
reversed roles.
This insight into the dynamics of first-order semantics will be put to use
later by considering the behavior of pronominalization in a context involving
subgames. Conversely, what we shall find about those types of pronomina-
lization will support the diagnosis we have built into the rule (G.cond 1).
At this point, a skeptical reader may very well wonder how much real
difference the replacement of (G.if) (or(G.~ by (G.cond 1) really makes.
There are in fact some prima facie reasons for skepticism here. It can easily
be seen that on purely classical assumptions, including prominently the
stipulation that all function variables (of any type) range over all functions
of the appropriate type, the interchange of (G. ~) and (G.cond 1) does not in
fact make any difference to the truth of the sentences of formal first-order
languages. For purely classically (i.e., if myself is declared the winner if I
win in G(X) with roles reversed) X ~ Y is true iff ,.., X or Y is true, i.e., iff
I have a winning strategy either in G(Y) (call it ~o) or else in G( ,..,X) (call
it Co). Then I can respectively put either ~= ~o or (identically) (1)(C) = Co
in (4). Conversely, suppose that there are ~ and (1) in (4) such as to guarantee
my win. Then either I have a winning strategy in G(,.., X) or else for each
winning strategy Cin G(X) there is A. such that I win in G(Y) by playing A.
against any strategy 1] of Nature's. But I can have as much as one such
strategy classically only if Y is true.
However, even though formally and classically speaking there is little to
choose between (G.if) (or (G. ~ and (G .cond 1), there are further possibilities
that might seem to serve to drive a wedge between the two. In fact there are
two entirely different openings for a distinction here.
(a) The game-theoretical viewpoint strongly suggests that we restrict the
strategy sets of the two players to computable functions and functionals.
More accurately, we can restrict the strategies represented in (4) by functions
and functionals to computable ones.
This modification immediately changes the whole situation. It does so
already in the otherwise classical first-order case. The set of true sentences
will be affected by the change.
More generally, we might be inclined to admit suitable nonstandard
models in the sense of Henkin (1950) (see also the correction by Peter
Andrews), that is to say, allow function quantifiers to range over suitable
subsets of all arbitrary functions of the appropriate type. The most liberal
66 JAAKKO HINTIKKA AND LA URI CARLSON
policy here is to require merely that these subsets be closed with respect to
Boolean operations and projective operations.
It turns out, however, as Laurence Nemirow first pointed out to us, that
after a restriction to computable functions and functionals has been carried
out, the distinction between (G.:::;) and (G.cond 1) does not make any dif-
ference. By modifying slightly the argument for the classical case, on this
restriction (G.:::;) and (G.cond 1) can be shown to be equivalent. This equi-
valence may perhaps be considered a partial reason for the relative success
of a purely truth-functional analysis of conditionals - and for the absence
of any viable alternative in the earlier literature.
It also shows that the main reasons for the greater naturalness of(G.cond l )
as compared with (G.:::;) have to be sought for elsewhere.
There is a major change, however, that can result from restrictions
imposed on strategy sets. Such a restriction may imply that neither player
has a winning strategy in some of the seman tical games. Then there will be a
difference between asserting that a sentence is true, i.e., that I have a winning
strategy in the correlated game, and asserting that it is not false, i.e., that
Nature does not have a winning strategy in it. This in turn generates a
certain ambiguity, as the sentence can be thought of as asserting either.
If a conditional like "If X, Y" is given the latter of these two inter-
pretations, its force will be that of
which is the same as that of
This is related very closely to the so-called no-counter-example interpretation.

(For it, see Mostowski, 1966, Ch. 5; Per Martin-Lof, 1970, p. 12.)
(b) In natural language, there are certain phenomena which become
explainable as soon as the rule (G.cond 1) (or some other rule which likewise
involves subgames) is adopted. In order to see what they are, let us consider
an example. What kind of anaphoric relation do we have in the following
simple conditional?
(5) If Bill owns a donkey, he beats it.
Here "it" cannot refer to any particular donkey referred to earlier, for taken
as a whole (5) does not speak any more of one of Bill's donkeys than of
another one of them. Hence we do not have here an instance of the usual
function of pronouns (pronominal anaphora), viz. to recover an earlier

reference to a particular individual. Nor does the "it" in (5) serve as a
so-called "pronoun of laziness," that is, merely as a placeholder for its
grammatical antecedent "a donkey," for (5) is not synonymous with
(6) If Bill owns a donkey, he beats a donkey.
Sometimes it is said, in view of these facts and of the intended meaning
of (4), that in (5) "a" has a "generic" function, i.e., serves as a universal
quantifier rather than as an existential one. (We shall not try to criticize here
this use of the term "generic," even though it is in certain respects a mis-
leading one.) Why "a" should be generic in sentences like (5) has not been
explained, however, even though such an explanation is made highly de-
sirable by the fact that in many contexts the indefinite article "a(n)" must be
construed as a genuinely existential quantifier.
Moreover, the explanatory force of a mere postulation of a new sense of
"a(n)" with a "generic" force is greatly reduced by the fact that the truly
universal quantifier "every" is not admissible in the same context, for we
cannot say
(7) *If Bill owns every donkey, he beats it.
The inadmissibility of (7) may have a partial explanation in terms of the
relative order of the game rules for "every" and "if." However, that this
explanation is not completely satisfactory is seen by turning (5) around. For
we can say
(8) Bill beats every donkey if he owns it.
Moreover,
(9) Bill beats a donkey if he owns it,
is perhaps a little less natural than (5). Moreover, in so far as (9) is acceptable,
it seems ambiguous between an existential-quantifier and a universal-
quantifier reading, again unlike (5). In fact, a slight change in the example
makes the existential-quantifier reading almost mandatory as, e.g., in
Bill will beat a donkey if he finds it.
Hence we have in our hands a problem both about the behavior of "a" in
(5) and (9) and about the behavior of "every" in (7) and (8), over and above
the question of the nature and conditions of pronominalization in all these
different sentences.
The more general problem we are facing here concerns the conditions on
which a quantifier phrase can be the antecedent of a singular pronoun. What
we have just seen suggests that a satisfactory answer cannot be given, e.g.,
in terms of definiteness, for presumably "every" is more definite than "a,"
and is equally definite in (7) and (8). (Here we have one more indication of
the unsystematic and frequently misleading character of linguists' concept of
definiteness.)
It is not surprising that more complicated versions of these examples,
such as
(10) If Bill owns a donkey that he likes, he beats it,
have caused not inconsiderable difficulties in Montague-type grammars.
Further examples similar to (5), (7)-(9) are easily found. Here is one
bunch:
(11) If a member contributes, he will be praised.
(12) *If every member contributes, he will be praised.
(13) A member will be praised if he contributes.
(14) Every member will be praised if he contributes.
Notice also that the conversion which takes us from (5) to (9) and from
(7) to (8) might very well be expected to preserve not only meaning but
acceptability. After all, all that happens in this conversion is the replacement
of a sentence of the form "If X, Y" by "Y if X," together with a reversal of
the relations of pronominalization between X and Y. It is hard to think of
an operation which prima facie would seem likelier to preserve meaning and
acceptability (including degree of acceptability). Yet we have seen that the
latter sometimes changes in the operation, and later we shall find an example
in which the preferred reading of the sentence in question is also affected.
All this requires an explanation.
In order to begin to solve these problems, let us consider first (5). How do
we get hold of the individual donkey that is supposed to be picked out in
some sense by "it"? This question is well-nigh impossible to answer as long
as we think of conditionals along the lines of semantically indivisible
wholes as in (G.if). However, the basic idea underlying (G.cond 1) at once
throws new light on the situation. This basic idea is that in the game con-
nected with the conditional (1) I have to correlate with each of my strategies
in G(X), say C, a similar strategy of mine in G(Y). This correlation is needed
in the game iff Cwins in the subgame G(X) (cf. our flowchart for G(IfX, Y) ).
What does such a strategy look like in (5)? Here X = "Bill owns a donkey."
Understanding "a" in the most straightforward way as an existential
quantifier, my winning strategies in
(15) G(Bill owns a donkey)
are simply the different choices of a donkey owned by Bill. Thus in the
antecedent of the conditional (5) we are as it were considering Bill's donkeys
one by one. And this is obviously just what semantically speaking gives us
the foothold for pronominalization in (5). After we have chosen to consider
some one winning strategy of mine in (15), i.e., to consider a donkey owned
by Bill, we can in the consequent (5) refer pronominally to that very donkey
and say something about it. And this isjust what happens in (5). It is precisely
the consideration of my several strategies in (15) that leads us to consider a
particular beast which in the consequent of (5) can be recovered by a
pronominal reference.
Thus we see how it is that the subgame idea serves to explain why certain
quantifier phrases can serve as pronominal antecedents. They represent
choices made in an earlier, already concluded subgame.
Several further observations can be made here which support our dia-
gnosis. First,let us note that what we just saw is in effect an explanation why
the indefinite article "a(n)" comes to have a "generic" (universal-quantifier)
sense in conditionals like (5). This explanation has the merit of turning on the
assumption that the basic force of "a" in (5) is that of an existential quantifier
(in the precise game-theoretical sense marking my move in our seman tical
games). It thus dispenses with all assumptions of irreducibly different senses
or uses of the indefinite article in English.
We must hasten to add that there are other generic uses of the indefinite
article "a" which are also explainable in this way - but not without a
great deal of further argument. A case in point is, e.g.,
(16) A cat loves comfort.
However, there is further evidence to support our diagnosis of cases like
(5). The only thing we assumed of "a" in (15) was that it expresses existential
quantification (i.e., marks my move). But so does "some." Hence, by the
same token, there ought to be a kind of generic sense to the sentence with
"some" instead of "a" otherwise analogous with (5), i.e., to
(17) If Bill owns some donkey, he beats it.
The acceptability of this sentence is not obvious, but in so far as it is ac-

ceptable, "some" in it clearly has a "generic" (prima facie universal) force,
just like (5). The acceptability of (17) may in fact be improved greatly by
changing it into
(18) If Bill owns some donkey or other, he beats it.
Here we have an interesting indication of the strength of our explanation
why "a" has a generic force in (5). If it is replaced there by "some," which
normally does not exhibit any predilection for a generic sense, it is likewise
forced to the role of a generic quantifier, albeit a little awkwardly.
The same point is strikingly confirmed by the fact that even the blatantly
existential "there is" assumes the force of a universal quantifier in the antece-
dents of conditionals. In order to see this, witness examples like the following.
If there is a donkey that Bill owns, he beats it.
(Whether this is completely grammatical does not affect my present point,
which pertains to the existential-universal contrast.)
Although we have not yet uncovered the mechanics of the conversion from
(5) to (9) or from (7) to (8), it is of interest to see that "some" follows here
roughly the same pattern as "a(n)." Applied to (17) the conversion yields
(19) Bill beats some donkey if he owns it.
It is not clear whether this is acceptable, but in so far as it is, its preferred
reading is clearly different from that of (17). In so far as (19) is acceptable, it
seems to allege Bill's animus against some particular beast. Hence in (19)
"some" seems to have the force of an initial existential quantifier, not that
of a universal quantifier. This point is even clearer if we change the example
to read
Bill will beat some donkey if he finds it.
The explanation we gave above for the possibility of pronominalization
in (5) serves to explain also why (7) is not acceptable. The idea was that
(G.cond 1) invited the players to consider my different strategies in G(X). In
so far as certain individuals are produced in a play of the game by such a
strategy, they can be referred to again pronominally. Now the individuals so
produced (selected for special attention) are the ones that an existential
quantifier prompts me to select. In contrast, my strategy does not specify
which individuals Nature perhaps chooses as the values of a universally
quantified variable. Hence our theory yields the prediction that only an
existential-quantifier phrase can serve as an antecedent of a singular pronoun
in the kind of pronominalization (i.e., from the antecedent of a conditional
to its consequent) we have in (5).
This prediction is confirmed on its negative side by the un acceptability of
(7). The acceptability of the analogous sentence
(20) If Bill owns any donkey, he beats it
causes no problems here in view of the well-established ordering principle
(O.any) which among other things gives the game rule (G.any) a priority over
the rule for "if" (see Hintikka, 1975).
Our predictions concerning the conditions of admissible pronominaliza-
tion are confirmed by many examples on the positive side, too. Perhaps the
most interesting ones are those conditionals whose antecedent contains an
existential quantifier within the scope of a universal quantifier. The following
example is due essentially to Lauri Karttunen.
(24) If you give every child a present for Christmas, some child will
open it the same day.
Here a winning strategy of mine for the antecedent assigns to every child a
present. Hence when "some child" in the consequent invites us to pick out
one, he or she comes already with an associated present, recoverable by the
pronoun "it" in the consequent of (24).
Further explanation is needed to account for the unacceptability of the
corresponding plural sentence
(24)' *If you give every child presents for Christmas, some child
will open it the same day.
The explanation does not lie in any requirement of uniqueness, for the
following is an acceptable sentence:
(24)' , If you give every child at least one present for Christmas,
some child will open it the same day.
The right explanation seems to lie in some sort of congruence requirement
between the pronoun and its antecedent. This requirement is not satisfied
in (24)' where the pronoun is singular but its antecedent is in the plural. In
contrast, the acceptability of (24)' , is predicted by our theory, and so is the
acceptability of the following sentence:
(24)' , , If you give every child presents for Christmas, some child
will open at least one of them the same day.
Notice that "them" does not refer here to the gifts given to different children,
but to those given to that child intended by "some."
As a word of warning, it must be pointed out that there does not seem to
be any hard-and-fast connection between the subgame idea and the direction
of pronominalization.
An interesting class of examples is generated by conditionals in which
propositional moves are made in the first subgame. (Our attention was
drawn to these examples by Lauri Karttunen.) They include the following.
(25) If Jane owns a car or John has a bicycle, it is in the garage.
(26) *If Jane owns a car and John has a bicycle, it is in the garage.
The strategy which is "remembered" by the two players in the second sub-
game (viz. the one connected with the consequent; cf. the "flow chart"
above) specifies in (25) a unique individual. For each one of my strategies in
G(Jane owns a car or John has a bicycle) specifies first the choice of a disjunct
and then the choice of an individual corresponding to the indefinite article
"a" which occurs in the chosen disjunct. Hence it is predicted by our theory
that the pronoun "it" in (25) is acceptable, as it obviously is, because there
is in the second subgame a unique individual present for it to refer to.
In contrast, in (26) each one of my strategies in the game connected with
the antecedent specifies a choice of an individual corresponding to the inde-
finite article in either disjunct. In brief, it specifies two individuals, not one,
wherefore there is no unique individual for the pronoun "it" to stand for.
Hence we can understand why (26) is unacceptable while the following is
acceptable:
(27) If Jane owns a car and John has a bicycle, they are in the
garage.
This application of our theory of subgames throws interesting light on
certain wider issues. Pronouns of the kind we are discussing are usually dealt
with in terms of a relation of grammatical antecedence. Given a pronoun
(anaphorically used), the main question has been: What is its head? By
considering (25) we will see some of the limitations of what can be done by
the sole means of this antecedence relation. In it, the pronoun "it" has two
different possible antecedents between which there is no way of deciding.
Any treatment which relies principally on the antecedence relation will there-
fore have to declare (25) ambiguous. Yet in some obvious sense it is not in
the least ambiguous.
Moreover, the difference between (25) and (26) cannot be accounted for
by considering only the antecedence relations in them, for they are analogous
in the two sentences.
In contrast to approaches to anaphora where the main weight is put on
antecedence relations, our treatment explains why (25) can be unambiguous
even though the singular pronoun "it" has in it two possible antecedents.
Likewise, we can readily explain the difference between (25) and (26). In both
cases, the explanation is essentially seman tical. An anaphoric use of a singular
pronoun presupposes the presence of a unique individual for which it can
stand. We have found that this kind of uniqueness can only be decided by
reference to the actual plays of a semantical game. It cannot be decided by
considering grammatical (syntactical) relations of antecedence only.
This explanation of the kind of use of pronouns we find in (25) and (26) is
confirmed by the observation that as soon as some of the possible anaphoric
relations are ruled out by collateral evidence, the two players' strategies will
be affected correspondingly. This means that both the interpretation and the
acceptability conditions for a conditional will change. This is illustrated by
the following examples.
(28) If John buys a house and Jane inherits some money, it will be
used to buy new furniture.
(29) *If John buys a house or Jane inherits some money, it will be
used to buy new furniture.
(Again, these examples were first suggested to us by Lauri Karttunen.)
Collateral information tells us (and the players) here that "it" cannot be
John's house. (A newly bought house cannot be used to buy furniture.) This
changes the conditions on which (28)-(29) are true (acceptable) as compared
with (25)-(26), explaining the asterisk in (29) and the lack of one in (28).
Although this explanation, when fully worked out, is a pragmatical rather
than semantical one, it is firmly grounded in our semantical theory, and
hence firmly supports it.
But why should a conversion from (5) to (9) make a difference here? An
answer is not hard to find. It leads however to an interesting generalization.
We have seen that the clause-by-clause semantic unfolding which is charac-
teristic of conditionals in natural language is captured by the subgame idea.
Now how is the order of the different subgames determined? A priori, this
order could be determined in many different ways. However, it is not difficult
to guess that ceteris paribus it proceeds from the left to the right (from the
earlier to the later in speech). This generalization we shall call the Progression
Principle. It is in keeping with our psycho-linguistic intuitions as to how the
understanding of a sentence actually proceeds. It is closely connected with
the linearization phenomena studied in text linguistics.
From the Progression Principle it follows that the game rule for "Y if X"
cannot be the same as the game rule (G.cond 1) for "If X, Y." For in
(G.cond 1) the subgame G(X) connected with X is played before the subgame
G(Y) connected with Y, and the latter subgame depends on the former. In
the case of "Y if X" this order is ruled out by the Progression Principle. In
its stead, we have the rule embodied in the following flow chart. We shall
call this rule (G.cond2).
(G. cond2)
Iwin=lwin G(VifX)
::;rat,g'7 Nature wins =

Nature wins
G(V) G(Vif xl
Nature's " "

strategy s -~ ~t~;~~;: ~ / -
Nature
wins - - - - - -..... G(Xl with reversed roles
I win = I win G (V if X 1
The translational counterpart to (G.cond 2) can be seen to be
This is different from (4). We therefore obtain the interesting result that
"If X, Y" and "Y if X" are not completely synonymous in English. The
difference is due to the dynamic left-to-right preference expressed by the
Progression Principle.
A comparison between (G.cond 1) and (G.cond2) may be instructive at this
point. It is easily seen from the flow charts that the intuitive situation is some-
what different with the two. In our first flow chart, my strategy in G(Y) was
seen to depend only on Cbut not on ~. It is easily seen that the corresponding
reasons are somewhat weaker in the case of (G.cond 2). In other words, there
may be some reasons for making my strategy (in Nature's original role) in
G(X) dependent on ~ and not only on C. Then the representation would be,
not (30) but
However, (30) clearly is still more natural than (30)'. Even so, this observa
tion serves to explain why such sentences as (9) and (13) are acceptable even
with a universal-quantifier reading. For what (30)' means is that in the game
G(X) both a strategy of Natunl's and a strategy of mine in G(Y) are as it
were known. Hence pronominal reference can recover also individuals
specified by the latter and not only these specified by the former. This is what
happens in (9) and (13) on their universal-quantifier reading, which seems to
be a viable one.
In the same way as in connection with (G.cond 1) it can be seen that on
classical assumptions the difference between (4) and (30) or (30)' is nil, and
that the simple non-classical ones differences are not any greater. However,
the difference in the order of the subgames G(X) and G(Y) in (G.cond 1) and
(G.cond 2) implies that the openings that there are for pronominalization
(pronominal anaphora) in "If X, Y" and in "Y if X" are entirely different.
Hence it is not surprising that the conditions of acceptability for the two types
of sentences are entirely different. This is illustrated forcefully by the contrast
between (7) and (8).
Moreover, the difference between (7) and (8) is predictable on the basis of
the game rules (G.cond 1) and (G.cond 2). In sentences of this kind, prono-
minalization happens "by strategy": the pronoun refers back to an individual
picked out in an earlier (and already concluded) subgame. In (5) and (7), this
individual must be picked out by a strategy of mine (chosen by Nature) in
G(X), as shown by (4). This is possible with (5) but not with (7). In contrast,
(25) shows that in (8) the individual in question must be picked out by a
strategy of Nature's in G(Y). Now in (8) Nature does choose an individual,
which must be a donkey if the game is to be continued beyond the subgame
G(Y). Hence the prediction our theory yields is that (8) is acceptable, as it
in fact is.
More generally, if there is just one (unnegated) quantifier in Y, it can
(ceteris paribus) be an antecedent of a pronoun in X (in "Y if X") if and only
if it is a universal one.
Moreover, differences in pronominalization between "If X, Y" and "Y if
X" may make a semantical difference. In fact (18) and (19) are not synony-
mous. (This observation is given an extra poignancy by the fact that (19) is
made relatively acceptable by our general ordering principles which favor
higher clauses over lower ones and also favor left-to-right order. Both factors
argue for a larger scope for "some" than for "if" in (19), which seems to be
what makes it relatively acceptable.)
It is in keeping with this that in the converse forms of our sample con-
ditionals, i.e., in (8) and (9), it is now the "indefinite" individuals introduced
by universal quantifiers that can naturally be re-introduced by pronouns.
Predictably, (8) is felt to be better formed and clearer in meaning than (9).
Moreover, (9) and (19) can be given some semblance of meaning not so much
by the kind of "pronominalization by strategy" we have been studying as by
assuming that the existential quantifier "a" or "some" has an exceptionally
wide scope comprising the whole conditional (9) or (19), respectively. The
reason why this effect is less marked in the case of (9) than in the case of (19)
is that in (9) the other generic uses of the indefinite article "a(n)" than those
we have explained are also operative.
Another fact that now can be explained is that mirror-image examples
dual to (24) are acceptable, i.e., examples in which existential and universal
quantifiers exchange roles over and above the reversal of the order of X and
Y. The following is a case in point:
Some man will seduce every girl if she does not watch out.
At the same time we obtain an explanation of the fact - it seems to us an
unmistakable fact - that (8) is perceptibly less natural than (5). The explana-
tion lies in the fact that the strategies which make pronominalization possible
in (5) are as many choices of donkeys owned by Bill. These are the individuals
(5) intuitively speaking is about. They are of course just the individuals whose
choice by Nature in G(X) leads us to play G(Y).
In contrast to this, the "right" choices in (8) are donkeys not beaten by
Bill. This accounts for the "contrapositive" feeling we seem to have in trying to
understand (8) and also for the intuitive unclarity as to whether (8) is "about"
donkeys beaten by Bill or about those not beaten by Bill or about those
owned by him - or about each and every donkey. It is as if we in (8) first
said something about all donkeys and then subsequently qualified it by ex-
cluding those donkeys not owned by Bill. It is amusing to see how neatly
this feeling matches what happens in a play of the game connected with (8).
(Here we can incidentally also see how elusive and unsystematic a notion
"aboutness" is.)
Along these lines we can hence solve all the problems concerning (5)-(10),
(18)-(19) and their ilk. These problems include the following:
(i) The possibility of pronominalization in sentences like (5).
(ii) The universal-quantifier sense of "a" or "some" in examples like
(5) and (18), respectively.
(iii) The asymmetry between existential and universal quantifiers vis-a-vis
the kind of pronominalization illustrated by (5).
(iv) The sweeping effects of the prima facie innocuous conversion of (5)
to (9), (7) to (8), or (18) to (19).
(v) The (small but unmistakable) difference in the degree of acceptability
between (5) and (8).
(vi) The possibility of a universal-quantifier reading in sentences like (9)
and (13).
Our solution to these problems can be extended in several different direc-
tions, thus gaining further support. One such direction is the treatment of
other English particles that can be used in conditionalization. As an example,
we shall consider here the particle "unless." The extension is as straight-
forward as it is obvious. Sentences of the form 'Z unless Y' are treated
essentially in the same way as the sentences 'If neg+(Z), Y.' The difference
as compared with the plain "if" is that in the game rule for "unless" Nature's
strategies in G(Z) play the same role as my strategies in G(X) played in the
game rule for "if."
The relevant game rules - we shall call them (G.unlessl) and (G.unless2)-
appear from the following two diagrams (see pp. 78-79).
(G. unless!)
Iwin=
I win G (Z unless Y)
my
,Imlegy / -
c-/
f win=
G(Z) I win G (Z unlessY)
Nature's
strategy
~
S --~.
my
strategy ~ (~)
Nature
wins - - - -....~G(Y)
Nature's
strategyl'J -
Nature wins =
Nature wins G (Z unless Y)
(G. unlessl) is parallel to (G.cond2) and hence straightforward.

(G. unless2) is not parallel with (G.cond l), and hence may require an
explanation.
The leading idea on which (G.unless2) is based is that when I say,
Unless Y, Z,
what I have in mind is a dependence of the way in which Z fails to be true
depending on how Y turns out to be true. For instance, if! say,
Unless you give him a gift, he is unhappy
the intended way of avoiding his unhappiness depends on the way in which
the antecedent "you give him a gift" is made true. This dependence is what
(G.unless2) codifies.
I' win = I win G(unless Y, Z)
my
strategy~ 7 _/
I win =
G (Y) I win G (unless Y, Z)
Naturels~
strategy ~- ~
my
strategy 7 /
Nature wins -----tl..~ G (Z)
Na'",e'. " " -

strategy ~ (E) - ~
Nature wins =
Nature wins G(unlessY,Zl
The corresponding translations are
and, respectively,
If we check what these rules imply for our theory, we can see that they
preserve the roles of existential and universal quantifiers. Thus our explana-
tions will automatically cover the corresponding sentences with "unless,"
too. Examples show that this is precisely what happens. For comparison, we
repeat at the same time some of the earlier ones.
(5) If Bill owns a donkey, he beats it.
(31) Unless Bill likes a donkey, he beats it.
(8) Bill beats every donkey if he owns it.
(32) Bill beats every donkey unless he likes it.
(7) *If Bill owns every donkey, he beats it.

(33) *Unless Bill owns every donkey, he beats it.
(9) ?Bill beats a donkey if he owns it.
(34) ?Bill beats a donkey unless he owns it.
(17) If Bill owns some donkey or other, he beats it.
(35) Unless Bill likes some donkey or other, he beats it.
(19) Bill beats some donkey if he owns it.
(36) Bill beats some donkey unless he likes it.
Here the acceptability of the last six examples is not clear, and the precise
meaning of (9) and (34) is likewise problematic. What is absolutely clear,
however, is the parallelism between "if" and "unless." Notice in particular
that we have a very natural explanation here for the universal-quantifier force
of "a" in (31) and (34) and for the similar force (such as it is) of "some"
in (35).
Prima jacie, our theory does not square very well with the fact that the
presence of negation in the antecedent of a conditional does not reverse the
conditions of acceptability, as our explanation might seem to presuppose.
For instance, we can say
(37) If Bill doesn't like a donkey, he beats it
and perhaps also
(38) If Bill doesn't like some donkey or other, he beats it
but not
(39) *If Bill doesn't like every donkey, he beats it.
Again we can say
(40) Bill beats every donkey if he doesn't like it
with roughly the same meaning as (34), whereas
(41) Bill beats some donkey if he doesn't like it
is either unacceptable or else clearly non-synonymous with (37).

This all seems wrong, for negation changes my strategies into Nature's
and vice versa. Hence one might prima jacie expect (39) to be acceptable but
not (37). Yet the converse was just found to be the case.
It is nevertheless clear that some additional account of (37)-(41) will have
to be given in any case. For one thing, the antecedent of (37), viz.
(42) Bill doesn't like a donkey
has on one of its readings an entirely different force alone and in (37). Alone,
it says (on this particular reading) that Bill has no affection for anyone
donkey. Presumably its having a different role in (37) is what also makes
pronominalization possible there.
The explanation for these facts lies in the fact that the ordering principles
(scope conventions) governing the English indefinite article "a(n)" are
exceptionally fluid. This holds for instance for the relative order of the game
rules (G. an) and (G.not) (or (G.neg for "a(n)" and for negation, res-
pectively.1t also holds for the relative order of (G.an) and epistemic rules.
The latter fact is illustrated by the ubiquity of the de dicta-de re ambiguity.
(This ambiguity typically concerns just the relative order of a quantifier rule
like (G.an) and an epistemic rule.) The former fact is illustrated by the fact
that sentences like (42) have two readings, on which it has the logical
force of
(43) (ax) (x is a donkey A Bill does not like x)
or the force of
(44) ,..., (ax) (x is a donkey A Bill likes x)
It is the second of these two readings that was commented on briefly above.
This ambiguity of "a(n)" is one of the main sources of its universal-
quantifier uses.
What happens in problematical conditionals like (37) is that only one of
the two a priori possible rule orderings (in connection with the antecedent
of (37) ) enables us to interpret the pronominalization in (37). If the reading
adopted is (44), which in other circumstances is perhaps the preferred one,
it follows from our earlier arguments that pronominalization in (37) cannot
be given a reasonable semantical interpretation. On this reading, (37) will be
in the same boat with (7). No wonder, therefore, that this is not how (37) is
ordinarily understood.
However, if the other ordering is adopted (corresponding to the reading
(43) of the antecedent taken alone), (37) can be analyzed semantically just
like (6). The resulting reading has the same logical force as
(45) (x) (x is a donkey ::I (,..., Bill likes x ::I Bill beats x) ).
And this is in fact the force of (37) in English. Now we can see how it comes
about. The restraints on the seman tical interpretation of pronouns filter out
one of the two ways of processing the antecedent of (37). The remaining
order of the game rules yields (45)
In this particular case, the impossibility of the other, filtered-out reading
is also illustrated by the impossibility of expressing it in the usual logical
notation. In fact, it would have to be written out as something like the
following
(46) (ax) (x is a donkey A Bill does not like x) ::J Bill beats x
which is either ill-formed or has the last "x" dangling.
This line of thought receives further support from supplementary observa-
tions. One of them is that our treatment of (37) extends in a predictable way
to a large number of conditionals with an epistemic operator in their ante-
cedent. Consider, as an example, the following sentence
(47) If Bill believes that a thief has stolen one of his horses, Bill will
at once pursue him.
Here the "a" in "a thief" clearly has the force of a universal quantifier.
Moreover, the belief-context in (47) must clearly be understood de re, for
how else can we make sense of Bill's pursuing some one putative thief? (If
Bill merely opines as a purely existential judgement that someone or other
has stolen a horse, it is nonsense to suggest that Bill undertakes to pursue the
thief. For then there would not be any answer to the question: whom is he
pursuing?) Nevertheless the antecedent of (47) admits also a de dicto reading.
Why should the latter be filtered out in (47)? The answer lies in precisely the
same mechanism as served to explain the peculiarities of (37). Because of this
mechanism, only the de re reading of the antecedent of (47) makes it possible
to interpret the pronoun "it" in (47).
What was just said is not incompatible with saying that there is a reading
of (47) on which its antecedent has merely the force of an existential judge-
ment. From what has been said it follows that then the pronoun in the
consequent must be interpreted as a "pronoun of laziness." This reading
assigns to (47) roughly the same force as
(48) If Bill believes that a thief has stolen one of his horses, Bill will
at once pursue such a thief.
It is interesting to see that if one wants to paraphrase (47) by reversing the
CONDITIONALS,. GENERIC QUANTIFIERS AND SUBGAMES 83
order of the (logical) antecedent and consequent one will end up making the
de re character of the belief-construction blatant, over and above having to
switch from an existential into a universal quantifier:
(49) Bill will at once pursue every thief it Bill believes that he has
stolen one of Bill's horses.
Another apparent counter-example to OUI theory may be seen by compar-
ing the following two sentences.
(50) If some student did not flunk the test, he must have been
studying hard.
(51) *If not every student flunked the test, he must have been
studying hard.
Now my strategies in (50) and (51) are the same, except for a temporary
exchange of roles. This is reflected by the logical equivalence of (50) and (51).
Accordingly, it might be thought that any explanation why the anaphora
in (50) is a happy one which (like ours) turns .00 "pronominalization by
strategy" would yield a wrong prediction here. For it would apparently have
to predict that the anaphora in (51) is quite as happy as in (50). Yet (51) is
unacceptable. This is the same problem we were confronted by earlier when
we noted the unacceptability of (39).
A clue to an explanation of the unacceptability of (51) is seen from our
remarks above on the requirement of congruence between a pronoun and its
grammatical antecedent. These observations can be extended by requiring
that there be a coreferential antecedent for each pronoun in the first place.
This requirement is in some sense not satisfied by (51), for "every student"
there is not coreferential with "he" in (51).
It is not quite easy to see how this idea can be incorporated in our actual
treatment of sentences like (51). For the unanalyzed notion of coreference the
requirement just for~ulated relies on is not automatically available to us,
and in the actual game associated with (51) the individual whose name re-
places "every student" will in fact be also referred to by "he." So how can
we do justice to the observation which seemed to solve our problem?
It seems to us that the key to a solution of these problems lies in the need
of a non-classical game rule for negation. An explicitly negated sentence, say
neg+ [Xl, does not just describe a world in which X fails to be true. It first
describes a world in which X is true, and then says that this is not what the
world is like. In spite of being subsequently cancelled, the description of a
world in which X is true may open the door to pronominalization.

The game-theoretical counterpart to this idea is as follows. What happens
when G(neg+X) is true is that every one of my attempts to win in G(X) by
means of a strategy ; must give rise, through a constant functional <P, to a
winning strategy, = <PC;) of mine in G(neg+X).
In terms of the subgame idea this can be expressed as follows. When the
game has reached neg+X, play of G(X) is undertaken with roles reversed.
If I win it (playing what originally was Nature's role), I win G(neg+[X]). If
Nature wins (playing for myself, as it were), the game is continued. After
Nature has divulged her strategy ; in G(X), a new play of the same game
G(X) is undertaken, again with the roles of the two players reversed. Since I
now know Nature's original strategy;, my new strategy is a function <P(;)
of ;. In the new game Nature must again use ;. If! win this new subgame,
I win the whole game, and vice versa.
The flow chart that goes together with this game rule is the following.
I win = I win G (neg +X)
my -.._/
strategy / -
Iwin=
G (X) with roles reversed I win G (neg+X)
Nature's
strategy E-
Nature
. wins
------J..~ G (X) with roles reversed
Nature wins =
Nature wins G(neg+X)
The corresponding translation rule is

(52) (3:4(3:C)(~)(X(~, C) ::l '" XU,4>U
which can be seen to be equivalent with
(53) (3:4>)(~) '" XU, 4>(~
Now when this rule is being used, the first and only individual chosen in
the game connected with (51) is selected by Nature trying to falsify the
antecedent of (51), As was pointed out earlier, strategies on which such
moves are based are not "remembered" in the game connected with the
consequent of a conditional, and hence cannot support a pronoun occurring
there. Hence our theory predicts that the pronoun is out of place in (51), as
we have found it to be.
This confirmed prediction further supports our theory, and certainly does
not amount to a counter-example to it. It is perhaps worth observing that
there is some independent evidence for the uncertainty as to whether the
"he" in (51) is supposed to pick out an arbitrary student who flunked or an
arbitrarily selected student who did not. This uncertainty is a consequence of
our explanation for the unacceptability of (51) in that the arbitrarily selected
student in (51) only serves to highlight the speaker's comment on what not
all students are like. The same uncertainty shows up in another way in the
sact that it is not clear on linguistic grounds alone whether the following
fentences speak of flunking or nonflunking students or of students simpliciter.
(54) If not all students flunked, they must have studied quite hard
(55) Even if not all students flunked, they cannot have studied very
hard.
This strengthens further our explanation for the unacceptability of (51).
Our treatment of pronominalization in sentences like (6) is immediately
extended to a large class of relative clauses. The following are cases in point:
(56) Everyone who owns a donkey beats it.
(57) Everyone who owns a donkey that he likes beats it.
(58) Everyone who doesn't like a donkey beats it.
In all similar cases, we can explain why "a" has in them a universal-
quantifier sense. Again, this sense is not a separate meaning or separate use
of the indefinite article, but an inevitable consequence of the way it occurs
in such sentences as (56)-(58).
It is important to realize, furthermore, that the relevance of the subgame

idea is not restricted to conditionals. Changes similar to the transition from
(G.if) to (G.cond) are needed also in (G. and) and (G.or). For instance, and
quite importantly, in the game G(X and Y) the players first play G(X). Only
if the winner is myself do they move on to play G(Y). The winner of this
second subgame wins G(X and Y).
Thus a pronoun in the second conjunct Y of "X and Y" with a quantifier
phrase as its antecedent in X is admissible in essentially the same circum-
stances as in the consequent Y of "If X, Y." For instance, when there is just
one unnegated quantifier in the antecedent of the pronoun, we have an
acceptable conjunction if and only if this single quantifier is an existential
one. Thus we have examples like the following:
(59) Some soldier was given a rifle, and he immediately fired it.
(60) *Every soldier was given a rifle, and he immediately fired it.
Since the modification of (G.and) seems to be only a preferential one, some
speakers might want to have a question mark instead of an asterisk in (60).
This would not tell against our theory, however.
This theory also explains why the indefinite article is an existential one and
not a universal (generic) one in conjunctions like the following:
(61) Bill owns a donkey, and he beats it.
We have already explained why the pronoun is grammatical in (61). It picks
out an individual earlier chosen by a strategy of mine, just as in (5). However,
in (5) this strategy is chosen by Nature while playing what originally was my
role, whereas in (61) it is chosen by myself. Since it is Nature's role here that
turns a quantifier into a universal one, our theory predicts that the indefinite
article is generic in (5) but not in (61), as it obviously is. Once again we see
that the so-called generic force of "a(n)," is not always an irreducible pheno-
menon but can often be predicted by means of a suitable semantical theory.
In contrast to (61), the following sentence is not grammatical, just as our
theory predicts that it is not.
Bill owns every donkey, and he beats it.
Likewise, the subgame idea has often to be brought in to account for
pronominalization across disjuncts.
Although we shall not discuss the syntactical problems of pronominaliza-
tion in this essay, some light may be thrown on them by our observations.
Our account of the possibility of certain types of pronominalization turns

entirely on semantical concepts and seman tical conceptualizations, such as
the subgame idea which forms the gist of (G.cond 1), (G.cond 2) and (G.neg)
This heavy reliance on semantics suggests rather strongly, albeit somewhat
obliquely, that a full account of pre nominaliza1 ion is impossible along purely
syntactical (generative) lines.
This suggestion is reinforced further when counterparts to some of the
problems of pronominalization within a sentence are found in text semantics
(in sentence-to-sentence pronominalization). A full solution to such problems
cannot very well be expected in terms of the generation process of individual
sentences. We shaH in fact turn our attention to these problems next.
For the most sweeping extensions of our results we have to go back to the
basic idea of our new rules like (G.cond) and (G.neg). This idea is that some-
times the semantical behavior of a word or phrase in a certain context has to
be accounted for in terms of suitable semantical games which are supposed
to have been played to the end already when we come to interpret this word
or phrase. Often, these critical words are pronouns. In the case of the
pronouns in (4) and (27) the games that are needed to understand them are
subgames, that is, games occurring as parts of the more comprehensive
supergame associated with the sentence in which the pronoun occurs. How-
ever, this is not the only case of its kind. Perhaps the most interesting re-
percussions of one basic idea are the text-semantical ones. One of the most
important phenomena of text semantics and text grammar is that the se-
mantical interpretation of a text proceeds in order from sentence to sentence.
When we come to a given one, we can assume the semantical games con-
nected with the earlier sentences to have already been carried out. And since
the different sentences of a text are normaHy thought of as being combined
conjunctively, the earlier sentences are to be assumed to be true, for only if I
win in the earlier games do the players proceed to later ones.
These observations are but further applications of our Progression Principle
formulated earlier. It is obviously relevant to many interesting phenomena in
text grammar and text semantics. Its main bite is in fact found here, it seems
to us. Only a part of its force is brought to bear on sentence grammar and
sentence semantics in the form of the subgame idea which we have been
exploiting in this essay.
As was already hinted at, one of the most obvious applications of the
Progression Principle is to explain the semantical possibility of certain kinds
of sentence-to-sentence pronominalization. Many of these kinds of inter-
sentence pronominalization are closely related to similarly problematic

varieties of intrasentence pronominalization. The following examples
illustrate this.
(62) John just gave every child a present for Christmas. Some child
or other will open it already to-day.
(63) Every soldier has a loaded rifle. Some of them will fire it before
they are ordered to do so.
The great variety of sentence types creates of course an almost corres-
ponding variety of similar examples of text-grammatical pronominalization.
For instance, the second sentences of (62)-(63) could equally well be,
respectively,
(64) I forbid any child to open it until Christmas Eve.
(65) Has any soldier fired it?
The examples (64)-(65) and their ilk are especially interesting in that they
show that sentence-to-sentence pronominalization problems cannot be
reduced to intrasentential problems by the mere trivial device of conjoining
the different sentences in question. This becomes very unnatural when the
different sentences are of different kinds (declaratives, imperatives, questions,
etc.), and it becomes completely impossible when the sentences in question
are uttered by different speakers, as in the following example.
(66) Does every soldier have a rifle? Yes, even though some of them
received it only yesterday.
More complicated examples combine a modification to one of our game
rules with sentence-to-sentence pronominalization. Here is an example which
turns on the difference between (G.neg) and (G.not).
(67) John did not after all marry a girl who has lots of money. She
is quite pretty, however.
All these examples (62)-(67) allow for an explanation by means of the
Progression Principle. The full implications of the principle nevertheless need
a separate investigation. Let us note here only that some of the uses of our
theory are negative ones, to explain, at least partly, why certain types of
sentence-to-sentence pronominalizations are not feasible.
In order to find such applications, an observation supplementary to our
earlier ones is needed. Even though we cannot syntactically speaking conjoin
the different bits and pieces of a text into one long conjunction, semantically
speaking a text usually proceeds conjunctively: the successive sentences are
all intended to be true. If so, the conditions of sentence-to-sentence prono-
minalization will normally be the same as those of pronominalization be-
tween conjuncts. Even though this explanatory schema needs qualifications,
it has plenty of explanatory bite. For instance, witness the difference in
acceptability between the following pairs of sentences:
(68) Some solder has been given a rifle. Has he fired it?
(69) *Every soldier has been given a rifle. Has he fired it?
The relationship between (68) and (69) is not obvious unless we assume the
Progression Principle. Prima jacie, one might even expect that a universal
quantifier has a more definite reference than an existential one, and would
therefore be a better candidate for an antecedent of a pronoun. Yet a com-
parison between (68) and (69) shows that the opposite is the case.
In fact, this observation can be generalized. An earlier quantifier phrase
marking my move, e.g., "some X" or "a(n) Y" can ceteris paribus serve as
an antecedent of a pronoun in a later sentence, while a similar universal-
quantifier phrase usually cannot. What we have said serves as an explanation
for this phenomenon.
The explanatory force of our theory can be illustrated further by reference
to the following example:
(70) Every student held a tray. A girl had laden it with fruit.
Here either the tray of each student had been filled by a girl who need not be
a student, or else the second half speaks of only one girl and her tray. In the
latter case, the girl must clearly be one of the students. Why? Where does this
implication come from? It comes from the need of having an antecedent for
"it." According to our theory, this pronoun relies on a strategy of mine, and
it is readily seen that such a strategy provides an individual reference for "it"
only if the girl in question is one of the students.
Another phenomenon which becomes understandable is the use of the-
phrases anaphorically, that is to say, to pick out individuals introduced earlier
in the same text, perhaps even rather indefinitely. Such the-phrases need not
have a unique reference absolutely, only given certain plays of the games
associated with earlier sentences of the same text. Here is a sample narrative:
(71) A tall stranger appeared on the road. The stranger approached
90 JAAKKO HINTIKKA AND LA URI CARLSON
a farmhouse. He came to a door of the farmhouse. The tall

stranger knocked on the door ....
This illustrates an important difference between logicians' theories of definite

descriptions and their actual use in ordinary discourse (text). For any account
of the semantics of "the" which is like Russell's famous theory of definite
descriptions requires that the-phrases exhibit uniqueness absolutely, and not
just in relation to a given play of a certain game. At the same time, it suggests
that logicians' idea of unique reference has something to recommend itself,
if developed and applied appropriately.
Again, we can note that the individual which has been introduced earlier
in the text and to which a the-phrase ("definite description") refers can
typically be introduced by an existential-quantifier phrase but not by a
universal-quantifier phrase. Our subgame idea again explains why this should
be the case.
We cannot resist the temptation of casting a side glance here at attempts to
explain pronominalization in terms of definitization. What we have been
discussing is a type of context where the possibility of either process requires
an explanation. However much progress is achieved by reducing one to the
other, such a reduction accordingly cannot solve all the problems of pro-
nominalization.
These sample applications of our theory of subgames are probably enough
to whet the reader's appetite for further ones.
A general comment on what we have been doing in this essay may be in
order by way of conclusion. Some of the pronominalization phenomena we
have studied have been assimilated in the literature to the so-called "pronouns
of laziness," that is, to pronouns which merely serve as placeholders for their
antecedents, irrespective of questions of coreference. It should be obvious by
this time that the pronouns studied in this essay are not pronouns oflaziness'
On the contrary, they serve to recover a reference to an individual which ha.
somehow been introduced earlier. Their peculiarity lies rather in the fact thas
the antecedently introduced individual is somehow an "arbitrarily chosen't
or otherwise underdetermined individual. Thus a better slogan for the phe-
nomena studied here would be "coreference without reference." This label is
partly metaphoric, of course, and what we have been doing here is to spell
out what it really covers, without using the dubious notion of coreference.
Hintikka has pointed out earlier that some instances of "co reference without
reference" are essentially modal, that is, involve tracing one and the .same
individual from one possible world to another. Here we have been discussing
instances of "coreference without reference" that arise in apparently com-
pletely non-modal contexts.
Another interesting general remark prompted by our observations is the
following. We have offered an account of the reasons for the acceptability and
unacceptability of certain types of expressions in English. This account is in
terms of certain semantical regularities of English, indeed regularities which
can be generalized from sentence semantics to text semantics. It is therefore
in sharp contrast to the whole tenor of generative grammar, where accep-
tability, un acceptability, and differences in the degree of acceptability are
(hopefully) accounted for by means of the generation process of different
kinds of sentences. What are we to say of this contrast?
What has been said does not exclude a generative account of the same
phenomena. But what would such an account look like? Basically, it would
in the paradigm case of conditionals have to deal with the restraints on
forming "If X, Y" and "Y if X" from X and Y. It is not obvious that these
restraints can be incorporated in an effective (recursive) generative rule.
However, even if they can, what would a theoretical motivation of the result-
ing rule look like? It is quite obvious that there cannot be any purely syntac-
tical motivation forthcoming. For one thing, the governing regularity we
have found extends also to text grammar, and hence cannot conceivably be
accounted for in its full generality in terms of the way iIi which individual
sentences are generated.
We have noted, moreover, that the relevant text-semantical principle
cannot be reduced to its sentence-semantical counterpart by the tempting
device of thinking of a text as a conjunction of its constituent sentences.
In contrast, our account ensues perfectly naturally from certain semantical
ideas which are forced on us in any case by the non-truth-functional character
of conditionals, quite independently of any problems of pronominalization.
Moreover, in some obvious sense our account is also closely related to the
way in which we in fact process a sentence semantically. When a speaker
rejects (7) but accepts (8), he is scarcely relying, however implicitly, on the
processes by means of which these two strings could perhaps be generated.
Rather, he perceives what happens when he tries to analyse these two strings
semantically. In (7), but not in (8), he is confronted by a pronoun whose
reference has not yet been fixed in the context of our semantical games at the
time he comes to it. This explanation is in keeping with the basic idea of our
approach: to understand a sentence S is to know what happens in the
correlated game G(S).

Hence we have found an example of an essentially semantical explanation
of the facts which in the generative approach are paradigmatically explained
(in so far as they can be explained) in syntactical (generative) terms. This
casts serious doubts, not on the soundness of the research strategy of the
generativists, but on its scope.
One example perhaps does not carry much persuasion in this respect.
However, Hintikka has found another striking example to the same effect in
the any-every contrast in English (Hintikka, 1975). Nor is this the only inte-
resting recent example of relevant interplay between semantics and syntax.
It seems to us in fact that the interaction of semantics and (what is usually
taken to be) syntax is a much deeper and subtler phenomenon than linguists
have recently realized.
Academy of Finland and

Stanford University
BIBLIO G RAPHY
Andrews, Peter, 1972, "General Models and Extensionality," Journal of Symbolic Logic
37: 395-397.
Henkin, Leon, 1950, "Completeness in the Theory of Types," Journal of Symbolic Logic
15: 81-91.
Hintikka, Jaakko, 1974, "Quantifiers vs. Quantification Theory," Linguistic Inquiry 5:
153-177.
Hintikka, Jaakko, 1975, "On the Limitations of Generative Grammar," Proceedings of the
Scandinavian Seminar on Philosophy ofLanguage (Uppsala, Nov. 8-9, 1974), Filosofiska
Studier utgivna av Filosofiska forening och filosofiska institutionen vid Uppsala
Universitet 26: 1-92.
Hintikka, Jaakko, 1976, "Quantifiers in Logic and Quantifiers in Natural Languages," in:
Philosophy of Logic, S. Korner (ed.), Oxford, Blackwell, 208-232.
Hintikka, Jaakko, forthcoming, "Negation and Semantical Games."
Martin-Lof, Per, 1970, Notes on Constructive Mathematics, Stockholm, AImqvist and
Wiksell.
Mostowski, Andrzej, 1966, Thirty Years of Foundational Studies, Acta Philosophic a
Fennica 17, Oxford, Blackwell.
HELMUT SCHNELLE
CIRCUMSTANCE SENTENCES
1. Circumstance sentences are sentences which may be understood only in

the context of the circumstances of utterance. In the specific sense of the
term I have in mind Quine's occasion sentences and standing sentences -
i.e. sentences, whose correct or incorrect application is strictly determined by
the occasion of their utterance or an occasion preceding the utterance within
a tolerable time interval - are not circumstance sentences.
In the present contribution I shaIl confine my discussion to circumstance
sentences bearing a special relation to occasion sentences; they would be
occasion sentences if they were in present tense) More specificaIly even, I am
concerned with occasion sentences shifted into past tense. I think that such
sentences strictly depend on settings in the state of use of the current semantic
network as do, perhaps, all other simple sentences not in present indicative.
An example I am going to discuss is
(13) The door was open.
It may be considered to be derived from the occasion sentence

(4 ') That door is open
accompanied by an act of pointing. My claim is that the complete understand-

ing of an utterance of (13) requires at least the knowledge of a circumstance
answering the where and when of (13) and that the usual method to specify
the where and when by the scientific concepts of space and time is inappro-
priate for the analysis of the functioning of ordinary language and in partic-
ular of verb inflection. I think that it is precisely the control of the relation of
predication to the circumstances of predication that is the essential point of
verb inflection. An insight into this relation is barred by the usual analysis
developed for scientific language varieties. The reason is that these language
varieties require only the analysis of eternal sentences and occasion sentences.
In the analysis of the former circumstances are irrelevant and the "circum-
stance" relevant for the latter is the occasion present at the utterance of an
occasion sentence and therefore need not be specified. I think that in trying
to explain the functioning of circumstance sentences one may provide for a
93
Copyright <CI 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
94 HELMUT SCHNELLE
partial analysis of Bar-Hillel's emphasis on the role of context and co-text

in the functioning of ordinary language. 2
2. In order to introduce my own proposal for a treatment of circumstance

sentences I shall first discuss the principles of canonical regimentations or
formal reconstructions in connection with tense and circumstantial deter-
mination in general, trying to show where they go astray.
Usually, logicians and analytic philosophers feel rather confused by the
use of tense in natural language. Let me cite Word and Object as an example:
"Our ordinary language 'shows a tiresome bias in its treatment of time.
Relations of date are exalted grammatically as relations of position, weight
and color are not. This bias is of itself an inelegance, or breach of theoretical
simplicity .... Hence in fashioning canonical notations it is usual to drop
tense distinctions" (p. 170). Quine proposes to regiment a sentence like
(1) I telephoned him but he was sleeping
by
(2) I telephone-[tenseless] him then but he is-[tenseless] sleeping
then
Quine remarks, however, that " 'then' refers to some time implicit in the
circumstances of the utterance" (ibid.). This remark is taken as the starting
point of the solution proposed by the logician. Leaving the question how
circumstances of utterances are to be explicated unanswered (whether time
is indeed implicit in those circumstances, and if so in which way it is implicit),
the logician turns immediately to the problem how reference to time con-
tributes to the understanding of the sentence. In his scientific context of
explication he takes the spatialized time concept which represents time by
one or more time line(s) for granted (cf. Quine, p. 172), each time line being
a dense sequence of time points, and he assumes that the deictic particle
"then" in sentences like (2) refers to a time point or an interval on the time
lines. Two ways are then open to make use of this idea in semantic analysis.
One consists in transforming the sentence with the deictic particle into a
context-independent one by substituting the particle by adverbial phrases
containing expressions that render dates and clock-readings. This is the
method of eternalizing the sentence, as Quine says.
The other way - contextual analysis - consists in letting the interpretation
CIRCUMSTANCE SENTENCES 95
frame for the sentence contain, in addition to the specification of the form of
the language and of the general conceptual scheme, a value for a time para-
meter and to assign to the sentence a meta-linguistic entity which is a function
defined for all time points assigning truth-values to time points. In certain
cases it may be necessary to allow for functions over several time points. This
procedure is in the line of treatments by means of indexing suggested by
Reichenbach and following him by others.3
3. For further discussion it is useful to remark that the procedure of

eternalizing sentences usually requires a parallel operation with respect to
the nominal phrase. As an example Quine discusses the eternalization of
(4) The door is open
in order to provide a substitute for what is said that Tom believes in uttering
the sentence
(5) Tom believes [the door is open].
After having eternalized with respect to time by, say,
(6) The door is-[tenseless] open on Nov. 1st 1958 at 3.15 p.m.
US Pacific time
something more has to be added to eternalize the reference of "the door"
which, as it stands, can only be identified on the basis of our knowledge of
the circumstances of utterance of (4). What should we have to do in order to
eternalize the nominal phrase? The logical counterpart to the solution for
time would be to substitute "the door" by "there" obtaining
(7) There open then
or
(8) There open on Nov. 1st 1958 at 3.15 p.m.
and saying that "there" refers to some space-time region implicit in the
circumstances of the utterance, corresponding to having said above that
"then" refers to some time implicit in the circumstance of the utterance. In
effect, this is how Quine proceeded in some later articles though not motivat-
ing it by these very words. Again, the corresponding contextual procedure to
be applied to the analysis of (7) would be to assign to a predicate a function
96 HELMUT SCHNELLE
defined in the Cartesian product of space-time regions and time points.

Usually, analysts are not as bold as this, in assuming that objects are
merely circumstantial aspects contributing to an implicit indication of
space-time regions and that they may, therefore, be neglected just as all
other factors contributing to this implicit reference, though from a point of
view aiming at the language of science this might be quite all right. At most,
then, analysis would proceed by assuming the problem of the roots of
reference solved and trying to explicate instead of (6) a sentence like
(9) This is-[tenseless] open on Nov. 1st 1958 at 3.15 p.m.
where "this" is to refer to some object from an eternal universe of discourse-
objects, and may be substituted by an individual expression.
But, usually, analysts interested in natural language are not even as bold
as this. Eternalization follows more sophisticated lines. After remarking that
"the door" in (4) or (6) is not yet identifying a particular object it is found
that adding an attributive clause to the nominal phrase may achieve the job
of identification. The nominal phrase by itself gives only the category of the
object to be identified and an indication that a particular object is to be
selected. On the basis of this insight, the appropriate regimented paraphrase
seems to be
(10) The so-and-so door is-[tenseless] open on Nov. 1st 1958 at
3.15 p.m.
and by some substitution of the qualitative particle we might obtain
(11) The door in Quine's studio in the Center for Advanced Studies-
in-autumn-1958 is-[tenseless] open on Nov. 1st 1958 at
3.15 p.m.
The corresponding alternative of contextual analysis would take the word
combination "so-and-so door" to refer to a combination of properties whose
extension at some circumstance is a unit class such that the article "the"
somehow selects the element of the unit class.
The analysis of the nominal phrase teaches us quite a lot about the con-
tribution of the circumstance. The nominal phrase serves to point at a certain
object from a range of objects, the current objects of discourse. This deictic
factor is rendered by the definite article. However, as in most cases of deictic
pointing, the simple act of pointing is not sufficient. But, given the actual
circumstances, the specification of the category of the object is usually suffi-
CIRCUMST ANCE SENTENCES 97
cient. If it is not, the circumstance must be made explicit and, ordinarily, it

is made explicit by adding an attributive clause. Let us note in passing that
in complicated cases the attributive clause that has to be added may be rather
complex and comprise a whole story or description. There are, as we see,
four interacting factors: the universe of discourse-objects, the pointing by
means of the article, categorial characterization, and attributive circum-
stantial determination. If the circumstantial determination is left implicit, we
get a circumstance sentence with respect to the reference of the subject; if it is
made explicit, we get a circumstantially determined (eternalized) sentence,
which may be paraphrased in its most elaborate regimented form by
(12) be open (the [-object-pointed-at-or-marked] [by-help-oj-
circumstantial-reference-determiner] in the studio ... [among-
the-] door [-objects] [in-the-current-universe-oJ-discourse
objects]; on Nov. 1st 1958, 3.15 p.m.)
4. What happens if we now turn around and apply a corresponding

procedure to the analysis of tense rather than leaving in the dark the way it
refers to time? Let us discuss instead of (4) the sentence
(13) The door was open.
Again, we may paraphrase it into regimented style:
(14) The door is-[tenseless] open then-past.
If we substituted the same date and clock-reading as above, we would
again obtain sentence (6). Instead, I think that here, as above, the basis is a
linguistic act of pointing controlled equally by four factors: (1) a current
universe of discourse-occasions in which a circumstance sentence like (14)
might be applied; (2) a deictic particle pointing to an occasion - we may
take "then" as that particle (corresponding to "the" in the nominal case);
(3) a category of occasions pre-selecting from the universe of discourse-
occasions or determining somehow the direction (e.g. "past") in the field of
occasions in which the occasion meant can be found (corresponding to the
class of entities determined by the noun); and (4) the specific difference that
may serve to identify the occasion, i.e. the circumstantial occasion determiner.
The specific difference is given by an adverbial clause (explicit or implicit)
that interacts with tense in a way corresponding to the interaction of an
attributive clause with the noun. 4
98 HELMUT SCHNELLE
A very important point I want to make is the following: It is not necessary

at all that the circumstantial occasion determiner be an adverbial temporal
expression based on dates and clock-readings! I even think that this assump-
tion is rather misleading if the eternalized sentence should be taken to
express approximately the same content as does the circumstance sentence
itself. Since, usually, people do not know the dates and clock-readings of the
events to which they refer in the past but rather their relation to other events,
the circumstantial determiner should be also of this type rather than based
on dates and clock-readings. In the case of circumstance sentence (13) it may
be clear to both the speaker and the hearer froni what was said before that it
applies to an occasion immediately after some particular occasion where
Quine left his studio and immediately before some particular occasion in
which his cat rushed into the studio to sit there on the mat. In a given
universe of discourse-occasions this much of information may be com-
pletely sufficient to identify the appropriate occasion and to consider the
question whether what has been said is true or not. In terms of this analysis
we may give the circumstantially determined ("eternalized") sentence for (13)
in the explicit, regimented form as follows:
(15) be open (... as in (12) ... ; then [in-the-occasion-pointed-at-or-
marked] [by-help-of-circumstantial-occasion-determiner:] after
Quine leaves his studio and before Quine's cat rushes into it
[in-the-field-of-] past [-occasions] [in-the-current-universe-of-
discourse-occasions])
As you see, my analysis of the tense marker is that it serves as an aid to the
deictic pointing in much the same way as would the qualifications "before
me," "above me," "to my left" in space. These qualifications ("past,"
"future," etc.) predetermine the field of possible objects or occasions that
may be referred to in practically the same way as do relative category terms
such as "(my) friend," "(my) neighbor," which, in addition to pre-classifying
like the terms "man," "woman," "animal," "plant," relate the classified
objects to the speaker. These general terms serve as a basis for further
determination either already present in the context or made explicit by a
circumstantial determiner expression.
The simple deictic particle accompanying an act of pointing in an actually
present occasion is split into various factors contributing to the successful
reference in particular non-present contexts. We get the parallel sequence
(16)-(18) by adding one factor in each line. In (16) there is merely an indica-
tion of definite reference, in (17) preselection (by noun or tense marker) is

added, and in (18) the specific difference is made explicit (by attributive or
adverbial clause).
(16) a. that b. there c. then
(17) a. that man b. (at that) c. (at that) party
,..,the man field there then
a'. that friend b'. there before c'. then past from
of mine here now
,.., the friend
of mine
(= my friend)
(I 8) a. the man, b. (at that) field c. (at that) party
who p there, where p then, when p
a'. my friend, b'. there before c'. then past from
who p me, where p now, when p
5. I want to add a brief remark on attributes and adverbial phrases. As we

have seen, in some noun phrases attributes function as circumstantial deter-
miner expressions and in some sentences adverbial phrases function analo-
gously as circumstantial determiners. The former determines the object
referred to and the latter the occasion the sentence is to be applied to. These
attributes and adverbial phrases restrict the range of objects or occasions
categorially determined by the head of the noun phrase or the tense of the
verb to - ideally - a unique object or occasion. This is the reason why these
attributes - especially relative clauses - are called restrictive.
This is no110 say, however, that all attributes and adverbial phrases func-
tion in this way. On the contrary, often the attributes and adverbial phrases
give a further specification of what has been expressed by the "head" they ale
attached to rather than determine a condition for their successful application.
This specification may express an additional property of the object or an
additional specification of the occasion, as in the cases
Schliemann, who firmly believed that Troy existed, prepared
the excavations in Turkey
or
Schliemann prepared the excavations, since he firmly believed
that Troy existed.
100 HELMUT SCHNELLE
The specification may also combine in more complicated ways with the
noun phrase or the sentence to yield an expression of a related object or
situation, as in the cases of the
alleged murderer
or
Seemingly, he prepared ...
These differences show only partially in the form of the sentences, they are
more appropriately uncovered by semantic relations such as paraphrase,
entailment, etc. It is nevertheless important, I think, that these differences are
specified already as grammatical differences.
6. We started our discussion with reference to Quine's remark that sentences

containing "then," i.e. circumstance sentences, refer to some time implicit
in the circumstances of the utterance. If, however, the circumstance of
utterance is mainly determined by tense and circumstantial determiner, as
discussed so far, time is not yet referred to in the strict sense of implying dates
and clock-readings. But in most uses, it seems to me, time in this sense is not
implied at all. It is sufficient, in these cases, that the occasion for which the
circumstance sentence is to hold is identified within a field of occasions. This
requires that the field of occasions from which the appropriate occasion
should be selected be specified and that by means of a unique reference to
parts of the scheme we may localize the occasion in the field.
If we envisioned each field of occasions as a map with temporal relations
between occasions, we have to know in each circumstance which map (i.e.
field of circumstances) we are required to refer to and to which part of the
map. I think that the appropriate map is selected from the class of alternatives
by the tense marker. The appropriate location on the map is determined by
the circumstantial determiner; by referring to uniquely identifiable "tem-
poral" localities on the map - corresponding to events expressed by the
sentence or phrase that is part of the circumstantial determiner - it specifies
the relation of the appropriate location to these localities. Usually, however,
to stay in the picture, the map is more or less a topological map (at most with
very rough indications of distances). Even if our own location on the map is
marked - and it is by the time of utterance - we could not even compute
the time in terms of dates and clock-readings. But in many cases, such as in
fiction and novels, even the marking of our own location is without im-
portance, neither the date of writing nor the date of reading plays any
essential role. In this case the use of the map is completely detached from
time. It is like a map without absolute coordinates and without localities
whose location we know. s
I do not want to say that most uses of language are of this type, but only
that language and tensed sentences may function perfectly well without
reference to time whatsoever, provided we have a field of occasions available
to which we want to refer in our talk.
7. Adding circumstantial determiners to circumstance sentences does not

eternalize these sentences in the strict sense, since the interpretation of the
circumstantially determined sentences depends not only on the language and
its conceptual scheme but also on the universe of discourse-objects and on
the universe of discourse-occasions. This is in contrast to scientific language,
since there we may assume one and only one strictly fixed universe of
discourse-objects. Only in connection with certain treatments of modal logics
and propositional contexts some analysts consider a number of universes of
discourse-objects and universes of discourse-occasions, but all of them are
still fixed in advance. In many uses of ordinary language this does not seem
to be the case. For instance, in telling a story, we introduce the elements in
the universe of discourse-objects as well as the elements in the universe of
discourse-occasions and then later on in the discourse we refer to the objects
and occasions introduced.
I think that we have to admit that the use of linguistic expressions is
strictly context-related in two ways: On the one hand their analysis and
interpretation is context-dependent - as has been shown, for instance, for
circumstance sentences - and on the other hand they contribute to building
up and changing a context, within which further expressions will have to be
interpreted. Expressions for noun categories, attributes, and articles as well
as expressions for occasion fields like tense, circumstance determiners and
deictic particles interact in controlling these processes. This position is rather
widespread among linguists. One of those who made it particularly clear
was J. Lyons:
When we use language to communicate with one another, we do not produce sentences,
but utterances; those utterances are produced in particular contexts and cannot be under-
stood without a knowledge of the relevant c.ontextual features. Furthermore, in the course
102 HELMUT SCHNELLE
of a conversation (let us suppose that it is a conversation) the context is constantly develop-

ing, in the sense that it "takes into itself" from what is said and what is happening all that
is relevant to the production and understanding of further utterances (J. Lyons, 1968,
p.419).
The position is not as modern as the date of the quoted passage may make
believe. There is even a doctrine of the structure of a text stemming from the
Neo-Grammarian H. Paul as early as 1886 (cf. H. Paul, 19667 , p. 148). In
referring to certain types of narrative texts he says that in the text each
sentence provides to the following a temporal and sometimes a causal deter-
minatIOn. He adds that we may think of a clumsy manner of expression
according to which each sentence is expressed twice, once independently,
making an assertion, say, once dependently, determining the following sen-
tences. 6 In this way the two functions of the utterances, presenting a piece
of utterance for interpretation and contributing to the context, are se-
parated. 7 K. BUhler took this idea up in his Sprachtheorie (1934) and
elaborated it in saying that anaphoric particles contribute to controlling the
reference to those pieces of the context which determine the interpretation of
a given utterance of a sentence or even serve as introductory particles for
clauses which are explicitly used to recollect earlier material or to indicate
information still to come (cf. p. 391, also pp. 374-375, 390,405).
To sum up: Utterances of expressions have a double use (a) to trigger a
meaning depending on a given context and (b) to change the context.
This resembles pretty much the concept of an automaton in terms of a
switching function if we correlate contexts with states of the automaton,
utterances with inputs, and meanings with outputs. Since there is an
infinity of possible contexts the corresponding switching system cannot be a
finite but must rather be an infinite state system. In terms of this analogue,
then, the meaning of an utterance depends in general on the state in which
the interpretation system is. In terms of this metaphor, ordinary logical
analysis of sentences is of a very special case. If we consider perceptual data
given in the occasion of an utterance to be input data on a channel different
from the linguistic channel, utterances of sentences considered by logicians-
namely occasion-sentences or eternal sentences - neither depend on context
nor do they contribute to context. In this case the corresponding automaton
degenerates into a one-state automaton. Interpretation of circumstantial
sentences on the other hand needs to be rather different from interpretation
of eternal or occasion sentences.
8. The theoretical appeal of this contextual or infinite-states-system ap-

proach is rather weak as long as we are unable to make the notions of
meaning and context precise in a way which enables us to identify and to
distinguish meanings and contexts, and, it must be admitted, linguists did not
succeed in making it sufficiently precise.
What is it that we have to determine? There is the sound complex which is
possibly an utterance of an expression in some language, and there is the
automaton whose task it is to recognize which expression of which language
it is an utterance of and to determine its meaning as well as its contribution to
the contextual analysis of further sound complexes. What is it we have to
endow the automaton with if it is to execute the task? Obviously, the auto-
maton must be able to identify and distinguish the expressions of a language
and determine the ranges of their possible utterances, in short the automaton
must know the language. But there is more: In order to be able to interpret,
it must have command of a conceptual scheme connected with the expres-
sions of the language. Let us assume that it is a general conceptual scheme in
so far as it remains unchanged by utterances interpreted in terms of the
scheme. Let us assume in particular that the scheme consists of statements
that may be true or false - or believed to be true or false - and a finite
sequence of relations expressing dependencies of the truth and falsity of the
statements related. The relation comprises in particular negation, paraphrase
and entailment. Because of this system of semantic relations we call a con-
ceptual scheme of this type a semantic network. s
An interpretation of the utterance of a sentence consists in assigning to it a
position in the semantic network, i.e. a statement or a statement form in the
network. The statement or statement form assigned will be called the meaning
center of the utterance under that interpretation. Among those factors
determining the meaning of an utterance the meaning center is the one stem-
ming from the utterance alone. The meaning of an utterance is to consist of
those positions or statements to which the meaning center may be related in
specific ways together with the relations it has to them. This concept of
meaning is, obviously, sense-relational in the sense of Lyons and also
of Hiz.9
Note, however, that only for eternal sentences the meaning is completely
determined by the meaning center alone. Utterances of sentences which are
not eternal sentences may not be interpreted on the basis of the conceptual
scheme only. Let us consider the following sentence
104 HELMUT SCHNELLE
(19) Black captured White's rook by his king.

We may be able to provide a meaning center by noting that someone called
Black captures under a certain circumstance in a complex of events, which the
speaker assumes that the hearer remembers among the events he classifies as
past, a certain rook determined by the circumstance and possessively related
to someone called White, and that he succeeds in doing so by means of a
certain king determined uniquely in the circumstance and possessively related
to Black. However, this meaning center is much too vague to play any
psychologically relevant role in understanding. But it is clear how this mean-
ing center may become more concrete: First, the hearer should discover from
the words that the speaker is speaking about a game of chess, let us call this
the general topic-of-the-setting. This general topic-of-the-setting enables him
further to have access to his general knowledge-about-the-general-topic-of-
the-setting, the rules and strategic methods of chess as well as the conventions
for talking about chess. From this he gathers e.g. that Black and White are
not persons but roles in the game that certain persons may take on. More in
particular he should know that he is asked to refer to a particular game which
he is supposed to have available in the sense that he would be able to relate
the moves, at least the relevant ones; i.e. he must know the sequence ofmoves
of that game. It is either a game which is in the past in its entirety or, if the
game is an ongoing game, the moves he is referred to are the moves of the
game made so far, i.e. among the moves he may consider he should only
select those in the appropriate sector, the past, i.e. he must know the sector
referred to. Finally he must know a "pointer" pointing to a particular move
within the complex of moves. This is the circumstantial determination, i.e. the
circumstances for which it is claimed that (l9) holds. Note that in a game of
chess there may be several occasions for which (l9) may hold true since that
move can occur several times. The circumstantial determination must specify
to which occurrence we are referred to.
I assume that each of these informations may be rendered by twice marking
certain of the statements in the semantic network, first by the value true and
second by an actual relevance value. We determine a certain state of use of
the semantic network by this marking. If the semantic network is thus
marked, it is easy to imagine how the meaning center mentioned above is
used: It is combined with the circumstantial determination to form some
complex statement which is about a game of chess and which is a unique
substructure of the past moves of the game under consideration and thereby
localized in that game. If this has been done the sentence may be evaluated
for truth. But it may also be that the speaker is not arguing with the hearer
but rather informing him about what happened in that game. In this case the
hearer is not required to evaluate the truth of what the speaker says but
rather to enter a piece of information into his knowledge about the past of
the game, i.e. changing his context for the interpretation of following sen-
tences appropriately. Certainly he need not mark his information simply as
true, but depending on the confidence he pays to the speaker as true according
to the speaker who may be payed confidence in this matter to the degree x,
i.e. truth with confidence x. In any case, the situation of conveying informa-
tion or telling a story is typically one where change of context is very impor-
tant together with interpreting what was said.
The discussion of the example shows that the interpretation and change of
context of circumstance sentences requires more than knowledge of the
language and its conceptual scheme. Their interpretation needs further in-
formation. The further information we are going to make use of is a partition-
ing of the conceptual scheme on the one hand and a marking of positions in
the resulting sector~, or, more precisely, equivalence classes of such markings.
A particular marking of positions or statements determines a state of the
conceptual scheme. In this way meanings as well as contexts of interpretation
are determined by positions in the semantic network of a language. A meaning
of an utterance is now determined by the meaning center marked due to the
occurrence of the utterance and by the context represented by the marked
positions in the network. Operating on an utterance of a sentence in terms of
a semantic network or conceptual scheme consists in (a) finding the meaning
center and its relation to other positions in the network relativized by context
(among them the position by which the truth value of the utterance in the
network may be determined) and (b) changing the marks on positions deter-
mining the context; (a) is the process of interpretation and (b) the process of
changing the context of interpretation. Oldinarily, then, the meaning to be
assigned to an utterance in the semantic network is dependent on the marking
of the semantic network; it is contextually determined. Only for special
sentences is the interpretation independent of the state of the semantic net-
work yielding for each state the same meaning center to a given utterance of
the sentence; these sentences are eternal sentences or occasion sentences.
Determining contexts in this way makes it natural to consider them as
states of the semantic network. In terms of the metaphor used above, the
semantic network is the structural basis of the interpreting automaton and
106 HELMUT SCHNELLE
the marking of certain positions defines its states.
9. We may, however, look at the same proposal under a different angle: If

we identify our semantic network with a meta-language for semantics we may
consider interpretation to be a translation of a sentence of the object language
into a sentence of the meta-language. In the case of eternal sentences the
translation is fixed. In the case of circumstantial sentences the translation is
relativized to the context. We may assume that the context can be expressed
in the meta-language; therefore, interpretation result as well as context are
expressible in the meta-language.
The same holds for a truth theory for sentences,lo Circumstance sentences
are then those sentences which may only be analyzed by means of a relativized
truth predicate. However, the relativization need not have recourse to notions
that do not occur in the ordinary translations into the meta-language; it can
be expressed in the same language as the correlates of the object language. In
particular, index points for rendering context as proposed by some model-
theoreticians need not be expressed in terms of set-theoretical notions; they
may as well be replaced by texts and phrases.
Relativization may be recursively reduced, if the object language is con-
tained in the meta-language and the expressions rendering relativization of
the truth predicate are contained in the object language. If we read Ch as
"a case of chess," G as "game G," P as "in sector of past events," M as "the
subset M of moves," and C as "circumstance C," we might have a reduction
sequence like
(20) r(19)' is true at Ch, G, P, M, C, iff ..
(21) r(19) {at, after, before} 0 is true at Ch, G, P, M iff .. .
(22) r(19) {at, after, before} C in Ch, G, P, M' is true iff .. .
The phrase contained in corner quotes of (22) might be taken as an eternal
sentence. It shows clearly that since C and M may be arbitrarily complicated,
trying to base semantics on the treatment of eternal sentences may require a
treatment of sentences as complicated as texts.
Truth-theoretical interpretation as well as translational interpretation
apply in a peculiar way to the expressions ofa language. I am with Kasherll
in assuming that the translational or meta-linguistic counterparts of expres-
sions are only sentence radicals (plus eventually classes of sequences that may
or may not satisfy the sentence radicals). A lot of further information, not
only mood but also tense and deictic pointers, do not get translated but only
provide conditions for translation or for pairing of the two sides of the truth
predicate. 12
But let me note an important point: Neither the theory of interpretation as
translation nor the truth theory interpretation do consider change of context,
i.e. change of relativization by a text, as something to be treated. This is a
drawback in view of the importance of this factor.
10. Even more sensitive to determination by circumstances than ordinary

non-present sentences are utterances of words that may be taken as utterances
of incomplete sentences. Such utterances occur frequently in conversations.
H.C. Longuet-Higgins (1972) studied conversations in terms of a computer
programmed to answer questions on a certain ongoing party where the com-
puter is informed about incoming and leaving of people as well as their pre-
sence in the room. We may consider the following conversation, similar to
the ones he studied
(23) Q: When David comes in will Bert be in
the room? P: Yes
(24) Q: Will Ed? P: No
(25) Q: Will Charlie just have left? P: Yes
(26) Q: Did Derek just come in? P: Eh?
Obviously, in answering (23) the questioner assumes that the question
concerns the ongoing party about which he wants some information. The
computer, being devised in order to take the role of the answerer, is pro-
grammed on that assumption. So much for the topic. Interesting is the
treatment of (24). It is understood in the same way as if the question
(24') When David comes in will Ed be in the room?
had been uttered. The answerer simply keeps the specification "be in the
room" and the circumstantial determination "when David comes in" in
memory as possibly relevant for following questions. When does the rele-
vance of this information expire? The specification expires by substituting it
by another, such as in (25). The circumstantial information becomes irrelevant
by - change of tense: The answerer is at a loss in answering (26). The circum-
stantial determination is a determination within the range offuture events. It
cannot serve that purpose in another sector, the sector of past as referred to
108 HELMUT SCHNELLE
in sentence (26). The questioner might have meant "under the present
circumstances" but then the question should rather have been
(27) Q: Has Derek just come in?
The answerer might take (26) simply as a slip of the tongue or as an error
and take it as if (27) had been uttered only if the information about the
questioner does not exclude this assumption.
In any case, the example seems to show, that our memory is indeed
organized into sectors and that circumstantial determiners point only within
the appropriate sector for which they are formulated. One may consider this
as a "tiresome bias" on the part of ordinary language, as Quine did, but one
may also consider this as a help of keeping one's memory structured into
manageable pieces which are psychologically as well as communicatively
sufficiently homogeneous. Simplicity of natural science need not correlate with
psychological and communicative simplicity.
11. In the framework discussed so far, "time" occurs in two different modi-
fications, the "outward" time and "inward" time as it were. The "outward"
time is the sequential order of actual or possible utterances and of the cor-
responding interpretation frames applied to their interpretation. The
"inward" time is the temporal structure inherent in the interpretation frames
or in the semantic network which is their structural core. Only the content of
the inherent temporal structure is potentially changed by each interpreted
utterance but not its general structure. As has been discussed in this article,
the general structure consists in a partitioning of the semantic network into
sectors of different types of temporal "presence": the "memorized" or past
sector, actual present ("now"), the sector of planned events, the sector of
events hoped for or wished for, the sector of events "simply" expected, etc.
Finally there is a sector for non-temporal facts. In addition to the partition-
ing into sectors such that each event or fact belongs to just one sector, each
sector (except the sector of actually present events and the sector of non-
temporal facts) is ordered by a temporal order relation (such as the relation
before or the relation immediately before) or, correspondingly, by the binary
associative operation of a free semigroup (of strings) with a set of generators
(the alphabet). On the basis of such a general structure, frames uniquely
identifying each possible occurrence within a sector are easily definable. Such
frames are potential circumstance determiners for a sector. The sector itself
is indicated by a tense marker or a modal marker.

The "outward" time (i.e. the sequence of interpretation frames) must be
related to the "inward time." The "outward" relation is rendered by the
change of the interpretation frame partially or completely determined by an
utterance interpreted, i.e. by the switching function for the interpretation
frame. The "inward" relation is a representation of the outward relation in
an interpretation frame itself. If this is available the interpretation frame itself
records the fact that under such-and-such circumstances someone made
such-and-such an utterance and that this utterance changed the knowledge
present before the utterance into a different but equally specifiable know-
ledge. This latter recording of the "outward" time relation in an "inward"
one is obviously more involved, it is knowledge about change of knowledge
relevant for the interpretation of utterances and not simply change of
knowledge. 13
In contrast to the general time concept presented here, natural science uses
only one homogenous sector, as it wele the past or complete memory at the
end of times of a spirit having lived at all times, or the expectation sector of
an omniscient spirit at the beginning of times, if one wants a psychological
correlate to the scientific notion of time.
Augustine is one of the most famous philosophical analysts of time who
takes this position. For him, the physicalistic time concept would not be a
time concept at all; God is non-temporal for Augustine because he knows
the "order of things" and "things" are not distributed for him in past,
present, expected or planned or hoped for future in a way which absorbs
humans as finite beings into passions. Return to God is being freed from
temporal involvement: "At ego in tempora dissilui, quorum ordinem nescio,
et tumultuosis varietatibus dilaniantur cogitationes meae, intima viscera
animae meae, donec in te confluam purgatus et liquidus igne amoris tui."
(Augustine, Confessions XI, 29, 39.)
Augustine discusses time with relation to a process under control of an
actor. His paradigm case is the singing of a song (e.g. ibid., 28, 38). Before
starting to sing it, each of its constituent acts are completely in the sector of
planned events. During the song part of the constituent acts have already
been transferred to the sector of past - through the execution of the cor-
responding constituent acts - and part of these acts are still in the sector of
planning. Obviously the constituent acts are ordered both in the sector of
past and in the sector of the planned future. In the actual present there is just
my action controlled by attention. In listening to a known song the present
110 HELMUT SCHNELLE
is just the perceptions on which attention is focussed and the future is the
expected future rather than the planned future. The actual present is more-
over the transient sector through which constituent acts flow from planned
future to executed past: "praesens tamen adest attentio mea, per quam
traicitur quod erat futurum, ut fiat praeteritum" (ibid.). Gradually, then, the
past, as much as it refers to the song, becomes longer and the plan still to be
executed with respect to the song becomes shorter. This corresponds to our
concept of a partitioned frame of interpretation that changes, in the course
of "outward" time, by adding constantly to the sector of the past and re-
moving constantly from the sector of planned or expected future already
executed constituent acts or events.
We may now ask, however, whether our concepts of time are the most
primitive ones on the one hand, and whether they are sufficient for the
purposes of linguistic analysis on the other. Both questions have to be
answered in the negative. As soon as our language contains words like
"recently," "soon" etc., we have to analyze either more primitively or in a
more sophisticated way. Let us start by the first. I think that among the most
primitive relational concepts is the notion of contrast which, for some
perceptual modalities at least, seem to be innate.1 4 There are two basic types
of contrast having temporal relevance, the change of a predominant or focal
feature in the perceptual field (say vision) into a contrasting one or the switch
from a predominantly felt need to its satisfaction or the other way round.
An elementary example of the former is the change of "light" into "no more
light" as experienced by a small child in his bed when mother switches the
light off. An example of the former is the need for sucking changing into a
state of satisfaction by taking a soother or a thumb or the converse of this
sequence. Thus the basic feature expressed linguistically by present perfect
has its basis already in these early experiences which do not yet involve
memory in a structured sense. It is, however, obvious that these experiences
are at the root of the first partitioning of the frame of interpretation into the
two sectors of actual present and just past.
Which are the next steps in the development of time concepts? It seems
that an essential growth of memory correlates with the possibility to structure
the corresponding sectors. Whether the possibility of structuring enables that
growth or whether growth creates the need for structuring may be left as an
open question. In any case, the first step seems to be a simply associative
clustering of events in the sector of memorized events, planned events, etc.
The relation of association is probably still very partial and unspecific. Neither
symmetry nor transitivity, actually none of the common relational properties

except irreflexivity can be clearly decided. At this stage prepositions and
conjunctions will be expected to appear but it is also to be expected that they
be still used in very vague ways.
Only in the next stage do asymmetric and transitive relations such as
before, after, immediately after, etc. clearly emerge and are applied in each
sector (except in the sector of actual present and in the non-temporal sector).
This is the basis for a development of the developmental stage assumed in
this article and described above.
Further development will have to introduce the temporal notions of a
temporal duration and of a time line on which the ordered sectors of each
interpretation frame may be projected and which may, in consequence, be
used to integrate the sectors of the interpretation frames into one world with
one time line - the classical notion of world and time. This is the basis for
terms referring to durations either in the absolute sense - as "in the year
1910" - or in a relative sense - as "last year," "yesterday," etc. - as well
as terms for referring to dates and clock-readings. But linguistically, expres-
sions for a reference of this type get the structure of a prepositional phrase or
clauses introduced by conjunctions, i.e. by structural means that are already
available before the specific and unique time concept has been developed.
This shows that temporal durations, dates and clock-readings are introduced
into language simply as special cases and playa secondary role in this sense.
They have semantic import but only slight structural impact.
12. I shall conclude with a very brief outline of the formal aspects of some
model that may render the essential aspects of my approach here. In order
not to complicate too much it is sufficient to consider only a language for the
analysis of talks on occasions.
The semantic network is built up as follows: First we define non-tensed
sentence radicals of occasion sentences. Secondly we define occasion descrip-
tions analogous to Carnap's state descriptions, the states being occasions
representable by a set of occasion sentences. The language in which the
occasion descriptions are defined should not contain temporal expressions.
Next we build a secondary language on top of the primary language for the
definition of occasion descriptions. The sentential expressions of the secon-
dary language are finite strings of occasion descriptions of the primary
language.1 5
112 HELMUT SCHNELLE
The finite strings of occasion descriptions are to be considered as part of

the semantic network, namely that part in which particular information,
stories, etc. can be recorded. For each state of use of the semantic network I
assume that for each sector like past, expected future, planned future, etc. a
particular finite string of occasion descriptions is marked as the "story"
actually true in that state. One of these sectors is marked as relevant in that
state.
A circumstance within a sector may be defined in the same way as one
defines an occurrence in a string. The standard way is to define it by an
initial string in a string. This is, obviously, not the only method. The final
string will also do, as well as every string before or after the occasion whose
occurrence is to be marked, provided only that it is unique in the string. Even
any unique substring of a given string together with a positional relation
appropriately defined will do. Now, I should propose to take a circumstance
as the equivalence class of all strings defining the occurrence in this way.1 6
If we proceed in this way, the factors determining the state of use of a
semantic network are themselves expressions from which the network is
composed, i.e. they are marked positions in the network, i.e. stated. Since
they are counterparts to the linguistic expressions which are interpreted
relative to a state of use of a semantic network, their marking as true or
relevant can easily be done by sentences and texts. The process of inter-
pretation and the change of context which characterize the use of expressions
as well as the result of the first process which is the meaning of an expression
can therefore be explicated in a unified way.
In the proposal so far there has not been reference to time in terms of dates
and clock-readings. These notions may be introduced in two steps. The first
step maps each occasion description in a string onto a real number represent-
ing a duration of that occasion. Moreover, it is stipulated that the durations
of occasions whose descriptions are neighbors in a string add up to their
durations. In the second step we assign time intervals on time lines to such
occasions which are not fictional but rather to be considered as having
happened. If we assume the classical non-relativistic time concept there will
be just one time line. After this has been done, we may also introduce
temporal expressions in the specific sense. But now these expressions and
their interpretation, i.e. the scientific notion of time, plays a purely secondary
role in language and linguistic understanding, as indeed it does empirically.
Ruhr Universitiit Bochum

NOTES
I.e. simple present, present progressive, present perfect, etc. For an occasion sentence
the same holds as for an observation sentence: "Commonly, an observation sentence will
cease to be an observation sentence when we change only the tense of its verb" (W.V.
Quine, J.S. Ullian (1970), p. 19).
2 Cf. Bar-Hillel (1970), p. 207 ff.
3 Cf. e.g. Reichenbach (1947), p. 287 ff.; R. Montague (974), pp. 256-257; H. Kamp
(1968), a.o.; R. Thomason (1974), pp. 36, 39; L. Aquist (1975), Cocchiarella (1966).
4 That tense cannot be appropriately analyzed without taking into accout its interplay
with adverbial clauses has been shown convincingly on the basis of syntactic arguments
by Wunderlich (1970), pp. 33 (Hyp. 8), 185 ff., 209 ff.
5 This corresponds to a relative coordinate system in the sense of Carnap, cf. e.g. Carnap
(1971), p. 72 ff. Carnap introduces a distinction between absolute and relative coordinate
systems in order to solve Goodman's puzzle of induction. My own proposals may lead to
an argument with respect to Goodman's puzzle that resembles Carnap's, the difference
being that I can give an empirical reason whereas Carnap makes a proposal.
6 A sequence of two sentences "p.q" in ordinary language may correspond to the se-
quence "p.q because of p" or "p.q after p" and a number of further determinations. Cf.
e.g. T. Van Dijk (1977), especially Ch. 8 6.
7 Even more formal analyses have been based on this concept, in particular the approa-
ches of H.C. Longuet-Higgins (972), S. Isard (1974), (1975), T. Ballmer (1974), (1976)
a.o., T. Vennemann (1975) and Smaby (forthcoming). Extensive discussions of the em-
pirical aspects may be found in T.A. van Dijk (1977).
8 This concept of a semantic network closely corresponds to Hi:/:' conception as e.g. put
forward in Hi:/: (1968), pp. 244-245, and also (1969).
9 J. Lyons (1961), p. 427, and H. Hi:/: (1969), p. 447.
10 E.g. D. Davidson's such as (1967), (1973), etc.
H A. Kasher (1974), especially 7, p. 24.
12 Cf. also S. Weinstein (1974), p. 182.
13 We are obviously at the base of an iterative system in which we may consider in the
next step the "outward" sequence of the changes of knowledge of change of knowledge
which in turn may be represented "inwardly" etc.
14 This much is even granted by Quine insofar as it can be related to some primitive
innate scheme of perceptual similarities. Cf. Quine (1974), p. 26 a.o. Contrast is the most
important factor contributing to perceptual salience. Other primitive factors determining
salience are, according to Quine, brightness, gaudy color. They, obviously, function only
if there is contrast, i.e. much less brightness or gaudy color in other regions of the visual
field or in the same region before or after the event. Motion of spots of brightness or gaudy
color combine contrast in parts of the visual field - i.e. "space" - and sequential pre-
sentations of the visual field - i.e. "time." Depending on these factors the focal position
is adjusted. (Motion and focal position are the other factors mentioned by Quine.)
15 We stipulate, however, that the strings of occasion descriptions do not contain sub-
sequences consisting of repetitions of the same occasion description only. We consider
such repetitions of occasion descriptions as describing the same occasion.
16 For more details see H. Schnelle (1976).
114 HELMUT SCHNELLE
BIBLIOGRAPHY
Aqvist, L., 1975, "Formal Semantics for Verb Tenses as Analyzed by Reichenbach," in.
Pragmatics and Poetics, T. van Dijk (ed.), Amsterdam, North Holland.
Aqvist, L., Giinthner, F., 1976, "Fundamentals of a Theory of Verb Aspect and Events
within the Setting of an Improved Tense Logic," in: Studies in Formal Semantics,
F. Giinthner and C. Rohrer (eds.), Amsterdam, North Holland.
Ballmer, T., 1974, Sprachrekonstruktionssysteme, Kronberg, Scriptor.
Ballmer, T., 1976, "Logical Language Reconstruction and Reference," Chicago Linguistic
Society 12: 33-48.
Bar-Hillel, Y., 1970, Aspects of Language, Jerusalem, Magnes Press.
Biihler, L., 1934, Sprachtheorie, Stuttgart, Fischer.
Carnap, R., 1970, "A Basic System of Inductive Logic, Part I," in: Studies in Inductive
Logic and Probability, R. Carnap and R.C. Jeffrey (eds.), Berkeley, University of
California Press.
Cocchiarella, N., 1966, "Tense Logic: A Study of Temporal Reference," Diss., University
of California at Los Angeles.
Davidson, D., 1967, "Truth and Meaning," Synthese 17: 304-323.
Davidson, D., 1970, "Semantics for Natural Languages," in: Linguaggi nella societa e nella
tecnica, Milano, Edizioni di Comunita, pp. 177-188.
Davidson, D., 1973, "In Defense of Convention T," in: Truth, Syntax and Modality,
H. Leblanc (ed.), Amsterdam, North Holland.
Hit, H., 1968, "Computable and Uncomputable Elements of Syntax," in: Logic, Metho-
dology and Philosophy of Sciences III, B. van Rootselaar and J.F. Staal (eds.), Amster-
dam, North Holland, pp. 239-254.
Hit, H., 1969, "Aletheic Semantic Theory," The Philosophical Forum 1 (N.S.): 438-451.
Isard, S., 1974, "What Would You Have Done If ... ?" Theoretical Linguistics 1: 233-255.
Isard, S., 1975, "Changing the Context," in: Formal Semantics of Natural Language,
E. Keenan (ed.), London, Cambridge University Press, pp. 287-296.
Kamp, H., 1968, "Tense Logic and the Theory of Linear Order," Diss., University of
California at Los Angeles.
Kasher, A., 1974, "Mood Implicatures: A Logical Way of Doing Generative Pragmatics,"
Theoretical Linguistics 1: 6-38.
Longuet-Higgins, H.C., 1972, "The Algorithmic Description of Natural Language,"
Proceedings of the Royal Society of London, Series B, 182: 255-276.
Lyons, J., 1968, Introduction to Theoretical Linguistics, Cambridge, Cambridge University
Press.
Montague, R., 1974, Formal Philosophy, New Haven, Yale University Press.
Quine, W. V., 1960, Word and Object, Cambridge, Mass., MIT Press.
Quine, W.V., 1974, The Roots of Reference, La Salle, Ill., Open Court.
Quine, W.V., UIlian, J.S., 1970, The Web of Belief, New York, Random House.
Reichenbach, H., 1947, Symbolic Logic, London, Macmillan, 51.
Schnelle, H., 1976, "Circumstances and Circumstantial Expressions," in: Amsterdam
Papers in Formal Grammar I, J. Groenendijk and M. Stokhof (eds.), Amsterdam,
Centrale InterfakuIteit, Universiteit van Amsterdam.
Smaby, R., Forthcoming, "Consequence, Presupposition, and Coreference under

Containers," mimeogr. Linguistics Department, University of Pennsylvania, Dec. 1974.
Thomason, R., 1974, Introduction, in: R. Montague, 1974.
Van Dijk, T.A., 1977, Text and Context, London, Longman.
Vennemann, T., 1975, "Topics, Sentence Accent, Ellipsis: A Proposal for their Formal
Treatment," in: Formal Semantics oj Natural Language, E. Keenan (ed.), London,
Cambridge University Press, pp. 313-328.
Weinstein, S., 1974, "Truth and Demonstratives," Nails 8: 179-184.
Wunderlich, D., 1970, Tempus und ZeitreJerenz im Deutschen, Munchen, Hueber Verlag.
VICTOR RASKIN
COMMENTS:
IS THERE ANYTHING NON-CIRCUMSTANTIAL?
In this paper I shall raise three questions concerning the problem itself and
the way it is presented in Schnelle's paper, as well as make two points some-
what more loosely associated with the problem and concerned with more
general issues of philosophy of language. The general goal is to demonstrate
that the problem in question is an inseparable part of a much more general
framework amounting to a full-fledged semantic theory and, therefore, an
attempt to investigate it in any partial, fragmentary, way seems to be rather
arbitrary.
Question (1). What kind of a problem is this? Or, more correctly, part of
what more general problem is the problem in question? The more general
problem is that of complete understanding of the sentence. It exists under
many guises which cover it at least partially. In Raskin (1968) I tried to cope
with the whole problem which I termed "semantic recursion." It covered
cases of calling for additional information needed to understand the sentence.
Every call for such information was triggered by a recursion signal, i.e. a
certain word or phrase of the sentence and was served by a recursion opera-
tion. The device was recursive in that the meaning of the sentence heavily
depended on the meaning of the preceding sentence calculated in exactly the
same way (cf. Raskin, 1978, Ch. 4). A better known partial approach to the
same problem is represented in fact by presuppositional analysis. In order
for the sentence to be true or false, to be understood, to be "appropriate"
or "apt," all its presuppositions should be true. The complete set of
presuppositions of the sentence would then provide all answers to all ques-
tions which could be raised with respect to the sentence; in a sense, therefore,
it would provide all the missing details. This is exactly what Schnelle's
circumstantial determination device is supposed to do. Thus, presupposition
and circumstance determination turn out to be different approaches from
different angles to the same problem. If this is, indeed, the case, then the
notion of presupposition provides a broader and better explored framework
for the problem.
Question (2). What do circumstances include? We have just spoken of all
answers to all questions. Schnelle concentrates only on time in the paper. It
116
COMMENTS 117
is obvious, though he never says so explicitly, that time is just an example of

circumstances. What is less clear is what else the domain of circumstances
includes. I get the impression, probably false and not intended by Schnelle,
that he thinks of circumstances as covering only a part of the meaning of the
sentence, and the rest is the sentence radical, the prepositional content or
whatever you call it (David Lewis, 1972, treats the meaning of the sentence in
a similar way: his indices as well as his intensions, defined as functions from
indices to extensions, cover only part of it). If this is, indeed, the impression
Schnelle intends, then my guess is that he is consciously or subconsciously
prejudiced by Quine. In Quine (1960), which Schnelle quotes, Quine says in
one place that "reports and predictions of specific single events are
eternal ... , when times, places or persons concerned are objectively indicated
rather than left to vary with the references of first names, incomplete descrip-
tions, and indicator words" (pp. 193-194). My impression is that throughout
the paper (with a somewhat illusory exception of his treatment of the nominal
phrase) Schnelle tacitly accepts that circumstances do cover times, places,
persons. I would not dare to interpret Quine here, with Professor Quine
himself so dangerously close, but I prefer to treat this triple he mentions
("times, places, and persons") as an incomplete list of more easily observable
examples rather than an exhaustive enumeration of all it takes to eternalize
a sentence. I would like to venture a claim to the contrary, that virtually
everything in the meaning of the sentence can depend on circumstances or,
putting it differently, in Quine's terms, that an eternal sentence can be dis-
eternalized at any point of its meaning including the verb, the predicate (cf.
Schnelle's example 24 which, again, he used for discussion of time only). It
is a characteristic linguistic universal that natural languages provide pronoun-
type categories for virtually every part of speech they possess. A major
consequence for the proposal Schnelle makes here is that if everything in the
meaning of the sentence can be represented as circumstantial, then the solu-
tion, the formalism, he proposes for one particular circumstance, time,
should be ready to be extended to accomodate the whole semantics and
would thus amount to a complete semantic theory. Either Schnelle is pre-
pared to present his partial approach here as a candidate for a full-fledged
semantic theory or he has to start distinguishing between different types of
circumstances.
Question (3). What is complete understanding? Do we understand a
sentence completely when we can tell whether it is true or false? Do we
understand it completely when all the circumstances are clear? Neither of
118 VICTOR RASKIN
these possible interpretations is likely to make one happy. (Lewis, 1972,

again, says explicitly that his truth-value-based intensions can capture only
part of what meaning is, p. 182.) Thus, the sentence the man had a body
would be true, though neither the exact reference nor the moment of time in
the past are known. On the other hand, we may be simply not interested in
some details and claim that we understand a sentence completely even with-
out them, e.g. He was given a watch, where we would not be interested in the
donor. Our interest would depend on topic-comment and certain other
factors; these in their turn, would depend on the context and cotext and,
ultimately, on the circumstances. What follows, rather dramatically, is that
the notion of complete understanding is itself dependent on circumstances,
and then the whole problem becomes dangerously circular.
Now, for the two more general points associated with the problem.
Schnelle himself introduces two dimensions into his paper which he treats as
marginal.
Theftrst one is linguistics. Schnelle turns to it a couple of times in order to
draw an example which will merely illustrate his point. It might have been
perfectly all right to behave in this way before linguistics started being con-
cerned with the status and validity of its own judgments, before, in other
words, it started philosophizing in a systematic-methodological way about
itself. Nowadays, an example like the tense morpheme in the early Indo-
European languages which Schnelle uses in passing would not be of any use
to a linguist, since this is not a linguistic universal, and other families of
languages may display very different, in fact, contradicting, characteristics
with respect to time. Had Schnelle been more seriously concerned with
linguistic support of this sort, he might have discovered a very significant
linguistic universal bearing out his idea that natural language is concerned
with relative rather than with absolute timing in terms of precise c1ock-
reading. I mean the so-called transposition of tenses characteristic for very
many languages, under which you can use the present or even the future tense
for a narrative in the past, once the sequence of events is determined.
It may be noted here that there is a tendency in modern linguistic theory
to treat time and tense as an underlying verb (cf. Ross, 1967 and McCawley,
1971). The sentence John had been smoking pot is analysed in the latter work,
roughly, as
COMMENTS 119
John smoke pot be have + Past
It is not my purpose here to throw this representation at Schnelle since the

frame of reference is entirely different (though, if substantiated, this analysis
would render additional support to the arguments under Question (2)
above - if time can be represented as a verb then, sure enough, predicates
can also be circumstantial). What may be relevant, however, is the way in
which this or any other proposal is justified in linguistic theory. One com-
pares it to the alternative approaches and tries to find some ground to prefer
it on the basis of its using some recommended or better-founded devices (e.g.
transformations), etc., and ultimately, it leads to certain predictions about
grammaticality, ambiguity and paraphrase relations of certain sentences of
natural language which the native speaker is expected to corroborate better
in the case of the former approach than in the case of the alternative ones.
Now, apart from being extremely vague and ill-defined as well as constantly
reformulated, each of these three criteria involves a continuous scale of
values. What is worse, however, is that the native speaker does not normally
reveal his intuition easily but rather does it under pressure. The reason, in my
opinion, is that while the linguist hopes to get a piece of pure linguistic com-
petence, the native speaker is highly prejudiced and conditioned by perfor-
mance, and what one is getting from him instead is performance. Never-
theless, in spite of these serious philosophical and methodological obstacles,
the philosopher cannot ignore the fact that the native speaker is constantly,
120 VICTOR RASKIN
systematically and repeatedly, engaged in processing semantic information,

and reveals, or can be made to reveal, certain regularities in the process. He
has this ability independently of his intellectual or educational level. This
ability is highly congruous to the conception of meaning as use. Thus, asking
relevant questions seems to have much to do with determining presupposi-
tions of an utterance; making and judging summaries may shed light on
topic and comment; I myself am processing at the moment the results of an
experiment based on the native speaker's ability to pass judgment on co-
herence of certain sequences of sentences (see Raskin, 1978, Ch. 5). Passing
verdicts on the truth or falsity of utterances is also just one such ability.
In my opinion, instead of ignoring the three linguistic cornerstones of
justification, philosophy of language should substantiate and justify the
introduction of new criteria, in addition to the old ones, e.g. such criteria as
those which have just been enumerated. This would make many philoso-
phical claims about language look more substantial than those which are
based on truth values and/or occasional appeals to intuition. An approach
to the problem in question based on some such criteria would seem better
motivated than the present one.
The second dimension Schnelle introduces purely illustratively and meta-
phorically is the automaton, the computer, and this is my last point. I think
that much more serious parallels with artificial intelligence, at least with
respect to formalism, can be discovered at present (see Raskin, 1978, Ch.
6). It is true that much of artificial intelligence still remains very much
ad hoc and prefers dealing with cleverly constructed toy systems, but the
semantic reality catches up with authors of such systems just as it did with
early mechanical translation. Those people who are interested in serious
problems such as inference-making cannot help realizing that they have to go
deeper .into semantics unless they want to face a combinatorial explosion in
their systems, and this is something they fear most. The most popular
semantic tool at the moment is the notion of frame, or script, or daemon, or
conceptual dependency, or semantic network (the latter term is used homo-
nymously with the term Schnelle uses). All these denote trees for separate
notions with nodes for everything which is relevant for the notion. All the
circumstances, including time, have their corresponding slots which are
routinely filled in with explicit and inferred information.
Thus, the sentence Napoleon suffered final defeat at Waterloo would be
analysed by Simmons (1972) in terms of a semantic network for suffer as
COMMENTS 121
Suffer
Modal Loc1 Theme Loc 2
/
Tense: Past
/
Napoleon
\ Defeat
~ Waterloo
Mood: Declarative by def at del

Essence: Positve modal Theme
Mood: NP
I
Napoleon Waterloo
Manner: Final by def at def
where the terminal nodes provide concrete information to fill in the cor-
responding slots in the frames named by the higher nodes.
Si nce for the linguist and the philosopher the computer is a very important
criterion of formality and explicitness of his proposal and since computer
science is keen on optimization of formalism, it may be highly useful for the
philosopher, at least, to have a look at the way his problem is optimally
formalized in artificial intelligence .. There are more deeply-rooted useful
parallels which I am not going to mention here.
To sum up, I suggest that the problem in question would profit from
multi-dimensional treatment in a broader and more systematic framework.
What, to my mind, Schnelle fails to do in the paper, and does not even try,
is to prove the separate existence of his problem, which cannot, as I hope
my comment demonstrates, be taken for granted.
BIBLIOGRAPHY
Lewis, D., 1972, "General Semantics," in: Semantics of Natural Language, D. Davidson,
and G. Harman (eds.), Dordrecht-Boston, Reidel, pp. 169-218.
McCawley, J.D., 1978, "Tense and Time References in English," in: Studies of Linguistic
Semantics, Ch.J. Fillmore and D.T. Langendoen (eds.), New York, Holt, Rinehart
and Winston, pp. 97-113.
Quine, V.O., 1960, Word and Object, Cambridge, Mass., M.LT. Press.
Raskin, V.V., 1968, "0 semanti~eskoj rekursii" IOn Semantic Recursionj, in: Semanti-
122 VICTOR RASKIN
ceskie i fonologiceskie problemy prikladnoj lingvistiki, V.A. Zvegincev (ed.), Moscow

University Press, pp. 268-283.
Raskin, V., 1978, "Presuppositional Analysis of Russian, I," in: Slavica Hierosolymitana
2, V. Raskin and D. Segal (eds.), Jerusalem, Magnes Press.
Ross, J.R., 1967, Constraints on Variables in Syntax, Unpubl. diss., M.I.T.
Simmons, R.F., 1973, "Semantic Networks," in: Computer Models of Though! and
Language, R.C. Schank and K.M. Colby (eds.), San Francisco, Freeman, pp. 63-113.
MICHAEL DUMMETT
WHAT DOES THE APPEAL TO USE DO

FOR THE THEORY OF MEANING?
Consider the following style of argument. What would one say, e.g., 'Either
he is your brother or he isn't,' for? Well, it is tantamount to saying, 'There
must be a definite answer: there are no two ways about it.' We say this when
someone is shilly-shallying, behaving as if it were no more right to say the
one thing than the other: so the utterance of that instance of the law of
excluded middle is an expression of the conviction that the sentence, 'He is
your brother,' has a definite sense. That, therefore, is the meaning of the
sentence, 'Either he is your brother or he isn't': that is its use in the language.
No doubt everyone here would agree that that is a bad argument: but why
is it a bad argument? A superficial answer might be, 'It does not take account
of other uses that exist for uttering an instance of the law of excluded middle,
for example in the course of a deductive argument. Thus Littlewood proved
a theorem by showing that it followed both from the Riemann hypothesis
and from the negation of that hypothesis: so his proof might have started,
"Either the Riemann hypothesis is true or it is false." , This is a superficial
answer, because, although it is quite true that people do use instances of the
law of excluded middle in this way, they might, given classical logic, per-
fectly well not do so, and still be able to carry out all the deductive arguments
that they wanted to; and yet the philosophical argument with which I
started would still be a bad argument. The following explanation of this fact
is a great improvement. The recognition of the law of excluded middle as
valid hangs together with the admission of certain forms of inference as
valid, in particular, the dilemma or argument by cases:
If A, then B If not A, then B
Therefore, B
which underlies the proof of Littlewood's that was mentioned. It hangs
together with it in the sense that any reasonable general formulation of these
rules of inference, together with a few others that strike us as inescapable,
will result in our being able to deduce each sentence of the form 'A or not A'
from no hypotheses at all (for instance, by an argument whose last step is
123
124 MICHAEL DUMMETT
the dilemma, as above, with 'B' replaced by 'A or not A'). The notion of
truth is, of course, connected with that of a valid inference by the fact that
whatever follows by valid inferences from true premisses must be true: so
we are committed, if we accept the dilemma and related forms of argument,
to regarding a sentence 'A or not A' as true. Now the meaning of a sentence
is more closely connected with what, if anything, does or would render it
true than with what would prompt an actual utterance of it. Hence, an
understanding of a sentence of that form is to be sought by explaining those
meanings of the logical constants 'or' and 'not' which permit of its derivation
from the null set of hypotheses.
The second argument does not, like the first, tamely accept that the use of a
sentence, in the sense of the point that an utterance of it might have, deter-
mines its meaning, and then claim that some such uses have been overlooked.
Rather, it challenges that principle by giving reasons for thinking that we
must have a prior understanding of the sentence before we can be in a position
to ask what the point ofa particular utterance of it may be. The argument, as
stated, appeals to an already understood notion of truth, with a known con-
nection with our recognition of any given principle of inference. The pro-
ponent of the argument that is being criticised may feel that such a notio,n of
truth is spurious, and has available a well-known device for countering an
appeal to it: he declares that the whole explanation of 'true,' in its only
intelligible sense, is given by the principle that 'A' is equivalent to 'It is true
that A,' or by a definition that is just sufficient to yield this equivalence for
each case; the use, or meaning, of an assertion that a sentence is true will
then be precisely the same as that of an utterance of that sentence, and the
notion of truth will be impotent to yield any results about meaning not
previously obtained by inquiring into use. But the sense of 'true' required for
the second counter-argument is shown by that argument itself: what is
needed of a true sentence is that there should exist means of justifying an
assertion of it of a kind we are accustomed to accept elsewhere; and so the
word 'true' can be dropped from the argument, and a direct appeal made to
this notion. This is, indeed, to assume that we recognise certain general
principles for the justification of our assertions; but so we obviously do,
otherwise there would be no such thing as deductive argument. It is now open
to the proponent of the second counter-argument to concede that, given
classical logic, an instance of the law of excluded middle is so obvious that
the point of an assertion of it is scarcely ever to call attention to the fact that
it can be justified, without calling in question his own thesis that it is the
THE APPEAL TO USE AND THE THEORY OF MEANING 125
possibility of justification to which our primary understanding of the sentence

relates; indeed, he can even maintain that 'A or not A' would be true in
some, or all, cases in which the sense of 'A' is not definite, so that an assertion
of it, intended to have the point which the proponent of the original argument
rightly said that such assertions often have, would go awry otherwise than
by failing to be true.
Now I do not for one moment suggest that an argument of the style with
which we started out and with which some of us were made, during a period
now passed, wearisomely familiar, represents faithfully the notion of use
which Wittgenstein had in mind when he coined the slogan, 'Meaning is use.'
On the contrary, Wittgenstein's notion was a much more general one: it
comprised anything that could be counted as belonging to the role of the
sentence in the language-game, which certainly included, not only the com-
municatory function of an utterance of the sentence itself, but also that of an
utterance of a complex sentence of which it was a constituent and, as well,
such other features as were appealed to in the counter-argument I set out. To
say this, however, is to make the conception of meaning as use totally pro-
grammatic: any feature of our linguistic practice that relates to the sentence
may be cited as bearing on its meaning. There is, however, a reason why
Wittgenstein's later philosophy of language should have led to this mis-
application of the identification of meaning with use. This lies in Wittgen-
stein's repudiation of the Fregean distinction between sense and force and,
particularly, of Frege's idea that there is such a thing as assertoric force in
general. There are, on this conception of Frege's, three grades in under-
standing an assertoric utterance. First comes the grasp we have of the sense
of the sentence, of the thought expressed; and this consists in an under-
standing, which is derived in accordance with our apprehension of the con-
struction of the sentence out of its linguistic elements (let us say, inexactly,
its component words), of the condition which must obtain for the sentence
to be true. Secondly, there is our knowledge of the practice of assertion: the
speaker is not merely uttering a sentence with which is associated a certain

truth-condition, that is, expressing a certain thought, but saying (that is,
asserting) that that thought is true, as opposed to asking whether it is true,
supposing it to be true for the purposes of argument, declaring himself un-
willing to deny that it is true, advising his hearer to make it true, expressing
the wish that it were true, or the like. (Whether or not there is a non-circular
account of what it is to assert that a thought is true, that is, of what is effected
by an assertoric utterance of a sentence expressing that thought, is another
126 MICHAEL DUMMETT
matter.) And, finally, there is the divination of the speaker's particular inten-
tion in asserting that thought to be true on that particular occasion. Wittgen-
stein rejected this conception, on the ground that there is no such thing as
'the practice of assertion,' or as, in his terminology, the language-game of
assertion, considered as effected by the utterance, in assertoric mode, of any
sentence syntactically fitted to be used assertorically and to which we may
ascribe conditions for it (or a specific utterance of it) to be true or false.
Something happens - and then I make a noise. What for? Presumably in order to tell
what happens. - But how is telling done? When are we said to tell anything? - What is
the language-game of telling? - I should like to say: you regard it much too much as a
matter of course that one can tell anything to anyone. That is to say: we are so much
accustomed to communication [Mitteilung - the abstract noun cognate with the verb
used for 'to tell'] through language, in conversation, that it looks to us as if the whole
point of communication lay in this: someone else grasps the sense of my words - which
is something mental: he as it were takes it into his own mind. If he then does something
further with it as well, that is no part of the immediate purpose of language (Philosophical
Investigations, I: 363).
Now the fact is that it is difficult to obliterate the distinction between the
first grade of understanding and the second without thereby also obliterating
that between the second and the third. This is because our concept of truth
gets a large part of its point from the contrast that we wish to draw between
a statement's being true and any more primitive, or at least undifferentiated,
conception of its being appropriate: for instance, between its being true and
the speaker's having a sufficient warrant to take it as true, or between its being
true and the intention that the speaker had in asserting it to be true just then
being a just one, his having had a legitimate point in making it. Of course,
once we have any given conception of a particular sentence's being deter-
mined, in some objective manner, as true or as false, then these distinctions
arise naturally, indeed inevitably: the questions of interest are why we
introduce the notion of truth at all, and why, in doing so, we draw the line
between the condition for a statement to be true and the condition for a
speaker's being in the right in making it in these more general ways at just
the place we do, and not somewhere else. There are various correct answers
to these questions, one being the necessity of explaining the role of the
sentence when it figures as a constituent in more complex sentences; but this
is not our present concern. Another partial answer is, obviously, the depen-
dence of a speaker's point in making a statement on the context, something
which, if we are to attain a conception of the meaning of the sentence as a
type, we must either filter out or reduce to a definite rule (as we can explain
indexical expressions systematically). But this is not an important feature of
the objection to the account with which I started of sentences like 'Either he
is your brother or he isn't,' an utterance of which was claimed as only ever
having one kind of point; as I remarked, the account would be wrong even if
the claim were sound. Rather, in that case, what we appealed to was the
existence in the practice of the speakers of certain generally accepted proce-
dures for justifying statements, procedures which would always yield a justi-
fication of an instance of the law of excluded middle even if it was never in
fact invoked in such a case. This looks circular, since such procedures are for
the purpose of justifying a statement as true, rather than as making a sound
point. So it comes down to this: that our linguistic practice - the language-
games in which we participate - involves the process whereby those utter-
ances which we call assertions (and perhaps some others) are subject to
challenge by our hearers and the process of responding to such challenges;
and, if we were to try to give any account of these practices, a mastery of
which is certainly essential for the ability to engage in converse with others,
we shall be forced to distinguish between different types of such challenges,
according to the kind of response that is appropriate; and among these
are challenges as to the truth of what is said and challenges as to its point
(the latter of which doubtless further subdivide into challenges as to rele-
vance, as to implicature in Grice's sense, etc.). Here a challenge as to truth
is to be distinguished by the fact that, if successfully met, the challenger will
himself give assent to the statement (though he need not be prepared himself
to make that statement, since it may be objectionable in other ways, e.g. as
breaking a confidence or being insulting); hence Quine's properly placed em-
phasis upon the notions of assent and dissent. (A suspicion of circularity
arises here also, since an expression of assent is surely an expression of a
willingness to make the statement so far as its truth is concerned, that is, but
for possible objections to it which are not objections as to truth; but I will
not push the inquiry further. Of course, as I said, these distinctions are easy
to draw once we have the notion of truth and know its application to a given
sentence; but I have been concerned with what we need the notion for and
why we give it the application that we do.) This is not to say that the notion
of truth so arrived at will serve all the purposes for which we need it, for
instance to explain the behaviour ofa sentence when it is a constituent of a
more complex sentence.
Once we have the notion of truth and so can distinguish between the
128 MICHAEL DUMMETT
second and the third grade of understanding, the distinction between the first
and second grade is all but inevitable. If all utterances were assertoric, and
no sentence ever occurred as a constituent of another sentence, there would
indeed be no place for it, but this would make no difference to the present
argument: so long as we appealed to the notion of the truth-conditions of a
sentence as determining its particular content, we should, in explaining what
is effected by an utterance of a sentence, have to give a general description of
the linguistic practice of making assertions. But the notion of truth is precisely
what we need, or, rather, what is forced on us, if we wish to distinguish
between the second and third grades of understanding. Hence, if there is no
such thing as the general practice of making assertions - or, at least, as a
uniform description of what this practice consists in, for a sentence with an
arbitrarily given individual content - then there can be no distinction, at any
rate no general distinction, between the second and third grades of under-
standing either. And what this appears to mean is that any account of the
meaning of a given sentence must simultaneously explain every feature of the
significance of any possible utterance of it. This, indeed, is not particularly
difficult to do for anyone sentence, at least if we ignore utterances the point
of which is heavily context-dependent. What seems impossibly hard is to
construct a systematic theory of meaning for a language along these lines,
that is, one which would show the derivation of the significance of the sentence
in accordance with its composition: as soon as we begin to think about the
construction of such a theory, we at once start to segment the task it has to
accomplish, along the lines of the repudiated distinctions between truth-
conditions, force and point. Wittgenstein's repudiation of these distinctions
is expressed by his adherence to the redundancy theory of truth (expressed in
a characteristically sloppy manner in the Remarks on the Foundations of
Mathematics, I, App. I: 6 - "For what does a proposition's 'being true'
mean? 'p' is true =p. (That is the answer.)"). If the equivalence of' "Snow
is white" is true' with 'Snow is white,' and so on, constitutes the whole
explanation of the concept of truth, then the concept is useless in giving a
theory of meaning. It is because of his rejection of the concepts of assertion
and of truth as capable of playing any role in an account of how language
functions that Wittgenstein's identification of meaning with use lent itself to
the kind of misapplication with which I began.
In an earlier period, however, Wittgenstein had seized on the notion of the
justification of a statement as the key to an explanation of sense: "It is what
is regarded as the justification of an assertion that constitutes the sense of the
assertion" (Philosophical Grammar, I: 40).

It is natural to contrast the idea that meaning is given by determining what
justifies us in using a sentence to make an assertion with the idea that it is
given by determining the conditions under which the sentence is true; and
Wittgenstein certainly meant to convey a very sharp divergence from the
theory of meaning in terms of truth-conditions set out in the Tractatus. But
to say that, on the view of meaning Wittgenstein held in the intermediate
period, the meaning of a sentence is not determined by its truth-conditions,
is liable to misconstruction. If one holds that the meaning of a sentence is
given in terms of what has to hold for it to be true, it is open to one to say
that someone may understand a sentence although he has not yet learned by
what means we may recogmse it as true, nor, therefore, what justifies an
assertion of it. But there is an asymmetry here: thinking that the meaning of
a sentence is given by what justifies asserting it does not entitle one to suggest
that someone might understand a sentence without yet knowing the condition
that must obtain for it to be true. Rather, from such a standpoint one would
say that the only legitimate notion of truth is one that is to be explained in
terms of what justifies an assertion: a sentence is true if an assertion made by
means of it would be justified. (Or, possibly, if there is some recognisable
state of affairs such that, if a speaker knew of it, he would be justified in
making that assertion.) There is therefore a sense in which, even on the theory
of meaning which is opposed to that of the Tractatus, it remains the case that
the sense of a sentence is determined by its truth-conditions: the question is
what is the relation between the notion of truth and that of the justification
of an assertion.
The theory of meaning expressed by the remark I have quoted from the
Philosophical Grammar stands in opposition to any conception of meaning
under which the sense of a sentence is given in terms of a notion of truth taken
as objectively and determinately either attaching or not attaching to each
sentence independently of our knowledge, or capacity to know, whether or
not that sentence is true, a notion of truth which is therefore taken to be
grasped without reference, in all cases, to the means available to us for judg-
ing a sentence to be true. This conception plainly informs the Tractatus. That
work carries a fundamental commitment to the principle of bivalence, in-
tegral to the conception of meaning I have just characterised, since, if
bivalence did not hold, the truth-tables would not have the kind of importance
allotted to them in the Tractatus. Since the Tractatus also contends that our
understanding of our sentences involves the grasp of infinitary truth-
130 MICHAEL DUMMETT
operations, it is equally plain that a grasp of the meaning of a sentence is not

held to be in all cases related to, or given in terms of, the means available to
us for recognising it as true. Wittgenstein came to repudiate this conception
for a multitude of reasons. First, a theory of meaning of this kind is powerless
to explain how we come by our knowledge of the conditions which warrant
us in asserting a statement, the means by which we can establish a statement
as true. Granted that the meaning of a sentence is not, in the first place, given
in terms of how we recognise it as true, still our grasp of what is to count for
us as showing that it is true must be derived in some way from our knowledge
of its meaning: for, if not, then, even when the meaning of a sentence has
been fixed by determining its truth-conditions, there will still remain room
for decision as to what we shall choose to count as showing it to be true, and
this is counter-intuitive. But, once we have allowed the two notions, that of
truth and that of the means by which truth is recognised, to be sundered at
the outset, we shall never find a means to connect them up again, to explain
how the one is derived from the other. More generally. the theory violates
the intuitive connection between meaning and knowledge: two sentences may,
according to the Tractatus theory, express the same sense (because they make
the same division in logical space) without our perceiving that their senses
are the same; this is because sense has been thought of as given in terms of
what makes a sentence true, rather than in terms of how it is recognised as
true. But, on the contrary, we ought to say that the meaning of any expres-
sion is determined by what a speaker must know if he is to be said to under-
stand that expression; it follows that, if someone understands two expressions
that have the same meaning, he must know that their meaning is the same.
(It could be argued - and has been argued by me - that this is in part a
consequence of Wittgenstein's abandonment of the distinction between sense
and reference as drawn by Frege, and therefore not a necessary result of
holding a conception of meaning as given by truth-conditions, under a notion
of truth subject to bivalence and not directly connected, in all cases, with our
means of recognising truth. That is not to say that the more restlicted form
of the objection, that such a conception of meaning will not always allow an
explanation of how we derive the means of recognising the truth of a sentence
from the condition for it to be true, is met by appealing to a Fregean distinc-
tion between sense and reference.)
Secondly, the Tractatus conception leaves us unable to state informatively
the conditions for the truth of many of our sentences: an essential circularity
appears in any attempt to do this, a circulality which does not appear in a
characterisation of what justifies us in asserting a sentence. This circularity

then leads us to attribute to a speaker a capacity for immediate recognition of
certain qualities, objects, processes or states (the private ostensive definition),
which capacity can be no further explained. But now it is evident that this
attribution is idle: all would go on in just the same way if the speaker mis-
recognised the entity every time, or if there were nothing there to be recog-
nised; at least, It would do so provided that, whenever recognition by two or
more speakers was called for, they tended to make the same mistakes. The
conception of our apprehension of the truth-condition of a sentence, with its
attendant capacity for immediate recognition of the presence of the referents
of certain terms, therefore fails to be explanatory: what, ultimately, actually
justifies us in the assertions we make is the fact of agreement between
speakers, which need not be taken as resting on anything more basic. As for
the notion of truth, the circularity disappears when we cease to think of it,
as the Tractatus insists, as that in terms of which the meanings of our
sentences are given, and, instead, regard its application to a sentence as
explicable only after the sense of that sentence is known.
Thirdly, this same circularity attends our atteinpts to state the truth-
conditions of those sentences our alleged grasp of which transcends our
means of recognising them as true: our grasp of the condition for such a
sentence to be true cannot consist in our ability, in certain special cases, to
recognise it as true, just because it involves our awareness that it may still be
true even when we are unable to recognise it as such. But then to attribute to
us a grasp of the condition for a sentence to be true, under such a transcen-
dental notion of truth, violates the principle that meaning is use: for a
knowledge of the condition for the truth of the sentence of this kind cannot
be fully manifested by the use the speaker makes of it, that is, by the linguistic
and non-linguistic behaviour on his part that is connected with the utterance
of the sentence.
These arguments are all of a negative kind: in so fal as they are cogent,
they show the conception of meaning as given by truth-conditions, as found
in Frege, and, in a different form, in the Tractatus, to be inadmissible; but
they do not show the conception of it as given in terms of what justifies an
assertion, rather than of some other feature of the use of a sentence, to be
correct. In so far as they are merely negative arguments, they survive into
Wittgenstein's later period: but I have suggested that he came to adopt a
still more radical view, one involving a repudiation of the sense/force dis-
tinction in a way in which the idea of meaning as given in terms of the justifi-
132 MICHAEL DUMMETT
cation of an assertion does not. (Indeed, it is of importance that the formula-

tion I quoted from the Philosophical Grammar employs the notion of
assertion.) I cannot here attempt an evaluation of the negative arguments I
have just sketched. Instead, I want to argue that the thesis that what con-
stitutes a justification of an assertion determines, or shows, the sense of the
sentence asserted indicates very precisely the constraints which the identifica-
tion of meaning with use puts, and those it does not put, upon a theory of
meaning.
First, the constraints it does not impose. If someone has the idea that the
justification for an assertion is the key to its content, he is allowing himself
a much richer set of data to which to appeal than those which either Quine
or Davidson permits himself. For Quine, the relevant data consist solely of
the correlations between the sensory stimuli to which a speaker is sUbjected
and his readiness to assent to or dissent from a sentence. Davidson is more
generous in allowing correlations between a speaker's holding a sentence true
and prevailing conditions of any kind. Both, however, propose to construct
a translation manual or meaning-theory by appeal solely to data of the form
of answers to the question, 'When do the speakers hold, or acknowledge,
sentences as true or as false?' The reason for this limitation is made very
explicit by Quine, less so by Davidson. For both, the problem is of construct-
ing a translation-manual, or meaning-theory, for a language whose speakers
one may observe and perhaps even interact with, but which is previously
quite unknown. All that one has to go on is what one can see and hear ofthe
speakers' utterances and associated behaviour. Now perhaps one can make
a plausible identification, in behavioural terms, of the speakers' mode of
expressing assent and dissent. But to go any further would be illegitimate.
The notion of justification, as used by Wittgenstein, does not refer only to
the process of justifying an assertion, when challenged, as this takes place
within the language, since various things are involved in a full account of
what justifies certain assertions that the speakers tacitly take for granted and
would never explicitly cite. Nevertheless, it does involve all that would be
appealed to in justifying an assertion, as this occurs between speakers. But,
from the standpoint of Quine and Davidson, if I have understood them
aright, to appeal to such a complicated thing as the justifications which
speakers give of their assertions is out of the question, since, to become
aware of that, or even to recognise what constituted a demand for justifica-
tion and what a response to it, one would have already to understand a large
part of the language.
Philosophers, unlike historians, do not have to solve problems that are

clearly demarcated in advance; and so they make up their own problems -
set themselves tasks, and then try to perform them. Disputes over philoso-
phical methodology are largely about which are the right problems to set.
One can hardly prove that this or that is the right problem: that would be
possible only if, behind the problems philosophers try to solve, lay further
clearly defined problems, and the solution to the former were a means to the
solution of the latter. The question is only the vague one: by solving which
problems shall we gain philosophical illumination? Now what is the point of
posing the problem: how we should arrive at an interpretation of a language
hitherto quite unknown to us? It is, surely, to exclude from the description of
the interpretation or of the process of arriving at it any appeals to concepts
which covertly presuppose an understanding of the language. But the con-
sequence of so posing the problem is that we fasten on some feature of the
speakers' linguistic behaviour which can be described at the outset, before
any understanding of the language has been gained, and try to use it as the
basis for the entire interpretation. Language is, however, an enormously
complicated thing, and it is highly unlikely that a satisfactory interpretation
of it is accessible if we so restrict ourselves. Certainly our actual acquisition
of our mother-tongue proceeds by stages: some features of our linguistic
practice can be mastered only after others have already been mastered.
We are, at any rate, not now in the position of having to interpret some
radically foreign language: that is a practical problem, the solution to
which is not obviously necessary or sufficient for the kind of understanding
of how language functions which we, as philosophers, wish to attain. We
already have, in our language, expressions for various concepts which relate
to our use of language, among them that of the justification of an assertion.
What we want to arrive at is a model of that in which our understanding of
our language consists, a model which will be adequate to explain the entire
practice of speaking the language. Certainly that model must itself be des-
cribed in terms which do not presuppose a tacit understanding of terms, such
as 'assertion,' 'justification,' 'true,' etc., which relate to the practice of which
the model aims to provide an account, or it will, to that extent, fail to be
explanatory. But that does not mean that, in groping our way towards such
a model, we must eschew appeal to any ofthose concepts which are not to be
used in giving the model itself. It does not matter whether or not an outside
observer - a Martian, say, who communicated by means so different from
our own that he would not for a long time recognise human language as a
134 MICHAEL DUMMETT
medium of communication - could ever arrive at the model we hope to give:

all that matters is whether, once he had it, it would serve to make our
language intelligible to him.
Now for the constraints which the identification of meaning with use does
impose. Having expressed a point of basic methodological disagreement with
Quine, let me now record one of strong agreement with him. In his lecture
"Mind and Verbal Dispositions" (Mind and Language, ed. S. Guttenplan,
Oxford, 1975, pp. 83-96), Quine says, "when I define the understanding of a
sentence as knowledge of its truth conditions I am certainly not offering a
definition to rest with; my term 'knowledge' is as poor a resting-point as the
term 'understanding' itself," and goes on to ask, "In what behavioural dis-
disposition. .. does a man's knowledge of the truth conditions of the
sentence ... consist?" This is in full consonance with what I have myself
repeatedly insisted on, that a meaning-theory, being a theoretical representa-
tion of a practical ability, must not only say what a speaker must know in
order to know the language, but in what his having that knowledge consists,
that is, what constitutes a manifestation of it. (I disagree with Quine only if,
as I suspect, he wants to eliminate the notion of knowledge from the theory
of meaning altogether.) But, now, this requirement calls in question the
feasibility of any model of understanding, any theory of meaning, according
to which the understanding of a sentence consists, in general, of a knowledge
of its truth-conditions, when the notion of truth is construed as satisfying the
principle of bivalence and as, in general, given independently of our means of
recognising truth. Of the three arguments which I cited as contained in middle
and late Wittgenstein, it is the third I wish to stress here. Our language con-
tains many sentences for which we know no procedure, even in principle,
which will put us in a position to assert or deny that sentence, at least with
full justification. Indeed, for many such sentences, we have no ground for
supposing that there necessarily exists any means whereby we could recognise
the sentence as true or as false, even means of which we have no effective
method of availing ourselves. Hence a notion of truth for such a sentence,
taken as subject to the principle of bivalence, cannot be equated with the
existence of a means of justifying an assertion of it - an equation which
sufficed for the distinction between the meaning of a sentence, as a type, and
the point of a particular assertion of it. More importantly, a speaker's
knowledge of the condition which must, in general, hold for the sentence to
be true cannot be taken to consist in his ability to recognise it as true when-
ever those conditions obtain under which it may be so recognised, and as
false when it may be recognised as false, since, by hypothesis, it may be true

even in the absence of any such conditions, and he must know the condition
for it to be true in those cases also. Therefore, if meaning is use, that is, if the
knowledge in which a speaker's understanding of a sentence consists must be
capable of being fully manifested by his linguistic practice, it appears that a
model of meaning in terms of a knowledge of truth-conditions is possible
only if we construe truth in such a way that the principle of bivalence fails;
and this means, in effect, some notion of truth under which the truth of a
sentence implies the possibility, in principle, of our recognising its truth. It is
hard to swallow such a conclusion, because it has profound metaphysical
repercussions: it means that we cannot operate, in general, with a picture of
our language as bearing a sense that enables us to talk about a determinate,
objective reality which renders what we say determinately true or false
independently of whether we have the means to recognise its truth or falsity.
On the other hand, if the identification of meaning with use does not impose
on a theory of meaning the constraints I have suggested, I for one find it
difficult to see how it can impose any constraints whatever.
All Souls College

Oxford
EDNA ULLMANN-MARGALIT
COMMENTS
I hope I am not alone here in feeling that, whether or not Professor

Dummett's very rich paper was hard to swalIow, it was not alI that easy to
folIow. It is for this reason, and also for the fact that I am far from able to
join the battle of the giants that is taking place here, that I have decided to
be a glossator rather than a commentator.
Thus, I shaII content myself with the hope that the schematic guide to the
Dummett labyrinth I shaII proceed to offer will be just about accurate enough
to facilitate the discussion, not to hinder it.
Let me begin by drawing a chart which represents, I believe, the basic
ingredients of Dummett's position - indeed, program - as well as its central
theses.
The ingredients:
I
assertion of sentence SENTENCE I meaning is use
justification of the sense (content)

theory of meaning
assertion of sentence of sentence
The theses:
Claim I: What constitutes a justification of a sentence-assertion
"determines," "shows," or "is the key to" the sense of
the sentence asserted.
Claim II: The identification of meaning with use puts constraints
upon an adequate theory of meaning. (Also, there are
constraints it does not put.)
Claim III: The thesis of claim I (that what constitutes ajustification ...)
indicates "very precisely" the constraints of claim II.
The fuII chart, then, is this:
136
A. Marga!it (ed.), Meaning and Use, 136-140. All Rights Reserved.
COMMENTS 137
assertion of sentence SENTENCE meaning is use

I
-+f nconstraints I
( I.
what constItutes a )- Il"d ."
etermones,
:f
Isense (content)
~
(any adeq uate)
.
justification of the
assertIon of sentence
T".
"shows",
k "
....... t
0 sen ence
L-_Is_ety=to=:::~_ _~r--_--=-_-'
theory of meaning
m"indicates" ~
Now behind this chart, or above it, or anyway somewhere, hovers the
notion of truth: to justify an assertion is to justify it as true; the sense of a
sentence is inextricably involved with its truth conditions; any adequate
theory of meaning has to make the notion of truth a central notion (whether
in the weak or in the strong sense, as suggested and discussed by Dummett
elsewhere).!
If I am right in taking these to be the key notions and theses of the paper,
one would naturally expect to find in it answers to at least the following five
questions (I shall start with the right-hand side of the chart):
(1) In what sense is meaning identified with use?
(2) What is expected of a theory of meaning?
(3) What are the constraints upon the meaning theory imposed by
the identification of meaning with use?
(4) How does the justification of an assertion provide the key to
its content?
and, finally, of course
(5) How does the relation of "indication" between the answers to
the last two questions come about? That is, in what way does
the thesis that the justification for an assertion provides the key
to its content indicate the constraints upon the meaning theory
imposed by the identification of meaning with use?
I shall now proceed to elicit - briefly and schematically - what seems to
me to be the core of Dummett's position on these questions, Let me turn,
then, to the first three questions, relating to the right-hand side of the
diagram, and begin with the meaning-use identification.
In the first part of his paper Dummett rejects as a distortion ofWittgenstein
the interpretation according to which "the use of a sentence, in the sense of
the point that an utterance of it might have, determines its meaning." The
138 EDNA ULLMANN-MARGALIT
interpretation he does endorse of the thesis that meaning is use is that "the
knowledge in which a speaker's understanding of a sentence consists must be
capable of being fully manifested by his linguistic practice."
This formulation at once gives vent to Dummett's cognitivist conviction
that to understand the meaning of a sentence is to have knowledge of some
kind, and entails the requirements from (or, in his terminology, the con-
straints upon) any adequate theory of meaning. To wit, an adequate meaning-
theory, if I understand Dummett correctly, must essentially be comprised of
two components. First, the cognitive component: the theory must tell us
what one has to know in order to qualify as knowing the language; second,
the behavioral component: the theory must also tell us what constitutes the
overt manifestations of this cognitive knowledge of the language.
I wonder, at this point, if Dummett could elaborate on this; it is somewhat
mysterious to me what for him are overt manifestations of the knowledge of
the language, since I take it that he is not willing to accept a purely behavior-
istic account of them in terms of bodily responses to stimulation of re-
ceptors - and I am curious if he can do it in a way which does not hinge
on the notion of knowledge itself, thereby rendering the enterprise circular.
Be that as it may, it is here that I take Dummett to be sending arrows in
the direction of Davidson, as choosing to ignore the second (behavioral)
component, and against Quine (somewhat more hesitantly, I feel), as choos-
ing to ignore the first (cognitive) one. 2 As for both Davidson and Quine
lumped together, Dummett essentially charges (if I understand him correctly)
that they both choose. to operate within the context of discovery, which he
believes may ultimately prove barren, whereas he himself sees the promise of
philosophical illumination within the context of justification. Thus, he says
in this connection: "It does not matter whether or not an outside observer-
a Martian ... could ever arrive at the model we hope to give: all that
matters is whether, once he had it, it would serve to make our language
intelligible to him."
Now let us go back to the left-hand side of the chart, and consider the
claim that the meaning, or the sense, of a sentence is determined by what
constitutes a justification of its assertion.
This claim is contrasted with the central thesis of the early, the Tractatus,
Wittgenstein, according to which the meaning of a sentence is given by the
conditions that have to obtain for it to be true (or false). For momentary
convenience let us agree to refer to them as the justification and the truth-
conditions theses, respectively. However, as Dummett emphasizes, although
COMMENTS 139
in contrast, these two theses are not exclusive or incompatible or inde-

pendent. Indeed the first presupposes the second, in the following sense:
that for the proponent of the justification thesis, an understanding of the
sentence cannot be had without knowledge of its truth/falsity conditions.
But - and this I believe is the crux of the matter - for Dummett the
notion of truth is other than a realist one (I hesitate to say that it is a
straightforward idealist one, but it clearly is non-realist). We are not to
assume "a determinate, objective reality" which determines the truth or
falsity of our sentences independently of whether we have the means to
recognize - and hence to justify - them as true or false. Rather, the only
legitimate notion of truth, for Dummett, is one that involves no commitment
to classical logic or to the principle of bivalence, and which is inextricable
from that of an assertion-justification. The key quotation here is that "a
sentence is true if an assertion made by means of it would be justified."
Thus, Dummett's conception of the notion of truth lies after all at the root
of it all, rather than merely hovers from above. From it flows what will
constitute the sense, or the meaning, of a sentence, and from this his con-
ception of what is required of an adequate theory of meaning.
Finally, let me just say this. Being programmatic, Dummett's paper should
of course not be expected to spell out everything in detail. However, the
notion of truth so clearly constituting the underlying stratum of all the rest,
I for one would like to know more about it. By this I do not just mean to
throw a 'could-you-please-elaborate' operator at Dummett. Rather, my
point is that the price for abandoning the theories of 11 uth which are currently
on the market and which on the whole have not done so badly, is relatively
high. And it seems to me worthwhile to have in advance a somewhat more
detailed and algued-for idea as to what the alternative expenditure would
involve.
We are told that Dummett's notion of truth involves no commitment to
classical logic. An informed guess of mine is that what Dummett has in mind
is at least the possibility of basing the notion of truth on something like
intuitionistic logic, and consequently explicating it in terms of some method of
verification. Let me point out in passing, though, that - perhaps curiously -
the term 'verification' is mentioned not even once in this paper, nor are, for
that matter, the terms 'evidence' and 'reference.' Their absence, I suspect,
may be roaring, at least to some of you.
Now if this conjecture is anywhere near the mark, and to know the mean-
ing of a sentence is to know that such-and-such a method of verification is
140 EDNA ULLMANN-MARGALIT
associated with it, and some set of recursive rules is presumably provided for
taking care of complex sentences, then I would like to know more about these
rules: Where do they begin? That is, I would like to ask Dummett to indicate
what are the "primitive" sentences in his system, what will count as the
justification of their assertion, and why.
NOTES
1 In his William James lectures series, Harvard, 1976.

2 See, however, Dummett's 1974 paper "Postscript" (Synthese 27: 523-534) where he
takes Quine's position to resemble Frege's (cognitivist) one, in that for both the idea that
meaning has to do with knowledge is "a regulative principle for the theory of meaning"
(p.524).
AVISHAI MARGALIT
OPEN TEXTURE
1. The concept of open texture was invoked by Waismann to revoke

verificationism.1 Empirical terms, so his argument goes, are essentially
open-ended: no set of rules can determine their application for all possible
situations. The reason is that we cannot envisage all possible states of affairs
in which such application might take place. Thus, empirical sentences which
employ such terms cannot be verified conclusively, and hence complete veri-
fiability cannot be taken as an adequate criterion for meaningfulness.
Complete verificationism is perhaps dead. 2 But the notion of open texture
is very much alive and kicking - so much so that it kicks against a res-
pectable, ongoing concern, namely the doctrine of possible worlds semantics,
which holds that the meaning of a sentence is its truth conditions for all
possible worlds. But before we assign the notion of open texture to this
important philosophical task let us check its credentials.
2. Waismann is in the grip of a spatial picture with regard to both open

texture and vagueness. Vagueness is presented as the blurred zone between
two distinct areas, while the dividing line of a clear-cut concept is sharply
delineated. Open texture, on the other hand, is distinguished by the fact that
is leaves 'gaps' in various 'directions.'3
An improvement on Waismann's theme has been suggested by Rolston. 4
Whereas vagueness, for him, is represented on two dimensional space, open
texture is represented on a multi-dimensional space. Even if a concept is
distinctly delineated on two dimensions, it can still be open 'from above,' so
to speak. That is, the 'directions' in Waismann's picture become 'dimensions'
in Rolston's and the 'gaps' are transformed into unboundedness in some of
these dimensions.
I suggest to replace the spatial picture, in whatever version, by a picture of
possible worlds: a 'direction' /,dimension' is to be replaced by a possible
world. To say that a term is open textured is to point out that there are
possible worlds where the application of this term is indefinite. To say that
all empirical terms are open textured is to claim that the truth value of each
empirical statement is indeterminate in some possible world. Let us dis-
141
142 A VISHAI MARGALIT
tinguish between the case in which the application of a term is indecisive and
the case in which it is indeterminate. In the first case the blame lies with the
applier of the term, in the second with the term applied. Put differently,
indecisiveness is a pragmatic notion whereas indeterminacy is a semantic
one.
Now for Waismann vagueness is a case of indecisiveness which can be
lemedied by fiat. s Not so the case of open texture, which is a genuine case of
indeterminacy. In terms of the spatial picture this means that we can de-
lineate a boundary in some directions, but not in all: gaps are inevitable. This
reflects the fact that we cannot take into account ahead oftime all the possible
situations in which our words might be put into test.
The metaphors of 'directions,' 'gaps' and 'dividing lines' provide one
"explanation" of open texture. Yet a better one can be extracted from
Waismann's stock of examples. They include in the first place ordinary terms
like 'cat,' in an envisaged situation in which a cat could be revived from death;
'gold,' in an envisaged situation where gold emits a new sort of radiation;
'man,' when confronting a Methuselah who remembers King Darius and has
claims for immortality. That is, these examples pose the question of whether a
natural-kind term is applicable to an object with some unnatural, or extra-
ordinary, features, where these features constitute prima facie stigmas - i.e ..
seem to disqualify such application. Admittedly, Waismann's set of examples
is not ideal for driving this point home, but that, I believe, is what he has in
mind.
Waismann's examples, however, also include the epithet 'intelligent.' This
is so, apparently, because we cannot specify ahead of time all that is included
in the behavior of an intelligent person. Now, there are many senses in which
a dispositional term such as 'intelligent' might be taken as open-ended. I want
to claim, however, that they are not of the open texture variety. For consider:
there is a relative sense in which intelligent, say, is more open-ended than
'alcoholist.' 'Alcoholist' is more of a determinate term, whereas 'intelligent'
is determinable: there is basically one sort of behavior that is expected from
someone who is referred to by the former, and many different sorts of be-
havior in the case of the latter.6 Then there is a further sense in which all dis-
positional terms are open-ended, due to the fact that they are not definable
by eliminative definitions (which tell us how to avoid a term in all contexts).
The way in which we may adduce more and more reduction sentences for the
introduction of such terms manifests this sense of open-endedness. 7
In short, I shall distinguish between open-endedness and open texture. I
OPEN TEXTURE 143
shall take the first as the generic term, and reserve the term 'open texture' to
cover those cases in which some of our hard core beliefs are assumed to be
violated. To be sure, this might happen with 'intelligent' too. Balaam, as you
may remember, was reproachfully asked by his ass why he has beaten it.
Apparently finding nothing peculiar about a talking ass, the soothsayer
replied offhand that he would have killed the ass had he had a sword in his
hand. Did Balaam manifest intelligent behavior, or did he rather become a
foolish man (,ben-beor')?8
Faced with a talking ass, the term 'intelligent' is thus open textured too,
not just an "ordinary" open-ended one, in the sense that only normal con-
texts are taken into account with respect to the question of its true applicatIOn.
3. In Kafka's Metamorphosis a salesman is transformed into a gigantic

insect. Its shape is that of a cockroach: dome-like belly, numerous legs, and
all the rest. Yet it (he?) has human consciousness. What would we say: that
it is an insect, though with human consciousness, or that it is a person,
though shaped as an insect? According to Isaiah Berlin, Austin had reacted
to this case in the following way: "We should not know what to say. This is
when we say 'words fail us,' and mean this literally. We should need new
words. The old ones just would not fit. They aren't meant to cover this kind
of thing."9
It seems that we are caught between two incompatible stands.
(a) We seem to understand stories like Kafka's Metamorphosis. Moreover,
children seem to understand without difficulties stories in which pet mice
become horses and a dog turns into a fine footman. In short, we seem to
understand fairy tales, science fiction, and the like.
(b) We share (or at least some of us do) Austin's feeling that our words
might fail us in bizarre cases.
In what follows I shall try to resolve the apparent conflict between (a)
and (b). I shall argue that there are several strategies of understanding fairy
tales and science fiction, none of which undermines Austin's claim. In the
sense of 'understanding' which is germane to Austin's claim, however, we do
not really understand fairy tales and science fiction. I refer to that sense of
'understanding' in which truth conditions of sentences for possible worlds
play a major role. This, in turn, might prove fatal to possible worlds se-
mantics as an explanatory thesis for natural languages, provided that 'the
world of fairy tale' is taken as an admissible world.
I shall start with children's understanding of stories which tell about
144 AVISHAI MARGALIT
miraculous happenings and bewitched objects.

Hilary Putnam drew an argument from children's understanding of tales
against Carnap's account of theoretical terms. tO There is nothing special
about understanding terms that do not refer to observables, he argues, since
children can easily understand stories about tiny people that cannot be seen.
The force of Putnam's argument, if he is right, is clear: Do not dramatize
difficulties beyond necessity, especially when they are such that children over-
come them effortlessly. But do they? Again, I contend that in the relevant
sense of 'understanding' (relevant to Carnap's claim as well as to Austin's)
they do not.
I find the following type of "explanation" quite revealing to our discussion:
Question: What is wireless?
Answer: I shall explain wire first. Well, wire is like a mammoth dog
whose tail is in Boston and its mouth in New York. You
pull the tail in Boston and it barks in New York. That is
the way wire works. Now wireless is the same, but without
the dog.
Do not dismiss such "explanations" offhand: after all, explanations in which
Nothingness is just like Being but without existence are with us from time
immemorial.
It is my contention that a great deal of what is involved in children's
understanding of tales belongs with the "understanding" provided by the
wireless explanation. Being told of tiny people whom no one can see, children
simply drop the bit of information concerning invisibility and go on treating
them as ordinary people to all intents and purposes.
This is a sense of 'understanding' in which freakish facts, rather than being
taken seriously as constraints, are simply ignored, thereby familiarizing the
story and its 'world.' Put hyperbolically, I suggest that the farther and the
crazier the alleged possible world, the closer we tend to bring it to our world.
This is so for the simple reason that the more freakish the features, the more
our imagination fails us in taking them as serious constraints.
But, of course, there is more to understanding of bizarre stories than a
withholding-the-dog kind of mental operation. For highly structured strange
stories, such as science fiction, what I shall call syntactical understanding is of
special interest. Let us try to describe an experience that, as philosophy
students, you most certainly have had.
Try to remember a seminar in which a totally obscure passage was
OPEN TEXTURE 145
discussed - say, something of Fichte about the Absolute, or some such stuff.
You probably had a strong feeling that you do not understand a bit, as well
as a gnawing suspicion that neither do the others. But then, I venture to
assume, you were amazed that no one stopped the heated discussion by
saying that the king is naked. What added to your embarassment was that
there were students who were clever and seemed to talk to the point, while
others were clearly out of it. How come?
I think that while full blown semantical understanding is precluded in such
cases, apart maybe from a very metaphorical, rough and underinterpreted
one, what is at work here is to a large extent syntactical understanding. That
is, some of the students revealed skill in making syntactical inferences, in
which the crucial words are grossly underinterpreted and yet correctly
manipulated. Obviously, syntactical understanding does not stand on its own.
Understanding Fichte is not just understanding an uninterpreted calculus: it
is surely accompanied by some vague idea about the meaning of some under-
interpreted terms and, perhaps what is more important, by some kind of
metaphorical understanding. Metaphorical understanding seems to play an
important role in the general understanding of fairy tales and even of science
fiction.
By metaphorical understanding I do not mean merely the understanding of
metaphors, but also of other modes of speech which resemble metaphors in
the way their meaning is computed.
One type of metaphorical understanding involves our forming a very crude
model of that which we are supposed to be talking about. In such cases we
usually have afeeling of what is going on rather than genuine understanding.
Familiarity breeds the feeling of understanding. You find this, i.e., in the
talk about "tiny motes plus a stack of bedsprings."11 There is nothing wrong
with motes, of course, and at the same time there is nothing special about
billiard balls as objects that usually furnish our model of molecules.
The point is not the crudity of the chosen domain of objects for the model,
but the fact that the model is unspecified as to its structure, and many of its
features are left in the dark. No isomorphism between the model and the
modelled can be cogently supposed and hence the question of tlUth does
not arise, not even vicariously.
The difference between models for science fiction and overly crude models
in popular science is that whereas in popular science, as we have seen, the
model is crude and the modelled highly articulated, in science fiction it is just
the opposite. To illustrate the latter think of an Asimov-like science fiction
story in which a new kind of pill is invented: by swallowing one you learn
Latin. The pill, so we understand, 'causes' us to acquire knowledge, but not
as a result of study or of being taught, and not even through what we believe
to be a relevant experience (such as being exposed to conversation in Latin).
Due to the lack of such features the sense of 'learn' involved remains indeter-
minate. Can you say, e.g., that your pharmacist 'taught' you Latin by selling
you the pill?
Now as a kind of model for this think of the case of learning Latin from a
record. True, apart from the fact that both the pill and the record have disk-
like shapes there is not much else that can be compared between them. But
then there is much that can be said about the procedure of learning a language
from a record. I believe this record model does confer familiarity, or
'feeling' - even though not strict understanding - upon the case of the pill.
It is these strategies of understanding which refute the doctrine that our
words fail us when our hard core beliefs are violated. None of these types of
understanding hinges in an essential way on the possible worlds truth con-
ditions of the sentences involved in the pertinent stories. In fact they con-
stitute what can be called strategies of understanding that try to overcome
such lack of clear truth conditions.
4. Should our hard core beliefs be violated, would not we be able to predict
the linguistic behavior we shall adopt? Put differently, cannot we project
from the present application of terms to paradigm cases in normal contexts
to their future application in abnormal situations?
Fodor's answer is no:
To ask what we would say should certain of our current beliefs prove false involves
asking what new beliefs we would then adopt. But to answer this question we would now
have to be able to predict what theories would be devised and accepted were our current
theories to prove untenable. Clearly, however, it is unreasonable to attempt to predict what
theories would be accepted if our current theories were abandoned and, a fortiori, it is
unreasonable to attempt to make such prediction on the basis of an appeal to our current
linguistic intuitions. 12
I think Fodor is right. However, the expression 'current theories' should, I

maintain, be properly hedged by the yet-to-be explained phrase 'hard core.'
There is an alleged straightforward argument to counter that of Fodor.
It is based on a distinction between pragmatic oddity of an utterance and
semantic deviancy of the sentence uttered. Violation of our hard core beliefs
is a source of pragmatic oddity (i.e. inappropriateness of an utterance in a
OPEN TEXTURE 147
normal setting), but it has nothing to do with the meaning of the sentence
uttered. The evidence in support of the distinction is mustered from the way
technological changes affect our linguistic behavior. Dramatic technological
changes, so the argument goes, affect, if at all, our speech but not our
language. What was considered at one time an odd utterance may be taken
as perfectly normal at another, in a community that experiences the relevant
technological invention. All that may happen without any change in meaning:
gadgets leave meaning intact.
Though I think the conclusion is wrong I agree with the methodology on
which the argument rests. That is, I believe that the context of technological
changes can supply an empirical clue as to what might have happened to our
language had our hard core beliefs turned wrong. Because of that let us
examine more closely two examples - connected with technological in-
ventions - in support of the distinction between pragmatic oddity and
semantic deviancy.13
Suppose someone goes these days to T.W.A. and utters "May I get two
tickets for the flight to the Mars." His utterance is surely odd. But then try
to envisage a future time in which T.W.A. schedules regular flights to the
Mars. In these circumstances such an utterance would be perfectly acceptable,
while none of the words uttered undergoes any change of meaning. It seems
that this case involves a routine projection of the term 'flight' from a para-
digmatic case, say a flight from New York to London, through a non-
paradigmatic case - from Cape Kennedy to the Moon - to the case of the
Mars. The evidence now available about the projection of 'flight' to the case
of the moon is obviously a good basis for projecting further to the case of the
Mars.
But it could have been different. After all, the term 'flight' is contrasted
with 'sail,' over the contrasting feature 'through the air' versus 'on the water';
the assumption being that such a distinction has diagnostic value. It could
have been the case that the vehicle's having wings or lacking wings would
count as a contrasting feature too; and so also the feature of 'flight through
air' as opposed to 'flight through space.' That the term 'flight' should be
amenable to such routine "extensions" need not be a matter for a learned
guess.
Now take a case in which an utterance of a sentence was most certainly
odd at some point in the past, but not so now. Had Nelson said to Lady
Chatterly "I'm talking to you from two thousand miles away," previous to
the introduction of the telephone, it would have been rather odd. If Nelson
(Rockefeller) utters the very same sentence to Happy, there is nothing odd
about it. Here too the change from an utterance which is odd to an utterance
of the very same sentence which is not odd does not seem to hinge at all on a
semantic change in the term 'talking.' Pragmatic oddity should not be con-
Hated with semantic deviancy: in the case of 'talking' there was indeed no
change of meaning caused by the invention of the telephone. And yet, I would
like once again to point out that it could have easily been different, had the
distinction between 'talking directly' and 'talking through the phone' been
a marked one, i.e., had it had a diagnostic value.
Relative to the invention of the telephone, the introduction of the shower
in the 1880's seems uneventful. Yet 'taking a shower' is since then con-
trasted with 'taking a bath': the replacement of one by the other can change
the truth value of the sentences containing them.
The conclusion I draw from these two examples is that they do not sustain
the claim that technological inventions are pertinent at most to pragmatics,
never to semantics. This conclusion is based on the fact that it is quite easy to
see how the situation could have been different, i.e. how it could have been
a matter of semantic change rather than mere pragmatic acceptability. What
the technological change cases bring to the fore are the facts that our
interests, not only our beliefs, might be affected by unexpected inventions,
and that diagnostic value has to do with interests as much as with beliefs.
Since there is no way of telling ahead of time what interests we shall have in
view of assumed changes, projection of terms to such cases is to a large extent
indeterminate.
A case in point is Rolston's example of the history of pole vault. 14 The pole
used for pole-vaulting underwent a series of successive technological im-
provements, from hickory to bamboo and then to aluminium alloy. All these
changes went through rather smoothly. In the 60's, however, fiber-glass poles
were introduced. This created an outcry by those athletes who could not, or
would not, adapt themselves to the new pole and as a result were suddenly
left behind. The new poles came close to being banned altogether (in motor
car races, incidentally, cars with super-engines were effectively banished).
This illustrates a clash of interests that might have affected our use of the
term 'pole vault' as referring to an athletic contest. Since athletic contests are
defined by their rules, the pole material could in principle be sUbjected to the
rules of pole vault, even though it is not determined by these rules.
Wittgenstein was definitely right in pointing out that the rules of tennis
leave many aspects of the game uncovered: e.g., the permitted height of the
OPEN TEXTURE 149
ball's trajectory,15 A coach may advise his trainee about that aspect of the
game, but this is an advice as to how to play good tennis, not just tennis. In
the current jargon this means that this aspect belongs to the game's
performance, not to its constitutive rules (competence). However, radical
changes in the standards of a game's performance might be followed by
changes of interests, and these might lead to pressure for changing the rules of
the game. This is true of games, and it is also true of language games - or
indeed of our language in general.
Open texture is a phenomenon that may potentially affect all of our rule-
governed behavior, not just the projected use of our language. We saw it with
games - language games and other games - but it can be generalized even
further. Wittgenstein, for one, pointed out the intimate relation between open
texture in our conceptual frame on the one hand and in our legal system on
the other)6 Thus, consider the case mentioned by Locke of a monster with a
man's head and a hog's body. He then asks whether the bishop should be
consulted "Whether it were man enough to be admitted to the font or no?"17
It seems to me that we can answer Locke with Wittgenstein's "This law was
not given with such cases in view."18
The point is that the assumption of normal conditions is important if not
vital both for applying the law and for applying our terms.
5. Karl, Carnap's subject in his famous thought-experiment,19 was presented

with descriptions of strange animals, like half-man half-dog, etc. Karl had
to say whether he is ready to apply 'Mensch' to a given description, decline
to do it, or withhold judgment. The task then is to figure out from Karl's
reactions what is the intension of ' Mensch' for him. The leading idea is that
the intension of a predicate comprehends possible sort of objects to which it
applies: grasping the meaning of an expression consists in mastering its
application to possible objects, not just to actual ones. Karl, as Carnap is
first to admit, might find it difficult to put the world "in brackets": "Karl's
ignorance has the psychological effect that he has seldom if ever thought of
these kinds (unless he happens to be a student of mythology or a science
fiction fan) and therefore never felt an urge to make up his mind to which of
them to apply the predicate 'Mensch' " (p. 240). Carnap goes on to say that
because of Karl's lack of training in dealing with fancy creatures the linguist
questioning him might find a large indeterminate zone for the predicate
'Mensch.' But this open texture ("intensional vagueness"), Carnap believes,
does not mean that the meaning of 'Mensch' is not understood by Karl. "This
lack of clarity does not bother him [Karl] much because it holds only for
aspects which have very little practical importance for him" (ibid.).
It seems to me that Karl's difficulties stem neither from his not being a
science fiction fan nor from his not being hooked on mythology. The difficul-
ties are not even due to the fact that "the man on the street is unwilling to say
anything about non-existent objects" (ibid.). (If anything, the "man on the
street" is to my mind only too ready to express an opinion about non-
existent things.) Karl's difficulty is in fact not his own difficulty but rather
one he poses to the linguist. It is in fact not radically different from the
problem faced by Locke's learned bishop, or by Kafka's friend Max Brod.
The linguist cannot be justified in counting on the intuitions of anyone of
them as to what they would say had they encountered the gigantic cockroach
that answers to the rigid designator 'Samsa.'
The reason is that the existence of such an insect would force them to
change their beliefs in some natural laws, though it is not that easy to specify
which. But giving up a belief in a natural law might lead to giving up some of
our central tenets, since the events connected by any law are causally con-
nected with many other events. So the assumed change might cut deep.
Moreover, not all the consequences of giving up a natural law can be fore-
seen. It might affect our modes of speech: perhaps not the figurative or the
"loosely speaking" modes of speech, but rather the "strictly speaking" mode
of speech - and that, after all, is where the problem of truth arises.
Not all possible worlds are equal. The post-Carnapian semanticists who
continued his research program added a constraint: the intensions are to be
determined not by all possible worlds, but only by those possible worlds that
constitute alternatives to our actual one. I consider this a move in the right
direction. And a natural solution to the problem of open texture seems to be
the following: take as genuine alternatives to our world only those worlds
which are compatible with our hard core beliefs. But there is a price to pay
for adopting such a solution: for one thing, such possible worlds semantics
is so epistemically constrained that all de re statements should be taken as
de dicto ones. 20 What is even more of a problem is the fact that our set of
hard core beliefs (when made explicit and represented in statements) is a
fuzzy set, hence the alternative relation among possible worlds is fuzzy too.
If after all this the enterprise of possible worlds semantics as semantics of
natural languages still seems worth doing, so let it be.
6. I have so far said nothing about our hard core beliefs. It is clear that not
OPEN TEXTURE 151
every statement we believed to be true and which turns out to be false in-
fluences our concepts. However it is not at all clear how to demarcate the
hard core beliefs from the non-hard core ones.
Instead of offering a demarcation line I shall suggest how to rank order the
vulnerability of our kernel beliefs - from the periphery to the center, as it
were. In terms of this spatial picture, then, I shall order the threats to our
'protective belts.'21 The 'outer belt,' I maintain, is attacked by counter-
Jactuals which are compatible with the known laws of nature. The 'inter-
mediate belt' is constituted by beliefs which are threatened by counternomic
statements, i.e. those that assume a natural law to be false. The 'inner belt' is
that which is attacked by counteridentity statements of the 'If I were Roth-
schild' type.
I choose to stop here. One can however go further and add yet another
'belt,' to be attacked only by counter-logical statements. But I prefer to
postpone judgment as to the question 'Is Logic Empirical?'
Another problem is who are the 'we' referred to in "our hard core beliefs":
all the speakers of the language? - 'all' distributively or 'all' collectively? or
maybe the context determines the 'we'? I suggest to identify the 'we' through
Putnam's "linguistic division of labor."22 The 'we' there is that of the com-
munity of speakers as a whole, via the agency of its experts. Thus, when
seeing a monkey that has just finished typewriting Hamlet, Karl hesitated
whether to apply'Affe' to it. But he could have consulted Reichenbach, who
would have told him that such monkey business is possible, though most
improbable.
The thesis I advocated in my discussion is that open texture is not a freak
of language, although it has to do with freakish behavior of objects. Any
semantic theory of natural language which is worth its salt should, if not
account for this phenomenon, at least take it into account.
The Hebrew University oj Jerusalem
NOTES
.. I am grateful to Edna Ullmann-Margalit who greatly helped me in matters of form and

substance.
1 F. Waismann, "Verifiability," in his How I See Philosophy, R. Harre (ed.), London,
1968, pp. 39-66.
2 No one defends complete verificationism nowadays.
152 A VISHAI MARGALIT .
3 Waismann, p. 42.
4 H.L. Rolston, "Wittgenstein's Concept of Family Resemblance," UnpubJ. diss.,
Harvard University, 1971.
5 Waismann, p. 42.
6 G. Ryle, The Concept of Mind, Penguin Books, 1949 (esp. ch. 5).
7 R. Carnap, "Testability and Meaning," in: Readings in the Philosophy of Science.
H. Feigl and M. Brodbeck (eds.), New York, 1953, pp. 47-92.
8 Numbers 22.
9 1. Berlin, "Austin and the Early Beginnings of Oxford Philosophy," in: Essays on
J.L. Austin, Oxford, 1973, p. 11.
10 H. Putnam, "What Theories Are Not," in: Logic, Methodology and Philosophy of
Science, E. Nagel, P. Suppes and A. Tarski (eds.), Stanford, California, 1962, pp. 240-251.
11 W.V. Quine, Word and Object, Cambridge, Mass., M.LT., 1960, p. 15.
12 J.A. Fodor, "On Knowing What We Would Say," in: Readings in the Philosophy of
Language, J.F. Rosenberg and C. Travis (eds.), Englewood Cliffs, New Jersey, 1971,
p.133.
13 J.A. Fodor and J.J. Katz, "Introduction," The Structure of Language, Englewood
Cliffs, New Jersey, 1964, p. 15; H. Putnam, "Minds and Machines," in: Minds and
Machines, A.R. Anderson (ed.), Englewood Cliffs, N.J. 1964, pp. 88-94.
14 See Rolston, note 4.
15 Philosophical Investigation, tr. by G.E.M. Anscombe, Oxford, 1953, 68.
16 Zettel, G.E.M. Anscombe and G.H. von Wright (eds.), Oxford, 1967, 350.
17 An Essay Concerning Human Understanding, Book 3, Ch. 6, 27.
18 Zettel, 120.
19 R. Carnap, "Meaning and Synonymy in Natural Languages," in his Meaning and
Necessity, 2nd ed., Chicago, 1956, pp. 233-247.
20 N. Chomsky, Reflections on Language, New York, 1975, p. 48.
21 I. Lakatos, "Falsification and the Methodology of Scientific Research Programmes,"
in: Criticism and the Growth of Knowledge, I. Lakatos and A. Musgrave (eds.), Cambridge,
1970, pp. 130-137.
22 H. Putnam, "The Meaning of 'Meaning,''' in his Mind, Language and Reality,
Cambridge University Press, 1975.
MARCELO DASCAL
CONVERSATIONAL RELEVANCE
In everyday life, in science and in philosophy, we often express judgments of

relevance: fact a is relevant to fact b, theory c is relevant to action d, state-
ment e is irrelevant to belief J, etc. And we often have a great measure of
confidence in such judgments, since we are ready to base important decisions
upon them (e.g., in the courtroom, when an objection concerning the irre-
levance of a question is either sustained or rejected by the judge) and, when
necessary, we are willing to engage in serious disputes about the truth of such
judgments (e.g., a doctoral committee discussing about the relevance of the
second chapter of a dissertation).
In philosophy, the concept of relevance or its close relatives is often
encountered at the bottom of efforts to solve central philosophical problems
and to analyze fundamental concepts: induction, verifiability, synonymy,
knowledge, natural kinds.
But when used by philosophers, scientists or the layman, this concept is
seldom clearly defined or characterized. Some take it to be hopelessly vague
and therefore useless in serious endeavors. Others will consider the recourse
to such a concept in any theory as an avowal of failure. I believe relevance is
a fundamental notion for science, philosophy and everyday life, a notion
which is clamoring for attention and analysis of its several characteristics as
well as of its innumerable applications. The present paper focuses on one
such application.
H.P. Grice, in his widely circulated William James Lectures (1968, 1975),
makes use of the concept of relevance as one ofthe key elements in his account
of the 'logic of conversation.' According to him, conversation is governed by
a general principle of cooperation:
(CP) Make your conversational contribution such as is required, at
the stage at which it occurs, by the accepted purpose or
direction of the talk exchange in which you are engaged
(1975: 45).
Let us leave aside, for the moment, the question of the justification or need
for such a principle. On the assumption that it is in fact justifiable, Grice goes
153
154 MARCELO DASCAL
on to distinguish four 'categories,' Quantity, Quality, Relation and Manner,

into which the various specific maxims derived from the CP will be classified.
These categories can be represented by four 'supermaxims' which are
alleged corollaries of the CP (1975: 45-46):
(QN) Make your contribution as informative as is required (for the
current purposes of the exchange).
(QL) Try to make your contribution one that is true.
(R) Be relevant.
(M) Be perspicuous.
Whereas Grice is able to specify certain maxims associated with QN, QL
and M, thus clarifying somehow their meaning, R is left by him in a dis-
paragingly vague state, which the examples given of its use do not help
to dispel. Grice fully acknowledges this fact and indicates some of the
difficulties that a better formulation of R ought to overcome:
... its formulation conceals a number of problems which exercise me a good deal; ques-
tions about what different kinds and foci of relevance there may be, how these shift in the
course of a talk exchange, how to allow for the fact that subjects of conversation are legi-
timately changed, and so on. I find the treatment of such questions exceedingly difficult,
and I hope to revert to them in a later work (1975:46).
As far as I know, the hope here expressed was not yet fulfilled. With no
pretension to solve these really thorny problems, I will make an attempt to
pinpoint some aspects of the concept or concepts of relevance required in
Grice's framework. I must warn you that I will raise more problems than I
will be able to solve. Yet, I hope that, in so doing, not only will something be
gained by way of explicitating supermaxim R, but also that some contribu-
tion will be made to a better understanding and re-structuration of the whole
framework. For, as many of Grice's examples indicate, there is a sense in
which a certain concept of relevance, perhaps not identical with the one
required for R, governs the operation of the other supermaxims, as if the CP
itself were in fact a principle of 'relevance' rather than a principle of 'coopera-
tion.' It is probably this fact that led Grice to admit that" ... I am fairly
sure that I cannot reach it [i.e., a justification for the CP, M. D.] until I am
a good deal clearer about the nature of relevance and of the circumstances
in which it is required" (1975: 49). Although I will not be able, within the
limits of this paper, to fully develop this line of thought, I believe the dis-
tinctions I propose here are generalizable and, thus, useful for such a further
development.
CONVERSATIONAL RELEVANCE 155
My main contention will be that it is indispensable to distinguish several

types of relevance in order to provide an adequate account of the operation
of maxim R. In particular, these types should include two quite distinct
notions of relevance, a 'pragmatic' and a 'semantic' one. The former has to
to do with the relevance of speech acts to certain goals; its characterization
may thus be viewed as a specialization of the general notion of relevance of an
action to a goal which is an essential piece of the theory of goal directed
behavior, itself a part of the much desired general theory of action. The
latter concerns the relevance of certain linguistic, logic, or cognitive entities,
say, propositions, to other entities of the same type; its characterization,
which involves concepts such as reference, aboutness, meaning relations,
entailment, etc., is, to my mind, a fundamental task of semantic theory. The
interaction of these two types of relevance in the generation of conversational
implicatures via supermaxim R illustrates the intricate relationship that must
obtain between the semantic and the pragmatic components of an adequate
grammar of any natural language.!
Grice's main purpose in building his system of maxims is to put it to use
in order to account for the widespread occurrence of what he calls 'con-
versational implicatures.' Consider some of his examples (those involving,
in his account, use of the maxim R):
(1) A and B are talking about a mutual friend C who is working in
a bank. A asks B how C is getting on in his job, and B replies:
"Oh quite well, I think; he likes his colleagues, and he hasn't
been to prison yet."
Implicature: B implicates that C is potentially dishonest.
(2) A is standing by an obviously immobilized car and is ap-
proached by B.
A: "I am out of petrol."
B: "There is a garage round the corner."
Implicature: B implicates that the garage is (possibly) open
and has petrol to sell.
(3) A: "Smith doesn't seem to have a girl friend these days."
B: "He has been paying a lot of visits to New York lately."
Implicature: B implicates that Smith has, or may have, a girl
friend in New York.
(4) At a genteel tea-party, A says "Mrs. R is an old bag." There
is a moment of appalled silence, then B says "The wheather has
156 MARCELO DASCAL
been quite delightful this summer."

Implicature: B implicates that A's remark should not be
discussed (or, more specifically, that A has
committed a social gaffe).
Now, according to Grice, underlying each of these implicatures there is a
certain (deductive) argument, involving several references to the maxim R.
For example, for case 1, the implicature-generating argument is:
Implicature-generating (1) B has apparently violated the maxim "Be
argument for relevant";2 yet I have no reason to suppose
example (1) that he is opting out from the operation of the
CP; (2) given the circumstances, I can regard
his irrelevance as only apparent if and only if
I suppose him to think that C is potentially
dishonest; (3) B knows that I am capable of
working out step (2). So B implicates that Cis
potentially dishonest. ("I" here refers to the
A of the example).
We may discern here at least three judgments of relevance, which play an
essential role in the argument, two of them apparently semantic and one
pragmatic. The first, upon which the claim that B apparently violated the
maxim "Be relevant" is based, is a negative assessment of the relevance of
what was strictly said by B, namely that C has not been to prison yet (P),3
Notice that p itself must be really and not only apparently irrelevant (at least
for the hearer) in order to allow for the generation of the implicature. What
is only apparently irrelevant is not p, but B's utterance of p. It is in order to
show that B's utterance of p is not really irrelevant (which is required by the
hypothesis that B is not opting out of maxim R) that the implicature "C is
potentially dishonest" (q) is introduced as a part of the content of B's
utterance. The second judgment of relevance is, then, the positive assessment
of the relevance of q. This leads, finally, to the third judgment: B's utterance
of p is in fact relevant because it somehow conveys or contains q, which is
relevant. Unlike the two previous judgments of relevance, which concern
certain propositions or contents (p and q), and are, perhaps, judgments of
semantic relevance, the third one concerns an utterance (which includes an
act of implicating), i.e., an act, and must be, therefore, a judgment of prag-
matic relevance.
The considerations above offer only a very rough first sketch of the uses of
the concept of relevance in an implicature-generating argument, but they
already hint at the complexities involved. In particular, they suggest some
ways in which Grice's pattern for implicature-generating arguments might be
expanded. However, consider now what is obviously missing in that sketch:
it is formulated as if relevance were a monadic predicate. But it is
obvious that, whatever it may be, relevance is at least a dyadic predicate,
a relation. Therefore, it is only possible to understand a judgment of re-
levance if both relata it relates are clearly specified. In general, the first step
towards clarifying the nature of a relation is to specify its domain and its
range. So far only the domains of the two kinds of relevance involved in the
argument were considered (and all too briefly, for that matter): speech acts
and propositions. What about their ranges? Grice talks loosely about the
"accepted (local) purpose or direction of a conversation" as constituting that
to which B's utterance ought to be relevant. But the difficulties in identifying
such (local) purposes or directions with any degree of confidence are con-
siderable. They can range from very specific requests of information, similar
to a multiple-answer test or to a yes/no question, to very diffuse aims such
as "keeping the conversation alive."4 Furthermore, a conversation may be
highly structured, with a fairly definite subject-matter to which each local
purpose is hierarchically subordinated and with respect to which, ultimately,
the relevance of each contribution to the conversation must be judged, or
else it may lack practically any structure, its coherence being maintained
merely by any imaginable association linking the contributions of speaker
and hearer.
But even if we assume that somehow an accepted local purpose of a con-
versation can be properly identified, as well as the speech act that is supposed
to be relevant to such a purpose, i.e., even if we manage to identify the two
relata, it is still far from clear how or in what respect or by what criteria the
relevance of the former to the latter ought to be assessed. On this point,
Grice leaves the reader completely in the dark. To be sure, he is not very
helpful, either, in what concerns the 'accepted local purposes' of conver-
sation, for his remarks on this topic are extremely general and in his examples
he invariably transfers the burden of identifying the purposes in question to
the reader, by using the intransitive (monadic) formulation of the judgments
of relevance. But at least, in this case, he hints at what he has in mind, and
at some of the problems posed by such an identification. Not even such hints
are available to help us determine the types and criteria of relevance that
158 MARCELO DASCAL
should be used in general and in each particular example. Yet, without the
specification of such criteria, it seems that any judgment of relevance about
<
any ordered pair a, b> containing elements of the appropriate domain and
range, turns out to be trivially true and, thus, uninformative. For, such an
unspecified judgment may only mean "There is at least one respect in which
a is relevant to b" and, given the notorious vagueness of the (general) notion
of relevance, some respect in which a is relevant to bean always be found,
thus making the judgment in question always true. s Unless some way of
blocking such a conclusion is found, the notion of relevance is utterly useless
for the derivation of conversational i mplicatures: if all judgments of relevance
are trivially true, then no jUdgment of irrelevance is true and, therefore, you
cannot even get started in the process of generating an implicature, whose
first step is precisely a judgment of irrelevance! Using Goodman's (1972)
phrase, we could call the need for blocking the above conclusion the 'require-
ment of selectivity.'
The natural way to fulfill such a requirement seems, then, to be the speci-
fication of the set of legitimate respects in which a's relevance to b is to be
assessed in each case, the definition of criteria of assessment for each of these
respects, and the establishment of a procedure for determining which of them
is actually used in a given context. Now, the task of replacing the vague and
comfortable notion of relevance by something that even remotely approaches
the precision suggested by the term 'criterion' is notoriously difficult.
Attempts to do so in other areas were doomed to failure - the outstanding
example being the sad story of the 'criteria of empirical significance' (see
Dascal, 1971). In the light of these failures, one critic has affirmed that
"relevance is not a precise logical category" and that "the word is used to
convey an essentially vague idea" (Berlin, 1938-39: 21). No wonder that
Grice does not even try to formulate his maxim more precisely. But this
alleged vagueness of the notion of relevance stands in sharp contrast to the
undeniable fact that implicatures are generated in a very precise and generally
reliable way. It is precisely that particular q, from a myriad of imaginable
alternatives, that is implicated and recognized as such. 6 I have in fact tested
some of the examples with several people, including my ten-year old daughter,
and almost invariably they reach similar conclusions concerning what (if
anything) has been implicated by the speaker in each case. Furthermore,
implicatures are widely used in diplomatic exchanges as a very sophisticated
means of conveying carefully nuanced and precise messages (consider the
recent case of the White House's "condemnation of all foreign intervention
in Lebanon" accompanied by the spokesman's refusal to condemn explicitly

the Syrian intervention there).
The problem we face, then, is the following one: how can such precise
messages as the implicatures be conveyed via a reasoning that makes essential
use of such an imprecise notion as relevance. To be sure, there are several
ways to escape from this problem: for example, to deny that the implicatures
are in fact as precise as I have suggested in cases where the maxim R is
involved; or to deny that maxim R is at all involved in implicatures, explain-
ing its apparent uses by means of some other principle, which does not
contain the notion of relevance. I reject both escape routes. The first one,
because of my experience concerning the reliability and precision of the
implicatures; the second one, because my intuition as a speaker (and human
agent, in general) is that we (humans) base indeed our conclusions (about
implicatures, among other things) on implicit judgments of relevance and,
furthermore, that we are in fact very good at it. The task of theory -
pragmatic theory and semantic theory, in the case of linguistic actions - is
to make this competence of ours explicit. Admittedly, even though we do
perform judgments of relevance, some other explanation of the implicatures
apparently generated through maxim R might be offered. Nevertheless, not
only such other explanations are hardly available, but even if they were, they
should anyhow be connected with the phenomena we describe naturally in
terms of relevance. Hence, the role of this concept in this area of research is
worth exploring anyway.
In order to reduce the problem to manageable proportions, let us assume
that a 'conversation' is a two-utterance dialog between speakers A and B in
context C. Let us say that A's utterance sets up a certain conversational
demand (for B), and that B's utternace is B's reaction to that demand, in the
context C.
Such a description of the situation involves some further simplifying
assumptions. Let us say that something is topically relevant at time t for a
subject S if it is at the center or focus of S's field of attention at t. What is
not topically relevant, but is still somewhere in the field of attention of S,
say, in its 'horizon,' will be said to be marginally relevant for S at t.7 One
must, of course, add that, beyond the horizon of the field of attention, there
is a whole domain of stored data which might be referred to as the back-
ground, whose members are potentially relevant (in various degrees) for S at t.
It must be recalled that the relationship between these three levels of rele-
vance (and there might be more) is not static but dynamic, since data are
160 MARCELO DASCAL
constantly moving from one to the other.

Now, it seems natural to assume that the conversational demand set up by
A's utterance at t is topically relevant for B at t+ 1, whereas B's perception
of the other elements of the context is merely marginally relevant for him. It
seems also natural to suppose that what is topically relevant is what primarily
commands a subject's conscious reactions. Certainly an utterance is a con-
scious reaction. Therefore, B's utterance can be described (primarily) as a
reaction to the demand set up by A's utterance. It is also a reaction to the
context, to be sure, but only secondarily. And it is certainly influenced by
what we called the background, but one cannot say that it is commanded
primarily by it. Among other things, this predominance ofthe conversational
demand plays an important role in determining which of the marginally and
potentially relevant data will become, in turn, topically relevant, as elements
which are likely to be used in the leaction. What these sketchy, phenomeno-
logically oriented, distinctions allow one to do is, in Quine's words, to start
to "cut through the jungle of possible motives [I would say stimuli] to the
volunteering of sentences." This is only the beginning, though. We will have
to cut much more in order to be able to pair reactions with conversational
demands.
This beginning, however, seems to be well justified empirically, since it can
already be put to use to account for some implicatures. There is indeed a
general presumption that our reactions should be commanded by what is
topically relevant. Normally, conversational demands are topically relevant,
whenever they are present. That is to say, participation in a conversation
monopolizes or should normally monopolize our attention, other things
being equal. But this is not always the case. Suppose, for example, that
Socrates and Meno are engaged in a conversation in the agora and it suddenly
starts to rain. This new stimulus is likely to become topical for both of them,
replacing the current conversational demand. It would be expected, then, to
command the ensuing reaction. In that case, a comment like "the weather
has been so nice these days" would be perfectly appropriate as a reaction to
the non-conversational demand set up by the intrusion of the rain into the
conversation and it would not, therefore, generate any implicature. In other
words, a shift of subject (at least a momentary one) should be expected in this
case. If, on the contrary, Meno insists on producing a linguistic reaction which
is relevant to the previous conversational demand alone, deliberately and
ostensively disregarding the obvious interruption, this fact would be likely to
generate an implicature: "You see, Socrates, I am so interested in talking
with you that I don't care about getting wet." If the interruption were caused
by an utterance of some other speaker, Euthydemus, trying to intervene in
the conversation, and Meno (deliberately and ostensively) pursued the con-
versation reacting only to Socrates' utterance, this fact would be interpreted,
both by Socrates and by Euthydemus, as carrying an insulting implicature
towards the latter.
The interpretations offered above are correct just in case one further
assumes that no merely causal account of Meno's reaction is available.
Suppose, for example, that he is so engrossed in thought that he does not
evenfeel the drops (or, alternatively, that he does not even hear Euthydemus'
intervention). In that case, the fact that he does not react to them cannot be
said to convey an implicature of the type "You see, Socrates, I am engrossed
by your words .... " To be sure, Meno's behavior is semiotically related
to his engrossment, in the sense that it can be viewed as a sign of it. But it is
a sign of it in the same way as my pronunciation of certain Hebrew phonemes
is a sign of my South-American origin. Let us use the term 'indicate' in order
to refer to this kind of relationship. We should then say that, no matter what
I intend to convey, directly or via an implicature, when I utter certain words,
my utterance indicates that I come from South America. S I have no control
over this fact, and my (communicative) intentions play no role in making it
convey what it does in fact convey. It is precisely this feature - the lack of
intentionality - that distinguishes indications (in the sense just stipulated)
from implicatures. Thus, if Meno's disregard for the rain is unintentional, his
behavior can at most be said to indicate his engrossment, but not to implicate
it. This is why, when describing the implicature, I insisted on using the words
'deliberately and ostensively.'
In terms of the concepts we have so far introduced, the explanation for the
case just envisaged lies in the fact that the rain is not topically relevant for
Meno, but at most marginally relevant. Therefore according to our sugges-
tion, his conscious reaction cannot be described as a reaction to the rain.
Hence, its relevance cannot be assessed in terms of the demands set up by the
rain, so that the alleged irrelevance of his behavior relative to the rain cannot
be the starting point for the derivation of an implicature.
A further simplifying assumption which we shall make is that B correctly
identifies the conversational demand set up by A's utterance. This assump-
tion blocks the way for a certain class of explanations of the possible apparent
irrelevance of B's reaction, namely, those based on the claim that B in fact
reacted to some demand other than the one really set up by A's utterance. In
162 MARCELO DASCAL
other words, misunderstanding as an explanation of irrelevance is thereby

excluded: B understood perfectly well the point of A's utterance in all its
details. This includes disambiguation of homonyms (B understood that A
was talking about the financial institution), correct identification of referents
(the pronoun 'he' used by A refers to Smith) and also identification of the
(illocutionary) force of A's utterance (although A used the indicative mood,
B understood pefectly well that A did not assert anything but rather posed a
question to him). You may have guessed that the utterance of A's I am talk-
ing about is "I bet you know whether he went to the bank yesterday."
Notice that the latter assumption, unlike the former, is not a natural one
to make in general. For the hypothesis of misunderstanding is in some cases
the most natural explanation for an apparent irrelevance. Suppose, for exam-
ple, that A conveyed his conversational demand to B via a somewhat
complex implicature and that B's reaction seems to be irrelevant. A would
naturally think that B did not understand his point, and would, thus, take
B's reaction to be really irrelevant, that is, non-conducive to an implicature-
generating argument. Knowing this general fact about how A would first try
to explain the irrelevance of his reaction, B would understand that it would
be risky to try to convey his response to A via an implicature. This means
that a chain of implicatures in a conversation is very unlikely to occur,
because the chances of successfully implicating diminish considerably with
each reiteration. So, our assumption that no misunderstanding has occurred
is indeed a purely simplifying assumption.
In order to make clearer the significance of the assumptions made so far, I
would like to point out their close similarity to some interesting proposals
put forward by Davidson. According to him, "interpreting an agent's
intentions, his beliefs and his words are parts of a single project, no part of
which can be assumed to be complete before the rest is" (1973: 315). He
compares this project to the task of solving an equation with several variables.
If you assign values to all of the variables but one, you will be able to solve
the equation for this last variable. Otherwise, the solution will remain indeter-
minate. For reasons having to do with his broader project of devising a
Tarski-like semantics for natural languages, Davidson restricts his discussion
to three main variables: the fact that, in a given context, a speaker 'holds a
sentence true,' the belief(s) the speaker has in that context, and the inter-
pretation(s) assigned to the sentence in that context. 9 If we know that S holds
p for true and the correct interpretation of p, then we can correctly assign a
certain belief (or set of beliefs) to S (in the context in question). Similarly, if
CONVERSA TIONAL RELEVANCE 163
we hold constant the belief(s) of S and the sentence p known to be held true
by him, we can determine the value of the third variable, namely, the meaning
or interpretation of p. Apparently, the model for such an account was
provided by the theory of action. If we want to explain why S lifted his arm,
we can assign him the desire to call someone. But this explanation presup-
poses in fact the assignment of a certain (set of) belief(s) to him, namely, the
belief that by lifting his arm he is indeed calling someone. If we assigned him
a different desire (e.g., the desire to pick up an apple), we should assign him
a different (set of) belief(s), and vice versa.
The cases we have been considering here disclose a similar relationship
between three variables: what is topically relevant for S, the conversational
demand as it is understood by S, and the interpretation of S's utterance
(particularly, the determination of its implicature - if there is any). If any
pair of these variables is held constant, then the equation is solvable for the
third, whereas if two of them are left unspecified, then the solution for the
third remains indeterminate. What our two assumptions amount to, then, is
no more than an attempt to devise admittedly artificial conditions under
which the equation can be solved for the variable that interests us in the
present paper, namely, the implicatures (which are part of the interpretation
of S's utterance). The analogy with Davidson's account may be extended
beyond the mere structural fact that there are three-variable equations in
both cases. My second variable - the conversational demand as identified
by S - is obviously a particular case of Davidson's second variable, i.e., a
particular type of belief assigned to S. The inclusion of implicatures in the
scope of the third variable is a natural way of extending and rendering more
complete the required notion of interpretation. As for the first variable -
what is topically relevant for S at t - I think it might be construed within
the framework of an adequate theory of perception and attention, in such a
way as to share some of the essential properties of Davidson's 'holding a
sentence true.' But it would take us well beyond the scope of this paper to
try to substantiate this claim. Let us rather return to our main topic.
All the preliminary work so far done has been necessary in order to circum-
scribe the second relalum of the relation of (pragmatic) relevance, that to
which B's utterance is supposed to be relevant. But we do not know yet in
what ways it can be relevant (or not). As a matter of fact, we do not know yet
what precisely these conversational demands are. Some hints have been given
above when we talked about the different kinds of disambiguation B achieved
with respect to A's utterance. These concern, on the one hand, semantic dis-
164 MARCELO DASCAL
ambiguation, that is, determination of the content of A's utterance; and on

the other, pragmatic disambiguation, that is, determination of the force of
A's utterance.
The concept of 'force' I have in mind is fairly close to Austin's notion of
'illocutionary force,' although I think it should be construed in such a way as
to clear up the confusion between it and some neighboring concepts. In
particular, it should be carefully distinguished from the components of the
meaning of a sentence that Cohen (1974) calls 'semantic modality' (Hare's
tropic) and 'semantic force' (Hare's neustic). The former are normally conveyed
in English by auxiliary verbs like 'is,' 'may,' and 'ought,' by the use of
indicative or subjunctive moods, or by certain adverbs like 'probably' and
'possibly.' The latter are usually expressed by the choice between declarative,
imperative, interrogative, optative, exclamatory or performative forms. Both
are characteristics of the meanings of sentences, not of utterances, as Cohen
correctly stresses. The iIIocutionary forces, on the other hand, are properly
assigned only to utterances. This is why I insist on classifying them as belong-
ing to pragmatics. Yet, nothing in my argument hinges on the acceptance of
a well-defined borderline between semantics and pragmatics. What is needed
for my purposes here is only the recognition of the fact that the interpretation
of an utterance involves several layers of, say, meaning (in the most general
sense), each one of which mayor may not be relevant to a corresponding
layer of another utterance. Some of these layers are clearly semantic (e.g., the
different lexical meanings of 'bank'), others may belong to a semantico-
pragmatic region (e.g., the semantic force), whereas others still are clearly
pragmatic (e.g., the iIIocutionary force).
Each one of these layers of meaning has distinctive effects upon assess-
ments of relevance. The force of A's utterance, for example, imposes certain
constraints on the relevance or appropriateness of the possible forces of B's
utterance (issued in reaction to A's utterance). In other words, it is not the
case that utterances of any kind of force are equally relevant as reactions to
an utterance having a particular type of force. For example, an information
question is not appropriate as a reaction to an information question unless
it is a clarificatory question, that is, a request for clarification of the prior
question. This is a kind of force that is defined, to be sure, in terms of a
peculiar semantic relationship between the contents of B's and A's utter-
ances: the former must be about the meaning of some component of the
latter, or of some detail of its structure. A typical device for formulating this
kind of question is the operator: 'What do you mean by ?' where the
CONVERSA TIONAL RELEVANCE 165
blank is filled in by some word of the previous sentence. Since a clarification

question is a kind of reaction appropriate for every conversational demand,
the operator just mentioned could be used in any conversational context in
order to produce relevant reactions. This fact was exploited in the computer
program ELIZA which simulates a psychiatrist interviewing his patient
(Weizenbaum, 1966). But even in such a very characteristic setting, some
questions formed by means of the device in question would be apparently
irrelevant, thus generating implicatures. Suppose, for example, that the
program reacted "What do you mean by 'the' (or, 'to')?" to the patient's
statement "I used to go to the pool." On the other hand, a reaction with the
force 'answer to information reply' is, as far as its force is concerned, clearly
relevant to an utterance whose force is 'information question.' Whenever a
reaction is relevant to the force of an utterance, let us say that it is 'pragma-
tically relevant' to the demand.
The differences in the (illocutionary) forces of utterances, which charac-
terize the kinds of reaction they belong to, are sometimes conveyed linguis-
tically through intonation, in fairly standard ways, for example:
(5) A: What did he drink?
B: 3 Tea 1 ~
(kind of reaction: answer to information request)
(6) B: 3 Tea 1 t
(kind of reaction: incredulous guess)
(7) A: Nobody is in the house.
B: 3 John 1 t
(kind of reaction: dissent plus complementation)
(meaning: "You forget John") (cf. Gunter, 1974: 60 ff.)
(Notice that in all these cases the semantic force, as opposed to the illocu-
tionary force, is the same, namely, declarative, or assertive. Consider also
the following: correcting, amending, exploring, recapitulating, asserting,
justifying, exploring, threatening, warning. All of them have different forces
and constitute different kinds of reaction but they all have the same semantic
force). But intonation alone does not determine by itself the force of a given
reaction. It combines, in fact, with information about the semantic relation-
ships obtaining between the two utterances and about the nature of the
demand. Compare examples (8), (9) and (5).
166 MARCELO DASCAL
(8) A: John drank wine.

B: 3 Tea 1 .t.
(kind of reaction: dissent, with correction)! 0
(9) A: John drank tea.
B: 3 Tea 1 .t.
(kind of reaction: recapitulation)
All these reactions have the same intonation; therefore, the undeniable
differences between the forces are determined either by the differences in
semantic relations (TEA-TEA identity; TEA-WINE opposition; example
(8) versus example (9 or by differences in type of demand (example (5)
versus example (8) ).
Furthermore, pragmatic relevance alone does not settle the question of the
relevance or irrelevance of the reaction to the demand. In many cases, the
demand is much more specific. It requires a certain kind of reaction with a
certain (kind of) content. That is, it requires also semantic relevance, in one
or more of its components. Now, if both the force and the content of the
reaction, including all its required components, match the demand (in the
case of a specific demand such as the one set up, say, by a yes/no question),
then we can say that the reaction (as a whole) satisfies the demand. In this
case, it is maximally relevant to it, and no implicature will be generated. If,
on the other hand, no component of the reaction matches the demand, then
the former is totally irrelevant to the latter, in which case too no implicature
can be generated. Only if some components are irrelevant, whereas others
are relevant, can an implicature arise. Exactly which one will in fact occur is
something that depends, in an orderly way, on the discrimination of the
relevant and irrelevant components, respectively.
In the light of the preceding remarks, one might generalize the suggestions
made above concerning the 'equation-like' nature of the problem faced by
someone who has to get at some implicature via the maxim R. Instead of
three variables, there would be many more variables, interconnected not by a
single equation but by a system of equations. Ideally, we would like to have
a system in which each equation connected only two or more variables of the
same kind, related to the demand and to the reaction: (illocutionary) force
with (illocutionary) force, semantic force with semantic force, modality with
modality, etc. Unfortunately, as we have seen, this is not so, and the equa-
tions are likely to be not only of mixed types but also much less precise than
algebraic equations. Their solution, therefore, cannot be expected to be
obtained by the application of some neat algorithm, which the theory of

implicatures - i.e., the 'logic of conversation,' in Grice's phrase - is
supposed to discover. It seems to me more realistic to look at the problem in
a different way.
The hearer who faces a (possible) implicature can be compared to a player
in a game. His task is to discover the implicature, if there is any - and this
too he is supposed to discover. Now, an implicature is a hypothesis about the
speaker's intentions that explains away the apparent irrelevance of his
utterance, by explaining how the utterance is in fact relevant. So, the task of
the hearer is similar to the task of a scientist who looks for a theory that
explains given data, under certain broad theoretical assumptions (comparable
to the maxim R or the CP). But we all know that, at least for the time being,
there is no 'method of discovery' or 'abductive logic' that would help him to
find the appropriate hypothesis. Similarly, the 'computation' of an implica-
ture is not really a 'computation' or a valid deductive argument, in spite of
Grice's attempt to present it as such. l1 For the whole argument stands or
falls with the truth or falsity of premise (c) (in the pattern presented in note
11) or (2) in example 1, especially with its 'only if' clause. But an account or
justification of how the hearer hits upon that precise q which will ultimately
establish the missing link of relevance between his reaction and the conversa-
tional demand is unavailable in Grice's theory, and likely to be eternally
unavailable if what we look for is a (deductive) logical account. For, finding
q is neither a deductive nor an inductive problem, but precisely an abductive
one. But, again, without such an account, no explanation is given for the
basic fact that implicatures are used and in a fairly reliable way. Certainly a
mere trial and error procedure or brute force strategy could not explain that
fact, for too much time would be necessary in order to test all the hypotheses
that might occur to the hearer. He needs some procedure for pre-selecting
those hypotheses that are worth testing. And the speaker needs to know what
is the procedure - particularly, in what order the hypotheses are going to be
tested - in order to be able to convey the implicatures he wants to convey.
Such a procedure might be a heuristics. A set of (at least partially) ordered
rules that would provide the hearer (and the speaker) with the instruments
for guessing (rather than deducing) the implicatures, but for guessing them in
an educated or systematic, rather than haphazard, way. These rules should,
among other things, exploit the first step in the argument to the maximum
(the detection of the irrelevance of the reaction to the demand). Although this
step is not explicitly formulated in Grice's pattern, its role is fundamental in
168 MARCELO DASCAL
the argument. It triggers the whole process of searching for an implicature

(if there is any - and it is this first judgment that will decide whether there is
any or not). And in so doing, it triggers the process in a certain direction; it
creates, so to speak, guidelines for the search. Among other things, these
guidelines suggest what kind of contextual features to focus upon, thus
avoiding the need for a search through all of the infinitely many faces of the
'context.' Partly, this is achieved because this first assessment of irrelevance
is not just global and undiscriminated, but rather a quite detailed one, that
indicates precisely what aspects of B's reaction are irrelevant. The general
distinction between pragmatic and semantic factors, with the subsequent
necessary characterization of the various layers of the latter (semantic force,
modality, meaning connection, aboutness, etc.), should provide part of the
instruments for making such carefully discriminative irrelevance (or re-
levance) judgments.
Consider a few examples:
(10) A: Why did John beat Mary?
B: Why not?
(a) Identification of demand; "B is to fill in the gap-:
'John beat Mary because .. .'''
(b) Check for force: question.
Inadequate, unless it is a clarificatory question.
(c) Is it a clarificatory question?
(Condition: only if content relates to content of demand)
Check content: Content may be related to content of
demand in the form of a generalized
request of information about the class
of sentences that could be used in order
to fill in the gap.
Implicature 1: "What kind of explanation are you look-
ing for?" or "Why should it be explained
at all?" or "What is there to be ex-
plained?"
If this interpretation holds water (and it should be checked
against the intonational data), then the above implicature
is generated at this stage.1 2 .
(d) If the former interpretation is not satisfactory, check for
other possible forces.
CONVERSATION AL RELEVANCE 169
Implicature 2: Refusal to accept the presupposition that

there is something to explain in connec-
tion with A's question.13
(e) If that does not work, try another possible force.
Implicature 3: Refusal to answer (meaning: It is none of
your business).
Notice that one is led to such an implicature only if no
content connection is found.
Compare with:
B: Why are two and two four?
B's reply in this case cannot be a clarificatory question, given
the total lack of content connection between the demand and
the reaction. Yet, it conveys a different implicature. Something
like "The answer to your question is obvious." This was ob-
tained apparently through a transfer of the characteristics of
the second question to the first one. A kind of mimicry is
involved here.1 4
Consider finally:
B: What is the solution to Fermat's problem?
Here the implicature would be perhaps that it is impossible to
answer the question.
These examples show how the identification of the type of irrelevance
leads to educated guesses about the hypothesis that will explain away such
an irrelevance. The confirmation of the hypothesis depends on the context.
The need for such a confirmation guides the interpreter in his scanning of
contextual features (as the sense of a referential expression guides him into
looking for individuals of a certain kind and not another). In this way, the
range of contextual phenomena that must be scanned by the interpreter is
significantly reduced by the heuristic procedure.
It would be too much to claim that the heuristic rules are applied in a
strict order, but some sort of an order is certainly present.
Consider example (3). Grice does not offer an explanation for the genera-
tion of the implicature in this case, declaring it "obvious" in the light of
example (2). Yet, it seems to me that the two examples differ significantly.
170 MARCELO DASCAL
In (2) there is a clear semantic relation between the terms 'garage' and
'selling petrol.' If indeed there is an implicature in that case, it must have
been generated - in the absence of further details - through the mediation
of such a relation. (Actually I doubt whether it should be considered an
example of implicature, any more than when someone says that John is a
bachelor he implicates that John is unmarried. When the semantic connec-
tion is obvious, the relevance is obvious and there is no need to make use of
the principles of conversation in order to give a satisfactory explanation of
the utterance.) But in (3) there seems to be no semantic relation between
Smith's frequent visits to New York and his having a girl friend there. And
A's statement alone ("Smith doesn't seem to have a girl friend these days")
is not sufficient to assign to B's speech act the specific implicature mentioned
in the example. For, all the requirement of relevance-to-A 's-statement (which
sets up the 'current purpose of the conversation' or 'demand') allows one to
conclude is that B's statement has "something to do with" or "is somehow
connected with" the (alleged) fact that Smith does not have a girl friend
presently. Now, this connection might be, for example, an explanatory one
(Smith does not have a girl friend because he is very busy, going so often to
New York; or, in the other direction: Smith does not have a girl friend,
therefore he goes so often to New York to the prostitutes). In both cases, B is
taken to assent to A's remark and then to go on either to justify it (reinforce
it) better, or, taking it for granted, to use it, by exploring some of its con-
sequences. In Grice's account of the example in question, B's attitude is taken
to be of dissent (No. It is not true that he does not have a girl friend. He has
one in New York.), but justified or informative (B gives his own evidence,
probably unknown to A, for his counter-claim).
Now, it is remarkable that once anyone of these attitudes (or kinds of
force) is selected by the interpreter, the set of possible implicatures is con-
siderably reduced. If the attitude is of assent, for example, then Grice's
proposed implicature is no longer available. If the attitude is of dissent, on
the other hand, B cannot be taken to be either justifying or exploring A's
statement. If, then, in. the initial judgment of irrelevance the force of the
reaction is identifiedjirst (on intonational grounds, for example), this restricts
enormously the number of alternative hypotheses still open for speculation.
On the other hand, contrast (3) with (4): in (3) there is some semantic
relation; both utterances are about Smith. In (4), the absence of any semantic
relation blocks the way to identifying the response as having the force
'assent.'
To the heuristic rules hinted at here, one should add Grice's other super-
maxims, which are, to my mind, subordinated to the general maxim of
relevance. Furthermore, one could lift the simplifying assumptions made
above, thus making available further types of possible explanation of irre-
levance, which should be checked, so it seems, before the other types of
explanation we have been discussing so far. The new heuristic rules to be
introduced by lifting the simplifying assumptions would be, roughly:
(T) Check for topical relevance.
(M) Check for correct identification of demand.
These two rules would allow to explain implicatures generated in the case of
Socrates and Meno described above 15 as well as cases of 'misunderstanding,'
like Hilary Putnam's example:
(13) A: Why did you rob the bank?
B: Because that's where the money is!
In this case, the irrelevance of B's reply to the demand set up by A's
utterance (i.e., A's expectations concerning a possible explanation for what-
ever it is that he thinks should be explained) could be explained away by
means of rule M, whose application might lead to the result: "B misunder-
stood the demand by incorrectly 'reading' the stress, in A's utterance, upon
bank instead of upon rob."
In order to balance off the skepticism about the possibility of handling the
notion of relevance in a theoretically fruitful way with which I opened this
paper, let me conclude with a much more optimistic note:
Relevance is not something mystical; it is a product of the facts of sentences - facts that
are clearly within the domain of linguistics. Relevance is, moreover, the phenomenon
that permits humans to converse; thus it must be accounted of supreme importance in
the working of language in human affairs (Gunter, 1974: 53).
Although I do not agree with the simplification embodied in the claim that
relevance is in the Jacts oj sentences alone, I share the general optimism of
this remark: the task of constructing a satisfactory explication of the notion
of relevance, especially where confined to particular types and well-defined
areas, is not totally hopeless. I hope that this paper has at least posed some
problems and suggested some paths for such an explication to follow.
Tel-Aviv University, Israel

and Universidade Estadual de Campinas, Brazil
172 MARCELO DASCAL
NOTES
I wish to thank Professor Helmut Schnelle and the Deutsche Forschungsgemeinschaft

for the financial support which made possible the preparation of part of this work (Project
AZ Schn 183/10, "Semantische Relationen"). I am grateful also to Ruth Manor for her
helpful suggestions in the course of the preparation of this manuscript, and to Mark
Glouberman for his comments on an early draft.
.1 Moreover, the distinction between the two types of relevance and the consideration of
the multiple ways in which they may interact provide a vantage point from which several
problems in the philosophy of language (e.g., translation, synonymy), in the philosophy of
science (e.g., induction, explanation) and in the theory of knowledge (e.g., Gettier counter-
examples) can be fruitfully reassessed.
2 One might argue that it is the maxim of quantity that has been apparently violated in
this case: B has gone into too much detail, etc. This shows how closely R is related to the
other maxims.
3 By 'what is strictly said' when one utters a sentence p in context C, I mean the standard,
literal interpretation 'normally' assigned to the sentence p in C. The modifier 'strictly' is
required because 'what is said' is too broad in the sense that it may be used to describe
also what is implicated through the utterance of p in a given context.
4 Cf. the notion of dUfuse versus specific significance of social objects in general, intro-
duced by Parsons et ai. (1951: 59-60). The failure to pay attention to the distinction be-
tween the local and the general purpose of a conversation leads to some mistakes in
Grice's analyses (e.g., the analysis of example 4, 1975: 54).
5 In this respect, 'is relevant to' behaves very much like 'is similar to' (cf., for example,
Quine, 1969: 116-121).
6 It should be noted, however, that the degree of precision with which an implicature is
generated is not so high as to be viewed as the conclusion of a conclusive deductive argu-
ment. In particular, I think Grice is wrong in introducing an only if clause as one of the
premises of his implicature-generating arguments (see, for example, step (2) in the im-
plicature-generating argument for example 1). Certainly, explanations other than B's
believing precisely a given q for the apparent lack of relevance of B's utterance could be
available in the specified circumstances. The implicature-generating-argument should
rather be viewed as a case of loose reasoning or of reasoning with loose concepts, in any
case, not as a strictly deductive argument. I shall return to this point later.
7 The use of the pairs of terms 'topic' and 'comment,' 'thema' and 'rhema,' 'new' and
'old' (information) by linguists should not be confused with my terminology here. Whereas
these terms are used to describe features of the internal structure of an utterance (or even
of a sentence), mine refer to the field of attention or 'short term memory' of a speaker-
hearer. An utterance u as a whole, no matter what its internal structure is, may be topically
relevant at t for S; later on, say at t+ 1 the 'old' information (the 'topic') of u may become
marginally relevant for S; if, at the same time, S concentrates his attention on u's 'new'
information (the 'comment'), it then becomes topically relevant for him.
8 Some authors employ the term 'express' for what I call 'indicate.' Unfortunately, I can
think of no term that would refer only to the unintentional conveying of something, thus
being free from the ambiguity of 'indicate,' 'express' and the like.
CON VERSA TIONAL RELEVANCE 173
\I Davidson is right in claiming some sort of 'evidential priority' for 'holding a sentence
true' vis-a-vis interpreting it, on the grounds that one may know that someone holds a
sentence true without knowing how the sentence is to be interpreted. I pointed out
(independentlY) the same fact and discussed some of its implications in Dascal, 1975.
10 One might improve upon the present account by admitting some sort of internal
structure to the forces or kinds of reaction here described. Thus, for example, we might
consider the dissent in this case to be supervenient on the correction. The force might
then be described as 'correction, therefore dissent.'
11 The general pattern of such arguments is, according to Grice (with minor modi-
fications):
(a) he has said that p;
(b) there is no reason to suppose that he is not observing the maxims, or at least
the CP;
(c) he could not be doing this unless he though that q;
(d) he knows (and knows that I know that he knows) that q is required;
(e) he has done nothing to stop me thinking that q;
therefore
(f) he intends me to think, or at least is willing to allow me to think, that q;
and so
(g) he has implicated that q.
12 A further question is whether the presupposition of the clarificatory question, namely,
that A's original question is unclear, is also implicated. I have no definite intuitions about
this. What is beyond doubt, however, is the fact that the rejection of presuppositions of a
question is, in general, an appropriate reaction to the question. The fact that such a rejec-
tion can be conveyed either explicitly or by means of an implicature suggests another
possible interpretation of the apparent irrelevance of the 'why not?' reply. This will be
put to use in step d below.
13 Notice that in this case a semantic relation, namely, presupposition, combines with
considerations pertaining to the level of pragmatic relations in order to generate the
implicature.
14 A similar kind of 'implicature by transfer' occurs in the foIlowing case: at the beginning
of a footbaIl game, the TV commentator discusses the merits of the two teams, concluding
that the one has far better chances to win than the other; "but - he adds - the baIl is
round." Here, the particular tautology chosen-particularly the word 'round' - conveys
the desired implicature.
15 Strictly speaking, a negative result of the application of rule T would eventuaIly
trigger a search for causal factors of B's utterance. This might lead to the interpretation of
the utterance in terms of what it indicates rather than of what it implicates.
BIBLIOGRAPHY
Berlin, I., 1938-39, "Verification," in: The Theory 0/ Meaning, G.H.R. Parkinson (ed.),
Oxford, University Press.
Cohen, L.J., 1971, "Speech Acts," in: Current Trends in Linguistics 12, T.A. Sebeok (ed.),
174 MARCELO DASCAL
The Hague, Mouton.

Dascal, M., 1971, "Empirical Significance and Relevance," Philosophia 1: 81-106.
Dascal, M., 1975, "La Razon y los misterios de la fe segun Leibniz," Revista Latinoa-
americana de Filosofia 1: 193-226.
Davidson, D., 1973, "Radical Interpretation," Dialectica 27: 313-328.
Goodman, N., 1972, "About," in: Problems and Projects, Indianapolis, Bobbs-Merril.
Grice, H.P., 1968, "Logic and Conversation" (Unpublished lecture notes from William
James Lectures at Harvard, 1967).
Grice, H.P., 1975, "Logic and Conversation," in: Syntax and Semantics 3, P. Cole and
J.L. Morgan (eds.), New York, Academic Press (corresponds to the second lecture in
Grice, 1968).
Gunter, R., 1974, Sentences in Dialog, Columbia (South Carolina), Hornbeam Press.
Hare, R.M., 1970, "Meaning and Speech Acts," in: Practical Inferences, London, Mac-
millan Press.
Parsons, T., E.A. Shils, and J. Olds., 1951, "Values, Motives, and Systems of Action," in:
Toward a General Theory of Action, Talcott Parsons and Edward A. Shils (eds.), New
York, Harper and Row.
Quine, W.v.O., 1969, Ontological Relativity and Other Essays, New York, Columbia
University Press.
Weizenbaum, J., 1966, "ELIZA - A Computer Program for the Study of Natural
Language Communication between Man and Machine," Communications of the
Association for Computing Machinery 9: 36-45.
RUTH MANOR
COMMENTS
As commentator, I have the relatively simple task of saying something

relevant to Relevance. It seems that I cannot fail- whatever I say will either
be (semantically) relevant, or it may serve as an example of (semantic)
irrelevance, thus being (pragmatically) relevant.
The topic of relevance is intrinsically complicated and therefore Dascal's
paper should be viewed, as he himself suggested, as an attempt at laying only
a very general foundation for a theory of conversational relevance. In the
following I shall summarize the main theses as I understand them, and pro-
ceed by making some general points of criticism.
The paper deals with two related problems. The first consists in clarifying
Grice's Supermaxim (R) of conversational relevance, and the second is the
problem of identifying implicatures and their relations to the concept of
relevance. Assuming that conversations are governed by some principle
similar to (R), Dascal poses two questions regarding relevance:
(1) In what respects can a given speech act performed in the
conversation be relevant (to the conversation)?
(2) How, in what respects and by what criteria, ought these
"kinds" of relevance be assessed?
Dascal answers (1) by distinguishing between semantic and pragmatic
relevance. Relevance is (at least) a dyadic relation; its domainisa set of speech
acts or a set of propositions, while its range is (following Grice) "the local
purpose of the conversation." The different elements in the domain (speech
acts and propositions) bring about the distinction between semantic and
pragmatic relevance. Pragmatic relevance is a relation between speech acts
and certain goals. It should be explicated as part of a theory of goal-directed
behavior. Semantic relevance, on the other hand, is a relation between pro-
positions or sets of propositions, and its characterization is a fundamental
task of semantic theory.
In discussing the range of the relevance relation, Dascal points out several
difficulties involved in the concept of "the local purpose of the conversation,"
but he does not clarify it. However, when he limits the discussion to simple
175
176 RUTH MANOR
conversations of two consecutive speech acts performed by agents A and B

respectively, he seems to restrict the "local purpose" to the "conversational
demand" set by A's speech act. This "conversational demand" is determined
by the content, force, intonation and possibly other elements of A's utterance.
B's reaction is relevant to A's utterance if it complies with the "demand":
semantically, if its semantic force and propositional content "fit" the demand,
and pragmatically, if its illocutionary force, intonation, and the kind of
reaction it is, "fit" the demand.
Dascal does not provide a general answer to his question (2), beyond
indicating that the different kinds of relevance should be explicated within
different theories. He does, however, suggest that his treatment of the pro-
blem of identifying implicatures provides some suggestions towards answer-
ing the problem of finding criteria for relevance. He defines an implicature as
"a hypothesis about the speaker's intentions that explains away the apparent
irrelevance of his utterance by explaining how the utterance is in fact
relevant." Hence, since the process of detecting implicatures involves stages
in which we judge whether or not an utterance is relevant, and stages in which
we judge whether or not a proposition explains how an utterance is relevant,
Dascal seems to indicate that the problem of criteria is attacked there.
Now, implicatures are detected after we have decided that B's reaction is,
at least apparently, semantically or pragmatically irrelevant to the conver-
sational demand set by A's utterance. We try, by a trial-and-error procedure
to (educatedly) guess the implicature that may serve as the "explanatory
hypothesis." We try a certain hypothesis, ifit does not "work" we try another
hypothesis, until we either find one that "works" or, I presume, we get tired
and we decide that B's utterance is simply and actually irrelevant to the
conversation.
The most natural distinction between pragmatic and semantic relevance is,
I believe, among the important contributions to the clarification of the
concept of relevance. The distinction enables us to divide the vast problem
of characterization and of finding a criterion for relevance into two better
defined problems seeking their solution within different and limited theories.
Furthermore, it enables us to restrict the range of possible candidates for
"explanatory hypotheses," depending on whether the apparent irrelevance is
pragmatic or semantic. However, in Dascal's outline, these advantages are
gained with difficulty.
The concept of relevance is imprecise if not essentially vague and, as
Dascal says: "the task of replacing the vague and comfortable notion of
COMMENTS 177
relevance by something that even remotely approaches the precision sug-

gested by the term 'criterion' is notoriously difficult. Attempts to do so in
other areas were doomed to failure ... " Hence, he avoids dealing directly
with the problem of criteria and attempts, in this paper, to solve a lesser
puzzle, namely, "How can such precise messages as the implicatures be
conveyed via a reasoning that makes essential use of such an imprecise notion
as relevance." This attempt fails, since in the analysis of implicatures he
presupposes a solution to the "worst" problem of the criteria. He first makes
a "simplifying" assumption that A's utterance at time t is (topically) relevant
for B at t + 1 and, later on, he lifts this assumption adding the new step in his
heuristics (T): "Check for topical relevance." How this checking is to be
performed is not indicated. The term "Check" suggests he is assuming the
availability of precise criteria of relevance, thus presupposing a solution to
the "notoriously difficult" problem he tried to avoid.
Even if we follow Dascal and assume that the "local purpose of the
conversation" or the "conversational demand" set up by A's utterance is
given and that we can check whether B's reaction to it is relevant, still the
very concept of "conversational demand" as it is used in reference to the
examples given is problematic, and possibly even more problematic than that
of "the purpose of the conversation." Dascal restricts his examples to safe
grounds when in all of them A's utterance is a question. It may be safe to say
that questions normally set some demand in a similar sense to Dascal's.
Questions normally call for an answer, and therefore we can normally
characterize reactions to questions as answering, rejecting the question,
rejecting the asking of the question, refusal to answer (and possibly other re-
actions) as (pragmatically) relevant, while a reaction of ignoring the question
is (at least apparently) irrelevant. However, not all speech acts in general set
up such specific "conversational demands," and some do not set up any
demand for a relevant reaction (either pragmatically or semantically). For
instance, if I tell a joke I hope (but I do not demand) that the addressee's
reaction will consist of a roaring laugh, but any reaction whatsoever will be
relevant to the conversation. Moreover, if he does volunteer a "too (seman-
tically) relevant" reaction, performing a post-mortem on the joke, I may be
disappointed and he may be impolite. Again, if I throw all my personal
problems upon your shoulders by confessing to all my sins, there most
likely will be an implicit understanding between us that (for the sake of
friendship), whatever your reaction is, it should not be (semantically)
relevant. Finally, I may pass you by saying "It's a nice day," not having any
178 RUTH MANOR
expectations nor any demands upon your reaction - any action you will
take will be as relevant as is required by my utterance. Indeed, when greeting
some of my friends, I do not expect even acknowledgement, thus my utter-
ance in the context does not pose any demand for reaction.
The above examples suggest that the supermaxim (R) of relevance should
be qualified, namely, "be as (pragmatically and semantically) relevant as is
required," and not unqualified, as both Grice and Dascal suggest. But if this
observation is correct, then Dascal's task becomes even more complicated
than it was before. In order to find an implicature, we now have to determine
whether A's utterance requires any relevance, what kind and, possibly, to
what extent, before we can judge whether there is any implicature to be
sought for. Moreover, if there are cases where there is a "demand" for
(semantic) irrelevance, the following question is raised: If B's reaction is
(semantically) relevant to A's utterance in spite of the opposing "demand,"
can this lead to the creation of implicatures as well? It is possible that the
following examples involve such cases.
(1) A: "I don't know who killed Jones, and that's that."
B: "Who killed Jones?"
Implicature: "You (A) are lying."
(2) A passes B and chooses to utter something that will kill the
silence without starting a conversation.
A: "It's a nice day."
B: "You really think so?"
Implicature: Something like the famous (depending on
intonation) "If you have nothing to say, don't
say it."
I find Dascal's discussion of implicature plausible and intuitive. In par-
ticular, the description of the implicature sought as an explanatory hypo-
thesis, explaining away the apparent irrelevance, seems natural. However, I
do not think that in actual conversations we seek implicatures by a trial-and-
error process, but rather that the process is a one shot kill: If we decide that
some implicature exists and we "got" it, we attach greater or lesser degree of
confidence to our discovery; but in order to decide that we failed and should
try another hypothesis, it seems that some new information must be given.
But the method of trial-and-error is objectionable on theoretical grounds as
well, for it requires independent means of identifying the errors in the hypo-
theses we try. Indeed, the heuristics in the examples include a step of checking
COMMENTS 179
whether a given hypothesis works, whether "the interpretation is satis-

factory." This step, which is the step of "confirmation" of the hypothesis,
depends on the context. In other words, Dascal hints here that there is a test
(or tests) by which, using contextual data, we can determine whether a
hypothesis (which has already passed the previous steps in the implicature-
game) works or not. But how does such a test work? And if such a test exists,
does it not render the other steps redundant? Dascal does not elaborate, nor
does he describe how such a test might work in reference to his examples.
As a final point I wish to make a suggestion which I believe to be in
accordance with Dascal's aims and which may help overcome the main
difficulties in his theory. Let me first summarize the main objections.
Dascal's puzzle, that "so precise messages as the implicatures are conveyed
via a reasoning that makes essential use of such an imprecise notion as
relevance," remains a puzzle under his treatment. First, because when dealing
with implicatures he ignores the imprecise nature of relevance and presup-
poses the existence for a precise test for (topical) relevance. Secondly, even
under this assumption, the heuristic described is insufficient for the precise
determination of implicatures unless we employ an unspecified test of
"contextual confirmation." Thus, it seems that neither was the imprecision of
relevance removed or used, nor was the precision of implicatures conveyed.
It is possible that at least part of the imprecision of the notion of relevance
is due to the fact that it covers too many sorts of things, and that it could be
made much more precise once we divide it into separate "areas" (or con-
texts) and deal with each separately. One such move was made in distinguish-
ing semantic from pragmatic relevance. But this may not be sufficient.
Returning to Grice's terms, the accepted purpose or direction of a conversa-
tion may indicate that the context of the conversation may play an important
role in determining what is relevant to the conversation, a more essential role
than the concept of "conversational demand" suggests. More specifically,
conversations are governed by certain rules of different kinds besides being
governed by logical rules as the ones suggested by Grice. The applicability of
these rules is determined by the context of the conversation. For example,
certain sudden changes in the physical environment of the conversation may
call for, or at least allow for a change of the topic (recall Dascal's Meno
example). Similarly, conversations are governed by what we may call "rules
of etiquette," indicating what we can say (and expect other conversants to
say) and in which manner, in different situations. Certain things are not
mentioned in public (without risking being called "impolite," "rude," or the
180 RUTH MANOR
like). What I can say to or ask a stranger, and how I can express myself to
him (within the limits of polite behavior) is different from the manner I may
(politely) express myself to a friend. The point here is not that we are always
polite, but that we are normally expected to be polite and expect other people
to be the same. These expectations determine to some extent what is relevant,
both semantically and pragmatically, to the conversation and, often, devia-
tions generate implicatures. (For example, "Will you be so kind and open
the window, please?" is fine when I address a stranger, but when I address
my kids they know they have done something wrong again.)
Besides these rules of etiquette, the range of relevance to a conversation
may be determined also by implicit or explicit agreement between the partici-
pants. For example, it may be that within a semantic theory the proposition
that (P) Golda Meir is not bald, is judged to be semantically relevant to the
proposition (q) that Golda Meir is hairy. But suppose I hold a conversation
with a child who does not know the meaning of 'hairy,' and I am well aware
of this fact. Then, if he asks me whether (P) is true, my reaction affirming (q)
is (apparently) semantically irrelevant to the conversation. If the kid is bright,
he may identify the implicature and learn that whoever is hairy is not bald.
The above examples suggest that in general Relevance is at least a triadic
relation, adding the context as a third argument to both relations of semantic
and pragmatic relevance. But more significantly, these examples suggest that
if we consider the specific non-logical rules governing conversations in
different contexts, this limits tremenduously the candidates for implicatures.
Such an approach, if viable, may explain how such an imprecise notion as
relevance can be made precise in context and thus can be used in conveying
such precise messages as are implicatures (without the need to describe the
process as a trial-and-error process). \
Tel-A viv University

JOHN R. SEARLE
INTENTIONALITY AND THE USE OF LANGUAGE
I
It is possible to make a reasonably clear distinction between intensionality
with an s and intentionality with a t. Without putting too fine a point on it
one can say that intensionality-with-an-s is a property of a certain class of
sentences. A sentence is intensional if literal utterances of it have at least one
interpretation where they fail to satisfy one or more of the standard tests for
extensionality. The two tests most relevant to the present discussion are these:
if existential generalization over the occurrence of referring expressions is not
a valid form of inference or if the sentence fails to allow the substitution salva
veritate of expressions which normally have the same reference, then it is
intensional-with-an-s. Thus for example, the sentence "John is looking for
the lost city of Atlantis" has at least one literal use to make a statement which
does not entail that there is a lost city such that John is looking for it. And
the sentence "The sheriff believes that Mr. Howard is an honest man" has a
literal use to make a statement which together with the true statement that
Mr. Howard is identical with Jesse James does not entail that the sheriff
believes that Jesse James is an honest man. Because intensional sentences
normally derive their intensionality from the occurrence of certain expressions
it is possible to speak not only of intensional sentences, but also of intensional
verbs, intensional contexts, etc. It is also possible to speak of intensional
statements and intensional propositions. No doubt these two tests need some
refinement to enable them to cope with sentences other than those used to
make statements but the basic idea of intensionality-with-an-s seems
reasonably straightforward.
When it comes to intentionality-with-a-t it is much harder to say
exactly what it is we are talking about. It is common to say that inten-
tionality-with-a-t is not primarily a property of sentences, but a property
of some or perhaps even all mental phenomena. It is the property which
mental states have of being directed at objects or states of affairs. Thus for
example, a belief is always a belief that such and such is the case, a fear is
always - or at least in general - a fear of something, a desire is always a
181
Copyright ~ 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
182 JOHN R. SEARLE
desire to have something or that something be the case. Just so we have some
fairly clear idea of what it is we are talking about, I propose the following as
a rough preliminary test for intentionality-with-a-t. A mental state is an
intentional state if and only if the specification of the content of that mental
state requires the specification of some object or state of affairs which is not
identical with that mental state. On this test pains and aches and at least some
cases of anxiety are not intentional, whereas beliefs, hopes, expectations and
desires are. A specification of the content of my pain is just a further des-
cription of the pain; but the specification of the content of my belief, hope,
expectation or desire must specify what it is that I believe, what I hope for,
what I expect, and what I desire. All that a pain has to have to be a pain is to
have a certain feel to it, certain phenomenal qualities, but for belief, hope,
expectation, and desire, something more is required: the questions "what do
you believe, what do you hope for, what do you expect, what do you desire?"
must have answers, if the agent can be said to have a belief, hope, desire, or
expectation at all; and those answers will specify objects and states of affairs
that are not identical with the mental states.
In Wittgenstein's jargon, pains and aches have causes but not targets,
whereas love and hate, belief and desire have both causes and targets, and in
each case the cause mayor may not be identical with the target. I will call
such mental states intentional states, and the objects and states of affairs at
which they are directed intentional objects. The notion of an intentional
object is a frequent source of confusion in philosophy and I will have more
to say about it later on. We will see that for the so-called propositional atti-
tudes it is not really necessary at all. For the present it is important to note
that on this criterion not all mental states are intentional states; only those
which require the specification of an intentional object are intentional states.
It is also important to note that this criterion is not intended as an analysis of
the notion of intentionality. If so, it would be hopelessly inadequate since it
rests on several unexplained and obscure notions: in what sense does a mental
state have a "content" and what is meant by saying the "specification" of that
content "requires" the specification of something else?
The test for intentionality-with-a-t is designed to isolate those mental
states which are in some sense directed at objects and states of affairs from
those which are not; but even assuming that it is successful in doing that, it
raises but so far leaves unanswered the hard questions about intentionality.
The two most pressing questions raised by the criterion are these: the state-
ments specifying the object or state of affairs in the specifications of the
INTENTIONALITY AND THE USE OF LANGUAGE 183
intentional state are not like ordinary statements containing specifications of

objects and states of affairs because they do not require the existence of the
objects or states of affairs in order to be true. It can, for example, be true that
John expected it would rain and that he worships God without it being true
either that it rained or that God exists. Indeed, the sentences which specify
intentional states are themselves intensional-with-an-s on both our
criteria for intensionality-with-an-s. So our first question is: why is that so?
How does intentionality-with-a-t give rise to intensionality-with-an-s?
The second question is this: what exactly is the relation (or set of relations)
between intentional states and their objects, and what is it about this relation
that requires the specification of an intentional object in the specification of
an intentional state? I think an answer to the second question will enable us
to answer the first and to clarify the notion of an intentional object.
In discussing the second question, Wittgenstein frequently makes remarks
like the following: "If someone could see the expectation itself - he would
have to see what is being expected" and "The representation of a wish is
eo ipso the representation of its fulfillment."! Wittgenstein himself and
Kenny in his book on Wittgenstein claim that there is an "internal relation"
between the intentional state and its intentional object. But this characteriza-
tion is not much help to us unless we are told exactly what is meant by the
notion of an internal relation. If for example, someone tells us that there is
an internal relation between being a triangle and being three sided, it is
possible to make some fairly clear sense of the notion of an internal relation:
a logically necessary condition of being a triangle is being three sided. But in
that sense there clearly is no internal relation between my belief that it will
rain on Wednesday and the state of affairs of its raining on Wednesday. The
proposition that it will rain on Wednesday neither entails nor is entailed by
the proposition that I believe that it will rain on Wednesday. It seems that
the only sense we can give to the notion of an internal relation in this case is
one we have already spelled out in our test for intentional states - the
specification of an intentional mental state requires a specification of its
intentional object. But that still leaves us with our second question un-
answered. What is it exactly about intentional states that relates them to their
objects? Mackie poses our second question as sharply as anyone. "At last we
are beginning to get into focus the real puzzle about intentionality" he
writes, "Is it not strange that there should be one state of affairs (Tom's state
of believing, hoping, fearing, or whatever it may be) that requires for its
adequate description the partial, incomplete, selective, indeterminate, des-
184 JOHN R. SEARLE
cription of a quite different and so far merely possible state of affairs?"2 I

think there is a fairly simple answer to Mackie's question. To see this answer
imagine a parallel question about statements and speech acts generally.
Suppose someone said "It is not strange that there should be one state of
affairs (Tom's statement that it is raining or his order to Bill to leave the
room) that requires for its adequate description the partial, incomplete, etc.,
description of a quite different and so far merely possible state of affairs (the
state of affairs that it is raining or that Bill leaves the room). But there is
nothing puzzling about this requirement once you recognize that a statement
is precisely a representation of a state of affairs that is asserted to exist and
an order is a representation of a state of affairs which the hearer is ordered
to bring into existence. Any speech act with a propositional content contains
a representation of some object or state of affairs; and in the same way, I
wish to argue, all intentional states are representations of objects and states
of affairs. The reason that the specification of my belief that it is raining
requires a specification of the state of affairs that it is raining is exactly the
same reason that the specification of my statement that it is raining requires a
specification of the state of affairs that it is raining: in each case the one is a
representation of the other.
The answer to our second question then, how does the intentional state
relate to the intentional object and what is it about the relation that requires
that a specification of one involves a specification of the other, is simply that
the intentional state contains a representation of the intentional object in the
same sense that speech acts contain representations of objects and states of
affairs.
Because an intentional state contains a representation we can now give a
clear sense to the notion that it is "internally related" to the object it re-
presents: Any representation is internally related to its object in the sense
that it could not be that representation if it did not have that object. Thus,
for example, an identity criterion of my belief that it is raining is that it must
have as its intentional object the state of affairs that it is raining.
I do not much like having to invoke the notion of "representation" here
because it has so little explanatory power, but it is at least correct and it does
enable us to answer both our second and now our first question, How does
intentionality-with-a-t give rise to intensionality-with-an-s? Sentences
about intentional states are at least in part about representations. That being
the case, their truth conditions will sometimes depend on features of the
representation and not entirely on features of - or even on the existence of -
the object represented. To be looking for the lost city of Atlantis is to have a
representation that one seeks to instantiate and the statement that one has
such a representation can be true even where there is no such instantiation.
And similarly to believe that Mr. Howard is an honest man is in part to have
a mental representation associated with the name "Mr. Howard" which may
be quite different from the mental representation associated with the name
"Jesse James," hence the failure of the substitutability of such names even
where Mr. Howard is identical with Jesse James.
This conception of intentionality as representation will also enable us to
get clear about the nature of mtentional objects. People often talk as if
intentional objects had some peculiar ontological status and had to be dis-
tinguished from actual objects. But the intentional object of a mental state is
just the actual object or state of affairs represented by an intentional state. If
there is no such object or state of affairs then the intentional state does not
have an intentional object though it does still contain a representation. We
need therefore to distinguish the representative content of a mental state from
the intentional object of that mental state. If John loves Sally and believes
that it is raining then the intentional object of his love is Sally, the actual flesh
and blood Sally and not some mental phenomenon, and the intentional object
of his belief is the state of affairs in the world that it is raining and not the
proposition that it is raining. In order that his love should be of Sally and his
belief be that it is raining he must have some representation of Sally and of
the state of affairs that it is raining. But the object of his intentional states is
not these representations, rather the intentional states are directed at their
objects by way of their representative content. A proposition one might say
is not the object of a belief, it is the content of the belief. The oscillation be-
tween the extensional and the intensional reading of statements about in-
tentional states is precisely an oscillation in the extent to which the statement
is committed only to facts about the representative content or to facts about
the intentional objects. 3
The distinction between the representative content and the intentional
object is parallel to Frege's distinction between sense and reference. Just as a
definite description refers to an object in virtue of its sense, but does not
thereby refer to its sense, so an intentional state is directed at an object in
virtue of its representative content, but is not thereby directed at its represen-
tative content. Both sentences describing intentional states and sentences
describing acts of referring are subject to extensional and intensional inter-
pretations and both for the same reasons. Thus "John referred to the King
186 JOHN R. SEARLE
of France" and "John thought about the King of France" have both ex-
tensional and intensional readings, depending on whether they are construed
as about objects of representations or solely about the representations
themselves.
The reason that many authors have failed to see that the intentional object
is identical with the actual object is that the specification of the intentional
object is by way of the aspect under which it is represented by the represen-
tative content, and for that reason it is an intensional-with-an-s specifica-
tion. To put it crudely, it has seemed to them that the intentional object is
incomplete in ways that actual objects are not incomplete and that therefore
intentional objects can never be actual objects. How can the actual object be
identical with the intentional object when the actual object has all sorts of
features the intentional object does not have? Thus Davidson writes:
What is less obvious, at least until we attend to it is that the event whose occurrence makes
I turned on the light true cannot be caIled the object, however intensional, of I wanted to
turn on the light. If I turned on the light, then I must have done it at a precise moment,
in a particular way - every detail is fixed, but it makes no sense to demand that my want
be directed at an action performed at anyone moment or done in some unique manner.
Anyone of an indefinitely large number of actions would satisfy that want, and can be
considered equaIly eligible as its object. 4
But one might as well argue that a description can never be a description of
an actual event because events have all sorts of features not included in the
description, and any number of other possible events could have satisfied the
description. I want to argue on the contrary that since the specification of a
want will specify an intentional object under certain aspects, namely those
under which it is wanted, it follows only that the specification of the want is
intensional-with-an-s. For example, from the fact that I want E, and E is
identical with F, it does not follow that I want F.
But it does not follow from the fact that specifications of wants are in-
tensional that actual events and states of affairs cannot be the objects of
wants. To conclude otherwise is to confuse properties of the specification of
wants with properties of wants, a common confusion which we will explore
in a moment.
If we were to generalize the form of Davidson's argument in a way I am
sure he never intended, it would come out as an invalid derivation of a false
conclusion from two true premises.
1. An intentional state represents its intentional object under

certain aspects only.
2. Actual objects (states of affairs, etc.) have all sorts of features
under which they are not represented by intentional states.
3. .". Actual objects can never be intentional objects.
But 3. does not follow from 1. and 2. An exactly parallel argument would
be
1* A description of an object represents the described object only
under certain of its aspects.
2* Actual objects have all sorts of features under which they are
not represented by descriptions.
3* .". Actual objects can never be described objects.
But why has the first form of argument seemed so plausible in a way that
the second has not? And why in general do philosophers say such weird
things about intentional objects? A large part of the answer I think is that
discussions of intentionality suffer from a confusion of levels which is some-
what analogous to a use-mention confusion. Because sentences about in-
tentional states are (often) intensional-with-an-s, it is tempting to conclude
that somehow the intentional states themselves are intensional-with-an-s, that
somehow they are not really directed at objects but at their own representa-
tive content. But this is to confuse properties of the description of intentional
states with properties of the state described, to confuse the "mention" of the
mental state with its "use." Specifications of intentional mental states are
indeed (in general) intensional-with-an-s because they are about re-
presentations. But intentional mental states are not about representations,
they are representations. In exactly the same way the description of a des-
cription of an object may be intensional but it does not follow from that, nor
is it generally the case, that descriptions of objects are intensional. Just as
ground floor descriptions of objects are extensional, so ground floor inten-
tional states are extensional. That is, insofar as it makes sense to apply the
terms "intensional" and "extensional" to mental states,S there is nothing
inherently intensional about intentionality. Just as my statement that Caesar
was Emperor of Rome is extensional, so my belief that Caesar was Emperor
of Rome is extensional and for the same reasons. Of course your statement
that I stated that Caesar was Emperor of Rome is intensional; just as your
statement that I believe that Caesar is Emperor of Rome is intensional and
188 JOHN R. SEARLE
again for the same reason: the first pair represents a state of affairs, the second
pair represents representations of a state of affairs. This confusion between
representations and representations of representations is quite pervasive.
Thus, it is often said that propositions are intensional entities. But there is
nothing intensional about, say, the proposition that Caesar was Emperor of
Rome. It is as extensional as it can be. Of course the expression "The pro-
position that Caesar was Emperor of Rome" is intensional on both our
criteria. But it is a use-mention confusion to confuse features of the expres-
sion with features of what the expression is about. Sentences about proposi-
tions are intensional and sentences about mental states are intensional but
propositions are not in general intensional and mental states are not in
general intensional.
In favor of the view that there is something intrinsically intensional about
intentionality, it is sometimes pointed out that a man can have an intentional
mental state even though no object or state of affairs satisfies the content of
his intentional mental state. A man can expect rain and worship God even if
it does not rain and God does not exist. But analogously a man can predict
rain and assert the existence of God even if it does not rain and God does not
exist. Just as his expectation may be unfulfilled and his worship directed at
nothing, so his prediction and his assertion may be false. For the case of
so-called propositional intentional attitudes what corresponds to the truth
conditions of a statement is the state of affairs which satisfies the represen-
tative content of the attitude, what I have been calling the intentional object.
For the non-propositional cases, like love and hate, what corresponds to the
object one refers to in the use of a referring expression is the intentional object
that one's attitude is directed at. And just as the fact that one may make a
statement which is false does not show that one's statement is intensional
or one may fail to refer does not show that reference is intensional, so the
fact that one's intentional states may not have intentional objects does not
in the least show that they are intensional.
The picture of intentional states that is emerging from this discussion is
this: Every intentional state consists of a representative content in a certain
mode. The same content can occur in different modes, as for example, when I
believe it will rain, hope it will rain, want it to rain, etc. It is necessary to dis-
tinguish the representative content from the intentional object. An inten-
tional state will have an intentional object if and only if its representative
content is satisfied by an object or state of affairs. I have so far said nothing
about how these representative contents are realized, whether by words,
images, dispositions to behavior or neurophysiological states, or what not,

and 1 do not think it matters much for the purposes of the discussion so far.
On this account, there is nothing ontologically peculiar about intentional
objects; they are just ordinary objects and states of affairs at which our mental
states are directed. To say that Sally is the intentional object of Bill's love of
Sally is just like saying Sally is the described object of Bill's description of
Sally, it assigns no ontological peculiarity to Sally. Intentionality-with-a-t
and intensionality-with-an-s are not even remotely in the same line of
business. Intentionality-with-a-t is so to speak a ground floor property of
the mind. It is how the mind grasps other things. But intensionality-with-
an-s is primarily a property of sentences and other forms of representation.
Some though not all intensional sentences are about intentionality-with-
a-t.
Broadly speaking, we can divide intentional states into two kinds: those
like belief and desire that represent states of affairs (the so-called proposi-
tional attitudes) and those like love and hate that can also represent objects,
events, etc. 1 have so far been speaking of the intentional objects as if they
could be either particulars or states of affairs. But that can be misleading.
States of affairs are not particulars, nor are they anything like particulars. If
my belief that it is raining is correct, then what stands to my belief as
"intentional object" is the same as what stands to my true statement that it is
raining as truth conditions. We can if we like call that an object, but to do so
encourages a mistake analogous to the mistake of assimilating stating to
referring. John's love for Sally has a genuine object, viz., Sally. John's belief
that it is raining will be correct under certain conditions, but those conditions
are not an object like Sally, and it does not represent those conditions in the
way that "Sally" represents Sally, but rather in the way that the statement
that it is raining represents those conditions. We can continue to talk of
intentional objects of propositional attitudes but it is not essential to do so,
anymore than it is essential to talk of objects of statements or orders, and if
we do so we must avoid confusing these with particulars.
A brief digression about "desire" and "want" before we continue: Desires
are propositional attitudes; they always have states of affairs as "intentional
objects" and never simply objects. In the fOlmal mode: "desire" and "want"
are like "believe" and unlike "love" and "hate" in that they must take a
sentential complement as a grammatical direct object. There is a very simple
syntactical argument in favor of this conclusion. Consider the sentence "I
want your house next summer." What does "next summer" modify? It can-
190 JOHN R. SEARLE
not be "want," for the sentence does not mean "I next summer want your
house." It means "I now want your house next summer," and the meaning of
that can be represented as "I now want that I have your house next summer."
Since any sentence of the form "I want x" admits of similar adverbial mo-
difiers, it would seem that wanting requires a propositional representative
content.
II
Because there ar.e so many close connections between intentional states and
speech acts, it is tempting to suppose that the explanation of intentionality
must be linguistic. That is, it is tempting to suppose that the representations
which intentional states contain must be themselves sentences or at least part
of some language-like system of representation. It is in this spirit that various
philosophers have analyzed belief in terms of dispositions to assert or assent
to sentences and that others have proposed that to have an intention to do A
is to say to oneself "I will do A." On this view (a large part of) the philosophy
of mind is a branch of the philosophy of language. And clearly there is much
plausibility and indeed some truth to this approach: for all but the simplest
intentional states, a being could not even have the state unless he had the
linguistic capacity to give expression to the state, as Wittgenstein is constantly
reminding us. But it seems to me equally clear that this view cannot be right.
Beings without language and without language-like systems of representation
can have beliefs, intentions, desires, and expectations. Only those in the grip
of a philosophical theory would deny that dogs and small children can, say,
desire bones and milk respectively. Furthermore, the direction of explanation
seems to be wrong. Among other things what the possession of language
enables us to do is give expression to beliefs, desires, and other intentional
states. True, given the system of representation provided by language, we can
have vastly richer and more complex intentional states than we could without
language. But language does not create intentionality; rather, as I shall argue,
in an important sense intentionality provides the foundation for linguistic
acts. On this view the philosophy of language is a branch of the philosophy
of mind.
As a step toward arguing in favor of this view, let us begin by making
explicit several of the analogies and connections between intentional states
and linguistic acts.
First, the distinction within the theory of speech acts between propositional
content and illocutionary force carries over to intentional states. Just as we

can say that my prediction that you will shut the door and my order to you
to shut the door both contain the same propositional content, namely that
you will shut the door, so we can say that my belief that you will shut the
door and my desire that you should shut the door both contain the same
representative content, namely that you will shut the door. Furthermore, the
state of affairs that makes the statement true or the order obeyed is exactly
the same state of affairs that makes the belief correct and the desire fulfilled.
Let us introduce the notion of satisfaction to describe what is common to the
notions of an order being obeyed, a statement being true, a desire being ful-
filled, an expectation coming to pass, etc.: a mental state or linguistic act is
satisfied if and only if the state of affairs specified in the propositional content
actuaIIy exists. We can then say that the conditions of satisfaction of
speech acts are closely paraIIeled by the conditions of satisfaction of in-
tentional states. More about this paraIIelism in a moment.
Secondly, the distinction in direction of fit that applies to speech acts also
applies to intentional states. The distinction in direction of fit between an
order and a statement can be characterized by saying that in the case of an
order, it is the responsibility of the person ordered to make his behavior
match the propositional content of the order (and hence orders have the
world-to-word direction of fit);6 whereas in the case of a statement, it is the
responsibility of the speaker to make the propositional content of his state-
ment match the world (and hence statements have the word-to-world
direction of fit). Apologies, thanks, and congratulations have no direction of
fit because in their performance the speaker takes for granted that the pro-
positional content is already satisfied. When I apologize for stepping on
your toe, I neither assert that your toe has been stepped on, nor do I try to
get your toe stepped on. Rather, I presuppose that it has been stepped on.
These distinctions carryover exactly to intentional states. Belief, one might
say, has the mind-to-world direction of fit, because if my belief is mistaken it
is the belief and not the world which is at fault. Desire has the world-to-mind
direction of fit because if my desire is unfulfiIIed it is so to speak the world
and not the desire which is at fault. Sorrow, gratitude, and pleasure have no
direction of fit, rather each of them contains a corresponding bebef that the
propositional content is satisfied. Thus in order to be sorry for stepping on
your toe I must believe that I did in fact step on your toe.
A third connection between intentional states and linguistic acts underlies
the first two. In general, a speech act with a propositional content is an ex-
192 JOHN R. SEARLE
pression ofthe corresponding intentional state, and the propositional content

of the speech act is identical with the representative content of the intentional
state. Thus when one states that p one expresses (whether sincerely or
insincerely) the belief that p, when one orders that H does A one expresses a
wish or desire that H does A, when one promises to do A one expresses an
intention to do A, when one apologizes, thanks, or congratulates one ex-
presses sorrow, gratitude, or pleasure about the states of affairs for which one
is apologizing, thanking, or congratulating. Furthermore, the expression of
the intentional state is not a mere accompaniment: there is an internal con-
nection in the strict sense between the performance of the speech act and the
expression of the corresponding psychological state, as is shown by Moore's
paradox. One cannot say, "It's raining but I don't believe it's raining," "I order
you to leave but I don't want you to leave," "I promise to come but I don't
intend to come," "I apologize but I am not sorry," "thanks for giving me the
money, but I am not glad you gave me the money." In each case the absurdity
derives from the fact that one is denying the presence of the intentional state
which was expressed in the immediately preceding portion of the utterance.
Of course one may express an intentional state one does not have, and that is
how insincerity in speech acts is possible. Indeed, each of these intentional
states is the sincerity condition of the corresponding speech act, because its
presence or absence determines what constitutes sincerity or insincerity in the
performance of the speech act. When one makes a statement, for example,
one necessarily expresses a belief but one does not necessarily have the belief
one expresses.
These three connections show the close parallelism of speech acts and
intentional states. Both speech acts and intentional states present a proposi-
tional content in a certain mode - a psychological mode in the case of
intentional states and an illocutionary mode in the case of speech acts. But
in the performance of the speech act there are not two things going on - the
performance of the act and the expression of the corresponding psychological
state. Rather the performance of the act is eo ipso an eXPlession of the cor-
responding intentional state and the propositional content of act and state
are identical. Where there is a direction of fit, the speech act will be satisfied
if and only if the intentional state is satisfied. Thus the order that p will be
obeyed if and only if the desire that p is fulfilled, the statement that p will be
true if and only if the belief that p is correct, the promise that p will be kept
if and only if the intention that p is carried out, etc.
Intentionality-with-a-t thus infects nearly all uses of language and this
fact requires explanation. A tempting but obviously false explanation is to

say that speech acts just are expressions of psychological states and that is all
there is to it. While this explanation has some plausibility for the expressive
class of speech acts - such as apologizing, thanking, congratulating, etc.'--it
is obviously false for statements, orders, promises and a large number of
others, false because there is more to a statement than just an expression of
belief, more to a promise than just an expression of intention and more to an
order than just an expression of a desire. But what more is there? And what
is the relation between this "more" and the underlying intentional states?
As a way of getting at the answers to these questions, imagine a class of
beings who were capable of having intentional states like belief, desire, and
intention but who did not have a language. What more would they require in
order to be able to perfolm linguistic acts? Notice that there is nothing fan-
ciful in the supposition of beings in such a state, since as far as we know the
human species once was in that state. Notice also that the question is
conceptual and not historical or genetic. I am not asking what additions
would need to be made to their brains or how language did evolve in the
history of the human race.
When we have ascribed to our beings the capacity for having intentional
states we have already ascribed to them the capacity for relating their in-
tentional states to objects and states of affairs in the world. The reason for
this is that a being capable of having intentional states must be capable of an
awareness of the conditions under which its intentional states would be
satisfied. For example, a being capable of having desires must be capable of
an awareness of the satisfaction or frustration of its desires, and a being
capable of intentions must be capable of recognizing the fulfillment or frus-
tration of its intentions. And this can be generalized: For any intentional
state, a being that has that state must be able to distinguish the satisfaction
from the frustration of that state. 7 This follows from the fact that an inten-
tional state is a representation of the conditions of its satisfaction. This does
not mean that such beings will always or even most of the time get it right,
that they will not make mistakes; rather, it means that they must have an
understanding of what it would be to get it right.
Now back to our question: what more would such beings have to have in
order to have a language? The question needs to be made narrower, because
there are all sorts of features of actual languages that are irrelevant for our
present discussion. Presumably such beings would need a recursive device
capable of generating an infinite number of representations, they would need
194 JOHN R. SEARLE
quantifiers, logical connectives, modal and deontic operators, tenses, color

words, etc. The question I am asking is much narrower. What would they
need in order to get from having intentional states to performing illocutionary
acts?
The first thing that our beings would need to perform illocutionary acts is
some means for externalizing, for making publicly recognizable to others, the
expressions of their intentional states .. A being that can do that on purpose,
that is a being that does not just express its intentional states but performs
acts for the purpose of letting others know its intentional states, already has
a primitive form of a speech act. But it still has nothing as rich as our notions
of a statement, a request, or a promise, etc. A man who makes a statement
does more than let it be known that he believes something, a man who makes
a request does more than let it be known that he wants something, a man
who makes a promise does more than let it be known that he intends some-
thing. But again, what more? The primary extra-linguistic purpose of having
the institution of assertion is to give information, the main purpose of having
the institution of requesting (ordering, commanding, etc.) is to get people to
do things, and the main purpose of having the institution of promising is to
create stable expectations of people's behavior. These facts will, I think,
provide a clue to the relations between the speech acts and the corresponding
intentional states. As a preliminary formulation one might say that our
beings would be capable of making a primitive form of assertion when they
could perform actions which were expressions of belief for the purpose of
giving information, "requests" (in this primitive form) would be expressions
of desire for the purpose of getting other people to do things, "promises"
(again in primitive form) would be expressions of intention for the purpose of
creating expectations in others.
The next step would be to introduce conventional procedures for doing
each of these things. However, there is no way that these extra-linguistic
purposes can be realized by a conventional procedure. They all have to do
with the effects which our actions have on our audiences, and there is no way
that a conventional procedure can guarantee that such effects will be achieved.
The perlocutionary effects of our utterances cannot be included in the con-
ventions for the use of the device uttered, because an effect which is achieved
by convention cannot include the subsequent responses and behavior of our
audiences. What the conventional procedures can capture is so to speak the
illocutionary analogue of these various perlocutionary aims. Thus for
example, any conventional device for indicating that the utterance is to have
the force of a statement (for example, the indicative mood) will be one which
by convention commits the speaker to the existence of the state of affairs
specified in the propositional content. Its utterance therefore provides the
hearer with a reason for believing that proposition and expresses a belief by
the speaker in that proposition. Any conventional device for indicating that
the utterance is to have the force of a directive (request, order, command,
etc.) will be one which by convention counts as an attempt by the speaker to
get the hearer to do the act specified in the propositional content. Its utterance
therefore provides a reason for the hearer to do the act and expresses a desire
of the speaker that the hearer do the act. Any conventional device for indicat-
ing that the utterance is to have the force of a commissive (promise, vow,
pledge) counts as an undertaking by the speaker to do the act specified in the
propositional content. Its utterance therefore creates a reason for the speaker
to do the act, creates a reason for the hearer to expect him to do the act, and
expresses an intention by the speaker to do the act.
The steps then necessary to get from the possession of intentional states to
the performance of conventionally realized illocutionary acts are, first, the
deliberate expression of intentional states for the purpose of letting others
know that one has them, second, the performance of these acts for the achie-
vement of the extra-linguistic aims which illocutionary acts standardly serve
and, third, the introduction of conventional procedures which conven-
tionalize the illocutionary points that correspond to the various per-
locutionary aims.
III
Throughout this paper I have used the concept of representation as an
unanalyzed notion. But as we have already noticed statements about re-
presentations are intensional-with-an-s and the reason for that is that
representation is intentional-with-a-t. So, by describing intentional states
using the concept of representation I have described intentionality in inten-
tional terms. Is there any way out of this circle? I do not believe there is. I do
not believe there is a nonintentional explanation of intentionality. That is,
there is no analysis of intentionality into logically necessary and sufficient
conditions of the form "X is in intentional state S if and only if p, q, and r,"
where "p, q and r" makes no use of intentional notions with-a-t.
This does not mean that there is not a great deal more that can be said by
way of describing and explaining how intentionality works. The point is
196 JOHN R. SEARLE
rather that such descriptions and explanations will necessarily employ

intentional notions. Our characterizations, descriptions and explanations of
intentionality will themselves not be reductive, they will presuppose some
understanding of intentionality. We might call this pattern of explanation,
where intentionality is used in its own characterization, the circle of inten-
tionality or, simply, the intentional circle.
It is interesting to observe the forms which efforts to eliminate inten-
tionality take. The standard behavioristic analyses of mental states blandly
use the concept of intentional behavior as if it were somehow less mentalistic
than the other mental notions, but to say that a man is walking to the store
or eating a meal is to attribute to him mental states which are no less mental
than to say that he wants to get to the store or he believes that the stuff on
his plate is food. We have the illusion that behavior is not a mental notion
because we can observe bodily movements, but the bodily movements only
constitute human actions given the assumption of the appropriate intentions
and beliefs. Thus either the behavioristic analysis assumes intentionality in
the analysans, in which case it is another instance of the circle of intentiona-
lity; or it does not, in which case the analysis is inadequate because no
behavior, in the sense of human action, is mentioned in the analysis at all.
Behavioristic analyses of meaning suffer from this dilemma in an acute form.
There have, for example, been efforts to explain semantic notions in terms of
the notion of assenting to a sentence. But assenting and dissenting are full
blown illocutionary acts, every bit as rich as promising or stating. To use these
notions in the analysis of meaning, or of any other semantic notion, is to
presuppose an understanding of speech acts, and since the speech act con-
cepts contain the semantic notions that the theory was supposed to analyze,
the theory is hopelessly circular. Here the situation is worse than in the circle
of intentionality, because the notions of assent and dissent require more than
just intentionality, they contain the notion of meaning.
The problem is not, as is sometimes alleged, that sentences asserting that
someone assented or dissented are intensional-with-an-s, nor even that
assenting and dissenting are themselves intentional-with-a-t, but that assent-
ing and dissenting are both illocutionary acts and thus understanding them
presupposes an understanding of the crucial semantic notions, especially
meaning.
But to say there is no non-intentional explanation of intentionality is not to
say there is no non-linguistic explanation of language. If we allow ourselves
intentional notions in the analysans, then it seems to me, we can give an
analysis of the basic linguistic acts in terms which do not themselves employ
any semantic notions, but do indeed employ intentional notions. The key to
this analysis is that the intentional notions already contain the notion of their
own satisfaction, and we can graft our semantic notions onto the intentional
notions using this non-semantic notion of intentional satisfaction. I have not
given that analysis in this paper but have only sketched the direction it might
take. It seems to me not at all paradoxical that there should be non-linguistic
analyses of the linguistic but not non-intentional analyses of intentionality.
Speaking a language is, after all, a part of human behavior and of human
conscious life. It would be surprising if we could not describe it in terms
derived from human behavior and human conscious life. But in the way that
intentionality underlies the possibility of linguistic acts there is nothing that
conceptually underlies intentionality. Intentionality is precisely that feature
of mental states, human or otherwise, that enables those states to represent
other things.
The University of California at Berkeley
NOTES
1 Zettel, 56 and Remarks on Fraser, p. 5.

2 "Problems of Intentionality," in: Phenomenology and Philosophical Understanding,
E. Pivcevic (ed.), Cambridge University Press, 1975, p. 48.
3 It is a mistake to say that there are different senses of "belief," a "notional" and a
"relational" sense. Rather there are different kinds of reports of belief. One and the same
belief can be reported in several different ways, e.g., "John believes truly that fa," "John
believes ofa that fa," and "John believes that fa."
4 "Actions. Reasons, and Causes," reprinted in: The Philosophy of Action, A. White
(ed.), Oxford University Press, 1968.
S It is fairly easy to extend our criteria of intensionality to apply to intentional states. If
the satisfaction of the state requires the existence of the objects represented and if the
satisfaction value of the state is not altered by the substitution of other representations of
the same object, then the state is extensional.
6 For an explanation of the notion of direction of fit, see J.R. Searle, "A Taxonomy of
Illocutionary Acts," in: Minnesota Studies in the Philosophy of Science 7 (1975): Language,
Mind and Knowledge, K. Gunderson (ed.), pp. 344-369.
7 This looks as if it leads to an infinite regress but it does not. The understanding of the
conditions of satisfaction of our intentional states is not an additional intentional state.
HILARY PUTNAM
REFERENCE AND UNDERSTANDING
My thesis is that the theory of language understanding and the theory of

reference and truth have much less to do with one another than many philo-
sophers have assumed. I want to explain why this is so. I believe that if I am
right, this will shed light on a number of things in philosophy as for example,
why Wittgenstein had both a "picture" theory of meaning and a "use"
theory of meaning (at different times, of course!), and what was going on in
Nineteenth Century objections to a Correspondence Theory of Truth.
Understanding Language
It seems to me that the account according to which understanding a language
consists in being able to use it (or to translate it into a language one can use)
is the only account now in the field. Perhaps Michael Dummett will succeed
in developing an alternative account (I know he wants to); but at present I
know of no alternative. Secondly, I do not think "ability to use" a language
has to be thought of as coming from the learning of separate little playlets of
the kind Wittgenstein uses in the early pages of the Investigations to illustrate
the notion of a "language game." Some Wittgensteinians appear to think of
language in this way - as consisting of disconnected "uses" (e.g., the expres-
sion "that's a different language game" such persons sometimes use), but I do
not think Wittgenstein is guilty of this; and, in any case, it is not essential to
the doctrine.
While a true-ta-life model of the global use of a language is hardly to be
hoped for, an oversimplified model (for assertoriallanguage) is contained in
the work ofCarnap and Reichenbach: this is the model of the speaker/hearer
as possessing an inductive logic (e.g. a subjective probability metric -
although I do not myself think this is a very good way to view induction), a
deductive logic, a preference ordering (although I do not myself think this is
more than an idealized way of modelling human preference structures), and
a rule of action (e.g., "maximize the estimated utility" - although I myself
think this is a bad rule in many situations). Imagine a community of such
speakers/hearers who accept sentences they hear others utter (or assign them
a high probability) and who utter sentences themselves whose probability
199
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland
200 HILARY PUTNAM
exceeds a certain value. In even such a terribly oversimplified model, speech

will effect behavior in a rich variety of ways. And the better the inductive
logic, the better the deductive logic, the more realistic the utility function, the
more the behavior of these creatures will resemble "understanding a
language."
Such a model is not tied to an individualistic conception of knowledge or
language. Since speakers may acquire knowledge from each other (know-
ledge is a "contagious disease," in John McDowell's happy phrase) it is not
necessary that each speaker who has a word, say, "gold," in his vocabulary
be himself able to tell whether or not something is gold. Indeed, as long as
each speaker who has the word "gold" in his vocabulary possesses a standard
minimum amount of information about gold, he or she will be able to parti-
cipate in collective discussions about gold. Thus, the feature that I called
"the linguistic division of labor" (in "The Meaning of 'Meaning''')1 is
perfectly compatible with such a model. Also, another idea I put forward in
"The Meaning of 'Meaning' " - the importance of "stereotypes," conceived
of as minimum amounts of information associated with words - is readily
assimilated into such a model. Finally, such a model of a speaker-hearer is
essentially holistic - the conditions under which any particular sentence will
be uttered and the behavior that will result if any given sentence is uttered
do not depend upon any isolated thing that could be called the "sense" of the
sentence, but on the total program. Changing the inductive logic, or the
deductive logic, or the utility function will to some extent effect the utterance
conditions for and the behavior responses to every sentence of the language.
However, not every change in "use" (in the sense of rules, or in the sense of
utterance conditions for sentences/behavior responses to sentences) is a
change in meaning. Meaning, in my view, is a coarse grid laid over use. I think
there are different criteria for when we say there has been a change of mean-
ing in the case of different sorts of words; and for some words (e.g. logical
words) there is no sense to the change of meaning/change of theory dichotomy
at all. (That our criteria for "change of meaning" are as various and sloppy
as they are is explained by the different kinds of interests we have in connec-
tion with different topics and activities. In this sense language is a "motley" -
even if one can bring the motley under a uniform representation at some level
of abstraction.)
This is only a sketch of a "use" account of understanding, but I take it
that this sort of account is pretty familiar (except, perhaps, for the idea of
meaning as a coarse grid over use). What I want to shift to now are the topics
REFERENCE AND UNDERSTANDING 201
of realism, reference, and truth.
Realism and Correspondence

Nothing in this account of "use" says anything about a correspondence
between words and things, or sentences and states of affairs. But it does not
follow that such a correspondence does not exist. A number of tools have
this feature: that the instructions for use of the tool do not mention some-
thing that explains the successful use of the tool. For example, the instructions
for turning an electric light on and off - "just flip the switch" - do not
mention electricity. But the explanation of the success of switch-flipping as a
method for getting lights to go on and off certainly does mention electricity.
It is in this sense that reference and truth have less to do with understanding
language than philosophers have tended to assume, in my opinion.
On the model just sketched, one can use one's language, at least on an
"object language" level, without any sophisticated notion of truth. Of
course, one has to be able to assent and dissent; but that one has in the
Carnap/Reichenbach model. The instructions the mind follows, in this model,
do not presuppose notions of the order of "true": they are instructions for
assigning high weights to certain sentences when certain patterns are detected
by the sense organs, instructions for "printing out," instructions for carrying
out syntactic transformations, instructions for producing non-verbal
behavior (e.g. "lift arm when 'lifting arm has maximum estimated utility' is
computed"), etc. But the success of the "language using program" may well
depend on the existence of a suitable correspondence between the words of a
language and things, and between the sentences of the language and states of
affairs. The notions of truth and reference may be of great importance in
explaining the relation of language to the world without being as central to
meaning theory (in the sense of understanding-of-language theory) as they
are in, for example, theories that equate understanding with knowledge of
truth conditions.
It was this that I had in mind when I spoke of Wittgenstein's two theories
of meaning: the picture theory was wrong as a theory of understanding, for
reasons that Wittgenstein himself very well brought out; but not totally
wrong as a theory oflanguage functioning. It is essential to view our theories
as a kind of "map" of the world, realists contend, if we are to explain how
they help us to guide our conduct as they do. But the "use" theory is also
right as an account of how language is understood. And the insights are not
incompatible: a map, after all, is only a map by virtue of being employed in
202 HILARY PUTNAM
certain ways, but that insight does not contradict but only supplements the
fact that a map is only successful if it corresponds in an appropriate way to a
particular part of the earth, or whatever. Talk of use and talk of reference are
parts of the total story, just as talk of switch-flicking and talk of electricity
flowing through wires are parts of a total story.
Success
I have been employing the notion of "success." Let me try to unpack this
notion. What "succeeds" or "fails" is not, in general, linguistic behavior by
itself but total behavior. E.g., we say certain things, conduct certain reason-
ings with each other, manipulate materials in a certain way, and finally we
have a bridge that enables us to cross a river that we could not cross before.
And our reasoning and discussion is as much a part of the total organized
behavior-complex as is our lifting of steel girders with a crane. So what I
should really speak of is not the success or failure of our linguistic behavior,
but rather the contribution of our linguistic behavior to the success of our
total behavior.
Up to a point, every metaphysical position gives the same explanation of
this - of the contribution of linguistic behavior to the success of total
behavior - but only up to a point. The explanation is that certain kinds of
beliefs we have tend to be true (or whatever predicate the position in question
substitutes for "true," e.g., "warrantedly assertible"). Some philosophers have
been so incautious as to put forward the maxim that "most of a speaker's
beliefs are true" as a kind of a priori principle governing radical translation;
but this seems to me to go too far. (First of all, I do not know how to count
beliefs. So I do not know what it means to speak of most of a speaker's
beliefs. And second, most people's beliefs on some topics - e.g. philosophy-
are probably false.) But most people do have true beliefs about where they
live, what the neighborhood looks like, how to get from one place to another,
etc. And they have many true beliefs about how other people wiII react in
various circumstances, and many true beliefs about how to do and make
certain things. And every philosophical position yields roughly the following
story.
(l)People act (in general) in such a way that their goals wiII be obtained
(as well as possible in the given situation), or in such a way that their expecta-
tions wiII not be frustrated, or, in more concrete terms, so that they wiII get
food, lodging, etc., find friends and companions, etc., get from one place to
another successfully, avoid dangers, and so on, if their beliefs are true.
(2) Many beliefs (of the kinds I mentioned, and of other relevant kinds) are
true. (Although some are false - which is part of the explanation of people's
failures.) (3) So, as a consequence of (1) and (2), people have a tendency to
attain certain kinds of goals.
I think this much is non-controversial. Every account of truth tries to make
(1) and (2) correct, and thus to permit us to explain (3) via (1) and (2). For
example, a pragmatist or phenomenalist account interprets "true" so that a
true statement corresponds to conditional expectations which will be fulfilled
if the action which is the antecedent condition for the conditional expectation
is carried out. So (1) is taken care of - if one is acting on true beliefs, one's
expectations will not be frustrated. And all philosophical positions - idea-
lism, pragmatism, realism, etc. (except extreme scepticism) - hold that most
beliefs people have about where they live, etc., are true. So (2) is accepted.
So in all positions "truth" plays the same role in accounting for the contribu-
tion that linguistic behavior makes to the success of total behavior up to this
point - that one explains (3) via (1) and (2).
However, these are not all the desiderata that one wishes to impose on an
account of truth. For example, I have argued elsewhere2 that some familiar
positivist substitutes for the notion of truth (e.g., "is simple and leads to true
predictions") do not have the property that the conjunction of acceptable
theories (theories with the property in question) is acceptable. So acceptability
is not preserved by deductive 10gic.3 But scientists do regard logical con-
sequences of acceptable theories as acceptable. (In the sense that, if the con-
sequence is unacceptable, then one has to go back and revise one's decision
that all the premisses were acceptable.) So an account of this kind - an
account that says that what we seek is a kind of acceptability that lacks the
property of deductive closure - fails to justify norms of scientific practice.
It is precisely because there are further desiderata that an account of truth
has to satisfy, beyond giving us the minimal account of the contribution of
linguistic behavior to the success of total behavior (by which I mean (1),
(2)/(3) ), that one has grounds for rejecting some metaphysical positions on
the nature of truth and/or acceptability as inadequate.
Besides giving us the minimal account of the contribution of linguistic
behavior to the success of total behavior, and giving us the fact that truth
(or whatever notion of acceptability one may propose as a replacement for
truth if one objects to the notion of truth for some reason) is preserved by
(suitable)4 rules of deductive logic, there is one further desideratum in
particular that a satisfactory account of truth ought to fulfill: it ought to
204 HILARY PUTNAM
account for the reliability of our learning. Let me now say something about
this desideratum.
Realist Accounts of Epistemic Reliability

The minimal account of the contribution of linguistic behavior to the success
of total behavior employs the idea that most (or at least many) of our beliefs
on certain topics are true. But these beliefs are not a definite stock fixed once
and for all. They are constantly changing as a result of learning (using the
term in a wide sense to include perception). So what premiss (2) of the mini-
mal account really asserts is that certain sorts of learning are reliable - in
the sense of leading to a large number of true beliefs. How is this reliability
to be explained?
Of course, what sort of "explanation" one thinks is to be given will depend
upon features of one's philosophical position. Contemporary realists often
differ from realists of earlier centuries in rejecting the notion of a priori truth,
or at least in not putting very much "weight" on the notion. They also -
here I again take the risk of generalizing about philosophers who have many
disagreements among themselves - tend to be sceptical of the idea of a fixed
unchanging "scientific method." And if one does not regard the methods of
gaining knowledge as given a priori, and if one regards these methods as
enmeshed and evolving in history, then one will be unlikely to view any sharp
dichotomy between the method and the content of knowledge as defensible. 5
What our "methods" are in any domain will depend on what our beliefs are
concerning the subject matter of that domain and on beliefs in other domains.
From such a point of view, the problem of explaining the reliability of learn-
ing is not to show a priori that our learning methods are reliable, but to show
that one can understand that reliability as a fact of nature - that one can
explain it, or make progress towards explaining it, as one explains or makes
progress towards explaining other facts of nature. The possibility of doing
this may be viewed as a kind of "consistency check" (in an informal sense of
"consistency") on the metaphysical view as a whole.
To see how this program can be carried out, let us take the simple case of
visual perception: a normal speaker looks at an object, say a rug, and says
"the rug is green" or "the rug is not green." Why is he reliable?
Part of the story is the usual causal account of vision. What this yields, to
the extent that it has been carried out (and no one would deny that enormous
progress has been made in the understanding of color vision), is an explana-
tion of the following facts:
(4) If the rug is green, then the speaker probably accepts "the rug
is green."
(5) If the rug is not green, then the speaker probably accepts "the
rug is not green."
But what we want is:
(6) The speaker probably accepts whichever statement is true.
Now the realist also accepts some standard truth definition6 for the
language. Such a truth definition sets up a correspondence between things
and words (e.g., "the rug" refers to some contextually definite rug; "is green"
refers to green things: sentences of the form "TheNV's" are true if and only if
the object referred to by "The N" is referred to by V.) Such a truth definition
has the features that
(a) It satisfies Tarski's Criterion (T), e.g.,
"The rug is green" is true if and only if the rug is green,
And
(b) "True" commutes with truth functions and with such operators
as "probably."
By virtue of these features, (4) and (5) imply:
(4') If the rug is green, the speaker probably accepts the statement
which is true from the pair "The rug is green," "The rug is
not green."
(5') If the rug is not green, the speaker probably accepts the
statement (from the same pair) which is true.
And since the rug (in the situation envisaged) is either green or not green -
(6') The speaker probably accepts the statement which is true (from
the pair in question in the situation envisaged).
(6') Says that a certain form of learning (visual perception in the case of a
uniformly colored rug) is reliable. And we have sketched how a realist can
account for this reliability from within our total conceptual system (causal
theory of perception and language use plus semantic theory of truth), as he
reconstructs that conceptual system. Thus the "consistency check" is satis-
factory. (Of course all this is programmatic, but so is our explanation of most
206 HILAR Y PUTNAM
empirical facts.) In the case of more complicated kinds of learning, the causal
account of reliability is, of course, much more programmatic. But, unless we
want to jettison or ignore our entire body of natural science and scientific
speculation, no alternative account is even in the field today.
One interesting case is that of inference to a theory: here, as many people
have remarked, it seems necessary to assume that we have "weak a priori
knowledge" in the sense of having evolved with a "simplicity ordering" of
theories, or a "prior probability metric," which is not "too bad" in the sense
of not assigning a hopelessly low probability to the theories which are true
in the actual world. How this happened we cannot explain: but that it should
have happened is at least not inconsistent with our present accounts of the
development of the species, and much more work is certainly going to be
done both on the history of that development and on the structure of our
inferential capacities. The messiness of the situation is not a source of dismay
to an anti-a prioristic and scientifically minded realist; it is just what he would
expect at the present stage of research into a complex problem that cuts
across the fields of biology, psychology, and inductive logic!
The role played in this by the idea of a correspondence between language
items and extra-linguistic reality is not hard to see. The causal theory of
reliability tells us that, for example, when a certain state of affairs obtains
(the rug being green) the speaker utters a certain sentence ("the rug is green").
The semantic theory of truth tells us that the sentence is true just in case that
state of affairs obtains - the correspondence involved in the causal story is
exactly the correspondence set up by the truth definition. So we can be viewed
as systems that reliably produce true sentences when a certain variety of
states of affairs obtain. Assumption (2) of the minimal account of the con-
tribution of linguistic behavior to the success of total behavior is (sketchily,
programmatically) explained.
Another desideratum on any account of truth is that the correctness of
assumption (1) be explained (in the same sense). For most idealists this is no
problem - a "true" statement is one that corresponds to the right sort of
expectations about the consequences of behavior. But the idealist has his
problems too. If he is not a holist - if statements correspond one-by-one to
sets of expectations - then he has trouble explaining the actual character of
the development of theories which is decidedly "holistic." If, on the other
hand, he makes truth (or acceptability) a predicate of large systems of
statements - perhaps whole bodies of knowledge - and not single sentences,
then he has trouble accounting for the reliability of learning (indeed, I would
argue he cannot explain it - but that goes beyond the present paper).
For the realist, however, (1) is not trivial. "Snow is white" does not cor-
respond to a fixed set of expectations about the future, on his account, but
to snow being white. So how does it come about that when we believe certain
statements we tend to act in a way which will not frustrate our expectations
if those statements are true (i.e. if the corresponding states of affairs obtain)?
The realist's answer is that the connection between the state of affairs in
question obtaining and our goals being satisfied is itself something about
which we have many true beliefs. How does this come about? Well, these
connections are themselves learned - thus the (programmatic) causal
account of the reliability of learning also explains the existence of true beliefs
about connections between actions and goals in various situations.
It is this causal account of the correctness of both assumptions of the
minimal account of the contribution of linguistic behavior to the success of
total behavior that I had in mind when I said at the beginning that the
correspondence between words and things, between statements and states of
affairs, is what explains the success oflanguage using even if it is not referred
to in the "program" for language using.
So far I have left it seeming miraculous that the relation between states of
affairs and sentences described by the causal theory of perception, language
acquisition, etc., is also the one specified by a truth definition for the language.
But this too can be accounted for. Let C be an arbitrary correspondence
between sentences and states of affairs, and call a sentence TRUE(C) if the
state of affairs to which it bears the relation C obtains. Then there is in
general no reason why the property of being TRUE(C) should be preserved
by deductive logic. But it is part of our explanation of speakers' reliability
that one of the ways in which they acquire new beliefs is the use of deductive
logic and that deductive logic preserves truth. 7 If this explanation is to be a
part of the total explanation of reliability, then the correspondence C has to
be of a special kind, and the most natural choice is a correspondence which
is based on a satisfaction relation.
Once we pick a correspondence of this kind to define truth-in-L, where L
is the language in question, it will, of course, be possible to define the cor-
respondence in question in any meta-language of sufficient strength that
contains L itself "disquotationally" - that is, so that the Criterion (T) is
satisfied. Note that (T) has no significance in radical translation, however,
and that the truth definition for a natural language is underdetermined by
the sort of causal considerations discussed here, and that there may well be
208 HILARY PUTNAM
some interest-relativity to notions like "truth" and "reference" in the context

of radical translation. 8 The fact that explanations may be interest relative
does not mean that they cannot be correct, however - which is all that is
needed for the argument I am advancing for realism.
Is the Causal Explanation Philosophically Neutral?

I can imagine a philosopher responding to what I have just said by saying,
"Of course what you have sketched is the scientific explanation of the re-
liability of learning" - the thrust being that such a scientific explanation
must be compatible with any metaphysical position. Now, of course all meta-
physical positions claim to be compatible with all of science, but claiming it
does not make it so.
In order to see what the difficulties are that face the non-realist philosopher
who wishes to incorporate the causal explanation of the success of language
using into his world picture, I shall discuss one position in particular. This is
the position I described in "What is Realism"9 (a position suggested by some
ideas of Michael Dummett). In "What is Realism" I showed that by giving a
quasi-intuitionist meaning to the logical connectives we could preserve all of
extensional scientific discourse, and also preserve Tarskian semantics
(including Criterion (T)), while avoiding any commitment to the existence
of theoretical entities such as genes or molecules. So, at first blush it seems
that one could take over the entire causal explanation of the contribution of
linguistic behavior to the success of total behavior, without really being
committed to the existence of most of the entities science talks about, or the
existence of a correspondence between words and such entities. Causal
realism within science would appear to be compatible with an idealist re-
interpretation of total science.
But, I went on to argue, this is not so. For science as ordinarily understood
makes modal statements - statements about what is and is not possible -
e.g., "one cannot build a perpetual motion machine." And a reinterpretation
of science that is not to be a mutilation must also "translate" these statements
(preserving truth value, and preserving inductive and deductive relations).
But this, I claim, one cannot do.
To see why not, I first point out that "truth," in the quasi-intuitionist
picture, becomes a version of warranted assertibility. (It is, of course, quite
typical of idealist reconstructions that the predicates offered as notions of
acceptability to replace the classical notion of truth are versions of warranted
assertibility.)
Now, in the case of seeing what color a rug is, it is a part of the causal
explanation that there is room for error - it is physically possible that one
seem to see a green rug, etc., and the rug not be green. Thus:
(7) "The rug is green" might be warrantedly assertible even
though the rug is not green.
is a modal statement implied by our theory. But this shows truth cannot be
warranted assertibility!
My argument has been described by Dummett as a sort of "Naturalistic
Fallacy" argument (one might call it the "Idealistic Fallacy Argument"),
because of its obvious similarity to Moore's celebrated argument for the
indefinability of "good." What I am claiming is that for any predicate P the
idealist may want to substitute for "true" one can find a statement S such
that
(8) S might have property P and still not be true.
follows from our causal theory of learning. And this is so simply because the
"slop" between being warrantedly assertible, no matter how construed, and
being true (assuming only Criterion (T)) is itself "built in" to our causal
theory.
It might seem, however, that Dummett has an easy way out. Suppose
Might-statements are confirmed by the realist by describing a "model" in
which the statement in question holds (satisfying the specified constraints)
and showing that the model satisfies the laws of our theory. Very well, let the
non-realist accept this as the verification procedure (warranted assertibility
procedure) for Might-statements, and adjoin them to his language with this
same verification procedure.
Then with "true" (warrantedly assertible) extended to statements of the
form "p might be true even if q" in this way, it will also hold for the non-
realist that (7) is true, and even that
(9) "The rug is green" might be warrantedly assertible even though
"The rug is green" is not true.
What has gone wrong is not hard to show. "Truth" was reinterpreted as
warranted assertibility (in a certain sense). And asserting that p was reinter-
preted as asserting that p is warrantedly assertible. "The rug is green" is now
entailed by "The rug is green is warrantedly assertible." So if we introduce
Might-statements into the language as just suggested, we will be giving up
210 HILARY PUTNAM
the incompatibility betweenp entails q andp might be even ifnot q. We should

not have preserved the deductive relations between statements in the transla-
tion process.
Actually, the problem is even worse: if the new entailments are allowed
to be used in the scientific theory itself, it becomes inconsistent. Also, "true"
fails to commute with "might": "p might be even if q" fails to be equivalent
to "p might be true even if q is true."
I have discussed Dummett's position because it represents the most viable
non-realist position I know of. The difficulties it faces reveal how hard it is to
provide an idealist reinterpretation of our causal theories that preserves
deductive and inductive relations between our statements, including the
modal statements we accept, and at the same time avoids my Idealistic
Fallacy Argument.
The Nineteenth Century Revisited

The great nineteenth century argument against the Correspondence Theory
of Truth was that one cannot think of truth as correspondence to facts (or
"reality") because, so it was contended, thinking of truth in this way would
require one to be able to compare concepts directly with unconceptualized
reality - and philosophers were as fond then as they are now of pointing out
the absurdity of such a comparison.
Needless to say, many different things were going on in this argument -
too many to discuss here. (E.g., part of the background to the argument was
the general view that we cannot really perceive material objects - a view that
has come under heavy fire in the second half of the twentieth century.) But
one thing which may well have been in people's minds is the following argu-
ment: assume, as is plausible, that to understand a statement is to be ex-
plained as knowing its truth conditions. If truth is correspondence to reality,
it would seem as if knowledge of what the correspondence is is presupposed by
knowledge that such and such a statement stands in the relation in question
to anything, or does not stand in the relation in question to anything. And if
understanding the statement is equated with knowing what it is for it to be the
case that it stands or does not stand in the relation in question to appropriate
entities, then knowledge of what the correspondence is is presupposed in the
understanding of every statement. But in what could this knowledge - which
does not consist in the acceptance of any statements, because it is prior to the
understanding of all statements - consist?
The semantic theory underlying this form of the objection - the theory
that truth is prior to meaning (i.e., that truth must be the central concept in a
theory of understanding) 10 is familiar enough in the present day. What I am
suggesting in this paper is that we reject this view. If we view language un-
derstanding as the "internalization" of a program for language use - a
program consisting of "language entry rules" (procedures for subjecting
some sentences to stimulus control),ll procedures for deductive and inductive
inference, and "language exit rules" (procedures for, e.g., taking something
in one's hand when one's brain computes "it would be optimal to take this
in my hand") - then implicit knowledge of truth conditions is not pre-
supposed in any way by the understanding of the language. To put the point
more briefly: one does not need to know that there is a correspondence be-
tween words and extra-linguistic entities to learn one's language. But there is
such a correspondence nonetheless, and it explains the success of what one
is doing. After one has learned one's language one can talk about any thing-
including the correspondence in question. Wittgenstein's view in the
Tractatus that the correspondence in question cannot be described is correct,
but only in a limited sense: even if one follows Kripke rather than Tarski, so
that the relation of reference can be spoken of in the object language itself
rather than only in a meta-language, a wider relation of reference can still be
defined by going up a level of language.
But still, even this wider relation of reference can be spoken about and not
merely "shown." And for the purposes of the kind of causal explanation of
the reliability oflearning, etc., discussed above, the relation of reference that
we can speak of in the object language itself is perfectly sufficient. As Carnap
long ago emphasized, there is nothing "indescribable" in the relation of
language to the world.
A Realist Objection
I can imagine that some realists may well be uncomfortable with the
account of understanding sketched here, even as an idealized and terribly
oversimplified model. Is not this model a Verificationist one? And is not
Verificationism at bottom a form of Idealism?
To take the first charge first: I am saying that there was an important
insight hidden in Verificationism. The insight is that in any plausible model
of a speaker /hearer the assignment of a "probability" or some not-necessarily-
quantitative analogue of "probability" to sentences (though not necessarily
to every sentence, on observational evidence - language is also used to
discuss what would happen under circumstances that we might not be able to
212 HILAR Y PUTNAM
confirm if they existed, e.g., phenomena in the interior of black holes) is going
to playa central role. In Experience and Prediction, Reichenbach illustrates
the power of what he calls the "probability theory of Meaning" in the follow-
ing way: a traditional phenomenalist has to explain talk about unobserved
objects - say, unobserved trees - by "reducing" it to talk about observed
objects (and ultimately to sense-impressions) in the fashion of Mach or
Avenarius. But Reichenbach does not have to do this. A sentence about
unobserved trees (e.g., "There is a tree behind me") is understood if we know
how to assign it a "weight" - i.e., how to conclude to it inductively (Reichen-
bach leaves out the need for a utility function and a decison rule - these
come in in Carnap's later work on probability), how to deduce/induce con-
sequences from it, etc. For example, suppose I see a tree shadow (as if of a
tree behind my left shoulder). Previously 1 have confirmed: "Whenever there
is a tree shadow, there is a tree in such and such a spatial relation to the tree
shadow." From this inductively confirmed generalization and the observation
of the tree shadow in front of me 1 deduce (or induce, if the generalization is
itself only statistical) "There is a tree in such and such a spatial relation to
this tree shadow" and, finally, "There is a tree behind me." So I have now
accepted (or assigned a weight to) a sentence about an unobserved tree with-
out any Machian (or C.I. Lewisian, or Russellian, etc.) "reduction" of the
unobserved tree to a series of actual or possible observed trees or sense
impressions. (A similar account is given by Carnap in Testability and
Meaning.)
What seems right to me about this is that if we had no inductive logic at
all- if we only had pattern recognition and deductive logic - there would
be no basis for ascribing to us any concept of an "unobserved object." Our
linguistic behavior would fit the account" 'tree' means 'observed tree' " -
and, more generally, "'object' means 'observed object.'" In this sense,
our inductive logic is part of our concept of an unobserved object, and hence
of an object at all (which does not mean that every change in our inductive
logic is a change in our concept of an object - notions like "concept" and
"meaning" are a coarse grid over use). Similarly, our deductive logic is part
of our understanding of what a set or property is, as well as of our under-
standing of the quantifiers and connectives.
Secondly, it is not clear that this form of Verificationism (the "probability
theory of meaning") is incompatible with realism. (I think the "probability
theory of meaning" is wrong on quite different grounds, but that is not the
question here. The question I want to look at for a moment is whether the
"probability theory of meaning" is compatible with realism or not.)

It is clear why the "conclusive verifiability in principle" theory of meaning
is incompatible with realism. The realist believes that the truth or falsity of
our sentences depends (usually) on something external- i.e., on facts which
are not (logically equivalent to) experiential facts, facts which are not about
sensations (or about language-rules, etc.). But realism, if true, must be
expressible in some language or other. ("If you can't say it you can't whistle
it either.") Suppose LR is a language adequate for the expression of the
thesis of realism: e.g., in LR we might want to give the sort of account I have
been sketching, of how there is a correspondence between sentences and
extra-linguistic (and non-phenomenal) facts the existence of which explains
the contribution language using makes to the success of over-all behavior. If
Conclusive Verificationism is correct, there must be phenomenal truth con-
ditions (or conditions in "observable thing language," if the dispute is about
realism with respect to "theoretical entities" rather than with respect to
non-phenomenal entities; but I shall ignore this refinement) for every sentence
in every intelligible language. So, in particular, the sentences of LR which
purport to describe the allegedly non-phenomenal states of affairs cor-
responding to the sentences of, say, natural language, actually have pheno-
menal truth conditions - i.e., these states of affairs are not really non-
phenomenal after all. So realism is either false or (worse) inexpressible.
But no such argument can be offered against the compatibility of realism
with the "probability theory of meaning." Why should it be impossible from
a realist point of view that (1) every meaningful sentence have some weight
or other in some observable situation, and (2) every difference in meaning be
reflected in some difference in weight in some observable situation? (These
are, in my formulation, the two principles of Reichenbach's "probability
theory of meaning.") At least it should be an open question for the realist
qua realist whether this is so or not, and not something that realism rules
out.
However, it is very important in this connection that the "probability
theory of meaning" does not require that every meaningful sentence have a
high weight (be "verified") in some situation or other. For this would rule
out such sentences as "There is a gold mountain but this statement will never
have a high weight on the evidence." And to insist on a semantical connection
between existence and knowledge is unacceptable to a realist. Similarly, to
require that every meaningful sentence be "falsifiable in principle" (have a
low "weight" in some observable situation) would be unacceptable to a
214 HILARY PUTNAM
realist. But neither Carnap nor Reichenbach impose these requirements.

However, it is unacceptable to a realist to simply take the "probability
theory of meaning" as an explication of meaningfulness, as Reichenbach
does. Let SI and S2 be sentences describing two states of affairs which are
possible according to physical theory and not equivalent according to the
laws of the physical theory. Before a realist could accept the "probability
theory of meaning," one thing he would have to be shown is that every such
pair S1> S2 has the property that in some observable situation or other SI and
S2 differ in weight. (If we allow differences in weight on non-observational
evidence to count as differences in meaning, then this problem may disappear,
but then the theory is not Verificationist any more.) The realist would not be
willing to simply stipulate that SI and S2 count as the same situation just
because they get the same weight on all observational evidence.1 2
Moreover, the "probability theory of meaning" seems wrong to me as an
account of meaning anyhow. It cannot as it stands incorporate either the
linguistic division oflabor (the fact that confirmation procedures for being
gold, or being aluminum, or being an elm tree, or being David are not the
property of every speaker - speakers defer to experts for the fixing of re-
ference in a huge number of cases) or the contribution of the environment
(the fact that the extension of a term sometimes fixes its meaning and not vice
versa).1 3 In my view, the criteria used by experts to tell whether or not some-
thing is gold are not "part of the meaning" of gold (e.g., the word does not
change its meaning in the language if the experts shift to a different set of
tests for the same metal), yet they are part of a mechanism for fixing the
extension of gold. But if they are not part of the meaning of gold, then there
is nothing in the meaning to enable us individually to assign a "weight" to
"such and such is gold" - contrary to the "probability theory of meaning."
But I did not argue that we should accept the "probability theory of
meaning." I only argued that the Reichenbach-Carnap model is partly
correct as an account of use and understanding, not as an account of
meaning.
Use and Meaning

If the account I have offered here is correct, it is clear why we want a notion
of reference; but what is the need for a notion of meaning?
We need a notion of reference because the referent of a term is important
in many diffeIent situations. Buying and selling wine is quite a different
"language game" from discussing wine at the dinner table; but the extension
of the term "wine" may be the same in both contexts, and the truth condi-
tions for "One can buy Israeli wine in this country" may be the same in both
contexts. And it is because a sentence can have the same truth conditions in
different contexts that the same inductive and deductive logic can be used in
different contexts. But again, what is the need for a notion of meaning?
In my view,14 a language made up and used by a being who belonged to no
community would have no need for such a concept as the meaning of a term.
To state the reference of each (simple and defined) term and to describe what
the language user believes in connection with each term is to tell the whole
story. If the language changes so that the reference of any term changes, we
can say that this has happened; or if the speaker revises some of his beliefs we
can say that this has happened; but why say that some (but not all) of the
latter changes count as "changes of meaning"?
But as soon as the language becomes a communal instrument things
change. How could discussion take place if we could assume nothing about
what all speakers believe? Could I safely use the word "tiger" in talking to
you if, for all I knew, you believed that tigers are a kind of clam? Where
would conversation start?
In "The Meaning of 'Meaning'" I argued that meaning is a several-
component affair. I put forward the view that one component of meaning is
the reference (extension). (In my view, reference is fixed by meaning only in
the sense of being a component of meaning, but not in the sense that meaning
is a mechanism for fixing reference. The actual mechanisms for fixing
reference - e.g., the criteria used by experts to tell whether or not something
is gold - are not always part of meaning.) Another major component, in my
view, is stereotype - and stereotypes are nothing but standardized sets of
beliefs or idealized beliefs associated with terms (e.g., the belief that tigers
are typically striped orange and black is part of our stereotype of a tiger).
The need for stereotypes is not primarily to fix extensions: that can be and
often is done by experts using criteria that are not "part of the meaning" in
any sense. The stereotype associated with "gold," for example, is all but
worthless for fixing the extension of the word (or its extension in possible
worlds, for that matter). Language is not only used to verify and falsify and
classify; it is also used to discuss. The existence of standardized stereotypes,
and hence of meaning, is a necessity for discussion, not for classification.
, Incidentally, Wittgenstein was right in saying that language is a motley in
the sense that we have many different standards for different types of dis-
cussion - and this reflects itself in the fact that the amount of information
216 HILARY PUTNAM
contained in "meanings," the nature of the information, and its logical

"tightness" (approximation to analyticity), varies enormously from kind of
word to kind of word, and even between words of one kind. (Compare
"bachelor" and "gold," and also "gold" and "molybdenum." "Gold" and
"molybdenum" are both names of metals - but there is a rich stereotype of
gold and virtually no stereotype of molybdenum beyond the feature "metal".
Again, "tiger" and "weasel" are both names of animals - but the stereotype
(and, in my view, the meaning) of "tiger" includes the feature of having
orange and black stripes whereas I have no idea what color weasels are, and
a fortiori no idea by virtue of "knowing the meaning" of "weasel. ")
Morals
If there are any morals to be drawn from my discussion, they are perhaps
that (1) The notion that one learns one's native language by learning what
the truth conditions are for its various sentences has no presently intelligible
sense, at least for a realist; (2) that it does not follow that the realist's notions
of truth and reference are not important for the discussion of language - but
their importance is for the explanation of the contribution linguistic behavior
makes to the success of total behavior, not to a theory of understanding; and
(3) since (as I have argued elsewhere) the notion of "meaning" has neither
the nature nor the function philosophers believe it to have, the injection of
the word "meaning" into discussions of understanding and use is more likely
to confuse than to clarify issues.
Harvard University
NOTES
1 Cf. [5], chap. 12.

2 In [5], chap. 11.
3 However, a certain kind of "warranted assertibility" is preserved by a non-classical
logic. The possibility of replacing the realist notion of truth with such a notion of warranted
assertibility by interpreting the logic of natural language as non-classical was suggested by
Michael Dummett (this is more or less hinted at in his famous paper on Truth [2]) and is
discussed by me in "What is Realism?" [6].
4 I believe that it is possible to decide that classical logic is incorrect for realist rather than
intuitionist reasons (cf. Note 7, below). But in this case the theorist must show that the
correct logic (according to the new theory) has classical logic as a "special case" in many
situations, so that the appropriateness of using classical logic where it is successful can be
accounted for.
S An interesting defense of this version of realism is Richard Boyd's "A Causal Theory
of Evidence" [1].
6 I use the term "truth defini tion" in the sense of standard Tarskian semantics. An
interesting variant has recently been proposed by Saul Kripke in [3]. This variant permits
reference and truth to be talked of in the object language itself at the cost of modifying
Tarski's Criterion (T).
7 If my position on quantum mechanics is correct (in [4], chap. 10) and the actual logic of
the world is not classical logic but the modular logic described by Birkhoff and von
Neumann, or something similar, then we would have to say: "classical deductive logic
preserves truth in situations in which quantum mechanics can be ignored," instead of what
is in the text.
8 I discuss this in my John Locke Lectures [7].
9 See Bibliography [6]
10 Here I employ the terminology of Michael Dummett's (unpublished) William James
Lectures. Although our positions clash, I have been enormously stimulated by Dummett's
important work, and this paper is largely a response.
11 Again, the idea of "stimulus control" (Quine's "stimulus meaning") is only an over-
simplification or idealization. Acceptance of an observation sentence by the actual human
brain certainly depends upon attention and global theory in complex ways, not just built
in or learned routines of pattern recognition.
12 The presence of a kind of Godelian incompleteness in all formalized inductive logic is
also a great difficulty for the "probability theory of meaning." Which of the possible
formalized inductive logics are we to use in computing "weight"? (The Godelian incom-
pleteness ofinductive logics was first pointed out in my [4], chap. 17, and is also discussed
in chap. 18 of the same book.)
13 Cf. "The Meaning of 'Meaning.' "
14 For a detailed statement, see "The Meaning of 'Meaning.' "
BIBLIOGRAPHY
[1] Boyd, Richard, "A Causal Theory of Evidence," Nous 7 (1973): 1-12.
[2] Dummett, Michael, "Truth," Proceedings of the Aristotelian Society 59 (1958-59):
141-162.
[3] Kripke, Saul, "Outline of a Theory of Truth," Journal of Philosophy 72 (1975):
690-716.
[4] Putnam, Hilary, Philosophical Papers, Vol. 1: Mathematics, Matter, and Method,
Cambridge University Press, 1975.
[5] Putnam, Hilary, Philosophical Papers, Vol. 2: Mind, Language, and Reality, Cam-
bridge University Press, 1975.
[6] Putnam, Hilary, "What is Realism?" Proceedings of the Aristotelian Society (1976):
177-194.
[7] Putnam, Hilary, John Locke Lectures 1976, in: Meaning and the Moral Sciences,
London, Routledge and Kegan Paul, 1978.
MICHAEL DUMMETT
COMMENTS
Realism rests upon - or, better, consists in - an adherence to a truth-

conditional semantics for our language. Where 'semantics' is taken to mean
'theory of meaning,' and a theory of meaning is equated with a theory of
understanding - an account of a speaker's mastery of a language in terms of
what he has to know to know the language - there are acute difficulties in
such a semantics, which will involve attributing to a speaker, as that in which
his understanding of each sentence consists, a knowledge of the condition
that has to hold for it to be true. First, there is the problem of explaining how
we derive from our grasp of the meanings of our sentences the actual use we
make of them: in particular, if the meaning is not, in general, given in terms
of how we recognise a sentence as true, but of what it is for it to be true,
independently of our capacity to recognise it as such, an explanation is called
for, and is difficult to provide, of how our grasp of that meaning leads us to
treat this or that as conclusive evidence, or as evidence falling short of being
conclusive, for its truth. (The classical 'problem of induction' is one aspect of
this difficulty.) Secondly, there is a difficulty about what it even means to
attribute to a speaker a knowledge of the condition for the truth of a sentence.
At one end of the scale, where a grasp of reference cannot be thought of as
mediated by anything, this appears to involve ascribing to the speaker an
immediate association between word and object, the object becoming part of
his private world; at the other end, truth, taken as subject to the principle of
bivalence, appears to transcend our capacity for recognising it as attaching to
the sentence, and a grasp of the condition for the truth of such a sentence
therefore becomes something not exhaustively explicable in terms of actual
linguistic behaviour.
Hilary Putnam proposes a new defence of realism, by allotting to the truth-
conditional semantics a different role. No longer will it form any part of an
account of that in which understanding consists: such an account is, rather,
to be given in terms of actual use, what we actually learn to do when we learn
to employ language, namely to treat sentences as verified or confirmed to
varying degrees in appropriate (recognisable) circumstances, to engage in
deductive and inductive reasoning, to modify our behaviour in various ways
218
COMMENTS 219
according to the sentences we have accepted as true, etc. The truth-conditional

semantics is to be invoked only as part of an explanation of the success of
language, so described, that is, of why engaging in such a complex practice
should be of value in assisting us to achieve our individual and collective
goals. By invoking it at this level and in this role, Putnam hopes, we may
escape the objections that can be lodged against it as a theory of under-
standing.
I should like to comment, first, that this appears to me to be both the most
powerful and the most original defence of realism that I know. I recently had
occasion to write down some possible lines of argument in favour of realism,
and I have to confess that I entirely overlooked the argument that Hilary has
just expounded; although I had read his recent Aristotelian Society paper,
and had heard some of his John Locke lectures, I had not properly grasped
the whole tenor of the argument, nor appreciated its force. I think it quite
probable that, if realism can be defended (a question about which I have no
firm opinion), it will be along some such lines as these. For the present,
however, I have to say that I am not yet convinced that the case has been
made out.
First, Hilary is indeed quite right, in my view, in saying that,for the purpose
of explaining the success of language, a truth-conditional semantics has an
advantage over a semantics of any other kind: namely that, in so far as the
deductive logic which we are trained to employ - the employment of
which is part of the practice of speaking the language - is classical logic,
truth-conditional semantics yields a justification of that logic, while most
other semantic theories will justify only weaker logical systems. (Certainly,
no logic of an ordinary kind, whose rules of inference preserve truth, whether
taken as subject to bivalence or not, can be expected to preserve any given
degree of probability. When bivalence fails, the classical laws will not all
preserve truth.) I do not, however, regard this as showing that, for this ex-
planatory purpose, only the truth-conditional semantics will serve: after all,
adoption of, say, an intuitionistic position concerning mathematics does not
leave us powerless to explain the considerable success that classical mathe-
matics has achieved. We can still be highly successful even if our reasoning
goes wrong at certain points; especially so ifit never leads from true premisses
to afalse conclusion, rather than to one that merely fails to be true.
Hilary has, however, a further argument for the necessity of a truth-
conditional semantics for this explanatory purpose, namely his Idealistic
Fallacy Argument. Truth cannot be explained in terms of any other general
220 MICHAEL DUMMETT
property, because we can always suppose that a sentence might have that
property and yet not be true. Let me first defend Hilary's use of this argument
against an objection that might be brought. Someone might say that, to
obtain a non-truth-conditional semantics (where this is taken as meaning one
in which the principle of bivalence fails), all we need to do is to give a Tarski-
style truth-theory in a non-classical metalanguage: so truth will not be
explained in terms of anything else (save satisfaction), any more than in the
two-valued case. What is wrong with this, in my view, is that it assumes that
the mere imposition of some set of logical laws is enough to confer intelligible
meanings on the logical constants. On the contrary, what is required is a
semantic theory which displays the manner in which a complex sentence is
qetermined as true or otherwise in accordance with its composition: and the
only semantic theory which will do this without appealing to any auxiliary
notion is that which takes the semantic value of a sentence to consist simply
in its being true or not being true, namely the two-valued semantics. Any
other semantic theory will invoke some other notion, such as that of veri fica-
tion or some relativised notion of truth (truth in a possible world, etc.), and
will define the absolute notion of truth in terms of that auxiliary notion (e.g.
as the existence ofa verification, or as truth in the actual world). In particular,
a semantic theory developed in response to the objections to a truth-
conditional semantics as a theory of understanding will employ as an
auxiliary notion one more closely connected than truth is with our actual
use of sentences - just such a notion as verification, falsification or practical
consequences. Only in the truth-conditional semantics is truth, so to speak,
an irreducible notion: it is just from this that the objections to its use as a
theory of understanding take their origin.
Hilary's Idealistic Fallacy Argument is evidently strong against any identi-
fication of truth with a property that may be lost: in saying that a sentence
might have that property and still not be true, we mean precisely that the
property is impermanent. But I cannot see that it has any force against an
account of truth - a thesis about what, in general, makes a sentence true if
it is true at all- that does not have this feature. Against one who says that a
mathematical statement, if true at all, is true only in virtue of the existence of
a proof of it, the assertion that such a statement might lack a proof and still
be true (the necessary inversion, for this case, of the Idealistic Fallacy Thesis)
appears to me to be powerless. Of course, if you are a platonist, you will
agree: but the constructivist is not forced to agree; the thesis is precisely what
he denies.
COMMENTS 221
One reason I have given for not accepting Hilary's defence of realism is
that it is not clear that a truth-conditional semantics is required to explain the
success of language. Supposing it to be possible to construct a workable non-
realist semantics for our language - and, of course, there are grave difficulties
in this project, too - this might serve equally well to explain our success,
since any workable semantics must yield the result that a great deal of what
we take to be true is in fact true.
Perhaps I may here interpolate some comments about the metaphysical
position resulting from a non-realist semantics. First, there is no question of
thinking that we create the world, or that the world is our dream, i.e. that
there is no objective reality external to us. It is not up to us to decide, but
only to find out, how things are. Secondly, there is no question, either, of
repudiating the notion of reference. Within the reality we encounter, we may
discern and pick out by names, descriptions, demonstrative phrases, etc.,
objects (in the most general sense of that word). Among such objects will be
buildings, stars, revolutions, and, for that matter, numbers (photons intro-
duce special problems because of the instability of scientific theory); for the
rejection of realism does not entail reductionism in any strong sense. This is
not, therefore, a theory according to which there are really only pure minds
and their contents. Unlike Hilary, I should not take the correspondence
theory to be an integral ingredient of realism. Actually, there is not one cor-
respondence theory, but several, one of which is wrong because sentences do
not represent in the way that pictures do, and others of which are wrong
because, in Frege's terminology, and contrary to what Searle suggested, facts
are not to be taken as belonging to the realm of reference but to that of sense.
(In fact, Frege is a clear example of a realist who expressly rejected the cor-
respondence theory.) In so far as by 'the correspondence theory' is meant
only the principle that a statement cannot be true unless there is something
in virtue of which it is true, however, I should say that this is a principle that
must hold under any notion of truth; Thirdly, if the semantics is to allow the
practice of deductive inference as legitimate, it cannot take the true state-
ments to consist only of those that have been directly established as true.
Inferential reasoning allows us to establish the truth of a statement in-
directly - that is, by steps that do not reflect its composition - and hence
truth must be allowed to attach to statements that have not been directly
established, but only could have been so established, where the sense of
'could have' is shown by what is required for valid deduction. Fourthly, the
divergence from the realist picture of reality lies in the fact that that reality is
222 MICHAEL DUMMETT
not conceived of as fully determinate in the sense of determining every intel-

ligible statement as true or as false. What model do we have for such a
conception? The obvious models are those of feigned reality, such as a dream
or fiction: not every statement about the fictional world has a truth-value.
Just this analogy is what it is most dangerous to appeal to here: so let us
consider instead the case of a subjunctive conditional. 1 say, 'I wonder what
Harry would say if! were to tell him I had joined the Communist Party,' and
someone says, 'Try it and see,' so I do, with this or that result. We certainly
do not have to interpret the outcome as revealing what would have been the
answer to my question even if I had not made the experiment. Now, it may be
said, no one (well, hardly anyone) wants to maintain a realistic attitude to
counterfactual conditionals in general. No: but a realistic interpretation of
various categorical statements finds expression precisely in the conviction,
for certain ranges of counterfactuals with the same antecedent, that some
determinate one of them must be true. On a non-realistic interpretation, the
senses of those categorical statements have to be construed in the same way
as that of subjunctive conditionals in general: to know their sense is to know
the sort of thing required to establish them as true or as false, but they do
not have to be either, since there does not have to be anything such that, if
we knew it, we should recognise them as one or as the other.
The task of metaphysics is to say what, in general, the world (reality) is like
in itself, rather than as we apprehend it. And one of the tools in such an
inquiry is the distinction between hard facts and soft facts. Soft facts are ones
stated by forms of sentence the condition for whose truth - or for a justified
assertion of them - is either not fully determinate or else depends in part
upon facts about ourselves, e.g. about what we know and do not know, that
do not enter explicitly into the statement of the fact. Many facts which at
first sight appear hard are revealed upon reflection to be soft: Saul Kripke's
example iIIustrated beautifully one reason (there are others) for saying that
ascriptions of belief state only soft facts;1 another good example is that of
statements about causal relations or about what someone is free to do if he
chooses. The phenomenon of open texture, discussed by Avishai Margalit, is
another aspect of the softness of certain facts. Now our inclination is to try
to arrive at a realm of hard facts - those Saul Kripke referred to as what we
know when we 'know exactly what is going on' - as those which constitute
the substance of reality. If it be required of a hard fact that it should deter-
minately either obtain or not obtain, any rejection of the principle of biva-
lence for statements of a given kind wiII be construed as saying that those
COMMENTS 223
statements, when true, state only soft facts. Hence we feel more comfortable
when the rejection of realism takes the form of reductionism: the language
into which the reduction is made expresses the hard facts which constitute
reality as it is in itself. But to reject realism without espousing any reduc-
tionist thesis looks as though it means that we cannot struggle free of our own
perspective. 'Idealism,' in one sense, means the view that the only hard facts
are those which record our experience of the world: even though, ultimately,
there is nothing but our experience of reality, still, that we have such-and-such
experiences is something that holds good absolutely. But to deny that state-
ments for which bivalence fails can be translated into ones for which it holds
(a translation that will not preserve negation) appears a more radical repu-
diation of objective reality than idealism, since, if, to state a hard fact, a
statement must be subject to bivalence, it involves that reality cannot be
fully described by stating hard facts alone.
Now, to revert to Hilary's proposal. I have questioned whether a truth-
conditional semantics will really have a greater explanatory power than a
semantics of some non-realist variety. But we have also to ask: does a truth-
conditional semantics escape the objections which face it as a theory of
understanding by being transposed to an explanatory key? As I explained,
I had not fully grasped Hilary's proposal until I read the paper he has just
delivered; and, as a result, I have not had sufficient time to reflect on it
sufficiently to feel sure of the answer. I am inclined, however, to think that
the transposition does not enable a realist semantics to escape the objections
to it. For even if not so much as an implicit grasp of a realistic notion of truth
is required for a knowledge of the language, stilI, if it is by means of such a
notion that we are to explain the success of language, we must be able to
acquire the notion somehow. How, then, are we to acquire it? Hilary's view
is that a Tarskian truth-definition will suffice to introduce it. The objection
that such a truth-definition, to be understood, requires us already to under-
stand the language, is powerless in this context: according to Hilary's account,
we do already understand the language, namely by a mastery of its use. To
yield the required realistic semantics, the metalanguage in which the defini-
tion is given must have a classical logic; but there will be no difficulty in this,
on Hilary's account, since we do already understand the classical logical
constants simply by having been trained to perform deductions according to
the classical laws.
I am, indeed, fundamentally disinclined to suppose that a theory of under-
standing really can be given in terms of use in the way that Hilary suggests,
224 MICHAEL DUMMETT
that is, in semi-holistic terms, as consisting in a mastery of a number of inter-

locking practices connected by no unifying principle in terms of which it can
be stated in what a grasp of the content of a sentence consists, and by reference
to which the various linguistic practices, including that of deductive infer-
ence, can be accounted for and justified. I think that to attribute to a speaker
no more than a knowledge of how to play these interlocking language-games
is to make him a participant in an activity he cannot survey ('cannot see what
is going on,' in Saul Kripke's already quoted phrase), in just the sense in
which one may learn enough of what to do to take part in some social activity
in an alien culture without 'knowing what one is doing': such a speaker has,
as it were, no grasp of what a sentence says. But I have not chosen to try to
argue this; so let us go back to the truth-definition.
Let us suppose - what I should not grant - that a grasp of the classical
logical constants only requires a knowledge of the classical laws of logic,
rather than presupposing that we already possess that notion of truth in terms
of which the two-valued truth-tables are stated, showing us how a complex
sentence is determined as true or false in accordance with its composition.
We can still ask whether the truth-definition will really yield the two-valued
semantics that the realist requires. At first sight, it appears to, since we seem
to be able to derive, in the metalanguage, the principle of bivalence, the
truth-tables, and other principles of two-valued semantics. But I think not.
Other semantic theories than the two-valued one yield classical logic, e.g.,
any involving a relativised notion of truth and containing only constants
definable by relativised two-valued truth-operations (for which the truth-
value of a complex statement at a point depends only on the truth-values of
the constituents at that point). For instance, it is plausible to give such a
semantics for a language containing vague expressions: but the sentences of
that language will not satisfy bivalence. Still, if we give a Tarskian truth-
definition for that language, we shall in just the same way be able to 'prove'
the principle of bivalence: and this simply shows that the notion of truth
given by the truth-definition is not, after all, that notion of truth which we
need for a genuine semantic theory. In the case of the language involving
vagueness, we can easily see why no semantic notion of truth, satisfying
bivalence, is attainable for the language: we could not make the required
connection between any such notion and the actual use of the sentences. The
hypothesis that the speakers actually use the sentences so as to confer on them
determinate truth-conditions, even though they do not realise that they do,
would not be explanatory at all, but simply unintelligible.
COMMENTS 225
I may put the point as follows. For a semantic theory even to play the
explanatory role that Hilary desires, it must be a possible theory of under-
standing, even if not one that shows what in fact constitutes the speakers'
mastery of their language. Now the objections to a truth-conditional se-
mantics are objections to it as even an intelligible theory of understanding:
they say that, for a deeper reason than that stemming from vagueness, no
notion of truth satisfying bivalence could have the required connection with
use. And, in that case, the shift from representing the truth-conditional
semantics as an account of what in fact our understanding consists in to
representing it as an explanatory theory will not escape those objections.
HILARY PUTNAM
REPLY TO DUMMETT'S COMMENT
The problem with identifying truth with verification is that, apart from
mathematical sentences, verification (or warranted assertibility) is "a pro-
perty that may be lost." Michael concedes this is so in the case of sentences
about photons, on account of what he refers to as "the instability of scientific
theory"; but it seems to me to be equally so in the case of sentences about
tables and chairs. If "there is a table in the next room" is held to be verified
in experiential circumstances C, then, no matter how circumstances C may be
spelled out by the Verificationist (assuming we are given some finite descrip-
tion), we can easily tell a story consistent with physical theory about how we
might be caused to have those very experiences even though no table was
present in the next room. Worse still, for the Verificationist, once such a
story is told, we can easily think of ways that we could verify it ifit were true,
though not necessarily at the same time that the experiential circumstances C
obtain. So, on his own grounds, the Verificationist must regard it as possible
that this story should on some occasion be true, and hence regard it as possible
that the circumstances C obtained although there was no chair in the next
room.
Now there are two possibilities: the Verificationist can say that "there is a
chair in the next room" was verified at a certain time (when the circumstances
C obtained) although "there is a chair in the next room" was not true. But
then the Idealistic Fallacy argument has succeeded: being verified =1= being
true. Or he can look for another condition (different from C) for the sentence
being verified. But no matter what more expansive condition he may come
up with, it looks as if we can always repeat the argument. (A still more
radical possibility would be to say "there is a chair in the next room" was
true at the relevant time to, although "there was a chair in the next room
at to" is not now true; but this would totally modify the logic of tensed
statements as we have it.)
Michael suggests that the understanding-theoretic semantics of the
language - the description of what goes on "in OUI heads" when we under-
stand our words - be done in terms of the Verificationist notions of being
verified and falsified. I agree with Michael that method of verification is re-
226
REPLY TO DUMMETT'S COMMENT 227
levant to understanding on any intelligible account of understanding (and it

is one of the great meri.ts of Michael's work in philosophy of language to
have brought out the strength of the Verificationist position in semantics);
but I do not agree that the account of method of verification must proceed in
terms of the qualitative notion, verified/falsified. At least for an idealized
reconstruction, it seems better to take the quantitative notion of degree of
confirmation (or degree of verification) as the basis for the decription of use.
Since the trichotomy verified/falsified/neither is only a simplified version
(oversimplified, in the case where various possibilities of error and illusion is
under discussion) of degree of confirmation, I cannot agree that the two
accounts of use - the account based on a quantitative scheme of degree of
confirmation, and the account based on verification/falsification - differ
significantly in "surveyability." Of course, knowing how a sentence is used
is not "knowing what it says" - but Michael himself points out in his book
on Frege that accepting a verificationist semantics for a language does not
commit one to identifying the method of verification with what the sentence
is about. "What a sentence says" is an intra-theoretic (or sometimes an inter-
theoretic) question.
If the theory of use is done in terms of degree of confirmation, rather than
in terms of a dichotomous or trichotomous notion, and the explanation of
the success of our linguistic practice is done in terms of a realist notion of
truth, then it no longer seems as if two notions of truth are somehow floating
around: a realist one and a non-realist one. Each notion has its own place in
our total picture of language and its relation to the speaker's environment.
Now, Michael raises the following problem: suppose my language contains
some vague sentence, say, "The number of trees in Canada is even." And
suppose we accept classical logic (so we are committed to "the number of
trees in Canada is even or the number of trees in Canada is not even"). If we
define truth a la Tarski, we will get as a theorem "the sentence 'the number
of trees in Canada is even' is either true or false." Yet, viewing the situation
from the standpoint of a theory of how the words are used in the language,
we readily see that "the number of trees in Canada is even" has no truth-value,
and that the "internal" statement which appears to say that it does only
means that it would have a truth value if suitable conventions were laid down
(determining what counts as a "tree" and exactly where the border of
Canada is to the last micro-millimeter).
This observation does not contradict anything I wrote; indeed, in my
Aristotelian Society paper ("What is Realism"), I made a similar point,
228 HILARY PUTNAM
using an example based on Michael's work. My position is not that the laws
of logic determine our concept of truth, but that our entire theory, including
our theory of the relation of language to the speaker's environment, deter-
mines our concept of truth.
But Michael's rejoinder here will be that the entire language may have -
indeed does have - many interpretations, both in the model-theoretic sense
of "interpretation," and in the more radical sense in which the logical con-
stants themselves are allowed to be reinterpreted. And in what intelligible
sense of "intended" is just one of these interpretations the "intended" one?
In answer to this question, let me distinguish between internal realism and
metaphysical realism. Internal realism is a first order theory about the relation
of a language (actually, of the speakers of a language) to the speakers' en-
vironment. From within such a story, the notion of a "correspondence" be-
tween words and sets of things is as legitimate and meaningful as the notion
of a chair or a pain. What is well-defined and what is not, what has a truth-
value and what does not, how two-valued truth explains the success of
language-using, are all parts of the story (on all fours with the behavior of
electricity, or the properties of heat).
My position is that the notion of truth and classical logic have their proper
locus here: within such a theory.
Metaphysical realism is a picture (or a "model," in the sense in which
colliding billiard balls are a "model" for a gas) of the relation of any correct
theory to The World. The picture is that each term in a correct theory is a
label for a determinate piece (or kind of piece) of The World. Such a picture
faces many serious problems: the fact that different correct theories can be in
some intuitive sense "incompatible," so that we cannot say how The World
is; the fact that "translation" relations between different correct theories
(where they exist) are notoriously non-unique; and many others. Partly
because of Michael (and partly because of Nelson Goodman), I am inclined
to doubt the intelligibility of metaphysical realism. But this does not at all
undermine internal realism. Indeed, it permits an easy answer to the question
Michael raises: what is "vague" and what is not, what is an "intended" inter-
pretation and what is not, are questions with no absolute sense. They can
only be answered from the standpoint of one or another meta-theory.
Thus I am not dismayed, but rather excited and pleased, by the meta-
physical picture Michael paints of a world which cannot be described by
"hard facts alone." Internal realism is true; but perhaps this is itself a "soft
fact."
P.F. STRAWS ON
MAY BES AND MIGHT HAVE BEENS
My subject in this paper is particular possibilities: the may-bes and the

might-have-beens that relate essentially to particular individuals or situa-
tions. The topic is one which is apt to evoke very different responses from
different philosophers. Some detect, or think they detect, an intoxicating
scent of something more metaphysically interesting than either merely epis-
temic possibilities on the one hand or merely de dicto possibilities or neces-
sities on the other. My remarks will not give much satisfaction to them. Some,
on the other hand, are suspicious of modalities in general as being dubiously
coherent notions. But it is clear that whether we believe there are such things
as particular possibilities or not, we are in practice bound to take account of
them.
We are in practice bound to take account of them because our knowledge
of how things will turn out, or of how things have turned out, is imperfect.
We do not know that this will not happen, we do not know that that will not.
We do not know that this has not happened, we do not know that that has
not. When we find it worth while to reckon with our ignorance on these
points, then we are envisaging this and that as possibilities: this may (or
might) happen or that may (or might); this may (or might) have happened or
that may (or might).
If this were all, we could simply conclude that to say that something may
happen is merely to say that it is not certain that it will not; and to say that
something might have happened is simply to say that it is not certain that it
has not or did not. But this is not all. For we frequently use the language of
might-have-beens of what we know perfectly well not to have been. 'You
might have been killed,' for example, does not normally express uncertainty
as to whether you are still alive. Nevertheless I want to suggest that these
more philosophically interesting might-have-beens are also, fundamentally,
of an epistemic order.
How, then, are they related to the more obviously epistemic may-bes and
might-have-beens, the ones which express present uncertainty? I shall answer
that our philosophically interesting might-have-beens stand in a very simple
relation to our obviously epistemic may-bes. Just as 'He may' amounts
229
230 P. F. STRAWSON
roughly to (or at least includes) 'It is not certain that he won't,' so 'He might
have' amounts roughly to (or at least includes) 'It was not certain that he
wouldn't.' Just as 'He may,' in this sense, points forward from the present
time towards a now uncertain future, so 'He might have' - in the sense we
are interested in - points forward from a past time towards a then un-
certain future. Both require there to be a point in the history of the individual
concerned such that available knowledge regarding that individual at that
point does not exclude the future development which is problematically
affirmed of him or (for the negative case) does not guarantee what is pro-
blematically denied of him.
We shall see later on that this formula requires one quite radical amend-
ment. But the need for that amendment apart, the formula as it stands suffers
from another defect: in at least one important respect it is uncomfortably
vague or unclear. When we use 'may' or 'might' to express present uncer-
tainties about what is now future, the uncertainty is clearly relativized to a
time and, more or less clearly, to persons. The time is now; the persons our-
selves, the speaker and his circle and others he regards as authoritative,
perhaps. But if the interesting 'might have been' is to be related to past un-
certainties about what was then future, it is by no means clear how the
corresponding relativization should go.
Let us consider a few examples. Surveying through the window and the
storm the large, elderly, shallow-rooted tree on the edge of the wood, some-
one says: 'The tree may (might) fall on the house.' Trees are blown down in
gales. The height of this tree exceeds the distance between its base and the
house. The house-dwellers have not enough information about strains and
stresses, forces and directions, to calculate that the tree will fall or that it will
not, or to calculate that it will fall, if at all, on the house or to calculate that
it will fall, if at all, elsewhere. (Perhaps they prudently evacuate the house for
the duration of the storm.) After the storm, when the tree, say, either has not
fallen at all or has fallen, but not on the house, the same speaker or another
says: 'The tree might have fallen on the house.' The second remark differs
from the first only in tense. The first remark says: 'There is a non-negligible
chance that the tree will fall on the house.' The second says: 'There was a
non-negligible chance that the tree would fall on the house.' Both remarks
relate to the historical situation as it actually was during the storm. Neither
commits the speaker to the belief that the behaviour of the tree was anything
but fully determined by mechanical forces. Neither can be made sense of
without construing it as having an epistemic reference - a reference to the
MAY BES AND MIGHT HAVE BEENS 231
non-availability, at the relevant time, of adequate grounds for ruling out the
tree's falling on the house. Notice that a legitimate variant on the second
remark would be: 'There was a real possibility that the tree would fall on the
house.' So the real possibility that p does not exclude the causal necessity that
not-p, though it excludes knowledge, at the relevant time, of the causal
necessity that not-po
But whose knowledge? Things are not really so simple as the example
suggests. What contradicts 'There was a possibility that p' is 'There was no
possibility of that.' If someone says 'A might have been elected instead of B
to the Chair at X,' the reply might be 'If you consider the composition of the
electoral committee, you will see that there was no possibility of that.' This
does not imply, I think, that there was anyone at the time who, apprised of
the composition of the electoral committee, would then and there have been
able to infer, with practical certainty, that A would not be elected. It means,
perhaps, that there were enough facts distributed in different minds to make
this conclusion virtually certain; but not that they were contained in one
mind. We can see that after a certain time (the time at which the composition
of the board was settled) there no longer existed a certain possibility which,
however, did not seem rationally excluded to anyone at that time; because,
though the relevant facts (about, say, the attitudes and preferences of mem-
bers of the board) were then 'available,' no one then was, as we now are,
master of them all. Historians, I suppose, often think along these lines. It is
called having the benefit of hindsight. By then, they say, the issue of the battle
was certain: there was no longer any possibility of the Austrians' winning.
The verdict is not that of omniscience, which is not interested in possibility.
It is, rather, that of an ideal intelligence officer collecting reasonably full and
accurate reports from all parts of the battlefield. But of course there was no
such officer receiving such reports. The issue of the battle was certain before
anyone actually present was in a position to be rationally certain of it.
I think we go yet further in our verdicts on past possibilities, taking into
account not only the evidence, the particular facts, collectively available at
the time but at the time uncollected, but also general truths now known but
then unknown, and even particular truths relating to that time now known
but then unknown. The verdict, then, becomes not simply that of the ideal
contemporary intelligence officer, but of such an officer further endowed with
knowledge of relevant facts and laws which have only subsequently become
known. This addition cuts both ways: allowing us not only to exclude, with
reference to a particular past situation, some might-have-beens then envi-
232 P. F. STRAWS ON
saged, perhaps, as may-bes, but also to contemplate, with reference to a

particular past situation, some might-have-beens which no one then would
have been in a position to consider as may-bes.!
These remarks are designed to expand on the force of 'It was not certain'
in the formula 'It was not certain that it would not be the case that p.' Inter
alia they show something about how to contradict it - with reference to a
particular time. But of course contradicting a statement of this kind relati-
vized to a particular time leaves uncontradicted any corresponding statement
relating to an earlier time. Remarks of the form 'It might have been the case
that p' are, in general, more and more vulnerable to contradiction as they are
more and more closely related to the time at which events definitively falsify
the proposition that p; are correspondingly less and less vulnerable as their
reference is shifted back in time; and assume their least vulnerable shape if
they are construed generally, as amounting to 'There was a time at which it
was not certain that it would not be the case that p.' So even though by 6
o'clock the issue of the battle was, humanly speaking, certain, even though
there was by then no possibility of the Austrians' winning, it is stilI true, on
some temporal relativization of those words, that they might have won; what
is not true is that at 6 p.m. they might still have won. Similarly, A might have
been elected instead of B, even though, after the composition of the electoral
board was determined, this was no longer a possibility.
The formula we have before us, then, runs roughly as follows: some pro-
position to the effect that a might have d is true (acceptable) if and only if
there was some point in the history of the individual concerned such that
presently available knowledge regarding that point does not permit the
rational inference that a did not (or, in other words, the facts as we know
them left open at least a chance that a would ). Similarly, some proposition
to the effect that a might not have d is true (acceptable) if and
only if there was some time in the history of a such that our present know-
ledge of the facts as they were then does not permit the rational inference
that a did (i.e. the facts as we know them left open at least a chance that a
would not ). I have been saying something about how the phrase 'available
knowledge' should here be construed; and if this phrase stilI remains a little
indeterminate, I think there is no harm in that, for that is a feature also ofthe
idiom we are trying to characterize.2
The formula as we have it obviously restricts the reference of our might-
have-beens to times falling within the history of the individual (or individuals)
concerned, hence to times at which propositions concerning that individual
MAY DES AND MIGHT HAVE DEENS 233
would be in principle entertainable. But there is one subclass of our interesting

might-have-beens in respect of which we are obliged to lift this restriction on
the temporal reference of our might-have-beens, and hence, as I hinted
earlier, radically to amend our formula. For it seems clearly true, of any
particular individual that you like to name, that that individual might not
have existed; and so we have a whole host of true might-have-been proposi-
tions which either directly exhibit the form 'a might not have existed' or
entail another proposition which exhibits it. Thus Aristotle might never have
been born and so might never have existed. Now it is clear that if these pro-
positions are to be regarded as belonging to the class we are interested in -
we might call it the class of objective historical epistemic propositions - they
will often relate to a time anterior to any time which can reasonably be
regarded as falling within the history of the individual concerned.
If we now consider the kind of reason we might most naturally give in
support of such propositions, it becomes clear also that such propositions, so
supported, do belong to the class we are interested in. Thus, putative mothers-
to-be do sometimes miscarry. Even given that his father was a physician, it
was not certain that Aristotle's mother would not miscarry of the child she
became pregnant with when she became pregnant with Aristotle. There was,
moreover, a time when it was not certain that Aristotle's mother and his
father would mate or that, if they mated, their union would be fertile.
Aristotle's mother, then, might have miscarried in the first months of the
pregnancy which terminated in the delivery of Aristotle. Or Aristotle's parents
might never have enjoyed a fertile union. So Aristotle might not have
existed.
The then existing uncertainties which underlie these might-have-beens we
can express or lecord by referring to Aristotle, though they could not then be
so recorded or expressed. But this makes no matter. It is still a historical-
epistemic truth about the couple who were in fact Aristotle's parents that
thele was a time at which it was not certain that they would enjoy a fertile
union; hence that they might not have done so; hence that Aristotle might
not have existed.
And so we bring these existential might-not-have-beens into the fold; and
we do so by extending the historical reference of our might-have-beens back
into what we might call the pre-history of the individuals concerned. It is
easy enough to see how to do this in the case of individuals which come into
existence by a process of natural generation. We can reasonably take it that
there were many relevant uncertainties along the historical, the genealogical,
234 P. F. STRAWS ON
lines of transmission which in fact lead to their existence. Matters are not so
straightforward in the case of artefacts. Some things are simple enough. This
table might not have existed. At one time it was not certain that the wood of
which it is composed would go into the construction of a table instead of, say,
a chest of drawers. The craftsman's decisions as to what to make depend, say,
on the orders of customers, and there is a time at which it is not certain what
those orders will be, for the customers have not made up their minds. It is
idiomatically permissible to express such a possibility by saying that this
table might have been a chest of drawers, or even - if, say, there was, at some
time, quite a likelihood of one kind of order arriving before the other- that
this table might easily have been a chest of drawers. But we cannot para-
phrase the idiom by 'This table might have existed in the form of a chest of
drawers.' The correct paraphrase is as already indicated: 'The materials which
compose this table might have composed a chest of drawers,' which entails
'This table might not have existed.'
Could it be that this table might have been two feet shorter or that this
table might have been oval (instead of, as it is, rectangular)? Yes. But we feel
the need of a true story to back up these specific might-have-beens whereas
we are free to presume some story or other to back up the unspecific 'This
table might not have existed.' There was hesitation, perhaps, over the specifi-
cations; or some unforeseeable chance determined the final plan. Without
such a story these might-have-beens leave us fairly blank. We are not prepared
to deny them; but we have no reason to affirm them.
Might this table have been made of marble instead of wood? We are
strongly, and reasonably, inclined to deny this without waiting for a story.
For what we are looking for, in the prehistory of artefacts, is an analogue of
forebears in the prehistory of animals; and the materials of which the artefact
is composed - not just the general type of material, but the particular speci-
mens of those types - are the obvious analogues. Or at least they are in some
cases, such as the case of this particular table; an object, say, of a fairly
standard kind. Nothing less, or other, than the particular materials will serve
our turn here. But may not the case be different with more elaborate human
constructions? What of a great temple or a transatlantic liner? The gleat
building is conceived, we say, in the brain of such-and-such a designer or the
collective brain of a committee of designers. The quarries at A and the
quarries at B are equally capable of supplying stone of the desired type. There
are advantages and disadvantages in the choice of either source of supply.
The stone actually comes from quarry B. But there was a time at which it was
not certain that the stone would not come from quarry A. Perhaps it seemed
quite likely that it would. So the building in question - the Old Bodleian,
say - might have been built of (composed of) stone from quarry A instead
of being built, as it was, of stone from quarry B.
Will someone say: then it would not have been this building, but another
just like it? The retort seems insufficiently motivated. Before the building
existed, there existed a plan: a plan for a building on this site, for this purpose,
to be constructed of such-and-such type-materials according to such-and-such
architectural specifications. Here we have all the prehistory we need. The
building, this building, is begotten of a particular project rather than a
particular scattered part of the earth's material. If someone said: 'The QE II,
you know, might have been built of quite a different lot of steel from that
which it was actually built of' - and gave his reasons - would it not be
absurd to reply: 'In that case it wouldn't have been the QE II at all - the
QE II wouldn't have existed - it would have been a different ship of that
name.'?
If we can go so far, perhaps we can go farther. There is, say, a point in the
history of our building project at which it becomes uncertain that stone of
the desired type will be readily available from any source; or there is diffe-
rence of opinion as to what type of stone is desirable. The difficulties are
resolved and the building is built of Portland stone; but it might have been
built of Bath stone. Shall we say: 'Then we would not have had this building'?
Well, we can certainly say: 'We would not have had these stones.' (And
someone might mean just that by saying 'Then we would not have had this
building'; as, similarly, tapping the side of the ship, he might say: 'Then we
should not have had just this ship.')
I shall not pursue this question further - interesting though it is. Instead I
return to the primary purpose of getting clear - or clearer - about the
general character of particular might-have-beens. I summarise the main
points.
First, then, a particular 'might have been' statement of the kind which
concerns us is a historical statement. It relates to the past, and looks to what
was once the future. It relates either to a past recent enough, in the usual
case, to include the existence of the individual referred to (or the most recent
of those individuals if more than one is referred to); or at least to a past which
includes that individual's immediate or remoter 'progenitors' - in either the
literal or some appropriate analogical sense of that word. So the statement
relates to a past about which facts are available concerning the individual or his
236 P. F. STRAWS ON
progenitors. The general form of the claim made in such a statement, with an
understood reference to some such past time, is that the facts available then-
even reinforced with our later knowledge - were not such as to give sufficient
grounds for certainty, i.e. for practical or human certainty, that the then
future would not turn out in a way incompatible, in some respect relating to
the individual in question, with the way it did in fact turn out.
Such a statement normally relates-and when, as in non-degenerate cases,
it is backed by some specific particular story, it always does relate - to some
fairly definite time. These are the cases in which our might-have-beens most
obviously admit of some qualification of degree: it might very easily have
been so, there was a very good chance (a very real possibility) that it would be
so; or again, it was just possible that it would be so. The later the time our
might-have-been statements refer to, the more vulnerable, in general, they
become to the form of objection: 'By then there was no longer any possibility
of that.' As time advances, evidence accumulates and chance decays till a
point is reached at which it becomes humanly certain that it will not be the
case thatp.
When we contemplate future possibilities (may-bes), our interest is often
practical. When we contemplate past (and unfulfilled) possibilities (might-
have-beens), our interest is normally historical. Our may-bes and our might-
have-beens alike can be more, or less, serious. A 'serious' historical interest
in might-have-beens is generally an interest in might-have-beens which do
relate to a fairly definite situation at a fairly definite time, fairly close to the
time at which what might have been turned out not to be. But nothing com-
pels us to definiteness, as nothing compels us to seriousness. We can now, for
practical purposes, seriously consider likely future contingencies. But we can
also vaguely dream about quite hidden futures, rosily or gloomily; so the
parent, contemplating the newborn child, thinks: 'He may become Lord
ChanceJlor' or 'He may be killed in a war or a motor-accident,' when little
or nothing is available in the way of particular fact to favour or disfavour
either chance. Since all times were once hidden futures, we enjoy no less a
licence in our idle might-have-beens than in our idle may-bes. Indeed, in at
least one radical respect we enjoy a much greater freedom. For our idlest
speculations about what are now future possibilities relate to a definite
temporal base: the present of those speCUlations. They are constrained within
the limits set by history up to now. But our idle speCUlations on past possibi-
lities need relate to no definite temporal base. They can rove indefinitely back
through the history, and even the prehistory, of the individual concerned. The
greater the temporal range, the greater the range of chances. Time and
chance governeth all things; but, at any rate in the present context, time
governeth chance. For the chances we here speak of stand in contrast, not
with necessities, but with certainties. In philosophy, perhaps, we tend to lose
our sense of the continuity between our serious might-have-beens and our
relatively frivolous might-have-beens and thus run the risk of missing the
point ofthe idiom altogether. And that would be a pity. For here we have an
actual use of the possible. I do not suggest it is the only use.
Magdalen College
Oxford
NOTES
1 At the limit of this process we can consider might-have-beens in relation to a period at

which no one envisaged any may-bes at all because there was no one to envisage them. In
such a case it is, of course, later knowledge which supplies all the facts of which we hold
that they did not make it certain that such-and-such (which did not in fact happen) would
not subsequently happen, i.e. the 'available facts' in relation to which we may say that
such-and-such might have happened. But nothing in our saying this implies any indeter-
minism in what did happen.
2 Objection 1. Suppose a's cPing, or not cPing, is the sort of outcome which can rea-
sonably be regarded as exclusively subject, at least from the reference-time onwards, to
deterministic laws. In that case our knowledge that a did not in fact (or did in fact) cP
entitles us to infer, regarding the reference-time, that circumstances then obtained which
were sufficient to ensure a's not cPing (or cPing). So the condition which the analysis requires
for an acceptable 'might have been' in such a case, viz. that knowledge now available
regarding the reference-time is insufficient to warrant the rational inference that a did not
cP (or that a did cP), is never fulfilled. Indeed a convinced universal determinist would have
no use at all for the idiom in the sense expounded.
Reply. The objection construes 'now available knowledge' too widely. What is required
for denying a reasonably supported assertion that there was a chance that a would cP (or
that a would not cP) is some more specific knowledge of conditions obtaining at the relevant
time than can be derived from the premise that a did not in fact cP (or did in fact cP)
coupled with a general conviction that a's cPing or not cPing is subject to deterministic
laws. There is no reason to think that convinced universal determinists would (or do)
eschew the idiom in the sense expounded or confine themselves to denying others' uses
of it.
Objection 2. Suppose the reference-time is t. Then knowledge available at one time
(say 1970) of the relevant circumstances of a at t may seem to warrant the assertion that a
might have cPd, whereas additional knowledge available at a later time (say 1975) may seem
to warrant its contradiction. The analysis then requires us to say that the proposition
238 P. F. STRAWSON
that a might have c/Jd, relativized to t, is true in 1970 but false in 1975. This conclusion is
unattractive.
Replies: (a) There is not one proposition but two; for the analysis of the idiom brings
out a concealed indexical element in it, viz. a reference to the state of knowledge at the
time of utterance.
(b) The 'analysis' may be seen as giving not truth conditions but 'justified
assertibility' conditions. The notion of truth conditions is inappropriate to the idiom itself,
though not, of course, to the statement of justifying conditions.
Evidently replies (a) and (b) are mutually exclusive alternatives. I leave open the choice
between them.
(In formulating these objections and replies, I have been greatly helped by the contribu-
tions of Professors S. Kripke and H. Gaifman to the discussion of this paper.)
SAUL A. KRIPKE
A PUZZLE ABOUT BELIEF
In this paper I will present a puzzle about names and belief. A moral or two
will be drawn about some other arguments that have occasionally been ad-
vanced in this area, but my main thesis is a simple one: that the puzzle is a
puzzle. And, as a corollary, that any account of belief must ultimately come
to grips with it. Any speculation as to solutions can be deferred.
The first section of the paper gives the theoretical background in previous
discussion, and in my own earlier work, that led me to consider the puzzle.
The background is by no means necessary to state the puzzle: As a philoso-
phical puzzle, it stands on its own, and I think its fundamental interest for
the problem of belief goes beyond the background that engendered it. As I
indicate in the third section, the problem really goes beyond beliefs expressed
using names, to a far wider class of beliefs. Nevertheless, I think that the back-
ground illuminates the genesis of the puzzle, and it will enable me to draw
one moral in the concluding section.
The second section states some general principles which underlie our ge-
neral practice of reporting beliefs. These principles are stated in much more
detail than is needed to comprehend the puzzle; and there are variant for-
mulations of the principles that would do as well. Neither this section nor the
first is necessary for an intuitive grasp of the central problem, discussed in
the third section, though they may help with fine points of the discussion. The
reader who wishes rapid access to the central problem could skim the first
two sections lightly on a first reading.
In one sense the problem may strike some as no puzzle at all. For, in the
situation to be envisaged, all the relevant facts can be described in one termi-
nology without difficulty. But, in another terminology, the situation seems to
be impossible to describe in a consistent way. This will become clearer later.
I. PRELIMINARIES: SUBSTITUTIVITY
In other writings,1 I developed a view of proper names closer in many ways

to the old Millian paradigm of naming than to the Fregean tradition which
probably was dominant until recently. According to Mill, a proper name is,
so to speak, simply a name. It simply refers to its bearer, and has no other
239
A. Margalit (ed.) , Meaning and Use, 239-283. Dordrecht, D. Reidel. All Rights Reserved.
Copyright 1979 by Saul A. Kripke.
240 SAUL A. KRIPKE
linguistic function. In particular, unlike a definite description, a name does

not describe its bearer as possessing any special identifying properties.
The opposing Fregean view holds that to each proper name, a speaker of
the language associates some property (or conjunction of properties) which
determines its referent as the unique thing fulfilling the associated property
(or properties). This property(ies) constitutes the 'sense' of the name. Pre-
sumably, if ' .. .' is a proper name, the associated properties are those that
the speaker would supply, if asked, "Who is ' .. .'?" If he would answer
" ... is the man who ," the properties filling the second blank are
those that determine the reference of the name for the given speaker and
constitute its 'sense.' Of course, given the name of a famous historical figure,
individuals may give different, and equally correct, answers to the "Who
is ... ?" question. Some may identify Aristotle as the philosopher who taught
Alexander the Great, others as the Stagirite philosopher who studied with
Plato. For these two speakers, the sense of "Aristotle" will differ: in partic-
ular, speakers of the second kind, but not ofthe first kind, will regard "Aris-
totle, if he existed, was born in Stagira" as analytic. 2 Frege (and Russell)3
concluded that, strictly speaking, different speakers of English (or German!)
ordinarily use a name such as 'Aristotle' in different senses (though with the
same reference). Differences in properties associated with such names, strictly
speaking, yield different idiolects. 4
Some later theorists in the Frege-Russellian tradition have found this con-
sequence unattractive. So they have tried to modify the view by 'clustering'
the sense of the name (e.g., Aristotle is the thing having the following long
list of properties, or at any rate most of them), or, better for the present
purpose, socializing it (what determines the reference of 'Aristotle' is some
roughly specified set of community-wide beliefs about Aristotle).
One way to point up the contrast between the strict Millian view and
Fregean views involves - if we permit ourselves this jargon - the notion of
propositional content. If a strict Millian view is correct, and the linguistic
function of a proper name is completely exhausted by the fact that it names
its bearer, it would appear that proper names of the same thing are every-
where interchangeable not only salva veritate but even salva significatione:
the proposition expressed by a sentence should remain the same no matter
what name of the object it uses. Of course this will not be true if the
names are 'mentioned' rather than 'used': "'Cicero' has six letters"
differs from " 'Tully' has six letters" in truth value, let alone in content.
(The example, of course, is Quine's.) Let us confine ourselves at this stage to
A PUZZLE ABOUT BELIEF 241
simple sentences involving no connectives or other sources of intensionality.

If Mill is completely right, not only should "Cicero was lazy" have the same
truth value as "Tully was lazy," but the two sentences should express the
same proposition, have the same content. Similarly "Cicero admired Tully,"
"Tully admired Cicero," "Cicero admired Cicero," and "Tully admired
Tully," should be four ways of saying the same thing. 5
If such a consequence of Mill's view is accepted, it would seem to have
further consequences regarding 'intensional' contexts. Whether a sentence
expresses a necessary truth or a contingent one depends only on the proposi-
tion expressed and not on the words used to express it. So any simple sentence
should retain its 'modal value' (necessary, impossible, contingently true, or
contingently false) when 'Cicero' is replaced by 'Tully' in one or more places,
since such a replacement leaves the content of the sentence unaltered. Of
course this implies that coreferential names are substitutable in modal con-
texts salva veritate: "It is necessary (possible) that Cicero ... " and "It is
necessary (possible) that Tully ... " must have the same truth value no matter
how the dots are filled by a simple sentence.
The situation would seem to be similar with respect to contexts involving
knowledge, belief, and epistemic modalities. Whether a given subject believes
something is presumably true or false of such a subject no matter how that
belief is expressed; so if proper name substitution does not change the content
of a sentence expressing a belief, coreferential proper names should be inter-
changeable salva veritate in belief contexts. Similar reasoning would hold for
epistemic contexts ("Jones knows that ... ") and contexts of epistemic ne-
cessity (" Jones knows a priori that ... ") and the like.
All this, of course, would contrast strongly with the case of definite descrip-
tions. It is well known that substitution of coreferential descriptions in simple
sentences (without operators), on any reasonable conception of 'content,' can
alter the content of such a sentence. In particular, the modal value of a sen-
tence is not invariant under changes of coreferential descriptions: "The
smallest prime is even" expresses a necessary truth, but" Jones's favorite
number is even" expresses a contingent one, even if Jones's favorite number
happens to be the smallest prime. It follows that coreferential descriptions
are not interchangeable salva veritate in modal contexts: "It is necessary that
the smallest prime is even" is true while "It is necessary that Jones's favorite
number is even" is false.
Of course there is a 'de re' or 'large scope' reading under which the second
sentence is true. Such a reading would be expressed more accurately by
242 SAUL A. KRIPKE
"Jones's favorite number is such that it is necessarily even" or, in rough

Russellian transcription, as "One and only one number is admired by Jones
above all others, and any such number is necessarily even (has the property of
necessary evenness)." Such a de re reading, ifit makes sense at all, by defini-
tion must be subject to a principle of substitution salva veritate, since ne-
cessary evenness is a property of the number, independently of how it is
designated; in this respect there can be no contrast between names and des-
criptions. The contrast, according to the Millian view, must come in the de
dicto or "small scope" reading, which is the only reading, for belief contexts
as well as modal contexts, that will concern us in this paper. If we wish, we
can emphasize that this is our reading in various ways. Say, "It is necessary
that: Cicero was bald" or, more explicitly, "The following proposition is
necessarily true: Cicero was bald," or even, in Carnap's 'formal' mode of
speech,6 " 'Cicero was bald' expresses a necessary truth." Now the Millian
asserts that all these formulations retain their truth value when 'Cicero' is
replaced by 'Tully,' even though 'Jones's favorite Latin author' and 'the man
who denounced Catiline' would not similarly be interchangeable in these
contexts even if they are codesignative.
Similarly for belief contexts. Here too de re beliefs - as in "Jones believes,
of Cicero (or: of his favorite Latin author) that he was bald" do not concern
us in this paper. Such contexts, if they make sense, are by definition subject
to a substitutivity principle for both names and descriptions. Rather we are
concerned with the de dicto locution expressed explicitly in such formulations
as, "Jones believes that: Cicero was bald" (or: "Jones believes that: the man
who denounced Catiline was bald"). The material after the colon expresses
the content of Jones's belief. Other, more explicit, formulations are: "Jones
believes the proposition - that - Cicero - was - bald," or even in the
'formal' mode, "The sentence 'Cicero was bald' gives the content of a belief
of Jones." In all such contexts, the strict Millian seems to be committed to
saying that codesignative names, but not co designative descriptions, are
interchangeable salva veritate. 7
Now it has been widely assumed that these apparent consequences of the
Millian view are plainly false. First, it seemed that sentences can alter their
modal values by replacing a name by a codesignative one. "Hesperus is
Hesperus" (or, more cautiously: "If Hesperus exists, Hesperus is Hesperus")
expresses a necessary truth, while "Hesperus is Phosphorus" (or: "If Hes-
perus exists, Hesperus is Phosphorus"), expresses an empirical discovery,
and hence, it has been widely assumed, a contingent truth. (It might have
turned out, and hence might have been, otherwise.)

It has seemed even more obvious that codesignative proper names are not
interchangeable in belief contexts and epistemic contexts. Tom, a normal
speaker of the language, may sincerely assent to "Tully denounced Catiline,"
but not to "Cicero denounced Catiline." He may even deny tne latter. And
his denial is compatible with his status as a normal English speaker who
satisfies normal criteria for using both 'Cicero' and 'Tully' as names for the
famed Roman (without knowing that 'Cicero' and 'Tully' name the same
person). Given this, it seems obvious that Tom believes that: Tully de-
nounced Catiline, but that he does not believe (lacks the belief) that: Cicero
denounced Catiline. 8 So it seems clear that codesignative proper names are
not interchangeable in belief contexts. It also seems clear that there must be
two distinct propositions or contents expressed by 'Cicero denounced Cati-
line' and 'Tully denounced Catiline.' How else can Tom believe one and
deny the other? And the difference in propositions thus expressed can only
come from a difference in sense between 'Tully' and 'Cicero.' Such a con-
clusion agrees with a Fregean theory and seems to be incompatible with a
purely Millian view. 9
In the previous work mentioned above, I rejected one of these arguments
against Mill, the modal argument. 'Hesperus is Phosphorus,' I maintained,
expresses just as necessary a truth as 'Hesperus is Hesperus'; there are no
counterfactual situations in which Hesperus and Phosphorus would have
been different. Admittedly, the truth of 'Hesperus is Phosphorus' was not
known a priori, and may even have been widely disbelieved before appropriate
empirical evidence came in. But these epistemic questions should be se-
parated, I have argued, from the metaphysical question of the necessity of
'Hesperus is Phosphorus.' And it is a consequence of my conception of
names as 'rigid designators' that co designative proper names are inter-
changeable salva veritate in all contexts of (metaphysical) necessity and
possibility; further, that replacement of a proper name by a codesignative
name leaves the modal value of any sentence unchanged.
But although my position confirmed the Millian account of names in modal
contexts, it equally appears at first blush to imply a nonMillian account of
epistemic and belief contexts (and other contexts of propositional attitude).
For I presupposed a sharp contrast between epistemic and metaphysical
possibility: Before appropriate empirical discoveries were made, men might
well have failed to know that Hesperus was Phosphorus, or even to believe it,
even though they of course knew and believed that Hesperus was Hesperus.
244 SAUL A. KRIPKE
Does not this support a Fregean position that 'Hesperus' and 'Phosphorus'
have different 'modes of presentation' that determine their references? What
else can account for the fact that, before astronomers identified the two
heavenly bodies, a sentence using 'Hesperus' could express a common belief,
while the same context involving 'Phosphorus' did not? In the case of
'Hesperus' and 'Phosphorus,' it is pretty clear what the different 'modes of
presentation' would be: one mode determines a heavenly body by its typical
position and appearance, in the appropriate season, in the evening; the other
determines the same body by its position and appearance, in the appropriate
season, in the morning. So it appears that even though, according to my view,
proper names would be modally rigid - would have the same reference when
we use them to speak of counterfactual situations as they do when used to
describe the actual world - they would have a kind of Fregean 'sense' ac-
cording to how that rigid reference is fixed. And the divergences of 'sense'
(in this sense of 'sense') would lead to failures of interchangeability of co-
designative names in contexts of propositional attitude, though not in modal
contexts. Such a theory would agree with Mill regarding modal contexts but
with Frege regarding belief contexts. The theory would not be purely
Millian.l0
After further thought, however, the Fregean conclusion appears less
obvious. Just as people are said to have been unaware at one time of the fact
that Hesperus is Phosphorus, so a normal speaker of English apparently may
not know that Cicero is Tully, or that Holland is the Netherlands. For he
may sincerely assent to 'Cicero was lazy,' while dissenting from 'Tully was
lazy,' or he may sincerely assent to 'Holland is a beautiful country,' while
dissenting from 'The Netherlands is a beautiful country.' In the case of
'Hesperus' and 'Phosphorus,' it seemed plausible to account for the parallel
situation by supposing that 'Hesperus' and 'Phosphorus' fixed their (rigid)
references to a single object in two conventionally different ways, one as the
'evening star' and one as the 'morning star.' But what corresponding con-
ventional 'senses,' even taking 'senses' to be 'modes of fixing the reference
rigidly,' can plausibly be supposed to exist for 'Cicero' and 'Tully' (or
'Holland' and 'the Netherlands')? Are not these just two names (in English)
for the same man? Is there any special conventional, community-wide 'conno-
tation' in the one lacking in the other?l1 I am unaware ofany.1 2
Such considerations might seem to push us toward the extreme Frege-
Russellian view that the senses of proper names vary, strictly speaking, from
speaker to speaker, and that there is no community-wide sense but only a
community-wide reference. 13 According to such a view, the sense a given

speaker attributes to such a name as 'Cicero' depends on which assertions
beginning with 'Cicero' he accepts and which of these he regards as defining,
for him, the name (as opposed to those he regards as mere factual beliefs
'about Cicero'). Similarly, for 'Tully.' For example, someone may define
'Cicero' as 'the Roman orator whose speech was Greek to Cassius,' and
'Tully' as 'the Roman orator who denounced Catiline.' Then such a speaker
may well fail to accept 'Cicero is Tully' if he is unaware that a single orator
satisfied both descriptions (if Shakespeare and history are both to be be-
lieved). He may well, in his ignorance, affirm 'Cicero was bald' while rejecting
'Tully was bald,' and the like. Is this not what actually occurs whenever
someone's expressed beliefs fail to be indifferent to interchange of 'Tully' and
'Cicero'? Must not the source of such a failure lie in two distinct associated
descriptions, or modes of determining the reference, of the two names? If a
speaker does, as luck would have it, attach the same identifying properties
both to 'Cicero' and to 'Tully,' he will, it would seem, use 'Cicero' and 'Tully'
interchangeably. All this appears at first blush to be powerful support for the
view of Frege and Russell that in general names are peculiar to idiolects,
with 'senses' depending on the associated 'identifying descriptions.'
Note that, according to the view we are now entertaining, one cannot say,
"Some people are unaware that Cicero is Tully.' For, according to this view,
there is no single proposition denoted by the 'that' clause, that the community
of normal English speakers expresses by 'Cicero is Tully.' Some - for
example, those who define both 'Cicero' and 'Tully' as 'the author of De
Fato' - use it to express a trivial self-identity. Others use it to express the
proposition that the man who satisfied one description (say, that he de-
nounced Catiline) is one and the same as the man who satisfied another (say,
that his speech was Greek to Cassius). There is no single fact, 'that Cicero is
Tully,' known by some but not all members of the community.
If I were to assert, "Many are unaware that Cicero is Tully," I would use
'that Cicero is Tully' to denote the proposition that I understand by these
words. If this, for example, is a trivial self-identity, I would assert falsely,
and irrelevantly, that there is widespread ignorance in the community of a
certain self-identity.1 4 I can, of course, say, "Some English speakers use both
'Cicero' and 'Tully' with the usual referent (the famed Roman) yet do not
assent to 'Cicero is Tully.' "
This aspect of the Frege-Russellian view can, as before, be combined with
a concession that names are rigid designators and that hence the description
246 SAUL A. KRIPKE
used to fix the reference of a name is not synonymous with it. But there are
considerable difficulties. There is the obvious intuitive unpalatability of the
notion that we use such proper names as 'Cicero,' 'Venice,' 'Venus' (the
planet) with differing 'senses' and for this reason do not 'strictly speaking'
speak a single language. There are the many well-known and weighty objec-
tions to any description or cluster-of-descriptions theory of names. And is it
definitely so clear that failure of interchangeability in belief contexts implies
some difference of sense? After all, there is a considerable philosophical
literature arguing that even word pairs that are straightforward synonyms
if any pairs are - "doctor" and "physician," to give one example - are
not interchangeable salva veritate in belief contexts, at least if the belief
operators are iterated.l s
A minor problem with this presentation of the argument for Frege and
Russell will emerge in the next section: if Frege and Russell are right, it is
not easy to state the very argument from belief contexts that appears to sup-
port them.
But the clearest objection, which shows that the others should be given
their proper weight, is this: the view under consideration does not in fact
account for the phenomena it seeks to explain. As I have said eIsewhere,16
individuals who "define 'Cicero'" by such phrases as "the Catiline de-
nouncer," "the author of De Fato," etc., are relatively rare: their prevalence
in the philosophical literature is the product of the excessive classical learning
of some philosophers. Common men who clearly use 'Cicero' as a name for
Cicero may be able to give no better answer to "Who was Cicero?" than "a
famous Roman orator," and they probably would say the same (if anything!)
for 'Tully.' (Actually, most people probably have never heard the name
'Tully.') Similarly, many people who have heard of both Feynman and Gell-
Mann, would identify each as 'a leading contemporary theoretical physicist.'
Such people do not assign 'senses' of the usual type to the names that uniquely
identify the referent (even though they use the names with a determinate
reference). But to the extent that the indefinite descriptions attached or
associated can be called 'senses,' the 'senses' assigned to 'Cicero' and 'Tully,'
or to 'Feynman' and 'Gell-Mann,' are identica1.l 7 Yet clearly speakers of this
type can ask, "Were Cicero and Tully one Roman orator, or two different
ones?" or "Are Feynman and Gell-Mann two different physicists, or one?"
without knowing the answer to either question by inspecting 'senses' alone.
Some such speaker might even conjecture, or be under the vague false im-
pression, that, as he would say, 'Cicero was bald but Tully was not.' The
premise of the argument we are considering for the classic position of Frege
and Russell - that whenever two codesignative names fail to be inter-
changeable in the expression of a speaker's beliefs, failure of interchangea-
bility arises from a difference in the 'defining' descriptions the speaker
associates with these names - is, therefore, false. The case illustrated by
'Cicero' and 'Tully' is, in fact, quite usual and ordinary. So the apparent
failure of codesignative names to be everywhere interchangeable in belief
contexts, is not to be explained by differences in the 'senses' of these names.
Since the extreme view of Frege and Russell does not in fact explain the
apparent failure of the interchangeability of names in belief contexts, there
seems to be no further reason - for present purposes - not to give the other
overwhelmingprimafacie considerations against the Frege-Russell view their
full weight. Names of famous cities, countries, persons, and planets are the
common currency of our common language, not terms used homonymously
in our separate idiolects.1 8 The apparent failure of codesignative names to be
interchangeable in belief contexts remains a mystery, but the mystery no
longer seems so clearly to argue for a Fregean view as against a Millian one.
Neither differing public senses nor differing private senses peculiar to each
speaker account for the phenomena to be explained. So the apparent exis-
tence of such phenomena no longer gives a prima facie argument for such
differing senses.
One final remark to close this section. I have referred before to my own
earlier views in "Naming and Necessity." I said above that these views, inas-
much as they make proper names rigid and transparent 19 in modal contexts,
favor Mill, but that the concession that proper names are not transparent in
belief contexts appears to favor Frege. On a closer examination, however, the
extent to which these opacity phenomena really support Frege against Mill
becomes much more doubtful. And there are important theoretical reasons
for viewing the "Naming and Necessity" approach in a Millian light. In that
work I argued that ordinarily the real determinant of the reference of names
of a former historical figure is a chain of communication, in which the re-
ference of the name is passed from link to link. Now the legitimacy of such
a chain accords much more with Millian views than with alternatives. For
the view supposes that a learner acquires a name from the community by
determining to use it with the same reference as does the community. We
regard such a learner as using "Cicero is bald" to express the same thing the
community expresses, regardless of variations in the properties different
learners associate with 'Cicero,' as long as he determines that he will use the
248 SAUL A. KRIPKE
name with the referent current in the community. That a name can be trans-
mitted in this way accords nicely with a Millian picture, according to which
only the reference, not more specific properties associated with the name, is
relevant to the semantics of sentences containing it. It has been suggested that
the chain of communication, which on the present picture determines the
reference, might thereby itself be called a 'sense.' Perhaps so - if we wish2o -
but we should not thereby forget that the legitimacy of such a chain suggests
that it is just preservation of reference, as Mill thought, that we regard as
necessary for correct language learning. 21 (This contrasts with such terms as
'renate' and' cordate,' where more than learning the correct extension is need-
ed.) Also, as suggested above, the doctrine of rigidity in modal contexts
is dissonant, though not necessarily inconsistent, with a view that invokes
antiMillian considerations to explain propositional attitude contexts.
The spirit of my earlier views, then, suggests that a Millian line should be
maintained as far as is feasible.
II. PRELl MI N ARIES: SOME GENERAL PRIN CI P LES
Where are we now? We seem to be in something of a quandary. On the one

hand, we concluded that the failure of 'Cicero' and 'Tully' to be interchange-
able salva veritate in contexts of propositional attitude was by no means
explicable in terms of different 'senses' of the two names. On the other hand,
let us not forget the initial argument against Mill: If reference is all there is to
naming, what semantic difference can there be between 'Cicero' and 'Tully'?
And if there is no semantic difference, do not 'Cicero was bald' and 'Tully
was bald' express exactly the same proposition? How, then, can anyone
believe that Cicero was bald, yet doubt or disbelieve that Tully was?
Let us take stock. Why do we think that anyone can believe that Cicero
was bald, but fail to believe that Tully was? Or believe, without any logical
inconsistency, that Yale is a fine university, but that Old Eli is an inferior
one? Well, a normal English speaker, Jones, can sincerely assent to 'Cicero
was bald' but not to 'Tully was bald.' And this even though Jones uses
'Cicero' and 'Tully' in standard ways - he uses 'Cicero' in this assertion as
a name for the Roman, not, say, for his dog, or for a German spy.
Let us make explicit the disquotational principle presupposed here, con-
necting sincere assent and belief. It can be stated as follows, where 'p' is to
be replaced, inside and outside all quotation marks, by any appropriate
standard English sentence: "If a normal English speaker, on reflection,
sincerely assents to 'p,' then he believes that p." The sentence replacing 'p' is
to lack indexical or pronominal devices or ambiguities, that would ruin the
intuitive sense of the principle (e.g., ifhe assents to "You are wonderful," he
need not believe that you - the reader - are wonderful).22 When we suppose
that we are dealing with a normal speaker of English, we mean that he uses
all words in the sentence in a st~ndard way, combines them according to the
appropriate syntax, etc.: in short, he uses the sentence to mean what a
normal speaker should mean by it. The 'words' of the sentence may include
proper names, where these are part of the common discourse of the com-
munity, so that we can speak of using them in a standard way. For example,
if the sentence is "London is pretty," then the speaker should satisfy normal
criteria for using 'London' as a name of London, and for using 'is pretty' to
attribute an appropriate degree of pulchritude. The qualification "on reflec-
tion" guards against the possibility that a speaker may, through careless
inattention to the meaning of his words or other momentary conceptual or
linguistic confusion, assert something he does not really mean, or assent to a
sentence in linguistic error. "Sincerely" is meant to exclude mendacity, act-
ing, irony, and the like. I fear that even with all this it is possible that some
astute reader - such, after all, is the way of philosophy - may discover a
qualification I have overlooked, without which the asserted principle is
subject to counterexample. I doubt, however, that any such modification will
affect any of the uses of the principle to be considered below. Taken in its
obvious intent, after all, the principle appears to be a self-evident truth. (A
similar principle holds for sincere affirmation or assertion in place of assent.)
There is also a strengthened 'biconditional' form of the disquotational
principle, where once again any appropriate English sentence may replace
'p' throughout: A normal English speaker who is not reticent will be disposed
to sincere reflective assent to 'p' if and only if he believes that p.23 The bicondi-
tional form strengthens the simple one by adding that failure to assent
indicates lack of belief, as assent indicates belief. The qualification about
reticence is meant to take account of the fact that a speaker may fail to avow
his beliefs because of shyness, a desire for secrecy, to avoid offense, etc. (An
alternative formulation would give the speaker a sign to indicate lack of
belief - not necessarily disbelief - in the assertion propounded, in addition
to his sign of assent.) Maybe again the formulation needs further tightening,
but the intent is clear.
Usually below the simple disquotational principle will be sufficient for our
purposes, but once we will also invoke the strengthened form. The simple
250 SAUL A. KRIPKE
form can often be used as a test for disbelief, provided the subject is a speaker
with the modicum of 10 gicality needed so that, at least after appropriate
reflection, he does not hold simultaneously beliefs that are straightforward
contradictions of each other - of the forms 'p' and' '" p.'24 (Nothing in such
a requirement prevents him from holding simultaneous beliefs that jointly
entail a contradiction.) In this case (where 'p' may be replaced by any ap-
propriate English sentence), the speaker's assent to the negation of 'p'
indicates not only his disbelief that p but also his failure to believe that p,
using only the simple (unstrengthened) disquotational principle.
So far our principle applies only to speakers of English. It allows us to
infer, from Peter's sincere reflective assent to "God exists," that he believes
that God exists. But of course we ordinarily allow ourselves to draw con-
clusions, stated in English, about the beliefs of speakers of any language: we
infer that Pierre believes that God exists from his sincere reflective assent to
"Dieu existe." There are several ways to do this, given conventional transla-
tions of French into English. We choose the following route. We have stated
the disquotational principle in English, for English sentences; an analogous
principle, stated in French (German, etc.) will be assumed to hold for French
(German, etc.) sentences. Finally, we assume the principle of translation: If a
sentence ofone language expresses a truth in that language, then any translation
of it into any other language also expresses a truth (in that other language).
Some of our ordinary practice of translation may violate this principle; this
happens when the translator's aim is not to preserve the content of the sen-
tence, but to serve - in some other sense - the same purposes in the home
language as the original utterance served in the foreign language. 2s But if
the translation of a sentence is to mean the same as the sentence translated,
preservation of truth value is a minimal condition that must be observed.
Granted the disquotational principle expressed in each language, reasoning
starting from Pierre's assent to 'Dieu existe' continues thus. First, on the
basis of his utterance and the French disquotational principle we infer (in
French):
Pierre croit que Dieu existe.
From this we deduce,26 using the principle of translation:
Pierre believes that God exists.
In this way we can apply the dis quotational technique to all languages.
Even if I apply the disquotational technique to English alone, there is a

sense in which I can be regarded as tacitly invoking a principle of translation.
For presumably I apply it to speakers of the language other than myself. As
Quine has pointed out, to regard others as speaking the same language as I
is in a sense tacitly to assume a homophonic translation of their language into
my own. So when I infer from Peter's sincere assent to or affirmation of "God
exists" that he believes that God exists, it is arguable that, strictly speaking,
I combine the disquotational principle (for Peter's idiolect) with the principle
of (homophonic) translation (of Peter's idiolect into mine). But for most
purposes, we can formulate the disquotational principle for a single language,
English, tacitly supposed to be the common language of English speakers.
Only when the possibility of individual differences of dialect is relevant need
we view the matter more elaborately.
Let us return from these abstractions to our main theme. Since a normal
speaker - normal even in his use of 'Cicero' and 'Tully' as names - can
give sincere and reflective assent to "Cicero was bald" and simultaneously to
"Tully was not bald," the disquotational principle implies that he believes
that Cicero was bald and believes that Tully was not bald. Since it seems that
he need not have contradictory beliefs (even if he is a brilliant logician, he
need not be able to deduce that at least one of his beliefs must be in error),
and since a substitutivity principle for coreferential proper names in belief
contexts would imply that he does have contradictory beliefs, it would seem
that such a substitutivity principle must be incorrect. Indeed, the argument
appears to be a reductio ad absurdum of the substitutivity principle in
question.
The relation of this argument against substitutivity to the classical position
of Russell and Frege is a curious one. As we have seen, the argument can be
used to give primafacie support for the Frege-Russell view, and I think many
philosophers have regarded it as such support. But in fact this very argument,
which has been used to support Frege and Russell, cannot be stated in a
straightforward fashion if Frege and Russell are right. For suppose Jones
asserts, "Cicero was bald, but Tully was not." If Frege and Russell are right,
I cannot deduce, using the disquotational principle:
(1) Jones believes that Cicero was bald but Tully was not,
since, in general, Jones and I will not, strictly speaking, share a common
idiolect unless we assign the same 'senses' to all names. Nor can I combine
disquotation and translation to the appropriate effect, since homophonic
252 SAUL A. KRIPKE
translation of Jones's sentence into mine will in general be incorrect for the
same reason. Since in fact I make no special distinction in sense between
'Cicero' and 'Tully' - to me, and probably to you as well, these are inter-
changeable names for the same man - and since according to Frege and
Russell, Jones's very affirmation of (1) shows that for him there is some dis-
tinction of sense, Jones must therefore, on Frege-Russellian views, use one
of these names differently from me, and homophonic translation is illegi-
timate. Hence, if Frege and Russell are right, we cannot use this example in
the usual straightforward way to conclude that proper names are not sub-
stitutable in belief contexts - even though the example, and the ensuing
negative verdict on substitutivity, has often been thought to support Frege
and Russell!
Even according to the Frege-Russellian view, however, Jones can conclude,
using the disquotational principle, and expressing his conclusion in his own
idiolect:
(2) I believe that Cicero was bald but Tully was not.
I cannot endorse this conclusion in Jones's own words, since I do not share
Jones's idiolect. I can of course conclude, "(2) expresses a truth in Jones's
idiolect." I can also, if! find out the two 'senses' Jones assigns to 'Cicero' and
'Tully,' introduce two names 'X' and 'Y' into my own language with these
same two senses ('Cicero' and 'Tully' have already been preempted) and
conclude:
(3) Jones believes that X was bald and Y was not.
All this is enough so that we can still conclude, on the Frege-Russellian view,
that codesignative names are not interchangeable in belief contexts. Indeed
this can be shown more simply on this view, since codesignative descriptions
plainly are not interchangeable in these contexts and for Frege and Russell
names, being essentially abbreviated descriptions, cannot differ in this respect.
Nevertheless, the simple argument, apparently free of such special Frege-
Russellian doctrinal premises (and often used to support these premises), in
fact cannot go through if Frege and Russell are right.
However, if, pace Frege and Russell, widely used names are common
currency of our language, then there no longer is any problem for the simple
argument, using the disquotational principle, to (2). So, it appears, on pain of
convicting Jones of inconsistent beliefs - surely an unjust verdict - we must
not hold a substitutivity principle for names in belief contexts. If we used the
strengthened disquotational principle, we could invoke Jones's presumed lack
of any tendency to assent to 'Tully was bald' to conclude that he does not
believe (lacks the belief) that Tully was bald. Now the refutation of the sub-
stitutivity principle is even stronger, for when applied to the conclusion that
Jones believes that Cicero was bald but does not believe that Tully was bald,
it would lead to a straightout contradiction. The contradiction would no
longer be in Jones's beliefs but in our own.
This reasoning, I think, has been widely accepted as proof that codesigna-
tive proper names are not interchangeable in belief contexts. Usually the
reasoning is left tacit, and it may well be thought that I have made heavy
weather of an obvious conclusion. I wish, however, to question the reason-
ing. I shall do so without challenging any particular step of the argument.
Rather I shall present - and this will form the core of the present paper -
an argument for a paradox about names in belief contexts that invokes no
principle of substitutivity. Instead it will be based on the principles - ap-
parently so obvious that their use in these arguments is ordinarily tacit - of
disquotation and translation.
Usually the argument will involve more than one language, so that the
principle of translation and our conventional manual of translation must be
invoked. We will also give an example, however, to show that a form of the
paradox may result within English alone, so that the only principle invoked
is that of disquotation (or, perhaps, disquotation plus homophonic trans-
lation). It will intuitively be fairly clear, in these cases, that the situation of
the subject is 'essentially the same' as that of Jones with respect to 'Cicero'
and 'Tully.' Moreover, the paradoxical conclusions about the subject will
parallel those drawn about Jones on the basis of the substitutivity principle,
and the arguments will parallel those regarding Jones. Only in these cases, no
special substitutivity principle is invoked.
The usual use of Jones's case as a counterexample to the substitutivity
principle is thus, I think, somewhat analogous to the following sort of pro-
cedure. Someone wishes to give a reductio ad absurdum argument against a
hypothesis in topology. He does succeed in refuting this hypothesis, but his
derivation of an absurdity from the hypothesis makes essential use of the
unrestricted comprehension schema in set theory, which he regards as self-
evident. (In particular, the class of all classes not members ofthemselves plays
a key role in his argument.) Once we know that the unrestricted comprehen-
sion schema and the Russell class lead to contradiction by themselves, it is
254 SAUL A. KRIPKE
clear that it was an error to blame the earlier contradiction on the topological
hypothesis.
The situation would have been the same if, after deducing a contradiction
from the topological hypothesis plus the 'obvious' unrestricted comprehen-
sion schema, it was found that a similar contradiction followed if we replaced
.the topological hypothesis by an apparently 'obvious' premise. In both cases
it would be clear that, even though we may still not be confident of any
specific flaw in the argument against the topological hypothesis, blaming the
contradiction on that hypothesis is illegitimate: rather we are in a 'para-
doxical' area where it is unclear what has gone wrong,27
It is my suggestion, then, that the situation with respect to the inter-
changeability of codesignative names is similar. True, such a principle, when
combined with our normal disquotational judgments of belief, leads to
straightforward absurdities. But we will see that the 'same' absurdities can be
derived by replacingthe interchangeability principle by our normal practices
of translation and dis quotation, or even by disquotation alone.
The particular principle stated here gives just one particular way of
'formalizing' our normal inferences from explicit affirmation or assent to
belief; other ways of doing it are possible. It is undeniable that we do infer,
from a normal Englishman's sincere affirmation of 'God exists' or 'London
is pretty,' that he believes, respectively, that God exists or that London is
pretty; and that we would make the same inferences from a Frenchman's
affirmation of 'Dieu existe' or 'Londres est jolie.' Any principles that would
justify such inferences are sufficient for the next section. It will be clear that
the particular principles stated in the present section are sufficient, l:>ut in the
next section the problem will be presented informally in terms of our in-
ferences from foreign or domestic assertion to belief.
III. THE PUZZLE
Here, finally(!), is the puzzle. Suppose Pierre is a normal French speaker

who lives in France and speaks not a word of English or of any other langu-
age except French. Of course he has heard of that famous distant city,
London (which he of course calls 'Londres') though he himself has never left
France. On the basis of what he has heard of London, he is inclined to think
that it is pretty. So he says, in French, "Londres est jolie."
On the basis of his sincere French utterance, we will conclude:
(4) Pierre believes that London is pretty.

I am supposing that Pierre satisfies all criteria for being a normal French
speaker, in particular, that he satisfies whatever criteria we usually use to
judge that a Frenchman (correctly) uses' est jolie' to attribute pulchritude and
uses 'Londres' - standardly - as a name of London.
Later, Pierre, through fortunate or unfortunate vicissitudes, moves to
England, in fact to London itself, though to an unattractive part of the city
with fairly uneducated inhabitants. He, like most of his neighbors, rarely
ever leaves this part of the city. None of his neighbors know any French, so
he must learn English by 'direct method,' without using any translation of
English into French: by talking and mixing with the people he eventually
begins to pick up English. In particular, everyone speaks of the city, 'London,'
where they all live. Let us suppose for the moment - though we will see
below that this is not crucial - that the local population are so uneducated
that they know few of the facts that Pierre heard about London in France.
Pierre learns from them everything they know about London, but there is
little overlap with what he heard before. He learns, of course - speaking
English - to call the city he lives in 'London.' Pierre's surroundings are, as I
said, unattractive, and he is unimpressed with most of the rest of what he
happens to see. So he is inclined to assent to the English sentence:
(5) London is not pretty.
He has no inclination to assent to:
(6) London is pretty.
Of course he does not for a moment withdraw his assent from the French
sentence, "Londres est jo lie " ; he merely takes it for granted that the ugly city
in which he is now stuck is distinct from the enchanting city he heard about
in France. But he has no inclination to change his mind for a moment about
the city he stills calls 'Londres.'
This, then, is the puzzle. If we consider Pierre's past background as a
French speaker, his entire linguistic behavior, on the same basis as we would
draw such a conclusion about many of his countrymen, supports the con-
clusion ( (4) above) that he believes that London is pretty. On the other hand,
after Pierre lived in London for some time, he did not differ from his neigh-
bors - his French background aside - either in his knowledge of English
or in his command of the relevant facts of local geography. His English
256 SAUL A. KRIPKE
vocabulary differs little from that of his neighbors. He, like them, rarely
ventures from the dismal quarter of the city in which they all live. He, like
them, knows that the city he lives in is called 'London' and knows a few other
facts. Now Pierre's neighbors would surely be said to use 'London' as a
name for London and to speak English. Since, as an English speaker, he does
not differ at all from them, we should say the same of him. But then, on the
basis of his sincere assent to (5), we should conclude:
(7) Pierre believes that London is not pretty.
How can we describe this situation? It seems undeniable that Pierre once
believed that London is pretty - at least before he learned English. For at
that time, he differed not at all from countless numbers of his countrymen,
and we would have exactly the same grounds to say of him as of any of them
that he believes that London is pretty: if any Frenchman who was both
ignorant of English and never visited London believed that London is pretty,
Pierre did. Nor does it have any plausibility to suppose, because of his later
situation after he learns English, that Pierre should retroactively be judged
never to have believed that London is pretty. To allow such ex post facto
legislation would, as long as the future is uncertain, endanger our attribu-
tions of belief to all monolingual Frenchmen. We would be forced to say that
Marie, a monolingual who firmly and sincerely asserts, "Londres est jolie,"
mayor may not believe that London is pretty depending on the later vicissi-
tudes of her career (if later she learns English and ... , ...). No: Pierre, like
Marie, believed that London is pretty when he was monolingual.
Should we say that Pierre, now that he lives in London and speaks English,
no longer believes that London is pretty? Well, unquestionably Pierre once
believed that London is pretty. So we would be forced to say that Pierre has
changed his mind, has given up his previous belief But has he really done so?
Pierre is very set in his ways. He reiterates, with vigor, every assertion he has
ever made in French. He says he has not changed his mind about anything,
has not given up any belief. Can we say he is wrong about this? If we did not
have the story of his living in London and his English utterances, on the basis
of his normal command of French we would be forced to conclude that he
still believes that London is pretty. And it does seem that this is correct.
Pierre has neither changed his mind nor given up any belief he had in
France.
Similar difficulties beset any attempt to deny him his new belief. His
French past aside, he is just like his friends in London. Anyone else, growing
up in London with the same knowledge and beliefs that he expresses in

England, we would undoubtedly judge to believe that London is not pretty.
Can Pierre's French past nullify such a judgment? Can we say that Pierre,
because of his French past, does not believe that (5)? Suppose an electric
shock wiped out all his memories of the French language, what he learned
in France, and his French past. He would then be exactly like his neighbors
in London. He would have the same knowledge, beliefs, and linguistic capa-
cities. We then presumably would be forced to say that Pierre believes that
London is ugly if we say it of his neighbors. But surely no shock that destroys
part of Pierre's memories and knowledge can give him a new belief. If Pierre
believes (5) after the shock, he believed it before, despite his French language
and background.
If we would deny Pierre, in his bilingual stage, his belief that London is
pretty and his belief that London is not pretty, we combine the difficulties of
both previous options. We still would be forced to judge that Pierre once
believed that London is pretty but does no longer, in spite of Pierre's own
sincere denial that he has lost any belief. We also must worry whether Pierre
would gain the belief that London is not pretty if he totally forgot his French
past. The option does not seem very satisfactory.
So now it seems that we must respect both Pierre's French utterances and
their English counterparts. So we must say that Pierre has contradictory
beliefs, that he believes that London is pretty and he believes that London is
not pretty. But there seem to be insuperable difficulties with this alternative
as well. We may suppose that Pierre, in spite of the unfortunate situation in
which he now finds himself, is a leading philosopher and logician. He would
never let contradictory beliefs pass. And surely anyone, leading logician or
no, is in principle in a position to notice and correct contradictory beliefs if
he has them. Precisely for this reason, we regard individuals who contradict
themselves as subject to greater censure than those who merely have false
beliefs. But it is clear that Pierre, as long as he is unaware that the cities he
calls 'London' and 'Londres' are one and the same, is in no position to see,
by logic alone, that at least one of his beliefs must be false. He lacks informa-
tion, not logical acumen. He cannot be convicted of inconsistency: to do so
is incorrect.
We can shed more light on this if we change the case. Suppose that, in
France, Pierre, instead of affirming "Londres est jolie," had affirmed, more
cautiously, "Si New York est jolie, Londres est jolie aussi," so that he believed
that if New York is pretty, so is London. Later Pierre moves to London,
258 SAUL A. KRIPKE
learns English as before, and says (in English) "London is not pretty." So he
now believes, further, that London is not pretty. Now from the two premises,
both of which appears to be among his beliefs (a) If New York is pretty,
London is, and (b) London is not pretty, Pierre should be able to deduce by
modus tollens that New York is not pretty. But no matter how great Pierre's
logical acumen may be, he cannot in fact make any such deduction, as long as
he supposes that 'Londres' and 'London' may name two different cities. If he
did draw such a conclusion, he would be guilty of a fallacy.
Intuitively, he may well suspect that New York is pretty, and just this
suspicion may lead him to suppose that 'Londres' and 'London' probably
name distinct cities. Yet, if we follow our normal practice of reporting the
beliefs of French and English speakers, Pierre has available to him (among his
beliefs) both the premises of a modus tollens argument that New York is not
pretty.
Again, we may emphasize Pierre's lack of belief instead of his belief. Pierre,
as I said, has no disposition to assent to (6). Let us concentrate on this,
ignoring his disposition to assent to (5). In fact, if we wish we may change
the case: Suppose Pierre's neighbors think that since they rarely venture
outside their own ugly section, they have no right to any opinion as to the
pulchritude of the whole city. Suppose Pierre shares their attitude. Then,
judging by his failure to respond affirmatively to "London is pretty," we may
judge, from Pierre's behavior as an English speaker, that he lacks the belief
that London is pretty: never mind whether he disbelieves it, as before, or
whether, as in the modified story, he insists that he has no firm opinion on
the matter.
Now (using the strengthened disquotational principle), we can derive a
contradiction, not merely in Pierre's judgments, but in our own. For on the
basis of his behavior as an English speaker, we concluded that he does not
believe that London is pretty (that is, that it is not the case that he believes
that London is pretty). But on the basis of his behavior as a French speaker,
we must conclude that he does believe that London is pretty. This is a
contradiction. 28
We have examined four possibilities for characterizing Pierre while he is
in London: (a) that at that time we no longer respect his French utterance
('Londres est jolie'), that is that we no longer ascribe to him the corresponding
belief; (b) that we do not respect his English utterance (or lack of utterance);
(c) that we respect neither; (d) that we respect both. Each possibility seems
to lead us to say something either plainly false or even downright contradic-
tory. Yet the possibilities appear to be logically exhaustive. This, then, is the
paradox.
I have no firm belief as to how to solve it. But beware of one source of
confusion. It is no solution in itself to observe that some other terminology,
which evades the question whether Pierre believes that London is pretty, may
be sufficient to state all the relevant facts. I am fully aware that complete and
straightforward descriptions of the situation are possible and that in this
sense there is no paradox. Pierre is disposed to sincere assent to 'Londres est
jolie' but not to 'London is pretty.' He uses French normally, English nor-
mally. Both with 'Londres' and 'London' he associates properties sufficient
to determine that famous city, but he does not realize that they determine a
single city. (And his uses of'Londres' and 'London' are historically (causally)
connected with the same single city, though he is unaware of that.) We may
even give a rough statement of his beliefs. He believes that the city he calls
'Londres' is pretty, that the city he calls 'London' is not. No doubt other
straightforward descriptions are possible. No doubt some of these are, in a
certain sense, complete descriptions of the situation.
But none of this answers the original question. Does Pierre, or does he not,
believe that London is pretty? I know of no answer to this question that seems
satisfactory. It is no answer to protest that, in some other terminology, one
can state 'all the relevant facts.'
To reiterate, this is the puzzle: Does Pierre, or does he not, believe that
London is pretty? It is clear that our normal criteria for the attribution of
belief lead, when applied to this question, to paradoxes and contradictions.
One set of principles adequate to many ordinary attributions of belief, but
which leads to paradox in the present case, was stated in Section 2; and other
formulations are possible. As in the case of the logical paradoxes, the present
puzzle presents us with a problem for customarily accepted principles and a
challenge to formulate an acceptable set of principles that does not lead to
paradox, is intuitively sound, and supports the inferences we usually make.
Such a challenge cannot be met simply by a description of Pierre's situation
that evades the question whether he believes that London is pretty.
One aspect of the presentation may misleadingly suggest the applicability
of Frege-Russellian ideas that each speaker associates his own description or
properties to each name. For as I just set up the case Pierre learned one set
offacts about the so-called 'Londres' when he was in France, and another set
of facts about 'London' in England. Thus it may appear that 'what's really
going on' is that Pierre believes that the city satisfying one set of properties is
260 SAUL A. KRIPKE
pretty, while he believes that the city satisfying another set of properties is not
pretty.
As we just emphasized, the phrase 'what's really going on' is a danger
signal in discussions of the present paradox. The conditions stated may -
let us concede for the moment - describe 'what's really going on.' But they
do not resolve the problem with which we began, that of the behavior of
names in belief contexts: Does Pierre, or does he not, believe that London
(not the city satisfying such-and-such descriptions, but London) is pretty? No
answer has yet been given.
Nevertheless, these considerations may appear to indicate that descrip-
tions, or associated properties, are highly relevant somehow to an ultimate
solution, since at this stage it appears that the entire puzzle arises from the
fact that Pierre originally associated different identifying properties with
'London' and 'Londres.' Such a reaction may have some force even in the
face ofthe now fairly well-known arguments against 'identifying descriptions'
as in any way 'defining,' or even 'fixing the reference' of names. But in fact
the special features of the case, as I set it out, are misleading. The puzzle can
arise even if Pierre associates exactly the same identifying properties with
both names.
First, the considerations mentioned above in connection with 'Cicero' and
'Tully' establish this fact. For example, Pierre may well learn, in France,
'Platon' as the name of a major Greek philosopher, and later, in England,
learns 'Plato' with the same identification. Then the same puzzle can arise:
Pierre may have believed, when he was in France and was monolingual in
French, that Plato was bald (he would have said, "Platon etait chauve"), and
later conjecture, in English, "Plato was not bald," thus indicating that he
believes or suspects that Plato was not bald. He need only suppose that, in
spite of the similarity of their names, the man he caIls 'Platon' and the man
he calls 'Plato' were two distinct major Greek philosophers. In principle, the
same thing could happen with 'London' and 'Londres.'
Of course, most of us learn a definite description about London, say 'the
largest city in England.' Can the puzzle still arise? It is noteworthy that the
puzzle can still arise even if Pierre associates to 'Londres' and to 'London'
exactly the same uniquely identifying properties. How can this be? Well,
suppose that Pierre believes that London is the largest city in (and capital of)
England, that it contains Buckingham Palace, the residence of the Queen of
England, and he believes (correctly) that these properties, conjointly, uni-
quely identify the city. (In this case, it is best to suppose that he has never
seen London, or even England, so that he uses only these properties to

identify the city. Nevertheless, he has learned English by 'direct method.')
These uniquely identifying properties he comes to associate with 'London'
after he learned English, and he expresses the appropriate beliefs about
'London' in English. Earlier, when he spoke nothing but French, however,
he associated exactly the same uniquely identifying properties with 'Londres.'
He believed that 'Londres,' as he called it, could be uniquely identified as the
capital of England, that it contained Buckingham Palace, that the Queen of
England lived there, etc. Of course he expressed these beliefs, like most mono-
lingual Frenchmen, in French. In particular, he used 'Angleterre' for England,
'Ie Palais de Buckingham' (pronounced 'Bookeengam'!) for Buckingham
Palace, and 'la Reine d'Angleterre' for the Queen of England. But if any
Frenchman who speaks no English can ever be said to associate exactly the
properties of being the capital of England etc., with the name 'Londres,'
Pierre in his monolingual period did so.
When Pierre becomes a bilingual, must he conclude that 'London' and
'Londres' name the same city, because he defined each by the same uniquely
identifying properties?
Surprisingly, no! Suppose Pierre had affirmed, 'Londres est jolie.' If Pierre
has any reason - even just a 'feeling in his bones,' or perhaps exposure to a
photograph of a miserable area which he was told (in English) was part of
'London' - to maintain 'London is not pretty,' he need not contradict him-
self. He need only conclude that 'England' and 'Angle terre' name two
different countries, that 'Buckingham Palace' and 'Ie Palais de Buckingham'
(recall the pronunciation!), name two different palaces, and so on. Then he
can maintain both views without contradiction, and regard both properties as
uniquely identifying.
The fact is that the paradox reproduces itself on the level of the 'uniquely
identifying properties' that description theorists have regarded as 'defining'
proper names (and a fortiori, as fixing their references). Nothing is more
reasonable than to suppose that if two names, A and B, and a single set of
properties, S, are such that a certain speaker believes that the referent of A
uniquely satisfies all of S and that the referent of B also uniquely satisfies all
of S, then that speaker is committed to the belief that A and B have the same
reference. In fact, the identity of the referents of A and B is an easy logical
consequence of the speaker's beliefs.
From this fact description theorists concluded that names can be regarded
as synonymous, and hence interchangeable salva veritate even in belief con-
262 SAUL A. KRIPKE
texts, provided that they are 'defined' by the same uniquely identifying pro-
perties.
We have already seen that there is a difficulty in that the set S of pro-
perties need not in fact be uniquely identifying. But in the present para-
doxical situation there is a surprising difficulty even if the supposition of the
description theorist (that the speaker believes that S is uniquely fulfilled) in
fact holds. For, as we have seen above, Pierre is in no position to draw
ordinary logical consequences from the conjoint set of what, when we con-
sider him separately as a speaker of English and as a speaker of French, we
would call his beliefs. He cannot infer a contradiction from his separate
beliefs that London is pretty and that London is not pretty. Nor, in the modi-
fied situation above, would Pierre make a normal modus tollens inference
from his beliefs that London is not pretty and that London is pretty if New
York is. Similarly here, if we pay attention only to Pierre's behavior as a
French speaker (and at least in his monolingual days he was no different from
any other Frenchmen), Pierre satisfies all the normal criteria for believing
that 'Londres' has a referent uniquely satisfying the properties of being the
largest city in England, containing Buckingham Palace, and the like. (If Pierre
did not hold such beliefs, no Frenchman ever did.) Similarly, on the basis of
his (later) beliefs expressed in English, Pierre also believes that the referent of
'London' uniquely satisfies these same properties. But Pierre cannot combine
the two beliefs into a single set of beliefs from which he can draw the normal
conclusion that 'London' and 'Londres' must have the same referent. (Here
the trouble comes not from 'London' and 'Londres' but from 'England' and
'Angleterre' and the rest.) Indeed, ifhe did draw what would appear to be the
normal conclusion in this case and any of the other cases, Pierre would in
fact be guilty of a logical fallacy.
Of course the description theorist could hope to eliminate the problem by
'defining' 'Angleterre,' 'England,' and so on by appropriate descriptions also.
Since in principle the problem may rear its head at the next 'level' and at each
subsequent level, the description theorist would have to believe that an
'ultimate' level can eventually be reached where the defining properties are
'pure' properties not involving proper names (nor natural kind terms or
related terms, see below!). I know of no convincing reason to suppose that
such a level can be reached in any plausible way, or that the properties can
continue to be uniquely identifying if one attempts to eliminate all names and
related devices. 29 Such speculation aside, the fact remains that Pierre, judged
by the ordinary criteria for such judgments, did learn both 'Londres' and
'London' by exactly the same set of identifying properties; yet the puzzle
remains even in this case.
Well, then, is there any way out of the puzzle? Aside from the principles of
disquotation and translation, only our normal practice of translation of
French into English has been used. Since the principles of disquotation and
translation seem self-evident, we maybe tempted to blame the trouble on the
translation of'Londres est jolie' as 'London is pretty,' and ultimately, then,
on the translation of 'Londres' as 'London.'3o Should we, perhaps, permit
ourselves to conclude that 'Londres' should not, 'strictly speaking' be trans-
lated as 'London'? Such an expedient is, of course, desperate: the translation
in question is a standard one, learned by students together with other
standard translations of French into English. Indeed, 'Londres' is, in effect,
introduced into French as the French version of 'London.'
Since our backs, however, are against the wall, let us consider this desperate
and implausible expedient a bit further. If'Londres' is not a correct French
version of the English 'London,' under what circumstances can proper names
be translated from one language to another?
Classical description theories suggest the answer: Translation, strictly
speaking, is between idiolects; a name in one idiolect can be translated into
another when (and only when) the speakers of the two idiolects associate the
same uniquely identifying properties with the two names. We have seen that
any such proposed restriction, not only fails blatantly to fit our normal prac-
tices of translation and indirect discourse reportage, but does not even
appear to block the paradox.31
So we still want a suitable restriction. Let us drop the references to idiolects
and return to 'Londres' and 'London' as names in French and English, res-
pectively - the languages of two communities. If 'Londres' is not a correct
French translation of 'London,' could any other version do better? Suppose
I introduced another word into French, with the stipulation that it should
always be used to translate 'London.' Would not the same problem arise
for this word as well? The only feasible solution in this direction is the most
drastic: decree that no sentence containing a name can be translated except
by a sentence containing the phonetically identical name. Thus when Pierre
asserts 'Londres est jolie,' we English speakers can at best conclude, if any-
thing: Pierre believes that Londres is pretty. Such a conclusion is, of course,
not expressed in English, but in a word salad of English and French; on
the view now being entertained, we cannot state Pierre's belief in English at
all. 32 Similarly, we would have to say: Pierre believes that Angleterre is a
264 SAUL A. KRIPKE
monarchy, Pierre believes that Platon wrote dialogues, and the like. 33
This 'solution' appears at first to be effective against the paradox, but it is
drastic. What is it about sentences containing names that makes them - a
substantial class - intrinsically untranslatable, express beliefs that cannot
be reported in any other language? At best, to report them in the other
language, one is forced to use a word salad in which names from the one
language are imported into the other. Such a supposition is both contrary to
our normal practice of translation and very implausible on its face.
Implausible though it is, there is at least this much excuse for the 'solution'
at this point. Our normal practice with respect to some famous people and
especially for geographical localities is to have different names for them in
different languages, so that in translating sentences we translate the names.
But for a large number of names, especially names of people, this is not so:
the person's name is used in the sentences of all languages. At least the restric-
tion in question merely urges us to mend our ways by doing always what we
presently do sometimes.
But the really drastic character of the proposed restriction comes out when
we see how far it may have to extend. In "Naming and Necessity" I suggested
that there are important analogies between proper names and natural kind
terms, and it seems to me that the present puzzle is one instance where the
analogy will hold. Putnam, who has proposed views on natural kinds similar
to my own in many respects, stressed this extension of the puzzle in his com-
ments at the Conference. Not that the puzzle extends to all translations from
English to French. At the moment, at least, it seems to me that Pierre, if he
learns English and French separately, without learning any translation
manual between them, must conclude, ifhe reflects enough, that 'doctor' and
'medecin,' and 'heureux' and 'happy,' are synonymous, or at any rate,
coextensive;34 any potential paradox of the present kind for these word pairs
is thus blocked. But what about 'lapin' and 'rabbit,' or 'beech' and 'hhre'?
We may suppose that Pierre is himself neither a zoologist nor a botanist. He
has learned each language in its own country and the examples he has been
shown to illustrate 'les lapins' and 'rabbits,' 'beeches' and 'les hetres' are
distinct. It thus seems to be possible for him to suppose that 'lapin' and
'rabbit,' or 'beech' and 'Mtre,' denote distinct but superficially similar kinds
or species, even though the differences may be indiscernible to the untrained
eye. (This is especially plausible if, as Putnam supposes, an English speaker -
for example, Putnam himself - who is not a botanist may use 'beech' and
'elm' with their normal (distinct) meanings, even though he cannot himself
distinguish the two trees.3 5 Pierre may quite plausibly be supposed to wonder
whether the trees which in France he called 'les hetres' were beeches or elms,
even though as a speaker of French he satisfies all usual criteria for using
'les hetres' normally. If beeches and elms will not serve, better pairs of
ringers exist that cannot be told apart except by an expert.) Once Pierre is in
such a situation, paradoxes analogous to the one about London obviously
can arise for rabbits and beeches. Pierre could affirm a French statement with
'lapin,' but deny its English translation with 'rabbit.' As above, we are hard-
pressed to say what Pierre believes. We were considering a 'strict and philo-
sophical' reform of translation procedures which proposed that foreign
proper names should always be appropriated rather than translated. Now it
seems that we will be forced to do the same with all words for natural kinds.
(For example, on price of paradox, one must not translate 'lapin' as 'rabbit'!)
No longer can the extended proposal be defended, even weakly, as 'merely'
universalizing what we already do sometimes. It is surely too drastic a change
to retain any credibility.36
There is yet another consideration that makes the proposed restriction
more implausible: Even this restriction does not really block the paradox.
Even if we confine ourselves to a single language, say English, and to phone-
tically identical tokens of a single name, we can still generate the puzzle.
Peter (as we may as well say now) may learn the name 'Paderewski' with an
identification of the person named as a famous pianist. Naturally, having
learned this, Peter will assent to "Paderewski had musical talent," and we
can infer - using 'Paderewski,' as we usually do, to name the Polish
musician and statesman:
(8) Peter believes that Paderewski had musical talent.
Only the disquotational principle is necessary for our inference; no translation
is required. Later, in a different circle, Peter learns of someone called 'Pa-
derewski' who was a 'Polish nationalist leader and Prime Minister. Peter is
skeptical of the musical abilities of politicians. He concludes that probably
two people, approximate contemporaries no doubt, were both named
'Paderewski.' Using 'Paderewski' as a name for the statesman, Peter assents
to, "Paderewski had no musical talent." Should we infer, by the disquota-
tional principle,
(9) Peter believes that Paderewski had no musical talent
or should we not? If Peter had not had the past history of learning the name
266 SAUL A. KRIPKE
'Paderewski' in another way, we certainly would judge him to be using

'Paderewski' in a normal way, with the normal reference, and we would
infer (9) by the disquotational principle. The situation is parallel to the pro-
blem with Pierre and London. Here, however, no restriction that names
should not be translated, but should be phonetically repeated in the transla-
tion, can help us. Only a single language and a single name are involved. If
any notion of translation is involved in this example, it is homophonic
translation. Only the disquotational principle is used explicitly.37 (On the
other hand, the original 'two languages' case had the advantage that it would
apply even if we spoke languages in which all names must denote uniquely
and unambiguously.) The restriction that names must not be translated is
thus ineffective, as well as implausible and drastic.
I close this section with some remarks on the relation of the present puzzle
to Quine's doctrine of the 'indeterminacy of translation,' with its attendant
repudiation of intensional idioms of 'propositional attitude' such as belief
and even indirect quotation. To a sympathizer with these doctrines the pre-
sent puzzle may well seem to be just more grist for a familiar mill. The situa-
tion of the puzzle seems to lead to a breakdown of our normal practices of
attributing belief and even of indirect quotation. No obvious paradox arises
if we describe the same situation in terms of Pierre's sincere assent to various
sentences, together with the conditions under which he has learned the name
in question. Such a description, although it does not yet conform to Quine'S
strict behavioristic standards, fits in well with his view that in some sense
direct quotation is a more 'objective' idiom than the propositional attitudes.
Even those who, like the present writer, do not find Quine's negative attitude
to the attitudes completely attractive must surely acknowledge this.
But although sympathizers with Quine'S view can use the present examples
to support it, the differences between these examples and the considerations
Quine adduces for his own skepticism about belief and translation should not
escape us. Here we make no use of hypothetical exotic systems of translation
differing radically from the usual one, translating 'lapin,' say, as 'rabbit stage'
or 'undetached part ofa rabbit.' The problem arises entirely within our usual
and customary sytem of translation of French into English; in one case, the
puzzle arose even within English alone, using at most 'homophonic' transla-
tion. Nor is the problem that many different interpretations or translations
fit our usual criteria, that, in Davidson's phrase,38 there is more than one
'way of getting it right.' The trouble here is not that many views as to Pierre's
beliefs get it right, but that they all definitely get it wrong. A straightforward
application of the principles of translation and disquotation to all Pierre's

utterances, French and English, yields the result that Pierre holds inconsistent
beliefs, that logic alone should teach him that one of his beliefs is false. In-
tuitively, this is plainly incorrect. If we refuse to apply the principles to his
French utterances at all, we would conclude that Pierre never believed that
London is pretty, even though, before his unpredictable move, he was like
any other monolingual Frenchman. This is absurd. If we refuse to ascribe
the belief in London's pulchritude only after Pierre's move to England, we
get the counterintuitive result that Pierre has changed his mind, and so on.
But we have surveyed the possibilities above: the point was not that they are
'equally good,' but that all are obviously wrong. If the puzzle is to be used as
an argument for a Quinean position, it is an argument of a fundamentally
different kind from those given before. And even Quine, if he wishes to in-
corporate the notion of belief even into a 'second level' of canonical nota-
tion,39 must regard the puzzle as a real problem.
The alleged indeterminacy of translation and indirect quotation causes
relatively little trouble for such a scheme for belief; the embarrassment it
presents to such a scheme is, after all, one of riches. But the present puzzle
indicates that the usual principles we use to ascribe beliefs are apt, in certain
cases, to lead to contradiction, or at least, patent falsehoods. So it presents a
problem for any project, Quinean or other, that wishes to deal with the 'logic'
of belief on any level. 40
IV. CONCLUSION
What morals can be drawn? The primary moral - quite independent of any
of the discussion of the first two sections - is that the puzzle is a puzzle. As
any theory of truth must deal with the Liar Paradox, so any theory of belief
and names must deal with this puzzle.
But our theoretical starting point in the first two sections concerned proper
names and belief. Let us return to Jones, who assents to "Cicero was bald"
and to "Tully was not bald." Philosophers, using the disquotational principle,
have concluded that Jones believes that Cicero was bald but that Tully was
not. Hence, they have concluded, since Jones does not have contradictory
beliefs, belief contexts are not 'Shakespearean' in Geach's sense: co designative
proper names are not interchangeable in these contexts salva veritate. 41
I think the puzzle about Pierre shows that the simple conclusion was un-
warranted. Jones' situation strikingly resembles Pierre's. A proposal that
268 SAUL A. KRIPKE
'Cicero' and 'Tully' are interchangeable amounts roughly to a homophonic

'translation' of English into itself in which 'Cicero' is mapped into 'Tully'
and vice versa, while the rest is left fixed. Such a 'translation' can, indeed, be
. used to obtain a paradox. But should the problem be blamed on this step?
Ordinarily we would suppose without question that sentences in French with
'Londres' should be translated into English with 'London.' Yet the same
paradox results when we apply this translation too. We have seen that the
problem can even arise with a single name in a single language, and that it
arises with natural kind terms in two languages (or one: see below).
Intuitively, Jones' assent to both 'Cicero was bald' and 'Tully was not
bald' arises from sources of just the same kind as Pierre's assent to both
'Londres est jolie' and 'London is not pretty.'
It is wrong to blame unpalatable conclusions about Jones on substitutivity.
The reason does not lie in any specific fallacy in the argument but rather in
the nature of the realm being entered. Jones's case is just like Pierre's: both
are in an area where our normal practices of attributing belief, based on
the principles of dis quotation and translation or on similar principles,
are questionable.
It should be noted in this connection that the principles of disquotation and
translation can lead to 'proofs' as well as 'disproofs' of substitutivity in belief
contexts. In Hebrew there are two names for Germany, transliteratable
roughly as 'Ashkenaz' and 'Germaniah' - the first of these may be somewhat
archaic. When Hebrew sentences are translated into English, both become
'Germany.' Plainly a normal Hebrew speaker analogous to Jones might
assent to a Hebrew sentence involving 'Ashkenaz' while dissenting from its
counterpart with 'Germaniah.' So far there is an argument against substitu-
tivity. But there is also an argument for substitutivity, based on the principle
of translation. Translate a Hebrew sentence involving 'Ashkenaz' into
English, so that 'Ashkenaz' goes into 'Germany.' Then retranslate the result
into Hebrew, this time translating 'Germany' as 'Germaniah.' By the principle
of translation, both translations preserve truth value. So: the truth value of
any sentence of Hebrew involving 'Ashkenaz' remains the same when
'Ashkenaz' is replaced by 'Germaniah' - a 'proof' of substitutivity! A similar
'proof' can be provided wherever there are two names in one language, and a
normal practice of translating both indifferently into a single name of another
language. 42 (If we combine the 'proof' and 'disproof' of substitutivity in this
paragraph, we could get yet another paradox analogous to Pierre's: our
Hebrew speaker both believes, and disbelieves, that Germany is pretty. Yet
no amount of pure logic or semantic introspection suffices for him to discover

his error.)
Another consideration, regarding natural kinds: Previously we pointed
out that a bilingual may learn 'lapin' and 'rabbit' normally in each respective
language yet wonder whether they are one species or two, and that this fact
can be used to generate a paradox analogous to Pierre's. Similarly, a speaker
of English alone may learn 'furze' and 'gorse' normally (separately), yet
wonder whether these are the same, or resembling kinds. (What about 'rabbit'
and 'hare'?) It would be easy for such a speaker to assent to an assertion
formulated with 'furze' but withhold assent from the corresponding assertion
involving 'gorse.' The situation is quite analogous to that of Jones with
respect to 'Cicero' and 'Tully.' Yet 'furze' and 'gorse,' and other pairs of
terms for the same natural kind, are normally thought of as synonyms.
The point is not, of course, that codesignative proper names are inter-
changeable in belief contexts salva veritate, or that they are interchangeable
in simple contexts even salva significatione. The point is that the absurdities
that disquotation plus substitutivity would generate are exactly paralleled by
absurdities generated by disquotation plus translation, or even 'disquotation
alone' (or: disquotation plus homophonic translation). Also, though our
naive practice may lead to 'disproofs' of substitutivity in certain cases, it can
also lead to 'proofs' of substitutivity in some of these same cases, as we saw
two paragraphs back. When we enter into the area exemplified by Jones and
Pierre, we enter into an area where our normal practices of interpretation and
attribution of belief are subjected to the greatest possible strain, perhaps to
the point of breakdown. So is the notion of the content of someone's asser-
tion, the proposition it expresses. In the present state of our knowledge, I
think it would be foolish to draw any conclusion, positive or negative, about
substitutivity.43
Of course nothing in these considerations prevents us from observing that
Jones can sincerely assert both "Cicero is bald" and "Tully is not bald," even
though he is a normal speaker of English and uses 'Cicero' and 'Tully' in
normal ways, and with the normal referent. Pierre and the "other paradoxical
cases can be described similarly. (For those interested in one of my own
doctrines, we can still say that there was a time when men were in no epistemic
position to assent to 'Hesperus is Phosphorus' for want of empirical informa-
tion, but it nevertheless expressed a necessary truth.)44 But it is no surprise
that quoted contexts fail to satisfy a substitutivity principle within the quota-
tion marks. And, in our present state of clarity about the problem, we are in
270 SAUL A. KRIPKE
no position to apply a disquotation principle to these cases, nor to judge

when two such sentences do, or do not, express the same 'proposition.'
Nothing in the discussion impugns the conventional judgment that belief
contexts are 'referentially opaque,' if 'referential opacity' is construed so that
failure of coreferential definite descriptions to be interchangeable salva
veritate is sufficient for referential opacity. No doubt Jones can believe that
the number of planets is even, without believing that the square of three is
even, if he is under a misapprehension about the astronomical, but not the
arithmetical facts. The question at hand was whether belief contexts were
'Shakespearean,' not whether they were 'referentially transparent.' (Modal
contexts, in my opinion, are 'Shakespearean' but 'referentially opaque.')4S
Even were we inclined to rule that belief contexts are not Shakespearean,
it would be implausible at present to use the phenomenon to support a Frege-
RusseIIian theory that names have descriptive 'senses' through 'uniquely
identifying properties.' There are the well-known arguments against descrip-
tion theories, independent of the present discussion; there is the implausi-
bility of the view that difference in names is difference in idiolect; and finally,
there are the arguments of the present paper that differences of associated
properties do not explain the problems in any case. Given these considera-
tions, and the cloud our paradox places over the notion of 'content' in this
area, the relation ofsubstitutivity to the dispute between Millian and Fregean
conclusions is not very clear.
We repeat our conclusions: Philosophers have often, basing themselves on
Jones' and similar cases, supposed that it goes virtually without saying that
belief contexts are not 'Shakespearean.' I think that, at present, such a
definite conclusion is unwarranted. Rather Jones' case, like Pierre's, lies in
an area where our normal apparatus for the ascription of belief is placed
under the greatest strain and may even break down. There is even less warrant
at the present time, in the absence of a better understanding of the paradoxes
of this paper, for the use of alleged failures of substitutivity in belief contexts
to draw any significant theoretical conclusion about proper names. Hard
cases make bad law. 46
Princeton University
NOTES
"Naming and Necessity," in: The Semantics of Natural Languages, D. Davidson and
O. Harman (eds.), Dordrecht, Reidel, 1971, pp. 253-355 and 763-769. (Also forthcoming
as a separate monograph, pub. Basil Blackwell.) "Identity and Necessity" in: Identity and
Individuation, M. Munitz (ed.), New York University Press, 1971, pp. 135-164. Acquain-
tance with these papers is not a prerequisite for understanding the central puzzle of the
present paper, but is helpful for understanding the theoretical background.
2 Frege gives essentially this example as the second footnote of "On Sense and Refe-
rence." For the "Who is ... 7" to be applicable one must be careful to elicit from one's
informant properties that he regards as defining the name and determining the referent, not
mere well-known facts about the referent. (Of course this distinction may well seem
fictitious, but it is central to the original Frege-Russell theory.)
3 For convenience Russell's terminology is assimilated to Frege's. Actually, regarding
genuine or 'logically proper' names, Russell is a strict Millian: 'logically proper names'
simply refer (to immediate objects of acquaintance). But, according to Russell, what are
ordinarily called 'names' are not genuine, logically proper names, but disguised definite
descriptions. Since Russell also regards definite descriptions as in turn disguised notation,
he does not associate any 'senses' with descriptions, since they are not genuine singular
terms. When all disguised notation is eliminated, the only singular terms remaining are
logically proper names, for which no notion of 'sense' is required. When we speak of
Russell as assigning 'senses' to names, we mean ordinary names and for convenience we
ignore his view that. the descriptions abbreviating them ultimately disappear on analysis.
On the other hand, the explicit doctrine that names are abbreviated definite descriptions
is due to Russell. Michael Dummett, in his recent Frege (Duckworth and Harper and Row,
1973, pp. 110-111) denies that Frege held a description theory of senses. Although as far
as I know Frege indeed makes no explicit statement to that effect, his examples of names
conform to the doctrine, as Dummett acknowledges. Especially his 'Aristotle' example is
revealing. He defines 'Aristotle' just as Russell would; it seems clear that in the case of a
famous historical figure, the 'name' is indeed to be given by answering, in a uniquely
specifying way, the 'who is' question. Dummett himself characterizes a sense as a "cri-
terion ... such that the referent of the name, if any, is whatever object satisfies that
criterion." Since presumably the satisfaction of the criterion must be unique (so a unique
referent is determined), doesn't this amount to defining names by unique satisfaction of
properties, i.e., by descriptions? Perhaps the point is that the property in question need
not be expressible by a usual predicate of English, as might be plausible if the referent is
one of the speaker's acquaintances rather than a historical figure. But I doubt that even
Russell, father of the explicitly formulated description theory, ever meant to require that
the description must always be expressible in (unsupplemented) English.
In any event, the philosophical community has generally understood Fregean senses in
terms of descriptions, and we deal with it under this usual understanding. For present
purposes this is more important than detailed historical issues. Dummett acknowledges
(p. 111) that few substantive points are affected by his (allegedly) broader interpretation
of Frege; and it would not seem to be relevant to the problems of the present paper.
4 See Frege's footnote in "On Sense and Reference" mentioned in note 2 above and
272 SAUL A. KRIPKE
especially his discussion of 'Dr. Gustav Lauben' in" Der Gedanke." (In the recent Geach-
Stoothoff translation, "Thoughts," Logical Investigations, Oxford, Blackwell, 1977, pp.
11-12).
5 Russell, as a Millian with respect to genuine names, accepts this argument with respect
to 'logically proper names.' For example - taking for the moment 'Cicero' and 'Tully' as
'logically proper names,' Russell would hold that ifI judge that Cicero admired Tully, I am
related to Cicero, Tully, and the admiration relation in a certain way: Since Cicero is
Tully, I am related in exactly the same way to Tully, Cicero, and admiration; therefore I
judge that Tully admired Cicero. Again, if Cicero did admire Tully, then according to
Russell a single fact corresponds to all of 'Cicero admired Tully,' 'Cicero admired Cicero,'
etc. Its constituent (in addition to admiration) is the man Cicero, taken, so to speak,
twice.
Russell thought that 'Cicero admired Tully' and 'Tully admired Cicero' are in fact
obviously not interchangeable. For him, this was one argument that 'Cicero' and 'Tully'
are not genuine names, and that the Roman orator is no constituent of propositions (or
'facts,' or 'judgments') corresponding to sentences containing the name.
6 Given the arguments of Church and others, I do not believe that the formal mode of
speech is synonymous with other formulations. But it can be used as a rough way to
convey the idea of scope.
7 It may well be argued that the MilIian view implies that proper names are scopeless and
that for them the de dicta-de re distinction vanishes. This view has considerable plausibility
(my own views on rigidity will imply something like this for modal contexts), but it need
not be argued here either way: de re uses are simply not treated in the present paper.
Christopher Peacocke ("Proper Names, Reference, and Rigid Designation," in: Mean-
ing, Reference, and Necessity, S. Blackburn (ed.), Cambridge, 1975; see Section I), uses
what amounts to the equivalence of the de dicta-de re constructions in all contexts (or,
put alternatively, the lack of such a distinction) to characterize the notion of rigid designa-
tion. I agree that for modal contexts, this is (roughly) equivalent to my own notion, also
that for proper names Peacocke's equivalence holds for temporal contexts. (This is roughly
equivalent to the 'temporal rigidity' of names.) I also agree that it is very plausible to
extend the principle to all contexts. But, as Peacocke recognizes, this appears to imply a
substitutivity principle for co designative proper names in belief contexts, which is widely
assumed to be false. Peacocke proposes to use Davidson's theory of intensional contexts
to block this conclusion (the material in the 'that' clause is a separate sentence). I myself
cannot accept Davidson's theory; but even if it were true, Peacocke in effect acknowledges
that it does not really dispose of the difficulty (p. 127, first paragraph). (Incidentally, if
Davidson's theory does block any inference to the transparency of belief contexts with
respect to names, why does Peacocke assume without argument that it does not do so for
modal contexts, which have a similar grammatical structure?) The problems are thus those
of the present paper; until they are resolved I prefer at present to keep to my earlier more
cautious formulation.
Incidentally, Peacocke hints a recognition that the received platitude - that codesigna-
tive names are not interchangeable in belief contexts - may not be so clear as is generally
supposed.
8 The example comes from Quine, Word and Object, M.I.T. Press, 1960, p. 145. Quine's
conclusion that 'believes that' construed de dicto is opaque has widely been taken for
granted. In the formulation in the text I have used the colon to emphasize that I am speak-
ing of belief de dicto. Since, as I have said, belief de dicto will be our only concern in this
paper, in the future the colon will usually be suppressed, and all 'believes that' contexts
should be read de dicto unless the contrary is indicated explicitly.
9 In many writings Peter Geach has advocated a view that is nonMillian (he would say
'nonLockean') in that to each name a sortal predicate is attached by definition ('Geach,'
for example, by definition names a man). On the other hand, the theory is not completely
Fregean either, since Geach denies that any definite description that would identify the
referent of the name among things of the same sort is analytically tied to the name. (See,
for example, his Reference and Generality, Cornell, 1962, pp. 43-45.) As far as the present
issues are concerned, Geach's view can fairly be assimilated to Mill's rather than Frege's.
For such ordinary names as 'Cicero' and 'Tully' :will have both the same reference and the
same (Geachian) sense (namely, that they are names of a man). It would thus seem that
they ought to be interchangeable everywhere. (In Reference and Generality, Geach appears
not to accept this conclusion, but the prima facie argument for the conclusion will be
the same as on a purely Millian view.)
10 In an unpublished paper, Diana Ackerman urges the problem ofsubstitutivity failures
against the Millian view and, hence, against my own views. I believe that others may have
done so as well. (I have the impression that the paper has undergone considerable revision,
and I have not seen recent versions.) I agree that this problem is a considerable difficulty
for the Millian view, and for the Millian spirit of my own views in "Naming and Necessity."
(See the discussion of this in the text of the present paper.) On the other hand I would
emphasize that there need be no contradiction in maintaining that names are modally rigid,
and satisfy a substitutivity principle for modal contexts, while denying the substitutivity
principle for belief contexts. The entire apparatus elaborated in "Naming and Necessity"
of the distinction between epistemic and metaphysical necessity, and of giving a meaning
and fixing a reference, was meant to show, among other things, that a Millian substitutivity
doctrine for modal contexts can be maintained even if such a doctrine for epistemic con-
texts is rejected. "Naming and Necessity" never asserted a substitutivity principle for
epistemic contexts.
It is even consistent to suppose that differing modes of (rigidly) fixing the reference is
responsible for the substitutivity failures, thus adopting a position intermediate between
Frege and Mill, on the lines indicated in the text of the present paper. "Naming and
Necessity" may even perhaps be taken as suggesting, for some contexts where a conven-
tional description rigidly fixes the reference ('Hesperus-Phosphorus'), that the mode of
reference fixing is relevant to epistemic questions. I knew when I wrote "Naming and
Necessity" that substitutivity issues in epistemic contexts were really very delicate, due to
the problems of the present paper, but I thought it best not to muddy the waters further.
(See notes 43-44.)
After this paper was completed, I saw Alvin Plantinga's paper "The Boethian Com-
promise," The American Philosophical Quarterly 15 (April, 1978): 129-138. Plantinga
adopts a view intermediate between Mill and Frege, and cites substitutivity failures as a
principal argument for his position. He also refers to a forthcoming paper by Ackerman.
I have not seen this paper, but it probably is a descendant of the paper referred to above.
274 SAUL A. KRIPKE
11 Here I use 'connotation' so as to imply that the associated properties have an a priori
tie to the name, at least as rigid reference fixers, and therefore must be true of the referent
(if it exists). There is another sense of 'connotation,' as in 'The Holy Roman Empire,'
where the connotation need not be assumed or even believed to be true of the referent. In
some sense akin to this, classicists and others with some classical learning may attach
certain distinct 'connotations' to 'Cicero' and 'Tully.' Similarly, 'The Netherlands' may
suggest low altitude to a thoughtful ear. Such 'connotations' can hardly be thought of as
community-wide; many use the names unaware of such suggestions. Even a speaker aware
of the suggestion of the name may not regard the suggested properties as true of the object;
cJ. 'The Holy Roman Empire.' A 'connotation' of this type neither gives a meaning nor
fixes a reference.
12 Some might attempt to find a difference in 'sense' between 'Cicero' and 'Tully' on the
grounds that "Cicero is called 'Cicero' " is trivial, but "Tully is called 'Cicero' " may not
be. Kneale, and in one place (probably at least implicitly) Church, have argued in this vein.
(For Kneale, see "Naming and Necessity," p.283.) So, it may be argued, being called
'Cicero,' is part of the sense of the name 'Cicero,' but not part of that of 'Tully.'
I have discussed some issues related to this in "Naming and Necessity," pp. 283-286.
(See also the discussions of circularity conditions elsewhere in "Naming and Necessity.")
Much more could be said about and against this kind of argument; perhaps I will sometime
do so elsewhere. Let me mention very briefly the following parallel situation (which may
be best understood by reference to the discussion in "Naming and Necessity"). Anyone
who understands the meaning of ' is called' and of quotation in English (and that 'alienists'
is meaningful and grammatically appropriate), knows that "alienists are called 'alienists' "
expresses a truth in English, even if he has no idea what 'alienists' means. He need not
know that "psychiatrists are called 'alienists' " expresses a truth. None of this goes to
show that 'alienists' and 'psychiatrists' are not synonymous, or that 'alienists' has being
called 'alienists' as part of its meaning when 'psychiatrists' does not. Similarly for 'Cicero'
and 'Tully.' There is no more reason to suppose that being so-called is part of the meaning
of a name than of any other word.
13 A view follows Frege and Russell on this issue even ifit allows each speaker to associate
a cluster of descriptions with each name, provided that it holds that the cluster varies from
speaker to speaker and that variations in the cluster are variations in idiolect. Searle's view
thus is Frege-Russellian when he writes in the concluding paragraph of "Proper Names"
(Mind 67 (1958): 166-173), " 'Tully = Cicero' would, I suggest, be analytic for most
people; the same descriptive presuppositions are associated with each name. But of course
if the descriptive presuppositions were different it might be used to make a synthetic
statement. "
14 Though here I use the jargon of propositions, the point is fairly insensitive to differences
in theoretical standpoints. For example, on Davidson's analysis, I would be asserting
(roughly) that many are unaware-of-the-content-of the following utterance of mine:
Cicero is Tully. This would be subject to the same problem.
15 Benson Mates, "Synonymity," University of California Publications in Philosophy 25
(1950): 201-226; reprinted in: Semantics and the Philosophy of Language, L. Linsky (ed.),
University of Illinois Press, 1952. (There was a good deal of subsequent discussion. In
Mates's original paper the point is made almost parenthetically.) Actually, I think that
Mates's problem has relatively little force against the argument we are considering for the
Fregean position. Mates's puzzle in no way militates against some such principle as: If
one word is synonymous with another, then a sufficiently reflective speaker subject to no
linguistic inadequacies or conceptual confusions who sincerely assents to a simple sentence
containing the one will also (sincerely) assent to the corresponding sentence with the other
in its place.
It is surely a crucial part of the present 'Fregean' argument that codesignative names may
have distinct 'senses,' that a speaker may assent to a simple sentence containing one and
deny the corresponding sentence containing the other, even though he is guilty of no
conceptual or linguistic confusion, and of no lapse in logical consistency. In the case of two
straightforward synonyms, this is not so.
I myself think that Mates's argument is of considerable interest, but that the issues are
confusing and delicate and that, if the argument works, it probably leads to a paradox or
puzzle rather than to a definite conclusion. (See also notes 23, 28, and 46.)
16 "Naming and Necessity," pp. 291 (bottom)-293.
17 Recall also note 12.
18 Some philosophers stress that names are not words of a language, or that names are
not translated from one language to another. (The phrase 'common currency of our com-
mon language' was meant to be neutral with respect to any such alleged issue.) Someone
may use 'Mao Tse-Tung,' for example, in English, though he knows not one word of
Chinese. It seems hard to deny, however, that "Deutschland," "Allemagne," and "Ger-
many," are the German, French, and English names of a single country, and that one
translates a French sentence using "Londres" by an English sentence using "London."
Learning these facts is part of learning German, French, and English.
It would appear that some names, especially names of countries, other famous localities,
and some famous people are thought of as part of a language (whether they are called
'words' or not is of little importance). Many other names are not thought of as part of a
language, especially if the referent is not famous (so the notation used is confined to a
limited circle), or if the same name is used by speakers of all languages. As far as I can see,
it makes little or no semantic difference whether a particular name is thought of as part of
a language or not. Mathematical notation such as '< ' is also ordinarily not thought of as
part of English, or any other language, though it is used in combination with English
words in sentences of mathematical treatises written in English. (A French mathematician
can use the notation though he knows not one word of English.) 'Is less than,' on the other
hand, is English. Does this difference have any semantic significance?
I will speak in most of the text as if the names I deal with are part of English, French,
etc. But it matters little for what I say whether they are thought of as parts of the language
or as adjuncts to it. And one need not say that a name such as 'Londres' is 'translated' (if
such a terminology suggested that names have 'senses,' I too would find it objectionable),
as long as one acknowledges that sentences containing it are properly translated into
English using 'London.'
19 By saying that names are transparent in a context, I mean that co designative names
are interchangeable there. This is a deviation for brevity from the usual terminology,
according to which the context is transparent. (I use the usual terminology in the paper
also.)
276 SAUL A. KRIPKE
20 But we must use the term 'sense' here in the sense of 'that which fixes the reference,'
not 'that which gives the meaning,' otherwise we shall run afoul of the rigidity of proper
names. If the source of a chain for a certain name is in fact a given object, we use the name
to designate that object even when speaking of counterfactual situations in which some
other object originated the chain.
21 The point is that, according to the doctrine of "Naming and Necessity," when proper
names are transmitted from link to link, even though the beliefs about the referent asso-
ciated with the name change radically, the change is not to be considered a linguistic
change, in the way it was a linguistic change when 'villain' changed its meaning from
'rustic' to 'wicked man.' As long as the reference of a name remains the same, the asso-
ciated beliefs about the object may undergo a large number of changes without these
changes constituting a change in the language.
If Geach is right, an appropriate sortal must be passed on also. But see footnote 58 of
"Naming and Necessity."
22 Similar appropriate restrictions are assumed below for the strengthened disquotational
principle and for the principle of translation. Ambiguities need not be excluded if it is
tacitly assumed that the sentence is to be understood in one way in all its occurrences.
(For the principle of translation it is similarly assumed that the translator matches the
intended interpretation of the sentence.) I do not work out the restrictions on indexicals in
detail, since the intent is clear.
Clearly, the disquotational principle applies only to de dicta, not de re, attributions of
belief. If someone sincerely assents to the near triviality "The tallest foreign spy is a spy,"
it follows that he believes that: the tallest foreign spy is a spy. It is well known that it does
not follow that he believes, of the tallest foreign spy, that he is a spy. In the latter case, but
not in the former, it would be his patriotic duty to make contact with the authorities.
23 What if a speaker assents to a sentence, but fails to assent to a synonymous assertion?
Say, he assents to "Jones is a doctor," but not to "Jones is a physician." Such a speaker
either does not understand one of the sentences normally, or he should be able to correct
himself "on reflection." As long as he confusedly assents to 'Jones is a doctor' but not to
'Jones is a physician,' we cannot straightforwardly apply disquotational principles to con-
clude that he does or does not believe that Jones is a doctor, because his assent is not
"reflective. "
Similarly, if someone asserts, "Jones is a doctor but not a physician," he should be able
to recognize his inconsistency without further information. We have formulated the dis-
quotational principles so they need not lead us to attribute belief as long as we have
grounds to suspect conceptual or linguistic confusion, as in the cases just mentioned.
Note that if someone says, "Cicero was bald but Tully was not," there need be no
grounds to suppose that he is under any linguistic or conceptual confusion.
24 This should not be confused with the question whether the speaker simultaneously
believes of a given object, both that it has a certain property and that it does not have it.
Our discussion concerns de dicto (notional) belief, not de re belief.
I have been shown a passage in Aristotle that appears to suggest that no one can really
believe both of two explicit contradictories. If we wish to use the simple disquotational
principle as a test for disbelief, it suffices that this be true of some individuals, after reflec-
tion, who are simultaneously aware of both beliefs, and have sufficient logical acumen and
respect for logic. Such individuals, if they have contradictory beliefs, will be shaken in one
or both beliefs after they note the contradiction. For such individuals, sincere reflective
assent to the negation of a sentence implies disbelief in the proposition it expresses, so the
test in the text applies.
2S For example, in translating a historical report into another language, such as, "Patrick
Henry said, 'Give me liberty or give me death!' " the translator may well translate the
quoted material attributed to Henry. He translates a presumed truth into a falsehood,
since Henry spoke English; but probably his reader is aware of this and is more interested
in the content of Henry's utterance than in its exact words. Especially in translating fiction,
where truth is irrelevant, this procedure is appropriate. But some objectors to Church's
'translation argument' have allowed themselves to be misled by the practice.
26 To state the argument precisely, we need in addition a form of the Tarskian disquota-
tion principle for truth: For each (French or English) replacement for 'p,' infer" 'p'
is true" from "p," and conversely. (Note that" 'p' is true" becomes an English sentence
even if 'p' is replaced by a French sentence.) In the text we leave the application of the
Tarskian disquotational principle tacit.
27 I gather that Burali-Forti originally thought he had 'proved' that the ordinals are not
linearly ordered, reasoning in a manner similar to our topologist. Someone who heard the
present paper delivered told me that Konig made a similar error.
28 It is not possible, in this case, as it is in the case of the man who assents to "Jones is a
doctor" but not to "Jones is a physician," to refuse to apply the disquotational principle
on the grounds that the subject must lack proper command of the language or be subject
to some linguistic or conceptual confusion. As long as Pierre is unaware that 'London'
and 'Londres' are co designative, he need not lack appropriate linguistic knowledge, nor
need he be subject to any linguistic or conceptual confusion, when he affirms 'Londres est
jolie' but denies 'London is pretty.'
29 The 'elimination' would be most plausible if we believed, according to a Russellian
epistemology, that all my language, when written in unabbreviated notation, refers to
constituents with which I am 'acquainted' in Russell's sense. Then no one speaks a language
intelligible to anyone else; indeed, no one speaks the same language twice. Few today will
accept this.
A basic consideration should be stressed here. Moderate Fregeans attempt to combine a
roughly Fregean view with the view that names are part of our common language, and that
our conventional practices of interlinguistic translation and interpretation are correct. The
problems of the present paper indicate that it is very difficult to obtain a requisite socialized
notion of sense that will enable such a program to succeed. Extreme Fregeans (such as
Frege and Russell) believe that in general names are peculiar to idiolects. They therefore
would accept no general rule translating 'Londres' as 'London,' nor even translating one
person's use of 'London' into another's. However, if they follow Frege in regarding senses
as 'objective,' they must believe that in principle it makes sense to speak of two people
using two names in their respective idiolects with the same sense, and that there must be
(necessary and) sufficient conditions for this to be the case. If these conditions for sameness
of sense are satisfied, translation of one name into the other is legitimate, otherwise not.
The present considerations (and the extension of these below to natural kind and related
terms), however, indicate that the notion of sameness of sense, if it is to be explicated in
278 SAUL A. KRIPKE
terms of sameness of identifying properties and if these properties are themselves expressed
in the languages of the two respective idiolects, presents interpretation problems of the
same type presented by the names themselves. Unless the Fregean can give a method for
identifying sameness of sense that is free of such problems, he has no sufficient conditions
for sameness of sense, nor for translation to be legitimate. He would therefore be forced to
maintain, contrary to Frege's intent, that not only in practice do few people use proper
names with the same sense but that it is in principle meaningless to compare senses. A view
that the identifying properties used to define senses should always be expressible in a
Russellian language of 'logically proper names' would be one solution to this difficulty but
involves a doubtful philosophy of language and epistemology.
30 If any reader finds the term 'translation' objectionable with respect to names, let him
be reminded that all I mean is that French sentences containing 'Londres' are uniformly
translated into English with 'London.'
31 The paradox would be blocked if we required that they define the names by the same
properties expressed in the same words. There is nothing in the motivation of the classical
description theories that would justify this extra clause. In the present case of French and
English, such a restriction would amount to a decree that neither 'Londres,' nor any other
conceivable French name, could be translated as 'London.' I deal with this view imme-
diately below.
32 Word salads of two languages (like ungrammatical 'semisentences' ofa single language)
need not be unintelligible, though they are makeshifts with no fixed syntax. "If God did
not exist, Voltaire said, ilfaudrait l'inventer." The meaning is clear.
33 Had we said, "Pierre believes that the country he calls 'Angleterre' is a monarchy," the
sentence would be English, since the French word would be mentioned but not used. But
for this very reason we would not have captured the sense of the French original.
34 Under the influence of Quine's Word and Object, some may argue that such conclu-
sions are not inevitable: perhaps he will translate' medecin' as 'doctor stage,' or 'undetached
part of a doctor'! If a Quinean skeptic makes an empirical prediction that such reactions
from bilinguals as a matter of fact can occur, I doubt that he will be proved correct. (I
don't know what Quine would think. But see Word and Object, p. 74, first paragraph.) On
the other hand, if the translation of'medecin' as 'doctor' rather than 'doctor part' in this
situation is, empirically speaking, inevitable, then even the advocate of Quine's thesis will
have to admit that there is something special about one particular translation. The issue is
not crucial to our present concerns, so I leave it with these sketchy remarks. But see also
note 36.
35 Putnam gives the example of elms and beeches in "The Meaning of 'Meaning' " (in:
Language, Mind, and Knowledge, Minnesota Studies,in the Philosophy of Science 7; also
reprinted in Putnam's Collected Papers). See also Putnam's discussion of other examples
on pp. 139-143; also my own remarks on 'fool's gold,' tigers, etc., in "Naming and
Necessity," pp. 316-323.
36 It is unclear to me how far this can go. Suppose Pierre hears English spoken only in
England, French in France, and learns both by direct method. (Suppose also that no one
else in each country speaks the language of the other.) Must he be sure that 'hot' and
'chaud' are coextensive? In practice he certainly would. But suppose somehow his experience
is consistent with the following bizarre - and of course, false! - hypothesis: England and
France differ atmospherically so that human bodies are affected very differently by their
interaction with the surrounding atmosphere. (This would be more plausible if France
were on another planet.) In particular, within reasonable limits, things that feel cold in one
of the countries feel hot in the other, and vice versa. Things don't change their temperature
when moved from England to France, they just feel different because of their effects on
human physiology. Then 'chaud,' in French, would be true of the things that are called
'cold' in English! (Of course the present discussion is, for space, terribly compressed. See
also the discussion of 'heat' in "Naming and Necessity." We are simply creating, for the
physical property 'heat,' a situation analogous to the situation for natural kinds in the
text.)
If Pierre's experiences were arranged somehow so as to be consistent with the bizarre
hypothesis, and he somehow came to believe it, he might simultaneously assent to 'C'est
chaud' and 'This is cold' without contradiction, even though he speaks French and English
normally in each country separately.
This case needs much more development to see if it can be set up in detail, but I cannot
consider it further here. Was I right in assuming in the text that the difficulty could not
arise for 'medecin' and 'doctor'?
37 One might argue that Peter and we do speak different dialects, since in Peter's idiolect
'Paderewski' is used ambiguously as a name for a musician and a statesman (even though
these are in fact the same), while in our language it is used unambiguously for a musician-
statesman. The problem then would be whether Peter's dialect can be translated homo-
phonically into our own. Before he hears of 'Paderewski-the-statesman,' it would appear
that the answer is affirmative for his (then unambiguous) use of 'Paderewski,' since he did
not differ from anyone who happens to have heard of Paderewski's musical achievements
but not of his statesmanship. Similarly for his later use of 'Paderewski,' if we ignore his
earlier use. The problem is like Pierre's, and is essentially the same whether we describe it
in terms of whether Peter satisfies the condition for the disquotational principle to be
applicable, or whether homophonic translation of his dialect into our own is legitimate.
38 D. Davidson, "On Saying That," in: Words and Objections, D. Davidson and J. Hin-
tikka (eds.), Dordrecht, Reidel, 1969, p. 166.
39 In Word and Object, p. 221, Quine advocates a second level of canonical notation,
"to dissolve verbal perplexities or facilitate logical deductions," admitting the proposi-
tional attitudes, even though he thinks them "baseless" idioms that should be excluded
from a notation "limning the true and ultimate structure of reality."
40 In one respect the considerations mentioned above on natural kinds show that Quine's
translation apparatus is insufficiently skeptical. Quine is sure that the native's sentence
"Gavagai!" should be translated "Lo, a rabbit!", provided that its affirmative and negative
stimulus meanings for the native match those of the English sentence for the Englishman;
skepticism sets in only when the linguist proposes to translate the general term 'gavagai' as
'rabbit' rather than 'rabbit stage,' 'rabbit part,' and the like. But there is another possibility
that is independent of (and less bizarre than) such skeptical alternatives. In the geographical
area inhabited by the natives, there may be a species indistinguishable to the nonzoologist
from rabbits but forming a distinct species. Then the:'stimulus meanings,' in Quine's sense,
of'Lo, a rabbit!' and 'Gavagai!' may well be identical (to nonzoologists), especially if the
ocular irradiations in question do not include a specification of the geographical locality .
280 SAUL A. KRIPKE
('Gavagais' produce the same ocular irradiation patterns as rabbits.) Yet 'Gavagai!' and
'Lo, a rabbit!' are hardly synonymous; on typical occasions they will have opposite truth
values.
I believe that the considerations about names, let alone natural kinds, emphasized in
"Naming and Necessity" go against any simple attempt to base interpretation solely on
maximizing agreement with the affirmations attributed to the native, matching of stimulus
meanings, etc. The 'Principle of Charity' on which such methodologies are based was first
enunciated by Neil Wilson in the special case of proper names as a formulation of the
cluster-of-descriptions theory. The argument of "Naming and Necessity" is thus directed
against the simple 'Principle of Charity' for that case.
41 Geach introduced the term 'Shakespearean' after the line, "a rose / By any other name,
would smell as sweet."
Quine seems to define 'referentially transparent' contexts so as to imply that core-
ferential names and definite descriptions must be interchangeable salva veritate. Geach
stresses that a context may be 'Shakespearean' but not 'referentially transparent' in this
sense.
42 Generally such cases may be slightly less watertight than the 'London'-'Londres' case.
'Londres' just is the French version of 'London,' while one cannot quite say that the same
relation holds between 'Ashkenaz' and 'Germaniah.' Nevertheless:
(a) Our standard practice in such cases is to translate both names of the first language
into the single name of the second.
(b) Often no nuances of 'meaning' are discernible differentiating such names as
'Ashkenaz' and 'Germaniah,' such that we would not say either that Hebrew would have
been impoverished had it lacked one of them (or that English is improverished because it
has only one name for Germany), any more than a language is impoverished if it has only
one word corresponding to 'doctor' and 'physician.' Given this, it seems hard to condemn
our practice of translating both names as 'Germany' as 'loose'; in fact, it would seem that
Hebrew just has two names for the same country where English gets by with one.
(c) Any inclinations to avoid problems by declaring, say, the translation of 'Ashkenaz'
as 'Germany' to be loose should be considerably tempered by the discussion of analogous
problems in the text.
43 In spite of this official view, perhaps I will be more assertive elsewhere.
In the case of 'Hesperus' and 'Phosphorus' (in contrast to 'Cicero' and 'Tully'), where
there is a case for the existence of conventional community-wide 'senses' differentiating the
two - at least, two distinct modes of 'fixing the reference of two rigid designators' - it is
more plausible to suppose that the two names are definitely not interchangeable in belief
contexts. According to such a supposition, a belief that Hesperus is a planet is a belief that
a certain heavenly body, rigidly picked out as seen in the evening in the appropriate season,
is a planet; and similarly for Phosphorus. One may argue that translation problems like
Pierre's will be blocked in this case, that' Vesper' must be translated as 'Hesperus,' not as
'Phosphorus.' As against this, however, two things:
(a) We should remember that sameness of properties used to fix the reference does not
appear to guarantee in general that paradoxes will not arise. So one may be reluctant to
adopt a solution in terms of reference-fixing properties for this case if it does not get to the
heart of the general problem.
(b) The main issue seems to me here to be - how essential is a particular mode of fixing
the reference to a correct learning of the name? If a parent, aware of the familiar identity,
takes a child into the fields in the morning and says (pointing to the morning star) "That
is called 'Hesperus,' " has the parent mistaught the language? (A parent who says, "Crea-
tures with kidneys are called 'cordates,' definitely has mistaught the language, even though
the statement is extensionally correct.) To the extent that it is not crucial for correct
language learning that a particular mode of fixing the reference be used, to that extent
there is no 'mode of presentation' differentiating the 'content' of a belief about 'Hesperus'
from one about 'Phosphorus.' I am doubtful that the original method of fixing the reference
must be preserved in transmission of the name.
If the mode of reference fixing is crucial, it can be maintained that otherwise identical
beliefs expressed with 'Hesperus' and with 'Phosphorus' have definite differences of
'content,' at least in an epistemic sense. The conventional ruling against substitutivity
could thus be maintained without qualms for some cases, though not as obviously for
others, such as 'Cicero' and 'Tully.' But it is unclear to me whether even 'Hesperus' and
'Phosphorus' do have such conventional 'modes of presentation.' I need not take a definite
stand, and the verdict may be different for different particular pairs of names. For a brief
related discussion, see "Naming and Necessity," p. 331, first paragraph.
44 However, some earlier formulations expressed disquotationally such as "It was once
unknown that Hesperus is Phosphorus" are questionable in the light of the present paper
(but see the previous note for this case). I was aware of this question by the time "Naming
and Necessity" was written, but I did not wish to muddy the waters further than necessary
at that time. I regarded the distinction between epistemic and metaphysical necessity as
valid in any case and adequate for the distinctions I wished to make. The considerations in
this paper are relevant to the earlier discussion of the 'contingent a priori' as well; perhaps
I will discuss this elsewhere.
45 According to Russell, definite descriptions are not genuine singular terms. He thus
would have regarded any concept of 'referential opacity' that includes definite descriptions
as profoundly misleading. He also maintained a substitutivity principle for 'logically proper
names' in belief and other attitudinal contexts, so that for him belief contexts were as
'transparent,' in any philosophically decent sense, as truth-functional contexts.
Independently of Russell's views, there is much to be said for the opinion that the
question whether a context is 'Shakespearean' is more important philosophically - even
for many purposes for which Quine invokes his own concept - than whether it is 're-
ferentially opaque.'
46 I will make some brief remarks about the relation of Benson Mates's problem (see
note 15) to the present one. Mates argued that such a sentence as (*)'Some doubt that
all who believe that doctors are happy believe that physicians are happy,' may be true,
even though 'doctors' and 'physicians' are synonymous, and even though it would have
been false had 'physicians' been replaced in it by a second occurrence of 'doctors.' Church
countered that (*) could not be true, since its translation into a language with only one word
for doctors (which would translate both 'doctors' and 'physicians') would be false. If both
Mates's and Church's intuitions were correct, we might get a paradox analogous to
Pierre's.
Applying the principles of translation and disquotation to Mates's puzzle, however,
282 SAUL A. KRIPKE
involves many more complications than our present problem. First, if someone assents to
'Doctors are happy,' but refuses assent to 'Physicians are happy,' prima facie disquotation
does not apply to him since he is under a linguistic or conceptual confusion. (See note 23.)
So there are as yet no grounds, merely because this happened, to doubt that all who believe
that doctors are happy believe that physicians are happy.
Now suppose someone assents to 'Not all who believe that doctors are happy believe
that physicians are happy.' What is the source of his assent? If it is failure to realize that
'doctors' and 'physicians' are synonymous (this was the situation Mates originally en-
visaged), then he is under a linguistic or conceptual confusion, so dis quotation does not
clearly apply. Hence we have no reason to conclude from this case that (*) is true. Alterna-
tively, he may realize that 'doctors' and 'physicians' are synonymous; but he applies dis-
quotation to a man who assents to 'Doctors are happy' but not to 'Physicians are happy,'
ignoring the caution of the previous paragraph. Here he is not under a simple linguistic
confusion (such as failure to realize that 'doctors' and 'physicians' are synonymous), but
he appears to be under a deep conceptual confusion (misapplication of the disquotational
principle). Perhaps, it may be argued, he misunderstands the 'logic of belief.' Does his
conceptual confusion mean that we cannot straightforwardly apply dis quotation to his
utterance, and that therefore we cannot conclude from his behavior that (*) is true? I
think that, although the issues are delicate, and I am not at present completely sure what
answers to give, there is a case for an affirmative answer. (Compare the more extreme case
of someone who is so confused that he thinks that someone's dissent from 'Doctors are
happy' implies that he believes that doctors are happy. If someone's utterance, 'Many
believe that doctors are happy,' is based on such a misapplication of disquotation, surely
we in turn should not apply disquotation to it. The utterer, at least in this context, does not
really know what 'belief' means.)
I do not believe the discussion above ends the matter. Perhaps I can discuss Mates's
problem at greater length elsewhere. Mates's pro blem is perplexing, and its relation to the
present puzzle is interesting. But it should be clear from the preceding that Mates's argu-
ment involves issues even more delicate than those that arise with respect to Pierre. First,
Mates's problem involves delicate issues regarding iteration of belief contexts, whereas the
puzzle about Pierre involves the application of disquotation only to affirmations of (or
assents to) simple sentences. More important, Mates's problem would not arise in a world
where no one ever was under a linguistic or a conceptual confusion, no one ever thought
anyone else was under such a confusion, no one ever thought anyone ever thought
anyone was under such a confusion, and so on. It is important, both for the puzzle about
Pierre and for the Fregean argument that 'Cicero' and 'Tully' differ in 'sense,' that they
would still arise in such a world. They are entirely free of the delicate problem of applying
disquotation to utterances directly or indirectly based on the existence of linguistic con-
fusion. See notes 15 and 28, and the discussion in the text of Pierre's logical consistency.
Another problem discussed in the literature to which the present considerations may be
relevant is that of 'self-consciousness,' or the peculiarity of 'I.' Discussions of this problem
have emphasized that 'I,' even when Mary Smith uses it, is not interchangeable with
'Mary Smith,' nor with any other conventional singular term designating Mary Smith. If
she is 'not aware that she is Mary Smith,' she may assent to a sentence with 'I,' but dissent
from the corresponding sentence with 'Mary Smith.' It is quite possible that any attempt
to clear up the logic of all this will involve itself in the problem of the present paper. (For
this purpose, the present discussion might be extended to demonstratives and indexicals.)
The writing of this paper had partial support from a grant from the National Science
Foundation, a John Simon Guggenheim Foundation Fellowship, a Visiting Fellowship at
All Souls College, Oxford, and a sabbatical leave from Princeton University. Various
people at the Jerusalem Encounter and elsewhere, who will not be enumerated, influenced
the paper through discussion.
HILARY PUTNAM
COMMENTS
Just as a way of opening the discussion, I am going to say a little about the
philosophy of language and a little about the philosophy of mind, because I
think that the problem that Saul has raised touches on issues in both the
philosophy of language and the philosophy of mind.
First the philosophy of language. It has seemed to me for a long time, ever
since the 1950's when I started thinking about the theory of reference and
things of that kind, that the idea that words are understood by being asso-
ciated with necessary and sufficient conditions cannot be right. For one thing
it cuts against the principle that reference can be preserved across theory
change, which seems to me to be central to any realist philosophy of science -
the principle that Professor Quine has called the Principle of Charity, or that I
call the Principle of Benefit of the Doubt. Take a very old example - if you
take the name of a disease, e.g. multiple sclerosis, it is wrong to say that the
meaning of the term changes each time our theory changes. People are still
shifting back and forth on multiple sclerosis concerning the view that it is
just a name given, unfortunately given, to a great many different diseases -
that is one extreme view about multiple sclerosis. (On this view you have
multiple sclerosis if and only if you have a disease such that when you die it
turns out that there are little scars on the myelin sheath of your spinal
cord.)
On the other view, multiple sclerosis is caused by a specific virus. Now
suppose we discover that the second view is right (and it seems more and
more to be right). We may then say of someone who died without there being
any scleromata on the myelin sheath, that he died of multiple sclerosis because
we decide that his death was caused by this virus and this virus had an aty-
pical effect because of his atypical metabolism or something. So that we
would say this was a case of multiple sclerosis which we could not previously
have known to be multiple sclerosis.
Now, the problem is, then, what is the continuity in the meaning of a word
like 'multiple sclerosis' if it is neither that the scientists necessarily use the
same tests at different times, nor that they give the same theoretical account?
I think the continuity is 99 % sameness of reference. The "meaning" is the
284
Copyright 1979 by D. Reidel Publishing Company.
COMMENTS 285
same as long as by the light of our later theories we are still using 'multiple
sclerosis' as a name for the same disease, making that decision from within
our current theory and making the normal adjustments due to charity. We
do not postulate a meaning change if we are prepared to say it is still the same
disease that these tests are tests for, and this is a matter, I think, of attributing
certain conditional intentions to earlier speakers, like saying, "surely if those
scientists had known that those symptoms were caused by this virus they would
have also classified this case as a case of multiple sclerosis." 99 %of the same-
ness of meaning is sameness of reference. Certainly Frege's argument shows
meaning cannot just be reference, but there may be much more truth than
falsity to the view that meaning is reference.
I was very glad that Saul mentioned Mill because I discovered that Mill
had an inkling of this in the case of general names. In the case of the word
'gold,' although Mill tries to preserve the standard account in which gold has
a connotation, a Fregean intension, and this at least approximately fixes the
extension - he says, quite surprisingly, that this does not exactly fix the
extension of the word 'gold' and the job of extension fixing is completed by
the substance itself. I think actually 99 % of the extension of the word 'gold'
is fixed by the substance itself. If you leave out the modern period when we
know the atomic composition of gold and so on, and take 99 % of the time
people have talked about gold, what people had was only paradigms; even
the ability to pick out those paradigms reliably was most of the time pos-
sessed only by experts, not by the lay speaker. But 'gold' was used as a name
not only for the paradigms, but for whatever was of the same nature as the
paradigms - where this was not a completely open metaphysical concept
because there were some notions about what it is for a metal, at least, to have
a nature. For example, if the paradigm samples were not alloy, then an alloy
is not gold even ifit can be made to look like the paradigm samples. It made
sense for Archimedes to worry about whether that crown was really a gold
crown even though it passed the operational test. Yet, if 'gold' simply meant
whatever passed the operational test, then Archimedes could just have gone
to the king and said "I have read Bridgman, and something is gold if it
passes the operational tests; it passes the operational tests, so your crown is
gold" and then he would have had his head chopped off and the king would
have gone to a better scientist!
The point of all this is that if this is right then Saul's problem affects just as
much general names as it does proper names. Saul's example was of London
andLondres, but one could do it just as well with general names. If this sort of
286 HILARY PUTNAM
semantics is right, that is, if most of the meanings of many terms is simply to
be identified with their reference, if meaning is something like an ordered pair
of a reference and a stereotype, then it is possible for a speaker to have exact
synonyms in his total vocabulary or total collection of languages and not
know that they are synonyms. Although the principle is evident on Fregean
semantics, that if you have two synonyms in your language or collection of
languages, then you must know that they are synonyms, on a "referentialist"
account like the one just sketched, this is wrong. Suppose you do not know that
"Buche" in German is translated as "beech" in English, but you do know the
trees in the Boston Common are called "beeches." Perhaps you have been
told in German that there are not any beech trees in the Boston Common, so
then exactly the same difficulty Saul pointed out would arise, that the normal
practice of translation, applied to what you sincerely say in German should
lead us to say that you do not believe there are any beech trees in the Boston
Common. The normal practice of going from direct discourse to indirect
discourse applied to your English utterances should lead us to say that you
do believe there are beech trees in the Boston Common. And the same holds
for chrysos and gold, chrysos being the ancient Greek word for gold, and so
on and so on, one can multiply these examples - but the point is that if it
is right that many, many more words in the language function like names
than is allowed by the Fregean paradigm, then the problem Saul has raised
is of great generality. That is all I am going to say about the philosophy of
language.
Now for a very short remark on philosophy of mind. r believe something
which if r just trotted it out today to deal with this problem would sound ad
hoc. Fortunately r did put it forward in some lectures r gave a few weeks ago
r
in England without this problem in mind, so can at least claim there are
some independent reasons - they might be terrible reasons but they are
certainly independent reasons - for thinking that what r am about to say is
true.
r think a general and very important phenomenon in language is that whole
systems of descriptions, whole ways of talking which are formally incom-
patible - that is, if you treat them as being in the same language and then
just write down the first set of sentences and then the second set of sentences,
what you get is inconsistent - may be equivalent descriptions. This notion of
equivalent descriptions was introduced by Reichenbach in Experience and
Prediction and exploited by him very heavily.
Now r believe there are equivalent descriptions in common sense psycho-
COMMENTS 287
logy, that is, the description of states of affairs in ordinary language psycholo-
gical vocabulary is not unique, and there can be whole systems of sentences
which are formally incompatible which come to two ways of describing the
same facts. I think that the example that Saul gave (taking 'belief' as a relevant
ordinary language psychological term, or the constellation involving 'belief'
and 'mean') is an example of this. That is to say, I think here we have a choice
between deciding to say - and neither description is very nice - that Pierre
believes that London is ugly and that he does not know that London is
Londres; or we can say that he believes that London is pretty and he does not
know that he is living in London; I think probably the best thing is to abandon
the notion of having beliefs about London in the case of Pierre and to say
that Pierre believes that Lopdon is pretty under the description 'London' and
believes that it is ugly under the description 'Londres.' It does not mean that
we give up the idea of simply believing that London is pretty in the case of
most speakers. The point is, in a nutshell, that the whole rationale of transla-
tion between languages is that what a translation is supposed to do for you
is enable you to carryover ordinary language psychological explanations. The
whole purpose of translation is so we can say he means this, he believes this,
he wants this, he knows this and that is why he does that, and I think that the
stability of our way of using belief with indirect discourse depends on the fact
that cases like this one are relatively rare. I think if cases like this one were
the rule rather than the exception, we would have no choice but to fall back
on the notion of believing sentences and let belief within indirect discourse go.
But fortunately they are rather rare.
Now, I know I have taken a risk in putting forward any view at all. I am
one of those who believe that very often no view is better than some, but I
hope that this is not such a case.
Of course when one decides to say, here are two theories which appear to
be formally incompatible, but they are both true, they are equivalent descrip-
tions of the world, or of psychological fact, or whatever, then one is com-
mitted to saying that the reference of some terms is not the same. But I want
to argue that there may be no method for telling that from semantic theory. If
meaning were a mechanism for fixing reference, then one could always tell
that one was dealing with equivalent descriptions by looking at one's se-
mantic theory and picking a translation which preserved reference, instead of
using the "homophonic" translation (i.e. the identity mapping).
A quick example. A mathematical example: you have a theory according
to which mathematical heaven consists of objects called sets. The theory says
288 HILARY PUTNAM
that some of these sets are functions - in fact it says that those sets which are
sets of ordered pairs satisfying a certain functionality condition are functions,
and then it proceeds to translate your favorite textbook on calculus. The
other theory says that mathematical heaven consists of things called func-
tions, even zero, I, 2, 3 turn out to be functions, and it says that some of
these functions are sets - in fact those functions which take on only the
values zero and I are sets. Now, these are formally incompatible. Ifwe decide,
and I think that it would be insane not to, that these are equivalent descrip-
tions, then when we view the world from within the set theory, we will say
"the word 'set' used in the other theory does not refer to what I am calling
sets." Or we could do the other thing - speaking from within the function
theory, we could say "the word 'function' in the other theory does not refer
to what I am calling functions," and we could stipulate what it does refer to
in several different ways. The choice is not dictated, by the way, by linguistic
theory or semantic theory in any sense of "semantic theory" that I am able
to understand. A sophisticated realist should not be bothered by the collapse
of the "One True Theory" version of realism.
Harvard University
[The above is Hilary Putnam's comment on Saul Kripke's lecture at the Encounter,
which consisted mainly of the "the puzzle" - Section III here. Ed.]
INDEX
aboutness 77 155 168 pattern governed 32

Ackerman, D. 273 behaviouristic analysis 196
action belief 2225 148 150 151 163 184 185 190
rule of 199 191 195 197 202-204 207 215 222 230
theory of 163 239241 244245247-254256-259 261-
activities 263 266-269 276 277 281 287
rule-governed 58 context 82 241-244 246 247 251-253
adequacy condition 41 42 260 262 268-270 272 273 280-282,
Alston, W.P. 51 53 transparency of 272 281
ambiguity contradictory 251 257 276 277
de dicto-de re 81 de dicto-de re attribution of 242 273
of sentences 49 66 276
analytic 240 274 hard core 146 147 150 151
Andrews, P. 65 intensional 181 184
antecedence relations 73 logic of 27
appropriateness 42 126 operators 246
a priori 25 26 34 243 reporting 239
Aquist, L. 113 theory of 267
Aristotle 233 Berlin, I. 143 152 158
artificial intelligence 120 121 Birkhoff, G.D. 217
assent and dissent 21 127 132 201 bivalence
assertion9-13151719 22 53102124-129 principle of 129 130 134 135 139 218
131-140 188 194222 220222- 225
practice of 126 128 Blanshard, B. 48 53
sign 9 12 15 18 Bohnert, H. 14 20
attributes 99 Boolean operations 66
Augustine 109 BUhler, K. 102
Austin, J.L. 10 16 20 37 41 143 144 152 Burali-Forti, C. 277
164
Avenarius 212 Carnap, R. 52 111 113 114149 152 199
201 211 212 214 242
Ballmer, T. 113 change of meaning I change of theory
Bar-Hillel, Y. ix xi 9 1521 37385294 dichotomy 200
behaviour 38 49 87 131 132 133 138 142 Chomsky, N. 37 39 41 52152
143 146 147 149 155 161 175 189 191 Church, A. 272 274 277 281
194196200-208213 216 218 circularity conditions 274
linguistic 131 146 147 202203 204206 circumstance sentences 93 94 98 100 101
207 208 216218 102 106
289
290 INDEX
circumstances 1 3 4 34 43 94 95 96 107 cooperation

116117118120218226237 principle of 49 53 153 154
of utterance 95100109 coreference 83 90 91
circumstantial correspondence 201 205-208 210 211 213
determination 97 104 107 116 228
determiner 97-101 108 counterfactuals 57 151 222243 276
Cocchiarella, N. 113 counteridentity statements 151
Cohen, L.J. 53 164 counternomic statements 151
competence 12 39 40 42 45 46 47 51 52 Cresswell, M.J. 52
119 149 159
and performance distinction 39 Dascal, M. 158 173 175-179
conceptual Davidson, D. 21 224042 52 53 113 132
dependency 120 138162163173186266272 274 279
scheme 103 105205 deductive inference 221 224
conditionals 57-60 63 64 66 68-73 76 80 definition
81 82 85 86 91 222 eliminative 142
conditions private ostensive 131
necessary and sufficient 277 284 definitization 90
confirmation description 241 242 245 259 271 274 287
degree of 227 abbreviated 252
connotation 244 274 co designative 242 243 252
conscious conventional 273
states 29 30 coreferential 241, definite 270
consciousness 28 29 35 defining 247281
constructivist 220 definite 241 260271 273 280, 'referential
context 2 6 40 41 42 50 67 82 87 91 93 94 opacity' of 270 281, theories of 90
9598 101-107 112 118 126 128 142143 185
146 159 160 162 168 169 179 180 181 equivalent 286-288
215 identifying 256 260
and cotext 94 95 118 indefinite 246
attitudinal 281 systems of 286
epistemic 241 243 273 theories of 263 270 271, explicitly
opaque 281 formulated 271
'Shakespearian' 280 281 theorist 261 262 278
temporal 272 designators
transparent 275 280 rigid 243 245272 280
truth-functional 281 Dewey, J. 1
see also intensional-, modal context disambiguation 162 163
contextual analysis 95 96 103 pragmatic 164
convention 11-1429 81 194 195 227 semantic 163 164
conversation 153 157 159-162 167 170 disquotation 251 253 254269281 282
175-180 disquotational judgments 254
conversational demand 159-167 176- disquotational principle 248-252 263
179, reaction to 159-168170175177 265-268 270 276 277 279 281 282,
178 simple 249 250 276, strengthened
INDEX 291
(biconditional) 249 250 253 258 276, Geach, P.T. 15 16 18 20 267 273276280
Tarskian 277 Glouberman, M. 172
disquotational technique 250 251 Goodman, N. 113 158228
dream 25 26 33 34 Gower, E. 53
Dummett, M. 9-11 14-16 18 20 42 53 Grice, H.P. 47-49 53 127 153-158 167
136-140 199208-210216217226-228 170-173 175 178 179
271 Gunter, R. 171
entailment 100 103 155 Hare, R.M. 164

eternalization of sentences 94-96 101 117 Hempel, C.G. 4152
evidence 42 43 139 214 Henkin, L. 65
empirical 243 Hintikka, J. 20495053585971 9092279
explanation of reliability of learning Hiz, H. 53 103 113
causal 207-211 homophonic translation, see translation
scientific 208 Hume, D. 28
extension 117 215285 hypothesis
explanatory 167176178179
facts 272
hard and soft 222 223 228 idealistic fallacy argument 209 210 219
fairy tales 143 145 220226
Feyerabend, P. 26 identity mapping 287
Fichte, J.G. 145 illocutionary acts 41 43 194 195 196
Fodor, J.A. 146 152 implicature 38 4447484953 127155-170
force 9 10 16 18 20 57 58 80 81 125 128 172 173175-180
131 164 165 166 168 169 170 173 175 indecisiveness (as pragmatic notion) 142
195 indeterminacy (as semantic notion) 142
illocutionary 13 19 20 162 164 165 166 indexical expressions 37 49 127
176 191 indications (as opposed to implications)
semantic 164 165 166 168 176 161
Frege, G. 91215125130131140185221 institution 43 4446 51
227 239 240 244-247 251 252 259 270 institutional role 42 43 51
271 273-275 277 278 282 285 286 intension 14 117 118 149 150
functions and functionals 61-66 Fregean 285
intensional
Gabbay, D. 53 contexts 181 241 272
Gaifman, H. 238 idioms 266
Galilei, G. 21 22 sentence 181 183
game 23 24 27 28 34 58 60 61 87 91 104 intensionality 181 183 184186-189197
148149167 intention
admission 23-25 28 31-33 conditional 285
of objects (GO) 23-29 31-33 of speaker 13 126 161 162 167 176 195
perception and memory (pMG) 24-29 intentional
31-33 object 182-189
theory 61 state 182-195 197
see also semantics, game theoretical terms 30 31
292 INDEX
theory of language 44 Lewis, C.I. 212

intentionality 28 161 181-185 187-192 Lewis, D. 1420117
195-197241 Liar Paradox 267
internal relation 183 184 linguistic
interpretation 51 105 106 109-112 163 acts 190 191 193 197
164173179228 appropriateness relationship 42
intuition 39 120146150 behaviour 255
intuitionists 64 208 219 change 276
Isard, S. 113 division oflabour 4349151 200214
function 240
James, W. 28 institution 42 43 46 50
judgment 272 means 40 51
stroke 9 15 universal 117 118
justification linguistics 118
notion of 124125 131 132 133 136-140 literal
153 purpose 40 42-45 47 51 52
thesis 138 139 use 181
Littlewood, J.E. 123
Kafka, F. 150 Locke, J. 149 150
Kamp, H. 113 logic 39
Kant, I. 28 abductive 167
Karttunen, L. 71-73 classical 139 216 217 219 223 224 227
Kasher, A. 5253 106 113 228
Katz, J.J. 152 deductive 199 200 203 207212215219
Kenny, A. 183 inductive 199 200 206 212 215 217
Kneale, W. 274 intuitionistic 139
knowledge 113 134 139 140 153 200 204 modal 101
206 modular 217
Konig, J. 277 non-classical 60 216
Kripke, S. 43, 211 217 222 224 238 284- logical
288 connectives 29 194 208
entities 30
Lakatos, I. 152 Longuet-Higgins, H.C. 107 113
language 4 568 11 13 29 32394549 101 Lorenzen, P. 50
118 128 133 134 149 190 193 199 200 Lyons, J. 101 103 113
207 208 211 213 215 216218 224225
227228 Mach, E. 212
entry transitions 32 33 35 Mackie, J.L. 183 184
game 23 31-35 50 58 125-127 149 199 Manor, R. 172
214224 Margalit, A. 222
user 38 3941 215 Mates, B. 274275 281 282
see also object-, observation-, ordinary-, maxims 202
natural-, scientific language and supermaxims 48 49 154 155 156
Lappin, S. 52 158 171 172
law of excluded middle 123 124 127 may-bes 229 232 236 237
INDEX 293
McCawley, J.D. 118 names 4345 185239242243245-248252

McDowell, J. 200 253 259-261 264-268 270-277 280 281
meaning ix 1 2 6 13-15 1620 38 39 46 286
51 525758102103105112116117120 co designative 242 243 247 252272 275
123-139 140 141 147 149 155 163 164 277, interchangeability of 244 247
196 200 201 211-218 220 273 274 276 254275
280284-287 coreferential 241 280
autonomy ofiinguistic 13 general 245 285
behavioural doctrine of 1 2 genuine 271 272
centre 103 104 105 interchangeable 241-244 246-248 252
change 285 253 272 273 280
picture theory of 199 201 ordinary.271 273
probability theory of 212 213 214 217 proper, see proper names
sameness of 285 'temporal rigidity' of 272
stimulus 279 280 natural kinds 153 265 269
theory of 1 42 51 52 132 134-137 139 natural language ix 37 41 4245 49 51 52
201 218 576066739496117 119 143 150 151
use theory of 199 155 162207 213 216
mental state 181-185 187-189 191 196 naturalistic fallacy argument, see idealistic
metalanguage 31 333435 106207211220 fallacy argument
223224 necessity
might-have-beens 229 237, see also epistemic 241, and metaphysical 273
counterfactuals, may-bes 281
Mill, J.S. 3 239-244247 248 270-273 285 Nemirow, L. 66
modal Neumann, J. von 217
argument 243 nominalization 50
context 241-244 247 270 272 273, normal speaker 249 251
rigidity in 248
value 241 242 object 23-25 30-32 96 142 149 151 181-
modality 189210212218
epistemic 241 language 106 211 217
modes of presentation 255 "objective" idiom 266
modus tollens argument 258 262 observation language 26
Montague, R. 37 38 5268 113 open-endedness 142 143
mood 91016 17192044107162164 open texture 141-143 149-151 222
connection with theory of truth 9 15 operational test 285
indicator 15 16 operator 15 18 192250 139 164 194 205
of sentences 9-15, imperative 1114164, epistemic 82
indicative 10 11 12 14 18 21 22 162 ordinary language 61 93 94 101 108 113
164 195, interrogative 11 164, 287
optative 21 164 philosopher 57 60
relation to use 10 11 14 15
setter 18 19 20 paradigm 146 147 285 286
theory of 14 22 Fregean 286
Moore, G.E. 192209 Parson, T. 172
294 INDEX
Paul, H. 102 248266279

Peacocke, C. 272 content 45 184191 192195240
perception 25 26 30 31 33 34 163 204205 Putnam, H. 43 49 53144151152171 218
207 219 220 221 223 225 264 278 284
causal theory of 205
performance 119 149 quantifier 50 57 58 59 63 64 67-71 75-86
performatives 14 16 1841 8990 194212
perlocutionaryaims 194 195 Quine, W.V. 93 94 95100108113117132
Plantinga, A. 273 134 138 140 152 160 172217240 251
Pope, A. 52 266 267 272 278-281 284
possibilities 229 231 236, see also may-bes quotation 274
possibility direct and indirect 266 267
epistemic and metaphysical 243
possible worlds 44 50 143 144 146 150220 Ramsey, F.P. 39
pragmatic oddity 146-148 Raskin, V. 116 120
pragmatically relevant 165 rational inference 232
pragmatics 37-41 46 148 159 164 rationality
presupposition 37 53116120173 maxim of 48 49
descriptive 274 principles of 474849 51
presuppositional analysis 116 realism 201 208212213 217 218 219 221
Principle of Benefit of the Doubt 284 222228
Principle of Charity 280 284 reductionism 223
principle of effective means 47 48 53 reference 3 910159096130139155181
progression principle 74 75 87-89 185 188201202208211214-218221
projection 147 148 230 244-248 260 261 266 271 ,.276
projective operations 66 284-287
pronominalization 65 67-72 75-77 81 83- community-wide 244 245
91 rigid 244, rigidly fixing 273 274 276 280
proper names 240 244 246 247 261 262 281 287
264 265 267 270-272 276 278 280 285 theory of ix 199 284
codesignative 242 243 253 267 269 272, referent 262 269 271 273-276
interchangeability of 242-244 253 Reichenbach, H. 95 113 151 199201212-
coreferential 241 251 214286
'logically' 271 272 278 281 relevance 110 127 153-159 161 163-168
modally rigid 244 247 273 276 170-172 175-180
property 240 248 259 262 271 274 280 pragmatic 155 156 165 166 175 176 177
defining 262 278 179180
identifying 260-263 270 278 semantic 155 156 165 166 175 176 177
pure 262 179 180
proposition 161734155-157175181183 relevant
185 188 195 233 240 241 243 245 248 marginally 159-161 172
272 maximally 166
intensional 181 potentially 159 160
propositional topically 159-163 172
attitude 6 21 22 182 188 189 243 244 reliability of learning 204-208 211
INDEX 295
representation 44 45 134 183 184 185 187 epistemic 281

188189190193195197 Geachean 273
system of 190 objective 277
representative content 185-188 190 192 private and public 247
Riemann hypothesis 123 sense/force distinction 131
Rolston, H.L. 141 148 152 sentence 1-379-11 1617-2021 39 40 42
Ross, J.R. 118 4592101 116124-129134136137140
rule 164171 181 189200201 213 215218
constitutive 38 39 50 149 220 221
epistemic 81 cognitive equivalence of 3-8
generative 91 complex 126 140220
heuristic 167 169 171 177-179 eternal 93 98 102 103 105 106 117
recursive 91 140 observation 3 8 31 32 33 113 217
Russell, B. 3790212240242244-247251 occasion 3 4 5 6 7 8 21 93 102 105 111
252259270-272 274 277 278 281 113
class 253 standing 3 5 7 93
Ryle, G. 39 51 53 152 Shoemaker, S. 25 30
sign 41 161
satisfaction Simmons, R.F. 120
conditions of 191 Skolem function 63
Schnelle, H. 116-121 172 Smaby, R. 113
science fiction 143 144 145 149 150 space-time region 95 96
scientific language 93 101 speech acts ix 13 16 18 37 38 41 43-4547
Searle, J. 41 43 48 52 53 221 274 53155157170175176184190-194196
Sellars, W. 24 29 30 31 32 see also illocutionary acts
semantic Stafford, J. 48
deviancy 146 147 148 Stalnaker, R.C. 53
force 164 165 state of affairs 181-186 188 189 191 195
modality 164 201 207 213 214
network 103-106 108 III 112 120 statements ix 141 206210221
recursion 116 intensional 181 184
theory 287 288 synthetic 274
semantics ix 1 24816172021374246 see also counteridentity-, counternomic
515861659192106116117120148 statements
150 151 155 159 162 164 175 180 196 stereotype 200 215 216
205208210218-221223-227 stimulation 2-6 8 38
Fregean 286 stimulatory situation 1 3 4
game theoretical 58-60 63 stimulus-response 32
possible worlds 141 143 150 strategy 61-68 72 75 76 78 83 8689167
sense 2 9 10 15 16 125 128-132 137-139 Strawson, P.F. 21 37 53
185 200 221 243 244 246 248 252 271 subgame 63 64 72-87 90
274-278282 substitutivity 251 268-270273 281
community-wide 244 doctrine 273 .
conventional 244 280 principle 251 253 269 272 273 281
descriptive 270 proof/disproof of 268
296 INDEX
success (oflanguage) 202 219 221 necessary 241-243 269

synonym 1 2 6 7 286 tables 29 129224
cognitive 6 7 theory of 9 1542106107139199220
partial 2 value 3141617202195118120141
synonymy 208 222224227228 240241 280
cognitive 5 6 7 type
interchangeability criterion of 6 meaning as 127 134
relation of 2 5 153
syntax 37 46 92 UlIian, I.S. 113
systematic import 41 Ullmann-Margalit, E. 151
understanding 9394 112 124 127 133-135
Tarski, A. 9 42 162205 211 220227277 138 144-146 212 226 227
terms complete (of sentences) 117 118
dispositional 142 metaphorical 145
general 279 semantical 145
natural kind 50 142 262 264 268 277 strategies of 146
279280 syntactical 144 145
singular 271 281 282 theory of 199-201211 214216218-220
underinterpreted 145 223225-227
Thomason, R. 52 113 universals
time 94 101 108-112 116-118 120237 formal and substantive 41 46
time line 94 95 111 112 universe of discourse
time point 94-96 objects 96 97 101
translation 8 106 172 202 207 208 210 228 occasions 97 98 101
251-254 263 265 266 269 275 277-281 unrestricted comprehension schema 253
287 254
homophonic 251-253 266 268 269 279 usage 39 57
287 use ix 1 29 10384041 454649-525758
indeterminacy of 266 267 102112123-125 131 132 134-138 199-
manual 132 202212214216218224227237
practice of 286 and connection with mood 14 15
principle of 250 251 253 263 267 276 theories of 374251
purpose of 287 use/mention confusion 187 188
systems of 266 utterance 2 3 9-12 15-21 40 93 101-103
truth 9 124 126-131 134 135 137 139 150 105 107-109 120 124 125 127 128 131
201 203 206-211 216 220 221 223 224 132 146-148 156 159-165 167 170 172
226227228 176 178 181 192 194 195
a priori 25 31 33 204 circumstance of 94 95 100
conditions 19 21 30 51 125 128-131 conditions 200
134 135 137 138 141 143 146 184
188 189 201 210 211 213 215-218 vagueness 141 142 158 225
224225238 Van Dijk, T.A. 113
contingent 241 242 Vennemann, T. 113
correspondence theory of 199 210 221 verifiability 153
definition 205 207 217 223 224 verification 139 209 220 226 227
INDEX 297
verificationism 141 211-214 226 Wittgenstein, L. 39 4152125126128-132

134 137 138 148 149 152 182 183 190
Waismann, F. 141 142 151 152 199201 211 215
Warnock, G.J. 49 53 Wunderlich, D. 113
warranted assertibility 208 209 216 226
Weinstein, S. 113 Zemach, E.M. 31-35
Wilson, N. 280
SYNTHESE LIBRARY
Studies in Epistemology,
Logic, Methodology, and Philosophy of Science
Managing Editor:
JAAKKO HINTIKKA (Academy of Finland and Stanford University)
Editors:
ROBERT S. COHEN (Boston University)

DONALD DAVIDSON (University of Chicago)
GABRI~L NUCHELMANS (University of Leyden)
WESLEY C. SALMON (University of Arizona)
1. J.M. Bochenski, A Precis of Mathematical Logic. 1959, X + 100 pp.

2. P.L. Guiraud, Problemes et methodes de la statistique linguistique. 1960, VI + 146 pp.
3. Hans Freudenthal (ed.), The Concept and the Role of the Model in Mathematics and
Natural and Social Sciences. Proceedings of a Colloquium held at Utrecht, The Nether-
lands, January 1960. 1961, VI + 194 pp.
4. Evert W. Beth, Formal Methods. An Introduction to Symbolic Logic and the Study of
Effective Operations in Arithmetic and Logic. 1962, XIV + 170 pp.
5. B.H. Kazemier and D. Vuysje (eds.), Logic and Language. Studies Dedicated to
Professor Rudolf Carnap on the Occasion ofHis Seventieth Birthday. 1962, VI + 256 pp.
6. Marx W. Wartofsky (ed.), Proceedings of the Boston Colloquiumfor the Philosophy of
Science 1961-1962, Boston Studies in the Philosophy of Science (ed. hy Robert
S. Cohen and Marx W. Wartofsky), Volume I. 1963, VIII + 212 pp.
7. A.A. Zinov'ev, Philosophical Problems of Many-Valued Logic. 1963, XIV + 155 pp.
8. Georges Gurvitch, The Spectrum of Social Time. 1964, XXVI + 152 pp.
9. Paul Lorenzen, Formal Logic. 1965, VIII + 123 pp.
10. Robert S. Cohen and Marx W. Wartofsky (eds.), In Honor of Philipp Frank, Boston
Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W. War-
tofsky), Volume II. 1965, XXXIV + 475 pp.
11. Evert W. Beth, Mathematical Thought. An Imroduction to the Philosophy of Ma-
thematics. 1965, XII + 208 pp.
12. Evert W. Beth and Jean Piaget, Mathematical Epistemology and Psychology. 1966,
XII + 326 pp.
13. Guido Kling, Ontology and the Logistic Analysis of Language. An Enquiry into the
Contemporary Views on Universals. 1967, XI + 210 pp.
299
300
14. Robert S. Cohen and Marx W. Wartofsky (eds.), Proceedings of the Boston Collo-
quiumfor the Philosophy of Science 1964---1966, in Memory of Norwood Russell Hanson,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
W. Wartofsky), Volume 111. 1967, XLIX + 489 pp.
15. C.D. Broad, Induction, Probability, and Causation. Selected Papers. 1968, XI + 296
pp.
16. GUnther Patzig, Aristotle's Theory of the Syllogism. A Logical-Philosophical Study of
Book A of the Prior Analytics. 1968, XVII + 215 pp.
17. Nicholas Rescher, Topics in Philosophical Logic. 1968, XIV + 347 pp.
18. Robert S. Cohen and Marx W. Wartofsky (eds.), Proceedings of the Boston Colloquium
for the Philosophy of Science 1966-1968, Boston Studies in the Philosophy of Science
(ed. by Robert S. Cohen and Marx W. Wartofsky), Volume IV. 1969, VIII + 537 pp.
19. Robert S. Cohen and Marx W. Wartofsky (eds.), Proceedings of the Boston Colloquium
for the Philosophy of Science 1966-1968, Boston Studies in the Philosophy of Science
(ed. by Robert S. Cohen and Marx W. Wartofsky), Volume V. 1969, VIII + 482 pp.
20. J.W. Davis, D.J. Hockney, and W.K. Wilson (eds.), Philosophical Logic. 1969, V11r +
277 pp.
21. D. Davidson and J. Hintikka (eds.), Words and Objections: Essays on the Work of
W. V. Quine. 1969, VIII + 366 pp.
22. Patrick Suppes, Studies in the Methodology and Foundations of Science. Selected Papers
from 1911 to 1969. 1969, XII + 437 pp.
23. Jaakko Hintikka, Models for Modalities. Selected Essays. 1969, IX + 220 pp.
24. Nicholas Rescher et al. (eds.), Essays in Honor of Carl G. Hempel. A Tribute on the
Occasion of His Sixty-Fifth Birthday. 1969, VII + 272 pp.
25. P.V. Tavanec (ed.), Problems of the Logic of Scientific Knowledge. 1969, XII + 429 pp.
26. Marshall Swain (ed.), Induction, Acceptance, and Rational Belief. 1970, VII + 232 pp.
27. Robert S. Cohen and Raymond J. Seeger (eds.), Ernst Mach: Physicist andPhilosopher,
W. Wartofsky), Volume VI. 1970, VIII + 295 pp.
28. Jaakko Hintikka and Patrick Suppes, Information and Inference. 1970, X + 336 pp.
29. Karel Lambert, Philosophical Problems in Logic. Some Recent Developments. 1970,
VII + 176 pp.
30. Rolf A. Eberle, Nominalistic Systems. 1970, IX + 217 pp.
31. Paul Weingartner and Gerhard Zecha (eds.), Induction, Physics, and Ethics: Proceed-
ings and Discussions of the 1968 Salzburg Colloquium in the Philosophy of Science.
1970, X + 382 pp.
32. Evert W. Beth, Aspects of Modern Logic. 1970, XI + 176 pp.
33. Risto Hilpinen (ed.), Deontic Logic: Introductory and Systematic Readings. 1971,
VII + 182 pp.
34. Jean-Louis Krivine, Introduction to Axiomatic Set Theory. 1971, VII + 98 pp.
35. Joseph D. Sneed, The Logical Structure of Mathematical Physics. 1971, XV + 311 pp.
36. Carl R. Kordig, The Justification of Scientific Change. 1971, XIV + 119 pp.
37. MiIi~ Capek, Bergson and Modern Physics, Boston Studies in the Philosophy of
Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume VII. 1971,
XV + 414pp.
301
38. Norwood Russell Hanson, What I Do Not Believe, and Other Essays (ed. by Stephen
Toulmin and Harry Woolf). 1971, XII + 390 pp.
39. Roger C. Buck and Robert S. Cohen (eds.), PSA 1970. In Memory of RudolfCarnap,
W. Wartofsky), Volume VIII. 1971, LXVI + 615 pp. Also available as paperback.
40. Donald Davidson and Gilbert Harman (eds.), Semantics of Natural Language. 1972,
X + 769 pp. Also available as paperback.
41. Yehoshua Bar-Hillel (ed.), Pragmatics of Natural Languages. 1971, VII + 231 pp.
42. Soren Stenlund, Combinators, A-Terms and Proof Theory. 1972, 184 pp.
43. Martin Strauss, Modern Physics and Its Philosophy. Selected Papers in the Logic,
History, and Philosophy of Science. 1972, X + 297 pp.
44. Mario Bunge, Method, Model and Matter. 1973, VII + 196 pp.
45. Mario Bunge, Philosophy of Physics. 1973, IX + 248 pp.
46. A.A. Zinov'ev, Foundations of the Logical Theory of Scientific Knowledge (Complex
Logic), Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and
Marx W. Wartofsky), Volume IX. Revised and enlarged English edition with an
appendix, by G.A. Smirnov, E.A. Sidorenka, A.M. Fedina, and L.A. Bobrova. 1973,
XXII + 301 pp. Also available as paperback.
47. Ladislav Tondl, Scientific Procedures, Boston Studies in the Philosophy of Science
(ed. by Robert S. Cohen and Marx W. Wartofsky), Volume X. 1973, XII + 268 pp.
Also available as paperback.
48. Norwood Russell Hanson, Constellations and Conjectures (ed. by Willard C. Hum-
phreys, Jr.). 1973, X + 282 pp.
49. K.J.J. Hintikka, J.M.E. Moravcsik, and P. Suppes (eds.), Approaches to Natural
Language. Proceedings of the 1970 Stanford Workshop on Grammar and Semantics.
1973, VIII + 526 pp. Also available as paperback.
50. Mario Bunge (ed.), Exact Philosophy - Problems, Tools, and Goals. 1973, X + 214
pp.
51. Radu J. Bogdan and Ilkka Niiniluoto (eds.), Logic, Language, and Probability. A
Selection of Papers Contributed to Sections IV, VI, and XI of the Fourth International
Congress for Logic, Methodology, and Philosophy of Science, Bucharest, September
1971. 1973, X + 323 pp.
52. Glenn Pearce and Patrick Maynard (eds.), Conceptual Change. 1973, XII + 282 pp.
53. Ilkka Niiniluoto and Raimo Tuomela, Theoretical Concepts and Hypothetico-
Inductive Inference. 1973, VII + 264 pp.
54. Roland FraYsse, Course of Mathematical Logic - Volume 1: Relation and Logical
Formula. 1973, XVI + 186 pp. Also available as paperback.
55. Adolf Griinbaum, Philosophical Problems of Space and Time. Second, enlarged edition,
W. Wartofsky), Volume XII. 1973, XXIII + 884 pp. Also available as paperback.
56. Patrick Suppes (ed.), Space, Time, and Geometry. 1973, XI + 424 pp.
57. Hans Kelsen, Essays in Legal and Moral Philosophy, selected and introduced by Ota
Weinberger. 1973, XXVIII + 300 pp.
58. R.J. Seeger and Robert S. Cohen (eds.), Philosophical Foundations of Science. Pro-
ceedings of an AAAS Program, 1969, Boston Studies in the Philosophy of Science (ed.
302
by Robert S. Cohen and Marx W. Wartofsky), Volume XI. 1974, X + 545 pp. Also
available as paperback.
59. Robert S. Cohen and Marx W. Wartofsky (eds.), Logical and Epistemological Studies
in Contemporary Physics, Boston Studies in the Philosophy of Science (ed. by Robert
S. Cohen and Marx W. Wartofsky), Volume XIII. 1973, VIII + 462 pp. Also available
as paperback.
60. Robert S. Cohen and Marx W. Wartofsky (eds.), Methodological and Historical Essays
in the Natural and Social Sciences. Proceedings of the Boston Colloquium for the
Philosophy of Science 1969-1972, Boston Studies in the Philosophy of Science (ed.
by Robert S. Cohen and Marx W. Wartofsky), Volume XIV. 1974, VIII + 405 pp.
61. Robert S. Cohen, J.J. Stachel and Marx W. Wartofsky (eds.), For Dirk Struik.
Scientific, Historical and Political Essays in Honor of Dirk J. Struik, Boston Studies
in the Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky),
Volume XV. 1974, XXVII + 652 pp. Also available as paperback.
62. Kazimierz Ajdukiewicz, Pragmatic Logic, trans]. from the Polish by Olgierd
Wojtasiewicz. 1974, XV + 460 pp.
63. Soren Stenlund (ed.), Logical Theory and Semantic Analysis. Essays Dedicated to Stig
Kanger on His Fiftieth Birthday. 1974, V + 217 pp.
64. Kenneth F. Schaffner and Robert S. Cohen (eds.), Proceedings of the 1972 Biennial
Meeting, Philosophy of Science Association, Boston Studies in the Philosophy of
Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume XX. 1974, IX +
444 pp. Also available as paperback.
65. Henry E. Kyburg, Jr., The Logical Foundations of Statistical Inference. 1974, IX + 421
pp.
66. Marjorie Grene, The Understanding of Nature: Essays in the Philosphy of Biology,
W. Wartofsky), Volume XXIII. 1974, XII + 360 pp. Also available as paperback.
67. Jan M. Broekman, Structuralism: Moscow, Prague, Paris. 1974, IX + 117 pp.
68. Norman Geschwind, Selected Papers on Language and the Brain, Boston Studies in
the Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume
XVI. 1974, XII + 549 pp. Also available as paperback.
69. Roland Fralsse, Course of Mathematical Logic - Volume 2: Model Theory. 1974,
XIX + 192 pp.
70. Andrzej Grzegorczyk, An Outline of Mathematical Logic. Fundamental Results and
Notions Explained with All Details. 1974, X + 596 pp.
71. Franz von Kutschera, Philosophy of Language. 1975, VII + 305 pp.
72. Juha Manninen and Raimo Tuomela (eds.), Essays on Explanation and Understanding.
Studies in the Foundations ofHumanities and Social Sciences. 1976, VII + 440 pp.
73. Jaakko Hintikka (ed.), Rudolf Carnap, Logical Empiricist. Materials and Perspectil'es.
1975, LXVIII + 400 pp.
74. Milil: Capek (ed.), The Concepts of Space and Time. Their Structure and Their Deve-
lopment, Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and
Marx W. Wartofsky), Volume XXII. 1976, LVI + 570 pp. Also available as paper-
back.
303
75. Jaakko Hintikka and Unto Remes, The Method of Analysis. Its Geometrical Origin
and Its General Significance, Boston Studies in the Philosophy of Science (ed. by
Robert S. Cohen and Marx W. Wartofsky), Volume XXV. 1974, XVIII + 144 pp.
76. John Emery Murdoch and Edith Dudley Sylla, The Cultural Context of Medieval
Learning. Proceedings of the First International Colloquium on Philosophy, Science,
and Theology in the Middle Ages - September 1973, Boston Studies in the Philosophy
of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume XXVI. 1975,
X + 566 pp. Also available as paperback.
77. Stefan Amsterdamski, Between Experience and Metaphysics. Philosophical Problems
of the Evolution of Science, Boston Studies in the Philosophy of Science (ed. by
Robert S. Cohen and Marx W. Wartofsky), Volume XXXV. 1975, XVIII + 193 pp.
78. Patrick Suppes (ed.), Logic and Probability in Quantum Mechanics. 1976, XV + 541
pp.
79. Hermann von Helmholtz: Epistemological Writings. The Paul Hertz/Moritz Schlick
Centenary Edition of1921 with Notes and Commentary by the Editors. (Newly translated
by Malcolm F. Lowe. Edited with an Introduction and Bibliography, by Robert
S. Cohen and Yehuda Elkana), Boston Studies in the Philosophy of Science (ed. by
Robert S. Cohen and Marx W. Wartofsky), Volume XXXVII. 1977, XXXVIII + 204
pp. Also available as paperback.
80. Joseph Agassi, Science in Flux, Boston Studies in the Philosophy of Science (ed. by
Robert S. Cohen and Marx W. Wartofsky), Volume XXVIII. 1975, XXVI + 553 pp.
81. Sandra G. Harding (ed.), Can Theories Be Refuted? Essays on the Duhem-Quine
Thesis. 1976, XXI + 318 pp. Also available as paperback.
82. Stefan Nowak, Methodology of Sociological Research: General Problems. 1977,
XVIII + 504 pp.
83. Jean Piaget, Jean-Blaise Grize, Alina Szeminska, and Vinh Bang, Epistemology and
Psychology of Functions, Studies in Genetic Epistemology, Volume XXIII. 1977,
XIV + 205 pp.
84. Marjorie Grene and Everett Mendelsohn (eds.), Topics in the Philosophy of Biology,
W. Wartofsky), Volume XXVII. 1976, XUI + 454 pp. Also available as paperback.
85. E. Fischbein, The Intuitive Sources ofProbabilistic Thinking in Children. 1975, XIII +
204 pp.
86. Ernest W. Adams, The Logic of Conditionals. An Application of Probability to Deduc-
tive Logic. 1975, XIII + 156 pp.
87. Marian Przel",cki and Ryszard W6jcicki (eds.), Twenty-Five Years of Logical Metho-
dology in Poland. 1977, Vlll + 803 pp.
88. J. Topolski, The Methodology of History. 1976, X + 673 pp.
89. A. Kasher (ed.), Language in Focus: Foundations, Methods and Systems. Essays
Dedicated to Yehoshua Bar-Hillel, Boston Studies in the Philosophy of Science (ed.
by Robert S. Cohen and Marx W. Wartofsky), Volume XLIII. 1976, XXVIII + 679
304
90. Jaakko Hintikka, The Intentions ofIntentionality and 0 ther New Modelsfor Modalities
1975, XVIII + 262 pp. Also available as paperback.
91. Wolfgang Stegmiiller, Collected Papers on Epistemology, Philosophy of Science and
History of Philosophy, 2 Volumes. 1977, XXVII + 525 pp.
92. Dov M. Gabbay, Investigations in Modal and Tense Logics with Applications to
Problems in Philosophy and Linguistics. 1976, XI + 206 pp.
93. Radu J. Bogdan, Local Induction. 1976, XIV + 340 pp.
94. Stefan Nowak, Understanding and Prediction: Essays in the Methodology of Social and
Behavioral Theories. 1976, XIX + 482 pp.
95. Peter Mittelstaedt, Philosophical Problems of Modern Physics, Boston Studies in the
Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume
XVIII. 1976, X + 211 pp. Also available as paperback.
96. Gerald Holton and William Blanpied (eds.), Science and Its Public: The Changing
Relationship, Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen
and Marx W. Wartofsky), Volume XXXIII. 1976, XXV + 289 pp. Also available as
paperback.
97. Myles Brand and Douglas Walton (eds.), Action Theory. Proceedings of the Winnipeg
Conference on Human Action, Held at Winnipeg, Manitoba, Canada, 9-11 May 1975.
1976, VI + 345 pp.
98. Risto Hilpinen, Knowledge and Rational Belief 1979 (forthcoming).
99. R.S. Cohen, P.K. Feyerabend, and M.W. Wartofsky (eds.), Essays in Memory of
Imre Lakatos, Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen
and Marx W. Wartofsky), Volume XXXIX. 1976, XI + 762 pp. Also available as
paperback.
100. R.S. Cohen and J.J. Stachel (eds.), Selected Papers ofLeon Rosenfeld, Boston Studies
in the Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky),
Volume XXI. 1978, XXX + 927 pp.
101. R.S. Cohen, C.A. Hooker, A.C. Michalos, and J.W. van Evra (eds.), PSA 1974:
Proceedings of the 1974 Biennial Meeting of the Philosophy of Science Association,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
Wartofsky), Volume XXXII. 1976, XIII + 734 pp. Also available as paperback.
102. Yehuda Fried and Joseph Agassi, Paranoia: A Study in Diagnosis, Boston Studies in
L. 1976, XV + 212 pp. Also available as paperback.
103. Marian Przel~cki, Klemens Szaniawski, and Ryszard W6jcicki (eds.), Formal Methods
in the Methodology of Empirical Sciences. 1976, 455 pp.
104. John M. Vickers, Belief and Probability. 1976, VIII + 202 pp.
105. Kurt H. Wolff, Surrender and Catch: Experience and Inquiry Today, Boston Studies in
LI. 1976, XII + 410 pp. Also available as paperback.
106. Karel Kosik, Dialectics of the Concrete, Boston Studies in the Philosophy of Science
(ed. by Robert S. Cohen and Marx W. Wartofsky), Volume LII. 1976, VIII + 158
107. Nelson Goodman, The Structure of Appearance, Boston Studies in the Philosophy of
Science (ed. by Ro bert S. Cohen and MarxW. Wartofsky), Volume LIII.1977,L + 285 pp.
305
108. Jerzy Giedymin (ed.), Kazimierz Ajdukiewicz: The Scientific World-Perspective and
Other Essays, 1931-1963. 1978, LIII + 378 pp.
109. Robert L. Causey, Unity of Science. 1977, VIII + 185 pp.
110. Richard E. Grandy, Advanced Logic for Applications. 1977, XIV + 168 pp.
111. Robert P. McArthur, Tense Logic. 1976, VII + 84 pp.
112. Lars Lindahl, Position and Change: A Study in Law and Logic. 1977, IX + 299 pp.
113. Raimo Tuomela, Dispositions. 1978, X + 450 pp.
114. Herbert A. Simon, Models of Discovery and Other Topics in the Methods of Science,
Wartofsky), Volume L1V. 1977, XX + 456 pp. Also available as paperback.
115. Roger D. Rosenkrantz, Inference, Method and Decision. 1977, XVI + 262 pp. Also
116. Raimo Tuomela, Human Action and Its Explanation. A Study on the Philosophical
Foundations of Psychology. 1977, XII + 426 pp.
117. Morris Lazerowitz, The Language of Philosophy. Freud and Wittgenstein, Boston
Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
Wartofsky), Volume LV. 1977, XVI + 209 pp.
118. Tran Duc Thao, Origins of Language and Consciousness, Boston Studies in the
Philosophy of Science (ed. by Robert S. Cohen and Marx. W. Wartofsky), Volume
LVI. 1979 (forthcoming).
119. Jerzy Pelc, Semiotics in Poland, 1894-1969. 1978, XXVI + 504 pp.
120. Ingmar Porn, Action Theory and Social Science. Some Formal Models. 1977, X + 129
pp.
121. Joseph Margolis, Persons and Minds. The Prospects of Nonreductive Materialism,
Wartofsky), Volume LVII. 1977, XIV + 282 pp. Also available as paperback.
122. Jaakko Hintikka, Ilkka Niiniluoto, and Esa Saarinen (eds.), Essays on Mathe-
matical and Philosophical Logic. Proceedings of the Fourth Scandinavian Logic
Symposium and of the First Soviet-Finnish Logic Conference, Jyviiskylii, Finland,
1976. 1978, VIII + 458 pp. + index.
123. Theo A. F. Kuipers, Studies in Inductive Probability and Rational Expectation.
1978, XII + 145 pp.
124. Esa Saarinen, Risto Hilpinen, Ilkka Niiniluoto, and Merrill Provence Hintikka
(eds.), Essays in Honour of Jaakko Hintikka on the Occasion of His Fiftieth
Birthday. 1978, IX + 378 pp. + index.
125. Gerard Radnitzky and Gunnar Andersson (eds.), Progress and Rationality in
Science, Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen
and Marx W. Wartofsky), Volume LVIII. 1978, X + 400 pp. + index. Also
126. Peter Mittelstaedt, Quantum Logic. 1978, IX + 149 pp.
127. Kenneth A. Bowen, Model Theory for Modal Logic. Kripke Models for Modal
Predicate Calculi. 1978, X + 128 pp.
128. Howard Alexander Bursen, Dismantling the Memory Machine. A Philosophica
Investigation of Machine Theories of Memory. 1978, XIII + 157 pp.
SYNTHESE HISTORICAL LIBRARY
Texts and Studies

in the History of Logic and Philosophy
Editors:
N. KRETZMANN (Cornell University)

G. NUCHELMANS (University of Leyden)
L.M. DE RIJK (University of Leyden)
I. M.T. Beonio-Brocchieri Fumagalli, The Logic of Abelard. Translated from the Italian.
1969, IX + 101 pp.
2. Gottfried Wilhelm Leibniz, Philosophical Papers and Letters. A selection translated and
edited, with an introduction, by Leroy E. Loemker. 1969, XII + 736 pp.
3. Ernst Mally, Logische Schriften, ed. by Karl Wolf and Paul Weingartner. 1971, X +
340 pp.
4. Lewis White Beck (ed.), Proceedings of the Third International Kant Congress. 1972,
XI + 718 pp.
5. Bernard Bolzano, Theory of Science, ed. by Jan Berg. 1973, XV + 398 pp.
6. J.M.E. Moravcsik (ed.), Patterns in Plato's Thought. Papers Arising Out of the 1971
West Coast Greek Philosophy Conference. 1973, VIII + 212 pp.
7. Nabil Shehaby, The Propositional Logic of Avicenna: A Translation from al-Shifa:
al-Qiyas, with Introduction, Commentary and Glossary. 1973, XIII + 296 pp.
8. Desmond Paul Henry, Commentary on De Grammatico: The Historical-Logical
Dimensions of a Dialogue of St. Anselm's. 1974, IX + 345 pp.
9. John Corcoran, Ancient Logic and Its Modern Interpretations. 1974, X + 208 pp.
10. E.M. Barth, The Logic of the Articles in Traditional Philosophy. 1974, XXVII + 533 pp.
11. Jaakko Hintikka, Knowledge and the Known. Historical Perspectives in Epistemology.
1974, XII + 243 pp.
12. E.J. Ashworth, Language and Logic in the Post-Medieval Period. 1974, XIII + 304 pp.
13. Aristotle, The Nicomachean Ethics. Translated with Commentaries and Glossary by
Hypocrates G. Apostle. 1975, XXI + 372 pp.
14. R.M. Dancy, Sense and Contradiction: A Study in Aristotle. 1975, XII + 184 pp.
15. Wilbur Richard Knorr, The Evolution of the Euclidean Elements. A Study of the Theory
of Incommensurable Magnitudes and Its Significance for Early Greek Geometry. 1975,
IX + 374 pp.
16. Augustine, De Dialectica. Translated with Introduction and Notes by B. Darrell
Jackson. 1975, XI + 151 pp.
306
SYNTHESE LANGUAGE LIBRARY
Texts and Studies

in Linguistics and Philosophy
Managing Editors:
JAAKKO HINTIKKA (Academy of Finland and Stanford University)

STANLEY PETERS (The University of Texas at Austin)
Editors:
EMMON BACH (University of Massachusetts at Amherst)

JOAN BRESNAN (Massachusetts Institute of Technology)
JOHN LYONS (University of Sussex)
JULIUS M.E. MORAVCSIK (Stanford University)
PATRICK SUPPES (Stanford University)
DANA SCOTT (Oxford University)
1. Henry Hit (ed.), Questions. 1977, XVII + 366 pp.

2. William S. Cooper, Foundations of Logico-Linguistics. A Unified Theory of Information,
Language, and Logic. 1978, XVI + 249 pp.
3. Avishai Margalit (ed.), Meaning and Use. Papers Presented at the Second Jerusalem
Philosophical Encounter, April 1976. 1979 (forthcoming).
4. F. Guenthner and S.J. Schmidt (eds.), Formal Semantics and Pragmatics for Natural
Languages. 1978, VIlI + 374 pp. + index.
5. Esa Saarinen (ed.), Game-Theoretical Semantics. 1978, XIV + 300 pp. + index.
6. F.J. Pelletier (ed.), Mass Terms: Some Philosophical Problems. 1978, XIV + 379 pp. +
index.
307

(Studies in Linguistics and Philosophy. Synthese Language Library Volume 3) Avishai Margalit (Ed.) - Meaning and Use-D. Reidel Publishing Co. & The Magnes Press (1979)

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

(Studies in Linguistics and Philosophy. Synthese Language Library Volume 3) Avishai Margalit (Ed.) - Meaning and Use-D. Reidel Publishing Co. & The Magnes Press (1979)

Diunggah oleh

Hak Cipta:

Format Tersedia

TO THE MEMORY OF

TEXTS AND STUDIES IN

LINGUISTICS AND PHILOSOPHY

JAAKKO HINTIKKA, Academy of Finland and Stanford University

EMMON BACH, University of Massachusetts at Amherst

D. REIDEL PUBLISHING COMPANY

THE MAGNES PRESS, THE HEBREW UNIVERSITY

Jerusalem Philosophical Encounter, 2d, 1976.

(Synthese language library; v. 3)

Published by D. Reidel Publishing Company.

Sold and distributed in the U.S.A., Canada, and Mexico

Sold and distributed in Israel

All Rights Reserved

W. V. QUINE / Use and Its Place in Meaning

The second Jerusalem Philosophical Encounter was held in Jerusalem on

The Hebrew University of Jerusalem A.M.

to Yehoshua Bar-Hillel than this meeting of outstanding scholars who gather

The Hebrew University of Jerusalem Amos Tversky

USE AND ITS PLACE IN MEANIN G

The notion of meaning is stubborn. It does not submit readily to satisfactory

the behavioral doctrine of meaning contributes is theoretical. it purports to

comparing the stimulatory conditions for the volunteering of sentences, it

verdicts under somewhat unlike stimulations.

receptors. Next there is the relation of cognitive equivalence of occasion

MOODS AND PERFORMANCES

Frege held that an adequate account of language requires us to attend to

is no clue to the seriousness of a command that it is uttered in the imperative

in assertion. It is easy to see that merely speaking a sentence in the strength-

cannot say is, I have urged, that there is no such convention.

"Will you take off your shoes?" have a common element.

I am using these words merely to refer to a sentence or the proposition it

other part of speech. This was because he thought of it as a part of a longer

The University of Chicago

1 For a recent statement of my position, see "Radical Interpretation," Dialectica 27

hat' as in effect 'It is desirable that your hat be on,' hence:

How does a subject become aware of some (e.g., external) objects? An

character or typical features in PMG-GO. It is not the mysterious, dia-

The Hebrew University of Jerusalem

1. In his paper Zemach attempts to draw a distinction between two

However, initial positions can be arrived at in ways other than moves in

3. To specify that a position in the language game L is an initial position

The Hebrew University of Jerusalem

WHAT IS A THEORY OF USE?

I. INTRODUCTORY: BAR-HILLEL AND PRAGMATICS

Yehoshua Bar-Hillel was one of the pilgrim fathers of Pragmatics of Natural

II. NEW GOALS FOR THEORIES OF LANGUAGE USE

According to the traditional characterization of Pragmatics, an investigation

computer engineering outside the confines of arithmetic. Pragmatics is the

The pragmatic competence is that of using linguistic means for certain

III. ADEQUACY CONDITIONS: UNITY VS. VARIETY

Approaches to a theory of language use are of two different kinds, to be

(CA) For every context of utterance C, every sentence S of a natural

if and only if,

I shall call this adequacy condition 'Criterion A' of appropriateness.

(T) "Yored geshem" is true-in-Hebrew when spoken by x at time

IV. SPEECH AND ACTS: THE CASE OF RATIONALITY

It is one of the scandals of recent philosophy of language that the philoso-

retical achievements of the theory of conversational implicatures are derivable

suggestive but vague super-maxim of relation, commending "relevance."

v. USE AND MEANING: THE CASE OF IMPERFECT INFORMATION

Turning now to a refutation of the interactive view, according to which there

tikka's language-games with rules that would apply to indefinite expressions

The present proposal carries an important observation with respect to