YEHOSHUA BAR-HILLEL
SYNTHESE LANGUAGE LIBRARY
Managing Editors:
Editorial Board:
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
ISBN 14-020-3263-3
Printed in Israel
TO THE MEMORY OF
YEHOSHUA BAR-HILLEL
TABLE OF CONTENTS
PREFACE IX
ADDRESS XI
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
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
USE AND ITS PLACE IN MEANING 5
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
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
... 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
MOODS AND PERFORMANCES 13
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,
MOODS AND PERFORMANCES 17
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
MOODS AND PERFORMANCES 19
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.
NOTES
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
A. Margalit (ed.), Meaning and Use, 21-22. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
22 W. V. QUINE
Harvard University
EDDY M. ZEMACH
AWARENESS OF OBJECTS
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
AWARENESS OF OBJECTS 27
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
AWARENESS OF OBJECTS 29
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.
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.
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.
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.
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:
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
WHAT IS A THEORY OF USE? 45
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
WHAT IS A THEORY OF USE? 47
the chaff, I would like to discuss two examples in some detail and point out
the emerging intermediate view.
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
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
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
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
A. Margalit (ed.), Meaning and Use, 57-92. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
58 JAAKKO HINTIKKA AND LAURI CARLSON
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).
60 JAAKKO HINTIKKA AND LAURI CARLSON
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
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)
CONDITIONALS, GENERIC QUANTIFIERS AND SUBGAMES 63
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
64 JAAKKO HINTIKKA AND LAURI CARLSON
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-
CONDITIONALS, GENERIC QUANTIFIERS AND SUB GAMES 65
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
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
CONDITIONALS, GENERIC QUANTIFIERS AND SUB GAMES 69
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.
70 JAAKKO HINTIKKA AND LAURI CARLSON
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:
72 JAAKKO HINTIKKA AND LAURI CARLSON
(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.
CONDITIONALS, GENERIC QUANTIFIERS AND SUB GAMES 73
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.
74 JAAKKO HINTIKKA AND LAURI CARLSON
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)
I win = I win G (V if X 1
CONDITIONALS, GENERIC QUANTIFIERS AND SUB GAMES 75
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
76 JAAKKO HINTIKKA AND LAURI CARLSON
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
CONDITIONALS, GENERIC QUANTIFIERS AND SUBGAMES 77
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).
78 JAAKKO HINTIKKA AND LAURI CARLSON
(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)
my
strategy~ 7 _/
I win =
G (Y) I win G (unless Y, Z)
Naturels~
strategy ~- ~
my
strategy 7 /
Nature wins =
Nature wins G(unlessY,Zl
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.
80 JAAKKO HINTIKKA AND LAURI CARLSON
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
but not
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) ).
82 JAAKKO HINTIKKA AND LAURI CARLSON
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
84 JAAKKO HINTIKKA AND LAURI CARLSON
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)
CONDITIONALS, GENERIC QUANTIFIERS AND SUB GAMES 85
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
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
92 JAAKKO HINTIKKA AND LAURI CARLSON
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
93
A. Margalit (ed.), Meaning and Use, 93-115. All Rights Reserved.
Copyright <CI 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
94 HELMUT SCHNELLE
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
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-
CIRCUMSTANCE SENTENCES 99
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.
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.
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.
CIRCUMSTANCE SENTENCES 103
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
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.
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
CIRCUMSTANCE SENTENCES 109
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
CIRCUMSTANCE SENTENCES 111
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
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.
CIRCUMSTANCE SENTENCES 115
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
A. Margalit (ed.), Meaning and Use, 116-122. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
COMMENTS 117
Suffer
/
Tense: Past
/
Napoleon
\ Defeat
~ Waterloo
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
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
A. Margalit (ed.), Meaning and Use, 123-135. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
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
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
THE APPEAL TO USE AND THE THEORY OF MEANING 127
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
THE APPEAL TO USE AND THE THEORY OF MEANING 129
COMMENTS
I
assertion of sentence SENTENCE I meaning is use
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.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
COMMENTS 137
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
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
OPEN TEXTURE
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.
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
146 AVISHAI MARGALIT
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
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
148 AVISHAI MARGALIT
(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.
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.
NOTES
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
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
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
CONVERSATIONAL RELEVANCE 159
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
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
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.'
CONVERSATIONAL RELEVANCE 171
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.
NOTES
\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
COMMENTS
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
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). \
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
A. Margalit (ed.), Meaning and Use, 181-197. All Rights Reserved.
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
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.
INTENTIONALITY AND THE USE OF LANGUAGE 187
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,
INTENTIONALITY AND THE USE OF LANGUAGE 189
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
INTENTIONALITY AND THE USE OF LANGUAGE 191
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
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.
NOTES
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
A. Margalit (ed.), Meaning and Use, 199-217. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland
200 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.
REFERENCE AND UNDERSTANDING 203
(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.
(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
REFERENCE AND UNDERSTANDING 207
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
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
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
REFERENCE AND UNDERSTANDING 213
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
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
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
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
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
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
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
A. Margalit (ed.), Meaning and Use, 226-228. All Rights Reserved.
Copyright 1979 by D. Reidel Publishing Company, Dordrecht, Holland.
REPLY TO DUMMETT'S COMMENT 227
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
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
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
MAY BES AND MIGHT HAVE BEENS 235
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
MAY BES AND MIGHT HAVE BEENS 237
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
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
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
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
A PUZZLE ABOUT BELIEF 245
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
A PUZZLE ABOUT BELIEF 247
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.
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):
In this way we can apply the dis quotational technique to all languages.
A PUZZLE ABOUT BELIEF 251
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
A PUZZLE ABOUT BELIEF 253
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.
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
A PUZZLE ABOUT BELIEF 257
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-
A PUZZLE ABOUT BELIEF 259
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
A PUZZLE ABOUT BELIEF 261
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
A PUZZLE ABOUT BELIEF 263
'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
A PUZZLE ABOUT BELIEF 265
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
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
Princeton University
A PUZZLE ABOUT BELIEF 271
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
A PUZZLE ABOUT BELIEF 273
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
A PUZZLE ABOUT BELIEF 275
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
A PUZZLE ABOUT BELIEF 277
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
A PUZZLE ABOUT BELIEF 279
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.
A PUZZLE ABOUT BELIEF 281
(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
A PUZZLE ABOUT BELIEF 283
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
A. Margalit (ed.), Meaning and Use, 284-288. All Rights Reserved.
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
289
290 INDEX
(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
Studies in Epistemology,
Logic, Methodology, and Philosophy of Science
Managing Editor:
JAAKKO HINTIKKA (Academy of Finland and Stanford University)
Editors:
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
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.
Also available as paperback.
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
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.
Also available as paperback.
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.
Also available as paperback.
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.
Also available as paperback.
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx
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
pp. Also available as paperback.
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
the Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume
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
the Philosophy of Science (ed. by Robert S. Cohen and Marx W. Wartofsky), Volume
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
pp. Also available as paperback.
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
Wartofsky), Volume L1V. 1977, XX + 456 pp. Also available as paperback.
115. Roger D. Rosenkrantz, Inference, Method and Decision. 1977, XVI + 262 pp. Also
available as paperback.
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,
Boston Studies in the Philosophy of Science (ed. by Robert S. Cohen and Marx W.
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
available as paperback.
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
Editors:
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
Managing Editors:
Editors:
307