Also published in this series

Moral Concepts edited by Joel Feinberg
Theories 0/ Ethics edited by Philippa Foot
Knowledge and Belief, edited by A. PhiUips Griffiths
The Philosophy 0/ Mathematics edited by Jaakko Hintikka
The Philosophy 0/ Science edited by P. H. Nidditch
The Theory 0/ Meaning edited by G. H. R. Parkinson
Political Philosophy edited by Anthony Quinton
Philosophical Logic edited by P. F. Strawson
The Philosophy 0/ Perception edited by G. J. Warnock
The Philosoph)' 0/ Action edited by Alan R. White
Other volumes are in preparation
THE PHILOSOPHY
OF LANGUAGE
Edited by
J. R. SEARLE
OXFORD UNIVERSITY PRESS
1971
Oiford University Press, Ely House, London W.1
GLASGOW NEW YORK TORONTO MELBOURNE WELLINGTON
CAPE TOWN SALISBURY IBADAN NAIROBI DAR ES SALAAM LUSAKA ADDlS ABABA
BOMBAY CALCUTTA MADRAS KARACHI LAHORE DACCA
KUALA LUMPUR SINGAPORE HONG KONG TOKYO
© OXFORD UNIVERSITY PRESS 1971
FILMSET AND PRINTED IN GREAT BRITAIN
BY BUTLER & TANN ER LID., FROME AND LONDON
CONTENTS
INTRoDucnoN
I. PERFORMATIVE-CONSfATIVE
by J. L. Austin
II. INTENTION AND CONVENTION IN SPEECH Acrs
by P. F. Strawson
IIl. WHAT Is A SPEECH ACT?
by J. R. Searle
IV. UTTERERS MEANING, SENTENCE-MEANING, AND WORD-MEANING
by H. P. Grice
V. TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
(a) Assumptions and Goals
(b) The Theory of Transformational Generative Grammar
by Noam Chomsky
VI. THE PmLOSOpmCAL RELEVANCE OF LINGUISTIC THEORY
by Jerrold J. Katz
VII. SYMPOSIUM ON INNATE IDEAS
(a) Recent Contributions to the Theory of Innate Ideas
by Noam Chomsky
(b) The 'Innateness Hypothesis' and Explanatory Models in
Linguistics
by Hilary Putnam
(c) The Epistemological Argument
by Nelson Goodman
NoTES ON THE CON1RIBUTORS
BIBUOGRAPHY
INDEX OF NAMES
13
23
39
54
71
83
101
121
130
140
145
146
149
INTRODUCTION
I T is important to distinguish between the philosophy of language and linguistic
philosophy. Linguistic philosophy consists in the attempt to solve philosophical
problems by analysing the meanings of words, and by analysing logical relations
between words in natural languages. This may be done in order to solve such
traditional philosophical problems as those concerning determinism, scepticism, and
causation; or it may be done without special regard to traditional problems but as an
investigation of concepts for their own interest, as an inquiry into certain aspects of
the world by scrutinizing the classifications and distinctions we make in the language
we use to characterize or describe the world. The philosophy oflanguage consists in
the attempt to analyse certain general features of language such as meaning, reference,
truth, verification, speech acts, and logical necessity.
'The philosophy of language' is the name of a subject matter within philosophy;
'linguistic philosophy' is primarily the name of a philosophical method. But the two,
method and subject, are intimately connected. Both because some of the problems in
the philosophy of language can be profitably attacked by the methods of linguistic
philosophy (the problems concerning the nature of truth, for example, may be
regarded, at least in part, as questions about the analysis of the concept 'true') amI,
more importantly, because the methods linguistic philosophers employ in conducting
linguistic analyses depend crucially on their philosophy of language. The way a
linguistic philosopher conducts an actual investigation will depend on certain general
conceptions he has concerning how words mean and how they relate to the world. It
is only given some general theory of or approach to language that one can even get
started on a particular linguistic analysis. It is for this reason--among others--that
in the spread of analytical philosophy in the twentieth century the philosophy of
language has occupied such a central, some would say the central, place in the entire
enterprise of philosophy. Most of the influential philosophers of this century, for
example, RusselI, Wittgenstein, Carnap, Quine, Austin, and Strawson, have been in
varying degrees philosophers of language.
Though both the philosophy of language and linguistic philosophy are pursued
nowadays with more self-consciousness than ever before, both are in fact as old as
philosophy. When in the Euthyphro Plato asks what is piety, he may be regarded as
asking a question concerning the concept pious, and this, most contemporary
philosophers would claim, may be regarded as a question concerning the exact
meaning of the Greek word for pious, 'hosion', and its synonyms in other languages.
When in the Phaedo he advances the theory that general terms get their meaning by
2 THE PHILOSOPHY OF LANGUAGE
standing for the Forms he is advancing a thesis in the philosophy of language, a
thesis about how words mean.
In spite of a historical continuity stretching back to the Greeks, there are good
reasons for dating the beginnings of modern philosophy of language in the work of
the German philosoph er and mathematician, Gottlob Frege. Frege wanted to show
that mathematics was reducible to and founded on logic. To carry out this enter-
prise he had to invent a new logical system. In the course of his logical investigations
he found the prevailing nineteenth-century views about language--represented most
influentially by J. S. Mill---to be rather superficial and in many cases downright
false, so he was forced to develop a theory of his own.
In order to place contemporary work in the philosophy of language in some
historical perspective, and to give the reader a general idea of the background to the
essays in this volume, I shall now-all too briefly---sketch some of the more im-
portant developments in the philosophy of language, beginning with Frege. What I
say will be nothing more than a rough outline, and the reader should not consider the
simplicity of the account as reflecting any simplicity in the actual deve10pment of
this subject.
Frege's most impOItant single discovery in the philosophy of language was the
distinction between sense and reference. He elucidated this distinction in terms of the
following puzzle about identity statements: How is it possible for a true statement of
the form, a is identical with b, to contain any more factual information than a state-
ment of the form, ais identical with a? 1 Ifwe construe the statements as being about
the objects referred to by the names which replace 'a' and 'b', it seems that they must
say exactly the same uninformative thing, namely, that an object is identical with
itself. If, on the other hand, we construe identity statements as about the names used
in making them then it seems they must be arbitrary, since we can arbitrarily assign
any name to any object we like. Yet obviously the statement 'The Evening Star is the
Morning Star' does not mean the same as, and is more factually informative than,
the statement 'The Evening Star is the Evening Star'. How is this possible? Frege's
answer is that in addition to the name and the object it refers to, viz. its reference,
there is a third element, its sense (or as we rnight prefer to say in English: the meaning
or descriptive content) of the name in virtue of which and only in virtue of which it
refers to its reference. The sense provides the 'mode of presentation' of the object,
and referring to a reference is always achieved by way of sense. The reason why 'The
Evening Star is the Morning Star' can be more factually informative than 'The
Evening Star is the Evening Star', even though the reference is the same, is that the
sense of 'The Evening Star' is different from the sense of'The Morning Star', and the
statement conveys that one and the same object has the features specified in the
different senses of the two expressions. Frege regards this theory of sense and refer-
I Frege, G., 'Sense and Reference', Philosophical Wrilin!?s. trans. P. T. Geach and M. Black (Blackwell,
Oxford, 1952).
INTRODUCTION 3
ence as applying not only to definite descriptions like 'the man in a blue shirt' but also
to ordinary proper names like 'Chicago' and 'Winston Churchill'.
Frege then extends this distinction from singular referring expressions to predicate
expressions and to whole sentences. He says that in addition to expressing their
sense, predicate expressions refer to concepts, and he says of sentences (at least those
where the question oftruth or falsity arises) that they express a thought as their sense,
and have a truth value (i.e. the circumstance that they are true or the circumstance
that they are false) as their reference. This extension of the distinction between sense
and reference to predicates and whole senten ces is less compelling and has been
historically less influential than the original distinction for referring expressions. To
my rnind it loses the most brilliant insight of the original distinction, an insight which
reveals the connection between reference and truth: namely that an expression refers
to an object only because it conveys something true of that object. But a predicate
does not convey something true of a concept nor does a sentence convey something
true of a truth value. But one important feature of Frege's account of sentences has
survived. Frege says that we must distinguish between the thought expressed by
a sentence and the assertion of that thought. Thus, for example, the same thought
is expressed by the sentence 'Socrates is wise' and 'Is Socrates wise?', and again, this
same thought is expressed in the antecedent clause of'If Socrates is wise then he is a
philosopher'. But only in the first of these three sentences is that thought asserted.
This distinction between the thought or content (or as most philosophers nowadays
call it, the proposition) and the assertion of that proposition is important for the
arguments in my article and in Grice's article in this volume.
After Frege the next great work in the philosophy of language was done by Russell
in the years before the First W orld War and carried on by his student Wittgenstein in
the Tractatus Logico-Philosophicus. Both Russell and Wittgenstein, for different sets
of reasons, rejected the distinction between sense and reference. They thought that it
might appear to work for simple cases, but that under a really rigorous analysis of
language into its most elementary forms the relation between words and the world
would turn out to be different from Frege's account of it. I regard the rejection of the
theory of sense and reference oby these two as a major mistake, and the reasons they
give for rejecting it seem to me bad reasons.
2
However, having rejected it they
developed a theory of how words relate to the world which was quite different from
Frege's.
RusselP begins by considering a problem posed by sentences which contain definite
descriptions with no corresponding object, for example, 'The King of France is bald.'
This is obviously a meaningful sentence, but the puzzle is: how can it be meaning-
ful since there is no King of France, hence there is nothing for the proposition
1 See J. R. Searle. 'Russell's Rejection of Frege's Theory of Sense and Reference', Analysis. 1958, for a
critical exarnination of Russell's reasons.
"On Denoting', Mind (1905).
4 THE PHILOSOPHY OF LANGUAGE
expressed by the senten ce to be about, hence there is nothing for the predicate to
be either true or false of? How can a sentence be meaningful, if apparently the
proposition it expresses is neither true nor false? Frege's answer was that a sentence
could have sense even though the subject expression had no reference. The sentence
might lack a truth value but the lack of a truth value does not render the sentence
meaningless. To suppose mistakenly that it does is simply to confuse sense and
reference. But Russen, having already rejected Frege's theory of sense and reference,
gave the following different answer : the sentence appears to be of subject -predicate
form but is not really so. Its grammatical form is misleading as to its logical form, and
its logical form is that of a conjunction of statements--one of which is an existential
statement. Under analysis the sentence really says:
There is a King of France.
There is not more than one King of France.
Whatever is King of France is bald.
In this interpretation we are able to see that the original sentence is meaningful, and
the proposition expressed by it is false. The proposition is false since there is no King
of France.
,
It is important to see the difTerence between Frege and Russell here. Frege asks:
how do proper names and definite descriptions refer to things? His answer is: in
virtue of their sense. Russen rejects the question. He argues, in efTect, that neither
definite descriptions nor ordinary proper names (since ordinary proper names are
just disguised or abbreviated definite descriptions for hirn) refer at all. Rather, he says,
they have no meaning 'in isolation', but sentences containing them are to be ana-
lysed on the model of the analysis of sentences containing the expression 'The King
of France', an analysis in which they no longer occur in their original form at alt. Allthat
is left are predicates, logical constants, and expressions such as 'there is a', 'some-
thing', 'nothing', 'anything', 'whatever', etc., none ofwhich refer to particular objects.
But how do words hook on the the world for Russen, if definite description and
ordinary proper names do not re/er? Part of his answer is that there is a class of
expressions of unanalysable simplicity, which are the logically proper names. They
simply stand for entities without having any sense or meaning other than standing for
entities. This conception is most fully worked out by Wittgenstein in the Tractatus.
In the Tractatus the point of contact between words and the world is provided by
names. 'The name me ans the object. The object is its meaning.'4 An elementary
sentence of a language is simply an arrangement or concatenation of names, and
since all meaningful sentences of a language are, for Wittgenstein, ultimately analys-
able into truth functional combinations of the elementary sentences, these names play
the crucial role of relating words to things in the world. But if, as Wittgenstein says,
elementary sentences are arrangements of names, how does a sentence come to have
4 Tractatus, 3.203.
INTRODUCTION 5
asense? How does it differ from, for example, a list ofnames? Wittgenstein's answer
is that the arrangement of names in the sentence constitutes a picture of a fact. Dif-
ferent ways of arranging names in sentences make the sentences different pictures of
facts, and the sense of a sentence is provided by this picturing relationship. If there
actually is a fact such as the picture represents, the sentence will be true; if not, the
sentence will be false. The picturing relationship may be completely conventional.
Thus in order to" say that x is. on top of y, one does not have to place the letter 'x' on
top of the letter 'y'; one can, by convention, write 'x' followed by 'is on top of'
followed by'y'. Here, 'is on top of' is not another name but is part of a conventional
way of picturing the relationship between x and y when x is on top of y. The fact that
words are arranged in a sentence in a certain way pictures the fact that things are
arranged in the world in a certain way.
This version of the picture theory of meaning encounters enormous difficulties, and
in Wittgenstein's later work, the Philosophical Investigations, he rejects it com-
pletely. One difficulty is this: if the meaning of a name is literally the object for which
it stands, then it seems the existence of these objects cannot be a matter of ordinary
contingent fact. The reason for this is that the destruction of any contingently exist-
ing object such as a mountain or a car cannot destroy the meaning of any words,
because any change in contingent features of the world must still be describable in
words. Indeed, it seems it would make no sense to assert or deny the existence of the
objects named by names. As Wittgenstein says, 'Objects make up the substance of
the world' and 'Substance is what exists independently ofwhat is the case'.5 Unfortu-
nately he never gives us any examples of objects or names or e1ementary sentences
but says that sentences of ordinary language must be analysable into elementary
sentences if they are to be fully meaningful. Paradoxically though-and this is a
disadvantage from which Frege's philosophy of language did not suffer--ordinary
names like 'Winston Churchill' and 'San Francisco' are not on his account 'names',
and ordinary objects are not 'objects' in his sense.
One of the aims of the Tractatus was to deli mit the realm of meaningful from
meaningless discourse. In spite of its obscurity it exerted a profound influence in the
1920s and 30s, especially on the logical positivists, a group of empirically oriented
philosophers interested in science and mathematics. They developed, partly from the
Tractatus and partly from other remarks of Wittgenstein, a criterion of meaningful-
ness, the Verification Principle. This principle states that the meaning of a proposition
is its method of verification, or put less obscurely: all meaningful statements are
either analytic on the one hand or empirical and synthetic on the other. All analytic
truths are true be definition, whereas all true empirical statements are true in virtue
of the sense experiences which verify them; indeed, they are equivalellt in meaning
to the sets of basic propositions recording sense experiences which would constitute
their verification.
, Traclalus 2.021, 2.024.
6 THE PHILOSOPHY OF LANGUAGE
Among the many objections made to the Verification Theory of Meaning is this:
What is the status of the Verification Principle itself? Surely the principle is not
synthetic, for it cannot be just a contingent empirical trait of meaningful utterances
that they are verifiable. But if the principle is analytic, is it not just an arbitrary defini-
tion of meaningfulness, which one may or may not choose to accept? And, indeed, it
appears to be quite inconsistent with our ordinary definitions of 'meaningful' and
'meaningless'. Or is the verification principle itse1f meaningless, like the metaphysics
it was intended to destroy? These difficulties, along with that of finding aversion of
the Verification Principle that could be precise1y stated, continued to plague the
positivists in the years before the Second World War.
Notice that all these theories which I have sketched-Frege, the Tractatus, logical
positivism---have certain characteristics in common. They all assume that the only,
at any rate the primary, aim of language is to represent and communicate factual
information, that the part of language that really counts is the 'cognitive' part. The
aim of language, in short, is to communicate what can be true or false. Secondly, they
treat the elements of language-words, sentences, propositions---as things that
represent or things that are true or false, etc., apart from any actions or intentions of
speakers and hearers. The elements of the language, not the actions and intentions of
the speakers are what count. In the late thirties and especially after the Second W orld
War these assumptions came to be vigorously challenged, especially by Wittgenstein.
Wittgenstein argued that stating facts is only one of the countless jobs we do with
language and that the meaning of elements lies not in any relationship they have in the
abstract but in the use we make ofthem. 'Language', says Wittgenstein, 'is an instru-
ment'; 6 and 'for a large dass of cases, ... the meaning of a word is its use in the
language'.7
Austin, in a similar vein, called attention to a dass of utterances that do not even
set out to be true or false. For ex am pie, a man who says, 'I prornise I will come', is
not stating a fact about himself to the effect that he promises, but rather he is making
a prornise. Austin called such utterances 'performatives', contrasting them with
'constatives' .
There was a good deal of carelessness in the way philosophers in this period talked
about the use of expressions and-with the notable exception of Austin---they did
not always distinguish among the different sorts of 'use' to which expressions could
be put. Hut this approach has certain real advantages which have survived to the
present day and which influence several of the artides in this volume. Most im-
portantly it recasts the discussion of many problems in the philosophy of language
into the larger context of the discussion of human action and behaviour generally.
Questions of meaning and reference can be seen in the context of speakers meaning
something by something and referring to something in the utterance of an expression.
, Philosophical Investigations. para. 569.
1 Op. ci!., para. 43.
INTRODUCTION 7
Instead of seeing the relations between words and the world as something existing
in vacua, one now sees them as involving intentional actions by speakers, employing
conventional devices (words, sentences) in accordance with extremely abstract sets of
rules for the use of those devices. For example, the real strength of Frege's theory
of sense and reference as opposed to Russell's theory of definite descriptions emerges
in Strawson's conception of reference as a speech act. 8 Once one sees referring as an
action that is performed in the utterance of an expression with a particular sense
provided by the rules for the use of the expression, then it is easier to see that it is
subject to the sorts of error that plague actions generally (one can fail to refer to a
king of France for the same reason that one can fail to hit a king of France: there is
no such person), and on this ac count there is much less motivation for trying to
identify referring (one kind of speech act), with asserting an existential proposition
(quite another kind of speech act), as Russell in effect does.
Which brings us to the artides in this volume. Austin once expressed the beliefthat
the work being done by philosophers, grammarians, psychologists, linguists, etc., in
this century would eventually result in a science of language in much the same way
that nineteenth- and early twentieth-century work in logic by people of various disci-
plines 'resulted in a science of logic. We have by no means yet achieved that result,
but the articles in this volume, I hope, will indicate something of the direction from
which I think it may come. They divide roughly into two sections. The first four
artides centre around the notion of speech acts and the relation of meaning to speech
acts; the rest deal with the implications of generative grammar.
In the first four artides we proceed historically from Austin's attempts to revise
the performative-constative distinction to current research on speech acts. Austin
in 'Performative-Constative' attacks the very distinction he originally made famous,
that between performative and constative utterances. But he attacks it from an in-
teresting angle. The original distinction was supposed to be a distinction between
utterances which are sayings (statements, descriptions, etc.) and utterances which are
doings of some other kinds (e.g. promises, bets, warnings). It is supposed to be a
distinction between utterances which are not acts (constatives), and utterances which
are acts (performatives). But, as Austin argues in this artide and his posthumously
published lectures, How Ta Do Things with Words, constatives turn out to be speech
acts as weil. Making a statement or giving a description is as much performing a
speech act as making a promise or giving an order. So wh at were originally presented
as special cases of utterances (performatives) now are seen to swallow the general
cases (constatives), which turn out to be only one dass of speech acts among others.
Austin, in his subsequent terminology, calls these different kinds of complete speech
acts (statements, bets, warnings, promises, etc.) illocutionary acts, and contrasts
them with acts which involve achieving certain effects on hearers such as persuading,
'Strawson, P. F., 'On Referring', Mind, 1950, reprinted A. Flew (ed.), Essays in Conceplual Analysis
(Macmillan, London, 1956).
8 THE PHILOSOPHY OF LANGUAGE
convincing, frightening, boring, amusing, annoying, which he calls perlocutionary
acts.
Both Strawson and I, in quite ditTerent ways and with differing objectives, attempt
to interpret Austin 's account of the notion of an illocutionary act in terms of Grice's
analysis of meaning. Grice analyses meaning as intending to produce an etTect in a
hearer by getting him to recognize the intention to produce that effect.
9
Strawson
probes the extent to which illocutionary acts are matters of convention and intention
respectively in light of Austin's claim that illocutionary acts are essentially con-
ventional. He concludes that some acts are indeed essentially conventional in that
they require extra-Iinguistic conventions for their performance. For example a
marriage ceremony, calling someone 'out' in agame, bidding at bridge, etc., are all
essentially conventional in that they require some extra-Iinguistic conventions for
their performances; but the great bulk of fundamental types of illocutionary acts,
statements, questions, requests, etc., for Strawson are not conventional in that or any
other than the trivial sense that they can be performed with conventional devices and
they have conventional names. The contrast between the two types of cases emerges
when we see that the non-conventional acts will be successfully performed if the
'complex overt intention' of the speaker is grasped by the hearer, that is if he
understands. But the overt intended etTect of the speech act may still not be achieved
without any breach of rules or conventions. With the conventional act, on the other
hand, any failure of the speaker to achieve his overt intention must be attributable to
some breach of rule or convention. In the one case therefore (the conventional case)
the explicit performative form may be the name of the very act which is performed
if and only if the speaker's overt intention is etTective; but in the other case (the non-
conventional case) it cannot be the name of this act.
I think this distinction does not have the force that Strawson attributes to it, and
that it, as weil as Grice's original account of meaning, suffers from a failure to
appreciate adequate1y Austin's distinction between illocutionary uptake (i.e. under-
standing the utterance) and perlocutionary etTect. Strawson and Grice both think of
the 'overt intention' of the speaker in the non-conventional case as the eliciting of
some response or etTect in the hearer, such as, e.g., getting hirn to believe something
(the overt intended effect of statements) or getting him to do something (the overt
intended etTect ofrequests). But, I wish to claim, the intended etTect ofmeaning some-
thing is that the hearer should know the illocutionary force and propositional content
of the utterance, not that he should respond or behave in such and such ways. I hint
at this point in my article in this volume, and I argue for it more fully in Speech
Acts.
IO
Here, for the sake of brevity, I state it rather dogmatically.
9 Grice, H. P., 'Meaning', Philosophie'al Re,·;ew. 1957, reprinted P. F. Strawson (ed.), Philosophical
Logic (Oxford University Press, London, 1967).
IUSearle, J. R., Speech AclS. An Essay in lhe Philosophy 0/ LanKuage (Cambridge University Press,
London and New York, 1969), pp. 42-50.
INTRODUCTION 9
Once that point is seen, then the difTerence Strawson cites does not appear so
great. Compare making a statement (non-conventional) and bidding at bridge (con-
ventional). In both cases the intended effect of meaning something is that the hearer
should understand the utterance. In the bidding case this is facilitated by some
linguistic conventions to the efTect that certain bids are to be made under certain
conditions. In both cases the intended perlocutionary efTect may not be achieved,
e.g. the hearer may not believe my statement, or he may not believe I have as many
high cards as my five no-tmmp bid would indicate. And in neither case does he breach
any mies or conventions if he fails to believe me. Furthermore, even in those cases
where failure to sec ure the perlocutionary efTect does indeed involve a breach of
mies, this is only for the trivial reason that the mies are designed to enforce the
perlocutionary efTect. For example, if I fail to p e r f ~ r r n my duties under a contract I
do indeed destroy the intended perlocutionary efTect of the contract and thus violate
the law. But this coincidence involves no deep linguistic point, it merely illustrates
that the law in this case is designed to ensure the perlocutionary efTect once the
illocutionary performance has been satisfactorily done.
In both the statement case and the contract case the performative verb is the name
of the act which is performed if the overt intention is efTective, because in the one case
the overt intention is to make a statement and in the other it is to make a contract.
To suppose otherwise is to suppose what 1 am alleging is amistake, namely that the
overt intentions of meaning something are perlocutionary.
Furthermore-and this is a quite separate point---the question of to wh at extent
the 'nonconventional acts' of statement making, etc., are conventional is not con-
clusively settled. For both Grice and Strawson the basic speech acts are not
essentially conventional at all. They believe that it just so happens that we do have
linguistic conventions for achieving what are natural responses like beliefs and
actions. The picture presented in my article is that for some acts at least, e.g. state-
ments and promises, the acts can only be performed within systems of'constitutive'
mies and the particular linguistic conventions we have in particular naturallanguages
are simply conventional realizations ofthese underlying constitutive mies. This is one
of the most important unresolved controversies in contemporary philosophy of
language. It might be rephrased as folIows: to what extent can such basic illocu-
tionary notions as statement, request, promise, etc., be explicated without employing
such notions as mle or convention, but employing only such notions as intention,
response, procedure, etc.? The most promising work in the latter category is in
Grice's William James lectures, a portion ofwhich is published in this volume. In this
article he attempts to analyse certain fundamental features of language using only
'bmte' notions. My article attempts to explore the assumption that some types of
illocutionary acts are necessarily conventional, i.e. can only be performed given
certain constitutive mies. It is an assumption that I have by no means proven and
wh ich Grice may in the end succeed in disproving.
\0 THE PHILOSOPHY OF LANGUAGE
In Grice's artic1e he continues work on the project, undertaken in his 1957 artic1e,
of explaining what it is for a sentence to mean something, and what it is for a word
or phrase to mean something in terms of what it is for speakers to mean something
by doing something. The basic notion, as in the earlier artic1e, is the notion of a
speaker performing an action and meaning something by that performance. Here
Grice modifies his earlier definition of meaning in a way that partially, but not
entirely, avoids the objections I just made. He characterizes the intended effect of
imperative type utterances as being that the hearer should intend to do something
rather than that he should actually do it; and he characterizes the intended effect of
indicative type utterances as being, not that the hearer should believe something,
but that he should believe that the speaker believes something. But this characteriza-
tion still seems to me inadequate, for several reasons, of which here are two. First,
it seems to me that a speaker may utter sentences, me an what he utters and still not
have these intentions. For example, I may utter 'It's raining', mean U's raining by
that utterance, and still not care a hang whether my hearer believes that I believe it's
raining. The intended effect on the hearer of meaning it's raining when I say 'It's
raining' and mean it (as opposed to just uttering 'It's raining' as a pronunciation
practice, say) is that the hearer should know that he has been told (i.e. the statement
has been made to hirn) that it is raining. The intended effect of meaning, in short, is
understanding, which is an illocutionary not a perlocutionary effect.
Secondly, is is not c1ear how these definitions of meaning would discriminate
among the many different kinds of illocutionary acts. For example, how on these
definitions do we tell the difference between TB do it', meant as a promise, and
meant as a statement of intention? Both according to Grice involve the intention to
produce in the hearer the belief that the speaker believes he will do the act, but this
is not adequate to distinguish the two different meant illocutionary forces.
The last three sections are concerned with transformational generative grammar
and its importance for philosophy. Such grammars are called 'generative' because
they contain systems of rules for generating an infinite set of sentences and thus
attempt to account for the ability of speakers of natural languages to produce and
understand an indefinite number of completely new utterances, utterances of
sentences they have never heard before. They are called transformational, because
they contain transformational as weil as phrase-structure rules. That is, they contain
rules which transform the structure of sentences by rearranging or deleting elements
as weil as rules which structure sentences into such phrases as the noun phrase and
the verb phrase. Indeed, one of the early achievements of this type of linguistic theory
was its demonstration of the inadequacy of phrase-structure rules alone to account
for the syntactical complexity of natural human languages.
Chomsky's work, of which I am attempting to provide philosophical readers a
glimpse with the selections inc1uded in this volume, does not lend itself to the kind
of piecemeal philosophical sniping that I have been making at the other authors
INTRODUCTION 11
because of its global and systematic character. It is important, however, to em-
phasize that from a philosopher's point of view a great deal is assumed and a great
deal is left unclear. For example, Chomsky teils us as one of his 'uncontroversial'
assumptions that 'the competence of the speaker-hearer can ideally be expressed as a
system of rules that relate signals to semantic interpretations of those signals'. But
that mastering competence in a language is a matter of mastering a system of rules
as opposed to, say, acquiring a set of dispositions to behave, is a controversial--()r
at any rate challengeable--assumption. It happens to be a view I hold, but it is not an
obvious truth. Furthermore, when Chomsky teils us that the child who learns the
language acquires an 'internal representation' ofthese rules, it is not totally clear what
such an internal representation is supposed to be. 1s it supposed to mean that he has
'tacit knowledge' of the rules in the sense of Polanyi? 11 I rather think so. But the notion
of an internal representation would, as far as Chomsky teils us, allow us to interpret
it in other ways, e.g. as implying only that the idealized linguistic competence of
the speaker-hearer can be described by these rules in the same way that the laws of
physics describe the idealized behaviour of a falling body. In neither case, on such an
interpretation, would we have to interpret the body or the speaker as knowinganyrules.
Katz applies the semantic component of generative grammar in an attempt to solve
certain traditional problems in the philosophy of language. It is not always clear that
he solves as much as he claims to solve. For example, it is puzzling that he should
think his explication of the notion of analyticity in terms of the concepts of semantic
theory is an answer to Quine's doubts about analyticity, because Quine objects to any
account that relies on notions such as meaning or semantic rules. But Katz's
apparatus of semantic markers, projection rules, etc., is just as meaning-Iaden as
Carnap's notion of a semantic rule, which Quine was originally concerned to attack.
Katz presupposes intensional notions and does not explicate them in the extensional
and behavioural fashion that Quine would regard as acceptable.
Again Katz's remarks about logical form settle none of the outstanding issues
about logical form. The most serious contemporary philosophical dispute about
logical form concerns Russell's theory of descriptions. Russell, as we have seen,
maintains that sentences with adefinite description as a grammatical subject are
misleading as to their logical form. Grammatically, says RusselI, they are of subject-
predicate form, but their logical form is that of an existential sentence, not a subject-
predicate sentence at all. It is hard to see how a Katzian interpretation of this issue
would contribute anything at all to settling it except perhaps by begging the question
in favour of Strawson, since there is nothing in the underlying phrase marker of, say,
'The King of France is bald' to show the Russellian distinction between existential
and subject-predicate sentences.
The last section, a symposium between Chomsky, Putnam, and Goodman, con-
cerns the innateness hypothesis. Chomsky and his followers claim that information
11 Cf., e.g., Polanyi, M., Personal Knowledge (University of Chicago Press, 1960).
12 THE PHILOSOPHY OF LANGUAGE
currently available in empirical linguistic theory supports the traditional rationalist
conception of innate ideas, the theory that human beings have concepts implanted
in their minds prior to any experiences. The dispute over this claim breaks down into
two questions. First, a historicaI question : what did the rationaIists mean by their
doctrine of innate ideas and is it what Chomsky thinks they meant? And second, for
our purposes the more interesting question : what assumptions must we make about
the innate intellectual capacities of children to account for the way in which they
acquire languages? In this symposium at least it is my opinion that Chomsky
definitely has the better of the argument on the second question. Putnam and Good-
man offer arguments to counter some of the claims made by the innateness theorists,
but they do not come to grips with the more important of Chomsky's arguments. For
example Chomsky claims that the rules reIating deep and surface sentence structures
are so complex and so abstract as to defy any stimulus-response or associationist
account of their learning. Only by supposing that the child has the form of the
grammar programmed into his brain can we account for his intellectuaI feat in
acquiring these rules. This, at least in part, is an empiricaI issue: how complex and
abstract are the syntactical ruIes of naturallanguages? In order to answer his claims,
Chomsky's critics would have to show either that he is mistaken in his empiricaI
theory of the syntax of natural languages or that he has drawn the wrong conclusion
from that theory.
There are three main contemporary approaches to the philosophy of language :
the neo-positivist-symbolic logic approach represented most ably by Quine, the so-
called 'ordinary language' approach of Wittgenstein and Austin, and the generative
gram m ar approach of Chomsky and his followers. I think, as the seIections in this
volume indicate, that the future development of the subject is Iikely to come from
joining the two latter approaches.
I
PERFORMATIVE-CONSTATIVE
J. L. AUSTIN
Translator's Note: 'Performative-Constative' is a straightforward translation of Austin's paper
'Performatif·Constatif', which he wrote in French and presented at a (predominantly) Anglo-
French conference held at Royaumont in March 1958. The case ofthe discussion which follows
it I is somewhat more complex. The actuaI discussion at Royaumont was carried on in both
French and English. What appears in the published volume after Austin's text (Cahiers de
Royaumont, Philosophie No. IV, La Philosophie Analytique: Les Editions de Minuit, 1962,
pp. 27 I -304) is aversion of this, based on a transcript but substantiaIly cut and edited, in
which the contributions originaIly made in English were translated into French by M. Bera.
lt might have been possible, for the present publication, to procure copies at least of those
portions of the original transcript that were in English. However, it seemed to me preferable
simply to translate into English the entire French text, mainly for the reason that it is this
edited version, and this only, that aII those taking part are known to have seen and approved
for publication. G. J. WARNOCK
ONE can quite easily get the idea of the performative utterance-4hough the ex-
pression, as I am weil aware, does not exist in the French language, or anywhere else.
This idea was brought in to mark a contrast with that of the declarative utterance,
or rather, as I am going to call it, the constative utterance. And there we have straight
off what I want to call in question. Ought we to accept this Performative-Constative
anti thesis ?
The constative utterance, under the name, so dear to philosophers, of statemenr,2
has the property of being true or false. The performance utterance, by contrast, can
never be either: it has its own special job, it is used to perform an action. To issue
such an utterance
3
is to perform the action--an action, perhaps, which one scarcely
could perform, at least with so much precision, in any other way. Here are some
examples:
I name this ship Liberte.
I apologize.
I welcome you.
I advise you to do it.
From Philosophy and Ordinaty Language. edited by Charles E. Caton (University ofIllinois Press, Urbana,
1963), pp. 22-23. Reprinted by permission of Mrs. J. Austin, G. J. Warnock and the University of IIIinois
Press.
I [Not here reprinted. Ed.j
2 [The French term is 'assertion'. I am sure that 'statement' is the English term Austin would have used
here, and I have so translaled 'assertion' throughout. Trans.1
l ['Formuler un tel enonce'. The translation is supplied in a footnote by Austin himself. Trans.]
14 PERFORMA TIVE-CONSTA TIVE
Utterances of this kind are common enough: we find them, for instance, every-
where in wh at are called in English the 'operative' c\auses of a legal instrument.
4
Plainly, many of them are not without interest for philosophers: to say 'I promise
to .. .' --to issue, as we say, this performative utterance----:iust is the act of making
a promise ; not, as we see, at all a mysterious act. And it may seem at once quite
obvious that an utterance of this kind can't be true or false---notice that I say it can't
be true or false, because it may very weil imply that some other propositions are true
or are false, but that, if I'm not mistaken, is a quite different matter.
However, the performative utterance is not exempt from all criticism: it may very
weil be criticized, but in a quite different dimension from that of truth and falsity.
The performative must be issued in a situation appropriate in all respects for the act
in question : if the speaker is not in the conditions required for its performance (and
there are many such conditions), then his utterance will be, as we call it in general,
'unhappy' . 5
First, our performative, like any other ritual or ceremony, may be, as the lawyers
say, 'null and void'. If, for example, the speaker is not in a position to perform an act
of that kind, or if the object with respect to which he purports to perform it is not
suitable for the purpose, then he doesn't manage, simply by issuing his utterance, to
carry out the purported act. Thus a bigamist doesn't get married a second time, he
only 'goes through the form' of a second marriage; I can't name the ship ifI am not
the person properly authorized to name it; and I can't quite bring offthe baptism of
penguins, those creatures being scarcely susceptible of that exploit.
Second, a performative utterance may be, though not void, 'unhappy' in a different
way--if, that is, it is issued insincere(v. If I say 'I promise to .. .' without in the
least intending to carry out the promised action, perhaps even not believing that it is
in my power to carry it out, the promise is hollow. It is made, certainly; but still, there
is an 'unhappiness': I have abused the formula.
Let us now suppose that our act has been performed: everything has gone off
quite normally, and also, ifyou like, sincerely. In that case, the performative utterance
will characteristically 'take effect'. We do not mean by that that such-and-such a
future event is or will be brought abaut as an effect of this action functioning as a
cause. We mean rather that, in consequence of the performance of this act, such-and-
such a future event, if it happens, will be in order, and such-and-such other events,
ifthey happen, will not be in order. If I have said 'I promise', I shall not be in order
if I break my word; if I have said 'I welcome you', I shall not be in order ifI proceed
to treat you as an enemy or an intruder. Thus we say that, even when the performa-
tive has taken effect, there may always crop up a third kind of unhappiness, which we
'The clauses, that is to say, in which the legal act is actually perfonned, as opposed to those--the
'preamble'-which set out the circumstances of the transaction.
5 l'Unhappy' is a term Austin regularly used in this connection, and he supplies it himself in brackets after
the French 'malheureux'. Trans.]
PERFORMA TIVE-CONSTA TIVE 15
call 'breach of commitment'.6 We may note also that commitments can be more or
less vague, and can bind us in very different degrees.
There we have, then, three kinds of unhappiness associated with the performative
utterance. It is possible to make a complete classification of these unhappinesses;
but it must be admitted that, as practically goes without saying, the different kinds
may not always be sharply distinguishable and may even coincide.
7
Then we must
add that our performative is both an action and an utterance: so that, poor thing, it
can't help being liable to be substandard in all the ways in which actions in general
can be, as weil as those in which utterances in general can be. For example, the
performative may be issued under duress, or by accident; it may suffer from
defective grammar, or from misunderstanding; it may figure in a context not wholly
'serious', in a play, perhaps, or in apoern. We leave all that on one side--let us simply
bear in mi nd the more specific unhappiness ofthe performative, that is, nullity, abuse
(insincerity), and breach of commitment.
Weil, now that we have before us this idea of the performative, it is very natural
to hope that we could proceed to find some criterion, whether of grammar or of
vocabulary, which would make it possible for us to answer in every case the question
whether a particular utterance is performative or not. But this hope is, alas, exaggera-
ted and, in large measure, vain.
It is true that there exist two 'normal forms', so to speak, in which the performa-
tive finds expression. At first sight both of them, curiously enough, have a thoroughly
constative look. One of these normal forms is that which I have already made use of
in producing my examples: the utterance leads off with a verb in the first person
singular of the present indicative active, as in 'I promise you that .. .'. The other
form, which comes to exactly the same but is more common in utterances issued in
writing, employs by contrast a verb in the passive voice and in the second or third
person of the present indicative, as in 'Passengers are requested to cross the line by
the footbridge only'. If we ask ourselves, as sometimes we may, whether a given
utterance of this form is performative or constative, we may setde the question by
asking wh ether it would be possible to insert in it the word 'hereby' or some equiva-
lent-as, in French, the phrase 'par ces mots-ci'.
By way of putting to the test utterances which one might take to be performative,
we make use of a well-known asymmetry, in the case of what we call an 'explicit
performative' verb, between the first person singular of the present indicative, and
other persons and tenses of the same verb. Thus, 'I promise' is a formula which is
used to perform the act of promising; 'I promised', on the other hand, or 'he
promises', are expressions which serve simply to describe or report an act of
promising, not to perform one.
, ['Rap/are d'engagemen/'. Austin himself supplies the translation. Trans. [
7 ['ThaI is 10 say, a particular case of unhappiness mighl arguably, or even quile properly, be c1assifiable
under more Ihan one heading. Trans.1
16 PERFORMATIVE-CONSTA TIVE
However, it is not in the least necessary that an utterance, if it is to be performa-
tive, should be expressed in one of these so-called normal forms. To say 'Shut the
door', plainly enough, is every bit as performative, every bit as much the performance
of an act, as to say 'I order you to shut the door'. Even the word 'Dog' by itself can
sometimes (at any rate in England, a country more practical than ceremonious) stand
in place of an explicit and formal performative; one performs, by this little word, the
very same act as by the utterance 'I warn you that the dog is about to attack us', or
by 'Strangers are warned that here there is a vicious dog'. To make our utterance
performative, and quite unambiguously so, we can make use, in place of the explicit
formula, of a whole lot of more primitive devices such as intonation, for instance, or
gesture; further, and above all, the very context in which the words are uttered can
make it entireIy certain how they are to be taken--as a description, for example,
or again as a warning. Does this word 'Dog' just give us a bit of detail about the
local fauna? In the context--when confronted, that is, with the notice on the gate-
we just don't need to ask ourseIves that question at all.
All we can really say is that our explicit performative formula ('I promise ... ',
'I order you .. .', etc.) serves to make explicit, and at the same time more precise,
what act it is that the speaker purports to perform in issuing his utterance. I say 'to
make explicit', and that is not at all the same thing as to state. S Bending low before
you, I remove my hat, or perhaps I say 'Salaam'; then, certainly, I am doing
obeisance to you, not just engaging in gymnastics; but the word 'Salaam' does not,
any more than does the act of removing my hat, in any way state that I am doing
obeisance to you. It is in this way that our formula makes the issuing of the utterance
that action which it is, but does not state that it is that action.
The other forms of expression, those that have no explicit performative formula,
will be more primitive and less precise, one might almost say more vague. If I say
simply 'I will be there', there will be no telling, just by considering the words, whether
I am taking on a commitment, or declaring an intention, or making perhaps a fatal-
istic prediction. One may think of the precise formulae as a relativeIy recent
phenomenon in the evolution of language, and as going together with the evolution
of more complex forms of society and science.
We can't, then, expect any pureIy verbal criterion of the performative. We may
hope, all the same, that any utterance which is in fact performative will be reducible
(in some sense of that word) to an utterance in one or the other of our normal forms.
Then, going on from there, we should be able, with the heIp of a dictionary, to make
a list of all the verbs which can figure in one of our explicit formulae. Thus we will
achieve a useful classification of all the varieties of acts that we perform in saying
something (in one sense, at least, of that ambiguous phrase).
Wehave now brought in, then, the ideas of the performative utterance, of its un-
happinesses, and of its explicit formulae. But we have been talking all along as if
, l'Ajlirmer. I have translated this verb by 'state' throughout. Trans.]
PERFORMA TIVE-CONSTA TIVE 17
every utterance had to be either constative or performative, and as ifthe idea ofthe
constative at any rate was as dear as it is familiar. But it is not.
Let us note in the first place that an utterance which is undoubtedly a statement
of fact, therefore constative, can fail to get by9 in more than one way. It can be
untrue, to be sure; but it can also be absurd, and that not necessarily in some gross
fashion (by being, for instance, ungrammatical). I would like to take a doser look at
three rather more subtle ways ofbeing absurd, two ofwhich have only recently co me
to light.
(1) Someone says 'All John's children are bald, but [or 'and'] John has no
children'; or perhaps he says 'All John's children are bald', when, as a matter of fact,
John has no children.
(2) Someone says 'The cat is on the mat, but [or 'and'] I don't believe it is'; or
perhaps he says 'The cat is on the mat', when, as a matter offact, he does not believe
it iso
(3) Someone says 'All the guests are French, and some ofthem aren't': or perhaps
he says 'All the guests are French', and then afterwards says 'Some of the guests are
not French'.
In each of these cases one experiences a feeling of outrage, and it's possible each
time for us to try to express it in terms of the same word-'implication', or perhaps
that word that we always find so handy, 'contradiction'. But there are more ways of
killing the cat than drowning it in butter,IO and equally, to do violence to language
one does not always need a contradiction.
Let us use the three terms 'presuppose', 'imply', and 'entail' ll for our three cases
respectively. Then:
1. Not only 'John's children are bald', but equally 'John's children are not bald',
presupposes that John has children. To talk about those children, or to refer to them,
presupposes that they exist. By contrast, 'The cat is not on the mat' does not, equally
with 'The cat is on the mat', imply that I believe it is; and sirnilarly, 'None of the
guests is French' does not, equ8uy with 'All the guests are French', entail that it is
false that some of the guests are not French.
2. We can quite weil say 'It could be the case both that the cat is on the mat and
that I do not believe it is'. That is to say, those two propositions are not in the least
incompatible: both can be true together. What is impossible is to state both at the
same time: his stating that the cat is on the mat is what implies that the speaker
believes it iso By contrast, we couldn't say 'It could be the case both that John has no
• [The French phrase is 'peut ne pas jouer'. Austin himself sometimes used in English the coined term
'non-play' (see, e.g., How To Do Things with Words, pp. 18n. and 31), but in a more restricted sense than
would be appropriate here. Trans.1
10 English proverb. [ am told that this rather refined way of disposing of cats is not found in France.
11 [These three English terms are supplied in a footnote by Austin himself. Trans.1
18 PERFORMA TIVE-CONSTA TIVE
children and that his children are bald'; just as we couldn't say 'It could be the case
both that all the guests are French and that some of them are not French'.
3. If' AU the guests are French' entails 'It is not the case that some of the guests
are not French', then 'Some of the guests are not French' entails 'It is not the case
that all the guests are French'. It's a question here of the compatibility and incom-
patibility of propositions. By contrast, it isn't like this with presupposition: if'John's
children are bald' presupposes that John has children, it isn't true at all that 'John
has no children' presupposes that John's children are not bald. Similarly, if'The cat
is on the mat' implies that I believe it is, it isn't true at all that to say 'I don't believe
that the cat is on the mat' implies that the cat is not on the mat (not, at any rate, in
the same sense of 'implies' ; besides, we have already seen that 'implication', for us,
is not a matter of the incompatibility of propositions).
Here then are three ways in which a statement can fail to get by without being
untrue, and without being a sheer rigmarole either. I would like to call attention to the
fact that these three ways of failing to get by correspond to three of the ways in which
a performative utterance may be unhappy. To bring out the comparison, let's first
take two performative utterances:
4. 'I bequeath my watch to you, but tor 'and'] I haven't got a watch'; or perhaps
someone says 'I bequeath my watch to you' when he hasn't got a watch.
5. 'I promise to be there, but tor 'and'] I have no intention of being there'; or
perhaps someone says 'I promise to be there' when he doesn't intend to be there.
We compare case 4 with case 1, the case, that is, of presupposition. For to say
either 'I bequeath my watch to you' or 'I don't bequeath my watch to you' pre-
supposes equally that I have a watch; that the watch exists is presupposed by the
fact that it is spoken of or referred to, in the performative utterance just as much as
in the constative utterance. And just as we can make use here of the term 'presupposi-
tion' as employed in the doctrine of the constative, equally we can take over for that
doctrine the term 'void' as employed in the doctrine of the unhappinesses of the
performative. The statement on the subject of John's children is, we may say, 'void
for lack of reference', which is exactly what lawyers would say about the purported
bequest of the watch. So he re is a first instance in which a trouble that affiicts state-
ments turns out to be identical with one of the unhappinesses typical of the performa-
tive utterance.
We compare case 5 with case 2, that is, the case where something is 'implied'. Just
as my saying that the cat is on the mat implies that I believe it is, so my saying I
promise to be there implies that I intend to be there. The procedure of stating is
designed for those who honestly believe what they say, exactly as the procedure of
promising is designed for those who have a certain intention, namely, the intention to
do whatever it may be that they promise. If we don't hold the belief, or again don't
PERFORMA TIVE-CONST A TIVE 19
have the intention, appropriate to the content of our utterance, then in each case
there is lack of sincerity and abuse of the procedure. If, at the same time as we make
the statement or the promise, we announce in the same breath that we don't believe
it or we don't intend to, then the utterance is 'self-voiding', as we might call it; and
hence our feeling of outrage on hearing it. Another instance, then, where a trouble
which afflicts statements is identical with one of the unhappinesses which atHiet
performative utterances.
Let us look back, next, to case 3, the case of entailment among statements. Can
we find, in the case of performatives, some analogue for this as weil? When I make
the statement, for instance, 'All the guests are French', do I not commit myselfin a
more or less rigorous fashion to behaving in future in such-and-such a way, in
particular with respect to the statements I will make? If, in the sequel, I state things
incompatible with my utterance (namely, that aIl the guests are French), there will
be a breach of commitment that one might weil compare with that of the case in
which I say 'I welcome you', and then proceed to treat you as an enemy or an
intruder--and perhaps even better, with that of which one is guilty when one says
'I define the word thus' (a performative utterance) and then proceeds to use the word
with a different meaning.
So then, it seems to me that the constative utterance is every bit as liable to
unhappinesses as the performative utterance, and indeed to pretty much the same
unhappinesses. Furthermore, making use of the key provided by our list of un-
happinesses noted for the case of performatives, we can ask our,selves whether there
are not still more unhappinesses in the case of statements, besides the three we have
just mentioned. For example, it often happens that aperformative is void because the
utterer is not in astate, or not in a position, to perform the act which he purports to
perform ; thus, it's no good my saying 'I order you' if I have no authority over you:
I can't order you, my utterance is void, my act is only purported. Now people have,
I know, the impression that where astatement, a constative utterance, is in question,
the case is quite different: anybody at all can state anything at aIl. What if he's
ilI-informed? Weil then, one can be mistaken, that's all. It's a free country, isn't it?
To state what isn't true is one of the Rights of Man. However, this impression can
lead us into error. In reality nothing is more common than to find that one can state
absolutely nothing on some subject, because one is simply not in a position to state
whatever it may be-and this may come about, too, for more than one reason. I
cannot state at this moment how many people there are in the next room : I haven't
been to see, I haven't found out the facts. What ifl say, nevertheless, 'At this moment
there are fifty people in the next room'? You will allow, perhaps, that in saying that I
have made a guess,12 but you will not allow that I have made a statement, not at any
rate without adding 'but he had no right whatever to do so'; and in this case my 'I
12 [The French text has 'conjoncture' here, but this must surely be amisprint für 'conjecture'. Trans. I
20 PERFORMA TIVE-CONSTA TIVE
state .. .' is exactly on a par with our 'I order .. .', said, we remember, without any
right to give an order. Here's another example. You confide to me Tm bored', and I
quite coolly reply 'You're not'. You say 'What do you mean, I'm not? What right
have you to say how I feeI?' I say 'But what do you mean, what right have I? I'm just
stating what your feelings are, that's all. I may be mistaken, certainly, but what of
that? I suppose one can always make a simple statement, can't one?' But no, one
can't always: usually, I can't state what your feelings are, unless you have discIosed
them to me.
So far I have called attention to two things: that there is no purely verbal criterion
by which to distinguish the performative from the constative utterance, and that the
constative is liable to the same unhappinesses as the performative. Now we must
ask ourselves whether issuing a constative utterance is not, after all, the performance
of an act, the act, namely, of stating. Is stating an act in the same sense as marrying,
apologizing, betting, etc.? 1 can't plumb this mystery any further at present. But it
is already pretty evident that the formula 'I state that .. .' is cIosely similar to the
formula 'I warn you that ... '-a formula which, as we put it, serves to make explicit
what speech-act 13 it is that we are performing; and also, that one can't issue any
utterance whatever without performing some speech-act of this kind.
What we need, perhaps, is a more general theory of these speech-acts, and in this
theory our Constative-Performative antithesis will scarcely survive.
Here and now it remains for us to examine, quite briefly, this craze for being either
true or false, something which people think is peculiar to statements alone and ought
to be set up on a pedestal of its own, above the battle. And this time let's beg in with
the performative utterance: is it the case that there is nothing here in the least
analogous with truth?
To begin with, it is cIear that if we establish that aperformative utterance is not
unhappy, that is, that its author has performed his act happily and in all sincerity, that
still does not suffice to set it beyond the reach of all criticism. It may always be
criticized in a different dimension.
Let us suppose that I say to you 'I advise you to do it'; and let us allow that alI the
circumstances are appropriate, the conditions for success are fulfilled. In saying that,
I actually do advise you to do it--it is not that I state, truely or falsely, that I advise
you. It is, then, a performative utterance. There does still arise, all the same, a little
question: was the advice good or bad? Agreed, I spoke in alI sincerity, I believed that
to do it would be in your interest; but was I right? Was my belief, in these circum-
stances, justified? Or again--though perhaps this matters less-was it in fact, or as
things turned out, in your interest? There is confrontation of my utterance with the
situation in, and the situation with respect to which, it was issued. 1 was fully
justified perhaps, but was I right?
Many other utterances which have an incontestably performative flavour are
13 lAustin supplies this English term himself. It is in any case the term he regularly used. Trans.l
PERFORMA TIVE-CONSTA TIVE 21
exposed to this second kind of criticism. Allowing that, in dedaring the accused
guilty, you have reached your verdict properly and in good faith, it still remains to
ask whether the verdict wasjust, or fair. Allowing that you had the right to reprimand
him as you did, and that you have acted without malice, one can still ask whether
your reprimand was deserved. Here again we have confrontation with the facts,
induding the circumstances of the occasion of utterance.
That not all performative utterances without exception are liable to this quasi-
objective evaluation-which for that matter must here be left pretty vague and
multifarious-may very well be true.
There is one thing that people will be particularly tempted to bring up as an
objection against any comparison between this second kind of criticism and the kind
appropriate to statements, and that is this: aren't these questions about something's
being good, or just, or fair, or deserved entirely distinct from questions oftruth and
falsehood? That, surely, is a very simple black-and-white business: either the utter-
ance corresponds to the facts or it doesn't, and that's that.
Well, I for my part don't think it iso Even if there exists a well-defined dass of
statements and we can restrict ourselves to that, this dass will always be pretty wide.
In this dass we shall have the following statements:
France is hexagonal.
Lord Raglan won the batt1e of Alma.
Oxford is 60 miles from London.
It's quite true that for each of these statements we can raise the question 'true or
false'. But it is only in quite favourable cases that we ought to expect an answer yes
or no, once and for all. When the question is raised one understands that the utterance
is to be confronted in one way or another with the facts. Very well. So let's confront
'France is hexagonal' with France. What are we to say, is it true or not? The question,
plainly, oversimplifies things. Oh well, up to a point ifyou like, I see what you mean,
true perhaps for some purposes or in some contexts, that would do for the the man
in the street but not for geographers. And so on. It's a rough statement, no denying
that, but one can't just say straight out that it's false. Then Alma, a soldier's batde if
ever there was one; it's true that Lord Raglan was in command of the allied army,
and that this army to some extent won a confused sort of victory; yes, that would be
a fair judgement, even well deserved, for schoo1children anyway, though really it's a
bit of an exaggeration. And Oxford, well yes, it's true that that city is 60 miles from
London, so long as you want only a certain degree of precision.
Under the heading 'truth' what we in fact have is, not a simple quality nor a
relation, not indeed one anything, but rather a whole dimension of criticism. We can
get some idea, perhaps not a very dear one, of this criticism; wh at is dear is that
there is a whole lot of things to be considered and weighed up in this dimension alone
---the facts, yes, but also the situation of the speaker, his purpose in speaking, his
22 PERFORMA TIVE-CONSTA TIVE
hearer, questions of precision, etc. If we are content to restrict ourse!ves to state-
ments of an idiotic or ideal simplicity, we shall never succeed in disentangling the true
from the just, fair, deserved, precise, exaggerated, etc., the summary and the detail,
the full and the concise, and so on.
From this side also, then, from the side of truth and falsehood, we fee! ourselves
driven to think again about the Performative-Constative antithesis. What we need, it
seems to me, is a new doctrine, both complete and general, of what one is doing in
saying something, in all the senses of that ambiguous phrase, and of what I call the
speech-act, not just in this or that aspect abstracting from all the r ~ s t , but taken in
its totality.
II
INTENTION AND CONVENTION IN
SPEECH ACTS
P. F. STRA WSON
I
1 N this paper 1 want to discuss some questions regarding J. L. Austin's notions of the
illocutionary force of an utterance and of the illocutionary act which a speaker per-
forms in making an utterance.
'
There are two preliminary matters 1 must mention, if only to get them out of the
way. Austin contrasts what he calls the 'normal' or 'serious' use of speech with what
he calls 'etiolated' or 'parasitical' uses. His doctrine of illocutionary force relates
essentially to the normal or serious use of speech and not, or not directly, to
etiolated or parasitical uses; and so it will be with my comments on his doctrine. 1 am
not suggesting that the distinction between the normal or serious use of speech and
the secondary uses which he calls etiolated or parasitical is so c1ear as to call for no
further examination ; but 1 shall take it that there is such a distinction to be drawn
and 1 shall not here further examine it.
My second preliminary remark concerns another distinction, or pair of distinctions,
which AustilT draws. Austin distinguishes the illocutionary force of an utterance from
what he calls its 'meaning' and distinguishes between the illocutionary and the
locutionary acts performed in issuing the utterance. Doubts may be feit about the
second term of each of these distinctions. It may be feit that Austin has not made
c1ear just what abstractions from the total speech act he intends to make by means
of his notions of meaning and of locutionary act. Although this is a question on which
1 have views, it is not what the present paper is about. Whatever doubts may be
entertained about Austin's notions of meaning and of locutionary act, it is enough
for present purposes to be able to say, as 1 think we c1early can, the following about
their relation to the notion of illocutionary force. The meaning of a (serious) utter-
ance, as conceived by Austin, always embodies some limitation on its possible force,
and sometimes--as, for example, in some cases where an explicit performative
formula, like 'I apologize', is used---the meaning of an utterance may exhaust its
From The Philosophical Re"iew, vol. LXXIII. no. 4, Oetober 1964, pp. 439-60. Reprinted by permission
of the author and the editor of PhilosophicaJ Rel'iew.
I All references, un1ess otherwise indieated, are to How To Do Things with Words (Oxford, 1962).
24 INTENTION AND CONVENTION IN SPEECH ACTS
force; that is, there may be no more to the force than there is to the meaning; but
very often the meaning, though it limits, does not exhaust, the force. Similarly, there
may sometimes be no more to say about the illocutionary force of an utterance than
we already know if we know what locutionary act has been performed; but ,very
often there is more to know about the illocutionary force of an utterance than we
know in knowing what locutionary act has been performed.
So much for these two prelirninaries. Now I shall proceed to assemble from the
text some indications as to what Austin means by the force of an utterance and as to
what he means by an illocutionary act. These two notions are not so closely related
that to know the force of an utterance is the same thing as to know what illocutionary
act was actually performed in issuing it. For if an utterance with the illocutionary
force of, say, a warning is not understood in this way (that is, as a waming) by the
audience to which it is addressed, then (it is held) the illocutionary act of waming
cannot be said to have been actually performed. 'The performance of an illocutionary
act involves the securing of uptake'; that is, it involves 'bringing about the under-
standing of the meaning and of the force of the locution' (pp. 115-16).2 Perhaps we
may express the relation by saying that to know the force of an utterance is the same
thing as to know what illocutionary act, if any, was actually performed in issuing it.
Austin gives many examples and lists of words which help us to form at least a fair
intuitive notion of what is meant by 'illocutionary force' and 'illocutionary act'.
Besides these, he gives us certain general clues to these ideas, which may be grouped,
as folIows, under four heads:
l. Given that we know (in Austin's sense) the meaning of an utterance, there may
still be a further question as to how what was said was meant by the s p e a ~ e r , or as to
how the words spoken were used, or as to how the utterance was to be taken or ought
to have been taken (pp. 98-9). In order to know the illocutionary force ofthe utter-
ance, we must know the answer to this further question.
2. A locutionary act is an act of saying something; an iIlocutionary act is an act
we perform in saying something. It is what we do, in saying what we say. Austin does
not regard this characterization as by any means a satisfactory test for identifying
kinds of illocutionary acts since, so regarded, it would admit many kinds of acts
which he wishes to exclude from the class (p. 99 and Lecture X).
3. It is a sufficient, thought not, I think, a necessary, condition of a verb's being the
name of a kind of illocutionary act that it can figure, in the first person present indica-
tive, as what Austin calls an explicit performative. (This latter notion I shall assurne
to be familiar and perspicuous.)
4. The illocutionary act is 'a conventional act; an act done as conforming to a
convention' (p. 105). As such, it is to be sharply contrasted with the producing of
certain effects, intended or otherwise, by me ans of an utterance. This producing of
2 I refer later to the need for quaIification of this doctrine.
INTENTION AND CONVENTION IN SPEECH ACTS 25
effects, though it too can often be ascribed as an act to the speaker (his perlocu-
tionary act), is in no way a conventional act (pp. 120-1). Austin reverts many times
to the 'conventional' nature ofthe illocutionary act (pp. 103, 105, 108, 115, 120, 121,
127) and speaks also of 'conventions of illocutionary force' (p. 114). Indeed, he
remarks (pp. 120-]) that though acts which can properly be called by the same
names as illocutionary acts-for example, acts of warning--<:an be brought off
nonverbally, without the use of words, yet, in order to be properly called by these
names, such acts must be conventional nonverbal acts.
11
I shall assurne that we are clear enough about the intended application of Austin's
notions of illocutionary force and illocutionary act to be able to criticize, by reference
to cases, his general doctrines regarding those notions. It is the general doctrine I
listed last above--the doctrine that an utterance's having such and such a force is a
matter of convention-that I shall take as the starting point of inquiry. Usually this
doctrine is affirmed in a quite unqualified way. But just once there occurs an interest-
ingly qualified statement of it. Austin says, of the use of language with a certain illo-
cutionary force, that 'it may ... be said to be conventional in the sense that at least
it could be made explicit by the performative formula' (p. 103). The remark has a
certain authority in that it is the first explicit statement of the conventional nature of
the illocutionary act. I shall refer to it later.
Meanwhile let us consider the doctrine in its unqualified form. Why does Austin
say that the illocutionary act is a conventional act, an act done as conforming to a
convention? I must first mention, and neutralize, two possible sources of confusion.
(It may seem an excess of precaution to do so. I apologize to those who find it so.)
First, we may agree (or not dispute) that any speech act is, as such, at least in part
a conventional act. The performance of any speech act involves at least the observ-
ance or exploitation of some linguistic conventions, and every illocutionary act is a
speech act. But it is absolutely clear that this is not the point that Austin is making
in dec\aring the illocutionary act to be a conventional act. We must refer, Austin
would say, to linguistic conventions to determine what locutionary act has been per-
formed in the making of an utterance, to determine what the meaning of the utterance
iso The doctrine now before us is the further doctrine that where force is not exhausted
by meaning, the fact that an utterance has the further unexhausted force it has is also
a matter of convention; or, where it is exhausted by meaning, the fact that it is, is a
matter of convention. It is not just as being a speech act that an illocutionary act-
for example, of warning-is conventional. A nonverbal act of warning is, Austin
maintains, conventionally such in just the same way as an illocutionary---that is,
verbal-act of warning is conventionally such.
Second, we must dismiss as irrelevant the fact that it can properly be said to be a
26 INTENTION AND CONVENTION IN SPEECH ACTS
matter of convention that an act of, for example, warning is correctly called by this
name. For if this were held to be a ground for saying that illocutionary acts were
conventional acts, then any describable act whatever would, as correctly described,
be a conventional act.
The contention that illocutionary force is a matter of convention is easily seen to be
correct in a great number of cases. F or very many kinds of human transaction involv-
ing speech are govemed and in part constituted by what we easily recognize as
established conventions of procedure additional to the conventions governing the
meanings of our utterances. Thus the fact that the word 'guilty' is pronounced by the
foreman of the jury in court at the proper moment constitutes his utterance as the
act of bringing in a verdict; and that this is so is certainly a matter of the con-
ventional procedures of the law. Similarly, it is a matter of convention that if the
appropriate umpire pronounces a batsman 'out', he thereby performs the act of giving
the man out, which no player or spectator shouting 'Out!' can do. Austin gives other
examples, and there are doubtless many more which could be given, where there
clearly exist statable conventions, relating to the circumstances of utterance, such
that an utterance with a certain meaning, pronounced by the appropriate person in
the appropriate circumstances, has' the force it has as conforming to those con-
ventions. Examples of illocutionary acts of which this is true can be found not only
in the sphere of social institutions which have a legal point (like the marriage
ceremony and the law courts themselves) or of activities govemed by a definite set of
rules (like cricket and games generally) but in many other relations ofhuman lire. The
act of introducing, performed by uttering the words 'This is Mr. Smith', may be said
to be an act performed as conforming to a convention. The act of surrendering, per-
formed by saying 'Kamerad!' and throwing up your arms when confronted with a
bayonet, may be said to be (to have become) an act performed as conforming to an
accepted convention, a conventional act.
But it seems equally clear that, although the circumstances of utterance are always
relevant to the determination of the illocutionary force of an utterance, there are
many cases in which it is not as conforrning to an accepted convention of any kind
(other than those linguistic conventions which help to fix the meaning of the utter-
ance) that an illocutionary act is performed. It seems clear, that is, that there are
many cases in which the illocutionary force of an utterance, though not exhausted by
its meaning, is not owed to any conventions other than those which help to give it its
meaning. Surely there may be cases in which to utter the words 'The ice over there is
very thin' to a skater is to issue a warning (is to say something with the force of a
warning) without its being the case that there is any statable convention at all (other
than those which bear on the nature of the locutionary act) such that the speaker's
act can be said to be an act done as conforming to that convention.
Here is another example. We can readily imagine circumstances in which an
utterance of the words 'Don't go' would be correctly described not as arequest or an
INTENTION AND CONVENTION IN SPEECH ACTS 27
order, but as an entreaty. I do not want to deny that there may be conventional
postures or procedures for entreating: one can, for example, kneel down, raise one's
arms and say, 'I entreat you.' But 1 do want to deny that an act of entreaty can be
performed only as conforming to some such conventions. What makes X's words to
Y an entreaty not to go is something-complex enough, no doubt--relating to X's
situation, attitude to Y, manner, and current intention. There are questions here·which
we must discuss later. But to suppose that there is always and necessarily a con-
vention conformed to would be like supposing that there could be no love affairs
which did not proceed on lines laid down in the Roman de la Rose or that every
dispute between men must follow the pattern specified in Touchstone's speech about
the countercheck quarrelsome and the lie direct.
Another example. In the course of a philosophical discussion (or, for that matter,
a debate on policy) one speaker raises an objection to what the previous speaker has
just said. X says (or proposes) that p and Y objects that q. Y's utterance has the force
of an objection to X's assertion (or proposaI) that p. But where is the convention that
constitutes it an objection? That Y's utterance has the force of an objection may lie
partly in the character of the dispute and of X's contention (or proposal) and it
certainly lies partly, in Y's view of these things, in the bearing which he takes the
proposition that q to have on the doctrine (or proposal) that p. But although there
may be, there does not have to be, any convention involved. other than those
linguistic conventions which help to fix the meanings of the utterances.
1 do not think it necessary to give further examples. It seems perfectIy cIear that,
if at least we take the expressions 'convention' and 'conventional' in the most natural
way, the doctrine of the conventional nature of the iIlocutionary act does not hold
genen·lly. Some illocutionary acts are conventional; others are not (except in so far
as they are locutionary acts). Why then does Austin repeatedly affirm the contrary?
It is unlikely that he has made the simple mistake of generalizing from some cases to
all. It is much more likely that he is moved by some further, and fundamental, feature
of illocutionary acts, which it must be our business to discover. Even though we may
decide that the description 'conventional' is not appropriately used, we may presume
it worth our while to look for the reason for using it. Here we may recall that oddly
qualified remark that the performance of an iIlocutionary act, or the use of a sentence
with a certain illocutionary force, 'may be said to be conventional in the sense that at
least it could be made explicit by the performative formula' (p. 103). On this we may
first, and with justice, be incIined to comment that there is no such sense of 'being
conventional', that if this is a sense of anything to the purpose, it is a sense of'being
capable of being conventional'. But although this is a proper comment on the remark,
we should not simply dismiss the remark with this comment. Whatever it is that
leads Austin to call illocutionary acts in general 'conventional' must be cIosely con-
nected with whatever it is about such acts as warning, entreating, apologizing,
advising, that accounts for the fact that they at least couldbe made explicit by the use
28 INTENTION AND CONVENTION IN SPEECH ACTS
of the corresponding first-person perfonnative fonn. So we must ask what it is about
them that accounts for this fact. Obviously it will not do to answer simply that they
are acts which can be performed by the use of words. So are many (perlocutionary)
acts, like convincing, dissuading, alarming, and amusing, for which, as Austin points
out, there is no corresponding first-person peiformative formula. So we need some
further explanation.
111
I think a concept we may find helpful at this point is one introduced by H. P. Grice
in his valuable article on Meaning (PhiIosophieal Review, LXVII, 1957), namely, the
concept of someone's nonnaturally meaning something byan utteranee. The concept
does not apply only to speech acts-that is, to cases where that by which someone
nonnaturally means something is a linguistie utterance. It is of more general applica-
tion. But it will be convenient to refer to that by which someone, S, nonnaturally
means something as S's utteranee. The explanation of the introduced concept is given
in terms of the concept of intention. S nonnaturally means something by an utterance
x if S intends (i \) to produce by uttering x a certain response (r) in an audience A and
intends (i2) that A shall recognize S's intention (i\) and intends (i,) that this recogni-
tion on the part of A of S's intention (i) shall function as A 's reason, or a part of
his reason, for his response r. (The word 'response', though more convenient in some
ways than Grice's 'efTect', is not ideal. It is intended to cover cognitive and afTective
states or attitudes as weil as actions.) It is, evidently, an important feature of this
definition that the securing of the response r is intended to be mediated by the
securing of another (and always cognitive) efTect in A; namely, recognition of S's
intention to sec ure response r.
Grice's analysis of his concept is fairly complex. But I think a little reflection shows
that it is not quite complex enough for his purpose. Grice's analysis is undoubtedly
offered as an analysis of a situation in which one person is trying, in a sense of the
word 'communicate' fundamental to any theory of meaning, to communicate with
another. But it is possible to imagine a situation in which Grice's three conditions
would be satisfied by a person Sand yet, in this important sense of'communicate',
it would not be the case that S could be said to be trying to communicate by means
of his production of x with the person A in whom he was trying to produce the
response r. I proceed to describe such a situation.
S intends by a certain action to induce in A the belief that p; so he satisfies con-
dition (i J He arranges convincing-looking 'evidence' that p, in a place where A is
bound to see it. He does this, knowing that A is watching him at work, but knowing
also that A does not know that S knows that A is wate hing him at work. He realizes
that A will not take the arranged 'evidence' as genuine or natural evidence that p, but
realizes, and indeed intends, that A will take his arranging of it as grounds for thinking
INTENTION AND CONVENTION IN SPEECH ACTS 29
that he, S, intends to induce in A the belief that p. That is, he intends A to recognize
his (i I) intention. So S satisfies condition (i
2
). He knows that A has general grounds
for thinking that S would not wish to make hirn, A, think that p unless it were known
to S to be the case that p; and hence that A's recognition of his (Ss) intention to
induce in A the belief that p will in fact seem to A a sufficient reason for believing
that p. And he intends that A's recognition ofhis intention (il) should function injust
this way. So he satisfies condition (i,).
S, then, satisfies all Grice's conditions. But this is clearly not a case of attempted
communication in the sense which (I think it is fair to assurne) Grice is seeking to
elucidate. A will indeed take S to be trying to bring it about that A is aware of some
fact; but he will not take S as trying, in the colloquial sense, to 'let him know' some-
thing (or to 'tell' hirn something). But unless S at least brings it about that A takes
him (S) to be trying to let him (A) know something, he has not succeeded in com-
municating with A ; and if, as in our ex am pie, he has not even tried to bring this about,
then he has not even tried to communicate with A. It seems a minimum further
condition of his trying to do this that he should not only intend A to recognize his
intention to get A to think that p, but that he should also intend A to recognize his
intention to get A to recognize his intention to get A to think that p.
We might approximate more closely to the communication situation if we changed
the example by supposing it not only clear to both A and S that A was watching S
at work, but also clear to them both that it was clear to them both. 1 shall content
myself, however, with drawing from the actually considered example the conclusion
that we must add to Grice's conditions the further condition that S should have the
further intention (i 4) that A should recognize his intention (i J. It is possible that
further argument could be produced to show that even adding this condition is not
suflicient to constitute the case as one of attempted communication. But 1 shall rest
content for the moment with the fact that this addition at least is necessary.
Now we might have expected in Grice's paper an account of what it is for A to
understand something by an utterance x, an account complementary to the account
of what it is for S to mean something by an utterance x. Grice in fact gives no such
account, and 1 shall suggest a way of at least partially supplying this lack. 1 say'at
least partially' because the uncertainty as to the sufficiency of even the modified
conditions for S's nonnaturally meaning something by an utterance x is refiected in
a corresponding uncertainty in the sufficiency of conditions for A 's understanding.
But again we may be content for the moment with necessary conditions. I suggest,
then, that for A (in the appropriate sense of 'understand') to understand something
by utterance x, it is necessary (and perhaps sufficient) that there should be some
complex intention of the (i 2) form, described above, which A takes S to have, and that
for A to understand the utterance correctly, it is necessary that A should take S to
have the complex intention of the (i 2) form which S does have. In other words, if A
is to understand the utterance correctly, S'S (iJ intention and hence his (i 2) intention
30 INTENTION AND CONVENTION IN SPEECH ACTS
must be fulfilled. Of course it does not follow from the fulfilment of these intentions
that his (i.) intention is fulfilled; nor, consequently, that his (i ,) intention is fulfilled.
It is at this point, it seems, that we may hope to find a possible poinl of connection
with Austin's terminology of 'securing uptake'. If we do find such a point of connec-
tion, we also find a possible starting point for an at least partial analysis of the notions
of illocutionary force and of the illocutionary act. For to secure uptake is to secure
understanding of (meaning and) illocutionary force; and securing understanding of
illocutionary force is said by Austin to be an essential element in bringing off the illo-
cutionary act. It is true that this doctrine of Austin's may be objected to.
3
For sure1y
a man may, for example, actually have made such and such a bequest, or gift, even
if no one ever reads his will or instrument of gift. We may be tempted to say instead
that at least the aim, if not the achievement, of securing uptake is an essential
element in the performance ofthe illociItionary act. To this, too, there is an objection.
Might not a man really have made a gift, in due form, and take some satisfaction in
the thought, even if he had no expectations of the fact ever being known? But this
objection at most forces' on us an amendment to which we are in any case obliged
4
:
namely, that the aim, if not the achievement, of securing uptake is essentially a
standard, if not an invariable, element in the performance of the illocutionary act.
So the analysis of the aim of securing uptake remains an essential element in the
analysis of the notion of the illocutionary act.
IV
Let us, then, make a tentative identification-to be subsequently qualified and revised
--of Austin's notion of uptake with that at least partially analysed notion ofunder-
standing (on the part of an audience) which I introducedjust now as complementary
to Grice's concept of somebody nonnaturally meaning something by an utterance.
Since the notion of audience understanding is introduced by way of a fuller (though
partial) analysis than any which Austin gives of the notion of uptake, the identifica-
tion is equivalent to a tentative (lmd partial) analysis of the notion of uptake and
hence of the notions of illocutionary act and illocutionary force. If the identification
were correct, then it would follow that to say something with a certain illocutionary
force is at least (in the standard case) to have a certain complex intention ofthe (iJ
form described in setting out and modifying Grice's doctrine.
Next we test the adequacy and explanatory power of this partial analysis by seeing
how far it helps to explain other features of Austin's d.octrine regarding illocutionary
acts. There are two points at which we shall apply this test. One is the point at which
Austin maintains that the production of an utterance with a certain illocutionary
.1 I owe the objections which follow to Professor Hart.
4 For an illocutionary act may be performed altoKether unintentionally. See the example about
redoubling at bridge, p. 36 below.
INTENTION AND CONVENTION IN SPEECH ACTS 31
force is a conventional act in that unconventional sense of 'conventional' which he
glosses in terms of general suitabiIity for being made explicit with the help of an
explicitly performative formula. The other is the point at which Austin considers the
possibility of a general characterization of the illocutionary act as what we do, in
saying what we say. He remarks on the unsatisfactoriness of this characterization in
that it would admit as illocutionary acts what are not such; and we may see whether
the suggested analysis helps to explain the exclusion from the class of illocutionary
acts of those acts falling under this characterization which Austin wishes to exclude.
These points are closely connected with each other.
First, then, we take the point about the general suitability of an illocutionary act
for performance with the he1p ofthe explicitly performative formula for that act. The
explanation of this feature of iIIocutionary acts has two phases ; it consists of, first, a
general, and then a special, point about intention. The first point may be roughly
expressed by saying that in general a man can speak of his intention in performing an
action with a kind of authority which he cannot command in predicting its outcome.
What he intends in doing something is up to him in a way in which the results of his
doing it are not, or not only, up to hirn. But we are concerned not with just any in-
tention to produce any kind of effect by acting, but with a very special kind of case.
We are concerned with the case in which there is not simply an intention to produce
a certain response in an audience, but an intention to produce that response by means
of recognition on the part of the audience of the intention to produce that response,
this recognition to serve as part of the reason that the audience has for its response,
and the intention that this recognition should occur being itself intended to be
recognized. The speaker, then, not only has the general authority on the subject of
his intention that any agent has; he also has a motive, inseparable from the nature of
his act, for making that intention dear. For he will not have secured understanding
of the illocutionary force of his utterance, he will not have performed the act of
communication he sets out to perform, unless his complex intention is grasped. Now
clearly, for the enterprise to be possible at all, there must exist, or he must find, me ans
of making the intention dear. If there exists any conventional linguistic means of
doing so, the speaker has both a right to use, and a motive for using, those means.
One such means, available sometimes, which comes very elose to the employment of
the explicit performative form, would be to attach, or subjoin, to the substance of the
message what looks like a forcecelucidating comment on it, which may or may not
have the form of a self-ascription. Thus we have phrases like 'This is only a sug-
gestion' or 'I'm only making a suggestion'; or again 'That was a warning' or 'I'm
warning you'. For using such phrases, I repeat, the speaker has the authority that
anyone has to speak on the subject of his intentions and the motive that I have tried to
show is inseparable from an act of communication.
From such phrases as these-which have, in appearance, the character of com-
ments on utterances other than themselves-to the explicit performative formula the
32 INTENTION AND CONVENTION IN SPEECH ACTS
step is only a short one. My reason for qualijring the re mark that such phrases have
the character of comments on utterances other than themselves is this. We are con-
sidering the case in which the subjoined quasi-comment is addressed to the same
audience as the utterance on which it is a quasi-comment. Since it is part ofthe
speaker's audience-directed intention to make c\ear the character of his utterance
as, for example, a warning, and since the subjoined quasi-comment directly subserves
this intention, it is better to view the case, appearances not\\'ithstanding. /lot as a case
in which we have two utterances. one commenting on the other. but as a case of a
single unitary speech act. Crudely. the addition of the quasi-comment 'That was a
warning' is part of the total act of warning. The effect of the short step to the
explicitly performative formula is simply to bring appearances into line with reality.
When that short step is taken, we no longer have, even in appearance. two utter-
ances, one a comment on the other, but a single utterance in which the first-person
performative verb manifest(r has that peculiar logical character of which Austin
rightly made so much, and which we may express in the present context by saying
that the verb serves not exactly to ascribe an intention to the speaker but rather. in
Austin's phrase, to make explicit the type of communication intention with which the
speaker speaks, the type of force which the utterance has.
The above rnight be said to be a deduction of the general possibility and utility of
the explicitly performative formula for the cases of illocutionary acts not essentia11y
conventional. It may be objected that the deduction fails to show that the intentions
rendered explicit by the use of performative formulae in general must be of just the
complex form described, and hence fails to justify the claim that just this kind of
intention lies at the core of all i11ocutionary acts. And indeed we sha11 see that this
claim would be mistaken. But before discussing why, we shall make a further applica-
tion of the analysis at the second testing point I mentioned. That is, we shall see what
power it has to explain why some of the things we may be dOing, in saying what we
say, are not illocutionary acts and could not be rendered explicit by the use of the
performative formula.
Among the things mentioned by Austin which we might be doing in saying things,
but which are not illocutionary acts, I shall consider the two examples of (1) sho\\'ing
off and (2) insinuating. Now when we show off, we are certainly trying to produce
an effect on the audience: we talk, indeed, for effect; we try to impress, to evoke the
response of adrniration. But it is no part of the intention to secure the effect hy means
01 the recognition of the intention to sec ure it. It is no part of our total intention to
secure recognition of the intention to produce the effect at a11. On the contrary:
recognition of the intention rnight militate against securing the effect and promote an
opposite effect, for example, disgust.
This leads on to a further general point not explicitly considered by Austin, but
satisfactorily explained by the analysis under consideration. In saying to an audience
what we do say, we very often intend not only to produce the primary response r by
INTENTION AND CONVENTION IN SPEECH ACTS 33
me ans of audience recognition of the intention to produce that response, but to
produce further effects by means of the production of the primary response r. Thus
my further purpose in informing you that p (that is, aiming to produce in you the
primary cognitive response of knowledge or belief that p) may be to bring it about
thereby that you adopt a certain line of conduct or a certain attitude. In saying what
I say, then, part of what I am doing is trying to influence your attitudes or conduct in
a certain way. Does this part of what I am doing in saying what I say contribute to
determining the character of the i1locutionary act I perform ? And if not, why not?
If we take the first question strict1y as introduced and posed, the answer to it is 'No'.
The reason tor the answer follows from the analysis. We have no complex intention
(i4) that there should be recognition of an intention (i) that there should be recog-
nition of an intention (i I) that the further effect should be produced; for it is no part
of our intention that the further effect should be produced by way of recognition of
our intention that it should be; the production in the audience of belief that p is
intended to be itself the means whereby his attitude or conduct is to be inftuenced.
We secure uptake, perform the act of communication that we set out to perform, if
the audience understands us as injorming hirn that p. Although it is true that, in
saying what we say, we are in fact tryingto produce the further effect-this is part of
wh at we are doing, whether we succeed in producing the effect or not-yet this does
not enter into the characterization of the illocutionary act. With this case we have to
contrast the case in which, instead of aiming at a primary response and a further
effect, the latter to be secured through the former alone, we aim at a complex primary
response. Thus in the case where I do not simply inform, but warn, you that p, among
the intentions I intend you to recognize (and intend you to recognize as intended to be
recognized), are not only the intention to sec ure your belief that p, but the intention
to secure that you are on your guard against p-perils. The difference (one of the
differences) between showing off and warning is that your recognition of my intention
to put you on your guard may weil contribute to putting you on your guard, whereas
your recognition of my intention to impress you is not likely to contribute to my
impressing you (or not in the way I intended).5
Insinuating fails, for a different reason, to be a type of illocutionary act. An
essential feature of the intentions which make up the iIIocutionary complex is their
overtness. They have, one might say, essential avowability. This is, in one respect, a
logically embarrassing feature. We have noticed already how we had to meet the
threat of a counterexample to Grice's analysis of the communicative act in terms of
three types of intention--(i I)' (i 2)' and (i,)-by the addition of a further intention
(i4) that an intention (i2) should be recognized. We have no proof, however, that the
5 Perhaps trying to impress might sometimes have an ilIocutionary character. For I might try to impress
you with my e.f!ronlery, intending you to recognize this intention and intending your recognition of it to
function as part of your reason for being impressed, and so forth. But then I am not merely trying to impress
you; I am inviting you to be impressed. I owe this point to Mr. B. F. McGuinness.
34 INTENTION AND CONVENTION IN SPEECH ACTS
resulting enlarged set of conditions is a complete analysis. Ingenuity might show it
was not; and the way seems open to a regressive series of intentions that intentions
should be recognized. While I do not think there is anything necessarily objectionable
in this, it does suggest that the complete and rounded-off set of conditions aimed at
in a conventional analysis is not easily and certainly attainable in these terms. That is
why I speak of the feature in question in these terms. That is why I speak of the
feature in question as logically embarrassing. At the same time it enables us easily
to dispose of insinuating as a candidate for the status of a type of illocutionary act.
The whole point of insinuating is that the audience is to suspect, but not more than
suspect, the intention, for example, to induce or disclose a certain belief. The intention
one has in insinuating is essentially nonavowable.
Now let us take stock a little. We tentative1y laid it down as a necessary condition
of securing understanding of the illocutionary force of an utterance that the speaker
should succeed in bringing it about that the audience took hirn, in issuing his utter-
ance, to have a complex intention of a certain kind, name1y the intention that the
audience should recognize (and recognize as intended to be recognized) his intention
to induce a certain response in the audience. The suggestion has, as we have just
seen, certain explanatory merits. Nevertheless we cannot claim general application
for it as even a partial analysis of the notions of illocutionary f o r ~ e and illocutionary
act. Let us look at some reasons why not.
v
I remarked earlier that the words 'Don't go' may have the force, inter alia, either of
arequest or of an entreaty. In either case the primary intention of the utterance (if
we presume the words to be uttered with the sense 'Don't go away') is that ofinducing
the person addressed to stay where he iso His staying where he is is the primary
response aimed at. But the only other intentions mentioned in our scheme of partial
analysis relate directly or indirectly to recognition of the primary intention. So how,
in terms of that scheme. are we to account for the variation in illocutionary force
between requests and entreaties?
This question does not appear to raise a major difficulty for the scheme. The
scheme, it seems, mere1y requires supplementing and enriching. Entreaty, for
example, is a matter of trying to secure the primary response not merely through
audience recognition of the intention to secure it, but through audience recognition of
a complex attitude of which this primary intention forms an integral part. A wish that
someone should stay may be held in different ways: passionately or lightly, con-
fidently or desperately; and it may, for different reasons, be part of a speaker's in-
tention to see ure recognition of how he holds it. The most obvious reason, in the case
of entreaty, is the belief, or hope, that such a revelation is more like1y to seeure the
fulfilment of the primary intention.
INTENTION AND CONVENTION IN SPEECH ACTS 35
But one may not only request and entreat; one mayorder someone to stay where
he iso The words 'Don't go' may have the illocutionary force of an order. Can we so
simply accommodate in our scheme this variation in illocutionary force? Weil, we
can accommodate it; though not so simply. We can say that a man who issues an
order typically intends his utterance to secure a certain response, that he intends
this intention to be recognized, and its recognition to be a reason for the response,
that he intends the utterance to be recognized as issued in a certain social context
such that certain social rules or conventions apply to the issuing of utterances in this
context and such that certain consequences may follow in the event of the primary
response not being secured, that he intends this intention too to be recognized, and
finally that he intends the recognition of these last features to function as an element
in the reasons for the response on the part of the audience.
Evidently, in this case, unlike the case of entreaty, the scheme has to be extended
to make room for explicit reference to social convention. It can, with some strain, be
so extended. But as we move further into the region of institutionalized procedures,
the strain becomes too much for the scheme to bear. On the one hand, one of its basic
features-namely, the reference to an intention to sec ure a definite response in an
audience (over and above the seeuring of uptake)---has to be dropped. On the other,
the reference to social conventions of procedure assurnes a very much greater im-
portance. Consider an umpire giving a batsman out, a jury bringing in a verdict of
guilty, a judge pronouncing sentence, a player redoubling at bridge, a priest or a
civil officer pronouncing a couple man and wife. Can we say that the umpire's
primary intention is to secure a certain response (say, retiring to the pavilion) from a
certain audience (say, the batsman), the jurymen's to secure a certain response (say,
the pronouncing of sentence) from a certain audience (say, the judge), and then build
the rest of our account around this, as we did, with some strain, in the case of the
order? Not with plausibility. It is not even possible, in other than a formal sense, to
isolate, among all the participants in the procedure (trial, marriage, game) to which
the utterance belongs, a particular audience to whom the utterance can be said to be
addressed.
Does this mean that the approach I suggested to the elucidation of the notion of
illocutionary force is entirely mistaken? I do not think so. Rather , we must distinguish
types of case; and then see what, if anything, is common to the types we have dis-
tinguished. What we initially take from Grice--with modifications--is an at least
partially analytical account of an act of communication, an act which might indeed
be performed nonverbally and yet exhibit all the essential characteristics of a (non-
verbal) equivalent of an illocutionary act. We gain more than this. For the account
enables us to understand how such an act may be linguistically conventionalized
right up to the point at which i1locutionary force is exhausted by meaning (in Austin's
sense); and in this understanding the notion of wholly overt or essentially avowable
intention plays an essential part. Evidently, in these cases, the illocutionary act itself
36 INTENTION AND CONVENTION IN SPEECH ACTS
is not essentially a conventional act, an act done as conforming to a convention; it
may be that the act is conventional, done as conforming to a convention, only in so
far as the means used to perfonn it are conventional. To speak only of those con-
ventional means which are also linguistic means, the extent to which the act is one
done as conforming to conventions may depend solelyon the extent to which con-
ventional linguistic meaning exhausts iIlocutionary force.
At the other end of the scale-the end, we may say, from which Austin beg an-
we have iIIocutionary acts which are essentially conventional. The examples I
mentioned just now will serve-marrying, redoubling, giving out, pronouncing
sentence, bringing in a verdict. Such acts could have no existence outside the
rule- or convention-governed practices and procedures of which they essentially form
parts. Let us take the standard case in which the participants in these procedures
know the rules and their roles, and are trying to play the game and not wreck it. Then
they are presented with occasions on which they have to, or may, perform an illocu-
tionary act which forms part of, or furthers, the practice or procedure as a wh oIe ;
and sometimes they have to make adecision within a restricted range of alternatives
(for example, to pass or redouble, to pr on ounce sentence of imprisonment for some
period not exceeding a certain limit). Between the case of such acts as these and the
case of the iIIocutionary act not essentially conventional, there is an important like-
ness and an important difTerence. The likeness resides in the fact that, in the case
of an utterance belonging ,to a convention-governed practice or procedure, the
speaker's utterance is standardly intended to further, or afTect the course of, the
practice in question in some one of the alternative ways open, and intended to be
recognized as so intended. I do not mean that such an act could never be performed
unintentionally. A player might let slip the word 'redouble' without meaning to
redouble; but if the circumstances are appropriate and the play strict, then he has
redoubled (or he may be held to have redoubled). But a player who continually did
this sort of thing would not be asked to play again, except by sharpers. Forms can
take charge, in the absence of appropriale intention; but when they do, the case is
essentially deviant or nonstandard. There is present in the standard case, that is to
say, the same element of wholly overt and avowable intention as in the case of the act
not essentially conventional.
The difTerence is a more complicated affair. We have, in these cases, an act which
is conventional in two connected ways. First, if things go in accordance with the
rules of the procedure in question, the act of furthering the practice in the way in-
tended is an act required or permitted by those rules, an act done as falling under the
rules. Second, the act is identified as the act it is just because it is performed by the
utterance of a form of words conventional for the performance of that act. Hence the
speaker's utterance is not only intended to further, or afTect the course of, the practice
in question in a certain conventional way; in the absence of any breach of the con-
ventional conditions for furthering the procedure in this way, it cannot fail to do so.
INTENTION AND CONVENTION IN SPEECH ACTS 37
And here we have the contrast between the two types of case. In the case of an
illocutionary act of a kind not essentiaUy conventional, the act of communication is
performed if uptake is secured, if the utterance is taken to be issued with the complex
overt intention with which it is issued. But even though the act of communication is
performed, the wholly overt intention which lies at the core of the intention complex
may, without any breach ofrules or conventions, be frustrated. The audience response
(belief, action, or attitude) may simply not be forthcoming. It is different with the
utterance which forms part of a wholly convention-governed procedure. Granted that
uptake is secured, then any frustration of the wholly overt intention of the utterance
(the intention to further the procedure in a certain way) must be attributable to a
breach of rule or convention. The speaker who abides by the conventions can avow-
ably have the intention to further the procedure in the way to which his current
linguistic act is conventionally appropriated only if he takes it that the conventional
conditions for so furthering it are satisfied and hence takes it that his utterance will
not only reveal his intentions but give them effect. There is nothing parallel to this in
the case of the illocutionary act of a kind not essentially conventional. In both cases,
we may say, speakers assurne the responsibility for making their intentions overt.
In one case (the case of the convention-constituted procedure) the speaker who uses
the explicitly performative form also explicitly assurnes the responsibility for making
his overt intention effective. But in the other case the speaker cannot, in the speech
act itself, explicitly assurne any such responsibility. For there are no conditions which
can conventionally guarantee the effectiveness of his overt intention. Whether it is
effective or not is something that rests with his audience. In the one case, therefore,
the explicitly performative form may be the name of the very act which is performed
if and only if the speaker's overt intention is effective; but in the other case it cannot
be the name of this act. But of course--and I shall recur to this thought-the sharp
contrast I have here drawn between two extreme types of case must not blind us to
the existence of intermediate types.
Acts belonging to convention-constituted procedures of the kind I have just re-
ferred to form an important part of human communication. But they do not form the
whole nor, we may think, the most fundamental part. It would be amistake to take
them as the model for understanding the notion of iIIocutionary force in general, as
Austin perhaps shows some tendency to do when he both insists that the illocutionary
act is essentially a conventional act and connects this claim with the possibility of
making the act explicit by the use of the performative formula. It would equally be a
mistake, as we have seen, to generalize the account of illocutionary force derived
from Grice's analysis; for this would involve holding, falsely, that the complex overt
intention manifested in any illocutionary act always includes the intention to secure a
certain definite response or reaction in an audience over and above that which is
necessarily secured if the illocutionary force of the utterance is understood. Never-
theless, we can perhaps extract from our consideration of two contrasting types of
38 INTENTION AND CONVENTION IN SPEECH ACTS
case something which is common to them both and to all the other types which lie
between them. For the illocutionary force of an utterance is essentially something
that is intended to be understood. And the understanding of the force of an utterance
in all cases involves recognizing what may be called broadly an audience-directed
intention and recognizing it as wholly overt, as intended to be organized. It is perhaps
this fact which lies at the base of the general possibility of the explicit performaiive
formula; though, as we have seen, extra factors come importantly into play in the
case of convention-constituted procedures.
Once this common element in all iIIocutionary acts is cIear, we can readily
acknowledge that the types of audience-directed intention involved may be very
various and, also, that different types may be exemplified by one and the same
utterance.
I have set in sharp contrast those cases in which the overt intention is simply to
forward adefinite and convention-governed practice (for ex am pIe, a game) in a
definite way provided for by the conventions or rules ofthe practice and those cases
in which the overt intention incIudes that of securing a definite response (cognitive or
practical) in an audience over and above that which is necessarily secured if uptake
is secured. But there is something misleading about the sharpness of this contrast;
and it would certainly be wrong to suppose that all cases fall cIearly and neatly into
one or another of these two cIasses. A speaker whose job it is to do so may offer
information, instructions, or even advice, and yet be overtly indifferent as to
whether or not his information is accepted as such, his instructions followed, or his
advice taken. His wholly overt intention mayamount to no more than that of
making available-in a 'take it or leave it' spirit-4o his audience the information
or instructions or opinion in question ; though again, in some cases, he may be seen
as the mouthpiece, mereIy, of another agency to which may be attributed at least
general intentions of the kind that can scarcely be attributed, in the particular case,
to hirn. We should not find such complications discouraging; for we can scarcely
expect a general account of linguistic communication to yield more than 'schematic
outlines, which may almost be lost to view when every qualification is added which
fideIity to the facts requires.
III
WHAT IS A SPEECH ACT?
J. R. SEARLE
I. INTRODUCTION
IN a typical speech situation involving a speaker, a hearer, and an utterance by the
speaker, there are many kinds of acts associated with the speaker's utterance. The
speakef will characteristicaUy have moved his jaw and tongue and made noises. In
addition, he will characteristically have performed some acts within the c1ass which
inc1udes informing or irritating or boring rus hearers; he will further characteristically
have perfarmed some acts within the c1ass which inc1udes referring to Kennedy or
Khruschchev or the North Pole; and he will also have performed acts witrun the
c1ass which inc1udes making statements, asking questions, issuing commands, giving
reports, greeting, and warning. The members of this last c1ass are what Austin 1 called
illocutionary acts and it is with this c1ass that I shall be concerned in this paper, so
the paper might have been caUed 'What is an Illocutionary Act?' I do not attempt
to define the expression 'illocutionary act', although if my analysis of a particular
illocutionary act succeeds it may provide the basis for adefinition. Some of the
English verbs and verb phrases associated with illocutionary acts are: state, assert,
describe, warn, remark, comment, command, order, request, criticize, apologize,
censure, approve, we1come, promise, express approval, and express regret. Austin
c1aimed that there were over a thousand such expressions in English.
By way of introduction, perhaps I can say why I trunk it is of interest and im-
portance in the philosophy of language to study speech acts, or, as they are some-
times called, language acts or linguistic acts. I think it is essential to any specimen of
linguistic communication that it involve a linguistic act. It is not, as has generally been
supposed, the symbol or ward or sentence, or even the token ofthe symbol or word
or sentence, wruch is the unit of linguistic communication, but rather it is the
production of the token in the performance of the speech act that constitutes the
basic unit of linguistic communication. To put trus point more precisely, the
production of the sentence token under certain conditions is the illocutionary act, and
the illocutionary act is the minimal unit of linguistic communication.
From Philosophy in AmericQ, ed. Max Black (Allen & Unwin, 1965), pp. 221-39. Reprinted by pennission
of the author, (;grnell University Press and George Allen & Unwin Ud.
I J. L. Austin, How To Do Things with Words (Oxford, 1962).
40 WHAT IS A SPEECH ACT?
I do not know how to prove that linguistic communication essentially involves acts
but I can think of arguments with which one might attempt to convince someone who
was sceptical. One argument would be to call the sceptic's attention to the fact that
when he takes a noise or a mark on paper to be an instance of linguistic communica-
tion, as a message, one of the things that is involved in his so taking that noise or
mark is that he should regard it as having been produced by a being with certain
intentions. He cannot just regard it as a natural phenJmenon, like a stone, a water-
fall, or a tree. In order to regard it as an instance of linguistic communication one
must suppose that its production is what I am calling a speech act. It is a logical
presupposition, for example, of current attempts to decipher the Mayan hieroglyphs
that we at least hypothesize that the marks we see on the stones were produced by
beings more or less like ourselves and produced with certain kinds of intentions. If
we were certain the marks were a consequence of, say, water erosion, then the
question of deciphering them or even calling them hieroglyphs could not arise. To
constme them under the category of linguistic communication necessarily involves
construing their production as speech acts.
To perform illocutionary acts is to engage in a mle-governed form of behaviour.
I shall argue that such things as asking questions or making statements are mle-
governed in ways quite similar to those in which getting a base hit in baseball or
moving a knight in chess are rule-governed forms of acts. I intend therefore to
explicate the notion of an illocutionary act by stating a set of necessary and suf-
ficient conditions for the performance of a particular kind of illocutionary act, and
extracting from it a set of semantical mies for the use ofthe expression (or syntactic
device) which marks the utterance as an illocutionary act of that kind. If I am suc-
cessful in stating the conditions and the corresponding rules for even one kind of
illocutionary act, that will provide us with a pattern for analysing other kinds of acts
and consequently for explicating the notion in general. But in order to set the stage
for actually stating conditions and extracting mIes for performing an illocutionary
act I have to discuss three other preliminary notions : ru/es, propositions, and mean-
ing. I shall confine my discussion of these notions to those aspects which are essential
to my main purposes in this paper, but, even so, what I wish to say concerning each
of these notions, if it were to be at all complete, would require a paper for each;
however, sometimes it may be worth sacrificing thoroughness for the sake of scope
and I shall therefore be very brief.
11. RULES
In recent years there has been in the philosophy of language considerable dis-
cussion involving the notion of mies for the use of expressions. Some philosophers
have even said that knowing the meaning of the word is simply a matter of knowing
WHAT IS A SPEECH ACT? 41
the rules for its use or employment. One disquieting feature of such discussions is that
no philosopher, to my knowledge at least, has ever given anything like an adequate
formulation of the rules for the use of even one expression. If meaning is a matter
of rules of use, surely we ought to be able to state the rules for the use of expressions
in a way which would explicate the meaning of those expressions. Certain
other philosophers, dismayed perhaps by the failure of their colleagues to pro-
duce any rules, have denied the fashionable view that meaning is a matter of
rules and have asserted that there are no semantical rules of the proposed kind
at all. I am inclined to think that this scepticism is premature and sterns from a
failure to distinguish different sorts of rules, in a way which I shall now attempt
to explain.
I distinguish between two sorts of rules: Some regulate antecedently existing forms
of behaviour; for example, the rules of etiquette regulate interpersonal reiationships,
but these relationships exist independently of the rules of etiquette. Some rules on the
other hand do not merely regulate but create or define new forms of behaviour. The
rules of football, fOT example, do not mereiy regulate the game of football but as it
were create the possibility of or define that activity. The activity of playing football
is constituted by acting in accordance with these rules; football has no existence
apart from these rules. I call the latter kind of rules constitutive rules and the former
kind regulative rules. Regulative rules regulate a pre-existing activity, an activity
whose existence is logically independent of the existence of the rules. Constitutive
rules constitute (and also regulate) an activity the existence of which is logically
dependent on the rules. 2
Regulative rules characteristically take the form of or can be paraphrased as
imperatives, e.g. 'When cutting food hold the knife in the right hand', or 'Officers are
to wear ti es at dinner'. Some constitutive rules take quite a different form, e.g. a
checkmate is made if the king is attacked in such a way that no move will leave it
unattacked; a touchdown is scored when a player crosses the opponents' goal li ne in
possession of the ball while play is in progress. If our paradigms of rules are impera-
tive regulative rules, such non-imperative constitutive rules are likely to strike us as
extremely curious and hardly even as rules at all. Notice that they are almost tauto-
logical in character, for what the 'rule' seems to offer is a partial definition of'check-
mate' or 'touchdown'. But, of course, this quasi-tautological character is a necessary
consequence of their being constitutive rules: the rules concerning touchdowns must
define the notion of 'touchdown' in the same way that the rules concerning football
define 'football'. That, for example, a touchdown can be scored in such and such
ways and counts six points can appear sometimes as a rule, sometimes as an analytic
truth; and that it can be construed as a tautology is a elue to the fact that the rule
in question is a constitutive one. Regulative rules generally have the form '00 X' or
2 This distinction occurs in J. Rawls. 'Two Concepts of Rules', Philosophh'al Rel'iew, 1955, and J. R.
Searle, 'How to Derive "Ought" from "Is"', Philosophim/ Rel'iew, 1964.
42 WHAT 1S A SPEECH ACT?
'If Y do X'. Some members of the set of constitutive rules have this form but some
also have the form 'X counts as Y'.3
The failure to perceive this is of some importance in philosophy. Thus, e.g., some
philosophers ask 'How can a promise create an obligation?' A similar question would
be 'How can a touchdown create six points?' And as they stand both questions can
only be answered by stating a rule of the form 'X counts as Y'.
I am inclined to think that both the failure of some philosophers to state rules for
the use of expressions and the scepticism of other philosophers concerning the exist-
ence of any such rules stern at least in part from a failure to recognize the distinctions
between constitutive and regulative rules. The model or paradigm of a rule which
most philosophers have is that of a regulative rule, and if one looks in semantics for
purely regulative rules one is not likely to find anything interesting from the point
of view of logical analysis. There are no doubt social rules of the form 'One ought not
to utter obscenities at formal gatherings', but that hardly seems a rule of the sort
that is crucial in explicating the semantics of a language. The hypothesis that lies
behind the present paper is that the semantics of a language can be regarded as a
series of systems of constitutive rules and that illocutionary acts are acts performed
in accordance with these sets of constitutive rules. One of the aims of this paper is
to formulate a set of constitutive rules for a certain kind of speech act. And if what
I have said concerning constitutive rules is correct, we should not be surprised if not
a11 these rules take the form of imperative rules. Indeed we shall see that the rules fall
into several different categories, none of which is quite like the rules of etiquette. The
effort to state the rules for an illocutionary act can also be regarded as a kind of test
of the hypothesis that there are constitutive rules underlying speech acts. If we are
unable to give any satisfactory rule formulations, our failure could be construed as
partially disconfirming evidence against the hypothesis.
111. PROPOSITIONS
Different iIIocutionary acts often have features in common with each other. Consider
utterances of the following sentences :
(1) Will John leave the room?
(2) John will leave the room.
(3) John, leave the room!
(4) Would that John left the room.
(5) If John will leave the room, I will leave also.
Utterances of each of these on a given occasion would characteristically be perform-
ances of different illocutionary acts. The first would, characteristically, be a question,
the second an assertion about the future, that is, aprediction, the third arequest or
J The formulation 'x counts as Y' was originally suggested to me by Max Black.
WHAT IS A SPEECH ACT? 43
order, the fourth an expression of a wish, and the fifth a hypothetical expression of
intention. Yet in the performance of each the speaker would characteristically per-
form some subsidiary acts which are common to all five illocutionary acts. In the
utterance of each the speaker refers to a particular person John and predicates the
act of leaving the room of that person. In no case is that all he does, but in every
case it is apart of what he does. I shall say, therefore, that in each of these cases,
although the illocutionary acts are different, at least some of the non-illocutionary
acts of reference and predication are the same.
The reference to some person John and predication of the same thing of him in
each of these illocutionary acts inclines me to say that there is a common content in
each of them. Something expressible by the dause 'that John will leave the room'
seems to be a common feature of all. We could, with not too much distortion, write
each of these sentences in a way which would isolate this common feature: 'I assert
that John willleave the room', 'I ask whether John will leave the room', etc.
For lack of a better word I propose to call this common content a proposition, and
I shall describe this feature of these illocutionaty acts by saying that in the utterance
of each of (1)-(5) the speaker expresses the proposition that John will leave the
room. Notice that I do not say that the sentence expresses the proposition; I do not
know how sentences could perform acts of that kind. But I shall say that in the
utterance of the sentence the speaker expresses a proposition. Notice also that I am
distinguishing between a proposition and an assertion or statement of that proposi-
tion. The proposition that John willleave the room is expressed in the utterance of all
of (1)-(5) but only in (2) is that proposition asserted. An assertion is an illocutionary
act, but a proposition is not an act at all, although the act of expressing a proposition
is apart of performing certain illocutionary acts.
I might summarize this by saying that I am distinguishing between the illocutionary
act and the propositional content of an iUocutionary act. Of course, not all illocu-
tionary acts have a prqpositional content, for example, an utterance of 'Hurrah!' or
'Ouch!' does not. In one version or another this distinction is an old one and has been
marked in different ways by authors as diverse as Frege, Sheffer, Lewis, Reichenbach
and Hare, to mention only a few.
From a semantical point of view we can distinguish between the propositional
indicator in the sentence and the indicator of iUocutionary force. That is, for a large
dass of sentences used to perform illocutionary acts, we can say for the purpose of
our analysis that the sentence has two (not necessarily separate) parts, the proposi-
tion-indicating element and the function-indicating device.
4
The function-indicating
device shows how the proposition is to be taken, or, to put it in another way, what
illocutionary force the utterance is to have, that is, what illocutionary act the speaker
4 In the senten ce 'I promise that I will come' the function-indicating device and the propositional element
are separate. In the sentence 'I promise to come', which me ans the same as the first and is derived from it by
certain transformations, the two elements are not separate.
44 WHAT IS A SPEECH ACT?
is performing in the utterance ofthe sentence. Function-indicating devices in English
include word order, stress, intonation contour, punctuation, the mood ofthe verb, and
finally a set of so-called performative verbs: I may indicate the kind of illocutionary
act I am performing by beginning the senten ce with 'I apologize', 'I warn', 'I state',
etc. Often in actual speech situations the context will make it clear what the illocu-
tionary force of the utterance is, without its being necessary to invoke the appropriate
function indicating device.
If this semantical distinction is of any real importance, it seems likely that it should
have some syntactical analogue, and certain recent developments in transformational
grammar tend to support the view that it does. In the underlying phrase marker of a
sentence there is a distinction between those elements which correspond to the
function-indicating device and those which correspond to the propositional content.
The distinction between the function-indicating device and the proposition-
indicating device will prove very useful to us in giving an analysis of an illocutionary
act. Since the same proposition can be common to all sorts of illocutionary acts, we
can separate our analysis of the proposition from our analysis of kinds of illocu-
tionary acts. I think there are rules for expressing propositions, rules for such things
as reference and prediction, but those rules can be discussed independently of the
rules for function indicating. In this paper I shall not attempt to discuss propositional
rules but shall concentrate on rules for using certain kinds of function-indicating
devices.
IV. MEANING
Speech acts are characteristically performed in the utterance of sounds or the making
of marks. What is the difference between just uttering sounds or making marks and
performing a speech act? One difference is that the sounds or marks one makes in the
performance of a speech act are characteristically said to have meaning, and a second
related difference is that one is characteristically said to mean something by those
sounds or marks. Characteristically when one speaks one means something by what
one says, and what one says, the string of morphemes that one emits, is characteristi-
cally said to have a meaning. Here, incidentally, is another point at which our analogy be-
tween performing speech acts and playing games breaks down. The pieces in agame
like chess are not characteristically said to have a meaning, and furthermore when
one makes a move one is not characteristically said to mean anything by that move.
But what is it for one to me an something by what one says, and what is it for
something to have a meaning? To answer the first of these questions I propose to
borrow and revise some ideas of Paul Grice. In an article entitled 'Meaning',5 Grice
gives the following analysis of one sense of the notion of 'meaning'. To say that A
meant something by x is to say that 'A intended the utterance of x to produce some
'Philosophical Review, 1957.
WHAT IS A SPEECH ACT?
45
effect in an audience by means of the recognition ofthis intention'. This seems to me
a useful start on an analysis of meaning, first because it shows the elose reIationship
between the notion of meaning and the notion of intention, and secondly because
it captures something which is, I think, essential to speaking a language : In speaking
a language I attempt to communicate things to my hearer by means of getting hirn
to recognize my intention to communicate just those things. For example, character-
istically, when I make an assertion, I attempt to communicate to and convince my
hearer of the truth of a certain proposition; and the means I employ to do this are to
utter certain sounds, which utterance I intend to produce in him the desired effect
by means of his recognition of my intention to produce just that effect. I shall illus-
trate this with an example. I might on the one hand attempt to get you to believe that
I am French by speaking French all the time, dressing in the French manner, showing
wild enthusiasm for de Gaulle, and cultivating French acquaintances. But I might on
the other hand attempt to get you to beIieve that I am French by simply telling you
that I am French. Now, what is the difference between these two ways ofmy attempt-
ing to get you to beIieve that I am French? One crucial difference is that in the second
case I attempt to get you to believe that I am French by getting you to recognize that
it is my purported intention to get you to beIieve just that. That is one of the things
involved in telling you that I am French. But of course if I try to get you to beIieve
that I am French by putting on the act I described, then your recognition of my
intention to produce in you the beliefthat I am French is not the means I am employ-
ing. Indeed in this case you would, I think, become rather suspicious if you recognized
my intention.
However valuable this analysis of meaning is, it seems to me to be in certain
respects defective. First of all, it fails to distinguish the different kinds of effects-
perlocutionary versus illocutionary---that one may intend to produce in one's
hearers, and it further fails to show the way in wh ich these different kinds of
effects are related to the notion of meaning. A second defect is that it fails to account
for the extent to which meaning is a matter of rules or conventions. That is, this
account of meaning does not show the connection between one's meaning some-
thing by what one says and what that which one says actually means in the language.
In order to illustrate this point I now wish to present a counter-example to this
analysis of meaning. The point of the counter-ex am pie will be to illustrate the con-
nection between what a speaker means and what the words he utters mean.
Suppose that I am an American soldier in the Second World War and that I am
captured by Italian troops. And suppose also that I wish to get these troops to beIieve
that I am a German officer in order to get them to release me. What I would like to
do is to tell them in German or Italian that I am a German officer. But let us suppose
I don't know enough German or Italian to do that. So I, as it were, attempt to put on
a show of telling them that I am a German officer by reciting those few bits of
German that I know, trusting that they don't know enough German to see through
46 WHAT IS A SPEECH ACT?
my plan. Let us suppose 1 know only one line of German, which 1 remember from a
poem 1 had to memorize in a high-school German course. Therefore I, a captured
American, address my Italian captors with the following senten ce : 'Kennst du das
Land, wo die Zitronen blühen?' Now, let us describe the situation in Gricean terms.
1 intend to produce a certain effect in them, namely, the effect ofbelieving that 1 am
a German officer; and 1 intend to produce this effect by means of their recognition of
my intention. 1 intend that they should think that what 1 am trying to tell them is that
1 am a German offker. But does it follow from this account that when 1 say 'Kennst
du das Land .. .' etc., what 1 me an is, 'I am a German officer'? Not only does it not
follow, but in this case it seems plainly false that when 1 utter the German sentence
what 1 mean is 'I am a German officer', or even 'Ich bin ein deutscher Offizier',
because what the words mean is, 'Knowest thou the land where the lemon trees
bloom?' Of course, 1 want my captors to be deceived into thinking that what 1 mean
is 'I am a German officer', but part of what is involved in the deception is getting
them to think that that is what the words which 1 utter mean in German. At one
point in the Philosophical Inl'estigations Wittgenstein says 'Say "it's cold here" and
mean "it's warm here"'. 6 The reason we are unable to do this is that what we can
mean is a function of what we are saying. Meaning is more than a matter of intention,
it is also a matter of convention.
Grice's account can be amended to deal with counter-examples of this kind. We
have here a case where 1 am trying to produce a certain effect by means of the
recognition of my intention to produce that effect, but the device I use to produce
this effect is one which is conventionally, by the rules governing the use ofthat device,
used as a means of producing quite different illocutionary effects. We must therefore
reformulate the Gricean account of meaning in such a way as to make it c1ear that
one's meaning something when one says something is more than just contingently
related to what the sentence me ans in the Ianguage one is speaking. In our analysis
of illocutionary acts, we must capture both the intentional and the conventional
aspects and especially the relationship between them. In the performance of an
illocutionary act the speaker intends to produce a certain effect by me ans of getting
the hearer to recognize his intention to produce that effect, and furthermore, ifhe is
using words Iiterally, he intends this recognition to be achieved in virtue of the fact
that the rules for using the expressions he utters associate the expressions with the
production of that effect. It is this combination of elements which we shall need to
express in our analysis of the illocutionary act.
V. HOW TO PROMISE
I shall now attempt to give an analysis of the illocutionary act of promising. In order
to do this 1 shall ask what conditions are necessary and sufficient for the act of
'Philosophicallnvestigations (Oxford, 1953\ para. 510.
WHAT IS A SPEECH ACT? 47
promising to have been performed in the utterance of a given sentence. I shall attempt
to answer this question by stating these conditions as a set of propositions such that
the conjunction of the members of the set entails the proposition that a speaker made
a promise, and the proposition that the speaker made a promise entails this conjunc-
tion. Thus each condition will-be a necessary condition for the performance of the act
of promising and taken collectively the set of conditions will be a sufficient condition
for the act to have been performed.
If we get such a set of conditions we can extract from them a set of rules for the
use of the function-indicating device. The method here is analogous to discovering
the rules of chess by asking oneself what are the necessary and sufficient conditions
under which one can be said to have correctly moved a knight or castled or check-
mated a player, etc. We are in the position of someone who has learned to play chess
without ever having the rules formulated and who wants such a formulation. We
learned how.to play the game ofillocutionary acts, but in general it was done without
an explicit formulation of the rules, and the first step in getting such a formulation is
to set out the conditions for the performance of a particular illocutionary act. Our
inquiry will therefore serve a double philosophical purpose. By stating a set of con-
ditions for the performance of a particular illocutionary act we shall have offered a
partial explication of that notion and shall also have paved the way for the second
step, the formulation of the rules.
I find the statement of the conditions very difficult to do, and I am not entirely
satisfied with the list I am about to present. One reason for the difficulty is that the
notion of a promise, like most notions in ordinary language, does not have absolutely
strict rules. There are all sorts of odd, deviant, and borderline promises ; and counter-
examples, more or less bizarre, can be produced against my analysis. I am incIined
to think we shall not be able to get a set of knock-down necessary and sufficient
conditions that will exactly mirror the ordinary use of the word 'promise'. I am
confining my discussion, therefore, to the centre of the concept of promising and
ignoring the fringe, borderline, and partially defective cases. I also confine my dis-
cussion to full-blown explicit promises and ignore promises made by elliptical turns
of phrase, hints, metaphors, etc.
Another difficulty arises from my desire to state the conditions without certain
forms of circularity. I want to give a list of conditions for the performance of a certain
iIIocutionary act, which do not themselves mention the performance of any iIIocu-
tionary acts. I need to satisfy this condition in order to offer an explication of the
notion of an illocutionary act in general, otherwise I should simply be showing the
relation between different iIIocutionary acts. However, although there will be no
reference to iIIocutionary acts, certain iIIocutionary concepts will appear in the
analysans as weil as in the analysandum; and I think this form of circularity is un-
avoidable because of the nature of constitutive rules.
In the presentation of the conditions I shall first consider the case of a sincere
48 WHAT IS A SPEECH ACT?
promise and then show how to modify the conditions to aIlow for insincere promises.
As our inquiry is semantical rather than syntactical, I shall simply assume the exist-
ence of grammatically well-formed sentences.
Given that a speaker S utters a sentence Tin the presence of a hearer H, then, in the
utterance of T, S sincerely (and non-defectively) promises that p to H if and only if:
(1) Normal input and output conditions obtain.
I use the terms 'input' and 'output' to cover the large and indefinite range of condi-
tions under which any kind of serious linguistic communication is possible. 'Output'
covers the conditions for intelligible speaking and 'input' covers the conditions for
understanding. Together they include such things as that the speaker and hearer both
know how to speak the language; both are conscious of what they are doing; the
speaker is not acting under duress or threats; they have no physical impediments to
communication, such as deafness, aphasia, or laryngitis; they are not acting in a play
or telling jokes, etc.
(2) Sexpresses that p in the utterance of T.
This condition isolates the proposition al content from the rest of the speech act and
enables us to concentrate on the peculiarities of promising in the rest of the analysis.
(3) In expressing that p, S predicates afuture act A of S.
In the case of promising the function-indicating device is an expression whose scope
includes certain features of the proposition. In a promise an act must be predicated
of the speaker and it cannot be a past act. I cannot promise to have done something,
and I cannot promise that someone else will do something. (Although I can promise
to see that he will do it.) The notion of an act, as I am construing it for present
purposes, includes refraining from acts, performing series of acts, and mayaiso
include states and conditions: I may promise not to do something, I may promise to
do something repeatedly, and I may promise to be or remain in a certain state or
condition. I call conditions (2) and (3) the propositional content conditions.
(4) H would prefer S's doing A to his not doing A, and S believes H would prefer his
doing A to his not doing A.
One crucial distinction between promises on the one hand and threats on the other
is that a promise is a pledge to do something for you, not to you, but a threat is a
pledge to do something to you, not for you. A promise is defective if the thing
promised is something the promisee does not want done; and it is further defective if
the promisor does not believe the promisee wants it done, since a non-defective promise
must be intended as a promise and not as a threat or waming. I think both halves of
this double condition are necessary in order to avoid fairly obvious counter-examples.
One can, however, think of apparent counter-examples to this condition as stated.
WHAT IS A SPEECH ACT? 49
Suppose I say to a lazy student 'If you don't hand in your paper on time I promise
you I will give you a failing grade in the course'. Is this utterance a promise ? I am
inc1ined to think not; we would more naturally describe it as a warning or possibly
even a threat. But why then is it possible to use the locution 'I promise' in such a
case? I think we use it here because 'I promise' and 'I hereby promise' are among the
strongest function-indicating devices for commitment provided by the English
language. For that reason we often use these expressions in the performance of
speech acts which are not strictly speaking promises but in which we wish to
emphasize our commitment. To illustrate this, consider another apparent counter-
example to the analysis along different lines. Sometimes, more commonly I think in
the United States than in England, one hears people say 'I promise' when making an
emphatic assertion. Suppose, for example, I accuse you of having stolen the money.
I say, 'You stole that money, didn't you?' You reply 'No, I didn't, I promise you I
didn't'. Did you make a promise in this case? I find it very unnatural to describe your
utterance as a promise. This utterance would be more aptly described as an emphatic
denial, and we ean explain the occurrence of the function-indicating device 'I
promise' as derivative from genuine promises and serving here as an expression
adding emphasis to your denial.
In general the point stated in condition (4) is that if a purported promise is to be
non-defective the thing promised must be something the hearer wants done, or
considers to be in his interest, or would prefer being done to not being done, ete. ;
and the speaker must be aware of or believe or know, etc., that this is the ease. I
think a more elegant and exact formulation of this condition would require the
introduction of technical terminology.
(5) It is not obvious to both Sand H that S will do A in the nonnal course of events.
This condition is an instanee of a general condition on many different kinds of illoeu-
tionary acts to the effect that the act must have a point. For example, if I make a
request to someone to do something whieh it is obvious that he is already doing or is
about to do, then my request is pointless and to that extent defeetive. In an actual
speech situation, listeners, knowing the rules for performing illocutionary acts, will
assurne that this condition is satisfied. Suppose, for example, that in the course of a
publie speech I say to a member of my audience 'Look here, Smith, pay attention to
what I am saying'. In order to make sense of this utterance the audience will have
to assurne that Smith has not been paying attention or at any rate that it is not
obvious that he has been paying attention, that the question of his paying attention
has arisen in some way; because a condition for making arequest is that it is not
obvious that the hearer is doing or about to do the thing requested.
Similarly with promises. It is out of order for me to promise to do something that
it is obvious I am going to do anyhow. If I do seem to be making such a promise, the
only way my audience can make sense of my utterance is to assurne that I believe
50 WHAT IS A SPEECH ACT?
that it is not obvious that I am going to do the thing promised. A happily married
man who promises his wife he will not desert her in the next week is likely to
provide more anxiety than comfort.
Parenthetically I think this condition is an instance of the sort of phenomenon
stated in Zipf's law. I think there is operating in our language, as in most forms of
human behaviour, a principle of least effort, in this case a principle of maximum
illocutionary ends with minimum phonetic efTort; and I think condition (5) is an
instance of it.
I call conditions such as (4) and (5) preparatory conditions. They are sine quibus
non of happy promising, but they do not yet state the essential feature.
(6) S intends to do A.
The most important distinction between sincere and insincere promises is that in
the case of the insincere promise the speaker intends to do the act promised, in the
case of the insincere promise he does not intend to do the act. Also in sincere
promises the speaker believes it is possible for him to do the act (or refrain from
doing it), but I think the proposition that he intends to do it entails that he thinks it
is possible to do (or refrain from doing) it, so I am not stating that as an extra condi-
tion. I call this condition the sincerity condition.
(7) S intends that the uUerance of T will place him under an obligation to do A.
The essential feature of a promise is that it is the undertaking of an obligation to
perform a certain act. I think that this condition distinguishes promises (and other
members of the same family such as vows) from other kinds of speech acts. Notice
that in the statement of the condition we only specify the speaker's intention; further
conditions will make dear how that intention is realized. It is dear, however, that
having this intention is a necessary condition of making a promise ; for if a speaker
can demonstrate that he did not have this intention in a given utterance, he can prove
that the utterance was not a promise. We know, for example, that Mr. Pickwick did
not promise to marry the woman because we know he did not have the appropriate
intention.
I call this the essential condition.
(8) S intends that the utterance ofT will produce in Habeliefthat conditions (6) and
(7) obtain by means of the recognition ofthe intention to produce that belief, and
he intends this recognition to be achieved by means of the recognition of the
sentence as one conventionally used to produce such beliefs.
This captures our amended Gricean analysis of what it is for the speaker to mean to
make a promise. The speaker intends to produce a certain iIIocutionary effect by
means of getting the hearer to recognize his intention to produce that efTect, and he
also intends this recognition to be achieved in virtue of the fact that the lexical and
WHAT IS A SPEECH ACT? 51
syntactical character of the item he utters conventionally associates it with producing
that effeCt.
Strictly speaking this condition could be formulated as part of condition (1), but
it is of enQugh philosophical interest to be worth stating separately. I find it trouble-
some for the following reason. If my original objection to Grice is really valid, then
surely, one might say, all these iterated intentions are superfluous; all that is neces-
sary is that the speaker should seriously utter a sentence. The production of all these
effects is simply a consequence of the hearer's knowledge of what the sentence means,
which in turn is a consequence of his knowledge of the language, which is assumed
by the speaker at the outset. 1 think the correct reply to this objection is that condi-
ti on (8) explicates what it is for the speaker to 'seriously' utter the sentence, i.e. to
utter it and mean it, but I am not completely confident about either the force of the
objection or of the reply.
(9) The sernantical rules ofthe dialect spoken by Sand H are such that T is correctly
and sincerely uttered if and only if conditions (1 )--(8) obtain.
This condition is intended to make clear that the sentence uttered is one which by the
semantical mies of the language is used to make a promise. Taken together with
condition (8), it e1iminates counter-ex am pIes like the captured soldier ex am pie con-
sidered earlier. Exactly what the formulation of the rules is, we shall soon see.
So far we have considered only the case of a sincere promise. But insincere
promises are promises none the less, and we now need to show how to modify the
conditions to allow for them. In making an insincere promise the speaker does not
have all the intentions and beliefs he has when making a sincere promise. However,
he purports to have them. Indeed it is because he purports to have intentions and
beliefs which he does not have that we describe his act as insincere. So to allow for
insincere promises we need only to revise our conditions to state that the speaker
takes responsibility for having the beliefs and intentions rather than stating that he
actually has them. A clue that the speaker does take such responsibility is the fact
that he could not say without absurdity, e.g., 'I promise to do A but I do not intend to
do A'. To say 'I promise to do A' is to take responsibility for intending to do A, and
this condition holds whether the utterance was sincere or insincere. To allow for the
possibility of an insincere promise then we have only to revise condition (6) so that it
states not that the speaker intends to do A, but that he takes responsibility for intend-
ing to do A, and to avoid the charge of circularity I shall phrase this as folIows:
(6*) S intends that the utterance of T will make hirn responsible for intending to
doA.
Thus amended (and with 'sincerely' dropped from our analysandum and from condi-
tion (9», our analysis is neutral on the question whether the promise was sincere or
insincere.
52 WHAT IS A SPEECH ACT?
VI. RULES FOR THE USE OF THE FUNCTION-INDICATING DEVICE
Our next task is to extract from our set of conditions a set of rules for the use of the
function-indicating device. Obviously not all of our conditions are equally relevant to
this task. Condition (1) and conditions of the forms (8) and (9) apply generally to all
kinds of normal illocutionary acts and are not peculiar to promising. Rules for the
function-indicating device for promising are to be found corresponding to condi-
tions (2)-(7).
The semantical rules for the use of any function-indicating device P for promising
are:
Rufe 1. P is to be uttered only in the context of a sentence (or larger stretch of
discourse) the utterance of which predicates some future act A of the speaker S.
I call this the propositionaf-content rufe. It is derived from the propositional-content
conditions (2) and (3).
Rufe 2. P is to be utterred only if the hearer H would prefer S's doing A to his not
doing A, and S believes H would prefer S's doing A to his not doing A.
Rufe 3. P is to be uttered only if it is not obvious to both Sand H that S will do A
in the normal course of events.
I call rules (2) and (3) preparatory rufes. They are derived from the preparatory
conditions (4) and (5).
Rufe 4. P is to be uttered only if S intends to do A.
I call this the sincerity rufe. It is derived from the sincerity condition (6).
Rufe 5. The utterance of P counts as the undertaking of an obligation to do A.
I call this the essential rufe.
These rules are ordered: rules 2-5 apply only if rule 1 is satisfied, and rule 5 applies
only if rules 2 and 3 are satisfied as weil.
Notice that whereas rules 1-4 take the form of quasi-imperatives, i.e. they are of
the form: utter P only if x, rule 5 is of the form: the utterance of P counts as Y. Thus
rule 5 is of the kind peculiar to systems of constitutive rules which I discussed in
section 11.
Notice also that the rather tiresome analogy with games is holding up remarkably
weil. If we ask ourselves under what conditions a player could be said to move a
knight correct1y, we would find preparatory conditions, such as that it must be his
turn to move, as weil as the essential condition stating the actual positions the knight
can move to. I think that there is even a sincerity rule for competitive games, the rule
that each side tries to win. I suggest that the team which 'throws' the game is behav-
ing in a way c10sely analogous to the speaker who lies or makes false prornises. Of
course, there usually are no propositional-content rules for games, because games do
not, by and Iarge, represent states of affairs.
If this analysis is of any general interest beyond the case of promising then it would
seem that these distinctions should carry over into other types of speech act, and I
WHAT IS A SPEECH ACT? 53
think a little reflection will show that they do. Consider, e.g., giving an order. Thc
preparatory conditions include that the speaker should be in a position of authority
over the hearer, the sincerity condition is that the speaker wants the ordered act done,
and the essential condition has to do with the fact that the utterance is an attempt to
get the hearer to do it. For assertions, the preparatory conditions include the fact
that the hearer must have some basis for supposing the asserted proposition is true,
the sincerity condition is that he must believe it to be true, and the essential condition
has to do with the fact that the utterance is an attempt to inform the hearer and
convince hirn of its truth. Greetings are a much simpler kind of speech act, but even
here some of the distinctions apply. In the utterance of 'Hello' there is no proposi-
tional content and no sincerity condition. The preparatory conditon is that the
speaker must have just encountered the hearer, and the essential rule is that the
utterance indicates courteous recognition of the hearer.
A proposal for further research then is to carry out a similar analysis of other
types of speech acts. Not only would this give us an analysis of concepts interesting
in themselves, but the comparison of different analyses would deepen our under-
standing of the whole subject and incidentally provide a basis for a more serious
taxonomy than any of the usual facile categories such as evaluative versus descrip-
tive, or cognitive versus emotive.
IV
UTTERER'S MEANING, SENTENCE-MEANING,
AND WORD-MEANING'
H. P. GRICE
A. PROLEGOMENA
My aim in this paper is to throw light on the connection between (a) a notion of
meaning which I want to regard as basic, viz. that notion which is involved in saying
of someone that by (when) doing such-and-such he meant that so-and-so (in what I
have called a non-natural sense of the word 'meant'), and (b) the notions of meaning
involved in saying (i) that a given senten ce means 'so-and-so' (ii) that a given word
or phrase means 'so-and-so'. What I have to say on these topics should be looked
upon as an attempt to provide a sketch of what might, I hope, prove to be a viable
theory, rather than as an attempt to provide any part of a finally acceptable theory.
The account which I shall otTer of the (for me) basic notion of meaning is one which
I shall not today seek to defend; I should like its approximate correctness to be
assumed, so that attention may be focused on its utility, if correct, in the explication
of other and (I hope) derivative notions of meaning. This enterprise forms part of a
wider programme which I shall in a moment delineate, though its later stages lie
beyond the limits which I have set for this paper.
The wider programme just mentioned arises out of a distinction which, for purposes
which I need not here specify, I wish to make within the total signification ofaremark:
a distinction between what the speaker has said (in a certain favoured, and maybe in
some degree artificial, sense of 'said'), and what he has 'implicated' (e.g. implied,
indicated, suggested, etc.), taking into account the fact that wh at he has implicated
may be either conl'entionally implicated (implicated by virtue ofthe meaning ofsome
word or phrase which he has used) or non-conl'entionally implicated (in which casethe
specification ofthe implicature falls outside the specification ofthe conventional meaning
of the words used). The programme is directed towards an explication of the favoured
sense of 'say' and a clarification of its relation to the notion of conventional meaning.
The stages of the programme are as folIows:
(I) To distinguish between locutions of the form 'U (utterer) meant that .. .' (locu-
tions which specify what rnight be called 'occasion-meaning') and locutions of the
From Foundalions oJ Language. 4 (1968), pp. 1-18. Reprinted by permission of the author and the editor
of Foundations oJ Language.
I I hope that material in this paper, revised and re·arranged, will form part of a book to be published by
the Harvard University Press.
AND WORD MEANING 55
form 'X (utterance-type) means H ••• "'. In locutions of the first type, meaning is
specified without the use of quotation-marks, whereas in locutions of the second
type the meaning of a sentence, word or phrase is specified with the aid of quotation-
marks. This difference is semantically important.
(11) To attempt to provide a definiens for statements of occasion-meaning; more
precisely, to provide a definiens for 'By (when) uttering x, U meant that *p'. Some
explanatory comments are needed here.
(a) I use the term 'utter' (together with 'utterance') in an artificially wide sense, to
cover any case of doing x or producing x by the performance of which U meant that
so-and-so. The performance in question need not be a linguistic or even a con-
ventionalized performance. A specificatory replacement of the dummy 'x' will in
some cases be a characterization of a deed, in others a characterization of a product
(e.g. asound).
(b) '*' is a dummy mood-indicator, distinct from specific mood-indicators like 'I-'
(indicative or assertive) or '!' (imperative). More precisely, one may think of the
schema 'Jones meant that *p' as yielding a full English sentence after two trans-
formation al steps:
(i) replace '*' by a specific mood-indicator and replace 'p' by an indicative sentence.
One might thus get to
'Jones meant that I- Smith will go home'
or to 'Jones meant that! Smith will go horne'.
(ii) replace the sequence following the word 'that' by an appropriate clause in in-
direct speech (in accordance with rules specified in a linguistic theory). One might
thus get to
'Jones meant that Srnith will go horne'
'Jones meant that Srnith is to go horne'.
(111) To attempt to elucidate the notion of the conventional meaning of an
utterance-type; more precisely, to explicate sentences which make claims of the
form 'X (utterance-type) means "*''', or, in case X is a non-scntcntial utterancc-
type, claims of the form 'X means H ••• "', where the location is completed by a non-
sentential expression. Again, some explanatory comments are required.
(a) It will be convenient to recognize that what I shall call statements of timeless
meaning (statements of the type 'X means " ... "', in which the of
meaning involves quotation-marks) may be subdivided into (i) statements of timeless
'idiolect-meaning', e.g. 'For U (in U's idiolect) X means " ... '" and (ü) statements of
timeless 'Ianguage meaning', e.g. 'In L (language) X means " ... "'. It will be con-
venient to handle these separately, and in the order just given.
(b) The truth of a statement to the effect that X means ' .. .' is of course not
incompatible with the truth of a further statement to the effect that X me ans '--",
when the two lacunae are quite differently completed. An utterance-type rriay
have more than one conventional meaning, and any definiens which we offer must
56 UTTERER'S MEANING, SENTENCE MEANING
allow fOT this fact. 'X means " ... '" should be understood as 'One of the meanings
of X is " ... " '.
(IV) In view of the possibility of multiplicity in the timeless meaning of an utter-
ance-type, we shall need to notice, and to provide an explication of, what I shall call
the applied timeless meaning of an utterance-type. That is to say, we need a
definiens for the schema 'X (utterance-type) meant here " ... "', a schema the
specifications of which announce the correct reading of X for a given occasion of
utterance.
Comments. (a) We must be careful to distinguish the applied timeless meaning of
X (type) with respecf to a particular token x (belonging to X) from the occasion-
meaning of U's utterance of x. The following are not equivalent:
(i) 'When U uttered it, the sentence "Palmer gave Nickiaus quite a beat-
ing" meant "Palmer vanquished Nickiaus with some ease" [rather
than, say, "Palmer administered vigorous corporal punishment to
NickIaus."]'
(ii) 'When U uttered the sentence "Palmer gave NickIaus quite a beating"
U meant that Palmer vanquished NickIaus with some ease.'
U might have been speaking ironically, in which case he would very likely have meant
that NickIaus vanquished Palmer with some ease. In that case (ii) would c1early be
false; but nevertheless (i) would still have been true.
(b) There is some temptation to take the view that the conjunction of
(i) 'By uttering X, U meant that *p' and
(ii) 'When uttered by U, X meant "*p'"
provides a definiens for 'In uttering X, U said that *p'. Indeed, ifwe give considera-
tion only to utterance-types for which there are available adequate statements of
time1ess meaning taking the exemplary form 'X meant "*p'" (or, in the case of
applied time1ess meaning, the form 'X meant here "*p" '), it may even be possible to
uphold the thesis that such a coincidence of occasion-meaning and applied time1ess
meaning is a necessary and sufficient condition for saying that *p. But a litde refiec-
tion should convince us of the need to recognize the existence of statements of time-
less meaning which instantiate forms other than the cited exemplary form; there are,
I think, at least some sentences whose timeless meaning is not adequately specifiable
by a statement of the exemplary form. Consider the sentence 'Bill is a philosopher
and he is, therefore, brave' (S ,). It would be appropriate, I think, to make a partial
specification of the timeless meaning of S, by saying 'Part of one meaning of S, is
"Bill is occupationally engaged in philosophical studies" '. One might, indeed, give a
full specifu::ation of timeless meaning for S, by saying 'One meaning of S, inc1udes
"Bill is occupationally engaged in philosophie al studies" and "Bill is courageous"
and "That Bill is courageous follows from his being occupationally engaged in philo-
sophical studies", and that is all that is included'. fWe might re-express this as 'One
meaning of S, comprises "Bill is occupationally engaged (etc)", "Bill is courageous",
AND WORD MEANING 57
and "That Bill is eourageous follows (ete . .)".'] It will be preferable to speeify the
timeless meaning of S I in this way than to do so as folIows: 'One meaning of S I is
"Bill is occupationally engaged (etc.) and Bill is courageous and that Bill is eoura-
geous follows (ete.)" '; for this latter formulation at least suggests that SI is synony-
mous with the conjunctive sentence quoted in the formulation, whieh does not seem
to be the case.
Since it is true that another meaning of SI inc1udes 'Bill is addicted to general
reftections about life' (l'ice 'Bill is occupationally engaged (etc.)'), one could have
occasion to say (truly), with respect to a given utterance by U of SI' 'The meaning
of SI here comprised "Bill is oecupationally engaged (ete.)", "Bill is eourageous", and
"That Bill is courageous follows (ete.)"', or to say 'The meaning of S I here included
"That Bill is courageous follows (etc.)" '. It could also be true that when U uttered SI
he meant (part of what he meant was) that that Bill is eourageous follows (ete.).
Now I do not wish to allow that, in my favoured sense of'say', one who utters SI
will have said that Bill's being courageous follows from his being a philosopher,
though he may weil have said that Bill is a philosopher and that Bill is courageous.
I would wish to maintain that the semantic function of the word 'therefore' is to
enable a speaker to indicate, though not to say, that a certain eonsequenee holds.
Mutatis mutandis, I would adopt the same position with regard to words like 'but'
and 'moreover'. My primary reason for opting for this partieular sense of'say' is that
I expect it to be of greater theoretical utility than some other sense of'say' would be.
So I shall be committed to the view that applied timeless meaning and occasion-
meaning may eoincide, that is to say, it may be true both (i) that when U uttered X
the meaning of X inc1uded '*p' and (ii) that part of what U meant when he uttered X
was that *p, and yet be false that U has said, among other things, that *p. I would
like to use the expression 'conventionally meant that' in such a way that the fulfil-
ment of the two conditions just mentioned, while insufficient for the truth of 'U said
that *p' will be suffieient (and neeessary) for the truth of 'U conventionally meant
that *p'.
(V) This distinction between what is said and wh at is eonventionally meant creates
the task of specifying the eonditions in whieh what U eonventionally meant by an
utterance is also part of what U said. I have hopes of being able to discharge this task
by proceeding along the following lines:
(1) To specify conditions whieh will be satisfied only by a limited range of speech-
acts, the members of which will thereby be stamped as speeially central or funda-
mental.
(2) To stipulate that in uttering X, U will have said that *p, ifboth (i) U has Y-ed
that *p, where Y-inR is a eentral speech-act, and (ii) X embodies some eonventional
deviee the meaning of which is such that its presenee in X indicates that its utterer is
Y -ing that *p.
(3) To define, for each member Y of the range of central speech-aets, 'U has Y -ed
58 UTTERER'S MEANING, SENTENCE MEANING
that *p' in terms of occasion-meaning (meaning that ... ) or in terms of some impor-
tant elementes) involved in the already provided definition of occasion-meaning.
(VI) The fulfilment of the task just outlined will need to be supplemented by an
account of the elements in the conventional meaning of an utterance which are not
part of what has been said. This account, at least for an important sub-class of such
elements, might take the following shape:
(1) The problematic elements are linked with certain speech-acts which are
exhibited as posterior to, and such that their performance is dependent upon, some
member or disjunction of members of the central range; for example, the meaning
of 'moreover' would be linked with the speech-act of adding, the performance of
which would require the performance of one or other of the central speech-acts.
(2) If Z-ing is such a non-central speech-act, the dependence of Z-ing that *p upon
the performance of some central speech-act would have to be shown to be of a nature
which justifies a reluctance to treat Z-ing that *p as a case not merely of saying that
*p, but also of saying that = p, or of saying that = *p (where' = p', or ' = *p', is a
representation of one or more sentential forms specifically associated with Z-ing). Z
(3) The notion of Z-ing that *p (where Z-ing is non-central) would be explicated in
terms of the nation of meaning that (or in terms of some important elementes) in the
definition of that notion).
B. TREATMENT OF SOME OF THE PROBLEMS RAlSED
The problems which I shall consider in the remainder of this paper are those which
are presented by Stages II -IV of the programme just outlined.
Stage 11 I shall otTer, without arguing for it, a somewhat over-simplified account of
the notion of occasion-meaning, which (as I said at the outset) I should like to be
treated as if it were correct.
In my 1957 article on 'Meaning'3 I in etTect suggested, for the schema 'U meant
(non-naturally) something by uttering x', a three-clause definiens which may be com-
pendiously reformulated as 'For some audience A, U intended his utterance of x to
produce in A some effect (response) E, by means of A's recognition ofthat intention'.
As I wish to continue to use the central idea of this definition, I shall introduce an
abbreviation; 'U intends to produce in A effect E by means of A's recognition ofthat
intention' will be abbreviated to 'U M-intends to produce in A etTect E'. ('M' for
'meaning'.)
The point of divergence between my current account and my 1957 account lies in
the characterization of the M-intended effect (response). In the earlier account I took
the view that the M-intended effect is, in the case of indicative-type utterances, that the
, As 'moreover _' is specifically associated with the speech-ac! of adding.
3 IPhilosophical Review, LXVII (1957).1 •
AND WORD MEANING 59
hearer should helieve something, and, in the case of imperative-type utterances, that
the hearer should do something. 1 wish for present purposes to make twochangeshere.
(1) 1 wish to represent the M-intended efTect ofimperative-type utterances as being
that the hearer should intend to do something (with, of course, the ulterior intention
on the part of the utterer that the hearer should go on to do the act in question).
(2) 1 wish to regard the M-intended efTect common to indicative-type utterances
as being, not that the hearer should believe something (though there will frequently
be an ulterior intention to that effect), but that the hearer should think that the utterer
believes something.
The effect of the first change will be that the way is opened to a simplified treat-
ment of the M-intended efTect, as being always the generation of some propositional
attitude. The efTect of the second change (made in order to unify the treatment of
indicative-type utterances, some of which are, and some of which are not, cases of
informing or telling) will be to introduce a distinction between what 1 might call
exhibitive utterances (utterances by which the utterer U M-intends to impart a belief
that he (U) has a certain propositional attitude) and utterances which are not only
exhibitive but also what 1 might call protreptic (utterances by which U M-intends,
via imparting a belief that he (U) has a certain propositional attitude, to induce a
corresponding attitude in the hearer).
1 shall now try to reformulate the account in a generalized form. Let 'A' range over
audiences or hearers. Let the device '.ep' (read 'asterisk-sub-ep') be a dummy, which
represents a specific mood-indicator which corresponds to the propositional attitude
ep-ing (whichever that may be), as for example, 'I-' corresponds to believing (thinking)
and '!' corresponds to intending. I can, using this device, offer the following rough
definition:
0.1. 'By (when) uttering x U meant that .ljJp' = df. '( 3A) (U uttered x M-intending
(i) that A should think U to lj) that p and in some cases only ldepending on the identi-
l'ication of "* lj)P "I, (ii) that A should, via the fulfilment of (i), hirnself lj) that p)'.
It will be convenient to have an abbreviated version of this definiens. Let the device
'ep t, (read 'Ij!-dagger') be a dummy which operates as folIows: in some cases the
phrase 'that A should ep t that p' is to be interpreted as 'that A should think U to ep
that p'; in other cases this phrase is to be interpreted as 'that A should ep that p (via
thinking U to ep that p)'. Which interpretation is to be selected is determined by the
specification of '.epp'. We may now reformulate 0.1 as folIows:
0.1 '. 'By (when) uttering x U meant that .epp' = df. '( 3A) (U uttered x M-intending
that A should ep t that p)'.
To meet all the difficulties to which my 1957 account (which was only intended as
a model) is exposed, a very much more complicated definition is required. But as the
examples which force the introduction of this complexity involve relatively sophisti-
cated kinds of communication or linguistic performance, 1 hope that, for working
purposes, the proffered definition will be adequate.
60 UTTERER'S MEANING, SENTENCE MEANING
Stage III (Step (1): timeless meaning for unstructured utterance-types)
It is, I think, extremely important to distinguish two problems.
(1) What is the relation between timeless meaning (for complete utterance-types)
and occasion-meaning?
(2) In the case of syntactically structured (Iinguistic) utterance-types, how is the
timeless meaning of a complete (sententiaI) utterance-type related to the timeless
meanings of its non-complete structured and unstructured elements (approximately,
phrases and words), and what ac count is to be given of timeless meaning for non-
complete utterance-types?
If we do not treat these problems separately, we shall have only ourselves to blame
for the confusion in which we shall find ourselves. So initially 1 shall restrict myself
to examining the notion of timeless meaning in its application to unstructured
utterance-types. My main example will be a gesture (a signal), and it will be con-
venient first to consider the idea of its timeless meaning for an individual (within a
signalling idiolect, so to speak); and only afterwards to consider the extension of this
idea to groups of individuals. We shall thus preserve for the time being the possibility
of keeping distinct the ideas of having an established meaning and of having a
conventional meaning.
Suppose that a particular sort of hand-wave (to be referred to as H-W) for a
particular individual U (within U's idiolect) means 'I know the route'. We are to look
for an explication of the sentence 'For U, H-W means "I know the route'" which
will relate timeless meaning to occasion-meaning. As a first shot one might suggest
something Ijke 'It is U's policy (practice, habit) to utter H-W in order to mean that U
knows the route' (where 'mean that' is to be analysed in accordance with D.l.); or,
more perspicuously, 'It is U's policy (practice, habit) to utter H-W ifTU is making an
utterance by which U means that U knows the route'.
If we apply D.l. to this suggested definiens, we shall get the following expanded
definiens: 'It is U's policy (practice, habit) to utter H-W ifT U is making an utterance
by means ofwhich (for some A) U M-intends to efTect that A thinks U to think that U
knows the route'. Now, whether or not this definiens is otherwise acceptable, I wish
to argue that the notion of M-intention is otiose here, and that only the notion of
simple intention need be invoked; if U's policy (practice, habit) is such that his use of
H-W is tied to the presence of a simple intention to afTect an audience in the way
described, it will follow that when, on a given occasion, he utters H-W, he will do so,
on that occasion, M-intending to affect his audience in that way.
Suppose that, using only the notion of simple intention, we specify U's policy as
folIows: 'I (that is, utterer U) shall utter H-W ifT I intend (want) some A to think that
I think I know the route'. Now, if U is ever to have the particular intentions which
will be involved in every implementation of this policy, he must (Iogically) be in a
position, when uttering H-W, to suppose that there is at least some chance that these
intentions will be realized; for such a supposition to be justified, as U weil knows, a
AND WORD MEANING
given audience A must be aware of U's policy and must suppose it to apply In Ihl'
utterance of H-W with which U has presented hirn. U, then, when uttering 11 W Oll
a particular occasion, must expect A to think (or at least to be in a position to think)
as folIows: 'U's policy for H-W is such that he utters H-W now with the intention
that 1 should think that he thinks that he knows the route; in that case, 1 take it that
he does think that he knows the route'. But to utter H-W expecting A to respond in
such a way is to utter H-W M-intending that A should think that U thinks that U
knows the route. So a formulation of U's policy of H-W in terms of the notion of
simple intention is adequate to ensure that, by a particular utterance of H-W, U will
mean that he knows the route.
We may, then, suggest a simplified definition: 'For U, H-W me ans "I know the
route'" = df. '11 is U's policy (practice, habit) to utter H-W itT, for some A, U intends
(wants) A to think that U thinks U knows the route'. This definition, however, is
doubly unacceptable. (1) For U, H-W may have a second meaning; it may also mean
'I am about to leave you'. 1fthat is so, U's policy (etc.) cannot be to utter H-W only if
U wants some A to think that U thinks U knows the route; sometimes he will be
ready to utter H-W wanting some A to think that U thinks that U is about to leave A.
(2) U may have other ways of getting an A to think that U thinks that U knows the
route (such as saying 'I know the route'), and may be ready, on occasion, to employ
them. That being so, U's policy (etc.) cannot be to utter H-W if (i.e. whenever) U
wants an A to think that U thinks U knows the route.
To cope with these difficulties, 1 think 1 need some such idea as that of 'having
a certain procedure in one's repertoire'. This idea seems to me to be intuitively fairly
intelligible and to have application outside the realm of linguistic, or otherwise com-
municative, performances, though it could hardly be denied that it requires furt her
explication. A faintly eccentric lecturer might have in his repertoire the following
procedure: if he sees an attractive girl in his audience, to pause for half aminute and
then take a sedative. His having in his repertoire this procedure would not be in-
compatible with his also having two further procedures : (a) if he sees an attractive
girl, to put on a pair of dark spectacles (instead of pausing and taking asedative);
(b) to pause and take asedative when he sees in his audience not an attractive girl,
but a particularly distinguished colleague. Somewhat similarly, if U has in his reper-
toire the procedure of uttering H-W if he wants an audience A to think U thinks U
knows the route, this fact would not be incompatible with his having at least two
further procedures: (a) to say 'I know the route' if he wants some A to think U thinks
U knows the route; and (b) to utter H-W if U wants some A to think U thinks he is
about to leave A. So 1 propose the definition:
D.2. 'For U utterance-type X means (has as one of its meanings) "*!/Jp'" =df.
'U has in his repertoire the following procedure: to utter a token of X if U intends
(wants) A to I/J t that p'.
We may now turn from the idea of timeless meaning within an 'idiolect' to that of
62 UTTERER'S MEANING, SENTENCE MEANING
timeless meaning for a group or class ofindividuals. IfU utters H-W, his measure of
expectation of success as regards efTecting the intended response obviously depends
(as has a1ready been remarked) on A's knowledge of U's procedure; and in general,
unless the signal is to be explained to each A, on A's repertoire containing the same
procedure. So obviously each member of some group G (within which H-W is to be
a tool of communication) will want his procedure with respect to H-W to conform
to the general practice of the group. So I suggest the following rough definition:
0.3. 'For group G, utterance-type X me ans " * ~ ) p ' " = df. 'At least some (? many)
members of group G have in their repertoires the procedure of uttering a token of X
if, for some A, they want A to !fJ t that p; the retention of this procedure being for
them conditional on the assumption that at least some (other) members ofG have, or
have had, this procedure in their repertoires'. D.3. gets in the idea of aiming at con-
formity, and so perhaps (derivatively) also that of correct and incorrect use of X, as
distinct from the idea merely of usual or unusual use of X.
The explication of the notion of 'having a procedure in one's repertoire' is, to my
mind, a task of considerable difficulty. I have feit inclined to propose, as a make-shift
definition, the following:
'U has in his repertoire the procedure of .. .' = df. 'U has a standing readiness
(willingness, preparedness), in some degree, to .. .', a readiness (etc.) to do some-
thing being a member of the same family (a weaker brother, so to speak) as an in-
tention to do that thing. But this definition would clearly be inadequate as it stands;
it may weil be true that, for my exceedingly prim Aunt Matilda, the expression 'he is
a runt' me ans 'he is an undersized person', and yet quite false that she has any degree
ofreadiness to utter the expression in any circumstances whatsoever. What one seems
to need is the idea of her being equipped to use the expression, and the analysis of this
idea is also problematic.
So I shall for the present abandon the attempt to provide adefinition, and content
myself with a few informal remarks. There seem to me to be three main cases in
which one may legitimately speak of an established procedure in respect of utterance-
type X.
(1) That in which Xis current for some group G; that is to say, to utter X in such-
and-such circumstances is part of the practice of many members of G. In that case
my Aunt Matilda (a member of G) may be said to have a procedure for X even
though she herself would rat her be seen dead than utter X; for she knows that some
other members of G do have a readiness to utter X in such-and-such circumstances.
(2) That in which X is current only for U; it is only crs practice to utter X in such-
and-such circumstances. In this case U will have a readiness to utter X in such-and-
such circumstances.
(3) That in which X is not current at all, but the utterance of X in such-and-such
circumstances is part of some system of communication which U has devised, but
which has never been put into operation (like the new Highway Code which I invent
AND WORD MEANING 63
one day while lying in my bath). In that case U has a procedure for X in the attenua-
ted sense that he has envisaged a possible system of practices which wauld involve a
readiness to utter X in such-and-such circumstances.
Stage IV (Step (I): applied timeless meaning for unstructured utterance-types)
We are now in a position to define a notion of applied time1ess meaning which will
apply to H-W.
0.4. 'When U uttered X (type), X meant ".p'" =df. '( 3A) (3q) (U intended A to
recognize (? and to recognize that U intended A to recognize) what U meant
[occasion-meaning] by his uttering X, on the basis of A's knowledge (assumption)
that, for U, X means (has as one of its meanings} ".p" [as defined by 0.2.])'.
Or more fully:
Let '.' and '." both be dummy mood-indicators.
0.4'. 'When U uttered X, X meant ".!/Jp'" =df. '( 3A) (U meant by uttering X
that .'q; and U intended A to recognize (? and to recognize that he was intended to
recognize) that by uttering X U meant that .'q via A's knowledge (assumption) that
in U's repertoire is the procedure of uttering X if, for some A', U wants A' tO!/J+
that p)'. ['p' may, or may not, represent that propositional content to which indefinite
reference is made in the existential quantification of 'q'.]
0.4., and of course 0.4'., allow both for the case in which U meant by H-W that
he knew the route (coincidence ofmeaning ' ... ' and meaning that . .. ), and also for the
case in which, for example, U (a criminal) has lured a victim into his car and signals
(non-literally, so to speak) to his accomplice that he knows how to handle the victim.
In both cases it is expected by U that the audience's understanding of the utterance
of H-W will be based on its knowledge that U has a certain procedure (to utter H-W
if U wants an audience to think that U thinks U knows the route).
Stages 1/1 and IV (Step (2): timeless and applied timeless meaning for structured
utterance-types, complete and non-complete)
To deal with structure utterance-types and their elements, I think I need the follow-
ing apparatus.
(1) Let 'I, (I
2
)' (read 'I,-with-I
2
') denote a sentence of which I
2
is a sub-
sentence. Allow that a sentence is a sub-sentence of itself, and so that I
2
may = I,.
(2) Let vIII (I
2
)] (read 'v-of-I,-with-I
2
') be a particular utterance (token) of
I, (I
2
) uttered by U. v [I, (I
2
)] is to be a camplete utterance; that is, it is not to be
part of v[I
3
(I, (I
2
») (not e.g. to be the utterance of a disjunct).
(3) It is characteristic of sentences (a characteristic shared with phrases) that their
standard meaning is consequential upon the meaning of the elements (words, lexical
items) which enter into them. So I need the notion of a 'resultant procedure': as a
first approximation, one might say that a procedure for an utterance-type X will be
aresultant procedure if it is determined by (its existence is inferable from) a know-
64 UTTERER'S MEANING, SENTENCE MEANING
ledge of procedures (a) for particular utterance-types which are elements in X, and (b)
for any sequence of utterance-types which exemplifies a particular ordering of
syntactical categories (a particular syntactical form).
Now let us deal with the notion of timeless meaning in U's idiolect.
D.5. 'For U, L means "*rpp'" =df. 'U has a resultant procedure for L, viz. to utter
L if, far same A, U wants A to rp + that p.' [D.5. par allels D.2.)
An explication of timeless meaning in any language can, perhaps, be provided by
adapting D.3.; I shall not attempt this task now.
For applied timeless meaning laffer
D.6. 'L
2
in V[LI (L
2
») meant "*rpp'" =df. '(ijA) (ijq) (U meant by VLL
I
(L
2
)] that
*'q, and U intended A to recognize that U meant by V[LI (L
2
)] that *'q at least partly
on the basis of A's thought that U has a resultant procedure for L
2
, viz. (far suitable
A') to utter L
2
if U wants ~ ' to rp+ that p).' [D.6. par allels D.4'.)
So far (maybe) so good. But the nation of'resultant procedure' has been left pretty
unilluminated ; and if we are to shed any light on the nation of word-meaning, and its
connection with 'meaning that', we ought to look at the nature of the more funda-
mental procedures from which a resultant procedure descends. It would be nice to
give a general schema, to show the role of word-meanings (covering every type of
ward) in deterrnining (in combination) sentence meanings (covering sentences of any
syntactical structure). But this looks like a Herculean task (in our present state of
knowledge). The best we can hope for is a sketch, for a very restricted (but central)
range of ward-types and syntactical forms, of a fragment of what might be the kind
of theory we need. Let us take as our range all ar part of the range of affirmative
categorical (not necessarily indicative) sentences involving a noun (or definite descrip-
tion) and an adjective (or adjectival phrase).
The apparatus needed (for one such attempt) would be:
(1) Suppose a to be an indicative sentence. Then we need to be able to apply the
ideas of an indicative version of a (a itself), an imperative version of a, an optative
version of a etc. (mood variations). It would be the business of some linguistic theory
to equip us to apply such characterizations (so as philosophers of language we can
ass urne this as given).
(2) We need to be able to apply some such nation as a predication of ß (adjectival)
on (t (nominal). 'Smith is tactful', 'Smith, be tactful', 'Let Smith be tactful', 'Oh that
Smith may be tactful' would be required to count, all of them, as predications of
'tactful' on 'Smith'. It would again be the business of some linguistic theory to set up
such a sentential characterization.
(3) Suppose we, for a moment, take for granted two species of correlation, R-
correlation (referential) and D-correlation (denotational). We want to be able to speak
of some particular object as an R-correlate of a (nominal), and of each member of
some dass as being a D-correlate of ß (adjectival).
Now suppose that U has the following procedures:
AND WORD MEANING
65
P.l. To utter the indieative version of a if (for some A) U wants/intends A to think
that U thinks ... (the blank being filled by the infinitive version of a, e.g. 'Smith to be
taetful'.) (Also, for example P.I': obtained from P.I by substituting 'imperative'/
'indieative' and 'intend'/'think that U thinks'.) [Sueh proeedures set up eorre1ations
between moods and speeifieations of 'rp t'.l
P.2. To utter rpolo-eorre1ated [cf. P.l. and P.I'., ete.l predication of ß on a if(for
some A) U wants A to rp t a particular R-correlate of a to be one of a particular set
of D-corre1ates of ß.
Further suppose that, for U, the following corre1ations hold:
CI. Jones' dog is an R-corre1ate of 'Fido'.
C2. Any hairy-coated thing is a D-eorre1ate of 'shaggy'.
Given that U has the initial proeedures P.I. and P.2. we can infer that U has the
resultant procedure (determined by P.l. and P.2.): RPL To utter the indicative
version of a predieation of ß on a if U wants A to think U to think a particular R-
correlate of a to be one of a particular set of D-correlates of ß.
Given RP land Cl we can infer that U has
RP2. To utter the indicative version of a predication of ß on 'Fido' if U wants A
to think U to think Jones' dog to be one of a particular set of D-corre1ates of ß.
Given RP2 and C2, we can infer that U has
RP3. To utter the indieative version of a predication of 'shaggy' on 'Fido' if U
wants A to think U to think Jones' dog is one ofthe set ofhairy-eoated things (i.e. is
hairy-eoated).
And given the information from the linguist that 'Fido is shaggy' is the indicative
version of a predication of 'shaggy' on 'Fido' (assumed), we can infer U to have
RP4. To utter 'Fido is shaggy' if U wants A to think U to think that Jones' dog is
hairy-coated. And RP4. is an interpretant of 'For U, "Fido is shaggy" means
"Jones' dog is hairy-coated".'
I have not yet provided an explication for statements of time1ess meaning re1ating
to non-eomplete utteranee-types. I am not in a position to provide a definiens far 'X
[non-completel means " ... "'; indeed I am not certain that a general form of definition
can be provided for this schema; it may remain impossible to provide a definiens
until the syntactical category of X has been given. I can, however, provide a
definiens which may be adequate for adjectil'al X (e.g. 'shaggy').
D.7. 'For U, (adjectival) means " ... '" =df. 'U has this procedure: to utter a
rp o'-eorrelated predieation of X on a if (for some A) U wants A to rp t a partieular
R-corre1ate of a to be .. .' [where the two lacunae represented by dots are identically
completedl.
Any specifie procedure of the form mentioned in the definiens of D.7. can be
shown to be a resultant procedure; for example, if U had P.2. and also C2., it will
be inferable that he has the procedure of uttering a rp t-correlated predieation of
'shaggy' on a if (for some A) U wants A to rp t a partieular R-correlate of a to be
66 UTTERER'S MEANING, SENTENCE MEANING
one of the set of hairy-coated things, i.e. that for U 'shaggy' means 'hairy-coated'.
I can now ofTer adefinition of the notion of a complete utterance-type which has so
far been taken for granted.
0.8. 'X is complete' =df. 'A fully expanded definiens for "X means' ... '" contains
no explicit reference to correlation, other than that involved in speaking of an R-
correlate of some referring expression occurring within X'. [The expanded definiens
for the complete utterance-type 'He is shaggy' may be expected to contain the phrase
'a particular R-correlate of "he".)
Correlation. We must now stop taking for granted the notion of correlation. What
is it to mean to say that e.g. Jones' dog is the/a R-correlate of 'Fido'? One idea (build-
ing in as little as possible) would be to think of 'Fido' and Jones' dog as paired, in
some system of pairing in which names and objects form ordered pairs. But in one
sense of 'pair' any one name and any one object form a pair (an ordered pair, the
first member of which is the name, the second the object). We want a sense of
'paired' in which 'Fido' is paired with Jones' dog but not with Smith's cat. 'Se-
lected pair'? But what does 'selected' mean? Not 'selected' in the sense in which
an apple and an orange may be selected from a dish: perhaps in the sense in which
a dog may be selected (as something with which (to which) the selector intends
to do something). But, in the case of the word-thing pair, do what? And what is the
process of selecting?
I suggest we consider initially the special case in which linguistic and non-linguistic
items are explicitly correlated. Let us take this to consist in performing some act as
a result of which a linguistic item or a non-linguistic item (or items) come to stand in
a relation in which they did not previously stand, and in which neither stands to non-
correlates in the other reaIm. Since the act of correIation may be a verbal act, how
can this set up a relation between items?
Suppose U produces a particular utterance (token) V, which belongs to the
utterance-type 'shaggy: hairy-coated things'. To be able to say that U had by V
correlated 'shaggy' with each member of the set of hairy-coated things, we should
need to be able to say that there is some relation R such that:
(a) By uttering V, U efTected that 'shaggy' stood in R to each hairy-coated thing,
and only to hairy-coated things.
(b) U uttered V in order that, by uttering V he should effect this.
It is dear that condition (b), on which some will look askance because it intro-
duces a reference to U's intention in performing his act of correlation, is required, and
that condition (a) aIone would be inadequate. Certainly by uttering V, regardless of
his intentions, U has set up a situation in which a relation R holds exdusively be-
tween 'shaggy' and each hairy-coated thing Z, namely the relation which consists in
being an expression uttered by U on a particular occasion 0 in conversational juxta-
position with the name of a cIass to which Z belongs. But, by the same act, U has
AND WORD MEANING 67
also set up a situation in which another relation R' holds exdusively between 'shaggy'
and each non-hairy-coated thing Z', namely the relation which consists in being an
expression uttered by U on occasion 0 in conversational justaposition with the name
of the complement of a dass to which Z' belongs. We do not, however, for our
purposes, wish to think ofU as having correlated 'shaggy' with each non-hairy-coated
thing. The only way to ensure that R' is eliminated is to add condition (b), which
confines attention to a relationship which U intends to set up. It looks as if inten-
sionality is embedded in the very foundations of the theory of language.
Let us, then, express more formally the proposed account of correlation. Suppose
that V =utterance-token of type' "Shaggy" : hairy-coated things' (written). Then,
by uttering V, U has correlated 'Shaggy' with (and only with) each hairy-coated
thing= (3R) {(U efTected by V that (Vx) (R 'Shaggy' x= xy (y is a hairy-coated
thing») & (U uttered V in order that U efTect by V that (Vx) ... )1.
If so understood, U will have correlated 'shaggy' with hairy-coated things only if
there is an identifiable R' for which the condition specified in the definiens holds.
What is such an R'? I suggest R'xy == x is a (word) type such that V is asequencecon-
sisting of a token of x followed by a colon followed by an expression l'hairy-coated
things'l the R-correlate of which is a set ofwhich y is a member. R'xy holds between
'shaggy' and each hairy-coated thing given U's utterance of V. Any utterance V'
of the form exemplified by V could be uttered to set up R"xy (involving V' instead of
V) between any expression and each member of any set of non-linguistic items.
There are other ways of achieving the same effect. The purpose of making the
utterance can be specified in the utterance: V = utterance of 'To effect that, for some
R, "shaggy" has R only to each hairy-coated thing, "shaggy" : hairy-coated things'.
(The expression of the specified R will now have 'V is a sequence containing' vice
'V is a sequence consisling 0/ .. .'.) Or U can use the performative form: 'I correlate
"shaggy" with each hairy-coated thing'. Utterance of this form will at the same time
set up the required relation and label itself as being uttered with the purpose of setting
up such a relation.
But by whichever form an act of explicit correlation is effected, to say of it that it
is (or is intended to be) an act of correlation will always be to make an indefinite
reference to a relation(ship) which the act is intended to set up, and the specification
of the relation involved will in turn always involve a further use of the notion of
correlation (e.g. as above in speaking of a set which is the corre1ate (R-correlate) of
a particular expression (e.g. 'Hairy-coated things'». This seems to involve a regress
which might weil be objectionable; though 'corre1ation' is not used in definition of
corre1ation, it will be used in specification of an indefinite reference occurring in the
definition of corre1ation. It might be considered desirable (even necessary) to find a
way of stopping this regress at some stage. (Is this a characteristically empiricist
demand ?) If we don't stop it, can correlation even get started (if prior correlation is
presupposed ?). Let us try 'ostensive correlation'.
68 UTTERER'S MEANING, SENTENCE MEANING
(Acts 1, 2, 3, etc.) U ostends simultaneously with each ostension
a
J
,
uttering 'shaggy' (intending to ostend only objects which are hairy-coated). For the
combination of these acts to constitute a case of correlating 'shaggy' with each hairy-
coated thing, it must be the case that:
(!:IR) (U effected, and intended to effect, by acts 1, 2, 3, etc. that (Vy)
('Shaggy' has R to y if and only if y is hairy-coated».
How is the appropriate relation to be specified'! As folIows:
R'xy (for some F) Iviz. being hairy-coatedJ (U ostended and intended to
ostend only objects which are Fand, in acts 1,2,3, etc. accompanied each
ostension by uttering a token of x; and y is F).
Given the ostensions, R'xy holds between 'shaggy' and each hairy-coated thing,
and the specification of R'xy at least seems not to involve further reference to
correlation.
So far, we have been acting on the assumption that the correlations, which in
association with initial procedures yield further procedures, are explicit correlations;
that is to say, that they are correlations set up by some identifiable and dateable act
of correlating. But this assumption is clearly artificial. Many correlatio!1s, referential
as well as denotative, seem to grow rather than to be created. The situation seems to
be as follows:
(1) We need to be able to invoke such aresultant procedure as the following, which
we will call RPI2, namely, to predicate ß on 'Fido', when U wants A to I/J t that
Jones' dog is a D-correlate of ß; and we want to be able to say that at least some-
times such aresultant procedure may result from among other things a non-explicit
R-correlation of 'Fido' and Jones' dog.
(2) It is tempting to suggest that a non-explicit R-correlation of 'Fido' and Jones'
dog consists in the fact that U would, explicitly, correlate 'Fido' and Jones' dog.
(3) But to say that U would explicitly correlate 'Fido' and Jones' dog must be
understood as an elliptical way of saying something of the form 'U would explicitly
correlate "Fido" and Jones' dog, if p'. How is 'p' to be specified?
(4) Perhaps 'If U were asked to give an explicit correlation for "Fido" '. But ifU
were actually faced with arequest, he rnight quite weil take it that he is being asked to
make a stipulation in making which he would have an entirely free hand. If he is not
being asked for a stipulation, then it must be imparted to hirn that his explicit correla-
tion is to satisfy some non-arbitrary condition. But what condition can this be? Again
it is tempting to suggest that he is to make his explicit correlation such as to match
or fit existing procedures.
(5) In application to RPI2, this would seem to amount to imposing on U the
demand that he should make his explicit correlation such as to yield RP 12.
(6) In that case, RP12 results from a non-explicit correlation which consists in the
AND WORD MEANING 69
fact that U would explicitly correlate 'Fido' and Jones' dog if he wanted to make an
explicit correlation which would generate relevant existing procedures, viz. RPl2
itself. There is an apparent circularity here. Is this tolerable?
(7) It may be tolerable in as much as it may be a special case of a general pheno-
menon which arises in connection with the explanation oflinguistic practice. We can,
if we are lucky, identify 'linguistic rules', so called, which are such that our linguistic
practice is as if we accepted these rules and consciously followed them. But we want
to say that this is not just an interesting fact about our linguistic practice, but an
explanation of it; and this leads us on to suppose that 'in some sense', 'im pli city' , we
do accept these rules. Now the proper interpretation of the idea that we do accept
these rules becomes something of a mystery, if the 'acceptance' of the rules is to be
distinguished from the existence of the related practices; but it seems like a mystery
which, for the time being at least, we have to swallow, while recognizing that it
involves us in an as yet unsolved problem.
CONCLUDING NOTE
It will hardly have escaped notice that my account of the cluster of notions con-
nected with the term 'meaning' has been studded with expressions for such intensional
concepts as those of intending and of believing; and my partial excursions into
symbolic notation have been made partly with the idea of revealing my commitment
to the legitimacy of quantifying over such items as propositions. I shall make two
highly general remarks about this aspect of my procedure.
(l) I am not sympathetic towards any methodological policy which would restrict
one from the start in an attempt to formulate a theory of meaning in extensiona1
terms. It seems to me that one should at least start by giving oneself a free hand to
make use of any intensiona1 notions or devices which seem to be required in order to
solve one's conceptual problems, at least at a certain level, in ways which (meta-
physical bias apart) reason and intuition commend. If one denies oneself this
freedom, one runs a very serious risk of underestimating the richness and complexity
of the conceptua1 field which one is investigating.
(2) I said at one point that intensiona1ity seems to be embedded in the very founda-
tions of the theory of language. Even if this appearance corresponds with rea1ity, one
is not, I suspect, precluded from being, in at least one important sense, an exten-
tionalist. The psychological concepts which, in my view, are needed for the formula-
tion of an adequate theory of language may not be among the most primitive or
fundamenta1 psychological concepts (like those which apply not only to human
beings but to quite lowly anima1s as weil); and it may be possible to derive (in some
relevant sense of 'derive') the intensiona1 concepts which 1 have been using from
more primitive extensiona1 concepts. Any extensionalist has to deal with the problem
70 UTTERER'S MEANING, SENTENCE MEANING
of allowing for a transition from an extensionaI to a non-extension al language; and
it is by no means obvious to me that intensionality can be explained only via the idea
of conceaIed references to language, and so presupposes the concepts in terms of
which the use of language will have to be understood.
v
TOPICS IN THE THEORY OF
GENERATIVE GRAMMAR
NOAM CHOMSKY
(a) ASSUMPTIONS AND GOALS
My original intention was to use these lectures to present some recent work on
general linguistic theory and on the structure of English, within the general frame-
work of transformational generative grammar. However, a sequence of recent publi-
cations has indicated that many points that I had hoped to take for granted are widely
regarded as controversial, and has also indicated misunderstanding, on a rather
substantial scale, of the general framework I had expected to presuppose--in particu-
lar, a misunderstanding as to which elements of this framework express substantive
assumptions about the nature of language and are, therefore, matters of legitimate
controversy and rational discussion, and which, on the other hand, relate only to
questions of goals and interests and are therefore no more subject to debate than
the question : is chemistry right or wrong? In the light of this, it seems advisable to
change my original plan and to spend much more time on background assurnptions
and general questions of various sorts than I had at first intended. I still hope to be
able to incorporate an exposition (much abbreviateJ) of some recent work, but I will
lead up to it more slowly, in the following steps:
(I) discussion of general background assurnptions and goals that underlie and
motivate much of the work in generative gramm ar of the past decade;
(2) discussion of various objections to this general point of view that seem to me to
be based on eITor, misunderstanding, or equivocation of one sort or another;
(3) presentation of a theory of generative grammar of a sort exemplified, for
example, in N. Chomsky, Syntactic Structures (The Hague, 1957), R. B. Lees, The
Grammar 0/ English Nominalizations (Bloomington, 1960), M. Halle, 'Phonology in
a Generative Grammar', Word, 18, pp. 54-72 (1962), and J. Katz and J. Fodor,
'The Structure of a Semantic Theory', Lg. 39, pp. 170-210 (1963);
(4) discussion of various real inadequacies that have been exposed in this position
in work of the past half-dozen years; and
(5) sketch of a refined and improved version of this theory, designed to overcome
these difficulties.
From Topics in the Theory 0/ Generative Grammar (in Janua Linxuarum), (Mouton & Co., The Hague,
1966), pp. 7-24 and 51-75. Reprinted by permission of the author, the editor, and the publishers.
72 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
I will try to cover these points in the first three sections, concentrating large1y on
syntax. Section I will deal with the first point, section 11 with the second, and section
III with the third, fourth and fifth.*
In the final section I will discuss an approach to the study of sound structure that
has been gradually evolving since Chomsky, Halle, and F. Lukoff, 'On Accent and
Juncture in English', For Roman Jakobson, eds. M. Halle, H. Lunt, and H. MacLean,
pp. 65-80 (The Hague, 1956), and has been presented invariousstagesofdevelopment
in publications of Halle's and mine (listed in the bibliography below) since then, and will,
hopefully, soon emerge to fulliight of day in a book that is now in active preparation.
In the course of this presentation, I will also discuss a few criticisms ofthis approach.
The discussion of criticisms will be very brief, however, since Halle and I havediscussed
most of them, in so far as they are known to us, in considerable detail e1sewhere. I
In general, this essay contains no new or original material. it is intended only as an
informal guide to other books and papers,
2
in which questions touched on here are
dealt with more thoroughly, and as an attempt to clarify issues that have been raised
in critical discussion.
In the course of this paper I will also make a few remarks about historical back-
grounds for the position that will be outlined.
3
Quite a few commentators have
assumed that recent work in generative grammar is somehow an outgrowth of an
interest in the use of computers for one or another purpose, or that it has some other
engineering motivation, or that it perhaps constitutes some obscure branch of mathe-
matics. This view is incomprehensible to me, and it is, in any event, entire1y false.
Much more perceptive are those critics who have described this work as in large
measure areturn to the concems and often even the specific doctrines of traditional
linguistic theory. This is true--apparently to an extent that many critics do not
realize.
4
I differ from them only in regarding this observation not as a criticism, but
* This volume includes section land section 1II of the original marked (a) and (b) rcspectively led.l.
I In particular, see Chomsky, Current Issues in Linguistic Theory, 3 I, pp. 105-7 (The Hague, 1964),
which deals with criticisms in C. A. Ferguson's review of HaUe, The Sound Pattern Qf Russian (The Hague,
1959); and in Chomsky and Halle, 'Some Controversial Questions in Phonological Theory', Journal 0/
Linguistics, I, pp. 97-138 (1965), which deals with objections raised by F. W. Householder Jr., 'On Some
Rccent Claims in Phonological Theory', Journal 0/ Linguislics, I, pp. 13-34 (1965).
2 e.g. Katz and P. Postal, An Integrated Theory 0/ Linguislic Description (Cambridge, Mass., 1964);
Chomsky, Current Issues in Linguistic Theory, and Aspects 0/ the Theory 0/ Syntax (Cambridge, Mass.,
1965).
3 This matter is discussed in more detail in Chomsky, Currenllssues in Linguislic Theory, § I, in AspeclS
o/lhe Theory 0/ Syntax, Ch. I, § 8, and in Cartesian Linguislics (New York, 1966).
4 To cite just one example, consider A. Reichling, 'Principles and Methods of Syntax: Cryptanalytical
Formalism', Lingua 10, pp. 1-7 (1961), who asserts that obviously I could not 'be said to sympathize with
such a "mentalistic monster" as the "innere Sprachform '''. But in fact the work that he is discussing is quite
explicitly and selfconsciously mentalistic (in the traditional, not the Bloomfieldian, sense of this word-that is,
it is an attempt to construct a theory of mental processes), and it can, furthermore, be quite accurately
described as an attempt to develop further the Humboldtian notion of'form of language' and its implications
for cognitive psychology, as will surely be evident to anyone familiar both with Humboldt and with recent
work in generative grammar (for explicit discussion, see the references cited above).
TOPICS IN THE THEORY OF GENERATIVE GRAMM AR 7J
rather as adefinite merit of this work. That is, it seems to me that it is the modern
study of language prior to the explicit study of generative grammar that is seriously
defective in its failure to deal with traditional questions and, furthermore, to recognize
the essential correctness of many of the traditional answers and the extent to which
they provide a fruitful basis for current research.
A distinction must be made between what the speaker of a language knows im-
plicitly (what we may call his competence) and what he does (his performance). A
grammar, in the traditional view, is an account of competence. It describes and
attempts to account for the ability of a speaker to understand an arbitrary sentence
of his language and to produce an appropriate sentence on a given occasion. If it is
a pedagogic grammar, it attempts to provide the student with this ability; if a
linguistic grammar, it aims to discover and exhibit the mechanisms that make this
achievement possible. The competence of the speaker-hearer can, ideally, be ex-
pressed as a system of rules that relate signals to semantic interpretations of these
signals. The problem for the grammarian is to disco ver this system of rules; the
problem for linguistic theory is to discover general properties of any system of rules
that may serve as the basis for a human language, that is, to elaborate in detail what
we may call, in traditional terms, the general form of language that underlies each
particular realization, each particular natural language.
Performance provides evidence for the investigation of competence. At the same
time, a primary interest in competence entails no disregard for the facts of perform-
ance and the problem of explaining these facts. On the contrary, it is difficult to see
how performance can be seriously studied except on the basis of an explicit theory
of the competence that underlies it, and, in fact, contributions to the understanding of
performance have largely been by-products of the study of grammars that represent
competence.
5
Notice, incidentally, that a person is not generally aware of the rules that govern
sentence-interpretation in the language that he knows; nor, in fact, is there any reason
to suppose that the rules can be brought to consciousness. Furthermore, there is no
reason to expect hirn to be fully aware even of the empirical consequences of these
internalized rules---that is, of the way in which signals are assigned semantic inter-
pretations by the rules of the language that he knows (and, by definition, knows
perfectly). On the difficulties of becoming aware of one's own linguistic intuitions,
see the discussion in Chomsky, Aspecls of the Theory of Syntax, Ch. 1, § 4. It is
I will not consider Reichling's criticisms of generative grammar here. The cited remark is just one
illustration of his complete lack of comprehension of the goals, concerns, and specific content of the work
that he was discussing, and his discussion is based on such gross misrepresentation of this work that
comment is hardly called for.
'For discussion, see G. A. Miller and Chomsky, 'Finitary Models of Language Users', Handbook 0/
MalhemalicaIPsychology.vol.lI. eds. R. D. Luce, R. Bush, and E. Galanter (New York, 1963); Chomsky,
A specls of I he Theory of Synlax, Ch. I, ~ 2.
74 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
important to realize that there is no paradox in this; in fact, it is precisely what should
be expected.
Current work in generative grammar has adopted this traditional framework of
interests and concerns. It attempts to go beyond traditional grammar in a funda-
mental way, however. As has repeatedly been emphasized, traditional grammars
make an essential appeal to the intelligence of the reader. They do not actually
formulate the rules of the grammar, but rather give examples and hints that enable
the intelligent reader to determine the grammar, in some way that is not at all under-
stood. They do not provide an analysis of the 'faculte de langage' that makes this
achievement possible. To carry the study of language beyond its traditional bounds,
it is necessary to recognize this limitation and to develop means to transcend it. This
is the fundamental problem to which all work in generative grammar has been
addressed.
The most striking aspect of linguistic competence is what we may call the
'creativity of language', that is, the speaker's ability to produce new sentences,
sentences that are immediately understood by other speakers although they bear no
physical resemblance to sentences which are 'familiar'. The fundamental importance
of this creative aspect of normal language use has been recognized since the
seventeenth century at least, and it was at the core of Humboldtian generallinguistics.
Modem linguistics, however, is seriously at fault in its failure to come to grips with
this central problem. In fact, even to speak of the hearer's 'familiarity with sentences'
is an abusrdity. Normal use of language involves the production and interpretation of
sentences that are similar to sentences that have been heard before only in that they
are generated by the rules ofthe same gramrnar, and thus the only sentences that can
in any serious sense be called 'familiar' are diches or fixed formulas of one sort or
another. The extent to which this is true has been seriously underestimated even by
those linguists (e.g. O. Jespersen) who have given some attention to the problem of
creativity. This is evident from the common description of language use as a matter
of 'grammatical habit' (e.g. O. Jespersen, Philosophy 0/ Grammar (London, 1924».
It is important to recognize that there is no sense of 'habit' known to psychology in
which this characterization of language use is true Gust as there is no notion of
'generalization' known to psychology or philosophy that entitles us to characterize
the new sentences of ordinary linguistic usage as generalizations of previous perform-
ance). The familiarity of the reference to normallanguage use as a matter of'habit'
or as based on 'generalization' in some fundamental way must not blind one to the
realization that these characterizations are simply untrue if terms are used in any
technical or well-defined sense, and that they can be accepted only as metaphors-
highly misleading metaphors, since they tend to lull the linguist into the entirely
erroneous belief that the problem of accounting for the creative aspect of normal
language use is not after all a very serious one.
Returning now to the central topic, a generative grammar (that is, an explicit
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 75
grammar that makes no appeal to the reader's 'faculte de langage' but rather attempts
to incorporate the mechanisms of this faculty) is a system of rules that relate signals
to semantic interpretations of these signals. It is descriptil'ely adequate to the extent
that this pairing corresponds to the competence of the idealized speaker-hearer. The
idealization is (in particular) that in the study of grammar we abstract away from the
many other factors (e.g. memory !imitations, distractions, changes ofintention in the
course of speaking, etc.) that interact with underlying competence to produce actual
performance.
If a generative grammar is to pair signals with semantic interpretations, then the
theory of generative grammar must provide a general, language-independent means
for representing the signals and semantic interpretations that are interrelated by the
grammars of particular languages. This fact has been recognized since the origins of
linguistic theory, and traditionallinguistics made various attempts to develop theories
of universal phonetics and universal semantics that might meet this requirement.
Without going into any detail, I think it would be widely agreed that the general
problem of universal phonetics is fairly weil understood (and has been, in fact, for
several centuries), whereas the problems of universal semantics still remain veiled in
their traditional obscurity. We have fairly reasonable techniques of phonetic repre-
sentation that seem to approach adequacy for all known languages, though, of course,
there is much to leam in this domain. In contrast, the immediate prospects for
universal semantics seem much more dirn, though surely this is no reason for the
study to be neglected (quite the opposte conclusion should, obviously, be drawn).
In fact, recent work of Katz, Fodor, and PostaI, to which I return in the third section,
seems to me to suggest new and interesting ways to reopen these tradition al questions.
The fact that universal semantics is in a highly unsatisfactory state does not imply
that we must abandon the programme of constructing grammars that pair signals
and semantic interpretations. For although there is little that one can say about the
language-independent system of semantic representation, a great deal is known about
conditions that semantic representations must meet, in particular cases. Let us then
introduce the neutral technical notion of 'syntactic description', and take a syntactic
description of a sentence to be an (abstract) object of some sort, associated with the
senten ce, that uniquely determines its semantic interpretation (the latter notion being
left unspecified pending further insights into semantic theory)6 as weil as its phonetic
form. A particular linguistic theory must specify the set of possible syntactic descrip-
tions for sentences of a natural language. The extent to which these syntactic des-
criptions meet the conditions that we know must apply to semantic interpretations
provides one measure of the success and sophistication of the grammatical theory in
question. As the theory of generative grammar has progressed, the notion of syntactic
"Working in this framework then, we would regard a semantically ambiguous minimal element as con-
stituting two distinct lexical entries ; hence two syntactic descriptions might difTer only in that they contain
different members of a pair of homonymous morphemes.
76 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
description has been darified and extended. I will discuss below some recent ideas on
just what should constitute the syntactic description of a senten ce, if the theory of
generative gramm ar is to provide descriptively adequate grammars.
Notice that a syntactic description (henceforth, SO) may convey information
about a sentence beyond its phonetic form and semantic interpretation. Thus we
should expect a descriptively adequate grammar of English to express the fact that
the expressions (1)--(3) are ranked in the order given in terms of 'degree of deviation'
from English, quite apart from the question of how interpretations can be imposed on
them [in the case of (2) and (3)] :
(1)
(2)
(3)
the dog looks terrifying
the dog looks bar king
the dog looks lamb
A generative grammar, then, must at least determine a pairing of signals with SO's;
and a theory of generative grammar must provide a general characterization of the
dass of possible signals (a theory of phonetic representation) and the dass of possible
SO's. A grammar is descriptively adequate to the extent that it is factually correct
in a variety of respects, in particular, to the extent that it pairs signals with SO's that
do in fact meet empirically given conditions on the semantic interpretations that they
support. For example, if a signal has two intrinsic semantic interpretations in a
particular language [e.g. (4) or (5), in English), a grammar of this language will
approach descriptive adequacy if it assigns two SO's to the sentence, and, beyond
this, it will approach descriptive adequacy to the extent that these SO's succeed in
expressing the basis for the ambiguity.
(4)
(5)
they don't know how good meat tastes
what disturbed J ohn was being disregarded by everyone
In the case of (4), for example, a descriptively adequate grammar must not only
assign two SO's to the sentence but must also do so in such a way that in one of these
the grammatical relations of good, meat, and taste are as in 'meat tastes good', while
in the other they are as in 'meat which is good tastes Adjective' (where the notion
'grammatical relation' is to be defined in a general way within the linguistic theory
in question), this being the basis for the alternative semantic interpretations that may
be assigned to this sentence. Similarly, in the case of (5), it must assign to the pair
disregard-John the same grammatical relation as in 'everyone disregards John', in
one SO; whereas in the other it must assign this very same relation to the pair
disregard-what (disturbed John), and must assign no semantically functional gram-
matical relation at aIl to disregard-John. On the other hand, in the case of(6) and (7)
only one SO should be assigned by a descriptively adequately grammar. This SO
should, in the case of (6), indicate that John is related to incompetent as it is in 'John
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 77
is incompetent', and that John is related to regard (as incompetent) as it is in 'every-
one regards lohn as incompetent'. In the case of (7), the SD must indicate that our
is re1ated to regard (as incompetent) as us is related to regard (as incompetent) in
'everyone regards us an incompetent'.
(6)
(7)
what disturbed John was being regarded
as incompetent by everyone.
what disturbed John was our being regarded
as incompetent by everyone.
Similarly, in the case of (8), the grammar must assign four distinct SD's, each of
which specifies the system of grammatical relations that underlies one of the distinct
semantic interpretations of this senten ce:
(8) the police were ordered to stop drinking after midnight.
Examples such as these should suffice to illustrate what is involved in the problem of
constructing descriptively adequate generative grammars and developing a theory
of gramrnar that analyses and studies in fuH generality the concepts that appear in
these particular grammars. It is quite evident from innumerable examples of this sort
that the conditions on semantic interpretations are sufficiently clear and rich so that
the problem of defining the notion 'syntactic description' and developing descriptive1y
adequate gramrnars (relative to this notion of SD) can be made quite co,?crete, despite
the fact that the notion 'semantic interpretation' itself still resists any deep analysis.
We return to some recent ideas on semantic interpretation of SD's in section 111.
A grammar, once again, must pair signals and SD's. The SD assigned to a signal
must deterrnine the semantic interpretation ofthe signal, in some way which, in detail,
remains unclear. Furthermore, each SD must uniquely determine the signal of which
it is the SD (unique1y, that is, up to free variation). Hence the SD must (i) determine
a semantic interpretation and (ii) determine a phonetic representation. Let us define
the 'deep structure of a sentence' as that aspect of the SD that determines its
semantic interpretation, and the 'surface structure of a sentence' as that aspect of the
SD that determines its phonetic form. A gramrnar, then, must consist ofthree com-
ponents: a syntactic component, which generates SD's each of which consists of a
surface structure and a deep structure; a semantic component, which assigns a
semantic interpretation to a deep structure; a phonological component, which assigns
a phonetic interpretation to a surface structure. Thus the gramm ar as a whole will
associate phonetic representations and semantic interpretations, as required, this
association being mediated by the syntactic component that generates deep and
surface structures as elements of SD's.
The notions 'deep structure' and 'surface structure' are intended as explications of
the Humboldtian notions 'inner form of a sentence' and 'outer form of a sentence'
78 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
(the general notion 'form' is probably more properly to be related to the notion
'generative grammar' itself-----cf. Chomsky, Current Issues in Linguistic Theory, for
discussion). The terminology is suggested by the usage familiar in contemporary
analytic philosophy (cf., for example, Wittgenstein, Philosophical Investigations,
p. 168 (Oxford, 1953». C. F. Hockett has also used these terms (A Course in Modem
Linguistics, Ch. 29 (New York, 1958» in roughly the same sense.
There is good reason (see below, section IV) to suppose that the surface structure
of a sentence is a labelIed bracketing that segments it into its continuous constituents,
categorizes these, segments the constituents into further categorized constituents,
etc. Thus underlying (6), for example, is a surface structure that analyses it into its
constituents (perhaps, 'what disturbed John', 'was', 'being regarded as incompetent
by everyone'), assigning each of these to a certain category indicated by the label-
ling, then further segmenting each of these into its constituents (e.g. perhaps, 'what
disturbed John' into 'what' and 'disturbed John'), each of these being assigned to a
category indicated by the labelling, etc., until ultimate constituents are reached. In-
formation of this sort is, in fact, necessary to determine the phonetic representation
of this sentence. The labelIed bracketing can be presented in a tree-diagram, or in
other familiar notations.
It is c1ear, however, that the deep structure must be quite different from this surface
structure. For one thing, the surface representation in no way expresses the gram-
matical relations that are, as we have just observed. crucial for semantic interpreta-
tion. Secondly, in the case of an ambiguous sentence such as, for example, (5), only
a single surface structure may be assigned, but the deep structures must obviously
differ. Such examples as these are sufficient to indicate that the deep structure under-
lying a sentence cannot be simply a labelIed bracketing of it. Since there is good
evidence that the surface structure should, in fact, simply be a labelIed bracketing,
we conc1ude that deep structures cannot be identified with surface structures. The
inability of surface structure to indicate semantically signiticant grammatical relations
(i.e. to serve as deep structure) is one fundamental fact that motivated the develop-
ment of transformational generative grammar, in both its classical and modern
varieties.
In summary, a full generative grammar must consist of a syntactic, semantic, and
phonological component. The syntactic component generates SD's each of which
contains a deep structure and a surface structure. The semantic component assigns
a semantic interpretation to the deep structure and the phonological component
assigns a phonetic interpretation to the surface structure. An ambiguous sentence
has several SD's, differing in the deep structures that they contain (though the con-
verse need not be true).
So far I have said !ittle that is in any way controversial. This discussion has so far
simply delirnited a certain domain of interest and a certain c1ass of problems, and has
suggested a natural framework for dealing with these problems. The only substantive
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 79
comments (i.e. factual assertions) that I have so far made witrun the framework are
that the surface structure is a labelled bracketing and that deep structures must in
general be distinct from surface structures. The first of these assertions is weil
supported (see below), and would probably be widely accepted. The second is surely
much too obvious to require elaborate defence.
To go on from here to develop a substantive Iinguistic theory we must provide:
(9) (i) theories of phonetic and semantic representation
(ü) a general ac count of the notion 'syntactic description'
(iii) a specification of the dass of potential generative grammars
(iv) a general account of how these grammars function, that is, how they generate
SD's and assign to them phonetic and semantic interpretations, thus pairing
phonetically represented signals with semantic interpretations.
Before going on to discuss these substantive questions, let us reassure ourselves about
the uncontroversial character of what has preceded. Is there, in fact, anything in this
account to which exception can be taken? Surely there is no conceivable question
about the necessity for distinguisrung competence from performance in the way sug-
gested above. Having made this distinction, one may or may not choose to be
interested in the general question of accounting for linguistic competence. If one
chooses to concern rumself with trus question, he must immediately face the fact of
'creativity' and must therefore focus attention on the problem of constructing genera-
tive gramm ars. It is difficult to see how a fuH generative grammar can be regarded,
ultimately, as anything other than a system of rules that relate signals to semantic
interpretations; and, having set trus goal, one is immediately faced with the problem
of developing a rich enough notion of 'syntactic description' to support phonetic
interpretation, on the one side, and semantic interpretation, on the other. The
distinction between deep and surface structure emerges from even the most superficial
examination of real linguistic material. Hence the condusions outIined so far seem
inescapable if the the problem of studying linguistic competence is taken up. Notice
that a substantive linguistic theory involves a specification of (9iv) as weil as (9üi).
For example, an essential part of the theory of phrase-structure grammar is a
particular specification of how categories and relations are determined for generated
strings (see Chomsky, 'Logical Structure of Linguistic Theory', Cambridge, 1955,
Ch. VI), and such a specification has been presupposed whenever this theory has been
investigated. A change in trus specification is as much arevision of the theory as a
change in the specification of the dass (9iii) of potential grammars. Failure to under-
stand trus leads to immediate absurdities. Thus if one thinks of the theory of'phrase-
structure' with the technique of interpretation (9iv) left free, one can easily prove that
a phrase-structure grammar of the language L assigns to sentences of L the structural
descriptions assigned by some transformation al grammar of L, etc. This point should
be obvious without further discussion.
80 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
Suppose that one chooses not to study linguistic competence (and, concomitantly,
linguistic performance within the framework of a theory of competence). One might,
alternatively, choose to limit attention to performance, or to surface structures, or to
sound patterns in isolation from syntactic structure, or to voiced fricatives, or to first
halves of sentences. The only question that arises, if any of these proposals is
adopted, is whether any interesting result is likely to be attainable under such
arbitrary limitation of subject matter. In each of the cited cases it seems quite un-
unlikely. It is, in general, unclear why anyone should insist on studying an isolated
aspect of the general problem of grammatical description unIess there is some reason
to believe that tbis is not affected by the character of other aspects of grammar.,7
I have been discussing so far only the question of descriptive adequacyofgramrnars
and the problem of developing a linguistic theory that will provide the basis for the
construction of descriptively adequate grammars. As has been repeatedly empha-
sized, however (see, e.g., Chomsky, Syntactic Structures; 'Explanatory Models in
Linguistics', Logic. Methodology. and Philosophy of Science. eds. E. Nagel, P.
Suppes, and A. Tarski (Stanford, 1962), pp. 528-50; Current Issues in Linguistic
Theory; and Aspects of the Theory of Syntax), the goals of linguistic theory can be
set much higher than this; and, in fact, it is aprerequisite even for the study of des-
criptive adequacy that they be set higher than this. It is essential also to raise the
question of 'explanatory adequacy' of linguistic theory. The nature of this question
can be appreciated readily in terms of the problem of constructing a hypothetical
language-acquisition device AD that can provide as 'output' a descriptively adequate
gramm ar G for the language L on the basis of certain primary linguistic data from L
as an input; that is, a device represented schematically as (10):
(10) primary linguistic data .... M .... G
1 Perhaps this matter can be clarified by considering examples of the latter sort. Thus, for example, it is
quite reasonable to study semantics in isolation from phonology or phonology in isolation from semanties,
since, at the moment, there seems to be no non-trivial relation between the systems of phonological and
semantic interpretation and no significant way in which semantic considerations can playa role in phonology
or phonological considerations in semantics. Similarly, it seems quite reasonable to develop a theory of
syntactic structure with no primitive notions of an essentially semantic nature, since, at the moment, there
is no reason to ass urne that apriori semantic concepts play a role in determining tbe organization of the
syntactic component of a grammar. On the other hand, it would be absurd to study semantics (and
similarly, it seems to me, phonology) in isolation from syntax, since the syntactic interpretation of a sentence
(similarly, its phonetic interpretation) depends in an essential way on its deep (respectively, surface) structure.
And it would be absurd to develop general syntactic theory without assigning an absolutely crucial role to
semantic considerations, since obviously the necessity to support semantic interpretation is one of the
primary requirements that the structures generated by the syntactic component of a grammar must meet.
For discussion of these points, see Chomsky, Syntactic Structures and Current Issues in Linguistic Theory;
Lees. Review of Chomsky, SYlltartic Struclllres, I.g., pp. 375-408 (1957); Katz and Postal, An
Inlegraled Theory 01 Lingllistic Description, and many other rcfcrences.
F ar 100 little care has been taken in the discussion of these questions in modern linguistics. As a result,
there has been much confusion about them, and many dogmatic claims have been voiced and repeatedly
echoed with no attempt to justify or support them by serious argument. The issues are important; while no
answers to any of these questions can be given with any certainty, the tentative position that the linguist
accepts may have an important inftuence on the character of the work that he does.
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 81
We naturally want the device AD to be language-independent---that is, capable of
learning any human language and only these. We want it, in other words, to provide
an implicit definition of the notion 'human language'. Were we able to develop the
specifications for a language-acquisition device of tbis sort, we could realistically
daim to be able to provide an explanation for the linguistic intuition-the tacit com-
petence--of the speaker of a language. This explanation would be based on the
assumption that the specifications of the device AD provide the basis for language-
acquisition, primary linguistic data from some language providing the empirical
conditions under wbich the development of a generative grammar takes place. The
difficulties of developing an empirically adequate language-independent specification
of AD are too obvious to require extended discussion; the vital importance of raising
this problem and pursuing it intensively at every stage of linguistic investigation also
seems to me entirely beyond the possibility of debate (cf. the references cited above
for elaboration of this point).
To pursue the study of explanatory adequacy, we may proceed in two parallel
ways. First, we must attempt to provide as narrow a specification of the aspects of
linguistic theory listed in (9) as is compatible with the known diversity of languages-
we must, in other words, develop as rich a hypothesis concerning linguistic universals
as can be supported by available evidence. This specification can then be attributed
to the system AD as an intrinsic property. Second, we may attempt to develop a
general evaluation procedure, as an intrinsic property of AD, which will enable it to
select a particular member of the dass of grammars that meet the specifications (9)
(or, conceivably, to select a small set of alternatives, though tbis abstract possibility
is hardly worth discussing for the present) on the basis of the presented primary
linguistic data. This procedure will then enable the device to select one of the apriori
possible hypotheses--one of the permitted grammars---that is compatible with the
empirically given data from a given language. Having selected such a hypothesis,
it has 'mastered' the language described by this grammar (and it thus knows a
great deal beyond what it has explicitly 'learned'). Given a linguistic theory that
specifies (9) and an evaluation procedure, we can explain some of the aspect of the
speaker's competence whenever we can show with some plausibility that this
aspect of his competence is determined by the most highly valued grammar of the
permitted sort that is compatible with data of the kind to which he has actually been
exposed.
Notice that an evaluation procedure (simplicity measure, as it is often called in
technical discussion) is itself an empirical hypothesis concerning universal properties
of language; it is, in other words, a hypothesis, true or false, about the prerequisites
for language-acquisition. To support or refute this hypothesis, we must consider
evidence as to the factual relation between primary linguistic data and descriptively
adequate grammars. We must ask whether the proposed evaluation procedure in fact
can mediate tbis empirically given relation. An evaluation procedure, therefore, has
82 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
much the status of a physical constant; in particular, it is impossible to support or
reject a specific proposal on the basis of apriori argument.
Once again, it is important to recognize that there is nothing controversial in what
has just been said. One may or may not choose to deal with the problem of ex-
planatory adequacy. One who chooses to overlook this problem may (and, in my
opinion, surely will) find that he has eliminated from consideration one of the most
important sour ces of evidence bearing on the problems that remain (in particu1ar, the
problem of descriptive adequacy). 8 His situation, then, may be quite analogous to that
of the person who has decided to limit his attention to surface structures (to the
exclusion of deep structures) or to first halves of sentences. He must show that the
delimitation of interest leaves him with a viable subject. Hut, in any event, he surely
has no basis for objecting to the attempt on the part of other linguists to study the
general question of which he has (artificially, in my opinion) delimited one facet.
I hope that these remarks will be sufficient to show the complete pointlessness of
much of the debate over the specific evaluation procedures (simplicity measures)
that have been proposed as empirical hypotheses concerning the form of language in
the course of work in generative grammar. To mention just one example, consider
Householder's criticism (Householder, 'On Some Recent Claims in Phonological
Theory', Journal of Linguistics, I (1965), pp. 13-34) of several proposals ofHalle's
regarding an appropriate evaluation procedure for phonology. Halle presented a
certain theory of phonological processes, including, as an essential part, a certain
empirical hypothesis regarding a simplicity measure. A crucial aspect of this theory
was its complete reliance on distinctive features in the formulation of phonological
rules to the exclusion of any 'segmental' notation (e.g. phonemic notation) except as
an informal expository device. His evaluation measure involved minimization of
features in the lexicon and the phonological rules. In support of this theory he
showed that a variety of facts can be explained on these assumptions. He also dis-
cussed alternative theories that use segmental notation along with or instead of
feature notation and gave several arguments to show that under these assumptions
it is difficu1t to see how any empirically valid evaluation measure can be formu1ated
, The reason for this is quite simple. Choice of a descriptively adequate gramm ar for the language L is
always much underdetermined (for the linguist, that is) by data from L. Other relevant data can be adduced
from study of descriptively adequate gramm ars of other languages, but only if the linguist has an ex-
planatory theory of the sort just sketchcd. Such a theory can receive empirical support from its success in
providing descriptively adequate grammars for other languages. Furthermore, it prescribes. in advance, the
fonn of the grammar of Land the evaluation procedure that leads to the selection of this grammar, given
data. In this way, it permits data from other languages to playa role in justifying the grammar selected as
an empirical hypothesis conceming the speakers of L. This approach is quite natural. Following it, the
linguist comes to a conclusion about the speakers of L on the basis of an independently supported assumJT
tion about the nature of language in general--an assumption, that is, concerning the general 'faculte de
language' that makes language-acquisition possible. The general explanatory theory of language and the
specific theory of a particular language that results from application of the general theory to data each has
psychological content, the first as a hypothesis about innate mental structure, the second as a hypothesis
about the tacit knowledge that emerges with exposure to appropriate experience.
TOPICS IN THE THEOR Y OF GENERATIVE GRAMMAR 83
---in particular, he showed how various rather natural measures involving minimiza-
tion fail on empirical grounds.
Householder makes no attempt to refute these arguments but simply objects to
them because they fail to meet certain apriori conditions that he arbitrarily imposes
on any notion of 'evaluation procedure', in particular, the requirement that such a
procedure must favour grammars that use fewer symbols and that are easy for the
linguist to read. Since the grammars that Halle proposes, with their consistent reliance
on feature representation, require more symbols than grammars that use auxiliary
symbols as abbreviations for feature sets, and since Halle's grammars are (House-
holder claims) not easy to read, he concludes that the theory on which they are based
must be mistaken. Hut c1early apriori arguments of this sort have no bearing on an
empirical hypothesis about the nature of language (i.e. about the structure of a general
language-acquisition device of the sort described above). Consequently, House-
holder's critique has no relevance to any issue that Halle discusses. Unfortunately,
much of the criticism of re cent attempts to develop valid evaluation measures is based
on similar presuppositions.
Notice, incidentally, that there is an interesting but poorly understood sense in
which one can talk of the 'simplicity' or 'elegance' or 'naturalness' of a theory (of
language, of the chemical bond, etc.), but this 'absolute' sense of simplicity has no
c1ear relevance to the attempt to develop an evaluation measure (a simplicity
measure) as apart of a theory of grammar. Such a theory is an empirical hypothesis,
true or false, proposed to account for some domain of linguistic fact. The 'simplicity
measure' that it contains is a constituent part of this empirical hypothesis. This
distinction between 'simplicity' as an absolute notion of general epistemology and
'simplicity' as apart of a theory of grammar has been repeatedly emphasized; con-
fusion regarding this point has, nevertheless, been quite widespread. Failure to make
this distinction vitiates most of the criticism of evaluation procedures that has
appeared in 'recent years.
(b) THE THEORY OF TRANSFORMATIONAL
GENERATIVE GRAMMAR
HA VING now covered the first two parts of the outline given in the introductory
section, I would like to turn, much more briefly, to parts three, four, and five. These
are discussed in much more detail in Chomsky, Aspects 0/ the Theory 0/ Syntax, and
in the references cited there.
The earliest versions of transformational generative grammar made the following
general assumptions conceming syntactic structure. The syntactic component of a
84 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
grammar consists of two sorts of rules: rewriting rules and transformational rules.
The rewriting rules constitute a phrase-structure grammar (with, perhaps, a condition
of linear ordering imposed). Each rule is, in other words, of the form A ..... X (with a
possible restrietion to the context Z-W), where A is a category symbol and X, Z,
Ware strings of category or terminal symbols. The strings generated by this system
we may call base strings (an alternative term is C-tenninal strings). In the course
of generating astring, the system of rewriting rules (let us call this the base component
of the syntax) assigns to it a phrase-marker which we can call a base phrase-marker,
this being representable as a labelIed bracketing or a tree diagram with categories
labelling the nodes.
The transformational rules map phrase-markers into new, derived phase-markers.
Each transformational rule is defined by a structural analysis stating a condition on
the dass of phrase-markers to which it applies and specifying an analysis of the termi-
nal string of this phrase-marker into successive parts. The specification of the trans-
formation is completed by associating with this structural analysis a certain ele-
mentary transfonnation which is a formal operation on strings, of a certain narrow
dass. For details, see the references cited above. By defining the 'product' of two
phrase-markers as the new phrase-marker derived essentially by concatenation of the
labelled bracketings,9 we can app\y what have been called generalized (or double
base, tripie base ete.) transformations to a phrase-marker representing a sequence
of phrase-markers, mapping such a produet into a new phrase-marker by the same
apparatus as is required in the singulary ease. The transformations meet certain
ordering conditions (I return to these below), which must be stated in a separate part
of the grammar. These eonditions inc1ude a specification of certain transformation
as obligatory or obligatory relative to certain sequences of transformations. To
generate a sentence, we seleet a sequenee of (one or more) base phrase-markers and
apply singulary and generalized transformations to them, observing the ordering
and obligatoriness requirements, until the result is a single phrase-marker dominated
by S (the initial category, representing 'sentence'). If we select a single base phrase-
marker and apply only obligatory transformations, we eall the resulting sentence a
kernel sentence (a kernel sentence is not to be confused with the base string that
underlies it as weIl as possibly many other more complex sentences).
We can represent the system of transformations that apply in the process of
derivation as a transformation-marker (T-marker). To illustrate, consider the
sentence
(18) I expected the man who quit work to be fired.
9 Precise definitions of the notions mentioned here are provided in Chomsky, 'The Logical Structure of
Linguistic Theory', unpublished manuscript, Microfilm M.J.T. Library (Cambridge, Mass., 1955), and
descriptions of varying degrees of informality appear throughout the literature. In particular, a phrase·
marker is representable as a set of strings, and the 'product' of two phrase·markers is then the complex
product of the two sets (i.e. the set of all strings XY such that X is in the first set and Y in the second).
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 85
The transformational derivation of (18) might be represented by the T-Marker (19).
In this representation, BI' B 2 and B 3 are the three base phrase-markers that underlie
the (kernei) sentences (20i)-(20iii): 10
(19)
(20) (i) I expected it
(ü) someone fired the man
(iü) the man quit work
The interpretation of (19) is straightforward. It represents the fact that to form (18)
we take the three base structures underlying (20i-iii), and proceed as follows. First,
apply to B
3
the relative transformation T Rel that c.onverts it to 'who (the man) quit
work' (rather, to the abstract string that underlies this----d. n. 2), with its derived
phrase-marker. Call this new structure K \. At this point, apply the generalized
embedding transformation T Emb to the pair of structures (B 2' K I)' deleting the occur-
rence of the man in the embedded sentence in the process, giving the string 'some-
one fired the man who quit work' with its derived phrase-marker K
2
• To K
2
, apply
the passive transformation T Pass to give 'the man who quit work was fired by
someone', with the phrase-marker K 3' To this apply the deletion transformation
T Dei to give 'the man who quit work was fired', with the derived phrase-marker K
4

Now apply to the pair of structures (B I' K
4
) the generalized embedding transforma-
tion T Emb' giving 'I expected the man who quit work was fired' with the derived
phrase-marker K 5' To K 5' apply the singulary transformation T
to
giving the sentence
(18) with its derived phrase-marker K
6

10 Since I am presenting this merely as the basis for some revisions to be proposed below, I skip many
details. In particular, I am completely overlooking the question of how to describe the Auxiliary system, and
I have also supposed, for simplicity of exposition, that each of B ,-B, underlies a kernel sentence. Actually,
this is not necessary, and in the transformational gramm ars presented in Chomsky, 'The Logical Structure
of Linguistic Theory', Syntactic Structures. 'A Transformational Approach to Syntax', Lees, The Grammar
0/ English Nominalizations, and others, many of the base strings contain 'dummy symbols' (e.g. Camp. in
the case of the analysis of such sentences as (15» which are either deleted or filled in by sentence transforms
in one way or another. Thus B, might have a dummy symbol as Object, B, might have an unspecified
Subject, etc.
I am also assuming here a simpler analysis of the main (matrix) structure than was postulated in earlier
work. The reasons for this go weU beyond anything considered here. See P. Rosenbaum, 'A Grammar of
English Predieate Complement Construetions', unpublished Ph.D. dissertation (M.I.T., 1965), and, for
further related diseussion, Chomsky, Aspects 0/ the Theory 0/ Syntax. Ch, 1, § 4.
Throughout the deseription of these struetures, leite sentences as examples, inaccurately, instead of the
abstract strings that underlie them. It should be kept in mind that this is only an expository device.
86 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
I emphasize once again that only after all the transformations have been com-
pleted do we have an actual 'sentence' -----that is, a s t r i n ~ of elements that constitutes
an 'output' of the syntactic component of the grammar and an 'input' to the phono-
logical component.
Perhaps this example suffices to convey the conte nt of the notion 'T-marker' (for
further elaboration, see Chomsky, 'The Logical Structure of Linguistic Theory';
Katz and Postal, An Integrated Theory 0/ Linguistic Description). It should be dear,
from this, how any transformational derivation can be presented as aT-marker which
gives the fuH 'transformational history' of the derived sentence, induding, in particu-
lar, a specification of the base phrase-markers from which it is derived. In Chomsky,
'The Logical Structure of Linguistic Theory', a general theory of linguistic levels is
developed in an abstract and uniform way, with phrase structure and transformations
each constituting a linguistic level. On each level, markers are constructed that
represent a sentence. In particular, derived phrase-markers and T-markers fill this
function on the phrase-structure and transformational levels, respectively. Each level
is a system of representation in terms of certain primes (elementary atomic symbols
of this level). On the level of phrase structure, the primes are category and terminal
symbols. On the level of transformations, the primes are base phrase-markers and
transformations. A marker is astring of primes or a set of such strings. 80th phrase-
markers and transformation-markers can be represented in this way. Levels are
organized in a hierarchy, and we may think of the markers of each level as being
mapped into the markers of the next lowest level and as representing the lowest level
marker (that is, the phonetic representation which is the marker on the lowest,
phonetic level--the primes of this level being sets of features), which is associated
directly with an actual signal. We limit the discussion here to the levels of phrase
structure and transformational structure.
The general requirement on a syntactic theory is that it define the notions 'deep
structure' and 'surface structure', representing the inputs to the semantic and phono-
logical components of a gramm ar respectively (see above), and state precisely how a
syntactic description consisting of a deep and surface structure is gerrerated by the
syntactic rules. These requirements are met by the theory outlined above in the
following way. The rewriting rules of the base component and the rules governing
ordering and arrangement of transformations generate an infinite dass ofT-markers,
in the manner just sketched. We take aT-marker to be the deep structure; we take
the derived phrase-marker that is the final output of the operations represented in
the T-marker to be the surface structure. Thus in the case of (18), the deep structure
is the T-marker represented as (19), and the surface structure is what we designated
as K
6
• The phrase-marker K
6
, then, must contain an information relevant to deter-
mination of the form of the signal corresponding to (18) (i.e. it is to be mapped into a
phonetic representation of (18) by rules ofthe phonologie al component); theT-marker
(19)"is to contain all information relevant to the semantic interpretation of (18).
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 87
To complete the theory, we must add a description of the phonological and seman-
tic components that interpret surface and deep structures, respectively. I wiil discuss
the phonological component briefty in the fourth section, along lines suggested by
R. Jakobson, G. Fant and M. Halle, Preliminaries to Speech Analysis (Cambridge,
Mass., 1952); Chomsky, Halle and LukotT, 'On Accent and Juncture in English';
Halle, The Sound Pattern of Russian, 'Phonology in a Generative Grammar', 'On
the B a s ~ s of Phonology' (Structure of Language, eds. Fodor and Katz, pp. 324-33);
Chomsky, 'Explanatory Models in Linguistics'; and other related publications. The
theory of semantic interpretation is in a much less developed state, as noted above,
although recent work of Katz, Fodor and Postal has been quite encouraging and, as
we shall note directly, has had important consequenees for the theory of syntax
as weil.
A theory of semantic interpretation based on the syntactic model outlined above
would have to provide, first, a characterization of the notion 'semantie interpretation
of a sentence', and second, a system of rules for assigning such an object to a deep
structure, that is, aT-marker. Analogously a theory of phonetic interpretation must
specify the notion 'phonetic interpretation of a sentence' --ü must, in other words,
specify a universal phonetic alphabet--and must provide a system ofrules for assign-
ing such an object to a surfaee strueture, that is, the final derived phrase-marker of
a sentence. The notion 'semantic interpretation of a sentence' remains in a rather
primitive state, for the moment. Several important steps have been taken towards the
study of rules that assign semantic interpretations to deep structures, however.
First of all, it is evident that the grammatical relations among the elements of the
string representing a sentenee and the grammatieal functions (i.e. Subjeet, Objeet,
ete.) that these elements fulfil provide information that is fundamental for semantie
interpretation. Furthermore, it has been evident sinee the beginnings of recent work
on transformational grammar that it is the grammatical relations and grammatical
functions represented in the base phrase-markers underlying a sentence that are
critical for its semantic interpretation (for example, it is not the 'grammatieal subject'
of the passive but rather its 'logical subject' that is the subjeet in the sense relevant
to semantic interpretation). This is evident from consideration of the examples dis-
eussed throughout this paper. These examples were chosen primarily to ilIustrate this
fact, as is characteristic of expository papers in transformational grammar. As
emphasized above, it is examples of grammatical relations and functions that are
obscured in the surface representation (the IC analysis) that provide the primary
motivation for the rejection of all versions oftaxonomie syntax, and for the develop-
ment of the theory of transformational grammar.
To my knowledge, the first fairly explicit discussion of grammatical relations ofthe
deep strueture that are not represented in the aetual physieal form and organization
of the sentence, and the first general discussion of the importanee of these for sem an-
tie interpretation, is in the Grammaire generale et raisonnee of Port-Royal (1660).
88 TOPICS IN THE THEORY OF GENERATIVE GRAMM AR
For some brief references, see Chomsky, Current Issues in Linguistic Theory, § 1,
and for some further discussion, Chomsky, Cartesian Linguistics. In modern linguis-
tics, the same insight was expressed by Harris, in somewhat different terms, in his
early work on transformations, 1 1 and the point is also emphasized in Chomsky, 'The
Logical Structure of Linguistic Theory', Syntactic Structures, and in an subsequent
work on transformational grammar.
To go beyond this observation, it is necessary to define grammatical relations and
grammatical functions, and to show how the relations and functions of the base
phrase-markers playa role in determining the semantic interpretation ofthe sentence
that they underlie. A phrase-structure gramm ar is, in fact, a very natural device for
assigning a system of grammatical relations and functions to a generated string.
These notions are represented directly in the phrase-marker assigned to astring
generated by such rules, as has been frequently pointed out. Various ways of defining
these notions are discussed in Chomsky ('The Logical Structure of Linguistic
Theory'; Current Issues in Linguistic Theory; A spects of the Theory of Syntax) and
Postal (Constituent Structure). For concreteness, consider a highly oversimplified
phrase-structure grammar with the rules (21):
(21) VP
V NP
John, Bill

This grammar generates the string 'lohn saw Bill' with the phrase-marker (22):
(22)
NP .............. S--VP
I / \
John V NP
I
I
saw Bill
To the grammatical rule A XBY, we can associate the grammaticalfunction [B, Al.
Thus associated with the rules of (21) we have the grammatical functions [NP, Sl,
[VP, Sl, [V, VP], [NP, VPl. We may give these the conventional names Subject-oj,
Predicate-of, Main-Verb-oj, Object-of, respectively. Using the obvious definitions of
these notions, we can say, then, that with respect to the phrase-marker (22), John is
the Subject-of the sentence, saw Bill is the Predicate-ofthe sentence, saw is the Main-
Verb-of the Verb Phrase, and Bill is the Object-ofthe Verb Phrase. We can go on to
define grammatical relations (Subject-Verb, etc.) in terms of these and other notions
"e.g. Harris, 'Discourse Analysis', Lg., 28, pp. 18-23 (1952); 'Distributional Structure', Ward, 10,
pp. 146-62 (1954); 'Co·occurrence and Transformation in Linguistic Structure', Lg., 33, pp. 283-340
(1957).
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 89
and there are various ways in which one can attempt to formulate language-
independent definitions for the central concepts (for details, see the cited references).
The important point is that a phrase-structure grammar need not be supplemented in
any way for it to assign these properties to the strings it generates. Once we reeog-
nize the relational eharaeter of these notions, we see at onee they are alreadyassigned,
in the appropriate way, with no further elaboration of the rules.
Notiee that we might define the grammatieal funetions not in terms of the genera-
ting rules, but in terms of the phrase-marker itself, in an obvious way. Ifwe do this,
we will have amore general notion of 'grammatical function' that will apply to
derived phrase-markers as well as to the base phrase-markers. I do not go into this
here, sinee, in any event, it is only the funetions in the base phrase-markers that are
significant for semantie interpretation (but see Chomsky, Aspects 0/ the Theory 0/
Syntax, pp. 220, 221, for some diseussion of the role of 'surfaee funetions', so
defined).
The first attempt to develop a theory of semantie interpretation as an integral part
of an explieit (i.e. generative) grammar is in Katz and Fador, ('The Strueture of a
Semantie Theory'). This is the first study that goes beyond the assertion that the
base phrase-markers underlying a sentenee are, in some sense, the basic content
elements that determine its semantie interpretation. Basing themselves on the aeeount
of syntaetie strueture outlined above, Katz and F odor argue that the semantie eom-
ponent of a grammar should be a purely interpretive system of rules that maps a deep
structure (a T-marker) into a semantie interpretation, utilizing in the proeess three
sorts of information: (i) intrinsie semantie features of lexieal items; (ii) the gramma-
tieal funetions defined by the base rules; (iii) the strueture of the T-marker. The
semantic eomponent should have two sorts of'projeetion rules'. The first type assign
semantie interpretations ('readings') to eategories ofthe base phrase-markers in terms
of the readings previously assigned to the elements dominated by (belonging to) these
eategories, beginning with the intrinsie readings of the lexieal items and using the
grammatieal funetions defined by the eonfigurations of the base phrase-markers to
determine how the higher level readings are assigned; and, ultimately, assigning a
reading to the dominant eategory S. The projeetion mies of the seeond type utilize
the readings assigned in this way to base phrase-markers, and, in terms of the
elements and configurations represented in the T-marker, determine the semantie
interpretation of the full sentenee. Not much is said about type two mies; as we shall
see below, this is not a serious gap in their theory.
With this brief survey, we eonclude part three of the outline of the introductory
seeHon, having now sketched a eertain theory of generative grammar that in part
overeomes the fundamental inability of taxonomie syntax to provide an adequate
notion of deep strueture.
Turning now to part four of the outline, I would like to eonsider some of the defeets
that have been exposed in the theory just sketehed as it has been applied to linguistic
material.
90 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
In Lees, The Grammar 0/ English Nominalizations, it is shown that the negation
transformation of Chomsky (Syntactic Structures, 'A Transformational Approach
to Syntax')'2 is incorrectly formulated. He shows that there are syntactic arguments
in favour of an alternative formulation in which the negation e1em(:nt is not introduced
by a transformation but is, rather, an option al element introduced by rewriting rules
of the base, the transformation serving simply to place it in the correct position in the
senten ce. At about the same time, E. 'So Klima pointed out that the same is true of the
question transformations of Chomsky, Syntactic Structures, 'A Transformational
Approach to Syntax'. There are syntactic arguments in favour of assuming an
abstract 'question marker' as an element introduced by base rules, the question trans-
formations then being conditional on the presence of this marker (i.e. obligatory when
it appears in astring, and inapplicable otherwise). Further arguments in support of
this view, and further elaboration of it, are presented in Katz and Postal, An Inte-
grated Theory 0/ Linguistic Description. See now also Klima, 'Negation in English',
Structure 0/ Language: Readings in the Philosophy 0/ Language, eds. Fodor and
Katz, pp. 246-323.
In Katz and Postal, it is further observed that the same is true of the imperative
transformation of earlier work. In the light of this and other observations, Katz and
Post al then conclude that all singulary transformations which affect meaning are
conditional upon the presence of markers of this sort; in other words, the singulary
transformations in themselves need not be referred to by the rules of the semantic
component since whatever contribution they appear to make to the meaning of the
sentence can be regarded as an intrinsic property of the marker that determines their
applicability, and can therefore be handled in base structures by type 1 projection
rules. It folIows, then, that the function of type 2 projection rules is much more
restricted than Katz and Fodor were forced to assurne, since they need not take into
account the presence of singulary transformations in aT-marker.
Turning then to generalized transformations, Katz and Postal carry out a detailed
analysis of many examples described in earlier studies that seem to demonstrate a
contribution of generalized transformations to the semantic interpretation of the
generated sentence in some way that goes beyond mere 'amalgamation'. They 'argue
(quite convincingly, it seems to me) that in each such case, there are syntactic
grounds for regarding the description as in error; furthermore, that in each such case
the only function of the generalized transformation is to embed a sentence transform
in a position that is already in the underlying structure (let us say, by the presence of
a dummy symbol).
Generalizing upon these various observations, they conclude that the only function
of generalized transformations, so far as semantic interpretation is concemed, is to
interrelate the semantic interpretations of the phrase-markers on which they operate;
in other words, to insert the reading for the embedded phrase-marker in the position
12 Publication delays account for the discrepancy in dates, here, and in several other places.
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 91
already marked (by a dummy element) in the phrase-marker in which it is inserted.
Thus the only aspect of the T-marker that need be considered in semantic interpreta-
tion is the interrelation specified by the nodes where generalized transformations
appear in the representation. Beyond this, transformations appear to play no role in
semantic interpretation. Thus the function of type 2 rules is still further restricted.
This principle obviously simplifies very considerably the theory of the semantic
component as this was presented in Katz and Fodor, 'The Structure ~ f a Semantic
Theory'. It is therefore important to observe that there is no question-begging in the
Katz-Postal argument. That is, the justification for the principle is not that it simpli-
fies semantic theory, but rather that in each case in which it was apparently violated,
syntactic arguments can be produced to show that the analysis was in error on
internal, syntactic grounds. In the light of this observation, it is reasonable to formu-
late the principle tentatively as a general property of grammar.
Furthermore, it seems that there are good reasons for regarding even the passive
transformation as conditional upon the presence of an abstract marker in the under-
Iying string (see Chomsky, Aspects ofthe Theory 0/ Syntax, for a survey of syntactic
arguments in support of this), rather than as optional, as assumed in earlier work.
Consequently, it seems that all singulary transformations other than those that are
'purely stylistic' (cf. Chomsky, Aspects of the Theory of Syntax, pp. 221, 223, for
some discussion of this distinction-discussion, incidentally, which is far from
satisfactory, although it seems to me that areal and important distinction is in-
volved) are conditional upon markers in base strings, whether or not these trans-
formations effect semantic interpretation.
Independently of these developments, C. J. Fillmore pointed out that there are
many restrictions on the organization of T-markers beyond those that were assumed
in earlier attempts to formulate a theory of transformational grammar (Fillmore,
'The Position of Embedding Transformations in a Grammar', Word, pp. 19,208-31
(1963)). What his observations come to is essentially this: there is no ordering among
generalized transformations, although singulary transformations are ordered
(apparently linearly); there are no singulary transformations that must apply to a
matrix sentence before a constituent sentence is embedded in it by a generalized
embedding transformation,13 although there are many instances of singulary trans-
formations that must apply to a matrix sentence after embedding of a constituent
structure within it and to a constituent sentence before it is embedded; embedding
13 The terms 'matrix sentence' and 'constituent sentence' are due to Lees, The Grammar 01 English
Nominalizations; the matrix sentence is the one into which a constituent sentence is inserted by a generalized
transformation. The same not ion appears in the analysis of transformational processes in the Grammaire
generale et raisonnee, where the terms 'proposition essentielle' and 'proposition incidente' are used for 'matrix
sentence' and 'constituent sentence', respectively. Actually 'matrix proposition' and 'constituent proposition'
would, in any event, be preferable terms, since what is involved here is not an operation on sentences but
rat her on the abstract structures that underlie them and determine their semantic interpretation. This is the
way in wh ich these operations are interpreted, correctly, in the Grammaire generale et raisonnee.
92 TOPICS IN THE THEOR Y OF GENERATIVE GRAMMAR
should be regarded as substitution of a sentence transform for a 'dummy symbol'
rather than as insertion of this transform in a categorially unspecified position. The
last observation is further elaborated by Katz and PostaI, An Integrated Theory of
Linguistic Description, as noted above.
Returning now to the T-marker (19) used as an example above, we observe that it
has just the properties that Fillmore outlines. That is, singulary transformations are
applied to a matrix sentence only after embedding and the only ordering is among
singularies. But the earlier theory of T-markers left open the possibility for ordering
of a much more complex sort. It is therefore quite natural to generalize from these
empirical observations, and to propose as a general condition on T-markers that
they must always meet Fillmore's conditions and have the form illustrated in (19).
As just formulated, this principle appears to be quite ad hoc, but there is another
way of saying exactly the same thing that makes it seem entirely natural. Notice that
if no singulary transformations apply to a matrix phrase-marker before embedding,
and if, furthermore, all embedding involves the insertion of a constituent phrase-
marker in a position marked by a dummy element in the matrix structure, then we
can, in fact, dispense with generalized transformations entirely. Instead of intro-
ducing constituent phrase-markers by embedding transformations, we can permit
the rewriting rules of the base to introduce the initial category symbol S, i.e. we can
permit rewriting rules of the form A ---> ••• S ...
Wherever such a symbol is introduced, we can allow it to head a new base
derivation. In short, we can apply the linearly ordered system of base rewriting
rules in a cyc1ic fashion, returning to the beginning of the sequence each time we
come upon a new occurrence of S introduced by a rewriting rule. Proceeding in this
way, we construct what we can call a generalized phrase-marker.
We now apply the linear sequence of singulary transformations in the following
manner. First, apply the sequence to the most deeply embedded structure'dominated
by S in the generalized phrase-marker. Having completed the application of the rules
to each such structure, reapply the sequence to the 'next-higher' structure dorninated
by S in the generalized phrase-marker. Continue in this way, until, finally, the
sequence of transformations is applied to the structure dominated by the occurrence
of S which initiated the first application of base rules, i.e. to the generalized phrase-
marker as a whole. Notice that with this formulation, we have, in effect, established
the particular formal properties of the T-marker (19) as general properties of any
transformational derivation.
Let us now return to the example (18)-(20) in the light of these suggested re-
visions of the theory of transformational grammar. By the application of the re-
writing rules of the base, we construct the generalized phrase-marker (23) (ornitting
all but the central configurations, and many details).
The transformations indicated in (19) now apply, obligatorily, in the following
order. First, T Rel applies to the most deeply embedded structure. We then turn to the
TOPICS IN THE THEOR Y OF GENERATIVE GRAMMAR 93
next higher structure, i.e. the one dominated by the occurrence of S in the fourth
lineof (23). At this point, an inversion rule (not indicated in (19), though in fact also
needed in the earlier formulation) inverts the relative clause and the following N.
Next we apply the passive transformation and the subsequent deletion of the un-
specified subject, these operations now being obligatorily marked by the dummy
elements passive and 6, (standing for an unspecified category) in (23). Since no
further transformational rules apply at this point, we turn to the next high er structure
dominated by S--in this case, the full generalized phrase-marker. To this we apply
TIO' as before, giving (18). The transformations indicated in the T-marker (19) are
now obligatory and the structure of the T-marker (19) is fully determined by (23)
itself, given the general convention for cyclic application of transformations.
*-S-*
NP-----VP
I ~ ~
1 V NP
I /1"
expected '* S '*
-------
NP .YP
1 Y ------;J;-----D
I ____ ~
fi
ired Det N by passive
. / ~ I
the '* S * man
~
NP VP
thlmbn V ~ N P
I. I
qua work
Notice now that all of the information to the semantic interpretation of (18) is
contained in the generalized phrase-marker (23) that underlies (18). Furthermore,
the same will be true in all other cases, if the modifications suggested above are
correct. By the pl'inciple suggested by Katz and Postal, the singulary transformations
will not make an intrinsic contribution to meaning, and the generalized transforma-
tions will do so only in so far as they interrelate base phrase-markers. But we have
now eliminated generalized transformations in favour of a recursive operation in the
base. Consequently all information relevant to the operation of the interpretive
semantic component should be contained in the generalized phrase-marker generated
by base rules.
The advantages of this modification are obvious. It provides a more highly
94 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
structured theory wh ich is weaker in expressive power; in other words, it excludes
in principle certain kinds of derivational pattern that were permitted by the earlier
version of transformationa1 theory, but never actually found. Since the primary goa1
of linguistic theory is to account for specific properties of particular languages in
terms of hypotheses about language structure in genera1, any such strengthening of
the constraints on a general theory is an important advance. Furthermore, there is
good interna1 motivation for enriching the structure (and hence decreasing the ex-
pressive power) of transformational theory in this way, namely, in that this modifica-
tion permits us to eliminate the notion of 'genera1ized transformation' (and with it,
the notion 'T-marker') from the theory of syntax. Hence the theory is conceptua1ly
simpler. Finally, the theory of the semantic component can be simplified in that type
2 projection rules are no longer necessary at a11.
Recapitulating, we are proposing that the syntactic component of a grammar
consists of rewriting rules and transformationa1 rules. The rewriting rules are per-
mitted to introduce the initial symbol S. These rules apply in a linear sequence; if the
initial symbol S appears in a derivation, then the sequence ofrules reapplies to form
a subderivation dominated by this symbol, in the usual manner. The recursive
property of the grammar (its 'creative aspect', to return to terminology used above)
is restricted to the base component. In fact, the restriction may be still heavier than
this, since recursion may be limited to introduction of the symbol S, that is, to intro-
duction of 'propositiona1 content'. This is not a necessary property of a phrase-
structure grammar.
The base rules, applying in the way just outlined, form genera1ized phrase-
markers. The function of the transformationa1 rules is to map generalized phrase-
markers into derived phrase-markers. If the transformational rules map the genera1-
ized phrase-marker MD into the fina1 derived phrase-marker M s of the sentence X,
then MD is the deep structure of X and M
s
is its surface structure.
This approach to syntax forma1izes, in one specific way, the view that the phonetic
form of a sentence is determined by its actua1 labelIed bracketing, whereas its
semantic interpretation is determined by the intrinsic semantic properties of its lexica1
items and by a network of grammatica1 relations, not necessarily represented in the
surface structure, that interconnect these items (cf. (13». The underlying grammatica1
relations are determined by the base rules. This abstract system of categories and
relations is related to a labelIed bracketing of the actual sentence by transforma-
tiona1 rules and the interpretive rules of the phonological component. There is fairly
good reason to suppose that the base rules are rather narrowly constrained both in
terms of the symbols that may appear in them and in terms of the configurations of
these symbols, but I will not go into this further question here (see Chomsky, Aspects
01 the Theory 01 Syntax, for some discussion). In so far as information is presently
available about syntactic structure, and about the relation of signa1s to semantic
interpretations of these signa1s, this view seems compatible with it. It is worth
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 95
mentioning that a view very much like this is expressed in the Grammaire generale
et raisonnee, to which we have now had occasion to refer several times.
We might ask why a natural language should be constructed in this way; why, in
particular, should it not identify deep and surface structures and thus dispense with
the transformations that interrelate them. One would naturally attempt to find an
answer to this question on perceptual grounds. For some speculations that seem to
me worth pursuing further, see Mi1ler and Chomsky, 'Finitary Models of Language
Users', part II.
Observe that the base rules may form generalized phrase-markers that cannot be
mapped by any sequence of transformations into a surface structure. Far example,
suppose that we had chosen the phrase 'the boy' instead of 'the man' in the most
deeply embedded structure of (23). In this case, the generalized phrase-marker would
evidently constitute the deep structure of no sentence; there is no sentence for which
this structure provides the semantic interpretation. And in fact, the relative trans-
formation would block when applying to this structure, because of the lack of identity
between the Noun Phrases of the matrix and constituent sentences.
14
Hence not all
generalized phrase-markers underlie sentences and thus count as deep structures.
The deep structures are the generalized phrase-markers that are mapped into well-
formed surface structures by transformational rules. Thus the transformations serve
a 'filtering' function; in effect, they supply certain global constraints that a deep
structure must meet, constraints that are, in fact, entirely unstatable within the frame-
work of elementary rewriting rules that seem perfectly adequate for the generation of
base structures with the grammatical functions and relations that they express. For
further discussion of this property of transformations, see Chomsky, Aspects 0/ the
Theory 0/ Syntax, eh. 3.
In this way, we can construct a theory of grammatical transformations that is
conceptually simpler than the earlier version, described above, but still apparently
empirically adequate. In this modified formulation, the functions of the base rules
and the transformational rules are more clearly expressed, as are also the notions of
deep and surface structure. We have, correspondingly, a simplification of semantic
theory.15
I began this section by presenting a certain theory of grammar in outline. I have
now tried to show that this theory was too broad and rich in expressive power, and
that a much more restricted version of it (which is, furthermore, conceptually weIl
motivated) will suffice to account for what empirical data are now avaiIable. I would
14 What is involved here is a set of very general conventions on recoverability of deletion, in the trans-
formational component of a grammar. For discussion, see Chomsky, Current lssues in Linguistic Theory;
Aspects 0/ the Theory 0/ Syntax; Katz and Postal, An lntegraled Theory 0/ Linguislic Descriplion.
15 Incidentally, only embedding transformations werc considered here. It is also necessary to show how
various transformations that introduce coordinate structures (e.g. conjunction) can be developed within this
framework. For some remarks on this question, see Chomsky, Aspecls 0/ the Theory 0/ Syntax, and the
references cited there.
96 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
now like to turn to an inadequacy in earlier theory of the opposite sort, that is, to a
class of problems that show this theory to be too poor in expressive power, in a
certain way.
Let us limit our attention now to the base component of the syntax. The theory
outlined followed structuralist assumptions in supposing that the relation of lexical
items to the categories to which they belong is fundamentally the same as the relation
of phrases to categories of which they are members. Formally speaking, it was
assumed that lexical item X is introduced by rewriting rules of the form A --> X, where
A is a lexical category, exactly in the way that phrases are introduced.
16
However,
difficulties in this view quickly emerged. Shortly after the publication of the earliest
work in transformational generative grammar, it was pointed out by G. H. Matthews
that whereas the categorization of phrases is typically hierarchie, and therefore within
the bounds of phrase-structure grammar, lexical categorization typically involves
cross-classification, and therefore goes beyond these bounds. For example, a Noun
may be either Proper or Common, and, independently of this, may be either Animate
or Inanimate; a Verb may be Transitive or non-Transitive, and independently ofthis,
may or may not take non-Animate Subjects; etc. This fact is unstatable within the
framework of phrase structure grammar. Consequently, the theory of the base must
be extended in some way so as to provide an analysis of lexical categorization that is
different in fundamental respects from the analysis in terms of rewriting rules that
seem quite adequate above the level of lexical category. Similar observations were
made independently by StockweIl, Anderson, Schachter and Bach, and various
proposals have been made as to how to remedy this defect of the base component.
The general problem is studied in some detail in Chomsky, Aspects ofthe Theory of
Syntax, Ch. 2, where reference is also made to the earlier work just noted. I will
sketch briefly the proposals offered there for modification of the theory of the base
component.
Notice that the problem of lexical cross-classification is formally analogous to the
problem of phonological classification. Thus phonologie al elements are also typically
cross-classified with respect to the operation of various phonological rules. Certain
rules apply to the category of Voiced segments; others to the category ofContinuents;
membership of a segment in one of these categories is independent of its membership
in the other. This is, furthermore, the typical situation. This, in fact, is one major
reason for the view that segments (e.g. phonemes or morphophonemes) have no inde-
pendent linguistic status and are simply to be regarded as sets of features.
More generally, a lexical item can be represented phonologically as a certain set of
features, indexed as to position. Thus the lexical item bee can be represented by the
16 Notice that a1though this has been the view of all work in modern syntactic theory that has gone beyond
mere elaboration of terminology, the incorrectness of this view became obvious only when it was formalized
within the framework of an explicit theory of grammar. An essential reason for formalization and explicit-
ness is, of course, that it immediately exposes inadequacies that may otherwise be far from evident.
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 97
feature set [Consonantal
l
, Voiced
l
, non-Continuant
l
, ••• , Vocalic
2
, non-Gravez, ... l
indicating that its first 'segment' is consonantal, voiced, a non-continuant, ... , and
that its second 'segment' is vocalic, non-grave, .... Such a representation can be
given in matrix form in an obvious and familiar way. It provides a perfectlysatis-
factory solution to the cross-c1assification problem on the phonological level (and
furthermore relates very nicely to what seems to me to be for the present the most
satisfactory theory of universal phonetics namely, lakobson's theory of distinctive
features--I will presuppose acquaintance with this, in the form recently given it by
Halle, for the remainder of this paper).
Observe also that the semantic analysis of lexical items also apparently requires
a kind of feature theory, and that these features typically cross-c1assify lexical entries.
Thus Katz and Fodor, 'The Structure of a Semantic Theory', and Katz and Postal,
An Integrated Theory 0/ Linguistic Description, are led to the conc1usion, essentially,
that a lexical entry in its semantic aspect should consist of a set of semantic features.
These observations suggest that the problem of syntactic cross-c1assification be
dealt with in the same way, particularly, since it apparently involves only lexical
items and not phrase types. Adopting this rather natural proposal, let us revise the
theory of the base in the following way. The base consists of a system (presumably,
a linear sequence) of rewriting rules which we may call its categorial component.
Beyond this, it contains alexieon. The lexicon is an unordered set of lexieal entries.
Each lexical entry is simply a set of specified features. The features constituting the
lexical entry may be phonological (e.g. [± Voiced"l, where n is an integer indicating
position), semantic (e.g. [± Artifact]), or syntactic (e.g. [± Proper]). We limit OUT
attention here to the syntactic features. The categoriaI component of the base
generates no lexical items in strings (though it may introduce grammatical mor-
phemes). As a first approximation, we may think of each lexicaI category A (e.g.
Noun, Verb, etc.) as being involved only in rewriting rules ofthe form A ~ 6, where
6 is a fixed dummy symbol. Thus the final strings generated by the categorial com-
ponent (let us call these pre-terminal strings) are bascd on a 'vocabulary' (i.e. a set
of primes-see above, p. 86) consisting of grammatical morphemes and the
symbol6. The latter will occupy the position in which items from the lexicon will
be inserted, in a manner which we will describe directly. Apre-terminal string is
converted to a terminal string by insertion of an appropriate lexical item in each
position marked by 6.
RecaIl that the deep structures that determine semantic interpretation are generaI-
ized phrase-markers generated by the base component. As we noted above, it seems
plausible to develop semantic theory in terms of projection rules that assign readings
to successively higher nodes of the deep structure, basing this assignment on the
readings assigned to aIready interpreted nodes and the grammaticaI relations repre-
sented by the configuration in question. The grammaticaI relations and the order of
application of the interpretive projection rules are determined completely by the
98 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
categorial component of the base. The intrinsic semantic properties that provide the
initial readings for this process of semantic interpretation (i.e. the readings of the
lexical items that are the terminal elements of the generalized phrase-marker) are
provided completely by the lexicon. Thus the two separate aspects of the semantic
theory are mirrored in the subdivision of the base into a categorial and a lexical
component.
The functioning of the categorial component is c1ear; let us, therefore, consider
the lexicon in some further detail. The lexical entry for a certain item should contain
all information about idiosyncratic features of this lexical item, features that cannot
be predicted by general rule. Thus the fact that 'buy' begins with a Voiced non-
Continuant, that it is a transitive Verb, that it has irregular inflexions, that it involves
transfer of ownership, etc., must all be represented by features of the lexical entry.
Other properties (for example, that the intitial non-Continuant is non-Aspirated) can
be predicted by rule (in this case, a phonological rule). But there may be redundancy
ru/es of various kinds that operate on phonological, semantic, and syntactic features,
and that specify interrelations among features of the various types. In so far as
regularities concerning feature composition can be expressed by rule, the features in
question can be extracted from the lexical entry (for discussion of redundancy rules,
see Chomsky, Aspects of the Theory of Syntax, particularly Ch. 4, § 2.1). Normally,
a lexical item will be idiosyncratic in many respects. Since these can now be specified
in the lexical entry, they need no longer be represented in the rewriting rules. This
leads to an enormous simplification of the base component, as will be evident to
anyone who has ever attempted to construct a detailed grammatical descrii'tion.
Let us now consider the rule that inserts lexical items in pre-terminal strings.
Notice that this rule must take account of the structure of the phrase-marker in which
the item is being inserted. For example, when we say that a Verb is Transitive, we
are asserting that it can appear in the position-NP in a Verb Phrase. Therefore the
syntactic feature [+ Transitive] must specify some aspect of the phrase-marker in
which the item can be inserted. Let us call a feature ofthis sort a contextual Feature.
In contrast, we will call such features of Nouns as [± Human] non-contextual. The
degenerate case of a contextual feature is the feature [ ± Noun] itself, which indicates
a minimal aspect of the phrase-marker, namely, the category dominating the occur-
rence of 6. for which the item in question may be substituted. These degenerate
contextual features, we may call category features. For the category features, the
obvious notation is [± Al, where A is a lexical category. By convention, then, we
assert that an item with the category feature (+ A] can only replace an occurrence of
6. domina ted by the category symbol A.
Consider now the problem of a proper notation for the other contextual features,
e.g. transitivity. Clearly the best notation is simply an indication of the context in
which the item can occur. Thus the feature [+ Transitive] can be represented simply
[ + - NP]. Similarly, the fact that 'persuade' can be followed by a Noun Phrase
TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 99
and a following Prepositional Phrase (e.g. 'I persuaded John of the pointlessness of
his actions') can be indicated by assigning the contextual feature [+ - NP PPho the
lexical entry for 'persuade' (in fact, this is apparently the only contextual feature
needed to specify the frame in which 'persuade' can appear, all other forms being
derived by transformation-for discussion, see Chomsky, Aspects of the Theory of
Syntax). Contextual features of this sort, which specify the frame in which an item
can be substituted, we will call strict subcategorization features.
Alongside of strict subcategorization features, there are contextual features of a
radically different sort that we will call selectional features. Whereas the strict sub-
categorization features specify categorial frames in which an item may appear, the
selectional features of a lexical item X specify lexical features of the items with which
X enters into grammatical relations. Thus the selectional features for 'frighten' will
indicate that its Object must be specified as [+ Animate], the selectional features
for 'e1apse' will indicate that its Subject cannot be [ + Human] (and for true descrip-
tive adequacy, must obviously give a much narrower specification than this), etc.
Similarly, the selectional features for 'abundant' must indicate that it can be predica-
ted of 'harvest' but not 'boy', whereas the selectional features for 'clever' must contain
the opposite specification. We can represent selectional features by a notation very
much like that suggested above for strict subcategorization features.
Contextual features can be regarded as specifying certain substitution transforma-
tions. The context stated in the contextual feature specifies the condition that must
be met by the phrase-marker to which the transformation in question applies and
the manner in which this phrase-marker must be analysed for the purposes of this
transformation. Thus it defines the structural analysis of the transformation (see
above, p. 84. The elementary transformation that completes the definition of the
transformation states that the lexical item in question (i.e. the set of specified features
that constitutes the lexical entry) substitutes for the occurrence of!:::" thai" appears in
the position indicated in the structural analysis.
It is c1ear from the examples that there are many restrietions on the form of the
substitution transformations defined by contextual features. Thus the strict sub-
categorization features only involve 'Iocal contexts' ~ . e . contexts dorninated by the
phrase category that immediately dominates the lexical category for which the lexical
item is substituted. On the other hand, selectional features refer only to 'heads' of
grammatical related constructions. These restrictions can be made precise, and can
be shown to lead to certain interesting consequences concerning the possible con-
straints that may appear in a grammar. For discussion, see again Chomsky, Aspects
of the Theory of Syntax.
I have not discussed the problem of deviation from grammaticalness here. How-
ever, it is clear that whenever a grammatical rule exists, we may ask how a sentence
is interpreted that deviates from this rule. It seems that sentences deviating from
selectional rules are interpreted quite differently from those deviating from strict
100 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR
subcategorization rules. Deviation from selectional rules gives such examples as
'colourless green ideas sleep furiously', 'sincerity admires lohn', etc.; deviation from
strict subcategorization rules gives such examples as 'lohn persuaded to leave', 'lohn
found sad', etc. Sentences of the former type are often interpreted as somehow meta-
phorical; sentences of the latter type, if interpretable at all, must be looked at in an
entirely different way. Deviations from contextual rules involving category features
(see above, p. 98) are still different in interpretive potential. Thus the various types
of contextual feature- are rather different in the conditions that they im pose on
senten ce structures.
Notice incidentally that the ease with which sentences deviating from selectional
rules can be interpreted is not simply a result of the fact that 'Iow-Ievel' syntactic
features such as [± Human] or [takes Animate Object] are These features
can participate in rules that are not at all violable in the way in which selectional
rules may be (consider, for example, such expressions as 'the table who I scratched
with a knife', 'who I saw was lohn', 'a very barking dog', etc.). There is much to say
about this general problem; it is elear, however, that a nontrivial study of it demands
a rich and detailed understanding of the various types of grammatical process.
We assumed, in this brief account of syntactic features, that the features of a Noun
are inherent to it and that the features that selectionally relate Nouns to Verbs or
Nouns to Adjectives appear as contextual (selectional) features of the Verbs and
Adjectives. This was not an arbitrary decision; it can easily be justified on syntactic
grounds. For discussion of this question, and many of the other topics mentioned
briefly here, see Chomsky, Aspects 0/ Theory 0/ Syntax, Ch. 2.
With this, I conclude part 5 ofthe introductory outline. I have now briefly sketched
two major respects in which the first modem attempts to formulate a theory of gram-
maticaI transformations were shown to be defective by later work. The first defect
was one of excessive richness in expressive power. Wehave now been discussing a
defect of the opposite kind, namely, an inability to express certain aspects of gram-
matical structure, and have suggested a way to modify the theory so as to overcome
this. The theory of transformational generative grammar that results from these
modifications is conceptually quite simple, and is reasonably weil supported by what
empirical evidence is presently available. Each component of the theory has a well-
defined function; I see no way in which any of the postulated mechanisms can be
eliminated without sacrifice of descriptive adequacy, and know of no justification for
postulating a more complex structure and organization of the theory of the syntactic
component than what has been sketched in outline here. For the present, then, this
theory seems to me to constitute the most satisfactory hypothesis as to the form of
the syntactic component of a grammar.
VI
THE PHILOSOPHICAL RELEVANCE OF
LINGUISTIC THEORYl
JERROLD J. KATZ
1. INTRODUCT10N
T HIS paper defends the relevance of linguistics to philosophy on the grounds that
linguistic theory incorporates solutions to significant philosophical problems. The
particular thesis to be defended here is that certain philosophical problems can be
represented correctly as questions about the nature of language, and that, so repre-
sented, they can be solved on the basis of theoretical constructions that appear in
linguistic theory. 2
Synchronic linguistics involves two distinct but interrelated studies: a study of the
diversity in forms of linguistic communication and a study of the limits of such
diversity. In the former, linguists investigate what is unique about individual natural
languages and formulate such facts in what are called linguistic descriptions (or
generative grammars). In the latter, linguists investigate what is common to all
natural languages and formulate these more general facts about language in linguistic
theory. Linguistic theory, therefore, is a specification of the universals of language.
Given this notion of linguistic theory, the thesis of this paper asserts that theoretical
constructions initially devised by linguists to enable linguistic theory to systematically
state uniforrnities across natural languages also fulfil the conditions on solutions to
certain philosophical problems, owing to the nature of those problems. This thesis
should not be interpreted as asserting that the linguist's descriptions of natural
languages reveal philosophical insights that somehow must escape philosophers
looking at the same languages. This thesis does not concern either a philosopher's
From The Linguisfic Turn, ed. Richard Rorty (University of Chicago Press, 1967), pp. 340-55. !Cl 1967
by the University of Chicago. All rights reserved. Reprinted by permission ofthe author and the University
of Chicago Press.
I This paper is a revised and expanded version of my paper 'The Relevance of Linguistics to Philosophy',
The Journal 0/ Philosophy. This work was supported by a grant from the National Institute of Health,
No. MH-05120-04, to Harvard University, Center for Cognitive Studies. It was also supported in part by the
Joint Services Electronics Program (Contract DA 36-039-AMC-03200(E»; and in part by the U.S. Air
Force (Contract AF 19 (628)-2487), National Science Foundation (Grant GP-2495), The National Institutes
of Health (Grant MH-04737-05), and the National Aeronautics and Space Administration (Grant Ns
G-496).
2 In the final section of this paper, I will indicate another sort of relevance oflinguistic theory tophilosophy.
102 PHILOSOPHICAL RELEV ANCE OF LINGUISTIC THEORY
or a linguist's account of the facts about a natural language, but rather concerns
the more abstract matters that are dealt with in an account of the facts about
language in general.
If the defence of this thesis is successful, then linguistics is not incidentally pertinent
to philosophy, in the way that philosophy of science bears upon the c1arification of
methodology and theory construction in linguistics, but is directiy relevant in the
same way that philosophical theories themselves are. Since I have no apriori notions
about the apriori character of philosophical investigations, I am in no way disturbed
by the fact that we might have to know quite a lot about extra-logical matters in order
to solve certain philosophical problems.
Within the confines of a paper of this sort it is, of course, impossible to present
all the arguments on behalf of this thesis, nor is it possible to formulate those that
are presented in their full form. The present paper, then, is best regarded as a
presentation of the thesis itself, with a sketch of some of the arguments that can be
given for it.
2. THE RATIONALE FOR APPEALING TO L1NGUISTICS
At the outset, it is appropriate to ask why an appeal to linguistics is necessary; why,
that is, we find it necessary to go outside the boundaries of contemporary philosophy
to search for the solutions to philosophical problems. The answer is simply that the
approaches to the philosophical problems concerned that are available in' contem-
porary philosophy have not dealt successfully with these problems and, moreover,
that these approaches contain inherent difficuities which make it quite unlikely that
they can deal successfully with them. In other publications, particuiarly 'What's
Wrong with the Philosophy of Language?' and The Philosophy 0/ Language, 3 I tried
to show in some detail why the two major approaches in contemporary philosophy,
Logical Empiricism (Logical Positivism) and Ordinary Language Philosophy, are
inherently incapable of providing adequate, well-motivated solutions to the major
philosophical problems they tackled. It is neither necessary nor possible to repeat
my criticisms here. The general character of the difficuities, however, is this. Logical
empiricism confined its efforts to the construction of highly arbitrary and con-
ceptually impoverished theories about a c1ass of artificial languages whose structure
bears little similarity to that of natural languages. Ordinary language philosophy
preoccupied itself with unearthing the most minute and detailed facts about the use
of English locutions to the almost complete neglect of any concern with theory. Thus,
while the former offered us philosophically irrelevant theories, the latter failed to give
us any theory. Of course, both approaches prided themselves on their shortcomings,
turning their vices into alleged virtues. Logical empiricism prided itself on its ex-
J J. J. Katz and J. A. Fodor, 'What's Wrang with the Philosophy of Language?' lnquiry, V (1962),
pp. 197-237, and J. J. Katz, The Phi/osophy 01 Language (Harper & Row, New York, 1966).
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 103
clusive concern with artificial languages, claiming that natural languages are too
irregular, amorphous, and vague to provide a basis for the solution to philosophical
problems. Ordinary language philosophy prided itself on its avoidance of theory
construction, claiming that theories cause the very philosophical perplexities that
philosophy seeks to resolve by examination of the use of particular linguistic con-
structions. But the claim of the logical empiricists was never submitted to empirical
investigations, nor, on the other hand, did logical empiricism provide an alternative
standard for justifying a theory of artificial languages that might serve as a replace-
ment for the standard of conformity to the facts of language that had been eliminated.
Ordinary language philosophers never established that the theories in which they
located the source of certain philosophical problems were not just bad theories, and
so never seriously asked what a good theory would be able to do toward supplying
a conceptual systematization of the facts of language that might offer solutions for
the philosophical problems that arise in the course of ordinary, theoretically UD-
sophisticated, uses of language. Accordingly, the solutions that these approaches
presented were based either on unmotivated, and hence arbitrary, principles or on
particularistic analyses of locutions whose bearing on philosophical problems were
neither established nor made fully clear.
This unsatisfactory situation led a number of philosophers---with Quine as the
most notable example--to turn some of their attention to empirical linguistics.
Though, of course, I endorse this concern, I differ with them on a number of funda-
mental points. One is the question of what it is in linguistics for which philosophical
relevance ought to be claimed, and another is the nature and extent of the relevance
of linguistics to philosophy.
3. L1NGUISTIC THEORY
I have claimed that the part of linguistics that is relevant to philosophy is linguistic
theory. In this section, I will try to explain what this part of linguistics iso
As mentioned above, linguistic theory is a specification of the uni versals of
language, the principles of organization and interpretation that are invariant from
natural language to natural language. Linguistic theory expresses such invariants in
the form of a model of a linguistic description, of which each empirically successful
linguistic description must be an instance, exemplifying every aspect of the model.
particular linguistic descriptions account for the diverse ways in which different
naturallanguages realize the abstract structural pattern displayed in the model, while
the model itself describes the form of a system of empirical generalizations capable
of expressing and organizing the facts about a naturallanguage.
Accordingly, the construction of linguistic theory and linguistic descriptions are
strongly interdependent. Linguists can abstract out the common features from a set
of linguistic descriptions and so generalize from them to hypotheses about linguistic
104 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
universals. Alternatively, linguists can facilitate their task of describing a language
by using the model provided by linguistic theory as a pattern for their systematization
of the facts they uncover in field work. As a consequence, the justification of both
linguistic theory and individual linguistic descriptions have a common basis, viz. the
facts from natural languages on which linguistic descriptions depend for their
empirical support. Since putative linguistic universals are inductively extrapolated
generalizations, projected from known regularities cutting across the set of already
constructed linguistic descriptions, their empirical adequacy is thus a matter of
whether further facts, upon which newly constructed linguistic descriptions will
eventually be based, continue to support these generalizations. Thus, the same facts
that confirm or disconfirm particular linguistic descriptions also confirm or disconfirm
a linguistic theory. Notice, finally, that if the general form of a particular linguistic
description can be deduced from linguistic theory, then that linguistic description will
be far better conflrmed than were it to derive its support solely from the facts about
the language it describes, since it will also be supported by a wealth of evidence from
many natural languages via the connection through linguistic theory between their
linguistic descriptions and this one.
Linguistic theory consists of three sub-theories, each corresponding to one of the
three components of a linguistic description. The terms 'phonological theory', 'syntac-
tic theory', and 'semantic theory' refer to these sub-theories, and 'phonological com-
ponent', 'syntactic component', and 'semantic component' refer to the correspond-
ing parts of a linguistic description. The phonological, syntactic, and semantic
components are rule-formed descriptions of knowledge that a speaker has ac-
quired in attaining fluency. The first states the rules determining the phonetic struc-
ture of speech sounds in a language; the second states the rules determining how
speech sounds with a fixed phonetic shape are organized into sentential structures;
the third states the rules determining how such sentential structures are interpreted
as meaningful messages. At the level of linguistic theory, phonological, syntactic, and
semantic theory, jointly, characterize the form of the rules in a linguistic description,
specify the theoretical constructs utilized in writing actual rules in appropriate forms,
and determine the relations between rules within each component.
Linguistic theory also specifies the relations between these components that weid
them into an integrated linguistic description. The fundamental problem to which a
linguistic description addresses itself is that of explicating the common system of
rules on the basis of which different speakers of the same language can correlate the
same speech signal with the same meaningful message. The ability ofspeakers to trans-
mit their thoughts and ideas to one another through the vehicle of articulated speech
sounds presupposes that each speaker has mastered a common system of rules
within which each well-formed utterance receives a fixed semantic interpretation.
Linguistic communication takes place when the same associations between sounds
and meaning are made by different speakers in verbal interaction. Since the linguistic
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 105
description must formally simulate the sound-meaning correlations made by
speakers, its components must be related to one another in such a manner that the
representations given by the phonological and syntactic components of the phonetic
and syntactic character of a sentence are formally connected with the representation
given by the semantic component of its meaning. Tbe model of a linguistic description
offered by linguistic theory must show that the various schemes for making such
correlations that are found in different natural languages are instances of a general
formula which is the same for all natural languages.
On the current model of a linguistic description,4 this formula is embodied in the
following organizational pattern for linguistic descriptions. The syntactic component
is the generative source of the linguistic description. It generates abstract formal
objects which are the input to the phonological and semantic components. Tbeir
outputs are, respectively, phonetic representations and semantic interpretations. Both
these components are, therefore, purely interpretive systems. The output of the
syntactic component is a syntactic description of each sentence of the language which
consists of a set of phrase-markers, where a phrase-marker can be thought of as a
labelIed bracketing of the constituents of a sentence. Tbe bracketing teIls us that the
elements enclosed within a single bracket form a constituent and the labelling teIls
us the syntactic category to which the constituent belongs. Thus, two words, phrases,
or clauses are constituents of the same type if and only ifthey receive the same label.
Now, the set of phrase-markers which constitutes the syntactic description of a
sentence consists of a sub set of underlying phrase-markers and a single superjicial
phrase-marker; the number ofunderlying phrase-markers in the syntactic description
of a sentence indicates its degree of syntactic ambiguity (so that each underlying
phrase-marker represents a syntactical unique sentence). An underlying phrase-
marker describes that aspect of the syntax of a sentence of which its meaning is a
function, while a superficial phrase-marker describes that aspect which determines
its phonetic shape. Accordingly, the rules of the semantic component operate on the
underlying phrase-markers of a sentence to provide its semantic interpretations, and
the rules of the phonological component operate on the superficial phrase-marker to
provide its phonetic representation. The underlying phrase-markers in a syntactic
description are related to the superficial phrase-marker by virtue of the fact that this
same superficial phrase-marker is transformationally derived from each of them.
That is, transformations are syntactic rules that generate superficial phrase-markers
from underlying phrase-markers. Thus, the set of underlying phrase-markers in a
given syntactic description is automatically connected with the superficial phrase-
marker in that description because the latter is obtained from each of the former
by a specifiable (but in each case different) sequence of transformations and this
4 Cf. J. J. Katz and P. Postal, An Integrated Theory 0/ Linguistic Descriptions (Massachusetts Institute
of Technology Press, Cambridge, 1964) and N. Chomsky, Aspects o/the Theory 0/ Syntax (Massachusetts
Institute of Technology Press, Cambridge, 1965).
106 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
superficial phrase-marker is not transformationally obtainable from any underlying
phrase-marker outside this set. Thus, the linguistic description will correlate the
phonetic representation of a sentence with its semantic interpretation as desired,
the corre1ation being effected by the transformations in the syntactic component and
the manner in which the phonological and semantic components are organized to
operate, respectively, on a superficial phrase-marker and the underlying phrase-
marker from which it was transformationally derived.
4. ADVANTAGES OF LINGUISTIC THEORY AS A STARTING POINT
The advantages of starting with Iinguistic theory as a basis for the treatment of
philosophical problems are strictly complementary to the previously mentioned
disadvantages of the logical empiricist and ordinary language philosophy approaches.
First, instead of having to content ourselves with philosophical solutions that rest on
the arbitrary principles of some artificial language or on an assortment of comments
on the use of certain words or expressions from some natural language, if we base
our solutions to philosophical problems on Iinguistic theory we have a straightforward
empirical basis on which to justify such solutions in terms of the empirical evidence
that provides the support for successful linguistic descriptions. For the justification
of the theoretical constructs used in the solutions is provided by the very same
evidence that empirically warrants their introduction into linguistic theory. Hence, we
avoid the difficulties that stern from the absence of any empirical controls on a
solution to a philosophical problem and from the failure to clarify the relation be-
tween facts about the uses of words and expressions and the solutions on behalf of
which they are adduced. Second, instead of having to resort to oversimplified and
largely unexplained concepts or having no theory at all to appeal to, we can utilize
the rich stock of concepts that express the common structure of naturallanguages,
explicitly defined in a formalized theory of linguistic universals, in order to obtain
solutions to philosophical problems.
There is a further advantage of this starting point. Basing a solution to a philo-
sophical problem on linguistically universal principles avoids relativizing that solution
to one or another particular natural language. Plato and Aristotle wrote in Greek,
Descartes in French, Kant in German, and Hume in English. But the philosophical
problems about which they wrote were language-independent questions about a
common conceptual structure. Accordingly, if it is these problems, rather than some
specialized ones having to do exclusively with one language, for which philosophers
seek solutions, then we cannot narrow the scope of solutions to philosophical prob-
lems. It is just as absurd to say that the solution to a certain philosophical problem
is such-and-such in English as it is to say that a broken back is such-and-such a
condition among Chinese. By starting with linguistic theory, we can argue from the
concepts that underlie language in general to the solution of a philosophical problem,
PHILOSOPHICAL RELEVANCE OF LlNGUISTIC THEORY 107
instead of having to argue from concepts that, for all we can say, might underlie
only one or another particular language. Furthermore, we can better justify the
solution to a philosophical problem because we can adduce far stronger evidence
than ifwe were restricted to the data from one or another particular language (cf. § 3,
the concluding remarks of paragraph 3).
5. THE EPISTEMOLOGICAL AND THE PSYCHOLOGICAL
Given that linguistic theory is a formal reconstruction of the universal principles by
which speakers relate speech signals and meaningful messages, it is clearly an explica-
tion of a facet of a human ability. This makes it in some sense a psychological theory.
A philosophical problem, on the other hand, concerns the structure of concepts and
the grounds for the validity of cognitive and evaluative principles, which makes it
epistemological in the broad sense, not psychological. How, then, can linguistic
theory ofTer solutions to philosophical problems when the 'solutions' are apparently
not even addressed to the right type of problem?
This criticism rests on a failure to distinguish two senses of the term 'psycho-
logical'. The distinction depends on the difference between a speaker's linguistic
competence, what he tacitly knows about the structure of his language, and his
linguistic perfonnance, what he does with this knowledge. A theory in linguistics
explicates linguistic competence, not linguistic performance. It seeks to reconstruct
the logical structure ofthe principles that speakers have mastered in attaining tluency.
On the other hand, a theory of performance seeks to discover the contribution of
each of the factors that interplay to produce natural speech· with its various and
sundry deviations from ideal linguistic forms. Thus, it must consider such linguisti-
cally extraneous factors as memory span, perceptual and motor limitations, lapses
of attention, pauses, level of motivation, interest, idiosyncratic and random errors,
etc. The linguist, whose aim is to provide a statement of ideal linguistic form un-
adulterated by the influence of such extraneous factors, can be compared to the
logician, whose aim is to provide a statement ofideal implicational form unaldultera-
ted by extraneous factors that influence actual inferences men draw.
Hence, there are two senses of 'psychological': on one, the subject of a psycho-
logical theory is a competence, and in the other, a performance. The criticism cited
above applies to a proposed solution for a philosophical problem extracted from
a theory that is psychological in the latter sense. But it does not apply to one
extracted from a theory in linguistics that is psychological in the former. A theory
of performance cannot solve a philosophical problem such as that of formulating a,n
inductive logic that is a valid codification of the principles of nondemonstrative
inference in science and daily life. People can be quite consistent in drawing non-
demonstrative inferences according to invalid principles, and be inconsistent in their
practice of using valid ones. Because a theory of performance must accept such
108 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
behaviour at face value, it has no means of correcting for the acceptance of invalid
principles and the rejection of valid ones. In contrast, however, a theory of com-
petence does. Since it regards performance only as evidence for the construction of
an idealization, it sifts the facts about behaviour, factoring out the distorting
influences of variables that are extraneous to the logical structure of the competence.
Such a theory has built in a means for correcting itself in cases where invalid
principles were accepted or valid ones rejected. Therefore, linguistic theory cannot
be criticized as irrelevant to the solution of philosophical problems.
6. GRAMMATICAL FORM AND LOGICAL FORM
But to establish the re1evance of Iinguistic theory, it must be shown to offer solutions
to significant philosophical problems. One of the pervasive problems of modern
philosophy is that of distinguishing between the grammatical and logieal forms of
sentences. It has long been recognized that the phonetic or orthographic realization
of many sentences is such that no analysis of them in terms of traditional taxonomie
grammar can reveal the true conceptual structure of the proposition(s) that they
express. Almost invariably, however, this recognition has led twentieth-century
philosophers--Russell, the early Wittgenstein, Carnap, and Ryle, to mention some
notable examples--to seek a philosophical theory about the logical form of proposi-
tions. They assumed that gramm ar had done what it could but that its best is not
good enough, so that a philosophical theory of one sort or another is needed to exhibit
the conceptual relations unmarked in grammatical analysis.
This assumption is open to a serious challenge, even aside from the fact that such
philosophical theories have not achieved much success. From the same cases where
grammatical form and logical form do not coincide, one can conclude instead that
the traditional taxonomic theory of grammar, on which these philosophers' con-
ception of grammatical form is based, is too Iimited to reveal the underlying concep-
tual structure of a sentence. Suitably extended, grammar might weil reveal the facts
about logical form, too. Philosophers who accepted this assumption simply over-
looked the possibility that traditional taxonomic grammar might not be the last
word on grammar.
The alternative to a philosophical theory about logical form is thus a linguistic
theory about logical form. Support for this alternative has come recently from
Chomsky's work on syntactic theory which shows that traditional taxonomic
grammar is too Iimited and revises it accordingly. 5 The feasibility of this alternative
rests on whether Chomsky's criticism is directed at just the features of traditional
taxonomic gramm ar that make it incapable of handling logical form and on whether
the revision provides the theoretical machinery to handle it.
'Cf. N. Chomsky, Syntactic Structures (Mouton & Co., The Hague, 1957) and P. Posta!, Con·
stituent Structure, Publication Thirty of the Indiana University Research Centre in Anthropology, Folklore,
and Linguistics (Bloomington, 1964).
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 109
The traditional taxonomie description of an utterance type is a single labelIed
bracketing that segments it into continuous phonetic sketches and c1assifies them as
constituents of one, or another sort. Chomsky's basic criticism is that such descrip-
tion cannot mark a variety of syntactic features because it fails to go below the
surface structure of sentences. Consider the sentences : (i) 'lohn is easy to leave' and
(ii) 'lohn is eager to leave'. In a traditional taxonomic description, both receive the
same syntactic analysis:
«lohn)NP ((iS)([';:Url) A (to leave»vp)s
This analysis, which, on the terminology introduced in Section 3, is the superficial
phrase-marker for (i) and (ü), does not mark the logical difTerence that in (i) 'lohn'
is the object of the verb 'leave' whereas in (ii) 'lohn' is its subject. Consider, furt her,
a sentence like: (üi) 'lohn knows a kinder person than Bill.' The syntactic ambiguity
of (iii) cannot be represented in its taxonomie description beeause a single (super-
ficial) phrase-marker cannot explicate the different propositional structures under-
lying the terms of its ambiguity. Finally, eonsider a normal imperative such as:
(iv) 'help him!' Ellipsis, which in such cases absents the subject and future tense
auxiliary constituent, cannot be handled by a traditional taxonomie deseription
because it deals only with the phonetically or orthographically realized constituents
of a senten ce. 6
These diffieulties cannot be remedied by enriching the complexity of superficial
phrase-markers. More elaborate segmentation and subc1assification cannot overcome
the inherent inability of this form of description to represent relational information.
Rather, the superficial phrase-marker, as it stands, has a proper role to play in
syntactic description, viz. that of providing the most compact representation of the
syntactic information required to determine the phonetic shape of a sentence. What
is wrong is that the superficial phrase-marker, beeause it is the only type of descrip-
tion sanctioned by the tradition al taxonomie theory of grammar, is made to do work
that, in principle, it eannot do so long as it must still play its proper role. To right
this wrong, Chomsky introduced the conception of a grammar as a generative, trans-
formational system to supersede the eoneeption of a grammar as a set of segmenta-
tion and c1assification procedures. Within this new conception, Chomsky and others
developed the concept of an underlying phrase-marker,
7
a form of syntactie des-
cription in which semantically significant grammatical relations can be adequately
represented and shown to underlie the phonetic form of sentences on the basis of
transformational rules that derive superficial phrase-markers from appropriate under-
lying phrase-markers by formally specified operations.
6 For the syntactic motivation behind the claim that there are such phonetically unrealized constituents
in normal imperatives, cf. P. Postal, 'Underlying and Superficial Linguistic Structures', The Harvard
Educational Review, XXXIV (1964).
7 J. J. Katz and P. PostaI, op. ci!., and N. Chomsky, Aspects 0/ the Theory 0/ Syntax. The notion or an
underlying phrase-marker used here is the same as Chomsky's notion or a deep phrase-marker.
110 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEOR Y
The logical difference between (i) and (ii) noted above can be indicated with the
underlying phrase-markers.
8
(I) «(it) «one)NP «(leaves}y (John)NP)YP)s>NP(is)(easy)}yp)s
(II) «JOhn)NP «isX(eagerX(John)NP (leaves}yp)S> A}YP)S
The grammatical relations subject 0/ and object 0/ are defined in syntactic theory
in terms of subconfigurations of symbols in underlying phrase-markers. A simplified
version of their definitions is as folIows.
Given a configuration of the form «X) NP (Y)yp>s or «X)NP«Y)Y (Z)NP)YP)S' Xis
the subject 0/ the verb Y and Z is the object 0/ the verb y.9
By this definition, 'John' in (i) is marked as the object of the verb 'leaves' because
it occupies the Z-position and 'leaves' occupies the V-position in the appropriate
subconfiguration of (I), and 'John' in (ii) is marked as the subject of'leaves' because
it occupies the X-position and 'leaves' occupies the V-position in the appropriate
subconfiguration of (11).
Further, since a sentence can be assigned more than one underlying phrase-marker
in a transformational syntactic component, syntactic ambiguities like those in (iii)
can be represented in terms of appropriate1y different underlying phrase-markers
transformationally associated with the same superficial phrase-marker. Thus, the
superficial phrase-marker for (iii), viz.
«John) NP «knows) Y «aXkinder) (person)(than)(Bill» N ~ y ~ s
is associated with two underlying phrase-markers both of which have the general
form, 10
but where in one . . . is
(the person) Np((is)(more)(than)((Bill) Np((is)(kind) A) yp) s(kind) A) YP
while in the other . . . is
(the person)NP((isXmore)(than)((the)(Bill)NP((knows}y(the person)NP}Yp)S (per-
son» NP«is Xkind) A}YP) s(kind) A}YP
'For further discussion, cf. G. A. Miller and N. Chomsky, 'Finitary Models of Language Users',
Handbook of Mathematica/ Psych%gy, vol. II, ed. D. R. Luce, R. R. Bush, and E. Galanter (John Wiley &
Sons, New York, 1963), pp. 476-80.
9 Note that this definition reconstructs the intuitive notion that the subject is the noun phrase preceding
the verb in a simple sentence and that the object is the noun phrase following it. Restricting the definition
to underlying phrase-markers makes it possible to have a single definition because complex sentences are
then handled in terms of the simple sentences from which they are constructed.
10 For further discussion, cf. C. S. Smith, • A Class of Complex Modifiers in English', Language, XXXVII
(1961), pp. 342-65.
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 111
The former case underlies the term of the ambiguity on which the person that lohn
knows is kinder than Bill is, and the latter case underlies the term on which the person
that lohn knows is kinder than the person Bill knows.
Finally, in ellipsis phonetically unrealized constituents can be specifiedinunderlying
phrase-markers and deleted in the transformational derivation of the superficial
phrase-marker. This enables us to account for their syntactic relations and their
semantic contribution without falsely characterizing the phonetic shape of the
sentence, as would be required if we modified the superficial phrase-marker to
account for them.
But, although it is dear from these examples that the distinction between under-
lying and superficial syntactic structure is a significant step toward the philosopher's
distinction between logical form and grammatical form, even a fully developed trans-
formation al syntactic component would not provide all the theoretical machinery
necessary to deal adequately with logical form. Philosophers have rightly held that
an analysis of the logical form of a sentence should not only tell us about the formal
relations between its constituents, but should also tell us about the semantic properties
and relations of the proposition(s) expressed by it. In particular, an account of the
logical form of a sentence should specify whether it is (1) semanticallyanomalous
(i.e. whether it expresses any proposition at all), (2) semantically ambiguous (i.e.
whether it expresses more than one proposition, and if so, how many), (3) a para-
phrase of a given sentence (i.e. whether the two sentences express the same proposi-
tion), (4) analytic, (5) contradictory, (6) synthetic, (7) inconsistent with a given
senten ce, (8) entails or is entailed by a given sentence, (9) apresupposition of a given
sentence, and so on.
The fact that a transformational syntactic component does not suffice, by itself,
to deterrnine such semantic properties and relations has brought about the formula-
tion of a conception of a semantic component designed to determine them. 11 This
conception is based on the idea that a speaker's ability to produce and understand
sentences he has never before spoken or heard depends on his mastery of principles
according to which the meaning of new and unfamiliar sentences can be obtained by
a process in which the meaning of syntactically compound constituents is composed
out of the meanings of their parts. The semantic component formally reconstructs
these compositional principles. It has a dictionary that contains an account of the
meaning of each syntactically atomic constituent in the language, i.e. representations
of the senses of lexical items, and a set of projection rules that provide the combina-
torial machinery for representing the senses of compound constituents on the basis
of representations ofthe senses ofthe lexical items that make them up. The dictionary
11 J. J. Katz and J. A. Fador, 'The Structure of a Semantic Theory', Language, XXXIX (1963),
pp. 170--210; reprinted in The Structure 01 Language: Readings in the Philosoph)' 01 Language, ed. J. A.
Fadar and J. J. Katz (Prentice-Hall Inc., Engelwoad Cliffs, 1964), pp. 479-518; J. J. Katz, 'Recent Issues
in Semantic Theory', Foundations 01 Language, val. 3, no. 2 (1967), pp. 124-94.
112 PHILOSOPHICAL RELEVANCE OF LlNGUlSTIC THEORY
is a list of entries, each of which consists of a lexical item written in phonological
form, a set of syntactic features, and a set of lexical readings. A lexical reading,
which represents one sense of a lexical item, consists of a set of semantic markers
and a selection restrietion.
A semantic marker is a theoretical term representing a dass of equivalent con-
cepts. For example, the semantic marker (Physical Object) represents the dass of
concepts of a material entity whose parts are spatially and temporally cO""ltiguous and
move in unison each of us has in mind when we distinguish the meanings of words
like 'chair', 'stone', 'man', 'building', etc., from the meanings of words like 'virtue',
'togetherness', 'shadow', 'after-image', etc. Semantic markers enable us to state
empirical generalizations about the senses of words (expressions, and sentences), for,
by induding the semantic marker (Physical Object) in a lexical reading for each of the
words in the former group and excluding it from the lexical readings for words in the
latter, we thereby express the generalization that the former words are similar in
meaning in this respect but that the latter are not. Aselection restriction states a
condition-framed in terms of a requirement about the presence or absence of certain
semantic markers---ilnder which a reading of a constituent can combine with
readings of other constituents to form derived readings representing conceptually
congruous senses of syntactically compound constituents.
The semantic component operates on underlying phrase-markers, converting them
into semantically interpreted underlying phrase-markers, which formally represent
all the information about the meaning of the sentences to which they are assigned.
Initially, each of the lexical items in an underlying phrase-marker receives a subset
of the lexical readings that it has in its dictionary entry. Then, the projection rules
combine lexical readings from sets assigned to different lexical items to form derived
readings, and these are combined to form further derived readings, and so on. Each
derived reading is assigned to the compound constituent whose parts are the con-
stituents whose readings were combined to form the derived reading. In this way
each constituent in the underlying phrase-marker, induding the whole sentence, is
assigned a set of readings that represents its senses. Thus, a semantically interpreted
underlying phrase-marker is an underlying phrase-marker each of whose brackets
is assigned a maximal set of readings (where by 'maximal' is meant that the set
contains every reading that can be formed by the projection rules without violating
a se1ection restriction).
We are now in a position to define the notions 'logical form' and 'grammatical
form': The logicaljonn oj a sentence is the set ojUs semantically interpreted under-
lying phrase-markers; the grammaticaljorm of a sentence is its superjicial phrase-
marker with its phonetic representation. Accordingly, the syntactic and semantic
components for a language comprise a theory of logical form for that language, while
the syntactic and phonological components for the language comprise a theory of
grammatical form for it. Similarly, syntactic theory and semantic theory comprise
PHILOSOPHICAL RELEV ANCE OF LlNO UISTIC THEOR Y 113
a theory of logical form in general, while syntactic theory and phonological theory
comprise a theory of grammatical form in general.
7. SEMANTIC PROPERTIES AND RELATIONS
However, semantic theory does much more than complete the account of the dis-
tinction between logical form and grammatical form. It also provides solutions to the
philosophical problems of explicating concepts such as (1) through (9). Definitions
of these concepts thus constitute further support for the relevance of the thesis that
I am defending.
Restricting our attention to syntactically unambiguous sentences, we can provide
a general idea of such definitions. First, a sentence is semantically anomalous just
in case the set of readings assigned to it is empty. This explicates the notion that
what prevents a sentence from having a meaningful interpretation are conceptual
incongruities between senses of its parts that keep these senses from compositionally
forming a sense for the whole sentence. Second, a sentence is semantically unique.
i.e. expresses exactly one proposition, just in case the set of readings assigned to it
contains one member. Third, a sentence is semantically ambiguous just in case the
set of readings assigned to it contains n members, for n > 1. Fourth, a sentence is a
paraphrase of another sentence just in case the set of readings assigned to them have
a member in common. Fifth, two sentences are full paraphrases just in case each is
assigned the same set of readings. Sixth, a sentence is analytic if there is a reading
assigned to it that is derived from a reading for its subject and a readiag for its verb
phrase such that the latter contains no semantic markers not already in the former.
1
2
Finally, a sentence entails another senten ce if each semantic marker in the reading
for the latter's subject is already contained in the reading for the former's subject
and each semantic marker in the reading for the latter's verb phrase is already con-
tained in the reading for the former's verb phrase.
13
The adequacy of these definitions as solutions to the philosophical problems to
which they are addressed is entirely a matter of their empirical justification. Since
such definitions are part of a semantic theory, which, in turn, is part of linguistic
theory, they must be justified on the same evidential basis as any other linguistic
12 This is a simplified version of the definition of analyticity given in J. J. Katz, 'Analyticity and
Contradiction in Natural Language', The Slructure 01 Language: Readings in the Philosophy 01 Language,
and in J. J. Katz, The Philosophy 01 Language. This concept of analyticity may be regarded as a linguistically
systematized version of Kant's concept of analyticity, with two refinements: one, that Kant's somewhat
vague and restricted notions of subject and predicate are replaced by the formally defined grammatical
relations subject 01 S and verb phrase 01 S, and two, that Kant's metaphorical notions of concept and of
containment are replaced by the formal analogues of a reading and the inclusion of a set of semantic markers
in another set. The semantic properties of contradiction and syntheticity can also be defined, as can
inconsistency and other related cases; their definitions, however, involve too many technicalities to be
given here.
13 A conditional sentence is analytic just in case its antecedent entails its consequent.
114 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
universal. Thus, their empirical evaluation consists in verifying the predictions to
which they lead about the semantic properties and relations of sentences from natural
languages. Given the semantically interpreted underlying phrase-marker for a
sentence S in a language L and the definition of a semantic property P or relation
R, we can deduce a prediction about whether S has P or bears R to some other
sentence. This deduction is merely a matter of determining whether or not the
semantically interpreted underlying phrase-marker(s) of S possess the formal features
required by the definition of P or R. Such predictions can be checked against the
ways that ftuent speakers of L sort sentences in terms of their naive linguistic
intuitions. Hence, the justification of these definitions depends on whether such
predictions accord with the judgements of ftuent speakers about c1ear cases from L.
To remove from these definitions the stigma that automatically attaches to defini-
tions of semantic properties and relations since Quine's attack on the analytic-
synthetic distinction, I will show how the above definition of analyticity avoids the
criticism he levelled against Carnap's explication of analyticity.14 This case is chosen
as our example because of its prominence in the literature, but what I am going to
say by way of a defence against Quine's criticism of this concept will apply directly
to similar criticisms of any of the other semantic properties and relations.
One of Quine's major criticisms was that Carnap's explication of analyticity,
contradiction, and related concepts merely defines one of these concepts in terms of
others, whose own definition quickly brings us back to the original one without off er-
ing a genuine analysis of any of them. The above definition of analyticity, however,
cannot be criticized on grounds of such circularity because it is not the case that any
of these related terms, or any others for that matter, were used to define it. The
unique feature of the above definitions is that the defining condition in each is stated
exclusively in terms of a different set of formal features in semantically interpreted
underlying phrase-markers. Moreover, no appeal to such definitions is made in the
process whereby a semantically interpreted underlying phrase-marker obtains the
formal features by virtue of which it satisfies such a defining condition. Further, Quine
criticizes Carnap for merely labeIling sentences as analytic without ever indicating
just what is attributed to a sentence so labelled. On Carnap's account, the term
'analytic' is just an unexplained label. But, on our account, labelling a sentence as
analytic depends on its semantic structure, as determined by its semantically inter-
preted underlying phrase-marker. Thus, labelling a sentence as analytic attributes
to it that semantic structure formalized in the definition that introduces 'analytic'
into linguistic theory. Lastly, the definition of 'analytic', as weil as that of any of the
other semantic properties and relations defined in semantic theory, cannot be
criticized for being too particularistic because, as Quine requires, they are formulated
for variable S and L. This language-independent generaIity is guaranteed by the
14 W. V. Quine, 'Two Dogmas of Empiricism', From a Logica/ Point 01 View (Harvard University Press,
Cambridge, 1953).
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 115
fact that they are given in Jinguistic theory and that their defining conditions are
formulated exclusively in terms of semantically interpreted underlying phrase-
markers, which are associated with each sentence in any linguistic description.
15
8. SEMANTIC CATEGORIES
The last philosophical problem whose solution I here want to treat within linguistic
theory is that of semantic categories, the most general classes into which the con-
cepts from all fields of knowledge divide. The most influential treatment of semantic
categories was certainly Aristotle's. Aristotelian categories claim to be the most
abstract c1assificational divisions under which ideas of any sort can be subsumed.
They are the ultimate, unanalysable, maximally general set of natural kinds that are
given in naturallanguages. Aristotle enumerated ten (perhaps eight) such categories:
substance, quantity, quality, relation, place, time, post ure, possession, action, and
passivity (with the last two of somewhat questionable status). But he did not explain
how he chose these categories nor how he decided that no others belong to the list.
The criterion he mentions falls short of providing a satisfactory principle for
categoryhood.
Aristotle's criterion comes to this: each category is the most general answer to
a question of the form 'What is X T. Thus, substance qualifies as a category because
'a substance' is the most general answer to such questions as 'What is Socrates?'.
Likewise, quality is a category because 'a quality' is the most general answer to such
questions as 'What is green ?'. There is, of course, much room for doubt about these
answers, but, even if this criterion were fairly successful in picking out cases that
we would intuitively regard as among the most abstract classificational divisions in
our conceptual system, it would tell us nothing about the nature of the categories
it sorted out. Its application relies on intuitive judgements about what are and what
are not the most general answers to the test questions, without clarifying either for
ourselves or our informants just what makes these judgements appropriate judge-
ments about the concepts concerned. Consequently, assuming that those things that
are proper answers to such questions are just those things that are the most general
genera for classifying concepts, still, we know no more about the idea of a category
than we did before having obtained its extension in this manner. The criterion itself
presupposes our intuitive understanding of the notion maximal generality in the
domain 0/ possible concepts as a condition for its application, but it does not provide
any analysis of this notion.
If we can embed the theory of semantic categories into linguistic theory, we can
obtain an analysis of this not ion and thereby arrive at a clearer understanding of
l' For a more detailed and complete account of how my explication of analyticity avoids Quine's
criticisms of Camap's explication, see J. J. Katz, 'Some Remarks on Quine on Analyticity', The Journal 0/
Phi/osophy, vol. LXIV (1967), pp. 36--52.
116 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
semantic categories. To do this, we have to ask by what means can we distinguish
two sets of semantic markers: first, the subset of the set of semantic markers appear-
ing in the dictionary of a linguistic description of some particular language whose
members represent concepts having the required degree of generality in that language,
and, second, the subset of the set ofuniversal semantic markers (as given in semantic
theory) whose members represent concepts having the required degree of generality
for language in general. Thus, we must provide some empirically motivated way for
determining the semantic categories of a particular language and the semantic
categories of natural language.
As we have characterized the entries for lexical items in the dictionary of a
semantic component, lexical readings contain a semantic marker for each inde-
pendent conceptual component of the sense that they represent. Formulated in this
way, almost every lexical reading would exhibit a high degree of redundancy in the
manner in which it specifies the semantic information in asense. For exam!1le, the
semantic markers (Physical Object) and (Human) that appear in the lexical readings
for the words 'bachelor', 'man', 'spinster', 'child', etc., are subject to a regularity
governing their occurrence with respect to one another in the dictionary, viz. when-
ever (Human) occurs in a lexical reading, so does (Physical Object). Hence, the
occurrence of (Physical Object) in lexical readings that contain (Human) is actually
redundant, by virtue of the generalization that the occurrence of (Physical Object)
is determined by the occurrence of (Human). However, as we have so far described
the dictionary, we have made no provision for the formulation of such generaliza-
tions. That is, at present, we have no way to express this regularity in the formalism
of semantic theory so that the redundancy of (Physical Object) for these cases is
forced on us in order that we be able to fully represent the senses of these words.
Without a means for expressing such regularities, linguistic descriptions that are
written in accord with semantic theory, as so far formulated, can be correctly
criticized for having missed an important generalization about their languages. With
a means of representing such regularities, the actual occurrence of the semantic
marker (Physical Object) is dispensable because its occurrence in the lexical readings
for 'bachelor', 'man', 'spinster', etc., is predictable from the occurrence of (Human)
in these lexical readings and the generalization that says that (Human) never occurs
in a lexical reading unless (Physical Object) occurs. Moreover, this case is not an
isolated one. Not only is there a broader regularity covering the occurrence of
(Physical Object), viz. whenever (Human), (Anima\), (Artifact), (Plant), etc., occurs
in a lexical reading, so does (Physical Object), but other semantic markers besides
(Physical Object) are redundant in the same way, and are similarly predictable from
generalizations expressing the appropriate regularities, viz. (AnimaI) occurs whenever
(MammaI) does. Hence, from the viewpoint of the whole dictionary with its thousands
of entries, there will be an incredible amount of redundancy in the specification of the
senses of lexical items unless we provide some way to eliminate such unnecessary
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 117
occurrences of semantic markers by finding some formalism to express these lexical
generalizations.
The obvious way to make dictionary entries more economical and to provide a
means of expressing these regularities is to extend the conception of the dictionary
presented in Section 6 so that the dictionary indudes rules wbich state the appro-
priate generalizations and thereby enable us to exdude redundant semantic markers
from lexical readings. In general, these rules will be of the form
where (M J is distinct from each (MX 1 ~ i ~ n and where v is the symbol for dis-
junction. In the case discussed above, we have an example of a rule of tbis type.
(MI) v (M
z
) v ... v (Human) v (Anima!) v (Artifact) v (Plant) v ... v (M.)-+
(Physical Object)
Adding tbis rule to the dictionary enables us to capitalize on the regularity noted
above and economize lexical readings that contain one of the semantic markers
(MI)' (M
z
)' •.. , (Human), (Animal), (Artifact), (Plant) ... , (Mn) by dropping
the occurrence of the semantic marker (Physical Object) from those lexical readings.
Such rules will comprise a new component of the dictionary, whose list of entries
can now contain only lexical readings in maximally reduced form. These rules thus
function to compress the readings in dictionary entries, making the dictionary a more
economical formulation of the lexical information in the language.
So much for the formalism. The redundancy rules not only simplify the statement
of the dictionary and state significant lexical generalizations, they also represent
indusion relations among the concepts represented by semantic markers. For such
rules can be interpreted as saying that the concepts represented by the semantic
markers on the left-hand side of the arrow are induded in, or subsumed under, the
concept represented by the semantic marker on the right-hand side. Hefe, then, is
where the application of tbis formalism to the question of semantic categories comes
in. Using the redundancy rules in the dictionary of a linguistic description for a
language L, we can formally determine wbich of the semantic markers in that
linguistic description represents the semantic categories of L. We dejine a semantic
category 0/ L to be any concept represented by a semantic marker that occurs on
the right-hand side 0/ some rufe in the redundancy rufes in the dictionary 0/ the
linguistic description 0/ L, but does not occur on the feft-hand side 0/ any rufe in
that set 0/ redundancy rufes. Thus, to find the semantic categories of a particular
language, we simply check over the list of redundancy rules in the linguistic descrip-
tion of L and pick each semantic marker for which there is a rule that says that
that marker subsumes other markers, and for wbich there is no rule that says that
that marker is subsumed under other markers. The significance of tbis definition is
118 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
two-fold. First,_ it makes it possible for us to formally determine the semantic
categories for a given language with respect to a linguistic description for it. Second,
it makes it possible to justify empirical claims to the etTect that such-and-such con-
cepts are the semantic categories for a given language. Such justification for a set of
putative semantic categories is a matter of empirically establishing that no simpler
formulation of the lexical readings in the dictionary of the language is provided by
redundancy rules other than those which, by the above definition of semantic
categories of L, yield the set of putative semantic categories in question. This is the
same sort of empirical justification appealed to in other branches of science when it
is claimed that some theoretical ac count is best because it employs the simplest laws
for describing the phenomena under study.
On the basis of these considerations, we can also formally determine the semantic
markers that represent semantic categories of language in general, i.e. the semantic
categories for all natural languages as opposed to the semantic categories of some
particular naturallanguage. We define the semantic categories of language to be those
concepts represented by the semantic markers belonging to the intersection of the
sets of semantic categories for each particular natural language LI' L
2
, ••• , Ln'
as obtainedfrom the redundancy rules in the dictionaries ofthe linguistic descriptions
for LI' L
2
, ••• , Ln in the manner just described. That is, a semantic category
of language is a concept represented by a semantic marker that is found in each and
every set of semantic categories for particular natural languages. The significance
of this definition is parallel to that of t,he previous one. First, it provides us with a
formal means to determine the categories of language, and second, it provides us with
a clear-cut empirical basis for deciding what is a semantic category of language.
The justification for a claim that some concept is a semantic category of language
can be given on the basis of the same evidence that warrants the claims that that
concept is a semantic category of LI' L 2' • • • , Ln'
Notice, finally, that the unexplicated notion of maximal generality on which the
Aristotelian notion of categories is based is here explicated formally, in terms of
membership in the set of semantic markers that comprises the intersection of the sets
of semantic markers that are semantic categories for the natural languages, where
each of the semantic markers in these latter sets is obtained by the condition that
it appears on the right-hand side of a redundancy rule but not on the left-hand side
of any.
9. THE SCOPE OF PHILOSOPHICAL RELEVANCE
If the considerations put forth in this paper on behalf of the thesis that linguistic
theory is relevant to the solution of philosophical problems are convincing, then it
is quite natural to ask to what phIlosophical problems is linguistic theory irrelevant.
That there are problems to which linguistic theory is irrelevant need not bequestioned,
PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 119
for the philosophy of mathematics and the philosophy of science provide abundant
examples. Consequently, it would be highly desirable to have some handy criterion
by which to dt;Cide whether a philosophical problem is essentially about the under-
Iying conceptual structure of natural languages or about something else. But I find it
hard to believe that we can have such a criterion, for not only do philosophical
problems not come to us ear-marked as either linguistic or non-linguistic but, even
given a fully developed linguistic theory, it would require considerable further inquiry
to discover whether some portion of that theory is relevant to some particular
philosophical problem and much further argument to establish that the relevance is
such that linguistic theory provides an authentie solution for it.
We have so far considered only one sense in which linguistic theory can be
relevant to philosophie al questions. In condusion, I would like to consider another
way in which it can have philosophical relevance, one that does not depend on the
theory offering us the concepts that answer the philosophie al question.
The problem of innate ideas, the crux of the controversy between empiricists and
rationalists, is a case to which linguistic theory has a significant application in a
somewhat different way than the one in which it applies to the problems discussed
above.
16
This problem can be recast as the question of whether the acquisition of
a natural language can be explained better on the basis of the empirieist hypothesis
that the mind starts out as a tabula rasa, or on the basis of the rationalist hypothesis
that the mind starts out with a rich stock of innately fixed principles which deter-
mine the general form of the rules for a natural language. Given that the child
obtains his inner representation of the rules of a language from the linguistic data to
which he is exposed during his formative years, we may consider the child's mind
to be a black box whose input is such linguistic data and whose output is an internali-
zation of the linguistic description of the language. Accordingly, we ask whether the
empirieist hypothesis that this internalization is obtained by processing sensory data
on the basis of principles of associative learning, or the rationalist hypothesis that
this internalization is obtained by a specialization of the innate system of principles
when they are activated by appropriate sensory stimulation, is the better account of
how the black box converts its input into its output. We have a fairly dear idea ofthe
associative principles with which the empirieist is willing to credit the child's mind
prior to experience, but it is by no me ans dear what are the innately fixed principles
concerning the general form of language on the rationalist's account. Here the
relevance of linguistic theory is, then, that it provides a statement of the principles
required to formulate the rationalist hypothesis in specific terms. The question is,
therefore, whether we must assume as rich a conception of innate structure as is given
by linguistic theory's account of the universals of language in order to explain
language acquisition.
16 For a more complete discussion of the problem of innate ideas, cf. eh. 5 of my book, The Philosophy
0/ Language.
120 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY
Notice, however, that although this question, which reformulates the problem
of innate ideas, can only be raised in an explicit form when linguistic theory supplies
the conception of innate structure for the rationalist hypothesis, its answer is not
given by linguistic theory. Linguistic theory does not validate the rationalist position
in its controversy with the empiricist position, since it is outside the scope of
linguistic theory to decide which of these two positions is best supported by the facts
about the linguistic information available to the child and about how he copes with
them.
This case was introduced not only to show that the relevance of linguistic theory
to philosophy goes beyond solving philosophical problems, but also to show that it
can be relevant in the specific sense of providing the means by which a philosophical
problem can be reformulated in a manner that makes it more susceptible to solution.
Whether linguistic theory is relevant to philosophical investigation in still other ways
must remain a matter for further philosophical and linguistic inquiry.
VII
SYMPOSIUM ON INNATE IDEAS
(a) RE CENT CONTRIBUTIONS
TO THE THEOR Y OF INNATE IDEAS
Summary 0/ Oral Presentation
NOAM CHOMSKY
I THINK that it will be useful to separate two issues in the discussion of our present
topic---one is the issue of historical interpretation, namely, what in fact was the
content of the classical doctrine of innate ideas, let us say, in Descartes and Leibniz;
the second is the substantive issue, namely, in the light of the information presently
available, what can we say about the prerequisites for the acquisition of knowledge
-what can we postulate regarding the psychologically apriori principles that deter-
mine the character of learning and the nature of what is acquired.
These are independent issues; each is interesting in its own right, and I will have
a few things to say about each. What I would like to suggest is that contemporary
research supports a theory of psychological apriori principles that bears a striking
resemblance to the classical doctrine of innate ideas. The separateness of these issues
must, nevertheless, be kept clearly in mind.
The particular aspect of the substantive issue that I will be concemed with is the
problem of acquisition of language. I think that a consideration of the nature of
linguistic structure can shed some light on certain cIassical questions concerning the
origin of ideas.
To provide a framework for the discussion, let us consider the problem of design-
ing a model of language-acquisition, an abstract 'Ianguage-acquisition device' that
duplicates certain aspects ofthe achievement ofthe human who succeeds in acquiring
linguistic competence. We can take this device to be an input-output system
data -+ I LA I -+ knowledge
To study the substantive issue, we first attempt to determine the nature of the out-
put in many cases, and then to determine the character of the function relating input
From Boston Studies in the Philosophy 01 Science, vol. III (The Humanities Press, New York, 1968),
pp. 81-107. Reprinted by permission of the authors, New York: Humanities Press Inc., and D. Reidel
Publishing Company.
122 SYMPOSIUM ON INNA TE IDEAS
to output. Notice that this is an entirely empirical matter; there is no place for any
dogmatic or arbitrary assumptions about the intrinsie, innate structure of the device
LA. The problem is quite analogous to the problem of studying the innate principles
that make it possible for a bird to acquire the knowledge that expresses itself in
nest-building or in song-production. On apriori grounds, there is no way to determine
the extent to which an instinctual component enters into these acts. To study this
question, we would try to determine from the behaviour of the mature animal just
what is the nature of its competence, and we would then try to construct a second-
order hypothesis as to the innate principles that provide this competence on the
basis of presented data. We might deepen the investigation by manipulating input
conditions, thus extending the information bearing on this input-output relation.
Similarly, in the case of language-acquisition, we can carry out the analogous study
of language-acquisition under a variety of different input conditions, for example,
with data drawn from a variety of languages.
In either case, once we have developed some insight into the nature of the resulting
competence, we can turn to the investigation of the innate mental functions that
provide for the acquisition of this competence. Notice that the conditions of the
problem provide an upper bound and a lower bound on the structure that we may
suppose to be innate to the acquisition device. The upper bound is provided by the
diversity of resulting competence--in our case, the diversity oflanguages. We cannot
impose so much structure on the ~ e v i c e that acquisition of some attested language is
ruled out. Thus we cannot suppose that the specific rules of English are innate to the
device and these alone, since this would be inconsistent with the observation that
Chinese can be learned as readily as English. On the other hand, we must attribute
to the device a sufficiently rich structure so that the output can be attained within the
observed limits of time, data, and access.
To repeat, there is no reason for any dogmatic assumptions about the nature of
LA. The only conditions we must meet in developing such a model of innate mental
capacity are those provided by the diversity of language, and by the necessity to
provide empirically attested competence within the observed empirical conditions.
When we face the problem of developing such a model in a serious way, it be-
comes immediately apparent that it is no easy matter to formulate a hypothesis
about innate structure that is rich enough to meet the condition of empirical
adequacy. The competence of an adult, or even a young child, is such that we must
attribute to him a knowledge of language that extends far beyond anything that he
has leamed. Compared with the number of sentences that a child can produce or
interpret with ease, the number of seconds in a lifetime is ridiculously small. Hence
the data available as input are only aminute sampie ofthe linguistic material that has
been thoroughly mastered, as indicated by actual performance. Furthermore, great
diversity of input conditions does not lead to a wide diversity in resulting com-
petence, so far as we can detect. Furthermore, vast differences in intelligence have
SYMPOSIUM ON INNATE IDEAS 123
only a small effect on resulting competence. We observe further that the tremendous
intellectual accomplishment of language-acquisition is carried out at aperiod of life
when the child is capable of little else, and that this task is entirely beyond the
capacities of an otherwise intelligent ape. Such observations as these lead one to
suspect, from the start, that we are dealing with a species-specific capacity with a
largely innate component. It seems to me that this initial expectation is strongly
supported by a deeper study of linguistic competence. There are several aspects of
normal linguistic competence that are crucial to this discussion.
I. CREATIVE ASPECT OF LANGUAGE USE
By this phrase I refer to the ability to produce and interpret new sentences in inde-
pendence from 'stimulus control'-i.e. external stimuli or independently identifiable
intern al states. The normal use of language is 'creative' in this sense, as was widely
noted in traditional rationalist linguistic theory. The sentences used in everyday
discourse are not 'familiar sentences' or 'generalizations of familiar sentences' in
terms of any known process of generalization. In fact, even to speak of 'familiar
sentences' is an absurdity. The idea that sentences or sentence-forms are learned by
association or conditioning or 'training', as proposed in recent behaviourist specula-
tions, is entirely at variance with obvious fact. More generally, it is important to
realize that in no technical sense of these words can language use be regarded as
a matter of 'habit' or can language be regarded as 'a complex of dispositions to
respond'.
A person's competence can be represented by a grammar, which is a system of
rules for pairing semantic and phonetic interpretations. Evidently, these rules operate
over an infinite range. Once a person has mastered the rules (unconsciously, of
course), he is capable, in principle, of using them to assign semantic interpretations
to signals quite independently of whether he has been exposed to them or their parts,
as long as they consist of elementary units that he knows and are composed by the
rules he has intemalized. The central problem in designing a language acquisition
device is to show how such a system of rules can emerge, given the data to which the
child is exposed. In order to gain some insight into thi's question, one naturally turns
to a deeper investigation of the nature of grammars. I think real progress has been
made in recent years in our understanding of the nature of grammatical rules and
the manner in which they function to assign semantic interpretations to phonetically
represented signals, and that it is precisely in this area that one can find results that
have some bearing on the nature of a language-acquisition device.
124 SYMPOSIUM ON INNATE IDEAS
11. ABSTRACTNESS OF PRINCIPLES OF SENTENCE INTERPRETATiON
A grammar consists of syntactic rules that generate certain underlying abstract
objects, and rules of semantic and phonological interpretation that assign an intrinsic
meaning and an ideal phonetic representation to these abstract objects.
Concretely, consider the sentence 'The doctor examined John'. The phonetic form
of this sentence depends on the intrinsic phonological character of its minimal items
('The', 'doctor', 'examine', 'past tense', 'John'), the bracketing of the sentence (that
is, as [[[thel [doctorll [[examinedl [Johnll]), and the categories to which the
bracketed elements belong (that is, the categories 'Sentence', 'Noun-Phrase', 'Verb',
'Noun', 'Deterrniner', in this case). We can define the 'surface structure' of an utter-
ance as its labelled bracketing, where the brackets are assigned appropriate cate-
gorial labels from a fixed, universal set. It is transparent that grammatical relations
(e.g. 'Subject-of, Object-of', etc.) can be defined in terms of such a labelIed bracket-
ing. With terms defined in this way, we can assert that there is very strong evidence
that the phonetic form of a sentence is determined by its labelIed bracketing by
phonological rules that operate in accordance with certain very abstract but quite
universal principles of ordering and organization.
The meaning of the sentence 'the doctor exarnined John' is, evidently, determined
from the meanings of its minimal items by certain general rules that make use of the
grammatical relations expressed by the labelIed bracketing. Let us define the 'deep
structure' of a sentence to be that labelIed bracketing that determines its intrinsic
meaning, by application of these rules of semantic interpretation. In the example just
given, we would not be far wrong ifwe took the deep structure to be identical with the
surface structure. But it is obvious that these cannot in general be identified. Thus
consider the slightly more complex sentences: 'John was examined by the doctor';
'someone persuaded the doctor to examine John'; 'the doctor was persuaded to
examine John'; 'John was persuaded to be exarnined by the doctor'. Evidently, the
grammatical relations among doctor, examine, and John, as expressed by the deep
structure, must be the same in all of these examples as the relations in 'the doctor
examined John'. But the surface structures will differ greatly.
Furthermore, consider the two sentences :
someone expected the doctor to examine J ohn
someone persuaded the doctor to examine John.
It is c1ear, in this case, that the similarity of surface structure masks a significant
difference in deep structure, as we can see, immediately, by replacing 'the doctor
to exarnine John' by 'John to be exarnined by the doctor' in the two cases.
So far, I have only made a negative point, namely, that deep structure is distinct
from surface structure. Much more important is the fact that there is very strong
evidence for a particular solution to the problem of how deep and surface structures
SYMPOSIUM ON INNATE IDEAS 125
are related, and how deep and surface structures are formed by the syntactic com-
ponent of the grammar. The details ofthis theory need not concem us for the present.
A crucial feature of it, and one which seems inescapable, is that it involves formal
manipulations of structures that are highly abstract, in the sense that their relation
to signals is defined by a long sequence of formal rules, and that, consequently, they
have nothing remotely like a point by point correspondence to signals. Thus sentences
may have very similar underlying structures despite the great diversity of physical
form, and diverse underlying structures despite similarity of surface form. A theory
of language acquisition must explain how tbis knowledge of abstract underlying forms
and the principles that manipulate them comes to be acquired and freely used.
111. UNIVERSAL CHARACTER OF LlNGUISTlC STRUCTURE
So far as evidence is available, it seems that very heavy conditions on the form of
grammar are universal. Deep structures seem to be very similar from language to
language, and the rules that manipulate and interpret them also seem to be drawn
from a very narrow class of conceivable formal operations. There is no apriori
necessity for a language to be organized in this highly specific and most peculiar
way. There is no sense of 'simplicity' in which this design for language can be
intelligibly described as 'most simple'. Nor is there any content to the claim that tbis
design is somehow 'logical'. Furthermore, it would be quite impossible to argue that
this structure is simply an accidental consequence of 'common descent'. Quite apart
from questions of historical accuracy, it is enough to point out that this structure
must be rediscovered by each child who leams the language. The problem is,
precisely, to determine how the child determines that the structure of his language
has the specific characteristics that empirical investigation of language leads us to
postulate, given the meagre evidence available to hirn. Notice, incidentally, that the
evidence is not only meagre in scope, but very degenerate in quality. Thus the child
learns the principles of sentence formation and sentence interpretation on the basis
of a corpus of data that consists, in large measure, of sentences that deviate in form
from the idealized structures defined by the gramrnar that he develops.
Let us now return to the problem of designing a language-acquisition device. The
available evidence shows that the output of tbis device is a system of recursive rules
that provide the basis for the creative aspect of language use and that manipulate
highly abstract structures. Furthermore, the underlying abstract structures and the
rules that apply to them have highly restricted properties that seem to be uniform
over languages and over different individuals speaking the same language, and that
seem to be largely invariant with respect to intelligence and specific experience. An
engineer faced with the problem of designing a device meeting the given input-output
conditions would naturally conclude that the basic properties of the output are a
126 SYMPOSIUM ON INNATE IDEAS
consequence of the design of the device. Nor is there any plausible alternative to
this assumption, so far as I can see. More specifically, we are led by such evidence
as I have mentioned to suppose that this device in some manner incorporates: a
phonetic theory that defines the dass of possible phonetic representations ; a semantic
theory that defines the dass of possible semantic representations; a schema that
defines the dass of possible grammars ; a general method far interpreting grammars
that assigns a semantic and phonetic interpretation to each sentence, given a
grarnrnar; a method of evaluation that assigns some measure of 'complexity' to
grammars.
Given such a specification, the device might proceed to acquire knowledge of a
language in the following way: the given schema for grammar specifies the dass
of possible hypotheses ; the method of interpretation perrnits each hypothesis to be
tested against the input data; the evaluation measure selects the highest valued
grammar compatible with the data. Once a hypothesis-a particular grammar-is
selected, the leamer knows the language defined by this grammar; in particular, he
is capable of pairing semantic and phonetic interpretations over an indefinite range
of sentences to which he has never been exposed. Thus his knowledge extends far
beyond his experience and is not a 'generalization' from his experience in any
significant sense of 'generalization' (except, trivially, the sense defined by the intrinsic
structure of the language-acquisition device).
Proceeding in this way, one can seek a hypothesis conceming language-acquisition
that falls between the upper and lower bounds, discussed above, that are set by the
nature of the problem. Evidently, for language leaming to take place the dasS' of
possible hypotheses-the schema for grarnrnar-must be heavily restricted.
This account is schematic an4 idealized. We can give it content by specifying the
language-acquisition system along the lines just outlined. I think that very plausible
and concrete specifications can be given, along these lines, but this is not the place
to pursue this matter, which has been elaborately discussed in many publications on
transforrnational generative grammar.
I have so far been discussing only the substantive issue of the prerequisites for
acquisition of knowledge of language, the apriori principles that determine how and
in what form such knowledge is acquired. Let me now try to place this discussion in
its historical context.
First, I mentioned three crucial aspects of linguistic competence: (1) creative
aspect of language use; (2) abstract nature of deep structure; (3) apparent universality
of the extremely special system of mechanisms formalized now as transforrnational
grammar. It is interesting to observe that these three aspects of language are dis-
cussed in the rationalist philosophy of the 17th century and its aftermath, and that the
linguistic theories that were developed within the framewark of this discussion are,
in essence, theories of transformational grammar.
Consequently, it would be historically accurate to describe the views regarding
SYMPOSIUM ON INNATE IDEAS 127
language structure just outlined as a rationalist conception of the nature of language.
Furthermore, I employed it, again, in the classical fashion, to support what might
fairly be called a rationalist conception of acquisition of knowledge, if we take the
essence of this view to be that the general character of knowledge, the categories in
which it is expressed or internally represented, and the basic principles that underlie
it, are determined by the nature of the mind. In our case, the schematism assigned
as an innate property to the language-acquisition device determines the form of
knowledge (in one of the many traditional senses of 'form'). The role of experience
is only to cause the innate schematism to be activated, and then to be ditTerentiated
and specified in a particular mann er.
In sharp contrast to the rationalist view, we have the classical empiricist assump-
tion that what is innate is (1) certain elementary mechanisms of peripheral processing
(a receptor system), and (2) certain analytical mechanisms or inductive principles
or mechanisms of association. What is assumed is that a preliminary analysis of
experience is provided by the peripheral processing mechanisms and that one's
concepts and knowledge, beyond this, are acquired by application of the innate
inductive principles to this initially analysed experience. Thus only the procedures
and mechanisms for acquisition of knowledge consititute an innate property. In the
case of language-acquisition, there has been much empiricist speculation about what
these mechanisms may be, but the only relatively clear attempt to work out some
specific ac count of them is in modern structural linguistics, which has attempted to
elaborate a system of inductive analytic procedures of segmentation and classification
that can be applied to data to determine a grammar. It is conceivable that these
methods might be somehow refined to the point where they can provide the surface
structures of many utterances. It is quite inconceivable that they can be developed
to the point where they can provide deep structures or the abstract principles that
generate deep structures and relate them to surface structures. This is not a matter of
further refinement, but of an entirely ditTerent approach to the question. Similarly,
it is difficult to imagine how the vague suggestions about conditioning and associa-
tive nets that one finds in philosophical and psychological speculations of an
empiricist cast might be refined or elaborated so as to provide for attested com-
petence. A system of rules for gene rating deep structures and relating them to surface
structures, in the manner characteristic of naturallanguage, simply does not have the
properties of an associative net or ahabit family; hence no elaboration of principles
for developing such structures can be appropriate to the problem of designing a
language-acquisition device.
I have said nothing explicit so far about the doctrine that there are innate ideas
and innate principles of various kinds that determine the character of what can be
known in what may be a rather restricted and highly organized way. In the traditional
view a condition for these innate mechanisms to become activated is that appropriate
stimulation must be presented. This stimulation provides the occasion for the mind to
128 SYMPOSIUM ON INNATE IDEAS
apply certain innate interpretive principles, certain concepts that proceed from 'the
power of understanding' itself, from the faculty of thinking rather than from external
objects. To take a typical example from Descartes (Reply to Objections, V):
' ... When first in infancy we see a triangular figure depicted on paper, this figure
cannot show us how areal tri angle ought to be conceived, in the way in which geo-
metricians consider it, because the true triangle is contained in this figure, just as the
statue of Mercury is contained in a rough block of wood. But because we already
possess within us the idea of a true triangle, and it can be more easily conceived by
our mind than the more complex figure ofthe tri angle drawn on paper, we, therefore,
when we see the composite figure, apprehend not it itself, but rather the authentic
triangle' (Haldane and Ross, vol. 11, p. 227). In this sense, the idea of triangle is
innate. For Leibniz what is innate is certain principles (in general, unconscious), that
'enter into our thoughts, of which they form the soul and the connection'. 'Ideas and
truths are for us innate as inclinations, dispositions, habits, or natural potentialities.'
Experience serves to elicit, not to form, these innate structures. Similar views are
elaborated at length in rationalist speculative psychology.
It seems to me that the conclusions regarding the nature of language-acquisition,
discussed above, are fully in accord with the doctrine of innate ideas, so understood,
and can be regarded as providing a kind of substantiation and further development of
this doctrine. Of course, such a proposal raises non-trivial questions of historical
interpretation.
What does seem to me fairly clear is that the present situation with regard to the
study of language learning, and other aspects of human intellectual achievement of
comparable intricacy, is essentially this. We have a certain amount of evidence about
the gramm ars that must be the output of an acquisition model. This evidence shows
clearly that knowledge of language cannot arise by application of step-by-step in-
ductive operations (segmentation, classification, substitution procedures, 'analogy',
association, conditioning, and so on) of any sort that have been developed or dis-
cussed within linguistics, psychology, or philosophy. Further empirieist speculations
contribute nothing that even faintly suggests a way of overcoming the intrinsic
limitations of the methods that have so far been proposed and elaborated. Further-
more, there are no other grounds for pursuing these empiricist speculations, and
avoiding what would be the normal assumption, unprejudiced by doctrine, that one
would formulate if confronted with empirical evidence of the sort sketched above.
There is, in particular, nothing known in psychology or physiology that suggests
that the empiricist approach is weil motivated, or that gives any grounds for
scepticism concerning the rationalist alternative sketched above.
For further discussion of the question of historical interpretation, see Chomsky,
Aspects of the Theory of Syntax (1965), Ch. 1, and Cartesian Linguistics (1966).
For further discussion of matters touched on here, see also Chomsky, 'Explanatory
Models in Linguistics', in Logic, Methodology and Philosophy o[ Science, ed. by
SYMPOSIUM ON INNATE IDEAS 129
E. Nagel, P. Suppes, and A. Tarski (1962); J. Katz, The Philosophy 0/ Language
(1966); P. M. Postal, Review of A. Martinet, Elements 0/ General Linguistics
(1966); and the selections in section VI of The Structure 0/ Language, Readings in
the Philosophy 0/ Language, ed. by J. Fodor and J. Katz (1964).
(b) THE 'INNATENESS HYPOTHESIS' AND
EXPLANATORY MODELS IN LINGUISTICS
HILARY PUTNAM
I. THE INNATENESS HYPOTHESIS
THE 'innateness hypothesis' (henceforth, the 'I.H.') is a daring-or apparently daring;
it may be meaningless, in which case it is not daring-hypothesis proposed by Noam
Chomsky. I owe a debt of gratitude to Chomsky for having repeatedly exposed me
to the I.H.; I have relied heavily in what follows on oral communications from hirn;
and I beg his pardon in advance if I misstate the I.H. in any detail, or misrepresent
any of the arguments for it. In addition to relying upon oral communications from
Chomsky, I have also relied upon Chomsky's paper 'Explanatory Models in
Linguistics', in which the I.H. plays a considerable role.
To beg in, then, the I.H. is the hypothesis that the human brain is 'programmed'
at birth in some quite specijic and structured aspects of human natural language.
The details of this programming are spelled out in some detail in 'Explanaiory Models
in Linguistics'. We should assurne that the speaker has 'built in' 1 a function which
assigns weights to the grammars GI' G 2' G J' ••• in a certain dass 1.: of transforma-
tional grammars. 2,' is not dass of all possible transformation al grammars ; rather
all the members of 1: have some quite strong similarities. These similarities appear
as 'Iinguistic universals' --i.e. as characteristics of all human natural languages. If
intelligent non-terrestrial life-say, Martians--exists, and if the 'Martians' speak a
language whose grammar does not belong to the subdass 1: of the dass of all trans-
formational grammars, then, I have heard Chomsky maintain, humans (except
possibly for a few geniuses or linguistic experts) would be unable to leam Martian;
a human child brought up by Martians would fail to acquire language; and Martians
would, conversely, experience similar difficulties with human tongues. (Possible
I What 'built in' means is highly unc1ear in this context. The weighting function by itself determines only
the relative ease with which various grammars can be leamed by a human being. If a grammar GI can be
leamed more easily than a grammar G , ~ , then doubtless this is 'innate' in the sense of being a fact about
human learning potential, as opposed to a fact about what has been leamed. But this sort of fact is what
leaming theory tries to account for; not the explanation being sought. It should be noticed that Chomsky
has never offered even a schematic account of the sort of device that is supposed to be present in the brain,
and that is supposed to do the job of selecting the highest weighted grarnmar compatible with the data. But
only a description, or at least a theory, of such a device could properly be caIled an innateness hypothesis
at all.
SYMPOSIUM ON INNATE IDEAS 131
difficulties in pronunciation are not at issue here, and may be assumed not to exist
for the purposes of this argument.) As examples ofthe similarities that all grammars
of the subdass L are thought to possess (above the level of phonetics), we may
mention the active-passive distinction, the existence of a non-phrase-structure portion
of the grammar, the presence of such major categories as concrete noun, verb taking
an abstract subject, etc. The project of delimiting the dass L may also be described
as the project of defining a nonnal form for grammars. Conversely, according to
Chomsky, any non-trivial normal form for grammars, such that correct and per-
spicuous grammars of all human languages can and should be written in that normal
form, 'constitutes, in effect, a hypothesis concerning the innate intellectual equip-
ment of the child'.
2
Given such a highly restricted dass L of grammars (highly restricted in the sense
that grammars not in the dass are perfectly conceivable, not more 'complicated' in
any absolute sense than grammars in the dass, and may weH be employed by non-
human speakers, if such there be), the performance of the human child in learning
his native language may be understood as folIows, according to Chomsky. He may
be thought of as operating on the following 'inputs' 3: a list of utterances, containing
both grammatical and ungrammatical sentences ; a list of corrections, which enable
him to dassify the input utterances as grammatical or ungrammatical; and some
information concerning which utterances count as repetitions of earlier utterances.
Simplifying slightly, we may say that, on this model, the child is supplied with a
list of grammatical sentence types and a list of ungrammatical sentence types. He
then 'selects' the grammar in L compatible with this information to which his
weighting function assigns the highest weight. On this scheme, the general form of
grammar is not learned from experience, but is 'innate', and the 'plausibility
ordering' of grammars compatible with given data of the kinds mentioned is likewise
'innate'.
So muchfor a statement of the l.H. lf I have left the IR. vague at many points,
I believe that this is no accident-for the I.H. seems to me to be essentially and
irreparably vague-but this much of a statement may serve to indicate what belief
it is that I stigmatize as irreparably vague.
A couple of remarks may suffice to give some idea ofthe role that I.H. is supposed
to play in linguistics. Linguistics relies heavily, according to Chomsky, upon 'intui-
tions' of grammaticality. But what is an intuition of 'grarnmaticality' an intuition
of? According to Chomsky, the sort of theory-construction programmatically out-
lined above is what is needed to give this question the only answer it can have or
deserves to have. Presumably, then, to 'intuit' (or assert, or conjecture, etc.) that a
sentence is grammatical is to 'intuit' (or assert, or conjecture, etc.) that the sentence
is generated by the highest-valued Gi in the dass L which is such that it generates
2 'Explanatory Models in Linguistics', p. 550.
3 'E. M. in L.', pp. 530-531.
132 SYMPOSIUM ON INNATE IDEAS
a1l the grammatical sentence types with which we have been supplied by the 'input'
and none of the ungrammatical sentence types listed in the 'input'.4
Chomsky also says that the Gi which receives the highest value must do more than
agree with 'intuitions' of grammaticality; it must account for certain ambiguities, for
example.
5
At the same time, unfortunateIy, he Iists no semantical information in the
input, and he conjectures 6 that a child needs semantical information only to 'provide
motivation for language learning', and not to arrive at theformal grammar of its
language. Apparently, then, the fact that a grarnmar which agrees with a sufficient
amount of 'input' must be in the cIass .E to be 'seIected' by the child is what mIes out
grammars that generate all and only the grammatical sentences of a given natural
language, but fail to correctly 'predict'·7 ambiguities (cf. 'E. M. in L.', p. 533).
In addition to making cIear what it is to be grammatical, Chomsky believes that
the I.H. confronts the linguist with the following tasks: To define the normal form for
grammars described above, and to define the weighting function. In Syntactic
Structures Chomsky, indeed, gives this as an objective for linguistic theory: to give
an effective procedure for choosing between rival grammars.
Lastly, the I.H. is supposed to justify the claim that what the linguist provides is
'a hypothesis about the innate intellectual equipment that a child brings to bear in
language learning'.8 Of course, even if language is wholly learned, it is still true that
linguistics 'characterizes the linguistic abilities of the mature speaker',
9
and that a
grammar 'could properly be called an explanatory model of the linguistic intuition of
the native speaker'.lo However, one could with equal truth say that a driver's manual
'characterizes the car-driving abilities of the mature driver' and that a calculus text
provides 'an explanatory model of the calculus-intuitions of the mathematician'.
Clearly, it is the idea that these abilities and these intuitions are close to the human
essen ce, so to speak, that gives linguistics its 'sex appeal', for Chomsky at least.
4 I doubt that the c1uld really is told which sentences it hears or utters are ungrammatical. At most it is told
which are del'iant-but it may not be told which are deviant for syntactical and which for semantical
reasons.
'Many of these---e.g. the aUeged 'ambiguity' in 'the shooting of the elephants was heard'-require
coaching to detect. The claim that grammar 'explains the ability to recognize ambiguities' thus lacks the
impressiveness that Chomsky believes it to have. I am grateful to Paul Ziff and Stephen Leeds for calling
this point to my attention.
• 'E. M. in L', p. 531, n. 5.
7 A grammar 'predicts' an ambiguity, in Chomsky's formalism, whenever it assigns two or more structural
descriptions to the same sentencc.
"E. M. in L', p. 530.
"E. M. in L', p. 530.
10 'E. M. in L', p. 533.
SYMPOSIUM ON INNATE IDEAS 133
11. THE SUPPOSED EVIDENCE FOR THE I.H.
A number of empirical facts and alleged empirical facts have heen advanced to
support the I.H. Since limitations of space make it impossible to describe all of them
here, a few examples will have to suffice.
(a) the ease of the child's originallanguage learning. 'A young child is able to gain
perfect mastery of a language with incomparably greater ease [than an adult-Ho P.l
and without any explicit instruction. Mere exposure to the language, and for a
remarkably short period, seems to be all that the normal child requires to develop
the competence of the native speaker.' 11
(b) The fact that reinforcement, 'in any interesting sense', seems to he unnecessary
for language learning. Some children have apparently even learned to speak without
talking, 12 and then displayed this ability at a relatively late age to startled adults
who had given them up for mutes.
(c) Tbe ability to 'develop the competence ofthe native speaker' has been said not
to depend on the intelligence level. Even quite low I.Q.'s 'internalize' the grammar of
their native language.
(d) Tbe 'Iinguistic universals' mentioned in the previous section are allegedly
accounted for by the I.H.
(e) Lastly, of course, there is the 'argument' that runs 'what else could account for
language learning?' The task is so incredibly complex (analogous to learning, at least
implicitly, a complicated physical theory, it is said), that it would he miraculous if
even one tenth of the human race accomplished it without 'innate' assistance. (This
is like Marx's 'proof of the Labour Tbeory of Value in Capital, vol. IH, which runs,
in essence, 'What else could account for the fact that commodities have different
value except the fact that the labour-content is different?')
III. CRITICISM OF THE ALLEGED EVIDENCE
A. The Irrelevance of Linguistic Universals
1. Not surprising on any theory
Let us consider just how surprising the 'Iinguistic universals' cited above really are.
Let us assume for the purpose a community of Martians whose 'innate intellectual
equipment' may he supposed to he as different from the human as is compatible with
their heing able to speak a language at all. What could we expect to find in their
language?
If the Martians' brains are not vastly richer than ours in complexity, then they,
11 'E. M. in L.', p. 529.
12 Macaulay's jirst words, it is said, were: 'Thank you, Madam, the agony has somewhat abated' (to a
lady who had spilled hot tea on him).
134 SYMPOSIUM ON IN NA TE IDEAS
like us, will find it possible to employ a practically infinite set of expressions only if
those expressions possess a 'grammar' ---i.e. if they are built up by recursive rules
from a limited stock of basic forms. Those basic forms need not be built up out of a
short list of phonemes---the Martians might have vastly greater memory capacity
than we do--but if Martians, like humans, find rote learning difficult, it will not be
surprising if they too have short lists of phonemes in their languages.
Are the foregoing reflections arguments for or against the LH. ? I find it difficult
to tell. If belief in 'innate intellectual equipment' is just that, then how could the I.H.
be false? How could something with no innate intellectual equipment leam anything?
To be sure, human 'innate intellectual equipment' is relevant to language learning;
if this means that such parameters as memory span and memory capacity playa
crucial role. But what rank Behaviourist is supposed to have ever denied t his? On the
other hand, that a particular mighty arbitrary set L of grammars is 'built in' to the
brain of both Martians and Humans is not a hypothesis we would have to invoke to
account for these basic similarities.
But for wh at similarities above the level of phonetics, where constitutional factors
playa large role for obvious reasons, would the LH. have to be invoked save in the
trivial sense that memory capacity, intelligence, needs, interests, etc., are all relevant
to language learning, and all depend in part, on the biological make-up of the
organism? If Martians are such strange creatures that they have no interest in
physical objects, for example, their language will contain no concrete nouns; but
would not this be more, not less surprising, on any reasonable view, than their having
an interest in physical objects? (Would it be surprising if Martian contained devices
for forming truth-functions and for quantification?)
Two more detailed points are relevant here. Chomsky has pointed out that no
natural language has a phrase-structure grammar. But trus too is not surprising. The
sentence 'John and Jim came horne quickly' is not generated by a phrase-structure
rule, in Chomsky's formalization of English gramm3.f. But the sente'nce 'John came
horne quickly and Jim came horne quickly' is generated by a phrase-structure rule in
the grammar of mathematical logic, and Chomsky's famous 'and-transformation'
is just an abbreviation rule. Again, the sentence 'That was the lady I saw you witli last
night' is not generated by a phrase-structure rule in English, or at least not in
Chomsky's description of English. But the sentence 'That is IX (X is a lady and I saw
you with X last night)' is generated by a phrase-structure rule in the grammar of
mathematical logic. And again the idiomatic English sentence can be obtained from
its phrase-structure counterpart by a simple rule of abbreviation. Is it really sur-
prising, does it really point to anything more interesting than general intelligence,
that these operations which break the bounds of phrase-structure grammar appear in
every natural language? 13
13 Another example of a transformation is the 'active-passive' transformation (cf. S)7ltactic Structures).
But (a) the presence of this, if it is apart of the grammar, is not surprising-why shouId not there be a
SYMPOSIUM ON INNATE IDEAS 135
Again, it may appear startling at first blush that such categories as noun, verb,
adverb, etc., have 'universal' application. But, as Curry has pointed out, it is too easy
to multiply 'facts' here. If a language contains nouns-that is, a phrase-structure
category which contains the proper names-it contains noun phrases, that is, phrases
which occupy the environments of nouns. If it contains noun phrases it contains
verb phrases--phrases which when combined with a noun phrase by a suitable con-
struction yield sentences. If it contains verb phrases, it contains adverb p h r a s e ~ ­
phrases which when combined with a verb phrase yield a verb phrase. Similarly,
adjective phrases, etc., can be defined in terms ofthe two basic categories 'noun' and
'sentence'. Thus the existence of nouns is all that has to be explained. And this
reduces to explaining two facts: (I) The fact that all naturallanguages have a large
phrase-structure portion in their grammar, in the sense just illustrated, in spite of the
effect ofwhat Chomsky calls 'transformations'. (2) The fact that all naturallanguages
contain proper names. But (1) is not surprising in view of the fact that phrase-
structure rules are extremely simple algorithms. Perhaps Chomsky would reply that
'simplicity' is subjective here, but this is just not so. The fact is that all the natural
measures of complexity of an algorithm---size of the machine table, length of com-
putations, time, and space required for the computation--lead to the same result
here, quite independently of the detailed structure of the computing machine
employed. Is it surprising that algorithms which are 'simplest' for virtually any
computing system we can conceive of are also simplest for naturally evolved 'com-
puting systems'? And (2)---the fact that all natural languages contain proper names
-----is not surprising in view of the utility of such names, and the difficulty of always
finding adefinite description which will suffice instead.
Once again, 'innate' factors are relevant to be sure-----if choosing simple algorithms
as the basis of the grammar is 'innate', and ifthe need for identifying persons rests on
something innate-but what Behaviourist would or should be surprised? Human
brains are computing systems and subject to some of the constraints that atTect all
computing systems; human beings have a natural interest in one another. If that is
'innateness', weIl and good!
systematic way of expressing the conrerse of a relation ?-and (b) the argument for the existence of such
a 'transformation' at all is extremely slim. It is contended that a gramm ar which 'defines' active and passive
forms separately (this can be done by even a phrase-structure grammar) fails to represent something that
every spcaker knows, viz_ that active and passive forms are related. But why must evcry relation be
mirrored by syntax? Every 'speaker' of the canonical languages of mathematical logic is aware that each
sentence (X) (FX::O GX) is related to a sentence (X) (o'r::o FX); yet the definition of'well formed formula'
fails to mirror 'what every speaker knows' in this respect, and is not inadequate on that account.
136 SYMPOSIUM ON INNATE IDEAS
2. Linguistic uni versals could be accounted for, even if surprising, without
invoking the /.H.
Suppose that language-using human beings evolved independently in two or more
places. Then, if Chomsky were right, there should be two or more types of human
beings descended from the two or more original populations, and normal children of
each type should fail to learn the languages spoken by the other types. Since we do
not observe this, since there is only one c1ass I: built into all human brains, we have
to conc1ude (if the I.H. is true) that language-using is an evolutionary 'leap' that
occurred only once. But in that case, it is overwhelmingly likely that all human
languages are descended from a single originallanguage, and that the existence today
of what are called 'unrelated' languages is accounted far by the great lapse of time
and by countless historical changes. This is, indeed, likely even if the I.H. is false,
since the human race itse1f is now generally believed to have resulted from a single
evolutionary 'leap', and since the human population was extremely small and con-
centrated for millennia, and only gradually spread from Asia to other continents.
Thus, even if language-using was learned or invented rather than 'built in', or even if
only some general dispositions in the direction of language using are 'built in', 14 it
is likely that some one group of humans first developed language as we know it, and
then spread this through conquest or imitation to the rest of the human population.
lndeed, we do know that this is just how alphabetic writing spread. In any case, I
repeat, this hypothesis--a single origin for human language---:is certainly required
by the I.H., but much weaker than the I.H.
But just this consequence of the I.H. is, in fact, enough to account for 'linguistic
universals' ! For, if all human languages are descended from a common parent, then
just such highly useful features of the common parent as the presence of some kind
of quantifiers, proper names, nouns, and verbs, etc., would be expected to survive.
Random variation may, indeed, alter many things; but that it should fail to strip
language of proper names, or common nouns, or quantifiers, is not so surprising as
to require the I.H.
B. The 'Ease' of Langua/?e Learning is not Clear
Let us consider somewhat c10sely the 'ease' with which children do learn their native
language. A typical 'mature' college student seriously studying a foreign language
spends three hours a week in lectures. In fourteen weeks of term he is thus exposed
to forty-two hours of the language. In four years he may pick up over 300 hours of
the language, very little of which is actuallistening to native informants. By contrast,
direct-method teachers estimate that 300 hours of direct-method teaching will enable
14 It is very difficult to account for such phenomena as the sponlaneous babbling of infants without Ihis
much 'innateness'. But this is not 10 say that a dass I and a funclionfare 'built in', as required by the I.H.
SYMPOSIUM ON INNATE IDEAS 137
one to converse fiuently in a foreign language. Certainly 600 hours---say, 300 hours
of direct-method teaching and 300 hours of reading-will enable any adult to speak
and read a foreign language with ease, and to use an incomparably larger vocabulary
than a young child.
It will be objected that the adult does not acquire a perfect accent. So what? The
adult has been speaking one way all of his life, and has a huge set of habits to unleam.
What can equally weil be accounted for by learning theory should not be cited as
evidence for the I.H.
Now the child by the time it is four or five years old has been exposed to I'astly
more than 600 hours of direct rnethod instruction. Moreover even if 'reinforcernent'
is not necessary, most children are consciously and repeatedly reinforced by adults
in a host of ways--e.g. the constant repetition of simple one-ward sentences ('cup',
'doggie') in the presence of babies. Indeed, any foreign adult living with the child
for those years would have an incomparably better grasp of the language than the
child does. The child indeed has a better accent. Also, the child's gramrnatical
mistakes, which are numerous, arise not from carrying over previous language habits,
but from not having fully acquired the first set. But it seems to rne that this 'evidence'
for the I.H. stands the facts on their head.
C. Reinforcement another Issue
As Chornsky is aware, the evidence is today slirn that any learning requires reinforce-
ment 'in any interesting sense'. Capablanca, for example, learned to play chess by
sirnply watching adults play. This is comparable to Macaulay's achievement in leam-
ing language without speaking. N on-geniuses normally do require practice both to
speak correctly and to play chess. Yet probably anyone could leam to speak or to
play chess without practice if rnuffied, in the first case, or not allowed to play, in the
second case, with sufficiently prolonged observation.
D. Independence of Intelligence Level an Artifact
Every child leams to speak the native language. What does this mean? If it means
that children do not rnake serious grarnmatical blunders, even by the standards of
descriptive as opposed to prescriptive grammar, this is just not true for the young
child. By nine or ten years of age this has ceased to happen, perhaps (I speak as a
parent), but nine or ten years is enough time to becorne pretty darn good at anything.
What is more serious is what 'grammar' means here. It does not include mastery of
vocabulary, in which even many adults are deficient, nor ability to understand com-
plex constructions, in which rnany adults are also deficient. It means pure1y and
simply the ability to leam what every normal adult leams. Every normal adult learns
what every adult leams. What this 'argument' reduces to is 'W ow! How cornplicated
138 SYMPOSIUM ON INNATE IDEAS
a skill every normal adult learns. What else could it be but innate.' Like the preceding
argument, it reduces to the 'What Else' argument.
But what of the 'What Else?' argument? Just how impressed should we be by the
failure of current learning theories to account for complex learning processes such as
those involved in the learning of language? If Innateness were a general solution,
perhaps we should be impressed. But the I.H. cannot, by its very nature, be generaI-
ized to handle aIl complex learning processes. Consider the following puzzle (calIed
'jump'):
* * *
* * *
* * * * * * *
* * *
* * *
To begin with, aIl the holes but the centre one are filled. The object of the game
is to remove all the pegs but one by 'jumping' (as in checkers) and to end with the
one remaining peg in the centre. A clever person can get the solution in perhaps eight
or ten hours of experimentation. A not so clever person can get a 'near-solution'-
two pegs Ieft--in the same time. No programme exists, to my knowledge, that would
enable a computer to solve even the 'near solution' problem without running out of
both time and space, even though the machine can spend the equivaIent of many
human lifetimes in experiment at ion. When we come to the discovery of even the
simplest mathematical theorem the situation is even more striking. The theorems of
mathematics, the solutions to puzzles, etc., cannot on any theory be indil'idually
'innate'; what must be 'innate' are heuristics, i.e. learning strategies. In the absence
of any knowledge of what general multipurpose learning strategies might even look
like, the assertion that such strategies (which absolutely must exist and be employed
by all humans) cannot account for this or that leaming process, that the answer or
an answer schema must be 'innate', is utterly unfounded.
I will be told, of course, that el'eryone learns his native language (as weIl as every-
one does), and that not everyone solves puzzles or proves theorems. But everyone
does learn pattern recognition, automobile driving, etc., and everyone in fact can
solve many problems that no computer can solve. In conversation Chomsky has
repeatedly used precisely such skiIls as these to support the idea that humans have
an 'innate conceptuaI space'. WeIl and good, if true. But that is no help. Let a
complete seventeenth-century Oxford Unil'ersity education be innate if you like; still
the solution to 'jurnp' was not innate; the Prime Number Theorem was not innate;
SYMPOSIUM ON INNATE IDEAS 139
and so on. Invoking 'Innateness' only postpones the problem o/leaming; it does not
solve it. Until we understand the strategies which make general learning possible-
and vague talk of 'classes of hypotheses' and 'weighting functions' is utterly useless
here-no discussion of the limits of learning can even begin.
(c) THE EPISTEMOLOGICAL ARGUMENT
NELSON GOODMAN
(Jason has brought back from the nomads of Outer Cantabridgia something that
Anticus suspects is more fleece than golden.)
ANTI C U s: Tell me about the resurrection.
JASON: After some centuries, the theory of Innate Ideas has been disinterred, and
enthroned as the only adequate explanation for some striking facts conceming human
linguistic proficiency.
A: What facts?
J: In the first place, that all natural languages, however diverse in origin and in
superficials, have certain remarkable properties in common.
A: But is it remarkable that the elements of any collection have some remarkable
properties in common? Surely we can find throughout the random deals of a pack
of cards during an evening some very special uniformities; but we do not take them
as posing a problem.
J: The claim is of course much stronger: that any language a human being can
acquire has the properties in question.
A: I can imagine having a good deal of trouble mastering a language with an
alphabet of a million letters and no word less than a million letters long. But does this
call for elaborate explanation?
J: The properties in question are more interesting properties of grammatical form
and of meaning.
A: Then the claim is indeed material and testable. I suppose these nomads have
constructed languages lacking the properties in question, and found that eamest
efforts to teach them to human beings fail. This seems to me not only remarkable
but incredible; for the human mind strikes me as agile enough to leam, with appro-
priate instruction and explanation, almost any transformation or distortion of an
already familiar language.
J: I have done them an injustice. They hold only that no language lacking the
properties in question can be acquired by a human being as an initiallanguage. Onee
one language is available and can be used for giving explanation and instruction, the
limitations are transcended.
A: That answers my objection; but now I am puzzled as to how they propose to
SYMPOSIUM ON INNATE IDEAS 141
examine the claim experimentally. Can they really take an infant at birth, isolate it
from all the influences of our language-bound culture, and attempt to inculcate it
with one of the 'bad' artificial languages?
J: No. They readily admit this cannot be done. They regard their claim as a
hypothesis not subject to such direct experimental test, but supported by ancillary
considerations and evidence.
A: Very weil; the claim is that certain statements about the properties of languages
that can be initially acquired are plausible, and a certain explanation illuminating.
But so far we have been speaking vaguely of 'certain properties' or 'the properties in
question'. If we are to judge plausibility, we must surely have a clearer formulation
or illustrations of what these properties are.
J: My informants are not always very explicit about this. They cite some general
grammatical properties now and then; but I know you would say that each of these
has been tailored to fit the known natural languages and derives rather from the
conceptual apparatus we impose upon these languages than from any remarkable
affinities among them. One case that might carry some weight with you, though,
concerns a concocted language called Grubleen. It differs from ordinary English only
in that it contains the predicates 'grue' (for 'examined before t and green or not so
examined and blue') and 'bleen' (for 'examined before t and blue or not so examined
and green') instead of the predicates 'green' and 'blue'. The claim is that while a
user of ordinary English might be taught to use Grubleen, no human being could
acquire Grubleen as an initial language.
A: Though, as you say, experimental support for this can hardly be expected. But
I have another worry. Let us assurne that we now have before us an example of a
language that cannot be so acquired. Still, what in general is the difference between
Grubleen-like and English-like languages? I see by your gesture that you are painfully
aware of the difficulties of answering that question. So far we seem to have concluded
first that the claim we are discussing cannot be experimentally tested even when we
have an acknowledged example of a 'bad' language, and second that the claim has
not even been formulated to the extent of citation of a single general property of
'bad' languages.
J: Nevertheless, important conjectures often cannot in the early stages be either
precisely stated or actually tested. What you have said does not convince me that the
claim ought to be rejected. If it is suggestive or promising enough, we ought rather
to help examine and deve10p it.
A: You are right in principle; but I am not moved to try in this case, since the
claim seems to me discredited by antecedent considerations.
J: Such as?
A: What we call a language is a fairly elaborate and sophisticated symbolic
system. Don't you think, Jason, that before anyone acquires a language, he has had
an abundance of practice in developing and using rudimentary prelinguistic symbolic
142 SYMPOSIUM ON INNATE IDEAS
systems in which gest ures and sensory and perceptual occurrences of all sorts
function as signs?
J: Yes; but language-acquisition is what is at issue.
A: You remember, though, that the real issue is over initial acquisition of
languages, since once some language is available, acquisition of others is relativeiy
easy.
J: True; but surely you do not call those rudimentary systems languages.
A: No; but I submit that our facility in going from one symbolic system to another
is not much atTected by whether each or either or neither is called a language; that
acquisition of an initial language is acquisition of a secondary symbolic system; and
that as we find no interesting limitations upon wh at we can acquire as a secondary
language, we have no ground for believing that there are such limitations upon what
we can acquire as a secondary symbolic system. In other words, when initial-
language acquisition is seen as secondary-symbolic-system acquisition, the claim
that there are rigid limitations upon initial-language acquisition is deprived of
plausibility by the fact that there are no such limitations upon secondary-language
acquisition.
J: I am afraid that what .you say undermines also a second claim; that initial-
language acquisition is astonishingly fast.
A: Yes. If the language were the first symbolic system acquired, and the process
of acquisition considered to begin with the first overt use of words, I suppose we
rnight manage to work up some astonishment. But if acquisition of the first language
is mereiy passage from a symbolic system already acquired to another that we are
taught, that is a much easier step. On the other hand, if the process of acquiring the
first language is thought of as beginning with the first use of symbols, then it must
begin virtually at birth and takes a long time.
J: Does not all this just move the question back from the nature of languages that
can be initially acquired to the nature of symbolic systems that can be so acquired?
I suspect we would find remarkable uniformities and astonishing speed of acquisition
here.
A: We'd certainly have an even harder time doing it. Little of the unimpressive
evidence adduced with respect to languages would be pertinent here; and obviously
we cannot argue back from uniformity of language to uniformity of preiinguistic
system. We'd have to examine symbols that are not overt and articulate but rather
inaccessible and ill-defined. And since the systems are likely to be
fragmentary as weil as rudimentary, we'd have trouble deciding when a system is
acquired. And experimentation under all these difficulties would have to begin with
symbol-using from the moment of birth. But I hardly have to refute your suspicions.
Rather than facts crying for a theory, the theory is crying for the facts.
J: Your objections are more telling against my inadequate presentation than
against the spirit and substance of what I am trying to present. All sophistry aside, is
SYMPOSIUM ON INNATE IDEAS 143
there nothing in human behaviour you find striking enough to demand special
explanation?
A: I can think of some remarkable behavioural facts that call for no such
explanation as a theory of innate ideas.
J: For example?
A: Well, I leamed instantly to fall when dropped, and moreover to fall, no matter
where dropped, precisely toward the centre of the earth.
J: And for this remarkable fact we do need a theory---the theory of gravitation.
A: A set of laws subsuming this behaviour under a very general description; but
I am not inclined to attribute knowledge of these laws to the falling objects.
J: But this is mechanical behaviour, common to animate and inanimate objects
alike. Living things obey more special laws framed in terms of other notions. And
human beings, in their cognitive behaviour, obey still more special laws that require
reference to innate ideas.
A: Your speed there is remarkable enough. Let us take it more slowly. Are you
saying that human cognition is explicable only by supposing that the mind is supplied
at the start with the interpretation of certain symbols? If that means only that it
responds in a fixed way to certain stimuli, this suggests a view of mind we would both
reject. What seems to me notable is not the fixity but rather the flexibility of the mind;
its ability to adapt, adjust, transform ; its way of achieving unity in variety, constancy
amid instability, of inventing rather than obeying. The mind does not merely kick
when tapped; it gropes. The groping and grasping, the seeking and finding, seem to
me more characteristic than any mere programme-reading.
J: You Berkeleyans always overstress the groping.
A: And you Leibnizians overstress the predetermination.
J: We go from peuifogging analysis to loose metaphor, and now name-calling!
But seriously, I think it is just those capacities of the mind that you praise that can
be accounted for only by the instrumentation of innate ideas.
A: We have been paying much less attention to what the theory is than to what
it is supposed to explain. Let us now assume that for certain remarkable facts I have
no alternative explanation. Of course, that alone does not dictate acceptance of
whatever theory may be offered; for that theory might be worse than none. InabiJity
to explain a fact does not condemn me to accept an intrinsically repugnant and
incomprehensible theory. Now I gather that the theory here proposed is that certain
ideas are implanted in the mind as original equipment.
J: Roughly that.
A: And being ideas, they are in consciousness?
J: No, not necessarily; not even usually.
A: Then they are in the subconscious mind, operating upon cognitive processes,
and capable of being brought into full consciousness?
J: Not even that. I may have no direct access to them at all. My only way of
144 SYMPOSIUM ON INNATE IDEAS
discovering them in my own mind may be by the same methods that someone else
might use to infer that I have them, or I to infer that he does.
A: Then I am puzzled. You seem to be saying that these innate ideas are neither
innate nor ideas.
J: What is innate are not concepts, images, formulae, or pictures, but rather
'inclinations, dispositions, habits, or natural potentialities'.
A: But I thought the ideas were posited to explain the capacities. If aB that is
claimed is that the mind has certain inclinations and capacities, how can you justify
caBing these ideas?
J: The justification is historical. Descartes and Leibniz used the term 'innate idea'
in just this sense. But after all, it is the theory that counts, not the term 'innate idea'.
A: In that case, why all the effort at historical justification? And why, after admit-
ting the term is controversial and claiming it is unnecessary, do these people go on
using it? F or a very compelling, but not very good, reason: that until the term 'innate
idea' is applied, what is advocated is the rather trivial truth that the mind has
certain capacities, tendencies, limitations. Once we apply the term, in anything like
its normal use, the thesis becomes far from obvious; but unfortunately, it becomes
false or meaningless. John Locke made aB this acutely clear.
J: Again I am afraid I have not been careful enough. Rather than identify the
innate ideas with capacities, etc., I probably should have said that these ideas exist
as or are 'innate as' such capacities.
A: A few minutes aga you accused me of sophistry; but I bow before the subtlety
of that last statement. Go again, Jason, and bring back to me aB the mysteries of
ideas being innate as capacities. Then, if you like, we can talk again about un-
substantiated conjectures that cry for explanation by implausible and untestable
hypotheses that hypostatize ideas that are innate in the mind as non-ideas.
NOTES ON THE CONTRIBUTORS
J. L. AUSTlN was White's Professor of Moral Philosophy at Oxford from 1953 until his early
death in 1960. His Philosophical Papers were published in 1961, Sense and Sensi bilia in 1962,
and How To Do Things with Words also in 1962. The last -named is a version, not fuUy
completed by Ausrin, of his William James Lectures at Harvard in 1955.
P. F. STRAWSON is Waynflete Professor of Metaphysical Philosophy at Oxford. Among his
publications are An Introduction to Logical Theory (1952), Individuals (1959), and The
Bounds 0/ Sense (J 966). He is the editor of Philosophical Logic in the present series.
J. R. SEARLE (the editor of the present volume) is a member ofthe Department ofPhilosophy
of the University of Califomia at BerkeIey. He has contributed many valuable articIes to
philosophical periodicals, and his book Speech Acts was published in 1969.
H. P. GRICE, formerly a FeUow of St. John's College, Oxford, is now a member of the
Department of Philosophy at Berkeley. He was William James Lecturer at Harvard in 1966.
NOAM CHOMSKY, of the Massachusetts Institute ofTechnology, is perhaps the most influential
of contemporary writers on linguistic theory. Among his publications are Syntactic Structures
(1957), Current lssues in Linguistic Theory (1964), Aspects 0/ the Theory 0/ Syntax (1965),
and Cartesian Linguistics (1966). He was John Locke Lecturer in Oxford in 1969.
JERROLD J. KATZ. of the Massachusetts Institute ofTechnology, is the author of The Problem
o/Induction and its Solution (1962), The Philosophy 0/ Language (J 966), and of many articIes.
He edited, with J. A. Fodor, The Structure 0/ Language (1964).
HILARY PUTNAM, of the Department of Philosophy, Harvard University, has written many
valuable papers on the philosophy of science and mathematics, as weil as on linguistic theory.
He edited, with Paul Benacerraf, Philosophy 0/ Mathematics: Selected Readings (1964).
NELSON GOODMAN, of Brandeis University, was John Locke Lecturer in Oxford in 1962.
Besides many important papers, he is the author of The Structure 0/ Appearance (J 951),
Fact, Fiction, and Forecast (1955), and Languages 0/ Art (1968).
BIBLIOGRAPHY
(not including material in this volume)
I. BOOKS
(I) Modern classics in the philosphy o/language
FREGE, G.: Philosophical writings, trans. P. T. Geaeh and M. Blaek (BlackweU, Oxford,
1952).
R USSELL, B.: Lectures on the philosophy o/logical atomism, reprinted in RusselI, Logic
and knowledge, ed. R. C. Marsh (AUen & Unwin, London, 1956).
WITIGENSTEIN, L.: Tractatus logico-philosophicus, trans. D. F. Pears and B. F.
McGuinness (Routledge & Kegan Paul, London, 1961).
Philosophical im'estigations (Blackwell, Oxford, 1953).
(2) Some recent books
ALSTON, W.: The philosophy o/language (Prentice Hall, New Jersey, 1964).
AUSTIN, J. L.: How to do things with words (Clarendon Press, Oxford, 1962).
C HOMSKY, N.: Aspects 0/ the theory 0/ syntax (M.I.T. Press, Cambridge, Mass., 1965).
GEACH, P. T.: Reference and generality (Cornell University Press, Ithaea, 1962).
KATZ, J. J.: The phi/osophy o/language (Harper & Row, New York, 1966).
QUINE, W. V. 0.: Word and object (Teehnology Press and John Wiley & Son, New
York and London, 1960).
RUSSELL, B.: An inquiry into meaning and truth (Allen & Unwin, London, 1948).
SEARLE, J. R.: Speech acts (Cambridge University Press, London and New York, 1969).
STRAWSON, P. f.: Individuals (Methuen, London, 1959).
ZIFF, P.: Semantic analysis (CorneU University Press, Ithaea, 1960).
11. ARTICLES ON SELECTED TOPICS
(I) Re/erence and the theory 0/ descriptions
DONELLAN, K. S.: 'Referenee and definite descriptions', Phi/osophical review (1966).
FREGE, G.: 'Sense and referenee', Philosophical writings, trans. P. T. Geaeh and M. Blaek
(Blaekwell, Oxford, 1962).
R USSELL, B.: 'On Denoting', Mind (1905), reprinted Readings in philosophical analysis,
ed. H. Feigl and W. Sellars (Appleton-Century-Crofts Ine., New York, 1949).
'Mr. Strawson on referring', Mind (1957).
STRAWSON, P. F.: 'On referring', Mind (1950), reprinted Essays in conceptual analysis,
ed. A. G. N. Flew (Maernillan, London, 1956).
'Identifying reference and truth values', Theoria (1964).
(2) Speech acts and propositions
ALSTON, W.: 'Linguistie aets', American philosophical quarterly (1965).
AUSTIN, J. L.: 'Performative utteranees', Philosophical papers (Clarendon Press,Oxford,
1961).
B1BLlOGRAPHY 147
CARTWRIGHT, R.: 'Propositions', Analytical philosophy, ed. R. J. Butler (Blackwell,
Oxford, 1962).
COHEN, L J.: '00 iIlocutionary forces exist?', Philosophical quarterly (1964).
GEACH, P.: 'Assertion', Philosophical rel'iew (1965).
LEMMON, E. J.: 'Sentences, statements and propositions', British analytical philosophy,
ed. B. A. O. Williams and A. C. Montefiore (Routledge & Kegan Paul, London,
1966).
SEARLE, J. R.: 'Austin on locutionary and iUocutionary acts', Philosophical review (1968).
TEICHMAN, J.: 'Propositions', Philosophical review (1961).
(3) Meaning
CAVELL, S.: 'Must we mean what we say?', Inquiry (1958), reprinted Ordinary language,
ed. V. Chappell (Prentice Hall, Englewood Cliffs, N.J., 1964).
DAVIDSON, D.: 'Truth and meaning', Synthese (1967).
GRlCE, H. P.: 'Meaning', Philosophical rel'iew (1957), reprinted Philosophical logic,
ed. P. F. Strawson (Oxford University Press, London, 1967).
R YLE, G.: 'The theory of meaning', British philosophy in mid-century, ed. C. A. Mace
(Macmillan, New York, 1957).
SHWAYDER, D.: 'Uses of language and uses of words', Theoria (1960), reprinted The
theory 0/ meaning, ed. G. H. R. Parkinson (Oxford University Press, London, 1968).
STAMPE, D. W.: 'Toward a grammar of meaning', Philosophical review (1968).
(4) Truth
AUSTIN, J. L: 'Truth', Proceedings 0/ the Aristotelian Society, Supp. vol. (1950),
reprinted Truth, ed. G. W. Pitcher (Prentice Hall, Englewood Cliffs, N. J., 1964).
DUMMETT, M.: 'Truth', Proceedings 0/ the Aristotelian Society (1958-9), reprinted Philo-
sophicallogic, ed. P. F. Strawson (Oxford University Press, London, 1967).
STRAWSON, P. F.: 'Truth', Proceedings 0/ the Aristotelian Society, Supp. vol. (1950),
reprinted in Truth, ed. G. W. Pitcher (Prentice Hall, Englewood Cliffs, N.J., 1964).
'Truth: a reconsideration of Austin's views', Philosophical quarterly(1965).
T ARSKI, A.: 'The semantic conception of truth', Philosophy and phenomelogical research
(1944), reprinted H. Feigl and W. Sellars, Readings in philosophical analysis (Apple-
ton-Century-Crofts Inc., New York, 1949).
(5) Philosophical aspects 0/ generative grammar
BOYD, J. and THORNE, J. P.: 'The semantics of modal verbs', Journal 0/ linguistic studies,
1969.
CHOMSKY, N.: 'Current issues in linguistic theory', The structure 0/ language, ed.
J. J. Katz and J. A. Fodor (Prentice Hall, Englewood Cliffs, N.J., 1964).
HARMAN, G.: 'Psychological aspects of the theory of syntax', The journal 0/ philosophy
(1967).
KATZ., J. J.: 'Analyticity and contradiction in natural languages', The struclure 0/
language, ed. J. J. Katz and J. A. Fodor (Prentice Hall, Englewood Cliffs, N.J.,
1964).
KIPARSKY, P. and KIPARSKY, c.: 'Fact', Recent advances in Linguistics, eds. Bierwisch
and Heidolph (Mouton, The Hague, 1969).
(6) Analytic propositions
BENNETT, J.: 'Analytic-synthetic', Proceedings 0/ the Aristotelian Society (1958-9).
148 BIBLIOGRAPHY
PUTNAM, H.: 'The analytic and the synthetic', Minnesota studies in the philosophy 0/
science, vol. III (University of Minn. Press, Minneapolis, 1962).
QUINE, W. V. 0.; 'Two dogmas of empiricism', From a logical point 0/ view (Harvard
University Press, Cambridge, Mass., 1953).
STRAWSON, P. F.: 'Propositions, concepts, and logical truth', Philosophical quarterly
(1957).
STRAWSON, P. F. and GRlCE, H. P.: 'In defense of a dogma', Philosophical review (1956).
(7) Miscellaneous
BLACK, M.: 'Metaphor', Models and metaphors (Cornell University Press, New York,
1961).
GEACH, P. T.: 'Subject and predicate', Mind (1950).
TOULMIN, S. and BAIER, K.: 'On describing', Mind (1952).
INDEX OF NAMES
(not including authors mentioned only in the Bibliography)
Aristotle, 106, I 15
Austin, J. L., I, 6-8, 12, 23-7, 30-2, 35-7,
39
Black, M., 42n.
Bush, R., 73n., I IOn.
Carnap, R., I, II, 108, 114, II5n.
Chomsky, N., 10-12, 71-3, 78-80, 83,
84n., 85n., 86-91,94-6,98-100, 105n.,
108, 109, II On., 128-32, 134-8
Curry, R. B., 135
Descartes, 106, 128, 144
Fant, G., 87
Ferguson, C. A., 72n.
Fillmore, C. J., 91, 92
Fodor, J., 71, 75, 87, 89-91, 97, 102n.,
Illn., 129m.
Frege, G., 2-7,43
Galanter, E., 73n., I IOn.
Goodman, N., 11, 12
Grice, H. P., 3, 8-10, 28-30, 33, 35, 37,
44, 46, 51
Halle, M., 71, 72, 82, 83, 87
Hare, R. M., 43
Harris, Z. S., 88
Hart, H. L. A., 30n.
Hockett, C. F., 78
Ilouseholder, F. W., 72n., 82, 83
Hume, 106
Jakobson, R., 87
.Icspersen, 0., 74
Kant, 106, Il3n.
Katz, J., 11, 71, 72n., 75, 80n., 86, 87,
89-93, 95n., 97, 102n., 105n., 108n.,
I1I n., Il3n., 115n., 129n.
Klima, E. S., 90
Lccds, S., 132n.
Lees, R. 8., 71, 80n., 90, 91n.
Leibniz, 144
Lewis, C. 1., 43
Locke, 144
Luce, R. D., 73n., 110n.
Lukoff, F., 72, 87
Lunt, H., 72
MacLean, H., 72
Martinet, A., 129n.
Matthews, G. H., 96
McGuinness, 8. F., 33n.
MiII, J. S., 1
Miliar, G. A., 73n., 95, IlOn.
Nagel, E., 80, 129n.
Plato, 1, 106
Polanyi, M., I I
Postal, P., 72n., 75, 80n., 86-8, 90-3, 95n.,
97, 105n., 108n., 129n.
Putnam, H., 11, 12
Quine, W. V. 0.,1,11,12,103,114, 115n.
Rawls, J., 41n.
Reichenbach, H., 43
Reichling, A., 72n.
Rosenbaum, P., 85n.
RusselI, B., I, 3, 4, 7, I I, 108
Ryle, G., 108
Schachter, P., 96
Searle, J. R., 3n., 8n., 41n.
Sheffer, H. M., 43
Smith, C. S., I IOn.
StockweIl, R., 96
Strawson, P. F., 1, 7-9, 11
Suppes, P., 80, 129n.
Tarski, A., 80, 129n.
Wittgenstein, L., I, 3--6, 12, 46, 78, 108
Ziff, P., 132n.

Also published in this series Moral Concepts edited by Joel Feinberg Theories 0/ Ethics edited by Philippa Foot Knowledge and Belief, edited by A. PhiUips Griffiths The Philosophy 0/ Mathematics edited by Jaakko Hintikka The Philosophy 0/ Science edited by P. H. Nidditch The Theory 0/ Meaning edited by G. H. R. Parkinson Political Philosophy edited by Anthony Quinton Philosophical Logic edited by P. F. Strawson The Philosophy 0/ Perception edited by G. J. Warnock The Philosoph)' 0/ Action edited by Alan R. White Other volumes are in preparation

THE PHILOSOPHY OF LANGUAGE
Edited by

J. R. SEARLE

OXFORD UNIVERSITY PRESS 1971

Ely House..Oiford University Press.1 GLASGOW CAPE TOWN BOMBAY NEW YORK IBADAN CALCUTTA TORONTO NAIROBI MADRAS MELBOURNE KARACHI WELLINGTON LUSAKA TOKYO ADDlS ABABA DACCA SALISBURY DAR ES SALAAM HONG KONG LAHORE KUALA LUMPUR SINGAPORE © OXFORD UNIVERSITY PRESS 1971 FILMSET AND PRINTED IN GREAT BRITAIN BY BUTLER & TANN ER LID. FROME AND LONDON . London W.

THE PmLOSOpmCAL RELEVANCE OF LINGUISTIC THEORY 83 by Jerrold J. Searle IV. AND WORD-MEANING 39 by H. R. P. INTENTION AND CONVENTION IN SPEECH Acrs 13 by P. SENTENCE-MEANING. L. TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 54 71 (a) Assumptions and Goals (b) The Theory of Transformational Generative Grammar by Noam Chomsky VI.CONTENTS INTRoDucnoN I. SYMPOSIUM ON INNATE IDEAS 101 (a) Recent Contributions to the Theory of Innate Ideas by Noam Chomsky (b) The 'Innateness Hypothesis' and Explanatory Models in Linguistics by Hilary Putnam (c) The Epistemological Argument by Nelson Goodman NoTES ON THE CON1RIBUTORS 121 130 140 145 146 149 BIBUOGRAPHY INDEX OF NAMES . Strawson IIl. PERFORMATIVE-CONSfATIVE by J. Austin II. F. WHAT Is A SPEECH ACT? 23 by J. Grice V. UTTERERS MEANING. Katz VII.

.

Wittgenstein. speech acts. he may be regarded as asking a question concerning the concept pious. It is for this reason--among others--that in the spread of analytical philosophy in the twentieth century the philosophy of language has occupied such a central. Though both the philosophy of language and linguistic philosophy are pursued nowadays with more self-consciousness than ever before. have been in varying degrees philosophers of language. for example. Most of the influential philosophers of this century. because the methods linguistic philosophers employ in conducting linguistic analyses depend crucially on their philosophy of language. for example. or it may be done without special regard to traditional problems but as an investigation of concepts for their own interest. and Strawson. and causation. as questions about the analysis of the concept 'true') amI. Carnap. 'hosion'. This may be done in order to solve such traditional philosophical problems as those concerning determinism. place in the entire enterprise of philosophy. Quine. The philosophy oflanguage consists in the attempt to analyse certain general features of language such as meaning. 'linguistic philosophy' is primarily the name of a philosophical method. Austin. Both because some of the problems in the philosophy of language can be profitably attacked by the methods of linguistic philosophy (the problems concerning the nature of truth. verification. more importantly. as an inquiry into certain aspects of the world by scrutinizing the classifications and distinctions we make in the language we use to characterize or describe the world. may be regarded as a question concerning the exact meaning of the Greek word for pious. RusselI. and this. and logical necessity. both are in fact as old as philosophy. But the two. some would say the central. and by analysing logical relations between words in natural languages. reference. It is only given some general theory of or approach to language that one can even get started on a particular linguistic analysis. and its synonyms in other languages. Linguistic philosophy consists in the attempt to solve philosophical problems by analysing the meanings of words.INTRODUCTION I T is important to distinguish between the philosophy of language and linguistic philosophy. are intimately connected. The way a linguistic philosopher conducts an actual investigation will depend on certain general conceptions he has concerning how words mean and how they relate to the world. truth. scepticism. method and subject. most contemporary philosophers would claim. When in the Euthyphro Plato asks what is piety. at least in part. When in the Phaedo he advances the theory that general terms get their meaning by . may be regarded. 'The philosophy of language' is the name of a subject matter within philosophy.

'Sense and Reference'. and is more factually informative than. trans. The reason why 'The Evening Star is the Morning Star' can be more factually informative than 'The Evening Star is the Evening Star'. P. . If.2 THE PHILOSOPHY OF LANGUAGE standing for the Forms he is advancing a thesis in the philosophy of language. Geach and M. Black (Blackwell. viz. the statement 'The Evening Star is the Evening Star'. and the statement conveys that one and the same object has the features specified in the different senses of the two expressions. so he was forced to develop a theory of his own. In the course of his logical investigations he found the prevailing nineteenth-century views about language--represented most influentially by J. on the other hand. Gottlob Frege. namely. even though the reference is the same. to contain any more factual information than a statement of the form. Frege regards this theory of sense and referI Frege. it seems that they must say exactly the same uninformative thing. and the reader should not consider the simplicity of the account as reflecting any simplicity in the actual deve10pment of this subject. Oxford. beginning with Frege. since we can arbitrarily assign any name to any object we like. In order to place contemporary work in the philosophy of language in some historical perspective. To carry out this enterprise he had to invent a new logical system. 1952). ais identical with a? 1 Ifwe construe the statements as being about the objects referred to by the names which replace 'a' and 'b'. and referring to a reference is always achieved by way of sense. G. Frege's most impOItant single discovery in the philosophy of language was the distinction between sense and reference. Frege wanted to show that mathematics was reducible to and founded on logic. Philosophical Wrilin!?s. a is identical with b. there are good reasons for dating the beginnings of modern philosophy of language in the work of the German philosoph er and mathematician.. its reference. In spite of a historical continuity stretching back to the Greeks. He elucidated this distinction in terms of the following puzzle about identity statements: How is it possible for a true statement of the form. Mill---to be rather superficial and in many cases downright false. a thesis about how words mean. The sense provides the 'mode of presentation' of the object. How is this possible? Frege's answer is that in addition to the name and the object it refers to. there is a third element. and to give the reader a general idea of the background to the essays in this volume. I shall now-all too briefly---sketch some of the more important developments in the philosophy of language. S. that an object is identical with itself. its sense (or as we rnight prefer to say in English: the meaning or descriptive content) of the name in virtue of which and only in virtue of which it refers to its reference. Yet obviously the statement 'The Evening Star is the Morning Star' does not mean the same as. What I say will be nothing more than a rough outline. we construe identity statements as about the names used in making them then it seems they must be arbitrary. is that the sense of 'The Evening Star' is different from the sense of'The Morning Star'. T.

this same thought is expressed in the antecedent clause of'If Socrates is wise then he is a philosopher'. and again. the proposition) and the assertion of that proposition is important for the arguments in my article and in Grice's article in this volume. To my rnind it loses the most brilliant insight of the original distinction. But only in the first of these three sentences is that thought asserted. This extension of the distinction between sense and reference to predicates and whole senten ces is less compelling and has been historically less influential than the original distinction for referring expressions. for example. an insight which reveals the connection between reference and truth: namely that an expression refers to an object only because it conveys something true of that object. "On Denoting'. for example.' This is obviously a meaningful sentence. Thus. the circumstance that they are true or the circumstance that they are false) as their reference. Searle. hence there is nothing for the proposition 1 See J. predicate expressions refer to concepts. and have a truth value (i. . R. After Frege the next great work in the philosophy of language was done by Russell in the years before the First W orld War and carried on by his student Wittgenstein in the Tractatus Logico-Philosophicus. I regard the rejection of the theory of sense and reference oby these two as a major mistake. 'The King of France is bald. 'Russell's Rejection of Frege's Theory of Sense and Reference'. Analysis. They thought that it might appear to work for simple cases. rejected the distinction between sense and reference.INTRODUCTION 3 ence as applying not only to definite descriptions like 'the man in a blue shirt' but also to ordinary proper names like 'Chicago' and 'Winston Churchill'. but the puzzle is: how can it be meaningful since there is no King of France. Frege then extends this distinction from singular referring expressions to predicate expressions and to whole sentences. and he says of sentences (at least those where the question oftruth or falsity arises) that they express a thought as their sense. for a critical exarnination of Russell's reasons.e. But one important feature of Frege's account of sentences has survived. Both Russell and Wittgenstein. But a predicate does not convey something true of a concept nor does a sentence convey something true of a truth value. but that under a really rigorous analysis of language into its most elementary forms the relation between words and the world would turn out to be different from Frege's account of it. Mind (1905). Frege says that we must distinguish between the thought expressed by a sentence and the assertion of that thought. for different sets of reasons. having rejected it they developed a theory of how words relate to the world which was quite different from Frege's. and the reasons they give for rejecting it seem to me bad reasons. He says that in addition to expressing their sense. This distinction between the thought or content (or as most philosophers nowadays call it. 2 However. RusselP begins by considering a problem posed by sentences which contain definite descriptions with no corresponding object. 1958. the same thought is expressed by the sentence 'Socrates is wise' and 'Is Socrates wise?'.

In this interpretation we are able to see that the original sentence is meaningful. these names play the crucial role of relating words to things in the world. The sentence might lack a truth value but the lack of a truth value does not render the sentence meaningless. in efTect. for Wittgenstein. But if. how does a sentence come to have 4 Tractatus. Russen rejects the question.'4 An elementary sentence of a language is simply an arrangement or concatenation of names. The proposition is false since there is no King of France. an analysis in which they no longer occur in their original form at alt. and since all meaningful sentences of a language are. and its logical form is that of a conjunction of statements--one of which is an existential statement. . gave the following different answer : the sentence appears to be of subject-predicate form but is not really so. and expressions such as 'there is a'. 3.4 THE PHILOSOPHY OF LANGUAGE expressed by the senten ce to be about. if apparently the proposition it expresses is neither true nor false? Frege's answer was that a sentence could have sense even though the subject expression had no reference. But how do words hook on the the world for Russen. elementary sentences are arrangements of names. Under analysis the sentence really says: There is a King of France. But Russen. . Whatever is King of France is bald. none ofwhich refer to particular objects. He argues. 'nothing'. Rather. but sentences containing them are to be analysed on the model of the analysis of sentences containing the expression 'The King of France'. etc. he says. 'whatever'. logical constants. which are the logically proper names. The object is its meaning. having already rejected Frege's theory of sense and reference. It is important to see the difTerence between Frege and Russell here.203. if definite description and ordinary proper names do not re/er? Part of his answer is that there is a class of expressions of unanalysable simplicity. In the Tractatus the point of contact between words and the world is provided by names. This conception is most fully worked out by Wittgenstein in the Tractatus. as Wittgenstein says. 'something'.. Its grammatical form is misleading as to its logical form. 'anything'. There is not more than one King of France. They simply stand for entities without having any sense or meaning other than standing for entities. 'The name me ans the object. Frege asks: how do proper names and definite descriptions refer to things? His answer is: in virtue of their sense. they have no meaning 'in isolation'. Allthat is left are predicates. hence there is nothing for the predicate to be either true or false of? How can a sentence be meaningful. that neither definite descriptions nor ordinary proper names (since ordinary proper names are just disguised or abbreviated definite descriptions for hirn) refer at all. To suppose mistakenly that it does is simply to confuse sense and reference. ultimately analysable into truth functional combinations of the elementary sentences. and the proposition expressed by it is false.

whereas all true empirical statements are true in virtue of the sense experiences which verify them. In spite of its obscurity it exerted a profound influence in the 1920s and 30s. One difficulty is this: if the meaning of a name is literally the object for which it stands. on top of y. he rejects it completely. partly from the Tractatus and partly from other remarks of Wittgenstein. it seems it would make no sense to assert or deny the existence of the objects named by names. or put less obscurely: all meaningful statements are either analytic on the one hand or empirical and synthetic on the other. and ordinary objects are not 'objects' in his sense. the sentence will be false. Indeed. All analytic truths are true be definition. because any change in contingent features of the world must still be describable in words. The fact that words are arranged in a sentence in a certain way pictures the fact that things are arranged in the world in a certain way.INTRODUCTION 5 asense? How does it differ from. write 'x' followed by 'is on top of' followed by'y'. a group of empirically oriented philosophers interested in science and mathematics. the sentence will be true. Traclalus 2. then it seems the existence of these objects cannot be a matter of ordinary contingent fact. Thus in order to" say that x is. 'Objects make up the substance of the world' and 'Substance is what exists independently ofwhat is the case'. If there actually is a fact such as the picture represents. . and in Wittgenstein's later work. if not. a criterion of meaningfulness. 2. indeed.021. They developed. The picturing relationship may be completely conventional. the Verification Principle. As Wittgenstein says. and the sense of a sentence is provided by this picturing relationship. Here. Paradoxically though-and this is a disadvantage from which Frege's philosophy of language did not suffer--ordinary names like 'Winston Churchill' and 'San Francisco' are not on his account 'names'. This principle states that the meaning of a proposition is its method of verification. especially on the logical positivists. This version of the picture theory of meaning encounters enormous difficulties. for example. 'is on top of' is not another name but is part of a conventional way of picturing the relationship between x and y when x is on top of y. the Philosophical Investigations. One of the aims of the Tractatus was to deli mit the realm of meaningful from meaningless discourse. one can. by convention. The reason for this is that the destruction of any contingently existing object such as a mountain or a car cannot destroy the meaning of any words. they are equivalellt in meaning to the sets of basic propositions recording sense experiences which would constitute their verification. . a list ofnames? Wittgenstein's answer is that the arrangement of names in the sentence constitutes a picture of a fact. one does not have to place the letter 'x' on top of the letter 'y'.024. Different ways of arranging names in sentences make the sentences different pictures of facts.5 Unfortunately he never gives us any examples of objects or names or e1ementary sentences but says that sentences of ordinary language must be analysable into elementary sentences if they are to be fully meaningful.

says Wittgenstein. etc. apart from any actions or intentions of speakers and hearers. 569. Hut this approach has certain real advantages which have survived to the present day and which influence several of the artides in this volume. they treat the elements of language-words. Questions of meaning and reference can be seen in the context of speakers meaning something by something and referring to something in the utterance of an expression. There was a good deal of carelessness in the way philosophers in this period talked about the use of expressions and-with the notable exception of Austin---they did not always distinguish among the different sorts of 'use' to which expressions could be put. contrasting them with 'constatives' . especially by Wittgenstein. Or is the verification principle itse1f meaningless. not the actions and intentions of the speakers are what count. in short. is it not just an arbitrary definition of meaningfulness. . propositions---as things that represent or things that are true or false.. it appears to be quite inconsistent with our ordinary definitions of 'meaningful' and 'meaningless'. logical positivism---have certain characteristics in common. ci!. called attention to a dass of utterances that do not even set out to be true or false. 'is an instrument'. 43. Philosophical Investigations. aim of language is to represent and communicate factual information. For ex am pie. 6 and 'for a large dass of cases. But if the principle is analytic. . which one may or may not choose to accept? And. but rather he is making a prornise.. is not stating a fact about himself to the effect that he promises..6 THE PHILOSOPHY OF LANGUAGE Among the many objections made to the Verification Theory of Meaning is this: What is the status of the Verification Principle itself? Surely the principle is not synthetic. Secondly. Notice that all these theories which I have sketched-Frege. at any rate the primary. They all assume that the only. 'Language'. that the part of language that really counts is the 'cognitive' part. sentences. Austin called such utterances 'performatives'. The elements of the language. para.. Most importantly it recasts the discussion of many problems in the philosophy of language into the larger context of the discussion of human action and behaviour generally. along with that of finding aversion of the Verification Principle that could be precise1y stated. 1 Op. . Wittgenstein argued that stating facts is only one of the countless jobs we do with language and that the meaning of elements lies not in any relationship they have in the abstract but in the use we make ofthem. the meaning of a word is its use in the language'. in a similar vein. a man who says. The aim of language. is to communicate what can be true or false. continued to plague the positivists in the years before the Second World War. para. In the late thirties and especially after the Second W orld War these assumptions came to be vigorously challenged. the Tractatus. for it cannot be just a contingent empirical trait of meaningful utterances that they are verifiable. 'I prornise I will come'. indeed. like the metaphysics it was intended to destroy? These difficulties.7 Austin.

descriptions. promises. Austin.and early twentieth-century work in logic by people of various disciplines 'resulted in a science of logic. the real strength of Frege's theory of sense and reference as opposed to Russell's theory of definite descriptions emerges in Strawson's conception of reference as a speech act. which turn out to be only one dass of speech acts among others. reprinted A. But he attacks it from an interesting angle. grammarians. calls these different kinds of complete speech acts (statements. It is supposed to be a distinction between utterances which are not acts (constatives). 8 Once one sees referring as an action that is performed in the utterance of an expression with a particular sense provided by the rules for the use of the expression. In the first four artides we proceed historically from Austin's attempts to revise the performative-constative distinction to current research on speech acts.. constatives turn out to be speech acts as weil. etc. bets.) illocutionary acts. that between performative and constative utterances. 1956). employing conventional devices (words.INTRODUCTION 7 Instead of seeing the relations between words and the world as something existing in vacua. in his subsequent terminology. 'Strawson. We have by no means yet achieved that result. I hope. will indicate something of the direction from which I think it may come. Flew (ed.g.. one now sees them as involving intentional actions by speakers. . the rest deal with the implications of generative grammar. sentences) in accordance with extremely abstract sets of rules for the use of those devices. with asserting an existential proposition (quite another kind of speech act). For example. 'On Referring'. The original distinction was supposed to be a distinction between utterances which are sayings (statements. linguists. psychologists. as Russell in effect does. etc. Essays in Conceplual Analysis (Macmillan. Which brings us to the artides in this volume. but the articles in this volume. and on this ac count there is much less motivation for trying to identify referring (one kind of speech act). then it is easier to see that it is subject to the sorts of error that plague actions generally (one can fail to refer to a king of France for the same reason that one can fail to hit a king of France: there is no such person). promises. bets.) and utterances which are doings of some other kinds (e. Mind. But. The first four artides centre around the notion of speech acts and the relation of meaning to speech acts. P. warnings. in this century would eventually result in a science of language in much the same way that nineteenth. as Austin argues in this artide and his posthumously published lectures. etc.). Austin once expressed the beliefthat the work being done by philosophers. Making a statement or giving a description is as much performing a speech act as making a promise or giving an order. and utterances which are acts (performatives). and contrasts them with acts which involve achieving certain effects on hearers such as persuading. London. F. Austin in 'Performative-Constative' attacks the very distinction he originally made famous. So wh at were originally presented as special cases of utterances (performatives) now are seen to swallow the general cases (constatives). They divide roughly into two sections. 1950. How Ta Do Things with Words. warnings).

I wish to claim. 'Meaning'. London.g.e. amusing. . but the great bulk of fundamental types of illocutionary acts..). and I argue for it more fully in Speech Acts. 1967). F. requests. 1957. P. such as. reprinted P. I state it rather dogmatically. An Essay in lhe Philosophy 0/ LanKuage (Cambridge University Press. boring. questions. The contrast between the two types of cases emerges when we see that the non-conventional acts will be successfully performed if the 'complex overt intention' of the speaker is grasped by the hearer. statements. any failure of the speaker to achieve his overt intention must be attributable to some breach of rule or convention. on the other hand. Grice analyses meaning as intending to produce an etTect in a hearer by getting him to recognize the intention to produce that effect. London and New York. for the sake of brevity. in quite ditTerent ways and with differing objectives. In the one case therefore (the conventional case) the explicit performative form may be the name of the very act which is performed if and only if the speaker's overt intention is etTective. etc. IO Here. R. Both Strawson and I. pp.. the intended etTect ofmeaning something is that the hearer should know the illocutionary force and propositional content of the utterance. for Strawson are not conventional in that or any other than the trivial sense that they can be performed with conventional devices and they have conventional names. But. and that it. calling someone 'out' in agame. bidding at bridge. I hint at this point in my article in this volume. getting hirn to believe something (the overt intended effect of statements) or getting him to do something (the overt intended etTect ofrequests). etc. IUSearle. understanding the utterance) and perlocutionary etTect.·. Strawson and Grice both think of the 'overt intention' of the speaker in the non-conventional case as the eliciting of some response or etTect in the hearer. H. For example a marriage ceremony. 1969). He concludes that some acts are indeed essentially conventional in that they require extra-Iinguistic conventions for their performance. J.. Strawson (ed. But the overt intended etTect of the speech act may still not be achieved without any breach of rules or conventions. not that he should respond or behave in such and such ways. Philosophical Logic (Oxford University Press. annoying. suffers from a failure to appreciate adequate1y Austin's distinction between illocutionary uptake (i. that is if he understands. 42-50.. frightening. are all essentially conventional in that they require some extra-Iinguistic conventions for their performances. 9 Strawson probes the extent to which illocutionary acts are matters of convention and intention respectively in light of Austin's claim that illocutionary acts are essentially conventional. 9 Grice. but in the other case (the nonconventional case) it cannot be the name of this act. e. Speech AclS. Philosophie'al Re. I think this distinction does not have the force that Strawson attributes to it.8 THE PHILOSOPHY OF LANGUAGE convincing. as weil as Grice's original account of meaning.. attempt to interpret Austin 's account of the notion of an illocutionary act in terms of Grice's analysis of meaning. which he calls perlocutionary acts. With the conventional act.ew.

In this article he attempts to analyse certain fundamental features of language using only 'bmte' notions.. the acts can only be performed within systems of'constitutive' mies and the particular linguistic conventions we have in particular naturallanguages are simply conventional realizations ofthese underlying constitutive mies. . if I fail to perf~rrn my duties under a contract I do indeed destroy the intended perlocutionary efTect of the contract and thus violate the law. request. For example. are conventional is not conclusively settled.? The most promising work in the latter category is in Grice's William James lectures. In both the statement case and the contract case the performative verb is the name of the act which is performed if the overt intention is efTective. The picture presented in my article is that for some acts at least. a portion ofwhich is published in this volume. namely that the overt intentions of meaning something are perlocutionary. In both cases the intended perlocutionary efTect may not be achieved. be explicated without employing such notions as mle or convention. But this coincidence involves no deep linguistic point. etc. To suppose otherwise is to suppose what 1 am alleging is amistake.e. It might be rephrased as folIows: to what extent can such basic illocutionary notions as statement. promise. the hearer may not believe my statement. In the bidding case this is facilitated by some linguistic conventions to the efTect that certain bids are to be made under certain conditions. or he may not believe I have as many high cards as my five no-tmmp bid would indicate. even in those cases where failure to sec ure the perlocutionary efTect does indeed involve a breach of mies. etc. statements and promises. procedure. but employing only such notions as intention. My article attempts to explore the assumption that some types of illocutionary acts are necessarily conventional. because in the one case the overt intention is to make a statement and in the other it is to make a contract. Furthermore. Furthermore-and this is a quite separate point---the question of to wh at extent the 'nonconventional acts' of statement making. In both cases the intended effect of meaning something is that the hearer should understand the utterance. For both Grice and Strawson the basic speech acts are not essentially conventional at all. can only be performed given certain constitutive mies. i.g. it merely illustrates that the law in this case is designed to ensure the perlocutionary efTect once the illocutionary performance has been satisfactorily done. e. then the difTerence Strawson cites does not appear so great. e. It is an assumption that I have by no means proven and wh ich Grice may in the end succeed in disproving. Compare making a statement (non-conventional) and bidding at bridge (conventional). And in neither case does he breach any mies or conventions if he fails to believe me.g..INTRODUCTION 9 Once that point is seen. this is only for the trivial reason that the mies are designed to enforce the perlocutionary efTect. They believe that it just so happens that we do have linguistic conventions for achieving what are natural responses like beliefs and actions. response. etc. This is one of the most important unresolved controversies in contemporary philosophy of language.

is understanding. The last three sections are concerned with transformational generative grammar and its importance for philosophy. I may utter 'It's raining'. which is an illocutionary not a perlocutionary effect. as in the earlier artic1e. say) is that the hearer should know that he has been told (i. and meant as a statement of intention? Both according to Grice involve the intention to produce in the hearer the belief that the speaker believes he will do the act. but this is not adequate to distinguish the two different meant illocutionary forces. The basic notion. but that he should believe that the speaker believes something. He characterizes the intended effect of imperative type utterances as being that the hearer should intend to do something rather than that he should actually do it. Chomsky's work. and what it is for a word or phrase to mean something in terms of what it is for speakers to mean something by doing something. mean U's raining by that utterance. Here Grice modifies his earlier definition of meaning in a way that partially. Such grammars are called 'generative' because they contain systems of rules for generating an infinite set of sentences and thus attempt to account for the ability of speakers of natural languages to produce and understand an indefinite number of completely new utterances. it seems to me that a speaker may utter sentences. of which I am attempting to provide philosophical readers a glimpse with the selections inc1uded in this volume. meant as a promise. But this characterization still seems to me inadequate. The intended effect on the hearer of meaning it's raining when I say 'It's raining' and mean it (as opposed to just uttering 'It's raining' as a pronunciation practice. how on these definitions do we tell the difference between TB do it'.\0 THE PHILOSOPHY OF LANGUAGE In Grice's artic1e he continues work on the project. of explaining what it is for a sentence to mean something. but not entirely. That is. one of the early achievements of this type of linguistic theory was its demonstration of the inadequacy of phrase-structure rules alone to account for the syntactical complexity of natural human languages. For example. they contain rules which transform the structure of sentences by rearranging or deleting elements as weil as rules which structure sentences into such phrases as the noun phrase and the verb phrase. and he characterizes the intended effect of indicative type utterances as being. me an what he utters and still not have these intentions. undertaken in his 1957 artic1e. avoids the objections I just made. First. The intended effect of meaning. because they contain transformational as weil as phrase-structure rules. for several reasons. not that the hearer should believe something. the statement has been made to hirn) that it is raining. and still not care a hang whether my hearer believes that I believe it's raining. of which here are two. utterances of sentences they have never heard before. is is not c1ear how these definitions of meaning would discriminate among the many different kinds of illocutionary acts. in short. is the notion of a speaker performing an action and meaning something by that performance. For example.e. They are called transformational. does not lend itself to the kind of piecemeal philosophical sniping that I have been making at the other authors . Secondly. Indeed.

on such an interpretation. But that mastering competence in a language is a matter of mastering a system of rules as opposed to... Personal Knowledge (University of Chicago Press. e. they are of subjectpredicate form. say. projection rules.INTRODUCTION 11 because of its global and systematic character. . as we have seen. etc. however. a symposium between Chomsky. It is not always clear that he solves as much as he claims to solve. to emphasize that from a philosopher's point of view a great deal is assumed and a great deal is left unclear.g. Putnam. concerns the innateness hypothesis. M. In neither case. as far as Chomsky teils us. It is important.. But the notion of an internal representation would. Furthermore. 'The King of France is bald' to show the Russellian distinction between existential and subject-predicate sentences. Again Katz's remarks about logical form settle none of the outstanding issues about logical form. The last section. acquiring a set of dispositions to behave.g. Katz applies the semantic component of generative grammar in an attempt to solve certain traditional problems in the philosophy of language. it is puzzling that he should think his explication of the notion of analyticity in terms of the concepts of semantic theory is an answer to Quine's doubts about analyticity. For example. say. it is not totally clear what such an internal representation is supposed to be. and Goodman. but their logical form is that of an existential sentence. Chomsky teils us as one of his 'uncontroversial' assumptions that 'the competence of the speaker-hearer can ideally be expressed as a system of rules that relate signals to semantic interpretations of those signals'. is a controversial--()r at any rate challengeable--assumption. maintains that sentences with adefinite description as a grammatical subject are misleading as to their logical form. It happens to be a view I hold. The most serious contemporary philosophical dispute about logical form concerns Russell's theory of descriptions. is just as meaning-Iaden as Carnap's notion of a semantic rule. Katz presupposes intensional notions and does not explicate them in the extensional and behavioural fashion that Quine would regard as acceptable. 1960). not a subjectpredicate sentence at all. would we have to interpret the body or the speaker as knowinganyrules.. For example. It is hard to see how a Katzian interpretation of this issue would contribute anything at all to settling it except perhaps by begging the question in favour of Strawson. says RusselI. when Chomsky teils us that the child who learns the language acquires an 'internal representation' ofthese rules. because Quine objects to any account that relies on notions such as meaning or semantic rules. e. allow us to interpret it in other ways. But Katz's apparatus of semantic markers. as implying only that the idealized linguistic competence of the speaker-hearer can be described by these rules in the same way that the laws of physics describe the idealized behaviour of a falling body. Grammatically. 1s it supposed to mean that he has 'tacit knowledge' of the rules in the sense of Polanyi? 11 I rather think so. but it is not an obvious truth. Chomsky and his followers claim that information 11 Cf. Russell. since there is nothing in the underlying phrase marker of. which Quine was originally concerned to attack. Polanyi.

Putnam and Goodman offer arguments to counter some of the claims made by the innateness theorists. The dispute over this claim breaks down into two questions. the theory that human beings have concepts implanted in their minds prior to any experiences. Only by supposing that the child has the form of the grammar programmed into his brain can we account for his intellectuaI feat in acquiring these rules. the socalled 'ordinary language' approach of Wittgenstein and Austin. First. . that the future development of the subject is Iikely to come from joining the two latter approaches. but they do not come to grips with the more important of Chomsky's arguments. Chomsky's critics would have to show either that he is mistaken in his empiricaI theory of the syntax of natural languages or that he has drawn the wrong conclusion from that theory. I think. a historicaI question : what did the rationaIists mean by their doctrine of innate ideas and is it what Chomsky thinks they meant? And second. This. There are three main contemporary approaches to the philosophy of language : the neo-positivist-symbolic logic approach represented most ably by Quine. and the generative gram mar approach of Chomsky and his followers. as the seIections in this volume indicate. For example Chomsky claims that the rules reIating deep and surface sentence structures are so complex and so abstract as to defy any stimulus-response or associationist account of their learning. is an empiricaI issue: how complex and abstract are the syntactical ruIes of naturallanguages? In order to answer his claims. at least in part. for our purposes the more interesting question : what assumptions must we make about the innate intellectual capacities of children to account for the way in which they acquire languages? In this symposium at least it is my opinion that Chomsky definitely has the better of the argument on the second question.12 THE PHILOSOPHY OF LANGUAGE currently available in empirical linguistic theory supports the traditional rationalist conception of innate ideas.

it seemed to me preferable simply to translate into English the entire French text. which one scarcely could perform. I advise you to do it. I am sure that 'statement' is the English term Austin would have used here. it is used to perform an action. I welcome you. To issue such an utterance 3 is to perform the action--an action. so dear to philosophers. to procure copies at least of those portions of the original transcript that were in English. that aII those taking part are known to have seen and approved for publication. From Philosophy and Ordinaty Language. mainly for the reason that it is this edited version. and this only. Trans. Trans. La Philosophie Analytique: Les Editions de Minuit. or rather. at least with so much precision.1 l ['Formuler un tel enonce'. or anywhere else. I apologize. perhaps.j [The French term is 'assertion'. [Not here reprinted. Here are some examples: I name this ship Liberte. 1962. which he wrote in French and presented at a (predominantly) AngloFrench conference held at Royaumont in March 1958. the constative utterance. WARNOCK ONE can quite easily get the idea of the performative utterance-4hough the expression. J. G. This idea was brought in to mark a contrast with that of the declarative utterance. J. The case ofthe discussion which follows it I is somewhat more complex. Philosophie No. pp. The performance utterance. Warnock and the University of IIIinois Press. G. as I am going to call it. 27 I -304) is aversion of this. Urbana. Caton (University ofIllinois Press. Ought we to accept this Performative-Constative anti thesis ? The constative utterance. pp. The translation is supplied in a footnote by Austin himself. Reprinted by permission of Mrs. as I am weil aware. Ed. based on a transcript but substantiaIly cut and edited.I PERFORMATIVE-CONSTATIVE J. 22-23. of statemenr. can never be either: it has its own special job. 1963). The actuaI discussion at Royaumont was carried on in both French and English. IV. What appears in the published volume after Austin's text (Cahiers de Royaumont. and I have so translaled 'assertion' throughout. J. in which the contributions originaIly made in English were translated into French by M. L. Bera. under the name. for the present publication. AUSTIN Translator's Note: 'Performative-Constative' is a straightforward translation of Austin's paper 'Performatif·Constatif'. And there we have straight off what I want to call in question. by contrast. Austin. in any other way. edited by Charles E. However. lt might have been possible.] I 2 . does not exist in the French language.2 has the property of being true or false.

. 'unhappy' . our performative. I shall not be in order ifI proceed to treat you as an enemy or an intruder. as we see. I can't name the ship ifI am not the person properly authorized to name it. but still. may be. . such-andsuch a future event. it is issued insincere(v. this performative utterance----:iust is the act of making a promise . If. those creatures being scarcely susceptible of that exploit. but that. as we call it in general. and I can't quite bring offthe baptism of penguins. there may always crop up a third kind of unhappiness. as we say. like any other ritual or ceremony. for instance. the performative utterance is not exempt from all criticism: it may very weil be criticized. sincerely. simply by issuing his utterance. Let us now suppose that our act has been performed: everything has gone off quite normally. 'unhappy' in a different way--if. but in a quite different dimension from that of truth and falsity. I shall not be in order if I break my word. will not be in order. because it may very weil imply that some other propositions are true or are false. everywhere in wh at are called in English the 'operative' c\auses of a legal instrument. as opposed to those--the 'preamble'-which set out the circumstances of the transaction. ifthey happen. The performative must be issued in a situation appropriate in all respects for the act in question : if the speaker is not in the conditions required for its performance (and there are many such conditions). that is to say. Thus a bigamist doesn't get married a second time. which we 'The clauses. and also.' without in the least intending to carry out the promised action. as the lawyers say. ifyou like. Thus we say that. is a quite different matter. will be in order. 'null and void'. there is an 'unhappiness': I have abused the formula. and such-and-such other events. though not void.' --to issue. then he doesn't manage. he only 'goes through the form' of a second marriage. in consequence of the performance of this act. then his utterance will be. to carry out the purported act. And it may seem at once quite obvious that an utterance of this kind can't be true or false---notice that I say it can't be true or false. if it happens. However. If I say 'I promise to . certainly. a performative utterance may be. that is. Second.. Trans. in which the legal act is actually perfonned. or if the object with respect to which he purports to perform it is not suitable for the purpose.] . not. many of them are not without interest for philosophers: to say 'I promise to . and he supplies it himself in brackets after the French 'malheureux'. if I'm not mistaken.5 First. even when the performative has taken effect. We mean rather that. 4 Plainly. 5 l'Unhappy' is a term Austin regularly used in this connection. perhaps even not believing that it is in my power to carry it out. In that case. It is made. . at all a mysterious act. the promise is hollow.14 PERFORMA TIVE-CONSTA TIVE Utterances of this kind are common enough: we find them. the performative utterance will characteristically 'take effect'. If I have said 'I promise'. the speaker is not in a position to perform an act of that kind. We do not mean by that that such-and-such a future event is or will be brought abaut as an effect of this action functioning as a cause. if I have said 'I welcome you'. for example.

poor thing. be c1assifiable 7 under more Ihan one heading. that is. on the other hand. we may setde the question by asking wh ether it would be possible to insert in it the word 'hereby' or some equivalent-as. as practically goes without saying.'. There we have. it may suffer from defective grammar. and other persons and tenses of the same verb. as sometimes we may. or 'he promises'. not to perform one. curiously enough. abuse (insincerity). nullity. By way of putting to the test utterances which one might take to be performative. If we ask ourselves. and breach of commitment. Trans. exaggerated and. [ ['ThaI is 10 say. it is very natural to hope that we could proceed to find some criterion. then. in large measure. . Weil. three kinds of unhappiness associated with the performative utterance. It is true that there exist two 'normal forms'. so to speak. 'I promise' is a formula which is used to perform the act of promising.PERFORMA TIVE-CONSTA TIVE 15 call 'breach of commitment'. 'I promised'. in which the performative finds expression. the performative may be issued under duress. whether of grammar or of vocabulary. Thus. . in French. or from misunderstanding. employs by contrast a verb in the passive voice and in the second or third person of the present indicative.6 We may note also that commitments can be more or less vague. The other form.. Trans. or by accident. But this hope is. perhaps. or in apoern. We leave all that on one side--let us simply bear in mi nd the more specific unhappiness ofthe performative. between the first person singular of the present indicative. One of these normal forms is that which I have already made use of in producing my examples: the utterance leads off with a verb in the first person singular of the present indicative active. as in 'I promise you that . which would make it possible for us to answer in every case the question whether a particular utterance is performative or not. For example. we make use of a well-known asymmetry. and can bind us in very different degrees. alas. in the case of what we call an 'explicit performative' verb. are expressions which serve simply to describe or report an act of promising. It is possible to make a complete classification of these unhappinesses. in a play. it may figure in a context not wholly 'serious'. the different kinds may not always be sharply distinguishable and may even coincide. as in 'Passengers are requested to cross the line by the footbridge only'.1 . ['Rap/are d'engagemen/'. a particular case of unhappiness mighl arguably. which comes to exactly the same but is more common in utterances issued in writing. but it must be admitted that.7 Then we must add that our performative is both an action and an utterance: so that. now that we have before us this idea of the performative. Austin himself supplies the translation. or even quile properly. whether a given utterance of this form is performative or constative. as weil as those in which utterances in general can be. vain. have a thoroughly constative look. it can't help being liable to be substandard in all the ways in which actions in general can be. At first sight both of them. the phrase 'par ces mots-ci'.

I say 'to make explicit'. To say 'Shut the door'.] . If I say simply 'I will be there'. by this little word. but does not state that it is that action. in any way state that I am doing obeisance to you. or again as a warning. the very same act as by the utterance 'I warn you that the dog is about to attack us'. or making perhaps a fatalistic prediction. of that ambiguous phrase). l'Ajlirmer. We can't. But we have been talking all along as if . whether I am taking on a commitment. we can make use. the ideas of the performative utterance. and quite unambiguously so. and at the same time more precise. then. S Bending low before you. Wehave now brought in. what act it is that the speaker purports to perform in issuing his utterance. should be expressed in one of these so-called normal forms. with the heIp of a dictionary. Thus we will achieve a useful classification of all the varieties of acts that we perform in saying something (in one sense. and above all. of its unhappinesses. of a whole lot of more primitive devices such as intonation. One may think of the precise formulae as a relativeIy recent phenomenon in the evolution of language.. going on from there. To make our utterance performative. or perhaps I say 'Salaam'. but the word 'Salaam' does not. Then. 'I order you . in place of the explicit formula. we should be able. then. it is not in the least necessary that an utterance. Trans.. those that have no explicit performative formula. one performs. The other forms of expression. with the notice on the gatewe just don't need to ask ourseIves that question at all. We may hope. to make a list of all the verbs which can figure in one of our explicit formulae. and as going together with the evolution of more complex forms of society and science. will be more primitive and less precise. certainly. for instance. Even the word 'Dog' by itself can sometimes (at any rate in England. and that is not at all the same thing as to state. and of its explicit formulae. Does this word 'Dog' just give us a bit of detail about the local fauna? In the context--when confronted. all the same. further. for example. I have translated this verb by 'state' throughout. It is in this way that our formula makes the issuing of the utterance that action which it is. or gesture. All we can really say is that our explicit performative formula ('I promise . is every bit as performative. '. just by considering the words.) serves to make explicit. any more than does the act of removing my hat. I am doing obeisance to you. the very context in which the words are uttered can make it entireIy certain how they are to be taken--as a description. one might almost say more vague.16 PERFORMATIVE-CONSTA TIVE However. or by 'Strangers are warned that here there is a vicious dog'. that any utterance which is in fact performative will be reducible (in some sense of that word) to an utterance in one or the other of our normal forms. at least. plainly enough. a country more practical than ceremonious) stand in place of an explicit and formal performative.'.. or declaring an intention. . there will be no telling. every bit as much the performance of an act. not just engaging in gymnastics. that is. I remove my hat. then. if it is to be performative. expect any pureIy verbal criterion of the performative. etc. as to say 'I order you to shut the door'.

John has no children. I would like to take a doser look at three rather more subtle ways ofbeing absurd. two ofwhich have only recently co me to light. as a matter of fact. It can be untrue. What is impossible is to state both at the same time: his stating that the cat is on the mat is what implies that the speaker believes it iso By contrast. 18n. or to refer to them. Trans. pp. but [or 'and'] John has no children'. 2. we couldn't say 'It could be the case both that John has no • [The French phrase is 'peut ne pas jouer'. In each of these cases one experiences a feeling of outrage. (2) Someone says 'The cat is on the mat. and 31).1 . To talk about those children. Austin himself sometimes used in English the coined term 'non-play' (see. ungrammatical). By contrast. But it is not. or perhaps that word that we always find so handy. presupposes that they exist. and sirnilarly. as a matter offact. when. How To Do Things with Words. (1) Someone says 'All John's children are bald. Not only 'John's children are bald'. and 'entail' l l for our three cases respectively. and it's possible each time for us to try to express it in terms of the same word-'implication'. 'None of the guests is French' does not. or perhaps he says 'The cat is on the mat'. 'imply'. presupposes that John has children. therefore constative. to be sure. can fail to get by 9 in more than one way. to do violence to language one does not always need a contradiction.1 10 English proverb. he does not believe it iso (3) Someone says 'All the guests are French.IO and equally. 11 [These three English terms are supplied in a footnote by Austin himself. imply that I believe it is. equ8uy with 'All the guests are French'. those two propositions are not in the least incompatible: both can be true together. We can quite weil say 'It could be the case both that the cat is on the mat and that I do not believe it is'. 'The cat is not on the mat' does not. and that not necessarily in some gross fashion (by being. entail that it is false that some of the guests are not French. but equally 'John's children are not bald'. equally with 'The cat is on the mat'. Let us note in the first place that an utterance which is undoubtedly a statement of fact. for instance. when. and as ifthe idea ofthe constative at any rate was as dear as it is familiar. and then afterwards says 'Some of the guests are not French'. 'contradiction'.. but it can also be absurd.g. But there are more ways of killing the cat than drowning it in butter. e. or perhaps he says 'All John's children are bald'. Trans.PERFORMA TIVE-CONSTA TIVE 17 every utterance had to be either constative or performative. Let us use the three terms 'presuppose'. Then: 1. That is to say. [ am told that this rather refined way of disposing of cats is not found in France. but in a more restricted sense than would be appropriate here. and some ofthem aren't': or perhaps he says 'All the guests are French'. but [or 'and'] I don't believe it is'.

If' AU the guests are French' entails 'It is not the case that some of the guests are not French'. but tor 'and'] I have no intention of being there'. For to say either 'I bequeath my watch to you' or 'I don't bequeath my watch to you' presupposes equally that I have a watch. but tor 'and'] I haven't got a watch'. let's first take two performative utterances: 4. which is exactly what lawyers would say about the purported bequest of the watch. equally we can take over for that doctrine the term 'void' as employed in the doctrine of the unhappinesses of the performative. To bring out the comparison. it isn't like this with presupposition: if'John's children are bald' presupposes that John has children. If we don't hold the belief. so my saying I promise to be there implies that I intend to be there. 'void for lack of reference'. Just as my saying that the cat is on the mat implies that I believe it is. for us. So he re is a first instance in which a trouble that affiicts statements turns out to be identical with one of the unhappinesses typical of the performative utterance. the case where something is 'implied'. that is. just as we couldn't say 'It could be the case both that all the guests are French and that some of them are not French'. it isn't true at all that to say 'I don't believe that the cat is on the mat' implies that the cat is not on the mat (not.18 PERFORMA TIVE-CONSTA TIVE children and that his children are bald'. And just as we can make use here of the term 'presupposition' as employed in the doctrine of the constative. if'The cat is on the mat' implies that I believe it is. 5. at any rate. besides. that the watch exists is presupposed by the fact that it is spoken of or referred to. By contrast. the intention to do whatever it may be that they promise. It's a question here of the compatibility and incompatibility of propositions. in the performative utterance just as much as in the constative utterance. The procedure of stating is designed for those who honestly believe what they say. or perhaps someone says 'I promise to be there' when he doesn't intend to be there. I would like to call attention to the fact that these three ways of failing to get by correspond to three of the ways in which a performative utterance may be unhappy. The statement on the subject of John's children is. We compare case 5 with case 2. and without being a sheer rigmarole either. or perhaps someone says 'I bequeath my watch to you' when he hasn't got a watch. exactly as the procedure of promising is designed for those who have a certain intention. the case. or again don't . We compare case 4 with case 1. 'I promise to be there. Here then are three ways in which a statement can fail to get by without being untrue. of presupposition. then 'Some of the guests are not French' entails 'It is not the case that all the guests are French'. namely. it isn't true at all that 'John has no children' presupposes that John's children are not bald. that is. 3. is not a matter of the incompatibility of propositions). we have already seen that 'implication'. in the same sense of 'implies' . we may say. Similarly. 'I bequeath my watch to you.

and in this case my 'I 12 [The French text has 'conjoncture' here. there will be a breach of commitment that one might weil compare with that of the case in which I say 'I welcome you'. it seems to me that the constative utterance is every bit as liable to unhappinesses as the performative utterance. where a trouble which afflicts statements is identical with one of the unhappinesses which atHiet performative utterances. we announce in the same breath that we don't believe it or we don't intend to. it often happens that aperformative is void because the utterer is not in astate. not at any rate without adding 'but he had no right whatever to do so'. too. Furthermore. I state things incompatible with my utterance (namely. Trans. to perform the act which he purports to perform . or not in a position. because one is simply not in a position to state whatever it may be-and this may come about. my act is only purported. It's a free country. then.selves whether there are not still more unhappinesses in the case of statements. Another instance. isn't it? To state what isn't true is one of the Rights of Man. the case of entailment among statements. do I not commit myselfin a more or less rigorous fashion to behaving in future in such-and-such a way. Can we find. that aIl the guests are French). I . I cannot state at this moment how many people there are in the next room : I haven't been to see. thus. making use of the key provided by our list of unhappinesses noted for the case of performatives. for instance. at the same time as we make the statement or the promise. in the case of performatives. it's no good my saying 'I order you' if I have no authority over you: I can't order you. For example. but this must surely be amisprint für 'conjecture'. this impression can lead us into error. one can be mistaken. appropriate to the content of our utterance. is in question. If. perhaps. in the sequel. Let us look back. the impression that where astatement. and hence our feeling of outrage on hearing it. I haven't found out the facts. next. we can ask our. my utterance is void. for more than one reason. 'At this moment there are fifty people in the next room'? You will allow. besides the three we have just mentioned. Now people have. that in saying that I have made a guess. and then proceed to treat you as an enemy or an intruder--and perhaps even better. However. then the utterance is 'self-voiding'. What if he's ilI-informed? Weil then. In reality nothing is more common than to find that one can state absolutely nothing on some subject. that's all. in particular with respect to the statements I will make? If. nevertheless.12 but you will not allow that I have made a statement. then in each case there is lack of sincerity and abuse of the procedure. the case is quite different: anybody at all can state anything at aIl. I know. What ifl say. a constative utterance.PERFORMA TIVE-CONST A TIVE 19 have the intention. with that of which one is guilty when one says 'I define the word thus' (a performative utterance) and then proceeds to use the word with a different meaning. and indeed to pretty much the same unhappinesses. 'All the guests are French'. as we might call it. So then. to case 3. some analogue for this as weil? When I make the statement.

in your interest? There is confrontation of my utterance with the situation in. and that the constative is liable to the same unhappinesses as the performative. but was I right? Many other utterances which have an incontestably performative flavour are 13 lAustin supplies this English term himself. It is.' is exactly on a par with our 'I order . You say 'What do you mean. Trans. but was I right? Was my belief. perhaps. a performative utterance. etc... You confide to me Tm bored'. and I quite coolly reply 'You're not'. in these circumstances.. or as things turned out. of stating. what right have I? I'm just stating what your feelings are. Here and now it remains for us to examine. It is in any case the term he regularly used. I believed that to do it would be in your interest. the conditions for success are fulfilled. 1 was fully justified perhaps. that still does not suffice to set it beyond the reach of all criticism. it is cIear that if we establish that aperformative utterance is not unhappy. So far I have called attention to two things: that there is no purely verbal criterion by which to distinguish the performative from the constative utterance. Now we must ask ourselves whether issuing a constative utterance is not. that I advise you. said. that is. unless you have discIosed them to me. '-a formula which. without any right to give an order. justified? Or again--though perhaps this matters less-was it in fact. In saying that. that's all.l . certainly.. What we need. something which people think is peculiar to statements alone and ought to be set up on a pedestal of its own. And this time let's beg in with the performative utterance: is it the case that there is nothing here in the least analogous with truth? To begin with. truely or falsely. the performance of an act. it was issued. but what of that? I suppose one can always make a simple statement. after all. . as we put it.? 1 can't plumb this mystery any further at present. Here's another example. and the situation with respect to which. apologizing.20 PERFORMA TIVE-CONSTA TIVE state . is a more general theory of these speech-acts. I actually do advise you to do it--it is not that I state. namely. Is stating an act in the same sense as marrying. that one can't issue any utterance whatever without performing some speech-act of this kind. betting. quite briefly. I spoke in alI sincerity. and let us allow that alI the circumstances are appropriate. .. we remember. and in this theory our Constative-Performative antithesis will scarcely survive. can't one?' But no.' is cIosely similar to the formula 'I warn you that . then. a little question: was the advice good or bad? Agreed. and also. serves to make explicit what speech-act 13 it is that we are performing. Let us suppose that I say to you 'I advise you to do it'. one can't always: usually. above the battle. the act. There does still arise. this craze for being either true or false. But it is already pretty evident that the formula 'I state that . I can't state what your feelings are. that its author has performed his act happily and in all sincerity. I'm not? What right have you to say how I feeI?' I say 'But what do you mean.'. I may be mistaken. all the same. It may always be criticized in a different dimension. .

and that this army to some extent won a confused sort of victory. true perhaps for some purposes or in some contexts. oversimplifies things. it's true that Lord Raglan was in command of the allied army. What are we to say.PERFORMA TIVE-CONSTA TIVE 21 exposed to this second kind of criticism. Allowing that you had the right to reprimand him as you did. It's quite true that for each of these statements we can raise the question 'true or false'. When the question is raised one understands that the utterance is to be confronted in one way or another with the facts. induding the circumstances of the occasion of utterance. that would be a fair judgement. once and for all. up to a point ifyou like. or fair. perhaps not a very dear one. though really it's a bit of an exaggeration. Here again we have confrontation with the facts. is it true or not? The question. Well. or just. you have reached your verdict properly and in good faith. I for my part don't think it iso Even if there exists a well-defined dass of statements and we can restrict ourselves to that. Then Alma. of this criticism. in dedaring the accused guilty. wh at is dear is that there is a whole lot of things to be considered and weighed up in this dimension alone ---the facts. so long as you want only a certain degree of precision. not a simple quality nor a relation. for schoo1children anyway. it still remains to ask whether the verdict wasjust. it's true that that city is 60 miles from London. It's a rough statement. Oh well. this dass will always be pretty wide. But it is only in quite favourable cases that we ought to expect an answer yes or no. and that's that. In this dass we shall have the following statements: France is hexagonal. Very well. And so on. surely. or fair. well yes. plainly. and that is this: aren't these questions about something's being good. is a very simple black-and-white business: either the utterance corresponds to the facts or it doesn't. but one can't just say straight out that it's false. a soldier's batde if ever there was one. or deserved entirely distinct from questions oftruth and falsehood? That. Allowing that. even well deserved. his purpose in speaking. That not all performative utterances without exception are liable to this quasiobjective evaluation-which for that matter must here be left pretty vague and multifarious-may very well be true. but also the situation of the speaker. yes. So let's confront 'France is hexagonal' with France. one can still ask whether your reprimand was deserved. I see what you mean. his . and that you have acted without malice. Under the heading 'truth' what we in fact have is. no denying that. We can get some idea. yes. not indeed one anything. And Oxford. but rather a whole dimension of criticism. that would do for the the man in the street but not for geographers. Lord Raglan won the batt1e of Alma. Oxford is 60 miles from London. There is one thing that people will be particularly tempted to bring up as an objection against any comparison between this second kind of criticism and the kind appropriate to statements.

. What we need. is a new doctrine. not just in this or that aspect abstracting from all the r~st. and of what I call the speech-act. the full and the concise. we fee! ourselves driven to think again about the Performative-Constative antithesis. the summary and the detail. questions of precision. etc. precise.. from the side of truth and falsehood. we shall never succeed in disentangling the true from the just. fair. and so on. it seems to me. of what one is doing in saying something.22 PERFORMA TIVE-CONSTA TIVE hearer. then. exaggerated. in all the senses of that ambiguous phrase. etc. From this side also. deserved. If we are content to restrict ourse!ves to statements of an idiotic or ideal simplicity. both complete and general. but taken in its totality.

1 am not suggesting that the distinction between the normal or serious use of speech and the secondary uses which he calls etiolated or parasitical is so c1ear as to call for no further examination . Oetober 1964. are to How To Do Things with Words (Oxford. 439-60. and sometimes--as. The meaning of a (serious) utterance. un1ess otherwise indieated. Whatever doubts may be entertained about Austin's notions of meaning and of locutionary act. 4. if only to get them out of the way. or not directly. L. LXXIII. is used---the meaning of an utterance may exhaust its From The Philosophical Re"iew. Austin distinguishes the illocutionary force of an utterance from what he calls its 'meaning' and distinguishes between the illocutionary and the locutionary acts performed in issuing the utterance. as 1 think we c1early can. to etiolated or parasitical uses. My second preliminary remark concerns another distinction. but 1 shall take it that there is such a distinction to be drawn and 1 shall not here further examine it. in some cases where an explicit performative formula. Reprinted by permission of the author and the editor of PhilosophicaJ Rel'iew. and so it will be with my comments on his doctrine. or pair of distinctions. Doubts may be feit about the second term of each of these distinctions. STRA WSON I 1N this paper 1 want to discuss some questions regarding J. . no. which AustilT draws. It may be feit that Austin has not made c1ear just what abstractions from the total speech act he intends to make by means of his notions of meaning and of locutionary act. always embodies some limitation on its possible force. like 'I apologize'.II INTENTION AND CONVENTION IN SPEECH ACTS P. 1962). vol. F. as conceived by Austin. I All references. Austin's notions of the illocutionary force of an utterance and of the illocutionary act which a speaker performs in making an utterance. Although this is a question on which 1 have views. Austin contrasts what he calls the 'normal' or 'serious' use of speech with what he calls 'etiolated' or 'parasitical' uses. it is not what the present paper is about. it is enough for present purposes to be able to say. for example. pp. His doctrine of illocutionary force relates essentially to the normal or serious use of speech and not. ' There are two preliminary matters 1 must mention. the following about their relation to the notion of illocutionary force.

(This latter notion I shall assurne to be familiar and perspicuous. . It is what we do. I think. 99 and Lecture X). does not exhaust. or as to how the utterance was to be taken or ought to have been taken (pp. A locutionary act is an act of saying something. though it limits. It is a sufficient. Austin does not regard this characterization as by any means a satisfactory test for identifying kinds of illocutionary acts since. but . then (it is held) the illocutionary act of waming cannot be said to have been actually performed. it involves 'bringing about the understanding of the meaning and of the force of the locution' (pp. he gives us certain general clues to these ideas. so regarded.) 4. condition of a verb's being the name of a kind of illocutionary act that it can figure. which may be grouped. it is to be sharply contrasted with the producing of certain effects.2 Perhaps we may express the relation by saying that to know the force of an utterance is the same thing as to know what illocutionary act. as folIows. there may still be a further question as to how what was said was meant by the spea~er. The illocutionary act is 'a conventional act. under four heads: l. if any. but very often the meaning. a necessary. These two notions are not so closely related that to know the force of an utterance is the same thing as to know what illocutionary act was actually performed in issuing it. a warning is not understood in this way (that is. 105). 3. in saying what we say. Now I shall proceed to assemble from the text some indications as to what Austin means by the force of an utterance and as to what he means by an illocutionary act.very often there is more to know about the illocutionary force of an utterance than we know in knowing what locutionary act has been performed. As such.24 INTENTION AND CONVENTION IN SPEECH ACTS force. 2. we must know the answer to this further question. Similarly. Given that we know (in Austin's sense) the meaning of an utterance. the force. that is. an act done as conforming to a convention' (p. say. an iIlocutionary act is an act we perform in saying something. there may sometimes be no more to say about the illocutionary force of an utterance than we already know if we know what locutionary act has been performed. was actually performed in issuing it. This producing of 2 I refer later to the need for quaIification of this doctrine. So much for these two prelirninaries. 'The performance of an illocutionary act involves the securing of uptake'. In order to know the illocutionary force ofthe utterance. as what Austin calls an explicit performative. 98-9). For if an utterance with the illocutionary force of. that is. thought not. intended or otherwise. there may be no more to the force than there is to the meaning. or as to how the words spoken were used. as a waming) by the audience to which it is addressed. by me ans of an utterance. 115-16). Austin gives many examples and lists of words which help us to form at least a fair intuitive notion of what is meant by 'illocutionary force' and 'illocutionary act'. in the first person present indicative. Besides these. it would admit many kinds of acts which he wishes to exclude from the class (p.

It is the general doctrine I listed last above--the doctrine that an utterance's having such and such a force is a matter of convention-that I shall take as the starting point of inquiry. two possible sources of confusion. where it is exhausted by meaning. yet. Why does Austin say that the illocutionary act is a conventional act. and neutralize. (It may seem an excess of precaution to do so. I apologize to those who find it so. The remark has a certain authority in that it is the first explicit statement of the conventional nature of the illocutionary act. of the use of language with a certain illocutionary force. It is not just as being a speech act that an illocutionary actfor example. 105. 127) and speaks also of 'conventions of illocutionary force' (p. 103). 121. 108. Indeed. the fact that an utterance has the further unexhausted force it has is also a matter of convention. Usually this doctrine is affirmed in a quite unqualified way. such acts must be conventional nonverbal acts. Austin would say. 115. to linguistic conventions to determine what locutionary act has been performed in the making of an utterance. I shall refer to it later. an act done as conforming to a convention? I must first mention. we may agree (or not dispute) that any speech act is. his general doctrines regarding those notions. 11 I shall assurne that we are clear enough about the intended application of Austin's notions of illocutionary force and illocutionary act to be able to criticize. The performance of any speech act involves at least the observance or exploitation of some linguistic conventions. that 'it may . Second. is in no way a conventional act (pp. Austin says. verbal-act of warning is conventionally such. Austin reverts many times to the 'conventional' nature ofthe illocutionary act (pp. in order to be properly called by these names. conventionally such in just the same way as an illocutionary---that is. 114). or. Austin maintains.. We must refer. is a matter of convention. as such. and every illocutionary act is a speech act. he remarks (pp. without the use of words. acts of warning--<:an be brought off nonverbally.INTENTION AND CONVENTION IN SPEECH ACTS 25 effects. though it too can often be ascribed as an act to the speaker (his perlocutionary act). the fact that it is. be said to be conventional in the sense that at least it could be made explicit by the performative formula' (p. by reference to cases. But it is absolutely clear that this is not the point that Austin is making in dec\aring the illocutionary act to be a conventional act. of warning-is conventional. to determine what the meaning of the utterance iso The doctrine now before us is the further doctrine that where force is not exhausted by meaning. 120. But just once there occurs an interestingly qualified statement of it.) First.. at least in part a conventional act. A nonverbal act of warning is. we must dismiss as irrelevant the fact that it can properly be said to be a . 103. Meanwhile let us consider the doctrine in its unqualified form. 120-]) that though acts which can properly be called by the same names as illocutionary acts-for example. 120-1).

although the circumstances of utterance are always relevant to the determination of the illocutionary force of an utterance. Similarly. performed by uttering the words 'This is Mr. where there clearly exist statable conventions. relating to the circumstances of utterance. that there are many cases in which the illocutionary force of an utterance. Thus the fact that the word 'guilty' is pronounced by the foreman of the jury in court at the proper moment constitutes his utterance as the act of bringing in a verdict. For if this were held to be a ground for saying that illocutionary acts were conventional acts. F or very many kinds of human transaction involving speech are govemed and in part constituted by what we easily recognize as established conventions of procedure additional to the conventions governing the meanings of our utterances. The act of surrendering. may be said to be an act performed as conforming to a convention. such that an utterance with a certain meaning. there are many cases in which it is not as conforrning to an accepted convention of any kind (other than those linguistic conventions which help to fix the meaning of the utterance) that an illocutionary act is performed. pronounced by the appropriate person in the appropriate circumstances. and that this is so is certainly a matter of the conventional procedures of the law. and there are doubtless many more which could be given. has' the force it has as conforming to those conventions. Surely there may be cases in which to utter the words 'The ice over there is very thin' to a skater is to issue a warning (is to say something with the force of a warning) without its being the case that there is any statable convention at all (other than those which bear on the nature of the locutionary act) such that the speaker's act can be said to be an act done as conforming to that convention. as correctly described. he thereby performs the act of giving the man out. warning is correctly called by this name. We can readily imagine circumstances in which an utterance of the words 'Don't go' would be correctly described not as arequest or an . then any describable act whatever would. Smith'. that is. Examples of illocutionary acts of which this is true can be found not only in the sphere of social institutions which have a legal point (like the marriage ceremony and the law courts themselves) or of activities govemed by a definite set of rules (like cricket and games generally) but in many other relations ofhuman lire. be a conventional act.26 INTENTION AND CONVENTION IN SPEECH ACTS matter of convention that an act of. may be said to be (to have become) an act performed as conforming to an accepted convention. which no player or spectator shouting 'Out!' can do. a conventional act. The act of introducing. performed by saying 'Kamerad!' and throwing up your arms when confronted with a bayonet. is not owed to any conventions other than those which help to give it its meaning. It seems clear. Austin gives other examples. But it seems equally clear that. for example. The contention that illocutionary force is a matter of convention is easily seen to be correct in a great number of cases. Here is another example. it is a matter of convention that if the appropriate umpire pronounces a batsman 'out'. though not exhausted by its meaning.

other than those linguistic conventions which help to fix the meanings of the utterances. But although this is a proper comment on the remark. the doctrine of the conventional nature of the iIlocutionary act does not hold genen·lly. others are not (except in so far as they are locutionary acts). kneel down. but as an entreaty. 103). we may presume it worth our while to look for the reason for using it. advising. 'may be said to be conventional in the sense that at least it could be made explicit by the performative formula' (p. Y's utterance has the force of an objection to X's assertion (or proposaI) that p. which it must be our business to discover. no doubt--relating to X's situation. any convention involved. manner. Even though we may decide that the description 'conventional' is not appropriately used. There are questions here·which we must discuss later. there does not have to be. or the use of a sentence with a certain illocutionary force. It is much more likely that he is moved by some further. attitude to Y. that if this is a sense of anything to the purpose. for example. Whatever it is that leads Austin to call illocutionary acts in general 'conventional' must be cIosely connected with whatever it is about such acts as warning.INTENTION AND CONVENTION IN SPEECH ACTS 27 order. feature of illocutionary acts. 1 do not think it necessary to give further examples. X says (or proposes) that p and Y objects that q. and with justice. a debate on policy) one speaker raises an objection to what the previous speaker has just said. we should not simply dismiss the remark with this comment. and fundamental. raise one's arms and say. 'I entreat you. that accounts for the fact that they at least couldbe made explicit by the use . for that matter. But to suppose that there is always and necessarily a convention conformed to would be like supposing that there could be no love affairs which did not proceed on lines laid down in the Roman de la Rose or that every dispute between men must follow the pattern specified in Touchstone's speech about the countercheck quarrelsome and the lie direct. in Y's view of these things. in the bearing which he takes the proposition that q to have on the doctrine (or proposal) that p. It seems perfectIy cIear that. I do not want to deny that there may be conventional postures or procedures for entreating: one can. Why then does Austin repeatedly affirm the contrary? It is unlikely that he has made the simple mistake of generalizing from some cases to all. But where is the convention that constitutes it an objection? That Y's utterance has the force of an objection may lie partly in the character of the dispute and of X's contention (or proposal) and it certainly lies partly. But although there may be.' But 1 do want to deny that an act of entreaty can be performed only as conforming to some such conventions. What makes X's words to Y an entreaty not to go is something-complex enough. entreating. it is a sense of'being capable of being conventional'. In the course of a philosophical discussion (or. apologizing. On this we may first. be incIined to comment that there is no such sense of 'being conventional'. Here we may recall that oddly qualified remark that the performance of an iIlocutionary act. Some illocutionary acts are conventional. and current intention. if at least we take the expressions 'convention' and 'conventional' in the most natural way. Another example.

111 I think a concept we may find helpful at this point is one introduced by H. as Austin points out. So we must ask what it is about them that accounts for this fact. so he satisfies condition (i J He arranges convincing-looking 'evidence' that p. P. Obviously it will not do to answer simply that they are acts which can be performed by the use of words. for which. 1957). to cases where that by which someone nonnaturally means something is a linguistie utterance. or a part of his reason. S. to communicate with another. So we need some further explanation. there is no corresponding first-person peiformative formula. Grice's analysis is undoubtedly offered as an analysis of a situation in which one person is trying. it would not be the case that S could be said to be trying to communicate by means of his production of x with the person A in whom he was trying to produce the response r. in a sense of the word 'communicate' fundamental to any theory of meaning.) that this recognition on the part of A of S's intention (i) shall function as A 's reason. S nonnaturally means something by an utterance x if S intends (i \) to produce by uttering x a certain response (r) in an audience A and intends (i2) that A shall recognize S's intention (i\) and intends (i. an important feature of this definition that the securing of the response r is intended to be mediated by the securing of another (and always cognitive) efTect in A. like convincing. in this important sense of'communicate'. dissuading. It is intended to cover cognitive and afTective states or attitudes as weil as actions. but realizes. and indeed intends. But I think a little reflection shows that it is not quite complex enough for his purpose. He realizes that A will not take the arranged 'evidence' as genuine or natural evidence that p. evidently. namely. but knowing also that A does not know that S knows that A is wate hing him at work. alarming. But it will be convenient to refer to that by which someone. namely.) It is. for his response r. and amusing. in a place where A is bound to see it. though more convenient in some ways than Grice's 'efTect'. is not ideal. He does this. The explanation of the introduced concept is given in terms of the concept of intention. recognition of S's intention to sec ure response r. But it is possible to imagine a situation in which Grice's three conditions would be satisfied by a person Sand yet. that A will take his arranging of it as grounds for thinking .28 INTENTION AND CONVENTION IN SPEECH ACTS of the corresponding first-person perfonnative fonn. Grice in his valuable article on Meaning (PhiIosophieal Review. It is of more general application. the concept of someone's nonnaturally meaning something byan utteranee. nonnaturally means something as S's utteranee. So are many (perlocutionary) acts. knowing that A is watching him at work. LXVII. The concept does not apply only to speech acts-that is. S intends by a certain action to induce in A the belief that p. (The word 'response'. I proceed to describe such a situation. Grice's analysis of his concept is fairly complex.

He knows that A has general grounds for thinking that S would not wish to make hirn. But again we may be content for the moment with necessary conditions. S. as in our ex am pie. then. So S satisfies condition (i2). S. I suggest. So he satisfies condition (i. But unless S at least brings it about that A takes him (S) to be trying to let him (A) know something. he has not even tried to bring this about. it is necessary that A should take S to have the complex intention of the (i 2) form which S does have. satisfies all Grice's conditions. but he will not take S as trying. A will indeed take S to be trying to bring it about that A is aware of some fact. 1 shall content myself. It seems a minimum further condition of his trying to do this that he should not only intend A to recognize his intention to get A to think that p. with drawing from the actually considered example the conclusion that we must add to Grice's conditions the further condition that S should have the further intention (i 4) that A should recognize his intention (i J. then he has not even tried to communicate with A. intends to induce in A the belief that p. that for A (in the appropriate sense of 'understand') to understand something by utterance x. if A is to understand the utterance correctly. Grice in fact gives no such account. but that he should also intend A to recognize his intention to get A to recognize his intention to get A to think that p. In other words. But 1 shall rest content for the moment with the fact that this addition at least is necessary. however. 1 say'at least partially' because the uncertainty as to the sufficiency of even the modified conditions for S's nonnaturally meaning something by an utterance x is refiected in a corresponding uncertainty in the sufficiency of conditions for A 's understanding. and if. then. in the colloquial sense. And he intends that A's recognition ofhis intention (il) should function injust this way. and hence that A's recognition of his (Ss) intention to induce in A the belief that p will in fact seem to A a sufficient reason for believing that p. It is possible that further argument could be produced to show that even adding this condition is not suflicient to constitute the case as one of attempted communication.INTENTION AND CONVENTION IN SPEECH ACTS 29 that he. and 1 shall suggest a way of at least partially supplying this lack. which A takes S to have.). an account complementary to the account of what it is for S to mean something by an utterance x. and that for A to understand the utterance correctly. to 'let him know' something (or to 'tell' hirn something). A. Now we might have expected in Grice's paper an account of what it is for A to understand something by an utterance x. he intends A to recognize his (i I) intention. it is necessary (and perhaps sufficient) that there should be some complex intention of the (i 2) form. But this is clearly not a case of attempted communication in the sense which (I think it is fair to assurne) Grice is seeking to elucidate. We might approximate more closely to the communication situation if we changed the example by supposing it not only clear to both A and S that A was watching S at work. described above. That is. but also clear to them both that it was clear to them both. think that p unless it were known to S to be the case that p. S'S (iJ intention and hence his (i 2) intention . he has not succeeded in communicating with A .

30 INTENTION AND CONVENTION IN SPEECH ACTS must be fulfilled. the identification is equivalent to a tentative (lmd partial) analysis of the notion of uptake and hence of the notions of illocutionary act and illocutionary force. It is at this point. too. if not the achievement. and securing understanding of illocutionary force is said by Austin to be an essential element in bringing off the illocutionary act. of securing uptake is essentially a standard. IV Let us. To this. actually have made such and such a bequest. then. in due form. it seems.1 4 . If we do find such a point of connection. make a tentative identification-to be subsequently qualified and revised --of Austin's notion of uptake with that at least partially analysed notion ofunderstanding (on the part of an audience) which I introducedjust now as complementary to Grice's concept of somebody nonnaturally meaning something by an utterance. There are two points at which we shall apply this test. there is an objection. that we may hope to find a possible poinl of connection with Austin's terminology of 'securing uptake'.) intention is fulfilled. Might not a man really have made a gift. We may be tempted to say instead that at least the aim. then it would follow that to say something with a certain illocutionary force is at least (in the standard case) to have a certain complex intention ofthe (iJ form described in setting out and modifying Grice's doctrine. Next we test the adequacy and explanatory power of this partial analysis by seeing how far it helps to explain other features of Austin's d. See the example about redoubling at bridge.octrine regarding illocutionary acts. if not the achievement. One is the point at which Austin maintains that the production of an utterance with a certain illocutionary I owe the objections which follow to Professor Hart. if not an invariable. for example. Since the notion of audience understanding is introduced by way of a fuller (though partial) analysis than any which Austin gives of the notion of uptake. that the aim. For to secure uptake is to secure understanding of (meaning and) illocutionary force. that his (i . consequently. Of course it does not follow from the fulfilment of these intentions that his (i. and take some satisfaction in the thought.) intention is fulfilled. . we also find a possible starting point for an at least partial analysis of the notions of illocutionary force and of the illocutionary act. For an illocutionary act may be performed altoKether unintentionally. element in the performance of the illocutionary act. or gift. p. It is true that this doctrine of Austin's may be objected to. 36 below. 3 For sure1y a man may. even if no one ever reads his will or instrument of gift. even if he had no expectations of the fact ever being known? But this objection at most forces' on us an amendment to which we are in any case obliged 4 : namely. So the analysis of the aim of securing uptake remains an essential element in the analysis of the notion of the illocutionary act. of securing uptake is an essential element in the performance ofthe illociItionary act. If the identification were correct. nor.

it consists of. available sometimes. inseparable from the nature of his act. For using such phrases. then. for the enterprise to be possible at all. in saying what we say. These points are closely connected with each other. and the intention that this recognition should occur being itself intended to be recognized. which may or may not have the form of a self-ascription.INTENTION AND CONVENTION IN SPEECH ACTS 31 force is a conventional act in that unconventional sense of 'conventional' which he glosses in terms of general suitabiIity for being made explicit with the help of an explicitly performative formula. would be to attach. or he must find. First. the character of comments on utterances other than themselves-to the explicit performative formula the . and we may see whether the suggested analysis helps to explain the exclusion from the class of illocutionary acts of those acts falling under this characterization which Austin wishes to exclude. or again 'That was a warning' or 'I'm warning you'. point about intention. From such phrases as these-which have. up to hirn. me ans of making the intention dear. One such means. The speaker. For he will not have secured understanding of the illocutionary force of his utterance. there must exist. a general. first. He remarks on the unsatisfactoriness of this characterization in that it would admit as illocutionary acts what are not such. The other is the point at which Austin considers the possibility of a general characterization of the illocutionary act as what we do. those means. I repeat. and then a special. then. Thus we have phrases like 'This is only a suggestion' or 'I'm only making a suggestion'. he will not have performed the act of communication he sets out to perform. The first point may be roughly expressed by saying that in general a man can speak of his intention in performing an action with a kind of authority which he cannot command in predicting its outcome. in appearance. for making that intention dear. If there exists any conventional linguistic means of doing so. or not only. We are concerned with the case in which there is not simply an intention to produce a certain response in an audience. the speaker has the authority that anyone has to speak on the subject of his intentions and the motive that I have tried to show is inseparable from an act of communication. but with a very special kind of case. which comes very elose to the employment of the explicit performative form. What he intends in doing something is up to him in a way in which the results of his doing it are not. But we are concerned not with just any intention to produce any kind of effect by acting. to the substance of the message what looks like a forcecelucidating comment on it. and a motive for using. this recognition to serve as part of the reason that the audience has for its response. the speaker has both a right to use. not only has the general authority on the subject of his intention that any agent has. Now clearly. unless his complex intention is grasped. or subjoin. he also has a motive. but an intention to produce that response by means of recognition on the part of the audience of the intention to produce that response. The explanation of this feature of iIIocutionary acts has two phases . we take the point about the general suitability of an illocutionary act for performance with the he1p ofthe explicitly performative formula for that act.

/lot as a case in which we have two utterances. but a single utterance in which the first-person performative verb manifest(r has that peculiar logical character of which Austin rightly made so much. one a comment on the other. but satisfactorily explained by the analysis under consideration. But it is no part of the intention to secure the effect hy means 01 the recognition of the intention to sec ure it. for example. When that short step is taken. but which are not illocutionary acts. but as a case of a single unitary speech act. My reason for qualijring the re mark that such phrases have the character of comments on utterances other than themselves is this. That is. It may be objected that the deduction fails to show that the intentions rendered explicit by the use of performative formulae in general must be of just the complex form described. we shall make a further application of the analysis at the second testing point I mentioned. to make explicit the type of communication intention with which the speaker speaks. the addition of the quasi-comment 'That was a warning' is part of the total act of warning. indeed. and hence fails to justify the claim that just this kind of intention lies at the core of all i11ocutionary acts. Among the things mentioned by Austin which we might be doing in saying things. I shall consider the two examples of (1) sho\\'ing off and (2) insinuating. even in appearance. Since it is part ofthe speaker's audience-directed intention to make c\ear the character of his utterance as. two utterances. The effect of the short step to the explicitly performative formula is simply to bring appearances into line with reality. On the contrary: recognition of the intention rnight militate against securing the effect and promote an opposite effect. in Austin's phrase. and which we may express in the present context by saying that the verb serves not exactly to ascribe an intention to the speaker but rather. we try to impress. one commenting on the other. to evoke the response of adrniration. a warning. we shall see what power it has to explain why some of the things we may be dOing. But before discussing why. we no longer have. It is no part of our total intention to secure recognition of the intention to produce the effect at a11. Crudely. we are certainly trying to produce an effect on the audience: we talk. disgust. appearances not\\'ithstanding. In saying to an audience what we do say. The above rnight be said to be a deduction of the general possibility and utility of the explicitly performative formula for the cases of illocutionary acts not essentia11y conventional. We are considering the case in which the subjoined quasi-comment is addressed to the same audience as the utterance on which it is a quasi-comment. Now when we show off.32 INTENTION AND CONVENTION IN SPEECH ACTS step is only a short one. This leads on to a further general point not explicitly considered by Austin. And indeed we sha11 see that this claim would be mistaken. are not illocutionary acts and could not be rendered explicit by the use of the performative formula. the type of force which the utterance has. in saying what we say. for example. and since the subjoined quasi-comment directly subserves this intention. it is better to view the case. we very often intend not only to produce the primary response r by . for effect.

in one respect. We have no proof. F. We have noticed already how we had to meet the threat of a counterexample to Grice's analysis of the communicative act in terms of three types of intention--(i I)' (i 2)' and (i. the production in the audience of belief that p is intended to be itself the means whereby his attitude or conduct is to be inftuenced. The difference (one of the differences) between showing off and warning is that your recognition of my intention to put you on your guard may weil contribute to putting you on your guard. we aim at a complex primary response. Although it is true that. why not? If we take the first question strict1y as introduced and posed. the answer to it is 'No'. I owe this point to Mr. for it is no part of our intention that the further effect should be produced by way of recognition of our intention that it should be. Does this part of what I am doing in saying what I say contribute to determining the character of the i1locutionary act I perform ? And if not. We secure uptake. aiming to produce in you the primary cognitive response of knowledge or belief that p) may be to bring it about thereby that you adopt a certain line of conduct or a certain attitude. one might say.f!ronlery. But then I am not merely trying to impress you. to be a type of illocutionary act. but to produce further effects by means of the production of the primary response r. whereas your recognition of my intention to impress you is not likely to contribute to my impressing you (or not in the way I intended). I am inviting you to be impressed. B. but the intention to secure that you are on your guard against p-perils. They have. For I might try to impress you with my e. Thus my further purpose in informing you that p (that is. and so forth. are not only the intention to sec ure your belief that p.)-by the addition of a further intention (i4) that an intention (i2) should be recognized. whether we succeed in producing the effect or not-yet this does not enter into the characterization of the illocutionary act. Thus in the case where I do not simply inform. that the 5 Perhaps trying to impress might sometimes have an ilIocutionary character. The reason tor the answer follows from the analysis. part of what I am doing is trying to influence your attitudes or conduct in a certain way. intending you to recognize this intention and intending your recognition of it to function as part of your reason for being impressed. With this case we have to contrast the case in which. a logically embarrassing feature. in saying what we say. we are in fact tryingto produce the further effect-this is part of wh at we are doing. for a different reason. then. . perform the act of communication that we set out to perform. you that p. We have no complex intention (i4) that there should be recognition of an intention (i) that there should be recognition of an intention (i I) that the further effect should be produced. essential avowability. This is. McGuinness.5 Insinuating fails. the latter to be secured through the former alone. if the audience understands us as injorming hirn that p. An essential feature of the intentions which make up the iIIocutionary complex is their overtness. In saying what I say. but warn.INTENTION AND CONVENTION IN SPEECH ACTS 33 me ans of audience recognition of the intention to produce that response. instead of aiming at a primary response and a further effect. among the intentions I intend you to recognize (and intend you to recognize as intended to be recognized). however.

or hope.34 INTENTION AND CONVENTION IN SPEECH ACTS resulting enlarged set of conditions is a complete analysis. inter alia. The intention one has in insinuating is essentially nonavowable. mere1y requires supplementing and enriching. for different reasons. . certain explanatory merits. At the same time it enables us easily to dispose of insinuating as a candidate for the status of a type of illocutionary act. and the way seems open to a regressive series of intentions that intentions should be recognized. in terms of that scheme. That is why I speak of the feature in question in these terms. A wish that someone should stay may be held in different ways: passionately or lightly. be part of a speaker's intention to see ure recognition of how he holds it. Now let us take stock a little. in issuing his utterance. it does suggest that the complete and rounded-off set of conditions aimed at in a conventional analysis is not easily and certainly attainable in these terms. Let us look at some reasons why not. So how. but through audience recognition of a complex attitude of which this primary intention forms an integral part. for example. to induce or disclose a certain belief. the intention. but not more than suspect. is a matter of trying to secure the primary response not merely through audience recognition of the intention to secure it. Ingenuity might show it was not. in the case of entreaty. The whole point of insinuating is that the audience is to suspect. While I do not think there is anything necessarily objectionable in this. either of arequest or of an entreaty. it seems. for example. and it may. to have a complex intention of a certain kind. are we to account for the variation in illocutionary force between requests and entreaties? This question does not appear to raise a major difficulty for the scheme. v I remarked earlier that the words 'Don't go' may have the force. But the only other intentions mentioned in our scheme of partial analysis relate directly or indirectly to recognition of the primary intention. Nevertheless we cannot claim general application for it as even a partial analysis of the notions of illocutionary for~e and illocutionary act. In either case the primary intention of the utterance (if we presume the words to be uttered with the sense 'Don't go away') is that ofinducing the person addressed to stay where he iso His staying where he is is the primary response aimed at. The most obvious reason. is the belief. Entreaty. The suggestion has. The scheme. We tentative1y laid it down as a necessary condition of securing understanding of the illocutionary force of an utterance that the speaker should succeed in bringing it about that the audience took hirn. name1y the intention that the audience should recognize (and recognize as intended to be recognized) his intention to induce a certain response in the audience. confidently or desperately. as we have just seen. That is why I speak of the feature in question as logically embarrassing. that such a revelation is more like1y to seeure the fulfilment of the primary intention.

retiring to the pavilion) from a certain audience (say. Does this mean that the approach I suggested to the elucidation of the notion of illocutionary force is entirely mistaken? I do not think so. and then see what. the jurymen's to secure a certain response (say. in other than a formal sense. a particular audience to whom the utterance can be said to be addressed. We can say that a man who issues an order typically intends his utterance to secure a certain response. Rather . Can we say that the umpire's primary intention is to secure a certain response (say. the judge). On the other. What we initially take from Grice--with modifications--is an at least partially analytical account of an act of communication. unlike the case of entreaty. the strain becomes too much for the scheme to bear. the illocutionary act itself . On the one hand. one of its basic features-namely.INTENTION AND CONVENTION IN SPEECH ACTS 35 But one may not only request and entreat. we must distinguish types of case. we can accommodate it. though not so simply. the batsman). Consider an umpire giving a batsman out. among all the participants in the procedure (trial. in the case of the order? Not with plausibility. and its recognition to be a reason for the response. and then build the rest of our account around this. and in this understanding the notion of wholly overt or essentially avowable intention plays an essential part. with some strain. in these cases. be so extended. a jury bringing in a verdict of guilty. the reference to social conventions of procedure assurnes a very much greater importance. an act which might indeed be performed nonverbally and yet exhibit all the essential characteristics of a (nonverbal) equivalent of an illocutionary act. Evidently. Can we so simply accommodate in our scheme this variation in illocutionary force? Weil. a priest or a civil officer pronouncing a couple man and wife. game) to which the utterance belongs. a player redoubling at bridge. as we did. that he intends this intention too to be recognized. and finally that he intends the recognition of these last features to function as an element in the reasons for the response on the part of the audience. It can. to isolate. We gain more than this. a judge pronouncing sentence. the pronouncing of sentence) from a certain audience (say. in this case. if anything. with some strain. For the account enables us to understand how such an act may be linguistically conventionalized right up to the point at which i1locutionary force is exhausted by meaning (in Austin's sense). But as we move further into the region of institutionalized procedures. one mayorder someone to stay where he iso The words 'Don't go' may have the illocutionary force of an order. that he intends this intention to be recognized. the reference to an intention to sec ure a definite response in an audience (over and above the seeuring of uptake)---has to be dropped. marriage. is common to the types we have distinguished. It is not even possible. Evidently. that he intends the utterance to be recognized as issued in a certain social context such that certain social rules or conventions apply to the issuing of utterances in this context and such that certain consequences may follow in the event of the primary response not being secured. the scheme has to be extended to make room for explicit reference to social convention.

but if the circumstances are appropriate and the play strict. and are trying to play the game and not wreck it. Hence the speaker's utterance is not only intended to further. in the absence of appropriale intention. Forms can take charge. but when they do. The likeness resides in the fact that. First. The examples I mentioned just now will serve-marrying. Let us take the standard case in which the participants in these procedures know the rules and their roles. A player might let slip the word 'redouble' without meaning to redouble. But a player who continually did this sort of thing would not be asked to play again. redoubling. At the other end of the scale-the end. the extent to which the act is one done as conforming to conventions may depend solelyon the extent to which conventional linguistic meaning exhausts iIlocutionary force. the case is essentially deviant or nonstandard. We have. it cannot fail to do so. the practice in question in a certain conventional way. perform an illocutionary act which forms part of. only in so far as the means used to perfonn it are conventional.to a convention-governed practice or procedure. done as conforming to a convention. and sometimes they have to make adecision within a restricted range of alternatives (for example. The difTerence is a more complicated affair. from which Austin beg anwe have iIIocutionary acts which are essentially conventional. there is an important likeness and an important difTerence. Between the case of such acts as these and the case of the iIIocutionary act not essentially conventional. if things go in accordance with the rules of the procedure in question. the act is identified as the act it is just because it is performed by the utterance of a form of words conventional for the performance of that act. the practice in question in some one of the alternative ways open. and intended to be recognized as so intended. Second. Such acts could have no existence outside the rule. we may say. the practice or procedure as a wh oIe . that is to say. To speak only of those conventional means which are also linguistic means. an act which is conventional in two connected ways. to pass or redouble. to pr on ounce sentence of imprisonment for some period not exceeding a certain limit). or afTect the course of. or furthers. an act done as falling under the rules. except by sharpers. the act of furthering the practice in the way intended is an act required or permitted by those rules. There is present in the standard case. .36 INTENTION AND CONVENTION IN SPEECH ACTS is not essentially a conventional act. pronouncing sentence. in these cases. or afTect the course of. I do not mean that such an act could never be performed unintentionally. in the case of an utterance belonging . an act done as conforming to a convention. bringing in a verdict.or convention-governed practices and procedures of which they essentially form parts. giving out. the speaker's utterance is standardly intended to further. it may be that the act is conventional. the same element of wholly overt and avowable intention as in the case of the act not essentially conventional. then he has redoubled (or he may be held to have redoubled). or may. Then they are presented with occasions on which they have to. in the absence of any breach of the conventional conditions for furthering the procedure in this way.

be frustrated. therefore. action. without any breach ofrules or conventions. the act of communication is performed if uptake is secured. as Austin perhaps shows some tendency to do when he both insists that the illocutionary act is essentially a conventional act and connects this claim with the possibility of making the act explicit by the use of the performative formula. But even though the act of communication is performed. It is different with the utterance which forms part of a wholly convention-governed procedure. or attitude) may simply not be forthcoming. in the speech act itself. It would be amistake to take them as the model for understanding the notion of iIIocutionary force in general. Whether it is effective or not is something that rests with his audience. to generalize the account of illocutionary force derived from Grice's analysis. For there are no conditions which can conventionally guarantee the effectiveness of his overt intention. for this would involve holding. But they do not form the whole nor. speakers assurne the responsibility for making their intentions overt. the most fundamental part. But of course--and I shall recur to this thought-the sharp contrast I have here drawn between two extreme types of case must not blind us to the existence of intermediate types. we may say. In both cases. But in the other case the speaker cannot. It would equally be a mistake. The speaker who abides by the conventions can avowably have the intention to further the procedure in the way to which his current linguistic act is conventionally appropriated only if he takes it that the conventional conditions for so furthering it are satisfied and hence takes it that his utterance will not only reveal his intentions but give them effect. Acts belonging to convention-constituted procedures of the kind I have just referred to form an important part of human communication.INTENTION AND CONVENTION IN SPEECH ACTS 37 And here we have the contrast between the two types of case. In one case (the case of the convention-constituted procedure) the speaker who uses the explicitly performative form also explicitly assurnes the responsibility for making his overt intention effective. the explicitly performative form may be the name of the very act which is performed if and only if the speaker's overt intention is effective. as we have seen. the wholly overt intention which lies at the core of the intention complex may. then any frustration of the wholly overt intention of the utterance (the intention to further the procedure in a certain way) must be attributable to a breach of rule or convention. There is nothing parallel to this in the case of the illocutionary act of a kind not essentially conventional. we may think. but in the other case it cannot be the name of this act. In the one case. Nevertheless. we can perhaps extract from our consideration of two contrasting types of . falsely. In the case of an illocutionary act of a kind not essentiaUy conventional. that the complex overt intention manifested in any illocutionary act always includes the intention to secure a certain definite response or reaction in an audience over and above that which is necessarily secured if the illocutionary force of the utterance is understood. The audience response (belief. explicitly assurne any such responsibility. Granted that uptake is secured. if the utterance is taken to be issued with the complex overt intention with which it is issued.

his instructions followed. or even advice. he may be seen as the mouthpiece. and it would certainly be wrong to suppose that all cases fall cIearly and neatly into one or another of these two cIasses. instructions. . It is perhaps this fact which lies at the base of the general possibility of the explicit performaiive formula. in some cases. mereIy. Once this common element in all iIIocutionary acts is cIear.38 INTENTION AND CONVENTION IN SPEECH ACTS case something which is common to them both and to all the other types which lie between them. For the illocutionary force of an utterance is essentially something that is intended to be understood. to hirn. His wholly overt intention mayamount to no more than that of making available-in a 'take it or leave it' spirit-4o his audience the information or instructions or opinion in question . that different types may be exemplified by one and the same utterance. of another agency to which may be attributed at least general intentions of the kind that can scarcely be attributed. a game) in a definite way provided for by the conventions or rules ofthe practice and those cases in which the overt intention incIudes that of securing a definite response (cognitive or practical) in an audience over and above that which is necessarily secured if uptake is secured. as intended to be organized. We should not find such complications discouraging. and yet be overtly indifferent as to whether or not his information is accepted as such. in the particular case. A speaker whose job it is to do so may offer information. as we have seen. And the understanding of the force of an utterance in all cases involves recognizing what may be called broadly an audience-directed intention and recognizing it as wholly overt. I have set in sharp contrast those cases in which the overt intention is simply to forward adefinite and convention-governed practice (for ex am pIe. or his advice taken. also. for we can scarcely expect a general account of linguistic communication to yield more than 'schematic outlines. though. But there is something misleading about the sharpness of this contrast. though again. we can readily acknowledge that the types of audience-directed intention involved may be very various and. extra factors come importantly into play in the case of convention-constituted procedures. which may almost be lost to view when every qualification is added which fideIity to the facts requires.

Some of the English verbs and verb phrases associated with illocutionary acts are: state. and an utterance by the speaker. the symbol or ward or sentence. so the paper might have been caUed 'What is an Illocutionary Act?' I do not attempt to define the expression 'illocutionary act'. command.III WHAT IS A SPEECH ACT? J.grnell University Press and George Allen & Unwin Ud. How To Do Things with Words (Oxford. The members of this last c1ass are what Austin 1 called illocutionary acts and it is with this c1ass that I shall be concerned in this paper. request. approve. and express regret. although if my analysis of a particular illocutionary act succeeds it may provide the basis for adefinition. we1come. I think it is essential to any specimen of linguistic communication that it involve a linguistic act. giving reports. Austin. and he will also have performed acts witrun the c1ass which inc1udes making statements. It is not. I J. Max Black (Allen & Unwin. Reprinted by pennission of the author. remark. L. criticize. promise. R. asking questions. or even the token ofthe symbol or word or sentence. (. wruch is the unit of linguistic communication. pp. assert. there are many kinds of acts associated with the speaker's utterance. 1965). apologize. perhaps I can say why I trunk it is of interest and importance in the philosophy of language to study speech acts. ed. The speakef will characteristicaUy have moved his jaw and tongue and made noises. a hearer. From Philosophy in AmericQ. SEARLE I. INTRODUCTION IN a typical speech situation involving a speaker. 1962). censure. the production of the sentence token under certain conditions is the illocutionary act. as they are sometimes called. Austin c1aimed that there were over a thousand such expressions in English. In addition. as has generally been supposed. 221-39. he will further characteristically have perfarmed some acts within the c1ass which inc1udes referring to Kennedy or Khruschchev or the North Pole. issuing commands. or. To put trus point more precisely. and warning. By way of introduction. . but rather it is the production of the token in the performance of the speech act that constitutes the basic unit of linguistic communication. and the illocutionary act is the minimal unit of linguistic communication. greeting. language acts or linguistic acts. describe. express approval. comment. warn. order. he will characteristically have performed some acts within the c1ass which inc1udes informing or irritating or boring rus hearers.

To constme them under the category of linguistic communication necessarily involves construing their production as speech acts. 11. even so. a waterfall. But in order to set the stage for actually stating conditions and extracting mIes for performing an illocutionary act I have to discuss three other preliminary notions : ru/es. If we were certain the marks were a consequence of. I intend therefore to explicate the notion of an illocutionary act by stating a set of necessary and sufficient conditions for the performance of a particular kind of illocutionary act. for example. I shall confine my discussion of these notions to those aspects which are essential to my main purposes in this paper. that will provide us with a pattern for analysing other kinds of acts and consequently for explicating the notion in general. like a stone. propositions. If I am successful in stating the conditions and the corresponding rules for even one kind of illocutionary act. water erosion. but. if it were to be at all complete. RULES In recent years there has been in the philosophy of language considerable discussion involving the notion of mies for the use of expressions. however. Some philosophers have even said that knowing the meaning of the word is simply a matter of knowing . would require a paper for each. He cannot just regard it as a natural phenJmenon.40 WHAT IS A SPEECH ACT? I do not know how to prove that linguistic communication essentially involves acts but I can think of arguments with which one might attempt to convince someone who was sceptical. of current attempts to decipher the Mayan hieroglyphs that we at least hypothesize that the marks we see on the stones were produced by beings more or less like ourselves and produced with certain kinds of intentions. One argument would be to call the sceptic's attention to the fact that when he takes a noise or a mark on paper to be an instance of linguistic communication. one of the things that is involved in his so taking that noise or mark is that he should regard it as having been produced by a being with certain intentions. and meaning. It is a logical presupposition. sometimes it may be worth sacrificing thoroughness for the sake of scope and I shall therefore be very brief. as a message. In order to regard it as an instance of linguistic communication one must suppose that its production is what I am calling a speech act. what I wish to say concerning each of these notions. and extracting from it a set of semantical mies for the use ofthe expression (or syntactic device) which marks the utterance as an illocutionary act of that kind. or a tree. I shall argue that such things as asking questions or making statements are mlegoverned in ways quite similar to those in which getting a base hit in baseball or moving a knight in chess are rule-governed forms of acts. then the question of deciphering them or even calling them hieroglyphs could not arise. say. To perform illocutionary acts is to engage in a mle-governed form of behaviour.

One disquieting feature of such discussions is that no philosopher. . sometimes as an analytic truth. 'Two Concepts of Rules'. for what the 'rule' seems to offer is a partial definition of'checkmate' or 'touchdown'. 1964. That. Constitutive rules constitute (and also regulate) an activity the existence of which is logically dependent on the rules. for example. an activity whose existence is logically independent of the existence of the rules. Notice that they are almost tautological in character. e.g. If our paradigms of rules are imperative regulative rules. If meaning is a matter of rules of use. fOT example. Regulative rules regulate a pre-existing activity. dismayed perhaps by the failure of their colleagues to produce any rules. I distinguish between two sorts of rules: Some regulate antecedently existing forms of behaviour. such non-imperative constitutive rules are likely to strike us as extremely curious and hardly even as rules at all. a checkmate is made if the king is attacked in such a way that no move will leave it unattacked. But.g. do not mereiy regulate the game of football but as it were create the possibility of or define that activity. 'How to Derive "Ought" from "Is"'. or 'Officers are to wear ti es at dinner'. Some rules on the other hand do not merely regulate but create or define new forms of behaviour. this quasi-tautological character is a necessary consequence of their being constitutive rules: the rules concerning touchdowns must define the notion of 'touchdown' in the same way that the rules concerning football define 'football'. surely we ought to be able to state the rules for the use of expressions in a way which would explicate the meaning of those expressions. 1955. in a way which I shall now attempt to explain. and J. 'When cutting food hold the knife in the right hand'. The activity of playing football is constituted by acting in accordance with these rules. of course. e.WHAT IS A SPEECH ACT? 41 the rules for its use or employment. and that it can be construed as a tautology is a elue to the fact that the rule in question is a constitutive one. Regulative rules generally have the form '00 X' or 2 This distinction occurs in J. a touchdown can be scored in such and such ways and counts six points can appear sometimes as a rule. 2 Regulative rules characteristically take the form of or can be paraphrased as imperatives. football has no existence apart from these rules. the rules of etiquette regulate interpersonal reiationships. Rawls. The rules of football. Some constitutive rules take quite a different form. Searle. Certain other philosophers. R. to my knowledge at least. for example. Philosophim/ Rel'iew. Philosophh'al Rel'iew. have denied the fashionable view that meaning is a matter of rules and have asserted that there are no semantical rules of the proposed kind at all. has ever given anything like an adequate formulation of the rules for the use of even one expression. but these relationships exist independently of the rules of etiquette. I am inclined to think that this scepticism is premature and sterns from a failure to distinguish different sorts of rules. I call the latter kind of rules constitutive rules and the former kind regulative rules. a touchdown is scored when a player crosses the opponents' goal li ne in possession of the ball while play is in progress.

I will leave also. And if what I have said concerning constitutive rules is correct. One of the aims of this paper is to formulate a set of constitutive rules for a certain kind of speech act. 111.42 WHAT 1S A SPEECH ACT? 'If Y do X'. Utterances of each of these on a given occasion would characteristically be performances of different illocutionary acts. John. Indeed we shall see that the rules fall into several different categories. The first would. There are no doubt social rules of the form 'One ought not to utter obscenities at formal gatherings'.. characteristically.g. The model or paradigm of a rule which most philosophers have is that of a regulative rule. we should not be surprised if not a11 these rules take the form of imperative rules. be a question. that is. the second an assertion about the future. PROPOSITIONS Different iIIocutionary acts often have features in common with each other. Some members of the set of constitutive rules have this form but some also have the form 'X counts as Y'. If we are unable to give any satisfactory rule formulations. aprediction. e. The hypothesis that lies behind the present paper is that the semantics of a language can be regarded as a series of systems of constitutive rules and that illocutionary acts are acts performed in accordance with these sets of constitutive rules. I am inclined to think that both the failure of some philosophers to state rules for the use of expressions and the scepticism of other philosophers concerning the existence of any such rules stern at least in part from a failure to recognize the distinctions between constitutive and regulative rules. Thus. but that hardly seems a rule of the sort that is crucial in explicating the semantics of a language.3 The failure to perceive this is of some importance in philosophy. the third arequest or J The formulation 'x counts as Y' was originally suggested to me by Max Black. leave the room! Would that John left the room. Consider utterances of the following sentences : (1) Will John leave the room? (2) (3) (4) (5) John will leave the room. The effort to state the rules for an illocutionary act can also be regarded as a kind of test of the hypothesis that there are constitutive rules underlying speech acts. If John will leave the room. none of which is quite like the rules of etiquette. . our failure could be construed as partially disconfirming evidence against the hypothesis. some philosophers ask 'How can a promise create an obligation?' A similar question would be 'How can a touchdown create six points?' And as they stand both questions can only be answered by stating a rule of the form 'X counts as Y'. and if one looks in semantics for purely regulative rules one is not likely to find anything interesting from the point of view of logical analysis.

but in every case it is apart of what he does. for example. to mention only a few. or. I shall say. the fourth an expression of a wish. etc. For lack of a better word I propose to call this common content a proposition. although the illocutionary acts are different. therefore. In the sentence 'I promise to come'. Something expressible by the dause 'that John will leave the room' seems to be a common feature of all. An assertion is an illocutionary act. Sheffer. In no case is that all he does. Reichenbach and Hare. the proposition-indicating element and the function-indicating device. but a proposition is not an act at all. In one version or another this distinction is an old one and has been marked in different ways by authors as diverse as Frege. that is. that in each of these cases. Of course. . We could. and I shall describe this feature of these illocutionaty acts by saying that in the utterance of each of (1)-(5) the speaker expresses the proposition that John will leave the room. That is. 4 The function-indicating device shows how the proposition is to be taken. for a large dass of sentences used to perform illocutionary acts. The reference to some person John and predication of the same thing of him in each of these illocutionary acts inclines me to say that there is a common content in each of them. write each of these sentences in a way which would isolate this common feature: 'I assert that John willleave the room'. we can say for the purpose of our analysis that the sentence has two (not necessarily separate) parts. what illocutionary force the utterance is to have. what illocutionary act the speaker 4 In the senten ce 'I promise that I will come' the function-indicating device and the propositional element are separate. Notice also that I am distinguishing between a proposition and an assertion or statement of that proposition. Lewis. at least some of the non-illocutionary acts of reference and predication are the same. to put it in another way. although the act of expressing a proposition is apart of performing certain illocutionary acts. 'I ask whether John will leave the room'. not all illocutionary acts have a prqpositional content. an utterance of 'Hurrah!' or 'Ouch!' does not. I might summarize this by saying that I am distinguishing between the illocutionary act and the propositional content of an iUocutionary act. I do not know how sentences could perform acts of that kind. Notice that I do not say that the sentence expresses the proposition. But I shall say that in the utterance of the sentence the speaker expresses a proposition. In the utterance of each the speaker refers to a particular person John and predicates the act of leaving the room of that person. The proposition that John willleave the room is expressed in the utterance of all of (1)-(5) but only in (2) is that proposition asserted. Yet in the performance of each the speaker would characteristically perform some subsidiary acts which are common to all five illocutionary acts. with not too much distortion. which me ans the same as the first and is derived from it by certain transformations. From a semantical point of view we can distinguish between the propositional indicator in the sentence and the indicator of iUocutionary force. the two elements are not separate.WHAT IS A SPEECH ACT? 43 order. and the fifth a hypothetical expression of intention.

rules for such things as reference and prediction. and what is it for something to have a meaning? To answer the first of these questions I propose to borrow and revise some ideas of Paul Grice. The distinction between the function-indicating device and the propositionindicating device will prove very useful to us in giving an analysis of an illocutionary act. In this paper I shall not attempt to discuss propositional rules but shall concentrate on rules for using certain kinds of function-indicating devices. Function-indicating devices in English include word order. What is the difference between just uttering sounds or making marks and performing a speech act? One difference is that the sounds or marks one makes in the performance of a speech act are characteristically said to have meaning. and a second related difference is that one is characteristically said to mean something by those sounds or marks. stress. we can separate our analysis of the proposition from our analysis of kinds of illocutionary acts. intonation contour. incidentally. IV. without its being necessary to invoke the appropriate function indicating device. Often in actual speech situations the context will make it clear what the illocutionary force of the utterance is. Characteristically when one speaks one means something by what one says. and furthermore when one makes a move one is not characteristically said to mean anything by that move. To say that A meant something by x is to say that 'A intended the utterance of x to produce some 'Philosophical Review. and what one says. Since the same proposition can be common to all sorts of illocutionary acts.5 Grice gives the following analysis of one sense of the notion of 'meaning'. The pieces in agame like chess are not characteristically said to have a meaning. is another point at which our analogy between performing speech acts and playing games breaks down. 'I state'. it seems likely that it should have some syntactical analogue. and finally a set of so-called performative verbs: I may indicate the kind of illocutionary act I am performing by beginning the senten ce with 'I apologize'. the string of morphemes that one emits. . punctuation. But what is it for one to me an something by what one says. but those rules can be discussed independently of the rules for function indicating. I think there are rules for expressing propositions.44 WHAT IS A SPEECH ACT? is performing in the utterance ofthe sentence. MEANING Speech acts are characteristically performed in the utterance of sounds or the making of marks. If this semantical distinction is of any real importance. Here. 1957. and certain recent developments in transformational grammar tend to support the view that it does. etc. In an article entitled 'Meaning'. 'I warn'. is characteristically said to have a meaning. In the underlying phrase marker of a sentence there is a distinction between those elements which correspond to the function-indicating device and those which correspond to the propositional content. the mood ofthe verb.

But let us suppose I don't know enough German or Italian to do that. then your recognition of my intention to produce in you the beliefthat I am French is not the means I am employing. Now. it fails to distinguish the different kinds of effectsperlocutionary versus illocutionary---that one may intend to produce in one's hearers. become rather suspicious if you recognized my intention. first because it shows the elose reIationship between the notion of meaning and the notion of intention. But I might on the other hand attempt to get you to beIieve that I am French by simply telling you that I am French. What I would like to do is to tell them in German or Italian that I am a German officer. That is one of the things involved in telling you that I am French. I attempt to communicate to and convince my hearer of the truth of a certain proposition. characteristically. and it further fails to show the way in wh ich these different kinds of effects are related to the notion of meaning. This seems to me a useful start on an analysis of meaning. And suppose also that I wish to get these troops to beIieve that I am a German officer in order to get them to release me. Suppose that I am an American soldier in the Second World War and that I am captured by Italian troops. this account of meaning does not show the connection between one's meaning something by what one says and what that which one says actually means in the language. it seems to me to be in certain respects defective. A second defect is that it fails to account for the extent to which meaning is a matter of rules or conventions. dressing in the French manner.WHAT IS A SPEECH ACT? 45 effect in an audience by means of the recognition ofthis intention'. I might on the one hand attempt to get you to believe that I am French by speaking French all the time. trusting that they don't know enough German to see through . I shall illustrate this with an example. and cultivating French acquaintances. But of course if I try to get you to beIieve that I am French by putting on the act I described. as it were. In order to illustrate this point I now wish to present a counter-example to this analysis of meaning. what is the difference between these two ways ofmy attempting to get you to beIieve that I am French? One crucial difference is that in the second case I attempt to get you to believe that I am French by getting you to recognize that it is my purported intention to get you to beIieve just that. The point of the counter-ex am pie will be to illustrate the connection between what a speaker means and what the words he utters mean. essential to speaking a language : In speaking a language I attempt to communicate things to my hearer by means of getting hirn to recognize my intention to communicate just those things. showing wild enthusiasm for de Gaulle. I think. Indeed in this case you would. That is. attempt to put on a show of telling them that I am a German officer by reciting those few bits of German that I know. For example. when I make an assertion. which utterance I intend to produce in him the desired effect by means of his recognition of my intention to produce just that effect. I think. However valuable this analysis of meaning is. and the means I employ to do this are to utter certain sounds. First of all. So I. and secondly because it captures something which is.

we must capture both the intentional and the conventional aspects and especially the relationship between them. or even 'Ich bin ein deutscher Offizier'. and furthermore. let us describe the situation in Gricean terms. Therefore I. a captured American. Grice's account can be amended to deal with counter-examples of this kind. 'Knowest thou the land where the lemon trees bloom?' Of course. 1 intend that they should think that what 1 am trying to tell them is that 1 am a German offker. But does it follow from this account that when 1 say 'Kennst du das Land . he intends this recognition to be achieved in virtue of the fact that the rules for using the expressions he utters associate the expressions with the production of that effect. used as a means of producing quite different illocutionary effects.' etc. address my Italian captors with the following senten ce : 'Kennst du das Land. and 1 intend to produce this effect by means of their recognition of my intention. Let us suppose 1 know only one line of German. 1953\ para. We have here a case where 1 am trying to produce a certain effect by means of the recognition of my intention to produce that effect. namely. ifhe is using words Iiterally. what 1 me an is. In the performance of an illocutionary act the speaker intends to produce a certain effect by me ans of getting the hearer to recognize his intention to produce that effect. because what the words mean is. 6 The reason we are unable to do this is that what we can mean is a function of what we are saying. 510. V. 1 intend to produce a certain effect in them. . wo die Zitronen blühen?' Now. it is also a matter of convention. but part of what is involved in the deception is getting them to think that that is what the words which 1 utter mean in German. Meaning is more than a matter of intention. 'I am a German officer'? Not only does it not follow.46 WHAT IS A SPEECH ACT? my plan. but the device I use to produce this effect is one which is conventionally. HOW TO PROMISE I shall now attempt to give an analysis of the illocutionary act of promising. In order to do this 1 shall ask what conditions are necessary and sufficient for the act of 'Philosophicallnvestigations (Oxford.. which 1 remember from a poem 1 had to memorize in a high-school German course. It is this combination of elements which we shall need to express in our analysis of the illocutionary act.. At one point in the Philosophical Inl'estigations Wittgenstein says 'Say "it's cold here" and mean "it's warm here"'. In our analysis of illocutionary acts. the effect ofbelieving that 1 am a German officer. We must therefore reformulate the Gricean account of meaning in such a way as to make it c1ear that one's meaning something when one says something is more than just contingently related to what the sentence me ans in the Ianguage one is speaking. but in this case it seems plainly false that when 1 utter the German sentence what 1 mean is 'I am a German officer'. by the rules governing the use ofthat device. 1 want my captors to be deceived into thinking that what 1 mean is 'I am a German officer'. .

otherwise I should simply be showing the relation between different iIIocutionary acts. We are in the position of someone who has learned to play chess without ever having the rules formulated and who wants such a formulation. If we get such a set of conditions we can extract from them a set of rules for the use of the function-indicating device. I also confine my discussion to full-blown explicit promises and ignore promises made by elliptical turns of phrase. I want to give a list of conditions for the performance of a certain iIIocutionary act. The method here is analogous to discovering the rules of chess by asking oneself what are the necessary and sufficient conditions under which one can be said to have correctly moved a knight or castled or checkmated a player. and the proposition that the speaker made a promise entails this conjunction. I find the statement of the conditions very difficult to do. metaphors. like most notions in ordinary language. etc. In the presentation of the conditions I shall first consider the case of a sincere . hints. can be produced against my analysis. deviant. However. Thus each condition will-be a necessary condition for the performance of the act of promising and taken collectively the set of conditions will be a sufficient condition for the act to have been performed. and borderline promises . and the first step in getting such a formulation is to set out the conditions for the performance of a particular illocutionary act.to play the game ofillocutionary acts. borderline. and counterexamples. By stating a set of conditions for the performance of a particular illocutionary act we shall have offered a partial explication of that notion and shall also have paved the way for the second step. and partially defective cases. I need to satisfy this condition in order to offer an explication of the notion of an illocutionary act in general. which do not themselves mention the performance of any iIIocutionary acts. does not have absolutely strict rules. and I am not entirely satisfied with the list I am about to present. therefore. more or less bizarre. etc. although there will be no reference to iIIocutionary acts. I shall attempt to answer this question by stating these conditions as a set of propositions such that the conjunction of the members of the set entails the proposition that a speaker made a promise. I am confining my discussion. the formulation of the rules. to the centre of the concept of promising and ignoring the fringe. We learned how. One reason for the difficulty is that the notion of a promise. and I think this form of circularity is unavoidable because of the nature of constitutive rules. Another difficulty arises from my desire to state the conditions without certain forms of circularity. certain iIIocutionary concepts will appear in the analysans as weil as in the analysandum. but in general it was done without an explicit formulation of the rules. There are all sorts of odd.WHAT IS A SPEECH ACT? 47 promising to have been performed in the utterance of a given sentence. I am incIined to think we shall not be able to get a set of knock-down necessary and sufficient conditions that will exactly mirror the ordinary use of the word 'promise'. Our inquiry will therefore serve a double philosophical purpose.

in the utterance of T. however.) The notion of an act. S predicates afuture act A of S. Together they include such things as that the speaker and hearer both know how to speak the language. such as deafness. or laryngitis. not for you. and mayaiso include states and conditions: I may promise not to do something. not to you. they are not acting in a play or telling jokes. (4) H would prefer S's doing A to his not doing A. One can. and it is further defective if the promisor does not believe the promisee wants it done. but a threat is a pledge to do something to you. (2) Sexpresses that p in the utterance of T. think of apparent counter-examples to this condition as stated. S sincerely (and non-defectively) promises that p to H if and only if: (1) Normal input and output conditions obtain. both are conscious of what they are doing. since a non-defective promise must be intended as a promise and not as a threat or waming. (3) In expressing that p. they have no physical impediments to communication. I use the terms 'input' and 'output' to cover the large and indefinite range of conditions under which any kind of serious linguistic communication is possible. then. One crucial distinction between promises on the one hand and threats on the other is that a promise is a pledge to do something for you. A promise is defective if the thing promised is something the promisee does not want done. I shall simply assume the existence of grammatically well-formed sentences. 'Output' covers the conditions for intelligible speaking and 'input' covers the conditions for understanding. and I may promise to be or remain in a certain state or condition. performing series of acts. This condition isolates the proposition al content from the rest of the speech act and enables us to concentrate on the peculiarities of promising in the rest of the analysis. I cannot promise to have done something.48 WHAT IS A SPEECH ACT? promise and then show how to modify the conditions to aIlow for insincere promises. etc. I may promise to do something repeatedly. as I am construing it for present purposes. and S believes H would prefer his doing A to his not doing A. and I cannot promise that someone else will do something. aphasia. I call conditions (2) and (3) the propositional content conditions. includes refraining from acts. I think both halves of this double condition are necessary in order to avoid fairly obvious counter-examples. (Although I can promise to see that he will do it. Given that a speaker S utters a sentence Tin the presence of a hearer H. As our inquiry is semantical rather than syntactical. the speaker is not acting under duress or threats. In the case of promising the function-indicating device is an expression whose scope includes certain features of the proposition. . In a promise an act must be predicated of the speaker and it cannot be a past act.

then my request is pointless and to that extent defeetive. etc. for example. ete. This utterance would be more aptly described as an emphatic denial. . pay attention to what I am saying'. more commonly I think in the United States than in England. (5) It is not obvious to both Sand H that S will do A in the nonnal course of events. consider another apparent counterexample to the analysis along different lines. It is out of order for me to promise to do something that it is obvious I am going to do anyhow. In an actual speech situation. and the speaker must be aware of or believe or know. I promise you I didn't'. I think a more elegant and exact formulation of this condition would require the introduction of technical terminology. for example. In order to make sense of this utterance the audience will have to assurne that Smith has not been paying attention or at any rate that it is not obvious that he has been paying attention. we would more naturally describe it as a warning or possibly even a threat. For that reason we often use these expressions in the performance of speech acts which are not strictly speaking promises but in which we wish to emphasize our commitment. I didn't. will assurne that this condition is satisfied. that this is the ease. Sometimes. didn't you?' You reply 'No. In general the point stated in condition (4) is that if a purported promise is to be non-defective the thing promised must be something the hearer wants done.WHAT IS A SPEECH ACT? 49 Suppose I say to a lazy student 'If you don't hand in your paper on time I promise you I will give you a failing grade in the course'. Smith. Suppose. But why then is it possible to use the locution 'I promise' in such a case? I think we use it here because 'I promise' and 'I hereby promise' are among the strongest function-indicating devices for commitment provided by the English language. that the question of his paying attention has arisen in some way. To illustrate this. one hears people say 'I promise' when making an emphatic assertion. because a condition for making arequest is that it is not obvious that the hearer is doing or about to do the thing requested. if I make a request to someone to do something whieh it is obvious that he is already doing or is about to do. and we ean explain the occurrence of the function-indicating device 'I promise' as derivative from genuine promises and serving here as an expression adding emphasis to your denial. listeners. Similarly with promises. For example. I say. knowing the rules for performing illocutionary acts. I accuse you of having stolen the money. This condition is an instanee of a general condition on many different kinds of illoeutionary acts to the effect that the act must have a point. If I do seem to be making such a promise. Is this utterance a promise ? I am inc1ined to think not.. or considers to be in his interest. or would prefer being done to not being done. Suppose. 'You stole that money. the only way my audience can make sense of my utterance is to assurne that I believe . Did you make a promise in this case? I find it very unnatural to describe your utterance as a promise. that in the course of a publie speech I say to a member of my audience 'Look here.

50

WHAT IS A SPEECH ACT?

that it is not obvious that I am going to do the thing promised. A happily married man who promises his wife he will not desert her in the next week is likely to provide more anxiety than comfort. Parenthetically I think this condition is an instance of the sort of phenomenon stated in Zipf's law. I think there is operating in our language, as in most forms of human behaviour, a principle of least effort, in this case a principle of maximum illocutionary ends with minimum phonetic efTort; and I think condition (5) is an instance of it. I call conditions such as (4) and (5) preparatory conditions. They are sine quibus non of happy promising, but they do not yet state the essential feature.
(6) S intends to do A.

The most important distinction between sincere and insincere promises is that in the case of the insincere promise the speaker intends to do the act promised, in the case of the insincere promise he does not intend to do the act. Also in sincere promises the speaker believes it is possible for him to do the act (or refrain from doing it), but I think the proposition that he intends to do it entails that he thinks it is possible to do (or refrain from doing) it, so I am not stating that as an extra condition. I call this condition the sincerity condition.
(7) S intends that the uUerance of T will place him under an obligation to do A.

The essential feature of a promise is that it is the undertaking of an obligation to perform a certain act. I think that this condition distinguishes promises (and other members of the same family such as vows) from other kinds of speech acts. Notice that in the statement of the condition we only specify the speaker's intention; further conditions will make dear how that intention is realized. It is dear, however, that having this intention is a necessary condition of making a promise ; for if a speaker can demonstrate that he did not have this intention in a given utterance, he can prove that the utterance was not a promise. We know, for example, that Mr. Pickwick did not promise to marry the woman because we know he did not have the appropriate intention. I call this the essential condition.
(8) S intends that the utterance ofT will produce in Habeliefthat conditions (6) and

(7) obtain by means of the recognition ofthe intention to produce that belief, and he intends this recognition to be achieved by means of the recognition of the sentence as one conventionally used to produce such beliefs.
This captures our amended Gricean analysis of what it is for the speaker to mean to make a promise. The speaker intends to produce a certain iIIocutionary effect by means of getting the hearer to recognize his intention to produce that efTect, and he also intends this recognition to be achieved in virtue of the fact that the lexical and

WHAT IS A SPEECH ACT?

51

syntactical character of the item he utters conventionally associates it with producing that effeCt. Strictly speaking this condition could be formulated as part of condition (1), but it is of enQugh philosophical interest to be worth stating separately. I find it troublesome for the following reason. If my original objection to Grice is really valid, then surely, one might say, all these iterated intentions are superfluous; all that is necessary is that the speaker should seriously utter a sentence. The production of all these effects is simply a consequence of the hearer's knowledge of what the sentence means, which in turn is a consequence of his knowledge of the language, which is assumed by the speaker at the outset. 1 think the correct reply to this objection is that conditi on (8) explicates what it is for the speaker to 'seriously' utter the sentence, i.e. to utter it and mean it, but I am not completely confident about either the force of the objection or of the reply.
(9) The sernantical rules ofthe dialect spoken by Sand H are such that T is correctly and sincerely uttered if and only if conditions (1 )--(8) obtain.

This condition is intended to make clear that the sentence uttered is one which by the semantical mies of the language is used to make a promise. Taken together with condition (8), it e1iminates counter-ex am pIes like the captured soldier ex am pie considered earlier. Exactly what the formulation of the rules is, we shall soon see. So far we have considered only the case of a sincere promise. But insincere promises are promises none the less, and we now need to show how to modify the conditions to allow for them. In making an insincere promise the speaker does not have all the intentions and beliefs he has when making a sincere promise. However, he purports to have them. Indeed it is because he purports to have intentions and beliefs which he does not have that we describe his act as insincere. So to allow for insincere promises we need only to revise our conditions to state that the speaker takes responsibility for having the beliefs and intentions rather than stating that he actually has them. A clue that the speaker does take such responsibility is the fact that he could not say without absurdity, e.g., 'I promise to do A but I do not intend to do A'. To say 'I promise to do A' is to take responsibility for intending to do A, and this condition holds whether the utterance was sincere or insincere. To allow for the possibility of an insincere promise then we have only to revise condition (6) so that it states not that the speaker intends to do A, but that he takes responsibility for intending to do A, and to avoid the charge of circularity I shall phrase this as folIows:
(6*) S intends that the utterance of T will make hirn responsible for intending to doA.

Thus amended (and with 'sincerely' dropped from our analysandum and from condition (9», our analysis is neutral on the question whether the promise was sincere or insincere.

52

WHAT IS A SPEECH ACT?
VI. RULES FOR THE USE OF THE FUNCTION-INDICATING DEVICE

Our next task is to extract from our set of conditions a set of rules for the use of the function-indicating device. Obviously not all of our conditions are equally relevant to this task. Condition (1) and conditions of the forms (8) and (9) apply generally to all kinds of normal illocutionary acts and are not peculiar to promising. Rules for the function-indicating device for promising are to be found corresponding to conditions (2)-(7). The semantical rules for the use of any function-indicating device P for promising are:
Rufe 1. P is to be uttered only in the context of a sentence (or larger stretch of discourse) the utterance of which predicates some future act A of the speaker S. I call this the propositionaf-content rufe. It is derived from the propositional-content conditions (2) and (3). Rufe 2. P is to be utterred only if the hearer H would prefer S's doing A to his not doing A, and S believes H would prefer S's doing A to his not doing A. Rufe 3. P is to be uttered only if it is not obvious to both Sand H that S will do A in the normal course of events. I call rules (2) and (3) preparatory rufes. They are derived from the preparatory conditions (4) and (5). Rufe 4. P is to be uttered only if S intends to do A. I call this the sincerity rufe. It is derived from the sincerity condition (6). Rufe 5. The utterance of P counts as the undertaking of an obligation to do A. I call this the essential rufe. These rules are ordered: rules 2-5 apply only if rule 1 is satisfied, and rule 5 applies only if rules 2 and 3 are satisfied as weil. Notice that whereas rules 1-4 take the form of quasi-imperatives, i.e. they are of the form: utter P only if x, rule 5 is of the form: the utterance of P counts as Y. Thus rule 5 is of the kind peculiar to systems of constitutive rules which I discussed in section 11. Notice also that the rather tiresome analogy with games is holding up remarkably weil. If we ask ourselves under what conditions a player could be said to move a knight correct1y, we would find preparatory conditions, such as that it must be his turn to move, as weil as the essential condition stating the actual positions the knight can move to. I think that there is even a sincerity rule for competitive games, the rule that each side tries to win. I suggest that the team which 'throws' the game is behaving in a way c10sely analogous to the speaker who lies or makes false prornises. Of course, there usually are no propositional-content rules for games, because games do not, by and Iarge, represent states of affairs. If this analysis is of any general interest beyond the case of promising then it would seem that these distinctions should carry over into other types of speech act, and I

Greetings are a much simpler kind of speech act. Consider.WHAT IS A SPEECH ACT? 53 think a little reflection will show that they do. giving an order.g. or cognitive versus emotive. the preparatory conditions include the fact that the hearer must have some basis for supposing the asserted proposition is true. In the utterance of 'Hello' there is no propositional content and no sincerity condition. but the comparison of different analyses would deepen our understanding of the whole subject and incidentally provide a basis for a more serious taxonomy than any of the usual facile categories such as evaluative versus descriptive. and the essential rule is that the utterance indicates courteous recognition of the hearer. e. The preparatory conditon is that the speaker must have just encountered the hearer. Not only would this give us an analysis of concepts interesting in themselves. Thc preparatory conditions include that the speaker should be in a position of authority over the hearer. the sincerity condition is that the speaker wants the ordered act done. but even here some of the distinctions apply. A proposal for further research then is to carry out a similar analysis of other types of speech acts. and the essential condition has to do with the fact that the utterance is an attempt to inform the hearer and convince hirn of its truth.. and the essential condition has to do with the fact that the utterance is an attempt to get the hearer to do it. the sincerity condition is that he must believe it to be true. For assertions. .

P. will form part of a book to be published by the Harvard University Press. taking into account the fact that wh at he has implicated may be either conl'entionally implicated (implicated by virtue ofthe meaning ofsome word or phrase which he has used) or non-conl'entionally implicated (in which casethe specification ofthe implicature falls outside the specification ofthe conventional meaning of the words used). PROLEGOMENA My aim in this paper is to throw light on the connection between (a) a notion of meaning which I want to regard as basic. if correct.' (locutions which specify what rnight be called 'occasion-meaning') and locutions of the From Foundalions oJ Language. Reprinted by permission of the author and the editor of Foundations oJ Language. suggested. The stages of the programme are as folIows: (I) To distinguish between locutions of the form 'U (utterer) meant that . SENTENCE-MEANING. The programme is directed towards an explication of the favoured sense of 'say' and a clarification of its relation to the notion of conventional meaning. indicated. I hope. and (b) the notions of meaning involved in saying (i) that a given senten ce means 'so-and-so' (ii) that a given word or phrase means 'so-and-so'. I wish to make within the total signification ofaremark: a distinction between what the speaker has said (in a certain favoured. 1-18. sense of 'said'). so that attention may be focused on its utility. 4 (1968). and what he has 'implicated' (e.g. implied. AND WORD-MEANING' H. that notion which is involved in saying of someone that by (when) doing such-and-such he meant that so-and-so (in what I have called a non-natural sense of the word 'meant'). and maybe in some degree artificial. The wider programme just mentioned arises out of a distinction which. revised and re·arranged. I I hope that material in this paper.). What I have to say on these topics should be looked upon as an attempt to provide a sketch of what might. pp. etc. I should like its approximate correctness to be assumed. though its later stages lie beyond the limits which I have set for this paper. GRICE A. for purposes which I need not here specify. The account which I shall otTer of the (for me) basic notion of meaning is one which I shall not today seek to defend. in the explication of other and (I hope) derivative notions of meaning. viz. prove to be a viable theory.. This enterprise forms part of a wider programme which I shall in a moment delineate. . . rather than as an attempt to provide any part of a finally acceptable theory.IV UTTERER'S MEANING.

g. (b) The truth of a statement to the effect that X means ' . to explicate sentences which make claims of the form 'X (utterance-type) means "*'''. In locutions of the first type. "'.. A specificatory replacement of the dummy 'x' will in some cases be a characterization of a deed. meaning is specified without the use of quotation-marks. to cover any case of doing x or producing x by the performance of which U meant that so-and-so. e. whereas in locutions of the second type the meaning of a sentence.. and in the order just given. Again. more precisely.AND WORD MEANING 55 form 'X (utterance-type) means H ••• " ' .. An utterance-type rriay have more than one conventional meaning. (111) To attempt to elucidate the notion of the conventional meaning of an utterance-type. or. claims of the form 'X means sentential expression. Some explanatory comments are needed here. e. when the two lacunae are quite differently completed. (11) To attempt to provide a definiens for statements of occasion-meaning.Smith will go home' or to 'Jones meant that! Smith will go horne'. The performance in question need not be a linguistic or even a conventionalized performance. One might thus get to 'Jones meant that Srnith will go horne' 'Jones meant that Srnith is to go horne'. U meant that *p'.g. '" and (ü) statements of timeless 'Ianguage meaning'. This difference is semantically important. in case X is a non-scntcntial utteranccwhere the location is completed by a nontype. . 'In L (language) X means " . (b) '*' is a dummy mood-indicator. More precisely. one may think of the schema 'Jones meant that *p' as yielding a full English sentence after two transformation al steps: (i) replace '*' by a specific mood-indicator and replace 'p' by an indicative sentence... It will be convenient to handle these separately. . some explanatory comments are required. (a) I use the term 'utter' (together with 'utterance') in an artificially wide sense. One might thus get to 'Jones meant that I.g. 'For U (in U's idiolect) X means " . in which the ~pecification of meaning involves quotation-marks) may be subdivided into (i) statements of timeless 'idiolect-meaning'. "'. more precisely.. and any definiens which we offer must H ••• " ' . (a) It will be convenient to recognize that what I shall call statements of timeless meaning (statements of the type 'X means " . asound).. distinct from specific mood-indicators like 'I-' (indicative or assertive) or '!' (imperative). to provide a definiens for 'By (when) uttering x. (ii) replace the sequence following the word 'that' by an appropriate clause in indirect speech (in accordance with rules specified in a linguistic theory).' is of course not incompatible with the truth of a further statement to the effect that X me ans '--". in others a characterization of a product (e. word or phrase is specified with the aid of quotationmarks.

Indeed.56 UTTERER'S MEANING."]' (ii) 'When U uttered the sentence "Palmer gave NickIaus quite a beating" U meant that Palmer vanquished NickIaus with some ease. there are. brave' (S .). 'X means " . "Bill is courageous". Comments. U meant that *p' and (ii) 'When uttered by U.. but nevertheless (i) would still have been true.. '" should be understood as 'One of the meanings of X is " . (a) We must be careful to distinguish the applied timeless meaning of X (type) with respecf to a particular token x (belonging to X) from the occasionmeaning of U's utterance of x. give a full specifu::ation of timeless meaning for S. "'. I think. and that is all that is included'. "Palmer administered vigorous corporal punishment to NickIaus. a schema the specifications of which announce the correct reading of X for a given occasion of utterance. (IV) In view of the possibility of multiplicity in the timeless meaning of an utterance-type. X meant "*p'" provides a definiens for 'In uttering X. " '. fWe might re-express this as 'One meaning of S. U said that *p'. what I shall call the applied timeless meaning of an utterance-type. In that case (ii) would c1early be false. at least some sentences whose timeless meaning is not adequately specifiable by a statement of the exemplary form. in the case of applied time1ess meaning. therefore. we shall need to notice. to make a partial specification of the timeless meaning of S. we need a definiens for the schema 'X (utterance-type) meant here " . the sentence "Palmer gave Nickiaus quite a beating" meant "Palmer vanquished Nickiaus with some ease" [rather than. by saying 'Part of one meaning of S. The following are not equivalent: (i) 'When U uttered it. say.. it may even be possible to uphold the thesis that such a coincidence of occasion-meaning and applied time1ess meaning is a necessary and sufficient condition for saying that *p. I think. is "Bill is occupationally engaged in philosophical studies" '. (b) There is some temptation to take the view that the conjunction of (i) 'By uttering X. indeed.. in which case he would very likely have meant that NickIaus vanquished Palmer with some ease. ifwe give consideration only to utterance-types for which there are available adequate statements of time1ess meaning taking the exemplary form 'X meant "*p'" (or. . and to provide an explication of.. the form 'X meant here "*p" '). Consider the sentence 'Bill is a philosopher and he is. That is to say..' U might have been speaking ironically. comprises "Bill is occupationally engaged (etc)". SENTENCE MEANING allow fOT this fact. by saying 'One meaning of S. It would be appropriate. One might. inc1udes "Bill is occupationally engaged in philosophie al studies" and "Bill is courageous" and "That Bill is courageous follows from his being occupationally engaged in philosophical studies". But a litde refiection should convince us of the need to recognize the existence of statements of timeless meaning which instantiate forms other than the cited exemplary form.

though not to say. while insufficient for the truth of 'U said that *p' will be suffieient (and neeessary) for the truth of 'U conventionally meant that *p'. whieh does not seem to be the case. Mutatis mutandis.)" '. I would wish to maintain that the semantic function of the word 'therefore' is to enable a speaker to indicate. in my favoured sense of'say'.)'). it may be true both (i) that when U uttered X the meaning of X inc1uded '*p' and (ii) that part of what U meant when he uttered X was that *p. among other things. one who utters SI will have said that Bill's being courageous follows from his being a philosopher.) and Bill is courageous and that Bill is eourageous follows (ete.). I would like to use the expression 'conventionally meant that' in such a way that the fulfilment of the two conditions just mentioned. and "That Bill is courageous follows (ete. ifboth (i) U has Y-ed that *p. though he may weil have said that Bill is a philosopher and that Bill is courageous. with respect to a given utterance by U of SI' 'The meaning of SI here comprised "Bill is oecupationally engaged (ete.'] It will be preferable to speeify the timeless meaning of S I in this way than to do so as folIows: 'One meaning of S I is "Bill is occupationally engaged (etc. that a certain eonsequenee holds. (3) To define. I would adopt the same position with regard to words like 'but' and 'moreover'.)" '. for this latter formulation at least suggests that SI is synonymous with the conjunctive sentence quoted in the formulation. or to say 'The meaning of S I here included "That Bill is courageous follows (etc. 'U has Y-ed . the members of which will thereby be stamped as speeially central or fundamental. Since it is true that another meaning of SI inc1udes 'Bill is addicted to general reftections about life' (l'ice 'Bill is occupationally engaged (etc. My primary reason for opting for this partieular sense of'say' is that I expect it to be of greater theoretical utility than some other sense of'say' would be. . one could have occasion to say (truly). It could also be true that when U uttered SI he meant (part of what he meant was) that that Bill is eourageous follows (ete.)". (2) To stipulate that in uttering X. U will have said that *p. that is to say. and (ii) X embodies some eonventional deviee the meaning of which is such that its presenee in X indicates that its utterer is Y-ing that *p. "Bill is eourageous".)"'. where Y-inR is a eentral speech-act.AND WORD MEANING 57 and "That Bill is eourageous follows (ete . for each member Y of the range of central speech-aets. that *p.)". I have hopes of being able to discharge this task by proceeding along the following lines: (1) To specify conditions whieh will be satisfied only by a limited range of speechacts. and yet be false that U has said. (V) This distinction between what is said and wh at is eonventionally meant creates the task of specifying the eonditions in whieh what U eonventionally meant by an utterance is also part of what U said. So I shall be committed to the view that applied timeless meaning and occasionmeaning may eoincide. Now I do not wish to allow that.

a three-clause definiens which may be compendiously reformulated as 'For some audience A. I shall introduce an abbreviation. at least for an important sub-class of such elements. TREATMENT OF SOME OF THE PROBLEMS RAlSED The problems which I shall consider in the remainder of this paper are those which are presented by Stages II -IV of the programme just outlined. (2) If Z-ing is such a non-central speech-act. for the schema 'U meant (non-naturally) something by uttering x'. but also of saying that = p. U intended his utterance of x to produce in A some effect (response) E. which (as I said at the outset) I should like to be treated as if it were correct. the performance of which would require the performance of one or other of the central speech-acts. Stage 11 I shall otTer. that the . and such that their performance is dependent upon. without arguing for it. In my 1957 article on 'Meaning'3 I in etTect suggested. LXVII (1957). 3 IPhilosophical Review. a somewhat over-simplified account of the notion of occasion-meaning. This account. 'U intends to produce in A effect E by means of A's recognition ofthat intention' will be abbreviated to 'U M-intends to produce in A etTect E'. ('M' for 'meaning'. might take the following shape: (1) The problematic elements are linked with certain speech-acts which are exhibited as posterior to.) The point of divergence between my current account and my 1957 account lies in the characterization of the M-intended effect (response). SENTENCE MEANING that *p' in terms of occasion-meaning (meaning that . Z (3) The notion of Z-ing that *p (where Z-ing is non-central) would be explicated in terms of the nation of meaning that (or in terms of some important elementes) in the definition of that notion). the meaning of 'moreover' would be linked with the speech-act of adding. or of saying that = *p (where' = p'.. some member or disjunction of members of the central range. In the earlier account I took the view that the M-intended effect is. the dependence of Z-ing that *p upon the performance of some central speech-act would have to be shown to be of a nature which justifies a reluctance to treat Z-ing that *p as a case not merely of saying that *p. As I wish to continue to use the central idea of this definition. is a representation of one or more sentential forms specifically associated with Z-ing).. or ' = *p'. by means of A's recognition ofthat intention'.1 • . for example. in the case of indicative-type utterances. (VI) The fulfilment of the task just outlined will need to be supplemented by an account of the elements in the conventional meaning of an utterance which are not part of what has been said. B. As 'moreover _ ' is specifically associated with the speech-ac! of adding. ) or in terms of some important elementes) involved in the already provided definition of occasion-meaning.58 UTTERER'S MEANING.

epp' = df. Let the device 'ep t.ep' (read 'asterisk-sub-ep') be a dummy. in other cases this phrase is to be interpreted as 'that A should ep that p (via thinking U to ep that p)'. offer the following rough definition: 0. 'I-' corresponds to believing (thinking) and '!' corresponds to intending. 1 shall now try to reformulate the account in a generalized form. It will be convenient to have an abbreviated version of this definiens. but that the hearer should think that the utterer believes something. and. '( 3A) (U uttered x M-intending (i) that A should think U to lj) that p and in some cases only ldepending on the identil'ication of "* lj)P "I. To meet all the difficulties to which my 1957 account (which was only intended as a model) is exposed. The efTect of the second change (made in order to unify the treatment of indicative-type utterances. But as the examples which force the introduction of this complexity involve relatively sophisticated kinds of communication or linguistic performance. 'By (when) uttering x U meant that . using this device. and some of which are not. in the case of imperative-type utterances. via the fulfilment of (i). the ulterior intention on the part of the utterer that the hearer should go on to do the act in question). The effect of the first change will be that the way is opened to a simplified treatment of the M-intended efTect.ljJp' = df. a very much more complicated definition is required.epp'.1 '.1. of course. (ii) that A should. Which interpretation is to be selected is determined by the specification of '. as being always the generation of some propositional attitude. Let the device '. for working purposes. some of which are. that the hearer should do something. cases of informing or telling) will be to introduce a distinction between what 1 might call exhibitive utterances (utterances by which the utterer U M-intends to impart a belief that he (U) has a certain propositional attitude) and utterances which are not only exhibitive but also what 1 might call protreptic (utterances by which U M-intends.AND WORD MEANING 59 hearer should helieve something. the proffered definition will be adequate. Let 'A' range over audiences or hearers. via imparting a belief that he (U) has a certain propositional attitude. not that the hearer should believe something (though there will frequently be an ulterior intention to that effect).1 as folIows: 0. (read 'Ij!-dagger') be a dummy which operates as folIows: in some cases the phrase 'that A should ep t that p' is to be interpreted as 'that A should think U to ep that p'. We may now reformulate 0. (2) 1 wish to regard the M-intended efTect common to indicative-type utterances as being. as for example. '( 3A) (U uttered x M-intending that A should ep t that p)'. hirnself lj) that p)'. I can. which represents a specific mood-indicator which corresponds to the propositional attitude ep-ing (whichever that may be). . (1) 1 wish to represent the M-intended efTect ofimperative-type utterances as being that the hearer should intend to do something (with. 1 wish for present purposes to make twochangeshere. to induce a corresponding attitude in the hearer). 1 hope that. 'By (when) uttering x U meant that .

Now. My main example will be a gesture (a signal). So initially 1 shall restrict myself to examining the notion of timeless meaning in its application to unstructured utterance-types. to this suggested definiens. more perspicuously. or. we shall get the following expanded definiens: 'It is U's policy (practice. and what ac count is to be given of timeless meaning for noncomplete utterance-types? If we do not treat these problems separately. a . habit) to utter H-W ifT U is making an utterance by means ofwhich (for some A) U M-intends to efTect that A thinks U to think that U knows the route'. we shall have only ourselves to blame for the confusion in which we shall find ourselves. (1) What is the relation between timeless meaning (for complete utterance-types) and occasion-meaning? (2) In the case of syntactically structured (Iinguistic) utterance-types. extremely important to distinguish two problems. As a first shot one might suggest something Ijke 'It is U's policy (practice. Now. SENTENCE MEANING Stage III (Step (1): timeless meaning for unstructured utterance-types) It is. We are to look for an explication of the sentence 'For U. 'It is U's policy (practice.l. If we apply D. we specify U's policy as folIows: 'I (that is. Suppose that a particular sort of hand-wave (to be referred to as H-W) for a particular individual U (within U's idiolect) means 'I know the route'. habit) is such that his use of H-W is tied to the presence of a simple intention to afTect an audience in the way described. I think. if U's policy (practice. so to speak). how is the timeless meaning of a complete (sententiaI) utterance-type related to the timeless meanings of its non-complete structured and unstructured elements (approximately. he must (Iogically) be in a position. as U weil knows. habit) to utter H-W ifTU is making an utterance by which U means that U knows the route'.60 UTTERER'S MEANING. when uttering H-W. using only the notion of simple intention. he will do so. and it will be convenient first to consider the idea of its timeless meaning for an individual (within a signalling idiolect. phrases and words). habit) to utter H-W in order to mean that U knows the route' (where 'mean that' is to be analysed in accordance with D. M-intending to affect his audience in that way. utterer U) shall utter H-W ifT I intend (want) some A to think that I think I know the route'. and that only the notion of simple intention need be invoked. I wish to argue that the notion of M-intention is otiose here. H-W means "I know the route'" which will relate timeless meaning to occasion-meaning.l. if U is ever to have the particular intentions which will be involved in every implementation of this policy. on a given occasion. for such a supposition to be justified. to suppose that there is at least some chance that these intentions will be realized. and only afterwards to consider the extension of this idea to groups of individuals. on that occasion.). he utters H-W. Suppose that. We shall thus preserve for the time being the possibility of keeping distinct the ideas of having an established meaning and of having a conventional meaning. it will follow that when. whether or not this definiens is otherwise acceptable.

2. performances. for some A. to put on a pair of dark spectacles (instead of pausing and taking asedative). 'U has in his repertoire the following procedure: to utter a token of X if U intends (wants) A to I/J t that p'. To cope with these difficulties. if U has in his repertoire the procedure of uttering H-W if he wants an audience A to think U thinks U knows the route. 'For U utterance-type X means (has as one of its meanings) "*!/Jp'" =df. Somewhat similarly. U intends (wants) A to think that U thinks U knows the route'. But to utter H-W expecting A to respond in such a way is to utter H-W M-intending that A should think that U thinks that U knows the route. it may also mean 'I am about to leave you'. suggest a simplified definition: 'For U. (1) For U. 1 think 1 need some such idea as that of 'having a certain procedure in one's repertoire'. however. whenever) U wants an A to think that U thinks U knows the route. H-W me ans "I know the route'" = df. U. A faintly eccentric lecturer might have in his repertoire the following procedure: if he sees an attractive girl in his audience.e. on occasion. by a particular utterance of H-W.) cannot be to utter H-W only if U wants some A to think that U thinks U knows the route. So a formulation of U's policy of H-W in terms of the notion of simple intention is adequate to ensure that. to employ them.AND WORD MEANING given audience A must be aware of U's policy and must suppose it to apply In Ihl' utterance of H-W with which U has presented hirn. and may be ready.) cannot be to utter H-W if (i. We may now turn from the idea of timeless meaning within an 'idiolect' to that of . This idea seems to me to be intuitively fairly intelligible and to have application outside the realm of linguistic. to pause for half aminute and then take a sedative. in that case. '11 is U's policy (practice. This definition. U's policy (etc. (b) to pause and take asedative when he sees in his audience not an attractive girl. His having in his repertoire this procedure would not be incompatible with his also having two further procedures : (a) if he sees an attractive girl. this fact would not be incompatible with his having at least two further procedures: (a) to say 'I know the route' if he wants some A to think U thinks U knows the route. but a particularly distinguished colleague. H-W may have a second meaning. is doubly unacceptable. then. though it could hardly be denied that it requires furt her explication. when uttering 11 W Oll a particular occasion. or otherwise communicative. That being so. U's policy (etc. must expect A to think (or at least to be in a position to think) as folIows: 'U's policy for H-W is such that he utters H-W now with the intention that 1 should think that he thinks that he knows the route. habit) to utter H-W itT. then. So 1 propose the definition: D. 1fthat is so. We may. sometimes he will be ready to utter H-W wanting some A to think that U thinks that U is about to leave A. (2) U may have other ways of getting an A to think that U thinks that U knows the route (such as saying 'I know the route'). U will mean that he knows the route. 1 take it that he does think that he knows the route'. and (b) to utter H-W if U wants some A to think U thinks he is about to leave A.

' = df. to .3. In that case my Aunt Matilda (a member of G) may be said to have a procedure for X even though she herself would rat her be seen dead than utter X. this procedure in their repertoires'. and content myself with a few informal remarks. So obviously each member of some group G (within which H-W is to be a tool of communication) will want his procedure with respect to H-W to conform to the general practice of the group. the following: 'U has in his repertoire the procedure of . unless the signal is to be explained to each A.. that is to say. . D. utterance-type X me ans "*~)p'" = df. In this case U will have a readiness to utter X in such-andsuch circumstances. There seem to me to be three main cases in which one may legitimately speak of an established procedure in respect of utterancetype X. for she knows that some other members of G do have a readiness to utter X in such-and-such circumstances. and yet quite false that she has any degree ofreadiness to utter the expression in any circumstances whatsoever. but the utterance of X in such-and-such circumstances is part of some system of communication which U has devised. gets in the idea of aiming at conformity. and so perhaps (derivatively) also that of correct and incorrect use of X. in some degree. the expression 'he is a runt' me ans 'he is an undersized person'. a task of considerable difficulty. SENTENCE MEANING timeless meaning for a group or class ofindividuals. (3) That in which X is not current at all. So I shall for the present abandon the attempt to provide adefinition. What one seems to need is the idea of her being equipped to use the expression. to utter X in suchand-such circumstances is part of the practice of many members of G. or have had. 'For group G. it is only crs practice to utter X in suchand-such circumstances. as distinct from the idea merely of usual or unusual use of X. (1) That in which Xis current for some group G. as a make-shift definition. The explication of the notion of 'having a procedure in one's repertoire' is.'. So I suggest the following rough definition: 0. preparedness). on A's repertoire containing the same procedure. and in general. they want A to !fJ t that p. for some A. 'U has a standing readiness (willingness. for my exceedingly prim Aunt Matilda. but which has never been put into operation (like the new Highway Code which I invent . But this definition would clearly be inadequate as it stands. I have feit inclined to propose. . (2) That in which X is current only for U. 'At least some (? many) members of group G have in their repertoires the procedure of uttering a token of X if. IfU utters H-W.. a readiness (etc.62 UTTERER'S MEANING. the retention of this procedure being for them conditional on the assumption that at least some (other) members ofG have. it may weil be true that. to my mind. so to speak) as an intention to do that thing.3. his measure of expectation of success as regards efTecting the intended response obviously depends (as has a1ready been remarked) on A's knowledge of U's procedure. and the analysis of this idea is also problematic.) to do something being a member of the same family (a weaker brother.

(I 2 ) uttered by U. Or more fully: Let '..AND WORD MEANING 63 one day while lying in my bath). (I 2 (not e. '( 3A) (U meant by uttering X that .4. (I 2 )] is to be a camplete utterance. and so that I 2 may = I.-with-I2 ') be a particular utterance (token) of I.'q via A's knowledge (assumption) that in U's repertoire is the procedure of uttering X if.'q. Allow that a sentence is a sub-sentence of itself. (I 2 )' (read 'I.] 0.g.p" [as defined by 0. v [I. ).4'. that is. and of course 0. for some A'.4'.. 'When U uttered X (type). complete and non-complete) To deal with structure utterance-types and their elements. (2) Let vIII (I 2 )] (read 'v-of-I. to be the utterance of a disjunct). . ['p' may. so to speak) to his accomplice that he knows how to handle the victim.2.p'" =df.])'. 0. it is not to be part of v[I 3 (I. lexical items) which enter into them. for example. Stages 1/1 and IV (Step (2): timeless and applied timeless meaning for structured utterance-types. U wants A' tO!/J+ that p)'... on the basis of A's knowledge (assumption) that.-with-I2') denote a sentence of which I 2 is a subsentence. represent that propositional content to which indefinite reference is made in the existential quantification of 'q'. Stage IV (Step (I): applied timeless meaning for unstructured utterance-types) We are now in a position to define a notion of applied time1ess meaning which will apply to H-W. for U. ' and meaning that . U (a criminal) has lured a victim into his car and signals (non-literally. X meant "." both be dummy mood-indicators.!/Jp'" =df.. and also for the case in which.' and '.. (3) It is characteristic of sentences (a characteristic shared with phrases) that their standard meaning is consequential upon the meaning of the elements (words. In that case U has a procedure for X in the attenuated sense that he has envisaged a possible system of practices which wauld involve a readiness to utter X in such-and-such circumstances.4. '( 3A) (3q) (U intended A to recognize (? and to recognize that U intended A to recognize) what U meant [occasion-meaning] by his uttering X. and U intended A to recognize (? and to recognize that he was intended to recognize) that by uttering X U meant that . So I need the notion of a 'resultant procedure': as a first approximation. 0. I think I need the following apparatus. or may not. (1) Let 'I. 'When U uttered X. X meant ". X means (has as one of its meanings} ". one might say that a procedure for an utterance-type X will be aresultant procedure if it is determined by (its existence is inferable from) a know- ») . allow both for the case in which U meant by H-W that he knew the route (coincidence ofmeaning ' . In both cases it is expected by U that the audience's understanding of the utterance of H-W will be based on its knowledge that U has a certain procedure (to utter H-W if U wants an audience to think that U thinks U knows the route).

for a very restricted (but central) range of ward-types and syntactical forms. to show the role of word-meanings (covering every type of ward) in deterrnining (in combination) sentence meanings (covering sentences of any syntactical structure). (3) Suppose we. of a fragment of what might be the kind of theory we need. We want to be able to speak of some particular object as an R-correlate of a (nominal). For applied timeless meaning laffer D. 'L 2 in V[LI (L 2») meant "*rpp'" =df. But the nation of'resultant procedure' has been left pretty unilluminated .2. Then we need to be able to apply the ideas of an indicative version of a (a itself). viz. and (b) for any sequence of utterance-types which exemplifies a particular ordering of syntactical categories (a particular syntactical form). 'For U.5.' [D. 'Smith. 'Oh that Smith may be tactful' would be required to count. The apparatus needed (for one such attempt) would be: (1) Suppose a to be an indicative sentence.3. 'Smith is tactful'.) An explication of timeless meaning in any language can. (far suitable A') to utter L 2 if U wants ~' to rp+ that p). be tactful'. D. Let us take as our range all ar part of the range of affirmative categorical (not necessarily indicative) sentences involving a noun (or definite description) and an adjective (or adjectival phrase).6.. take for granted two species of correlation. an imperative version of a. But this looks like a Herculean task (in our present state of knowledge). L means "*rpp'" =df. and U intended A to recognize that U meant by V[LI (L 2 )] that *'q at least partly on the basis of A's thought that U has a resultant procedure for L 2 .' [D. 'U has a resultant procedure for L. Rcorrelation (referential) and D-correlation (denotational). Now let us deal with the notion of timeless meaning in U's idiolect. SENTENCE MEANING ledge of procedures (a) for particular utterance-types which are elements in X. U wants A to rp + that p. (mood variations). It would again be the business of some linguistic theory to set up such a sentential characterization. viz. I shall not attempt this task now. It would be the business of some linguistic theory to equip us to apply such characterizations (so as philosophers of language we can ass urne this as given). an optative version of a etc. as predications of 'tactful' on 'Smith'. far same A. all of them.6. '(ijA) (ijq) (U meant by VLL I (L 2 )] that *'q. we ought to look at the nature of the more fundamental procedures from which a resultant procedure descends. and if we are to shed any light on the nation of word-meaning. It would be nice to give a general schema.4'.5.) So far (maybe) so good. par allels D. be provided by adapting D. to utter L if.64 UTTERER'S MEANING. (2) We need to be able to apply some such nation as a predication of ß (adjectival) on (t (nominal). and its connection with 'meaning that'. par allels D. for a moment. perhaps. and of each member of some dass as being a D-correlate of ß (adjectival). Now suppose that U has the following procedures: . 'Let Smith be tactful'. The best we can hope for is a sketch.

Given RP2 and C2.) (Also..l predication of ß on a if(for some A) U wants A to rp t a particular R-correlate of a to be one of a particular set of D-corre1ates of ß.e. and P.2. is an interpretant of 'For U.I': obtained from P.) [Sueh proeedures set up eorre1ations between moods and speeifieations of 'rp t'. (adjectival) means " .g.l P."'. for example. Given RP land Cl we can infer that U has RP2.'" =df.l.I by substituting 'imperative'/ 'indieative' and 'intend'/'think that U thinks'. Given that U has the initial proeedures P.7. Jones' dog is an R-corre1ate of 'Fido'.2. if U had P. D. C2.. for example P.l.' I have not yet provided an explication for statements of time1ess meaning re1ating to non-eomplete utteranee-types. (the blank being filled by the infinitive version of a.7. And RP4. can be shown to be a resultant procedure. 'Smith to be taetful'. To utter rpolo-eorre1ated [cf. .. "Fido is shaggy" means "Jones' dog is hairy-coated".. Any hairy-coated thing is a D-eorre1ate of 'shaggy'.I'. we can infer that U has the resultant procedure (determined by P.I. it will be inferable that he has the procedure of uttering a rp t-correlated predieation of 'shaggy' on a if (for some A) U wants A to rp t a partieular R-correlate of a to be . e.. And given the information from the linguist that 'Fido is shaggy' is the indicative version of a predication of 'shaggy' on 'Fido' (assumed). the following corre1ations hold: CI.): RPL To utter the indicative version of a predieation of ß on a if U wants A to think U to think a particular Rcorrelate of a to be one of a particular set of D-correlates of ß. To utter the indieative version of a predication of 'shaggy' on 'Fido' if U wants A to think U to think Jones' dog is one ofthe set ofhairy-eoated things (i. for U. To utter the indicative version of a predication of ß on 'Fido' if U wants A to think U to think Jones' dog to be one of a particular set of D-corre1ates of ß. 'For U. and P. we can infer that U has RP3. it may remain impossible to provide a definiens until the syntactical category of X has been given. indeed I am not certain that a general form of definition can be provided for this schema.. Further suppose that. To utter the indieative version of a if (for some A) U wants/intends A to think that U thinks .' [where the two lacunae represented by dots are identically completedl.2..AND WORD MEANING 65 P. ete. provide a definiens which may be adequate for adjectil'al X (e. P. 'shaggy'). is hairy-eoated). we can infer U to have RP4. I can. To utter 'Fido is shaggy' if U wants A to think U to think that Jones' dog is hairy-coated. and also C2.g.2. and P. however. 'U has this procedure: to utter a rp o'-eorrelated predieation of X on a if (for some A) U wants A to rp t a partieular R-corre1ate of a to be . Any specifie procedure of the form mentioned in the definiens of D.l.. I am not in a position to provide a definiens far 'X [non-completel means " ..

Since the act of correIation may be a verbal act. To be able to say that U had by V correlated 'shaggy' with each member of the set of hairy-coated things. is required. i.e.66 UTTERER'S MEANING. [The expanded definiens for the complete utterance-type 'He is shaggy' may be expected to contain the phrase 'a particular R-correlate of "he". in some system of pairing in which names and objects form ordered pairs. Certainly by uttering V. by uttering V he should effect this. What is it to mean to say that e. by the same act. Jones' dog is the/a R-correlate of 'Fido'? One idea (building in as little as possible) would be to think of 'Fido' and Jones' dog as paired. U has set up a situation in which a relation R holds exdusively between 'shaggy' and each hairy-coated thing Z. the second the object). We must now stop taking for granted the notion of correlation. Let us take this to consist in performing some act as a result of which a linguistic item or a non-linguistic item (or items) come to stand in a relation in which they did not previously stand. It is dear that condition (b). (b) U uttered V in order that. regardless of his intentions. and that condition (a) aIone would be inadequate. on which some will look askance because it introduces a reference to U's intention in performing his act of correlation. But..g. But. U has . 0. 'X is complete' =df. I can now ofTer adefinition of the notion of a complete utterance-type which has so far been taken for granted. and only to hairy-coated things. SENTENCE MEANING one of the set of hairy-coated things. We want a sense of 'paired' in which 'Fido' is paired with Jones' dog but not with Smith's cat. 'Selected pair'? But what does 'selected' mean? Not 'selected' in the sense in which an apple and an orange may be selected from a dish: perhaps in the sense in which a dog may be selected (as something with which (to which) the selector intends to do something). that for U 'shaggy' means 'hairy-coated'. do what? And what is the process of selecting? I suggest we consider initially the special case in which linguistic and non-linguistic items are explicitly correlated. how can this set up a relation between items? Suppose U produces a particular utterance (token) V.8.) Correlation. U efTected that 'shaggy' stood in R to each hairy-coated thing. and in which neither stands to noncorrelates in the other reaIm. But in one sense of 'pair' any one name and any one object form a pair (an ordered pair. 'A fully expanded definiens for "X means' . other than that involved in speaking of an Rcorrelate of some referring expression occurring within X'. we should need to be able to say that there is some relation R such that: (a) By uttering V. '" contains no explicit reference to correlation. which belongs to the utterance-type 'shaggy: hairy-coated things'. the first member of which is the name.. in the case of the word-thing pair. namely the relation which consists in being an expression uttered by U on a particular occasion 0 in conversational juxtaposition with the name of a cIass to which Z belongs.

AND WORD MEANING

67

also set up a situation in which another relation R' holds exdusively between 'shaggy' and each non-hairy-coated thing Z', namely the relation which consists in being an expression uttered by U on occasion 0 in conversational justaposition with the name of the complement of a dass to which Z' belongs. We do not, however, for our purposes, wish to think ofU as having correlated 'shaggy' with each non-hairy-coated thing. The only way to ensure that R' is eliminated is to add condition (b), which confines attention to a relationship which U intends to set up. It looks as if intensionality is embedded in the very foundations of the theory of language. Let us, then, express more formally the proposed account of correlation. Suppose that V =utterance-token of type' "Shaggy" : hairy-coated things' (written). Then, by uttering V, U has correlated 'Shaggy' with (and only with) each hairy-coated thing= (3R) {(U efTected by V that (Vx) (R 'Shaggy' x= xy (y is a hairy-coated thing») & (U uttered V in order that U efTect by V that (Vx) ... )1. If so understood, U will have correlated 'shaggy' with hairy-coated things only if there is an identifiable R' for which the condition specified in the definiens holds. What is such an R'? I suggest R'xy == x is a (word) type such that V is asequenceconsisting of a token of x followed by a colon followed by an expression l'hairy-coated things'l the R-correlate of which is a set ofwhich y is a member. R'xy holds between 'shaggy' and each hairy-coated thing given U's utterance of V. Any utterance V' of the form exemplified by V could be uttered to set up R"xy (involving V' instead of V) between any expression and each member of any set of non-linguistic items. There are other ways of achieving the same effect. The purpose of making the utterance can be specified in the utterance: V = utterance of 'To effect that, for some R, "shaggy" has R only to each hairy-coated thing, "shaggy" : hairy-coated things'. (The expression of the specified R will now have 'V is a sequence containing' vice 'V is a sequence consisling 0/ .. .'.) Or U can use the performative form: 'I correlate "shaggy" with each hairy-coated thing'. Utterance of this form will at the same time set up the required relation and label itself as being uttered with the purpose of setting up such a relation. But by whichever form an act of explicit correlation is effected, to say of it that it is (or is intended to be) an act of correlation will always be to make an indefinite reference to a relation(ship) which the act is intended to set up, and the specification of the relation involved will in turn always involve a further use of the notion of correlation (e.g. as above in speaking of a set which is the corre1ate (R-correlate) of a particular expression (e.g. 'Hairy-coated things'». This seems to involve a regress which might weil be objectionable; though 'corre1ation' is not used in definition of corre1ation, it will be used in specification of an indefinite reference occurring in the definition of corre1ation. It might be considered desirable (even necessary) to find a way of stopping this regress at some stage. (Is this a characteristically empiricist demand ?) If we don't stop it, can correlation even get started (if prior correlation is presupposed ?). Let us try 'ostensive correlation'.

68

UTTERER'S MEANING, SENTENCE MEANING

(Acts 1, 2, 3, etc.) U ostends

Objects{:~,} simultaneously with each ostension

a J, uttering 'shaggy' (intending to ostend only objects which are hairy-coated). For the combination of these acts to constitute a case of correlating 'shaggy' with each hairycoated thing, it must be the case that: (!:IR) (U effected, and intended to effect, by acts 1, 2, 3, etc. that (Vy) ('Shaggy' has R to y if and only if y is hairy-coated». How is the appropriate relation to be specified'! As folIows: R'xy (for some F) Iviz. being hairy-coatedJ (U ostended and intended to ostend only objects which are Fand, in acts 1,2,3, etc. accompanied each ostension by uttering a token of x; and y is F). Given the ostensions, R'xy holds between 'shaggy' and each hairy-coated thing, and the specification of R'xy at least seems not to involve further reference to correlation. So far, we have been acting on the assumption that the correlations, which in association with initial procedures yield further procedures, are explicit correlations; that is to say, that they are correlations set up by some identifiable and dateable act of correlating. But this assumption is clearly artificial. Many correlatio!1s, referential as well as denotative, seem to grow rather than to be created. The situation seems to be as follows: (1) We need to be able to invoke such aresultant procedure as the following, which we will call RPI2, namely, to predicate ß on 'Fido', when U wants A to I/J t that Jones' dog is a D-correlate of ß; and we want to be able to say that at least sometimes such aresultant procedure may result from among other things a non-explicit R-correlation of 'Fido' and Jones' dog. (2) It is tempting to suggest that a non-explicit R-correlation of 'Fido' and Jones' dog consists in the fact that U would, explicitly, correlate 'Fido' and Jones' dog. (3) But to say that U would explicitly correlate 'Fido' and Jones' dog must be understood as an elliptical way of saying something of the form 'U would explicitly correlate "Fido" and Jones' dog, if p'. How is 'p' to be specified? (4) Perhaps 'If U were asked to give an explicit correlation for "Fido" '. But ifU were actually faced with arequest, he rnight quite weil take it that he is being asked to make a stipulation in making which he would have an entirely free hand. If he is not being asked for a stipulation, then it must be imparted to hirn that his explicit correlation is to satisfy some non-arbitrary condition. But what condition can this be? Again it is tempting to suggest that he is to make his explicit correlation such as to match or fit existing procedures. (5) In application to RPI2, this would seem to amount to imposing on U the demand that he should make his explicit correlation such as to yield RP 12. (6) In that case, RP12 results from a non-explicit correlation which consists in the

AND WORD MEANING

69

fact that U would explicitly correlate 'Fido' and Jones' dog if he wanted to make an explicit correlation which would generate relevant existing procedures, viz. RPl2 itself. There is an apparent circularity here. Is this tolerable? (7) It may be tolerable in as much as it may be a special case of a general phenomenon which arises in connection with the explanation oflinguistic practice. We can, if we are lucky, identify 'linguistic rules', so called, which are such that our linguistic practice is as if we accepted these rules and consciously followed them. But we want to say that this is not just an interesting fact about our linguistic practice, but an explanation of it; and this leads us on to suppose that 'in some sense', 'im pli city' , we do accept these rules. Now the proper interpretation of the idea that we do accept these rules becomes something of a mystery, if the 'acceptance' of the rules is to be distinguished from the existence of the related practices; but it seems like a mystery which, for the time being at least, we have to swallow, while recognizing that it involves us in an as yet unsolved problem.

CONCLUDING NOTE

It will hardly have escaped notice that my account of the cluster of notions con-

nected with the term 'meaning' has been studded with expressions for such intensional concepts as those of intending and of believing; and my partial excursions into symbolic notation have been made partly with the idea of revealing my commitment to the legitimacy of quantifying over such items as propositions. I shall make two highly general remarks about this aspect of my procedure. (l) I am not sympathetic towards any methodological policy which would restrict one from the start in an attempt to formulate a theory of meaning in extensiona1 terms. It seems to me that one should at least start by giving oneself a free hand to make use of any intensiona1 notions or devices which seem to be required in order to solve one's conceptual problems, at least at a certain level, in ways which (metaphysical bias apart) reason and intuition commend. If one denies oneself this freedom, one runs a very serious risk of underestimating the richness and complexity of the conceptua1 field which one is investigating. (2) I said at one point that intensiona1ity seems to be embedded in the very foundations of the theory of language. Even if this appearance corresponds with rea1ity, one is not, I suspect, precluded from being, in at least one important sense, an extentionalist. The psychological concepts which, in my view, are needed for the formulation of an adequate theory of language may not be among the most primitive or fundamenta1 psychological concepts (like those which apply not only to human beings but to quite lowly anima1s as weil); and it may be possible to derive (in some relevant sense of 'derive') the intensiona1 concepts which 1 have been using from more primitive extensiona1 concepts. Any extensionalist has to deal with the problem

70 UTTERER'S MEANING. SENTENCE MEANING of allowing for a transition from an extensionaI to a non-extension al language. . and so presupposes the concepts in terms of which the use of language will have to be understood. and it is by no means obvious to me that intensionality can be explained only via the idea of conceaIed references to language.

Fodor. 170-210 (1963). in the following steps: (I) discussion of general background assurnptions and goals that underlie and motivate much of the work in generative gramm ar of the past decade. M. (Mouton & Co. therefore. 18. Halle. misunderstanding. relate only to questions of goals and interests and are therefore no more subject to debate than the question : is chemistry right or wrong? In the light of this. and which. the editor. Reprinted by permission of the author. 1966). B. Syntactic Structures (The Hague. a misunderstanding as to which elements of this framework express substantive assumptions about the nature of language and are. Lees. on the other hand. 54-72 (1962). Word. From Topics in the Theory 0/ Generative Grammar (in Janua Linxuarum). Chomsky. or equivocation of one sort or another. pp. 7-24 and 51-75. within the general framework of transformational generative grammar. on a rather substantial scale. (3) presentation of a theory of generative grammar of a sort exemplified.. matters of legitimate controversy and rational discussion. for example. I still hope to be able to incorporate an exposition (much abbreviateJ) of some recent work. 'The Structure of a Semantic Theory'. 1960). of the general framework I had expected to presuppose--in particular. pp.v TOPICS IN THE THEORY OF GENERATIVE GRAMMAR NOAM CHOMSKY (a) ASSUMPTIONS AND GOALS My original intention was to use these lectures to present some recent work on general linguistic theory and on the structure of English. designed to overcome these difficulties. The Hague. 39. a sequence of recent publications has indicated that many points that I had hoped to take for granted are widely regarded as controversial. (4) discussion of various real inadequacies that have been exposed in this position in work of the past half-dozen years. and has also indicated misunderstanding. . However. in N. Lg. The Grammar 0/ English Nominalizations (Bloomington. pp. 'Phonology in a Generative Grammar'. 1957). R. Katz and J. and (5) sketch of a refined and improved version of this theory. it seems advisable to change my original plan and to spend much more time on background assurnptions and general questions of various sorts than I had at first intended. and the publishers. but I will lead up to it more slowly. and J. (2) discussion of various objections to this general point of view that seem to me to be based on eITor.

who asserts that obviously I could not 'be said to sympathize with such a "mentalistic monster" as the "innere Sprachform '''. sense of this word-that is. Halle. however. Lingua 10. concentrating large1y on syntax. pp. in so far as they are known to us. H. This view is incomprehensible to me. as will surely be evident to anyone familiar both with Humboldt and with recent work in generative grammar (for explicit discussion. and H. 'Some Controversial Questions in Phonological Theory'. consider A. but volume includes section land section 1II of the original marked (a) and (b) rcspectively led.l. and F. In particular. I will also discuss a few criticisms ofthis approach. and section III with the third.. 1966). The Sound Pattern Qf Russian (The Hague. and has been presented invariousstagesofdevelopment in publications of Halle's and mine (listed in the bibliography below) since then. In the course of this paper I will also make a few remarks about historical backgrounds for the position that will be outlined.. in considerable detail e1sewhere. 3 This matter is discussed in more detail in Chomsky. 1959). I. This is true--apparently to an extent that many critics do not realize. pp. not the Bloomfieldian. 4 To cite just one example. Halle. W. 1964). it is an attempt to construct a theory of mental processes). 65-80 (The Hague. see Chomsky. entire1y false. 97-138 (1965). 4 I differ from them only in regarding this observation not as a criticism. But in fact the work that he is discussing is quite explicitly and selfconsciously mentalistic (in the traditional.. Much more perceptive are those critics who have described this work as in large measure areturn to the concems and often even the specific doctrines of traditional linguistic theory. Householder Jr. this essay contains no new or original material. 13-34 (1965). 2 in which questions touched on here are dealt with more thoroughly. and as an attempt to clarify issues that have been raised in critical discussion. section 11 with the second. 1965). Katz and P. and will.g. Journal 0/ Linguistics. 2 e. Mass. M. 3 I. or that it perhaps constitutes some obscure branch of mathematics. and in Cartesian Linguislics (New York. Current Issues in Linguistic Theory.* In the final section I will discuss an approach to the study of sound structure that has been gradually evolving since Chomsky. Reichling. it is intended only as an informal guide to other books and papers. Ferguson's review of HaUe. be quite accurately described as an attempt to develop further the Humboldtian notion of'form of language' and its implications for cognitive psychology. pp. Section I will deal with the first point. 3 Quite a few commentators have assumed that recent work in generative grammar is somehow an outgrowth of an interest in the use of computers for one or another purpose. Currenllssues in Linguislic Theory. Journal 0/ Linguislics. 1964). For Roman Jakobson. MacLean. Current Issues in Linguistic Theory. and it can. pp. which deals with criticisms in C. which deals with objections raised by F. 'On Accent and Juncture in English'. 105-7 (The Hague. and in Chomsky and Halle.72 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR I will try to cover these points in the first three sections. eds. or that it has some other engineering motivation. 'Principles and Methods of Syntax: Cryptanalytical Formalism'. and it is. A. hopefully. I. Ch. Postal. soon emerge to fulliight of day in a book that is now in active preparation. Lunt. I * This . I In general. pp. fourth and fifth. in any event. In the course of this presentation. Mass. § I. 'On Some Rccent Claims in Phonological Theory'. and Aspects 0/ the Theory 0/ Syntax (Cambridge. see the references cited above). in AspeclS o/lhe Theory 0/ Syntax. furthermore. Chomsky. The discussion of criticisms will be very brief. 1-7 (1961). I. Lukoff. 1956). An Integrated Theory 0/ Linguislic Description (Cambridge. § 8. since Halle and I havediscussed most of them.

in the traditional view. 5 Notice. is an account of competence. The competence of the speaker-hearer can. a primary interest in competence entails no disregard for the facts of performance and the problem of explaining these facts. It is I will not consider Reichling's criticisms of generative grammar here. and specific content of the work that he was discussing. the problem for linguistic theory is to discover general properties of any system of rules that may serve as the basis for a human language. Performance provides evidence for the investigation of competence. Chomsky. A specls of I he Theory of Synlax. concerns. I. Furthermore.TOPICS IN THE THEORY OF GENERATIVE GRAMM AR 7J rather as adefinite merit of this work. ~ 2. Miller and Chomsky.vol. Ch. R. and. the general form of language that underlies each particular realization. Handbook 0/ MalhemalicaIPsychology. It describes and attempts to account for the ability of a speaker to understand an arbitrary sentence of his language and to produce an appropriate sentence on a given occasion. Ch. The cited remark is just one illustration of his complete lack of comprehension of the goals. eds. it attempts to provide the student with this ability. 1. 'For discussion. D. it seems to me that it is the modern study of language prior to the explicit study of generative grammar that is seriously defective in its failure to deal with traditional questions and. there is no reason to expect hirn to be fully aware even of the empirical consequences of these internalized rules---that is. in fact. At the same time. On the contrary.lI. it is difficult to see how performance can be seriously studied except on the basis of an explicit theory of the competence that underlies it. and his discussion is based on such gross misrepresentation of this work that comment is hardly called for. A. of the way in which signals are assigned semantic interpretations by the rules of the language that he knows (and. by definition. nor. Aspecls of the Theory of Syntax. that is. 'Finitary Models of Language Users'. A grammar. ideally. see G. R. On the difficulties of becoming aware of one's own linguistic intuitions. it aims to discover and exhibit the mechanisms that make this achievement possible. incidentally. A distinction must be made between what the speaker of a language knows implicitly (what we may call his competence) and what he does (his performance). see the discussion in Chomsky. to elaborate in detail what we may call. in traditional terms. and E. Galanter (New York. If it is a pedagogic grammar. contributions to the understanding of performance have largely been by-products of the study of grammars that represent competence. be expressed as a system of rules that relate signals to semantic interpretations of these signals. to recognize the essential correctness of many of the traditional answers and the extent to which they provide a fruitful basis for current research. . that a person is not generally aware of the rules that govern sentence-interpretation in the language that he knows. The problem for the grammarian is to disco ver this system of rules. if a linguistic grammar. furthermore. each particular natural language. 1963). is there any reason to suppose that the rules can be brought to consciousness. in fact. Bush. That is. knows perfectly). Luce. § 4.

Modem linguistics. is seriously at fault in its failure to come to grips with this central problem. Returning now to the central topic. To carry the study of language beyond its traditional bounds. Jespersen) who have given some attention to the problem of creativity. This is the fundamental problem to which all work in generative grammar has been addressed. O. but rather give examples and hints that enable the intelligent reader to determine the grammar. Normal use of language involves the production and interpretation of sentences that are similar to sentences that have been heard before only in that they are generated by the rules ofthe same gramrnar. O. and thus the only sentences that can in any serious sense be called 'familiar' are diches or fixed formulas of one sort or another. The most striking aspect of linguistic competence is what we may call the 'creativity of language'. It is important to recognize that there is no sense of 'habit' known to psychology in which this characterization of language use is true Gust as there is no notion of 'generalization' known to psychology or philosophy that entitles us to characterize the new sentences of ordinary linguistic usage as generalizations of previous performance). It attempts to go beyond traditional grammar in a fundamental way. 1924». This is evident from the common description of language use as a matter of 'grammatical habit' (e. They do not actually formulate the rules of the grammar. They do not provide an analysis of the 'faculte de langage' that makes this achievement possible. In fact. The fundamental importance of this creative aspect of normal language use has been recognized since the seventeenth century at least. in some way that is not at all understood. The extent to which this is true has been seriously underestimated even by those linguists (e. even to speak of the hearer's 'familiarity with sentences' is an abusrdity. it is precisely what should be expected. an explicit . in fact. it is necessary to recognize this limitation and to develop means to transcend it. and it was at the core of Humboldtian generallinguistics. however.g. As has repeatedly been emphasized. a generative grammar (that is. and that they can be accepted only as metaphorshighly misleading metaphors. however. The familiarity of the reference to normallanguage use as a matter of'habit' or as based on 'generalization' in some fundamental way must not blind one to the realization that these characterizations are simply untrue if terms are used in any technical or well-defined sense.74 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR important to realize that there is no paradox in this. the speaker's ability to produce new sentences. sentences that are immediately understood by other speakers although they bear no physical resemblance to sentences which are 'familiar'. Current work in generative grammar has adopted this traditional framework of interests and concerns. traditional grammars make an essential appeal to the intelligence of the reader. that is. Philosophy 0/ Grammar (London.g. since they tend to lull the linguist into the entirely erroneous belief that the problem of accounting for the creative aspect of normal language use is not after all a very serious one. Jespersen.

seems to me to suggest new and interesting ways to reopen these tradition al questions. A particular linguistic theory must specify the set of possible syntactic descriptions for sentences of a natural language. that uniquely determines its semantic interpretation (the latter notion being left unspecified pending further insights into semantic theory)6 as weil as its phonetic form. hence two syntactic descriptions might difTer only in that they contain different members of a pair of homonymous morphemes. then the theory of generative grammar must provide a general.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 75 grammar that makes no appeal to the reader's 'faculte de langage' but rather attempts to incorporate the mechanisms of this faculty) is a system of rules that relate signals to semantic interpretations of these signals. Without going into any detail. The extent to which these syntactic descriptions meet the conditions that we know must apply to semantic interpretations provides one measure of the success and sophistication of the grammatical theory in question. to which I return in the third section. obviously.g. We have fairly reasonable techniques of phonetic representation that seem to approach adequacy for all known languages. the immediate prospects for universal semantics seem much more dirn. be drawn). in fact. Let us then introduce the neutral technical notion of 'syntactic description'. language-independent means for representing the signals and semantic interpretations that are interrelated by the grammars of particular languages. This fact has been recognized since the origins of linguistic theory. in particular cases. The fact that universal semantics is in a highly unsatisfactory state does not imply that we must abandon the programme of constructing grammars that pair signals and semantic interpretations. and take a syntactic description of a sentence to be an (abstract) object of some sort. If a generative grammar is to pair signals with semantic interpretations. distractions. memory !imitations. and PostaI. though surely this is no reason for the study to be neglected (quite the opposte conclusion should. recent work of Katz. we would regard a semantically ambiguous minimal element as constituting two distinct lexical entries . Fodor. of course. . the notion of syntactic "Working in this framework then. though. a great deal is known about conditions that semantic representations must meet.) that interact with underlying competence to produce actual performance. The idealization is (in particular) that in the study of grammar we abstract away from the many other factors (e. It is descriptil'ely adequate to the extent that this pairing corresponds to the competence of the idealized speaker-hearer. changes ofintention in the course of speaking. associated with the senten ce. In contrast. for several centuries). there is much to leam in this domain. As the theory of generative grammar has progressed. In fact. I think it would be widely agreed that the general problem of universal phonetics is fairly weil understood (and has been. etc. and traditionallinguistics made various attempts to develop theories of universal phonetics and universal semantics that might meet this requirement. For although there is little that one can say about the language-independent system of semantic representation. whereas the problems of universal semantics still remain veiled in their traditional obscurity.

Notice that a syntactic description (henceforth. and a theory of generative grammar must provide a general characterization of the dass of possible signals (a theory of phonetic representation) and the dass of possible SO's. indicate that John is related to incompetent as it is in 'John . it must assign to the pair disregard-John the same grammatical relation as in 'everyone disregards John'. quite apart from the question of how interpretations can be imposed on them [in the case of (2) and (3)] : (1) (2) (3) the dog looks terrifying the dog looks bar king the dog looks lamb A generative grammar. meat. this being the basis for the alternative semantic interpretations that may be assigned to this sentence. if a signal has two intrinsic semantic interpretations in a particular language [e. Thus we should expect a descriptively adequate grammar of English to express the fact that the expressions (1)--(3) are ranked in the order given in terms of 'degree of deviation' from English. in one SO. while in the other they are as in 'meat which is good tastes Adjective' (where the notion 'grammatical relation' is to be defined in a general way within the linguistic theory in question). whereas in the other it must assign this very same relation to the pair disregard-what (disturbed John). in the case of(6) and (7) only one SO should be assigned by a descriptively adequately grammar.76 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR description has been darified and extended. it will approach descriptive adequacy to the extent that these SO's succeed in expressing the basis for the ambiguity. SO) may convey information about a sentence beyond its phonetic form and semantic interpretation. must at least determine a pairing of signals with SO's. in the case of (5). for example. if the theory of generative gramm ar is to provide descriptively adequate grammars. (4) or (5). then. Similarly. in particular.g. to the extent that it pairs signals with SO's that do in fact meet empirically given conditions on the semantic interpretations that they support. in the case of (6). and must assign no semantically functional grammatical relation at aIl to disregard-John. A grammar is descriptively adequate to the extent that it is factually correct in a variety of respects. I will discuss below some recent ideas on just what should constitute the syntactic description of a senten ce. (4) (5) they don't know how good meat tastes what disturbed J ohn was being disregarded by everyone In the case of (4). On the other hand. in English). a grammar of this language will approach descriptive adequacy if it assigns two SO's to the sentence. a descriptively adequate grammar must not only assign two SO's to the sentence but must also do so in such a way that in one of these the grammatical relations of good. and. For example. This SO should. and taste are as in 'meat tastes good'. beyond this.

that is. which assigns a phonetic interpretation to a surface structure. Furthermore. the SD must indicate that our is re1ated to regard (as incompetent) as us is related to regard (as incompetent) in 'everyone regards us an incompetent'. each of which specifies the system of grammatical relations that underlies one of the distinct semantic interpretations of this senten ce: (8) the police were ordered to stop drinking after midnight. and that John is related to regard (as incompetent) as it is in 'everyone regards lohn as incompetent'. in detail. then. what disturbed John was our being regarded as incompetent by everyone. Hence the SD must (i) determine a semantic interpretation and (ii) determine a phonetic representation. in the case of (8). In the case of (7). must consist ofthree components: a syntactic component. a phonological component. Let us define the 'deep structure of a sentence' as that aspect of the SD that determines its semantic interpretation. each SD must uniquely determine the signal of which it is the SD (unique1y.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 77 is incompetent'. A gramrnar. Similarly. which assigns a semantic interpretation to a deep structure. as required.?crete. once again. The notions 'deep structure' and 'surface structure' are intended as explications of the Humboldtian notions 'inner form of a sentence' and 'outer form of a sentence' . Examples such as these should suffice to illustrate what is involved in the problem of constructing descriptively adequate generative grammars and developing a theory of gramrnar that analyses and studies in fuH generality the concepts that appear in these particular grammars. remains unclear. (6) (7) what disturbed John was being regarded as incompetent by everyone. It is quite evident from innumerable examples of this sort that the conditions on semantic interpretations are sufficiently clear and rich so that the problem of defining the notion 'syntactic description' and developing descriptive1y adequate gramrnars (relative to this notion of SD) can be made quite co. which generates SD's each of which consists of a surface structure and a deep structure. this association being mediated by the syntactic component that generates deep and surface structures as elements of SD's. a semantic component. Thus the gramm ar as a whole will associate phonetic representations and semantic interpretations. the grammar must assign four distinct SD's. in some way which. The SD assigned to a signal must deterrnine the semantic interpretation ofthe signal. up to free variation). A grammar. despite the fact that the notion 'semantic interpretation' itself still resists any deep analysis. We return to some recent ideas on semantic interpretation of SD's in section 111. must pair signals and SD's. and the 'surface structure of a sentence' as that aspect of the SD that determines its phonetic form.

categorizes these. Hockett has also used these terms (A Course in Modem Linguistics. p. is a surface structure that analyses it into its constituents (perhaps. we conc1ude that deep structures cannot be identified with surface structures. crucial for semantic interpretation. necessary to determine the phonetic representation of this sentence. each of these being assigned to a category indicated by the labelling. C. for example. simply be a labelIed bracketing. 'being regarded as incompetent by everyone').. Secondly.e. but the deep structures must obviously differ. An ambiguous sentence has several SD's. Such examples as these are sufficient to indicate that the deep structure underlying a sentence cannot be simply a labelIed bracketing of it. Chomsky.. only a single surface structure may be assigned. that the deep structure must be quite different from this surface structure. etc. Ch. The inability of surface structure to indicate semantically signiticant grammatical relations (i. 'was'. a full generative grammar must consist of a syntactic. for example. The labelIed bracketing can be presented in a tree-diagram. assigning each of these to a certain category indicated by the labelling. Since there is good evidence that the surface structure should. In summary. Current Issues in Linguistic Theory. The terminology is suggested by the usage familiar in contemporary analytic philosophy (cf. 29 (New York. in both its classical and modern varieties. in the case of an ambiguous sentence such as. in fact. The semantic component assigns a semantic interpretation to the deep structure and the phonological component assigns a phonetic interpretation to the surface structure. For one thing. differing in the deep structures that they contain (though the converse need not be true). 'what disturbed John'. then further segmenting each of these into its constituents (e. The syntactic component generates SD's each of which contains a deep structure and a surface structure. Thus underlying (6). for example.g. until ultimate constituents are reached. 'what disturbed John' into 'what' and 'disturbed John').78 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR (the general notion 'form' is probably more properly to be related to the notion 'generative grammar' itself-----cf. This discussion has so far simply delirnited a certain domain of interest and a certain c1ass of problems. section IV) to suppose that the surface structure of a sentence is a labelIed bracketing that segments it into its continuous constituents. to serve as deep structure) is one fundamental fact that motivated the development of transformational generative grammar. in fact. There is good reason (see below. The only substantive . 1958» in roughly the same sense. as we have just observed. Philosophical Investigations. semantic. and phonological component. the surface representation in no way expresses the grammatical relations that are. however. perhaps. Information of this sort is. for discussion). 1953». F. 168 (Oxford. So far I have said !ittle that is in any way controversial. etc. (5). or in other familiar notations. segments the constituents into further categorized constituents. Wittgenstein. and has suggested a natural framework for dealing with these problems. It is c1ear.

one can easily prove that a phrase-structure grammar of the language L assigns to sentences of L the structural descriptions assigned by some transformation al grammar of L. having set trus goal. It is difficult to see how a fuH generative grammar can be regarded. 'Logical Structure of Linguistic Theory'. Hence the condusions outIined so far seem inescapable if the the problem of studying linguistic competence is taken up. and. factual assertions) that I have so far made witrun the framework are that the surface structure is a labelled bracketing and that deep structures must in general be distinct from surface structures. and semantic interpretation. on the other. 1955.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 79 comments (i. The first of these assertions is weil supported (see below). let us reassure ourselves about the uncontroversial character of what has preceded. anything in this account to which exception can be taken? Surely there is no conceivable question about the necessity for distinguisrung competence from performance in the way suggested above. on the one side. For example. ultimately. an essential part of the theory of phrase-structure grammar is a particular specification of how categories and relations are determined for generated strings (see Chomsky. To go on from here to develop a substantive Iinguistic theory we must provide: (9) (i) theories of phonetic and semantic representation (ü) a general ac count of the notion 'syntactic description' (iii) a specification of the dass of potential generative grammars (iv) a general account of how these grammars function. and such a specification has been presupposed whenever this theory has been investigated. as anything other than a system of rules that relate signals to semantic interpretations. Having made this distinction. he must immediately face the fact of 'creativity' and must therefore focus attention on the problem of constructing generative gramm ars. one may or may not choose to be interested in the general question of accounting for linguistic competence. A change in trus specification is as much arevision of the theory as a change in the specification of the dass (9iii) of potential grammars. The second is surely much too obvious to require elaborate defence. Ch. in fact. Failure to understand trus leads to immediate absurdities.e. one is immediately faced with the problem of developing a rich enough notion of 'syntactic description' to support phonetic interpretation. . This point should be obvious without further discussion. The distinction between deep and surface structure emerges from even the most superficial examination of real linguistic material. how they generate SD's and assign to them phonetic and semantic interpretations. Notice that a substantive linguistic theory involves a specification of (9iv) as weil as (9üi). Cambridge. etc. thus pairing phonetically represented signals with semantic interpretations. Thus if one thinks of the theory of'phrasestructure' with the technique of interpretation (9iv) left free. Is there. VI). Before going on to discuss these substantive questions. that is. If one chooses to concern rumself with trus question. and would probably be widely accepted.

P. M . 'Explanatory Models in Linguistics'. there is no reason to ass urne that apriori semantic concepts play a role in determining tbe organization of the syntactic component of a grammar. G Perhaps this matter can be clarified by considering examples of the latter sort.. Suppes. The nature of this question can be appreciated readily in terms of the problem of constructing a hypothetical language-acquisition device AD that can provide as 'output' a descriptively adequate gramm ar G for the language L on the basis of certain primary linguistic data from L as an input. if any of these proposals is adopted.g. Chomsky. it would be absurd to study semantics (and similarly. since. for example. is whether any interesting result is likely to be attainable under such arbitrary limitation of subject matter. An Inlegraled Theory 01 Lingllistic Description. pp. and Aspects of the Theory of Syntax). concomitantly. And it would be absurd to develop general syntactic theory without assigning an absolutely crucial role to semantic considerations. Syntactic Structures. On the other hand. . and. Review of Chomsky. the tentative position that the linguist accepts may have an important inftuence on the character of the work that he does. it seems to me. pp. As has been repeatedly emphasized.g. e. Lees. that is. Logic. there seems to be no non-trivial relation between the systems of phonological and semantic interpretation and no significant way in which semantic considerations can playa role in phonology or phonological considerations in semantics. since. 528-50. It is. alternatively.80 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR Suppose that one chooses not to study linguistic competence (and. One might. For discussion of these points. phonology) in isolation from syntax. F ar 100 little care has been taken in the discussion of these questions in modern linguistics. 375-408 (1957). 1962). The only question that arises. surface) structure. E. The issues are important. Methodology.. while no answers to any of these questions can be given with any certainty. and many dogmatic claims have been voiced and repeatedly echoed with no attempt to justify or support them by serious argument. As a result. it is quite reasonable to study semantics in isolation from phonology or phonology in isolation from semanties. or to surface structures... at the moment. choose to limit attention to performance. and A. I. since the syntactic interpretation of a sentence (similarly. Katz and Postal. In each of the cited cases it seems quite ununlikely. at the moment. It is essential also to raise the question of 'explanatory adequacy' of linguistic theory. there has been much confusion about them.. and many other rcfcrences. unclear why anyone should insist on studying an isolated aspect of the general problem of grammatical description unIess there is some reason to believe that tbis is not affected by the character of other aspects of grammar. and Philosophy of Science. it is aprerequisite even for the study of descriptive adequacy that they be set higher than this.. its phonetic interpretation) depends in an essential way on its deep (respectively. a device represented schematically as (10): (10) 1 primary linguistic data . SYlltartic Struclllres. Current Issues in Linguistic Theory. in general. or to sound patterns in isolation from syntactic structure. see Chomsky. or to voiced fricatives. it seems quite reasonable to develop a theory of syntactic structure with no primitive notions of an essentially semantic nature.7 I have been discussing so far only the question of descriptive adequacyofgramrnars and the problem of developing a linguistic theory that will provide the basis for the construction of descriptively adequate grammars. since obviously the necessity to support semantic interpretation is one of the primary requirements that the structures generated by the syntactic component of a grammar must meet. or to first halves of sentences. Syntactic Structures and Current Issues in Linguistic Theory. however (see.. Similarly. in fact. Nagel. linguistic performance within the framework of a theory of competence). the goals of linguistic theory can be set much higher than this. Thus. eds. Tarski (Stanford.

This specification can then be attributed to the system AD as an intrinsic property. in other words.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 81 We naturally want the device AD to be language-independent---that is. To pursue the study of explanatory adequacy. We must ask whether the proposed evaluation procedure in fact can mediate tbis empirically given relation. To support or refute this hypothesis. This procedure will then enable the device to select one of the apriori possible hypotheses--one of the permitted grammars---that is compatible with the empirically given data from a given language. the vital importance of raising this problem and pursuing it intensively at every stage of linguistic investigation also seems to me entirely beyond the possibility of debate (cf. Were we able to develop the specifications for a language-acquisition device of tbis sort. about the prerequisites for language-acquisition. This explanation would be based on the assumption that the specifications of the device AD provide the basis for languageacquisition. An evaluation procedure. we must attempt to provide as narrow a specification of the aspects of linguistic theory listed in (9) as is compatible with the known diversity of languageswe must. to provide an implicit definition of the notion 'human language'. a hypothesis. in other words. Having selected such a hypothesis. which will enable it to select a particular member of the dass of grammars that meet the specifications (9) (or. as it is often called in technical discussion) is itself an empirical hypothesis concerning universal properties of language. it is. it has 'mastered' the language described by this grammar (and it thus knows a great deal beyond what it has explicitly 'learned'). We want it. we can explain some of the aspect of the speaker's competence whenever we can show with some plausibility that this aspect of his competence is determined by the most highly valued grammar of the permitted sort that is compatible with data of the kind to which he has actually been exposed. we may attempt to develop a general evaluation procedure. to select a small set of alternatives. conceivably. Notice that an evaluation procedure (simplicity measure. we could realistically daim to be able to provide an explanation for the linguistic intuition-the tacit competence--of the speaker of a language. capable of learning any human language and only these. the references cited above for elaboration of this point). primary linguistic data from some language providing the empirical conditions under wbich the development of a generative grammar takes place. Second. in other words. Given a linguistic theory that specifies (9) and an evaluation procedure. though tbis abstract possibility is hardly worth discussing for the present) on the basis of the presented primary linguistic data. as an intrinsic property of AD. we may proceed in two parallel ways. First. has . we must consider evidence as to the factual relation between primary linguistic data and descriptively adequate grammars. develop as rich a hypothesis concerning linguistic universals as can be supported by available evidence. therefore. true or false. The difficulties of developing an empirically adequate language-independent specification of AD are too obvious to require extended discussion.

I (1965). that is) by data from L. the second as a hypothesis about the tacit knowledge that emerges with exposure to appropriate experience. including. Choice of a descriptively adequate gramm ar for the language L is always much underdetermined (for the linguist. Furthermore. His evaluation measure involved minimization of features in the lexicon and the phonological rules. One may or may not choose to deal with the problem of explanatory adequacy. . as an essential part. it is important to recognize that there is nothing controversial in what has just been said. the fonn of the grammar of Land the evaluation procedure that leads to the selection of this grammar.82 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR much the status of a physical constant. it is impossible to support or reject a specific proposal on the basis of apriori argument. that is. in advance. 8 His situation. He must show that the delimitation of interest leaves him with a viable subject. This approach is quite natural. surely will) find that he has eliminated from consideration one of the most important sour ces of evidence bearing on the problems that remain (in particu1ar. but only if the linguist has an explanatory theory of the sort just sketchcd. 'On Some Recent Claims in Phonological Theory'. A crucial aspect of this theory was its complete reliance on distinctive features in the formulation of phonological rules to the exclusion of any 'segmental' notation (e. Hut. in any event. concerning the general 'faculte de language' that makes language-acquisition possible. the linguist comes to a conclusion about the speakers of L on the basis of an independently supported assumJT tion about the nature of language in general--an assumption. the problem of descriptive adequacy). Other relevant data can be adduced from study of descriptively adequate gramm ars of other languages. In this way. it permits data from other languages to playa role in justifying the grammar selected as an empirical hypothesis conceming the speakers of L. in my opinion. in my opinion) delimited one facet. 13-34) of several proposals ofHalle's regarding an appropriate evaluation procedure for phonology. he surely has no basis for objecting to the attempt on the part of other linguists to study the general question of which he has (artificially. phonemic notation) except as an informal expository device. may be quite analogous to that of the person who has decided to limit his attention to surface structures (to the exclusion of deep structures) or to first halves of sentences. Once again. a certain empirical hypothesis regarding a simplicity measure. Journal of Linguistics. In support of this theory he showed that a variety of facts can be explained on these assumptions. consider Householder's criticism (Householder. The reason for this is quite simple.g. it prescribes. Halle presented a certain theory of phonological processes. He also discussed alternative theories that use segmental notation along with or instead of feature notation and gave several arguments to show that under these assumptions it is difficu1t to see how any empirically valid evaluation measure can be formu1ated . I hope that these remarks will be sufficient to show the complete pointlessness of much of the debate over the specific evaluation procedures (simplicity measures) that have been proposed as empirical hypotheses concerning the form of language in the course of work in generative grammar. Following it. given data. To mention just one example. The general explanatory theory of language and the specific theory of a particular language that results from application of the general theory to data each has psychological content. in particular. One who chooses to overlook this problem may (and. Such a theory can receive empirical support from its success in providing descriptively adequate grammars for other languages. pp. then. the first as a hypothesis about innate mental structure.

in particular. This distinction between 'simplicity' as an absolute notion of general epistemology and 'simplicity' as apart of a theory of grammar has been repeatedly emphasized. Aspects 0/ the Theory 0/ Syntax. and five. much more briefly. much of the criticism of re cent attempts to develop valid evaluation measures is based on similar presuppositions. with their consistent reliance on feature representation. etc. The earliest versions of transformational generative grammar made the following general assumptions conceming syntactic structure.e. Householder makes no attempt to refute these arguments but simply objects to them because they fail to meet certain apriori conditions that he arbitrarily imposes on any notion of 'evaluation procedure'. Since the grammars that Halle proposes. he showed how various rather natural measures involving minimization fail on empirical grounds. Such a theory is an empirical hypothesis. Householder's critique has no relevance to any issue that Halle discusses. to parts three. Consequently. proposed to account for some domain of linguistic fact. that there is an interesting but poorly understood sense in which one can talk of the 'simplicity' or 'elegance' or 'naturalness' of a theory (of language. I would like to turn. incidentally. require more symbols than grammars that use auxiliary symbols as abbreviations for feature sets. and since Halle's grammars are (Householder claims) not easy to read. The syntactic component of a . Failure to make this distinction vitiates most of the criticism of evaluation procedures that has appeared in 'recent years. but this 'absolute' sense of simplicity has no c1ear relevance to the attempt to develop an evaluation measure (a simplicity measure) as apart of a theory of grammar. he concludes that the theory on which they are based must be mistaken. nevertheless.TOPICS IN THE THEOR Y OF GENERATIVE GRAMMAR 83 ---in particular. confusion regarding this point has. The 'simplicity measure' that it contains is a constituent part of this empirical hypothesis. These are discussed in much more detail in Chomsky.). about the structure of a general language-acquisition device of the sort described above). four. been quite widespread. (b) THE THEORY OF TRANSFORMATIONAL GENERATIVE GRAMMAR HA VING now covered the first two parts of the outline given in the introductory section. and in the references cited there. of the chemical bond. true or false. Notice. Hut c1early apriori arguments of this sort have no bearing on an empirical hypothesis about the nature of language (i. the requirement that such a procedure must favour grammars that use fewer symbols and that are easy for the linguist to read. Unfortunately.

The specification of the transformation is completed by associating with this structural analysis a certain elementary transfonnation which is a formal operation on strings. The transformational rules map phrase-markers into new. this being representable as a labelIed bracketing or a tree diagram with categories labelling the nodes. until the result is a single phrase-marker dominated by S (the initial category. the system of rewriting rules (let us call this the base component of the syntax) assigns to it a phrase-marker which we can call a base phrase-marker. Each transformational rule is defined by a structural analysis stating a condition on the dass of phrase-markers to which it applies and specifying an analysis of the terminal string of this phrase-marker into successive parts. and the 'product' of two phrase·markers is then the complex product of the two sets (i.9 we can app\y what have been called generalized (or double base.. . To illustrate. of a certain narrow dass..T. mapping such a produet into a new phrase-marker by the same apparatus as is required in the singulary ease. representing 'sentence'). derived phase-markers. The rewriting rules constitute a phrase-structure grammar (with.e. consider the sentence (18) I expected the man who quit work to be fired.) transformations to a phrase-marker representing a sequence of phrase-markers. a condition of linear ordering imposed). In the course of generating astring.. in other words. we seleet a sequenee of (one or more) base phrase-markers and apply singulary and generalized transformations to them. The strings generated by this system we may call base strings (an alternative term is C-tenninal strings). We can represent the system of transformations that apply in the process of derivation as a transformation-marker (T-marker). Microfilm M. The transformations meet certain ordering conditions (I return to these below). which must be stated in a separate part of the grammar. Each rule is. In particular.J. unpublished manuscript. For details. tripie base ete. we eall the resulting sentence a kernel sentence (a kernel sentence is not to be confused with the base string that underlies it as weIl as possibly many other more complex sentences). see the references cited above. 'The Logical Structure of Linguistic Theory'. By defining the 'product' of two phrase-markers as the new phrase-marker derived essentially by concatenation of the labelled bracketings.. Library (Cambridge. where A is a category symbol and X. Ware strings of category or terminal symbols. of the form A .. Z. 1955).84 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR grammar consists of two sorts of rules: rewriting rules and transformational rules. and descriptions of varying degrees of informality appear throughout the literature. To generate a sentence. If we select a single base phrasemarker and apply only obligatory transformations. X (with a possible restrietion to the context Z-W). observing the ordering and obligatoriness requirements. 9 Precise definitions of the notions mentioned here are provided in Chomsky. These eonditions inc1ude a specification of certain transformation as obligatory or obligatory relative to certain sequences of transformations. a phrase· marker is representable as a set of strings. perhaps. Mass. the set of all strings XY such that X is in the first set and Y in the second).

this is not necessary. Call this new structure K \. and others. BI' B 2 and B 3 are the three base phrase-markers that underlie the (kernei) sentences (20i)-(20iii): 10 (19) (20) (i) I expected it (ü) someone fired the man (iü) the man quit work The interpretation of (19) is straightforward. in the case of the analysis of such sentences as (15» which are either deleted or filled in by sentence transforms in one way or another. instead of the abstract strings that underlie them. apply the passive transformation T Pass to give 'the man who quit work was fired by someone'.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 85 The transformational derivation of (18) might be represented by the T-Marker (19). apply the generalized embedding transformation T Emb to the pair of structures (B 2' K I)' deleting the occurrence of the man in the embedded sentence in the process. leite sentences as examples. and in the transformational gramm ars presented in Chomsky. underlies a kernel sentence. I am completely overlooking the question of how to describe the Auxiliary system. might have a dummy symbol as Object. The reasons for this go weU beyond anything considered here. and proceed as follows. for simplicity of exposition.onverts it to 'who (the man) quit work' (rather.. Throughout the deseription of these struetures. etc.I. At this point. See P. Thus B. It should be kept in mind that this is only an expository device. 2). with its derived phrase-marker.-B.g. 'A Grammar of English Predieate Complement Construetions'. B. Chomsky. 'A Transformational Approach to Syntax'. Syntactic Structures. It represents the fact that to form (18) we take the three base structures underlying (20i-iii). unpublished Ph. Actually. that each of B . apply to B 3 the relative transformation T Rel that c. Ch. In this representation. Rosenbaum. and I have also supposed. n. with the derived phrase-marker K 4 • Now apply to the pair of structures (B I' K 4 ) the generalized embedding transformation T Emb' giving 'I expected the man who quit work was fired' with the derived phrase-marker K 5' To K 5' apply the singulary transformation T to giving the sentence (18) with its derived phrase-marker K 6 • 10 Since I am presenting this merely as the basis for some revisions to be proposed below. I am also assuming here a simpler analysis of the main (matrix) structure than was postulated in earlier work. The Grammar 0/ English Nominalizations. for further related diseussion. to the abstract string that underlies this----d. Camp. many of the base strings contain 'dummy symbols' (e. inaccurately.T.D. 'The Logical Structure of Linguistic Theory'. 1. dissertation (M. and. Aspects 0/ the Theory 0/ Syntax. giving the string 'someone fired the man who quit work' with its derived phrase-marker K 2• To K 2 . with the phrase-marker K 3' To this apply the deletion transformation T Dei to give 'the man who quit work was fired'. § 4. First. I skip many details. Lees. In particular. . might have an unspecified Subject. 1965).

. and state precisely how a syntactic description consisting of a deep and surface structure is gerrerated by the syntactic rules. the phonetic representation which is the marker on the lowest. a specification of the base phrase-markers from which it is derived. Thus in the case of (18). Each level is a system of representation in terms of certain primes (elementary atomic symbols of this level). a strin~ of elements that constitutes an 'output' of the syntactic component of the grammar and an 'input' to the phonological component. representing the inputs to the semantic and phonological components of a gramm ar respectively (see above). with phrase structure and transformations each constituting a linguistic level. A marker is astring of primes or a set of such strings. On the level of phrase structure. The rewriting rules of the base component and the rules governing ordering and arrangement of transformations generate an infinite dass ofT-markers. phonetic level--the primes of this level being sets of features). in the manner just sketched. and we may think of the markers of each level as being mapped into the markers of the next lowest level and as representing the lowest level marker (that is. from this. It should be dear.e. An Integrated Theory 0/ Linguistic Description). On each level. the deep structure is the T-marker represented as (19). markers are constructed that represent a sentence. the primes are category and terminal symbols. These requirements are met by the theory outlined above in the following way. derived phrase-markers and T-markers fill this function on the phrase-structure and transformational levels. must contain an information relevant to determination of the form of the signal corresponding to (18) (i. Levels are organized in a hierarchy. 'The Logical Structure of Linguistic Theory'. 80th phrasemarkers and transformation-markers can be represented in this way. In Chomsky.86 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR I emphasize once again that only after all the transformations have been completed do we have an actual 'sentence'-----that is. theT-marker (19)"is to contain all information relevant to the semantic interpretation of (18). On the level of transformations. and the surface structure is what we designated as K 6• The phrase-marker K 6. which is associated directly with an actual signal. induding. 'The Logical Structure of Linguistic Theory'. The general requirement on a syntactic theory is that it define the notions 'deep structure' and 'surface structure'. the primes are base phrase-markers and transformations. We take aT-marker to be the deep structure. we take the derived phrase-marker that is the final output of the operations represented in the T-marker to be the surface structure. in particular. We limit the discussion here to the levels of phrase structure and transformational structure. Katz and Postal. In particular. respectively. Perhaps this example suffices to convey the conte nt of the notion 'T-marker' (for further elaboration. see Chomsky. then. it is to be mapped into a phonetic representation of (18) by rules ofthe phonologie al component). how any transformational derivation can be presented as aT-marker which gives the fuH 'transformational history' of the derived sentence. a general theory of linguistic levels is developed in an abstract and uniform way.

and other related publications. the first fairly explicit discussion of grammatical relations ofthe deep strueture that are not represented in the aetual physieal form and organization of the sentence. Halle and LukotT. 'Phonology in a Generative Grammar'. Several important steps have been taken towards the study of rules that assign semantic interpretations to deep structures. Furthermore. for the moment. has had important consequenees for the theory of syntax as weil. Objeet.e. Jakobson. as noted above. G. as is characteristic of expository papers in transformational grammar. eds. is in the Grammaire generale et raisonnee of Port-Royal (1660). 'On Accent and Juncture in English'.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 87 To complete the theory. it is not the 'grammatieal subject' of the passive but rather its 'logical subject' that is the subjeet in the sense relevant to semantic interpretation). a system of rules for assigning such an object to a deep structure. These examples were chosen primarily to ilIustrate this fact. although recent work of Katz. First of all. pp. To my knowledge. it has been evident sinee the beginnings of recent work on transformational grammar that it is the grammatical relations and grammatical functions represented in the base phrase-markers underlying a sentence that are critical for its semantic interpretation (for example. however. that is. 'Explanatory Models in Linguistics'. Halle. respectively. in other words. as we shall note directly. Chomsky. This is evident from consideration of the examples diseussed throughout this paper. Chomsky. Subjeet. A theory of semantic interpretation based on the syntactic model outlined above would have to provide. first. 1952). 'On the Bas~s of Phonology' (Structure of Language. along lines suggested by R. a characterization of the notion 'semantie interpretation of a sentence'. Fant and M.) that these elements fulfil provide information that is fundamental for semantie interpretation. I wiil discuss the phonological component briefty in the fourth section. and the first general discussion of the importanee of these for sem antie interpretation. it is evident that the grammatical relations among the elements of the string representing a sentenee and the grammatieal functions (i. and for the development of the theory of transformational grammar. Analogously a theory of phonetic interpretation must specify the notion 'phonetic interpretation of a sentence'--ü must. we must add a description of the phonological and semantic components that interpret surface and deep structures. Preliminaries to Speech Analysis (Cambridge. The Sound Pattern of Russian. and second. Mass. The theory of semantic interpretation is in a much less developed state. ete.. Fodor and Postal has been quite encouraging and. The notion 'semantic interpretation of a sentence' remains in a rather primitive state. Fodor and Katz. that is. . it is examples of grammatical relations and functions that are obscured in the surface representation (the IC analysis) that provide the primary motivation for the rejection of all versions oftaxonomie syntax. aT-marker. 324-33). Halle. specify a universal phonetic alphabet--and must provide a system ofrules for assigning such an object to a surfaee strueture. As emphasized above. the final derived phrase-marker of a sentence.

[NP. VPl. Cartesian Linguistics. Ward. Sl. Object-of. For concreteness.. John is the Subject-of the sentence. In modern linguistics. saw Bill is the Predicate-ofthe sentence. we can say. 28. etc. and in an subsequent work on transformational grammar. We can go on to define grammatical relations (Subject-Verb.. § 1. a very natural device for assigning a system of grammatical relations and functions to a generated string.~ NP~ VP V NP John. 'The Logical Structure of Linguistic Theory'. Al. . Predicate-of. in fact. Thus associated with the rules of (21) we have the grammatical functions [NP. Various ways of defining these notions are discussed in Chomsky ('The Logical Structure of Linguistic Theory'. Sl.88 TOPICS IN THE THEORY OF GENERATIVE GRAMM AR For some brief references. Using the obvious definitions of these notions.. pp. Syntactic Structures. see Chomsky. Chomsky... Lg. as has been frequently pointed out. it is necessary to define grammatical relations and grammatical functions. saw is the MainVerb-of the Verb Phrase. A phrase-structure gramm ar is. A spects of the Theory of Syntax) and Postal (Constituent Structure). 18-23 (1952). pp. Main-Verb-oj.) in terms of these and other notions "e. 283-340 (1957). 'Discourse Analysis'. Current Issues in Linguistic Theory.. 10..g. and to show how the relations and functions of the base phrase-markers playa role in determining the semantic interpretation ofthe sentence that they underlie. 146-62 (1954). Lg. Current Issues in Linguistic Theory. respectively. [VP. 'Distributional Structure'. S--VP I John V / saw \ NP Bill I I To the grammatical rule A ~ XBY. that with respect to the phrase-marker (22). 33. pp.. [V. VP]... the same insight was expressed by Harris. then. These notions are represented directly in the phrase-marker assigned to astring generated by such rules. We may give these the conventional names Subject-oj.. 'Co·occurrence and Transformation in Linguistic Structure'. 1 1 and the point is also emphasized in Chomsky. and Bill is the Object-ofthe Verb Phrase. Bill V~saw This grammar generates the string 'lohn saw Bill' with the phrase-marker (22): (22) NP . in somewhat different terms.... in his early work on transformations. To go beyond this observation. Harris. we can associate the grammaticalfunction [B. consider a highly oversimplified phrase-structure grammar with the rules (21): (21) S~NP VP.. and for some further discussion.

The first attempt to develop a theory of semantie interpretation as an integral part of an explieit (i. in some sense. but in terms of the phrase-marker itself. the basic content elements that determine its semantie interpretation.e. for some diseussion of the role of 'surfaee funetions'. Turning now to part four of the outline. having now sketched a eertain theory of generative grammar that in part overeomes the fundamental inability of taxonomie syntax to provide an adequate notion of deep strueture. it is only the funetions in the base phrase-markers that are significant for semantie interpretation (but see Chomsky. The projeetion mies of the seeond type utilize the readings assigned in this way to base phrase-markers. we see at onee they are alreadyassigned. Not much is said about type two mies. The semantic eomponent should have two sorts of'projeetion rules'. 220. determine the semantie interpretation of the full sentenee. with no further elaboration of the rules. and. Notiee that we might define the grammatieal funetions not in terms of the generating rules. The important point is that a phrase-structure grammar need not be supplemented in any way for it to assign these properties to the strings it generates. as we shall see below. ultimately. we eonclude part three of the outline of the introductory seeHon. ('The Strueture of a Semantie Theory'). utilizing in the proeess three sorts of information: (i) intrinsie semantie features of lexieal items. This is the first study that goes beyond the assertion that the base phrase-markers underlying a sentenee are. in terms of the elements and configurations represented in the T-marker. 221. see the cited references). assigning a reading to the dominant eategory S. Ifwe do this. (iii) the strueture of the T-marker. in the appropriate way. sinee. Basing themselves on the aeeount of syntaetie strueture outlined above. and. I do not go into this here. beginning with the intrinsie readings of the lexieal items and using the grammatieal funetions defined by the eonfigurations of the base phrase-markers to determine how the higher level readings are assigned. this is not a serious gap in their theory. Once we reeognize the relational eharaeter of these notions. Katz and F odor argue that the semantie eomponent of a grammar should be a purely interpretive system of rules that maps a deep structure (a T-marker) into a semantie interpretation. . The first type assign semantie interpretations ('readings') to eategories ofthe base phrase-markers in terms of the readings previously assigned to the elements dominated by (belonging to) these eategories. in any event. Aspects 0/ the Theory 0/ Syntax. pp.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 89 and there are various ways in which one can attempt to formulate languageindependent definitions for the central concepts (for details. in an obvious way. generative) grammar is in Katz and Fador. (ii) the grammatieal funetions defined by the base rules. I would like to eonsider some of the defeets that have been exposed in the theory just sketehed as it has been applied to linguistic material. With this brief survey. we will have amore general notion of 'grammatical function' that will apply to derived phrase-markers as well as to the base phrase-markers. so defined).

it seems to me) that in each such case. furthermore. that the function of type 2 projection rules is much more restricted than Katz and Fodor were forced to assurne. . pp. See now also Klima. the singulary transformations in themselves need not be referred to by the rules of the semantic component since whatever contribution they appear to make to the meaning of the sentence can be regarded as an intrinsic property of the marker that determines their applicability. and inapplicable otherwise). an option al element introduced by rewriting rules of the base. since they need not take into account the presence of singulary transformations in aT-marker. In Katz and Postal. and in several other places. here. so far as semantic interpretation is concemed.e. Generalizing upon these various observations. Katz and Postal carry out a detailed analysis of many examples described in earlier studies that seem to demonstrate a contribution of generalized transformations to the semantic interpretation of the generated sentence in some way that goes beyond mere 'amalgamation'. Further arguments in support of this view. 246-323. In the light of this and other observations. It folIows. eds. then. An Integrated Theory 0/ Linguistic Description. Syntactic Structures. 'A Transformational Approach to Syntax'. there are syntactic grounds for regarding the description as in error. to insert the reading for the embedded phrase-marker in the position 12 Publication delays account for the discrepancy in dates. There are syntactic arguments in favour of assuming an abstract 'question marker' as an element introduced by base rules.90 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR In Lees. Turning then to generalized transformations. rather. At about the same time. in other words. E. the question transformations then being conditional on the presence of this marker (i. They 'argue (quite convincingly. and can therefore be handled in base structures by type 1 projection rules. by the presence of a dummy symbol). the transformation serving simply to place it in the correct position in the senten ce. 'A Transformational Approach to Syntax')'2 is incorrectly formulated. 'So Klima pointed out that the same is true of the question transformations of Chomsky. are presented in Katz and Postal. Katz and Post al then conclude that all singulary transformations which affect meaning are conditional upon the presence of markers of this sort. that in each such case the only function of the generalized transformation is to embed a sentence transform in a position that is already in the underlying structure (let us say. He shows that there are syntactic arguments in favour of an alternative formulation in which the negation e1em(:nt is not introduced by a transformation but is. 'Negation in English'. Fodor and Katz. it is shown that the negation transformation of Chomsky (Syntactic Structures. Structure 0/ Language: Readings in the Philosophy 0/ Language. it is further observed that the same is true of the imperative transformation of earlier work. in other words. is to interrelate the semantic interpretations of the phrase-markers on which they operate. obligatory when it appears in astring. they conclude that the only function of generalized transformations. The Grammar 0/ English Nominalizations. and further elaboration of it.

208-31 (1963)). which is far from satisfactory. the justification for the principle is not that it simplifies semantic theory. 221. it is reasonable to formulate the principle tentatively as a general property of grammar. Beyond this. the matrix sentence is the one into which a constituent sentence is inserted by a generalized transformation. it seems that all singulary transformations other than those that are 'purely stylistic' (cf. This is the way in wh ich these operations are interpreted. Furthermore. It is therefore important to observe that there is no question-begging in the Katz-Postal argument. 'The Structure ~f a Semantic Theory'. it seems that there are good reasons for regarding even the passive transformation as conditional upon the presence of an abstract marker in the underIying string (see Chomsky. J. be preferable terms. Actually 'matrix proposition' and 'constituent proposition' would. whether or not these transformations effect semantic interpretation. The Grammar 01 English Nominalizations. This principle obviously simplifies very considerably the theory of the semantic component as this was presented in Katz and Fodor. as assumed in earlier work. 19. there are no singulary transformations that must apply to a matrix sentence before a constituent sentence is embedded in it by a generalized embedding transformation. for some discussion of this distinction-discussion. Word. transformations appear to play no role in semantic interpretation.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 91 already marked (by a dummy element) in the phrase-marker in which it is inserted. correctly. In the light of this observation. embedding 13 The terms 'matrix sentence' and 'constituent sentence' are due to Lees. Fillmore pointed out that there are many restrictions on the organization of T-markers beyond those that were assumed in earlier attempts to formulate a theory of transformational grammar (Fillmore. Thus the function of type 2 rules is still further restricted. rather than as optional. Aspects ofthe Theory 0/ Syntax. in any event.13 although there are many instances of singulary transformations that must apply to a matrix sentence after embedding of a constituent structure within it and to a constituent sentence before it is embedded. although singulary transformations are ordered (apparently linearly). Independently of these developments. Aspects of the Theory of Syntax. Thus the only aspect of the T-marker that need be considered in semantic interpretation is the interrelation specified by the nodes where generalized transformations appear in the representation. but rather that in each case in which it was apparently violated. where the terms 'proposition essentielle' and 'proposition incidente' are used for 'matrix sentence' and 'constituent sentence'. Consequently. syntactic grounds. for a survey of syntactic arguments in support of this). pp. although it seems to me that areal and important distinction is involved) are conditional upon markers in base strings. pp. incidentally. 'The Position of Embedding Transformations in a Grammar'. C. What his observations come to is essentially this: there is no ordering among generalized transformations. 223. Chomsky. since what is involved here is not an operation on sentences but rat her on the abstract structures that underlie them and determine their semantic interpretation. respectively. That is. syntactic arguments can be produced to show that the analysis was in error on internal. . The same not ion appears in the analysis of transformational processes in the Grammaire generale et raisonnee. in the Grammaire generale et raisonnee.

obligatorily. Continue in this way. to the generalized phrasemarker as a whole.92 TOPICS IN THE THEOR Y OF GENERATIVE GRAMMAR should be regarded as substitution of a sentence transform for a 'dummy symbol' rather than as insertion of this transform in a categorially unspecified position. we construct the generalized phrase-marker (23) (ornitting all but the central configurations. returning to the beginning of the sequence each time we come upon a new occurrence of S introduced by a rewriting rule. established the particular formal properties of the T-marker (19) as general properties of any transformational derivation. In short. but there is another way of saying exactly the same thing that makes it seem entirely natural. and if. as noted above. Having completed the application of the rules to each such structure. It is therefore quite natural to generalize from these empirical observations. Notice that if no singulary transformations apply to a matrix phrase-marker before embedding. By the application of the rewriting rules of the base. dispense with generalized transformations entirely. we can permit the rewriting rules of the base to introduce the initial category symbol S. in effect. we construct what we can call a generalized phrase-marker. We now apply the linear sequence of singulary transformations in the following manner. Notice that with this formulation. Let us now return to the example (18)-(20) in the light of these suggested revisions of the theory of transformational grammar. But the earlier theory of T-markers left open the possibility for ordering of a much more complex sort. until. That is. Returning now to the T-marker (19) used as an example above. As just formulated. furthermore. then we can. we can allow it to head a new base derivation. An Integrated Theory of Linguistic Description. Wherever such a symbol is introduced. First.e. this principle appears to be quite ad hoc. The last observation is further elaborated by Katz and PostaI. Proceeding in this way. i. i. and many details).. we have. singulary transformations are applied to a matrix sentence only after embedding and the only ordering is among singularies. all embedding involves the insertion of a constituent phrasemarker in a position marked by a dummy element in the matrix structure. reapply the sequence to the 'next-higher' structure dorninated by S in the generalized phrase-marker. we can permit rewriting rules of the form A ---> ••• S . in fact.. apply the sequence to the most deeply embedded structure'dominated by S in the generalized phrase-marker. We then turn to the . we observe that it has just the properties that Fillmore outlines. finally. in the following order. First. Instead of introducing constituent phrase-markers by embedding transformations.e. we can apply the linearly ordered system of base rewriting rules in a cyc1ic fashion. The transformations indicated in (19) now apply. T Rel applies to the most deeply embedded structure. and to propose as a general condition on T-markers that they must always meet Fillmore's conditions and have the form illustrated in (19). the sequence of transformations is applied to the structure dominated by the occurrence of S which initiated the first application of base rules.

giving (18). The transformations indicated in the T-marker (19) are now obligatory and the structure of the T-marker (19) is fully determined by (23) itself. (standing for an unspecified category) in (23).----.YP ~ Y I ------.TOPICS IN THE THEOR Y OF GENERATIVE GRAMMAR 93 next higher structure. The advantages of this modification are obvious. the singulary transformations will not make an intrinsic contribution to meaning. To this we apply TIO' as before. an inversion rule (not indicated in (19). Furthermore.J. Since no further transformational rules apply at this point. It provides a more highly . But we have now eliminated generalized transformations in favour of a recursive operation in the base. given the general convention for cyclic application of transformations. we turn to the next high er structure dominated by S--in this case. the full generalized phrase-marker. At this point. the same will be true in all other cases. the one dominated by the occurrence of S in the fourth lineof (23). i. fiired Det /~ the S NP ~ '* * man VP I by passive thlmbn V~NP qua I. Next we apply the passive transformation and the subsequent deletion of the unspecified subject.e. Consequently all information relevant to the operation of the interpretive semantic component should be contained in the generalized phrase-marker generated by base rules. these operations now being obligatorily marked by the dummy elements passive and 6. if the modifications suggested above are correct. though in fact also needed in the earlier formulation) inverts the relative clause and the following N. work I Notice now that all of the information to the semantic interpretation of (18) is contained in the generalized phrase-marker (23) that underlies (18). By the pl'inciple suggested by Katz and Postal. and the generalized transformations will do so only in so far as they interrelate base phrase-markers.D ____ N . *-S-* NP-----VP I 1 V ~~ expected I NP NP 1------'* /1" S '* .

94

TOPICS IN THE THEORY OF GENERATIVE GRAMMAR

structured theory wh ich is weaker in expressive power; in other words, it excludes in principle certain kinds of derivational pattern that were permitted by the earlier version of transformationa1 theory, but never actually found. Since the primary goa1 of linguistic theory is to account for specific properties of particular languages in terms of hypotheses about language structure in genera1, any such strengthening of the constraints on a general theory is an important advance. Furthermore, there is good interna1 motivation for enriching the structure (and hence decreasing the expressive power) of transformational theory in this way, namely, in that this modification permits us to eliminate the notion of 'genera1ized transformation' (and with it, the notion 'T-marker') from the theory of syntax. Hence the theory is conceptua1ly simpler. Finally, the theory of the semantic component can be simplified in that type 2 projection rules are no longer necessary at a11. Recapitulating, we are proposing that the syntactic component of a grammar consists of rewriting rules and transformationa1 rules. The rewriting rules are permitted to introduce the initial symbol S. These rules apply in a linear sequence; if the initial symbol S appears in a derivation, then the sequence ofrules reapplies to form a subderivation dominated by this symbol, in the usual manner. The recursive property of the grammar (its 'creative aspect', to return to terminology used above) is restricted to the base component. In fact, the restriction may be still heavier than this, since recursion may be limited to introduction of the symbol S, that is, to introduction of 'propositiona1 content'. This is not a necessary property of a phrasestructure grammar. The base rules, applying in the way just outlined, form genera1ized phrasemarkers. The function of the transformationa1 rules is to map generalized phrasemarkers into derived phrase-markers. If the transformational rules map the genera1ized phrase-marker MD into the fina1 derived phrase-marker M s of the sentence X, then MD is the deep structure of X and M s is its surface structure. This approach to syntax forma1izes, in one specific way, the view that the phonetic form of a sentence is determined by its actua1 labelIed bracketing, whereas its semantic interpretation is determined by the intrinsic semantic properties of its lexica1 items and by a network of grammatica1 relations, not necessarily represented in the surface structure, that interconnect these items (cf. (13». The underlying grammatica1 relations are determined by the base rules. This abstract system of categories and relations is related to a labelIed bracketing of the actual sentence by transformationa1 rules and the interpretive rules of the phonological component. There is fairly good reason to suppose that the base rules are rather narrowly constrained both in terms of the symbols that may appear in them and in terms of the configurations of these symbols, but I will not go into this further question here (see Chomsky, Aspects 01 the Theory 01 Syntax, for some discussion). In so far as information is presently available about syntactic structure, and about the relation of signa1s to semantic interpretations of these signa1s, this view seems compatible with it. It is worth

TOPICS IN THE THEORY OF GENERATIVE GRAMMAR

95

mentioning that a view very much like this is expressed in the Grammaire generale et raisonnee, to which we have now had occasion to refer several times. We might ask why a natural language should be constructed in this way; why, in particular, should it not identify deep and surface structures and thus dispense with the transformations that interrelate them. One would naturally attempt to find an answer to this question on perceptual grounds. For some speculations that seem to me worth pursuing further, see Mi1ler and Chomsky, 'Finitary Models of Language Users', part II. Observe that the base rules may form generalized phrase-markers that cannot be mapped by any sequence of transformations into a surface structure. Far example, suppose that we had chosen the phrase 'the boy' instead of 'the man' in the most deeply embedded structure of (23). In this case, the generalized phrase-marker would evidently constitute the deep structure of no sentence; there is no sentence for which this structure provides the semantic interpretation. And in fact, the relative transformation would block when applying to this structure, because of the lack of identity between the Noun Phrases of the matrix and constituent sentences. 14 Hence not all generalized phrase-markers underlie sentences and thus count as deep structures. The deep structures are the generalized phrase-markers that are mapped into wellformed surface structures by transformational rules. Thus the transformations serve a 'filtering' function; in effect, they supply certain global constraints that a deep structure must meet, constraints that are, in fact, entirely unstatable within the framework of elementary rewriting rules that seem perfectly adequate for the generation of base structures with the grammatical functions and relations that they express. For further discussion of this property of transformations, see Chomsky, Aspects 0/ the Theory 0/ Syntax, eh. 3. In this way, we can construct a theory of grammatical transformations that is conceptually simpler than the earlier version, described above, but still apparently empirically adequate. In this modified formulation, the functions of the base rules and the transformational rules are more clearly expressed, as are also the notions of deep and surface structure. We have, correspondingly, a simplification of semantic theory.15 I began this section by presenting a certain theory of grammar in outline. I have now tried to show that this theory was too broad and rich in expressive power, and that a much more restricted version of it (which is, furthermore, conceptually weIl motivated) will suffice to account for what empirical data are now avaiIable. I would
14 What is involved here is a set of very general conventions on recoverability of deletion, in the transformational component of a grammar. For discussion, see Chomsky, Current lssues in Linguistic Theory; Aspects 0/ the Theory 0/ Syntax; Katz and Postal, An lntegraled Theory 0/ Linguislic Descriplion. 15 Incidentally, only embedding transformations werc considered here. It is also necessary to show how various transformations that introduce coordinate structures (e.g. conjunction) can be developed within this framework. For some remarks on this question, see Chomsky, Aspecls 0/ the Theory 0/ Syntax, and the references cited there.

96

TOPICS IN THE THEORY OF GENERATIVE GRAMMAR

now like to turn to an inadequacy in earlier theory of the opposite sort, that is, to a class of problems that show this theory to be too poor in expressive power, in a certain way. Let us limit our attention now to the base component of the syntax. The theory outlined followed structuralist assumptions in supposing that the relation of lexical items to the categories to which they belong is fundamentally the same as the relation of phrases to categories of which they are members. Formally speaking, it was assumed that lexical item X is introduced by rewriting rules of the form A --> X, where A is a lexical category, exactly in the way that phrases are introduced. 16 However, difficulties in this view quickly emerged. Shortly after the publication of the earliest work in transformational generative grammar, it was pointed out by G. H. Matthews that whereas the categorization of phrases is typically hierarchie, and therefore within the bounds of phrase-structure grammar, lexical categorization typically involves cross-classification, and therefore goes beyond these bounds. For example, a Noun may be either Proper or Common, and, independently of this, may be either Animate or Inanimate; a Verb may be Transitive or non-Transitive, and independently ofthis, may or may not take non-Animate Subjects; etc. This fact is unstatable within the framework of phrase structure grammar. Consequently, the theory of the base must be extended in some way so as to provide an analysis of lexical categorization that is different in fundamental respects from the analysis in terms of rewriting rules that seem quite adequate above the level of lexical category. Similar observations were made independently by StockweIl, Anderson, Schachter and Bach, and various proposals have been made as to how to remedy this defect of the base component. The general problem is studied in some detail in Chomsky, Aspects ofthe Theory of Syntax, Ch. 2, where reference is also made to the earlier work just noted. I will sketch briefly the proposals offered there for modification of the theory of the base component. Notice that the problem of lexical cross-classification is formally analogous to the problem of phonological classification. Thus phonologie al elements are also typically cross-classified with respect to the operation of various phonological rules. Certain rules apply to the category of Voiced segments; others to the category ofContinuents; membership of a segment in one of these categories is independent of its membership in the other. This is, furthermore, the typical situation. This, in fact, is one major reason for the view that segments (e.g. phonemes or morphophonemes) have no independent linguistic status and are simply to be regarded as sets of features. More generally, a lexical item can be represented phonologically as a certain set of features, indexed as to position. Thus the lexical item bee can be represented by the
16 Notice that a1though this has been the view of all work in modern syntactic theory that has gone beyond mere elaboration of terminology, the incorrectness of this view became obvious only when it was formalized within the framework of an explicit theory of grammar. An essential reason for formalization and explicitness is, of course, that it immediately exposes inadequacies that may otherwise be far from evident.

basing this assignment on the readings assigned to aIready interpreted nodes and the grammaticaI relations represented by the configuration in question. An Integrated Theory 0/ Linguistic Description.. lakobson's theory of distinctive features--I will presuppose acquaintance with this.g. let us revise the theory of the base in the following way.g. The lexicon is an unordered set of lexieal entries. ••• . . . Beyond this.. Thus the final strings generated by the categorial component (let us call these pre-terminal strings) are bascd on a 'vocabulary' (i. non-Gravez. since it apparently involves only lexical items and not phrase types. .g. The categoriaI component of the base generates no lexical items in strings (though it may introduce grammatical morphemes). Verb. in a manner which we will describe directly. [± Artifact]). Adopting this rather natural proposal. [± Voiced"l. The features constituting the lexical entry may be phonological (e. where n is an integer indicating position). Apre-terminal string is converted to a terminal string by insertion of an appropriate lexical item in each position marked by 6. The latter will occupy the position in which items from the lexicon will be inserted. 'The Structure of a Semantic Theory'.. The base consists of a system (presumably.. Thus Katz and Fodor. 86) consisting of grammatical morphemes and the symbol6. We limit OUT attention here to the syntactic features. [± Proper]). a non-continuant. As a first approximation. and that its second 'segment' is vocalic. it contains alexieon. essentially. particularly.. non-grave. . and Katz and Postal. The grammaticaI relations and the order of application of the interpretive projection rules are determined completely by the .TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 97 feature set [Consonantal l . are led to the conc1usion. a linear sequence) of rewriting rules which we may call its categorial component. etc. non-Continuant l . l indicating that its first 'segment' is consonantal. It provides a perfectlysatisfactory solution to the cross-c1assification problem on the phonological level (and furthermore relates very nicely to what seems to me to be for the present the most satisfactory theory of universal phonetics namely. Observe also that the semantic analysis of lexical items also apparently requires a kind of feature theory. Such a representation can be given in matrix form in an obvious and familiar way. Voiced l . p. and that these features typically cross-c1assify lexical entries.g. semantic (e. Noun. Each lexical entry is simply a set of specified features. it seems plausible to develop semantic theory in terms of projection rules that assign readings to successively higher nodes of the deep structure. or syntactic (e.. that a lexical entry in its semantic aspect should consist of a set of semantic features. a set of primes-see above. These observations suggest that the problem of syntactic cross-c1assification be dealt with in the same way. we may think of each lexicaI category A (e. voiced. in the form recently given it by Halle. As we noted above. RecaIl that the deep structures that determine semantic interpretation are generaIized phrase-markers generated by the base component. where 6 is a fixed dummy symbol.e. for the remainder of this paper)..) as being involved only in rewriting rules ofthe form A~ 6. Vocalic 2.

where A is a lexical category. see Chomsky. For the category features. transitivity. that the intitial non-Continuant is non-Aspirated) can be predicted by rule (in this case. Thus the two separate aspects of the semantic theory are mirrored in the subdivision of the base into a categorial and a lexical component. etc. a lexical item will be idiosyncratic in many respects. we are asserting that it can appear in the position-NP in a Verb Phrase. These degenerate contextual features. domina ted by the category symbol A. the obvious notation is [± Al. Thus the fact that 'buy' begins with a Voiced nonContinuant. as will be evident to anyone who has ever attempted to construct a detailed grammatical descrii'tion.1).NP]. Notice that this rule must take account of the structure of the phrase-marker in which the item is being inserted.e. In so far as regularities concerning feature composition can be expressed by rule. The functioning of the categorial component is c1ear. Aspects of the Theory of Syntax. the features in question can be extracted from the lexical entry (for discussion of redundancy rules. let us. Consider now the problem of a proper notation for the other contextual features. For example. they need no longer be represented in the rewriting rules. Similarly. and syntactic features. we may call category features. Normally. namely. Since these can now be specified in the lexical entry. But there may be redundancy ru/es of various kinds that operate on phonological. Therefore the syntactic feature [+ Transitive] must specify some aspect of the phrase-marker in which the item can be inserted. In contrast. The intrinsic semantic properties that provide the initial readings for this process of semantic interpretation (i. 4. for which the item in question may be substituted. must all be represented by features of the lexical entry. that it is a transitive Verb. consider the lexicon in some further detail.98 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR categorial component of the base. the fact that 'persuade' can be followed by a Noun Phrase . when we say that a Verb is Transitive. which indicates a minimal aspect of the phrase-marker. semantic.. then. particularly Ch. By convention.g. e. § 2. Let us now consider the rule that inserts lexical items in pre-terminal strings. Let us call a feature ofthis sort a contextual Feature. that it has irregular inflexions. the category dominating the occurrence of 6. This leads to an enormous simplification of the base component. Clearly the best notation is simply an indication of the context in which the item can occur. The degenerate case of a contextual feature is the feature [ ± Noun] itself. that it involves transfer of ownership. Thus the feature [+ Transitive] can be represented simply [ + . features that cannot be predicted by general rule. The lexical entry for a certain item should contain all information about idiosyncratic features of this lexical item. a phonological rule). therefore. we will call such features of Nouns as [± Human] non-contextual. we assert that an item with the category feature (+ A] can only replace an occurrence of 6. and that specify interrelations among features of the various types. Other properties (for example. the readings of the lexical items that are the terminal elements of the generalized phrase-marker) are provided completely by the lexicon.

g. Contextual features of this sort.NP PPho the lexical entry for 'persuade' (in fact. Whereas the strict subcategorization features specify categorial frames in which an item may appear. the selectional features for 'abundant' must indicate that it can be predicated of 'harvest' but not 'boy'. contexts dorninated by the phrase category that immediately dominates the lexical category for which the lexical item is substituted. see Chomsky. Thus the selectional features for 'frighten' will indicate that its Object must be specified as [+ Animate]. there are contextual features of a radically different sort that we will call selectional features. whereas the selectional features for 'clever' must contain the opposite specification. Contextual features can be regarded as specifying certain substitution transformations. which specify the frame in which an item can be substituted. we will call strict subcategorization features. These restrictions can be made precise.e. 'I persuaded John of the pointlessness of his actions') can be indicated by assigning the contextual feature [+ . I have not discussed the problem of deviation from grammaticalness here. the selectional features of a lexical item X specify lexical features of the items with which X enters into grammatical relations. selectional features refer only to 'heads' of grammatical related constructions. etc. The elementary transformation that completes the definition of the transformation states that the lexical item in question (i. the selectional features for 'e1apse' will indicate that its Subject cannot be [ + Human] (and for true descriptive adequacy. However. this is apparently the only contextual feature needed to specify the frame in which 'persuade' can appear. the set of specified features that constitutes the lexical entry) substitutes for the occurrence of!:::" thai" appears in the position indicated in the structural analysis. The context stated in the contextual feature specifies the condition that must be met by the phrase-marker to which the transformation in question applies and the manner in which this phrase-marker must be analysed for the purposes of this transformation. For discussion. it is clear that whenever a grammatical rule exists. On the other hand. Alongside of strict subcategorization features. Aspects of the Theory of Syntax).e.TOPICS IN THE THEORY OF GENERATIVE GRAMMAR 99 and a following Prepositional Phrase (e. It seems that sentences deviating from selectional rules are interpreted quite differently from those deviating from strict . must obviously give a much narrower specification than this). we may ask how a sentence is interpreted that deviates from this rule. 84. Thus it defines the structural analysis of the transformation (see above. Thus the strict subcategorization features only involve 'Iocal contexts' ~. p. Aspects of the Theory of Syntax. and can be shown to lead to certain interesting consequences concerning the possible constraints that may appear in a grammar. It is c1ear from the examples that there are many restrietions on the form of the substitution transformations defined by contextual features. all other forms being derived by transformation-for discussion. see again Chomsky. We can represent selectional features by a notation very much like that suggested above for strict subcategorization features. Similarly.

must be looked at in an entirely different way. The first defect was one of excessive richness in expressive power. etc. an inability to express certain aspects of grammatical structure. I conclude part 5 ofthe introductory outline. For discussion of this question. Wehave now been discussing a defect of the opposite kind. The theory of transformational generative grammar that results from these modifications is conceptually quite simple. With this. for example. and know of no justification for postulating a more complex structure and organization of the theory of the syntactic component than what has been sketched in outline here. Sentences of the former type are often interpreted as somehow metaphorical. p. I have now briefly sketched two major respects in which the first modem attempts to formulate a theory of grammaticaI transformations were shown to be defective by later work. We assumed. see Chomsky. These features can participate in rules that are not at all violable in the way in which selectional rules may be (consider. 2. and is reasonably weil supported by what empirical evidence is presently available. 'lohn found sad'.. Notice incidentally that the ease with which sentences deviating from selectional rules can be interpreted is not simply a result of the fact that 'Iow-Ievel' syntactic features such as [± Human] or [takes Animate Object] are involv~d. in this brief account of syntactic features. deviation from strict subcategorization rules gives such examples as 'lohn persuaded to leave'. This was not an arbitrary decision. then. this theory seems to me to constitute the most satisfactory hypothesis as to the form of the syntactic component of a grammar. Aspects 0/ Theory 0/ Syntax. Thus the various types of contextual feature. Deviations from contextual rules involving category features (see above. Each component of the theory has a welldefined function. . if interpretable at all. sentences of the latter type. it can easily be justified on syntactic grounds.100 TOPICS IN THE THEORY OF GENERATIVE GRAMMAR subcategorization rules. Deviation from selectional rules gives such examples as 'colourless green ideas sleep furiously'. etc. and many of the other topics mentioned briefly here. it is elear. Ch. and have suggested a way to modify the theory so as to overcome this. For the present. such expressions as 'the table who I scratched with a knife'. I see no way in which any of the postulated mechanisms can be eliminated without sacrifice of descriptive adequacy. 'a very barking dog'. that a nontrivial study of it demands a rich and detailed understanding of the various types of grammatical process. 'who I saw was lohn'. however.are rather different in the conditions that they im pose on senten ce structures. There is much to say about this general problem. 98) are still different in interpretive potential.). namely. 'sincerity admires lohn'. etc. that the features of a Noun are inherent to it and that the features that selectionally relate Nouns to Verbs or Nouns to Adjectives appear as contextual (selectional) features of the Verbs and Adjectives.

to Harvard University. !Cl 1967 by the University of Chicago. Linguistic theory. INTRODUCT10N T HIS paper defends the relevance of linguistics to philosophy on the grounds that linguistic theory incorporates solutions to significant philosophical problems. Center for Cognitive Studies. 2 In the final section of this paper. linguists investigate what is common to all natural languages and formulate these more general facts about language in linguistic theory. Air Force (Contract AF 19 (628)-2487). therefore. This thesis does not concern either a philosopher's From The Linguisfic Turn. and the National Aeronautics and Space Administration (Grant Ns G-496). so represented. they can be solved on the basis of theoretical constructions that appear in linguistic theory. 1967). KATZ 1. pp. Richard Rorty (University of Chicago Press. Reprinted by permission ofthe author and the University of Chicago Press. and that. The particular thesis to be defended here is that certain philosophical problems can be represented correctly as questions about the nature of language. In the latter. The Journal 0/ Philosophy. It was also supported in part by the Joint Services Electronics Program (Contract DA 36-039-AMC-03200(E». All rights reserved.VI THE PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORYl JERROLD J. the thesis of this paper asserts that theoretical constructions initially devised by linguists to enable linguistic theory to systematically state uniforrnities across natural languages also fulfil the conditions on solutions to certain philosophical problems. The National Institutes of Health (Grant MH-04737-05).S. ed. 2 Synchronic linguistics involves two distinct but interrelated studies: a study of the diversity in forms of linguistic communication and a study of the limits of such diversity. 340-55. This work was supported by a grant from the National Institute of Health. This thesis should not be interpreted as asserting that the linguist's descriptions of natural languages reveal philosophical insights that somehow must escape philosophers looking at the same languages. Given this notion of linguistic theory. linguists investigate what is unique about individual natural languages and formulate such facts in what are called linguistic descriptions (or generative grammars). In the former. I will indicate another sort of relevance oflinguistic theory tophilosophy. and in part by the U. . MH-05120-04. I This paper is a revised and expanded version of my paper 'The Relevance of Linguistics to Philosophy'. National Science Foundation (Grant GP-2495). No. is a specification of the universals of language. owing to the nature of those problems.

is best regarded as a presentation of the thesis itself. A. 'What's Wrang with the Philosophy of Language?' lnquiry. is this. J. Of course. pp. 2. Logical empiricism confined its efforts to the construction of highly arbitrary and conceptually impoverished theories about a c1ass of artificial languages whose structure bears little similarity to that of natural languages. If the defence of this thesis is successful. both approaches prided themselves on their shortcomings. 3 I tried to show in some detail why the two major approaches in contemporary philosophy. The answer is simply that the approaches to the philosophical problems concerned that are available in' contemporary philosophy have not dealt successfully with these problems and. . Logical Empiricism (Logical Positivism) and Ordinary Language Philosophy.102 PHILOSOPHICAL RELEV ANCE OF LINGUISTIC THEORY or a linguist's account of the facts about a natural language. J. that these approaches contain inherent difficuities which make it quite unlikely that they can deal successfully with them. 197-237. The general character of the difficuities. turning their vices into alleged virtues. 1966). The present paper. then. New York. Thus. Fodor. in the way that philosophy of science bears upon the c1arification of methodology and theory construction in linguistics. it is appropriate to ask why an appeal to linguistics is necessary. however. particuiarly 'What's Wrong with the Philosophy of Language?' and The Philosophy 0/ Language. nor is it possible to formulate those that are presented in their full form. Katz and J. Within the confines of a paper of this sort it is. then linguistics is not incidentally pertinent to philosophy. THE RATIONALE FOR APPEALING TO L1NGUISTICS At the outset. Katz. It is neither necessary nor possible to repeat my criticisms here. Logical empiricism prided itself on its exJ J. the latter failed to give us any theory. V (1962). Ordinary language philosophy preoccupied itself with unearthing the most minute and detailed facts about the use of English locutions to the almost complete neglect of any concern with theory. Since I have no apriori notions about the apriori character of philosophical investigations. but is directiy relevant in the same way that philosophical theories themselves are. moreover. well-motivated solutions to the major philosophical problems they tackled. with a sketch of some of the arguments that can be given for it. that is. but rather concerns the more abstract matters that are dealt with in an account of the facts about language in general. are inherently incapable of providing adequate. while the former offered us philosophically irrelevant theories. In other publications. we find it necessary to go outside the boundaries of contemporary philosophy to search for the solutions to philosophical problems. of course. impossible to present all the arguments on behalf of this thesis. I am in no way disturbed by the fact that we might have to know quite a lot about extra-logical matters in order to solve certain philosophical problems. why. and J. The Phi/osophy 01 Language (Harper & Row.

the solutions that these approaches presented were based either on unmotivated. L1NGUISTIC THEORY I have claimed that the part of linguistics that is relevant to philosophy is linguistic theory. Accordingly. In this section. of course. and vague to provide a basis for the solution to philosophical problems. 3. Though. on the other hand. claiming that theories cause the very philosophical perplexities that philosophy seeks to resolve by examination of the use of particular linguistic constructions. the principles of organization and interpretation that are invariant from natural language to natural language. of which each empirically successful linguistic description must be an instance. Linguistic theory expresses such invariants in the form of a model of a linguistic description. claiming that natural languages are too irregular. Accordingly. This unsatisfactory situation led a number of philosophers---with Quine as the most notable example--to turn some of their attention to empirical linguistics. uses of language. and another is the nature and extent of the relevance of linguistics to philosophy. exemplifying every aspect of the model.PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 103 clusive concern with artificial languages. the construction of linguistic theory and linguistic descriptions are strongly interdependent. I differ with them on a number of fundamental points. and hence arbitrary. I endorse this concern. One is the question of what it is in linguistics for which philosophical relevance ought to be claimed. linguistic theory is a specification of the uni versals of language. principles or on particularistic analyses of locutions whose bearing on philosophical problems were neither established nor made fully clear. nor. Ordinary language philosophy prided itself on its avoidance of theory construction. particular linguistic descriptions account for the diverse ways in which different naturallanguages realize the abstract structural pattern displayed in the model. amorphous. theoretically UDsophisticated. Linguists can abstract out the common features from a set of linguistic descriptions and so generalize from them to hypotheses about linguistic . did logical empiricism provide an alternative standard for justifying a theory of artificial languages that might serve as a replacement for the standard of conformity to the facts of language that had been eliminated. and so never seriously asked what a good theory would be able to do toward supplying a conceptual systematization of the facts of language that might offer solutions for the philosophical problems that arise in the course of ordinary. while the model itself describes the form of a system of empirical generalizations capable of expressing and organizing the facts about a naturallanguage. I will try to explain what this part of linguistics iso As mentioned above. Ordinary language philosophers never established that the theories in which they located the source of certain philosophical problems were not just bad theories. But the claim of the logical empiricists was never submitted to empirical investigations.

and determine the relations between rules within each component. 'syntactic component'. and 'semantic component' refer to the corresponding parts of a linguistic description. continue to support these generalizations. The terms 'phonological theory'. their empirical adequacy is thus a matter of whether further facts. and 'semantic theory' refer to these sub-theories. Notice. the second states the rules determining how speech sounds with a fixed phonetic shape are organized into sentential structures. viz. Since the linguistic . The phonological. Linguistic theory also specifies the relations between these components that weid them into an integrated linguistic description. Since putative linguistic universals are inductively extrapolated generalizations. and semantic theory. the facts from natural languages on which linguistic descriptions depend for their empirical support. Alternatively. As a consequence. each corresponding to one of the three components of a linguistic description. upon which newly constructed linguistic descriptions will eventually be based. finally. Thus. linguists can facilitate their task of describing a language by using the model provided by linguistic theory as a pattern for their systematization of the facts they uncover in field work. that if the general form of a particular linguistic description can be deduced from linguistic theory. syntactic. 'syntactic theory'. The ability ofspeakers to transmit their thoughts and ideas to one another through the vehicle of articulated speech sounds presupposes that each speaker has mastered a common system of rules within which each well-formed utterance receives a fixed semantic interpretation. phonological.104 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY universals. the third states the rules determining how such sentential structures are interpreted as meaningful messages. since it will also be supported by a wealth of evidence from many natural languages via the connection through linguistic theory between their linguistic descriptions and this one. characterize the form of the rules in a linguistic description. and 'phonological component'. specify the theoretical constructs utilized in writing actual rules in appropriate forms. At the level of linguistic theory. syntactic. projected from known regularities cutting across the set of already constructed linguistic descriptions. The first states the rules determining the phonetic structure of speech sounds in a language. then that linguistic description will be far better conflrmed than were it to derive its support solely from the facts about the language it describes. and semantic components are rule-formed descriptions of knowledge that a speaker has acquired in attaining fluency. Linguistic theory consists of three sub-theories. The fundamental problem to which a linguistic description addresses itself is that of explicating the common system of rules on the basis of which different speakers of the same language can correlate the same speech signal with the same meaningful message. the same facts that confirm or disconfirm particular linguistic descriptions also confirm or disconfirm a linguistic theory. the justification of both linguistic theory and individual linguistic descriptions have a common basis. jointly. Linguistic communication takes place when the same associations between sounds and meaning are made by different speakers in verbal interaction.

Accordingly. the set of underlying phrase-markers in a given syntactic description is automatically connected with the superficial phrasemarker in that description because the latter is obtained from each of the former by a specifiable (but in each case different) sequence of transformations and this 4 Cf. Thus. The output of the syntactic component is a syntactic description of each sentence of the language which consists of a set of phrase-markers. The syntactic component is the generative source of the linguistic description. Postal. J. Cambridge. Tbe model of a linguistic description offered by linguistic theory must show that the various schemes for making such correlations that are found in different natural languages are instances of a general formula which is the same for all natural languages. 1964) and N. Tbe bracketing teIls us that the elements enclosed within a single bracket form a constituent and the labelling teIls us the syntactic category to which the constituent belongs. An underlying phrasemarker describes that aspect of the syntax of a sentence of which its meaning is a function. . Cambridge. Aspects o/the Theory 0/ Syntax (Massachusetts Institute of Technology Press. An Integrated Theory 0/ Linguistic Descriptions (Massachusetts Institute of Technology Press. and the rules of the phonological component operate on the superficial phrase-marker to provide its phonetic representation. purely interpretive systems. or clauses are constituents of the same type if and only ifthey receive the same label. phonetic representations and semantic interpretations. where a phrase-marker can be thought of as a labelIed bracketing of the constituents of a sentence. 1965). Now. its components must be related to one another in such a manner that the representations given by the phonological and syntactic components of the phonetic and syntactic character of a sentence are formally connected with the representation given by the semantic component of its meaning. the rules of the semantic component operate on the underlying phrase-markers of a sentence to provide its semantic interpretations. The underlying phrase-markers in a syntactic description are related to the superficial phrase-marker by virtue of the fact that this same superficial phrase-marker is transformationally derived from each of them. while a superficial phrase-marker describes that aspect which determines its phonetic shape. phrases. the set of phrase-markers which constitutes the syntactic description of a sentence consists of a sub set of underlying phrase-markers and a single superjicial phrase-marker. Thus. Katz and P. J. respectively. transformations are syntactic rules that generate superficial phrase-markers from underlying phrase-markers. It generates abstract formal objects which are the input to the phonological and semantic components. Chomsky. the number ofunderlying phrase-markers in the syntactic description of a sentence indicates its degree of syntactic ambiguity (so that each underlying phrase-marker represents a syntactical unique sentence). On the current model of a linguistic description. Both these components are. two words.PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 105 description must formally simulate the sound-meaning correlations made by speakers. Tbeir outputs are. That is. therefore.4 this formula is embodied in the following organizational pattern for linguistic descriptions.

Descartes in French. the linguistic description will correlate the phonetic representation of a sentence with its semantic interpretation as desired. the corre1ation being effected by the transformations in the syntactic component and the manner in which the phonological and semantic components are organized to operate. we can utilize the rich stock of concepts that express the common structure of naturallanguages.106 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY superficial phrase-marker is not transformationally obtainable from any underlying phrase-marker outside this set. Basing a solution to a philosophical problem on linguistically universal principles avoids relativizing that solution to one or another particular natural language. But the philosophical problems about which they wrote were language-independent questions about a common conceptual structure. ADVANTAGES OF LINGUISTIC THEORY AS A STARTING POINT The advantages of starting with Iinguistic theory as a basis for the treatment of philosophical problems are strictly complementary to the previously mentioned disadvantages of the logical empiricist and ordinary language philosophy approaches. if it is these problems. for which philosophers seek solutions. Thus. First. instead of having to content ourselves with philosophical solutions that rest on the arbitrary principles of some artificial language or on an assortment of comments on the use of certain words or expressions from some natural language. in order to obtain solutions to philosophical problems. Accordingly. It is just as absurd to say that the solution to a certain philosophical problem is such-and-such in English as it is to say that a broken back is such-and-such a condition among Chinese. explicitly defined in a formalized theory of linguistic universals. Plato and Aristotle wrote in Greek. we avoid the difficulties that stern from the absence of any empirical controls on a solution to a philosophical problem and from the failure to clarify the relation between facts about the uses of words and expressions and the solutions on behalf of which they are adduced. we can argue from the concepts that underlie language in general to the solution of a philosophical problem. respectively. rather than some specialized ones having to do exclusively with one language. instead of having to resort to oversimplified and largely unexplained concepts or having no theory at all to appeal to. if we base our solutions to philosophical problems on Iinguistic theory we have a straightforward empirical basis on which to justify such solutions in terms of the empirical evidence that provides the support for successful linguistic descriptions. on a superficial phrase-marker and the underlying phrasemarker from which it was transformationally derived. Second. Kant in German. Hence. . and Hume in English. There is a further advantage of this starting point. For the justification of the theoretical constructs used in the solutions is provided by the very same evidence that empirically warrants their introduction into linguistic theory. then we cannot narrow the scope of solutions to philosophical problems. 4. By starting with linguistic theory.

what he tacitly knows about the structure of his language. Furthermore. The linguist. How. § 3. On the other hand. a performance. idiosyncratic and random errors. might underlie only one or another particular language. concerns the structure of concepts and the grounds for the validity of cognitive and evaluative principles. on the other hand. A theory in linguistics explicates linguistic competence. 5. People can be quite consistent in drawing nondemonstrative inferences according to invalid principles. Thus. and be inconsistent in their practice of using valid ones. lapses of attention.PHILOSOPHICAL RELEVANCE OF LlNGUISTIC THEORY 107 instead of having to argue from concepts that. level of motivation. THE EPISTEMOLOGICAL AND THE PSYCHOLOGICAL Given that linguistic theory is a formal reconstruction of the universal principles by which speakers relate speech signals and meaningful messages.n inductive logic that is a valid codification of the principles of nondemonstrative inference in science and daily life. not psychological. can be compared to the logician. But it does not apply to one extracted from a theory in linguistics that is psychological in the former. interest. can linguistic theory ofTer solutions to philosophical problems when the 'solutions' are apparently not even addressed to the right type of problem? This criticism rests on a failure to distinguish two senses of the term 'psychological'. Hence. Because a theory of performance must accept such . It seeks to reconstruct the logical structure ofthe principles that speakers have mastered in attaining tluency. not linguistic performance. there are two senses of 'psychological': on one. perceptual and motor limitations. etc. This makes it in some sense a psychological theory. The criticism cited above applies to a proposed solution for a philosophical problem extracted from a theory that is psychological in the latter sense. for all we can say. and his linguistic perfonnance. then. what he does with this knowledge. the subject of a psychological theory is a competence. we can better justify the solution to a philosophical problem because we can adduce far stronger evidence than ifwe were restricted to the data from one or another particular language (cf. A theory of performance cannot solve a philosophical problem such as that of formulating a. the concluding remarks of paragraph 3). it must consider such linguistically extraneous factors as memory span. which makes it epistemological in the broad sense. pauses. it is clearly an explication of a facet of a human ability. The distinction depends on the difference between a speaker's linguistic competence. whose aim is to provide a statement of ideal linguistic form unadulterated by the influence of such extraneous factors. and in the other. A philosophical problem. whose aim is to provide a statement ofideal implicational form unaldulterated by extraneous factors that influence actual inferences men draw. a theory of performance seeks to discover the contribution of each of the factors that interplay to produce natural speech· with its various and sundry deviations from ideal linguistic forms.

too. They assumed that gramm ar had done what it could but that its best is not good enough. It has long been recognized that the phonetic or orthographic realization of many sentences is such that no analysis of them in terms of traditional taxonomie grammar can reveal the true conceptual structure of the proposition(s) that they express. it sifts the facts about behaviour. Support for this alternative has come recently from Chomsky's work on syntactic theory which shows that traditional taxonomic grammar is too Iimited and revises it accordingly. This assumption is open to a serious challenge. Therefore. is too Iimited to reveal the underlying conceptual structure of a sentence. Philosophers who accepted this assumption simply overlooked the possibility that traditional taxonomic grammar might not be the last word on grammar. Con· stituent Structure. Folklore. factoring out the distorting influences of variables that are extraneous to the logical structure of the competence.108 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY behaviour at face value. GRAMMATICAL FORM AND LOGICAL FORM But to establish the re1evance of Iinguistic theory. Publication Thirty of the Indiana University Research Centre in Anthropology. and Linguistics (Bloomington. 5 The feasibility of this alternative rests on whether Chomsky's criticism is directed at just the features of traditional taxonomic gramm ar that make it incapable of handling logical form and on whether the revision provides the theoretical machinery to handle it. and Ryle. to mention some notable examples--to seek a philosophical theory about the logical form of propositions. 1964). Suitably extended. however. it must be shown to offer solutions to significant philosophical problems. Carnap. a theory of competence does. . Almost invariably. Such a theory has built in a means for correcting itself in cases where invalid principles were accepted or valid ones rejected. Chomsky. One of the pervasive problems of modern philosophy is that of distinguishing between the grammatical and logieal forms of sentences. however. Posta!. the early Wittgenstein. linguistic theory cannot be criticized as irrelevant to the solution of philosophical problems. The alternative to a philosophical theory about logical form is thus a linguistic theory about logical form. it has no means of correcting for the acceptance of invalid principles and the rejection of valid ones. From the same cases where grammatical form and logical form do not coincide. so that a philosophical theory of one sort or another is needed to exhibit the conceptual relations unmarked in grammatical analysis. 'Cf. Since it regards performance only as evidence for the construction of an idealization.. one can conclude instead that the traditional taxonomic theory of grammar. N. on which these philosophers' conception of grammatical form is based. In contrast. Syntactic Structures (Mouton & Co. 6. 1957) and P. even aside from the fact that such philosophical theories have not achieved much success. The Hague. this recognition has led twentieth-century philosophers--Russell. grammar might weil reveal the facts about logical form.

PostaI. Consider. Chomsky introduced the conception of a grammar as a generative. on the terminology introduced in Section 3. 'Underlying and Superficial Linguistic Structures'. op. Chomsky. is made to do work that. which in such cases absents the subject and future tense auxiliary constituent. as it stands. Chomsky's basic criticism is that such description cannot mark a variety of syntactic features because it fails to go below the surface structure of sentences. Postal. More elaborate segmentation and subc1assification cannot overcome the inherent inability of this form of description to represent relational information. 7 J. The Harvard Educational Review. What is wrong is that the superficial phrase-marker. Within this new conception. both receive the same syntactic analysis: «lohn)NP ((iS)(['. Consider the sentences : (i) 'lohn is easy to leave' and (ii) 'lohn is eager to leave'. ci!. Katz and P. does not mark the logical difTerence that in (i) 'lohn' is the object of the verb 'leave' whereas in (ii) 'lohn' is its subject.' The syntactic ambiguity of (iii) cannot be represented in its taxonomie description beeause a single (superficial) phrase-marker cannot explicate the different propositional structures underlying the terms of its ambiguity. eonsider a normal imperative such as: (iv) 'help him!' Ellipsis. cf. Finally. The notion or an underlying phrase-marker used here is the same as Chomsky's notion or a deep phrase-marker. which.:Url)A (to leave»vp)s This analysis. cannot be handled by a traditional taxonomie deseription because it deals only with the phonetically or orthographically realized constituents of a senten ce. it eannot do so long as it must still play its proper role.. in principle. In a traditional taxonomic description. beeause it is the only type of description sanctioned by the tradition al taxonomie theory of grammar. P. J.PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 109 The traditional taxonomie description of an utterance type is a single labelIed bracketing that segments it into continuous phonetic sketches and c1assifies them as constituents of one. 7 a form of syntactie description in which semantically significant grammatical relations can be adequately represented and shown to underlie the phonetic form of sentences on the basis of transformational rules that derive superficial phrase-markers from appropriate underlying phrase-markers by formally specified operations. To right this wrong. has a proper role to play in syntactic description. Aspects 0/ the Theory 0/ Syntax. . Rather. XXXIV (1964). a sentence like: (üi) 'lohn knows a kinder person than Bill. that of providing the most compact representation of the syntactic information required to determine the phonetic shape of a sentence. transformational system to supersede the eoneeption of a grammar as a set of segmentation and c1assification procedures. or another sort. is the superficial phrase-marker for (i) and (ü). and N. 6 For the syntactic motivation behind the claim that there are such phonetically unrealized constituents in normal imperatives. viz. the superficial phrase-marker. Chomsky and others developed the concept of an underlying phrase-marker. 6 These diffieulties cannot be remedied by enriching the complexity of superficial phrase-markers. furt her.

A. 10 For further discussion. 'Finitary Models of Language Users'. ed. R. the superficial phrase-marker for (iii). pp. Language.9 By this definition. 10 but where in one . G. New York. Thus. 8 (I) «(it) «one)NP «(leaves}y (John)NP)YP)s>NP(is)(easy)}yp)s (II) «JOhn)NP «isX(eagerX(John)NP (leaves}yp)S> A}YP)S The grammatical relations subject 0/ and object 0/ are defined in syntactic theory in terms of subconfigurations of symbols in underlying phrase-markers. . II.110 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEOR Y The logical difference between (i) and (ii) noted above can be indicated with the underlying phrase-markers. . Miller and N. 'John' in (i) is marked as the object of the verb 'leaves' because it occupies the Z-position and 'leaves' occupies the V-position in the appropriate subconfiguration of (I). R. XXXVII (1961). since a sentence can be assigned more than one underlying phrase-marker in a transformational syntactic component. and E. Luce. is (the person) Np((is)(more)(than)((Bill) Np((is)(kind) A) yp) s(kind) A) YP while in the other . cf. Given a configuration of the form «X) NP (Y)yp>s or «X)NP«Y)Y (Z)NP)YP)S' Xis the subject 0/ the verb Y and Z is the object 0/ the verb y. Handbook of Mathematica/ Psych%gy. vol. S. R. is (the person)NP((isXmore)(than)((the)(Bill)NP((knows}y(the person)NP}Yp)S (person» NP«is Xkind) A}YP) s(kind) A}YP 'For further discussion. Chomsky. viz. 476-80. Bush. A simplified version of their definitions is as folIows. Restricting the definition to underlying phrase-markers makes it possible to have a single definition because complex sentences are then handled in terms of the simple sentences from which they are constructed. C. syntactic ambiguities like those in (iii) can be represented in terms of appropriate1y different underlying phrase-markers transformationally associated with the same superficial phrase-marker. D. Smith. 342-65. . pp. and 'John' in (ii) is marked as the subject of'leaves' because it occupies the X-position and 'leaves' occupies the V-position in the appropriate subconfiguration of (11). «John) NP «knows) Y «aXkinder) (person)(than)(Bill» N~ y~ s is associated with two underlying phrase-markers both of which have the general form. •A Class of Complex Modifiers in English'. 9 Note that this definition reconstructs the intuitive notion that the subject is the noun phrase preceding the verb in a simple sentence and that the object is the noun phrase following it. . Galanter (John Wiley & Sons. . Further. 1963). cf.

Language. even a fully developed transformation al syntactic component would not provide all the theoretical machinery necessary to deal adequately with logical form. It has a dictionary that contains an account of the meaning of each syntactically atomic constituent in the language. whether it expresses more than one proposition. The dictionary 11 J. Engelwoad Cliffs. how many).PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 111 The former case underlies the term of the ambiguity on which the person that lohn knows is kinder than Bill is. reprinted in The Structure 01 Language: Readings in the Philosoph)' 01 Language. Katz (Prentice-Hall Inc. 124-94. representations of the senses of lexical items. pp. although it is dear from these examples that the distinction between underlying and superficial syntactic structure is a significant step toward the philosopher's distinction between logical form and grammatical form. by itself. XXXIX (1963). and a set of projection rules that provide the combinatorial machinery for representing the senses of compound constituents on the basis of representations ofthe senses ofthe lexical items that make them up.e. Finally. .e. J. 1964). Fador. This enables us to account for their syntactic relations and their semantic contribution without falsely characterizing the phonetic shape of the sentence. whether the two sentences express the same proposition). But. in ellipsis phonetically unrealized constituents can be specifiedinunderlying phrase-markers and deleted in the transformational derivation of the superficial phrase-marker. Foundations 01 Language. (4) analytic. (6) synthetic. (8) entails or is entailed by a given sentence. whether it expresses any proposition at all). 3. 2 (1967). A. (2) semantically ambiguous (i. and if so. The fact that a transformational syntactic component does not suffice. pp. an account of the logical form of a sentence should specify whether it is (1) semanticallyanomalous (i. 170--210. Katz. 'Recent Issues in Semantic Theory'. J. to deterrnine such semantic properties and relations has brought about the formulation of a conception of a semantic component designed to determine them. The semantic component formally reconstructs these compositional principles. no.. Philosophers have rightly held that an analysis of the logical form of a sentence should not only tell us about the formal relations between its constituents. but should also tell us about the semantic properties and relations of the proposition(s) expressed by it. (7) inconsistent with a given senten ce. 11 This conception is based on the idea that a speaker's ability to produce and understand sentences he has never before spoken or heard depends on his mastery of principles according to which the meaning of new and unfamiliar sentences can be obtained by a process in which the meaning of syntactically compound constituents is composed out of the meanings of their parts.e. val. J. 479-518. pp. i. and so on.e. 'The Structure of a Semantic Theory'. and the latter case underlies the term on which the person that lohn knows is kinder than the person Bill knows. A. J. ed. Fadar and J. In particular. J. (9) apresupposition of a given sentence. Katz and J. (3) a paraphrase of a given sentence (i. as would be required if we modified the superficial phrase-marker to account for them. (5) contradictory.

each of which consists of a lexical item written in phonological form. 'building'. A semantic marker is a theoretical term representing a dass of equivalent concepts. the grammaticaljorm of a sentence is its superjicial phrasemarker with its phonetic representation. which represents one sense of a lexical item. 'shadow'. Accordingly. Each derived reading is assigned to the compound constituent whose parts are the constituents whose readings were combined to form the derived reading. The semantic component operates on underlying phrase-markers.112 PHILOSOPHICAL RELEVANCE OF LlNGUlSTIC THEORY is a list of entries. while the syntactic and phonological components for the language comprise a theory of grammatical form for it. by induding the semantic marker (Physical Object) in a lexical reading for each of the words in the former group and excluding it from the lexical readings for words in the latter. We are now in a position to define the notions 'logical form' and 'grammatical form': The logicaljonn oj a sentence is the set ojUs semantically interpreted underlying phrase-markers. induding the whole sentence. 'man'. In this way each constituent in the underlying phrase-marker. a set of syntactic features. the projection rules combine lexical readings from sets assigned to different lexical items to form derived readings. Initially. which formally represent all the information about the meaning of the sentences to which they are assigned. and these are combined to form further derived readings. Aselection restriction states a condition-framed in terms of a requirement about the presence or absence of certain semantic markers---ilnder which a reading of a constituent can combine with readings of other constituents to form derived readings representing conceptually congruous senses of syntactically compound constituents. 'after-image'. converting them into semantically interpreted underlying phrase-markers. Similarly. etc. and so on. the syntactic and semantic components for a language comprise a theory of logical form for that language. for. a semantically interpreted underlying phrase-marker is an underlying phrase-marker each of whose brackets is assigned a maximal set of readings (where by 'maximal' is meant that the set contains every reading that can be formed by the projection rules without violating a se1ection restriction). 'togetherness'. A lexical reading. each of the lexical items in an underlying phrase-marker receives a subset of the lexical readings that it has in its dictionary entry. For example. and sentences). and a set of lexical readings. Semantic markers enable us to state empirical generalizations about the senses of words (expressions. is assigned a set of readings that represents its senses. the semantic marker (Physical Object) represents the dass of concepts of a material entity whose parts are spatially and temporally cO""ltiguous and move in unison each of us has in mind when we distinguish the meanings of words like 'chair'. 'stone'. we thereby express the generalization that the former words are similar in meaning in this respect but that the latter are not. etc. syntactic theory and semantic theory comprise .. Then. consists of a set of semantic markers and a selection restrietion. from the meanings of words like 'virtue'. Thus.

. a sentence is semantically unique. 7. a sentence entails another senten ce if each semantic marker in the reading for the latter's subject is already contained in the reading for the former's subject and each semantic marker in the reading for the latter's verb phrase is already contained in the reading for the former's verb phrase. 1 2 Finally. The Slructure 01 Language: Readings in the Philosophy 01 Language. Fifth. a sentence is a paraphrase of another sentence just in case the set of readings assigned to them have a member in common. 13 The adequacy of these definitions as solutions to the philosophical problems to which they are addressed is entirely a matter of their empirical justification. which. expresses exactly one proposition. two sentences are full paraphrases just in case each is assigned the same set of readings. Restricting our attention to syntactically unambiguous sentences. while syntactic theory and phonological theory comprise a theory of grammatical form in general. just in case the set of readings assigned to it contains one member. J. as can inconsistency and other related cases. a sentence is semantically ambiguous just in case the set of readings assigned to it contains n members. in turn. for n > 1. Fourth. Katz. Second. Definitions of these concepts thus constitute further support for the relevance of the thesis that I am defending. involve too many technicalities to be given here. and two. 13 A conditional sentence is analytic just in case its antecedent entails its consequent. they must be justified on the same evidential basis as any other linguistic 12 This is a simplified version of the definition of analyticity given in J. The Philosophy 01 Language. It also provides solutions to the philosophical problems of explicating concepts such as (1) through (9). This concept of analyticity may be regarded as a linguistically systematized version of Kant's concept of analyticity. a sentence is semantically anomalous just in case the set of readings assigned to it is empty. Third. i.PHILOSOPHICAL RELEV ANCE OF LlNO UISTIC THEOR Y 113 a theory of logical form in general. SEMANTIC PROPERTIES AND RELATIONS However. semantic theory does much more than complete the account of the distinction between logical form and grammatical form. we can provide a general idea of such definitions. however. that Kant's metaphorical notions of concept and of containment are replaced by the formal analogues of a reading and the inclusion of a set of semantic markers in another set. is part of linguistic theory. that Kant's somewhat vague and restricted notions of subject and predicate are replaced by the formally defined grammatical relations subject 01 S and verb phrase 01 S. Katz. Since such definitions are part of a semantic theory. and in J. 'Analyticity and Contradiction in Natural Language'. J. First. a sentence is analytic if there is a reading assigned to it that is derived from a reading for its subject and a readiag for its verb phrase such that the latter contains no semantic markers not already in the former.e. their definitions. Sixth. This explicates the notion that what prevents a sentence from having a meaningful interpretation are conceptual incongruities between senses of its parts that keep these senses from compositionally forming a sense for the whole sentence. The semantic properties of contradiction and syntheticity can also be defined. with two refinements: one.

'Two Dogmas of Empiricism'. This language-independent generaIity is guaranteed by the 14 W. however. 1953).114 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY universal. To remove from these definitions the stigma that automatically attaches to definitions of semantic properties and relations since Quine's attack on the analyticsynthetic distinction. The above definition of analyticity. Quine. as determined by its semantically interpreted underlying phrase-marker. the definition of 'analytic'. or any others for that matter. the term 'analytic' is just an unexplained label. One of Quine's major criticisms was that Carnap's explication of analyticity. From a Logica/ Point 01 View (Harvard University Press. whose own definition quickly brings us back to the original one without offering a genuine analysis of any of them.14 This case is chosen as our example because of its prominence in the literature. Thus. labelling a sentence as analytic attributes to it that semantic structure formalized in the definition that introduces 'analytic' into linguistic theory. But. Given the semantically interpreted underlying phrase-marker for a sentence S in a language L and the definition of a semantic property P or relation R. we can deduce a prediction about whether S has P or bears R to some other sentence. as weil as that of any of the other semantic properties and relations defined in semantic theory. Quine criticizes Carnap for merely labeIling sentences as analytic without ever indicating just what is attributed to a sentence so labelled. I will show how the above definition of analyticity avoids the criticism he levelled against Carnap's explication of analyticity. . on our account. contradiction. as Quine requires. their empirical evaluation consists in verifying the predictions to which they lead about the semantic properties and relations of sentences from natural languages. labelling a sentence as analytic depends on its semantic structure. This deduction is merely a matter of determining whether or not the semantically interpreted underlying phrase-marker(s) of S possess the formal features required by the definition of P or R. they are formulated for variable S and L. Lastly. were used to define it. On Carnap's account. Further. The unique feature of the above definitions is that the defining condition in each is stated exclusively in terms of a different set of formal features in semantically interpreted underlying phrase-markers. the justification of these definitions depends on whether such predictions accord with the judgements of ftuent speakers about c1ear cases from L. cannot be criticized on grounds of such circularity because it is not the case that any of these related terms. Moreover. cannot be criticized for being too particularistic because. V. and related concepts merely defines one of these concepts in terms of others. but what I am going to say by way of a defence against Quine's criticism of this concept will apply directly to similar criticisms of any of the other semantic properties and relations. Cambridge. Such predictions can be checked against the ways that ftuent speakers of L sort sentences in terms of their naive linguistic intuitions. no appeal to such definitions is made in the process whereby a semantically interpreted underlying phrase-marker obtains the formal features by virtue of which it satisfies such a defining condition. Thus. Hence.

LXIV (1967).PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 115 fact that they are given in Jinguistic theory and that their defining conditions are formulated exclusively in terms of semantically interpreted underlying phrasemarkers. we can obtain an analysis of this not ion and thereby arrive at a clearer understanding of l' For a more detailed and complete account of how my explication of analyticity avoids Quine's criticisms of Camap's explication. still. SEMANTIC CATEGORIES The last philosophical problem whose solution I here want to treat within linguistic theory is that of semantic categories. unanalysable. the most general classes into which the concepts from all fields of knowledge divide. quantity. but it does not provide any analysis of this notion. Aristotelian categories claim to be the most abstract c1assificational divisions under which ideas of any sort can be subsumed. There is. 'Some Remarks on Quine on Analyticity'. 15 8. Aristotle enumerated ten (perhaps eight) such categories: substance. Its application relies on intuitive judgements about what are and what are not the most general answers to the test questions. The most influential treatment of semantic categories was certainly Aristotle's. 36--52. . The criterion itself presupposes our intuitive understanding of the notion maximal generality in the domain 0/ possible concepts as a condition for its application. quality. But he did not explain how he chose these categories nor how he decided that no others belong to the list. post ure. it would tell us nothing about the nature of the categories it sorted out. of course. place. we know no more about the idea of a category than we did before having obtained its extension in this manner. action. Consequently. vol. Thus. substance qualifies as a category because 'a substance' is the most general answer to such questions as 'What is Socrates?'. time. Likewise. relation. The Journal 0/ Phi/osophy. much room for doubt about these answers. possession. The criterion he mentions falls short of providing a satisfactory principle for categoryhood. Katz. without clarifying either for ourselves or our informants just what makes these judgements appropriate judgements about the concepts concerned. see J. and passivity (with the last two of somewhat questionable status). J. but. quality is a category because 'a quality' is the most general answer to such questions as 'What is green ?'. pp. They are the ultimate. even if this criterion were fairly successful in picking out cases that we would intuitively regard as among the most abstract classificational divisions in our conceptual system. assuming that those things that are proper answers to such questions are just those things that are the most general genera for classifying concepts. maximally general set of natural kinds that are given in naturallanguages. Aristotle's criterion comes to this: each category is the most general answer to a question of the form 'What is X T. which are associated with each sentence in any linguistic description. If we can embed the theory of semantic categories into linguistic theory.

Hence. etc. As we have characterized the entries for lexical items in the dictionary of a semantic component. 'child'. etc. we have to ask by what means can we distinguish two sets of semantic markers: first.. viz. whenever (Human) occurs in a lexical reading. That is. (Artifact). by virtue of the generalization that the occurrence of (Physical Object) is determined by the occurrence of (Human). from the viewpoint of the whole dictionary with its thousands of entries. viz. is predictable from the occurrence of (Human) in these lexical readings and the generalization that says that (Human) never occurs in a lexical reading unless (Physical Object) occurs. 'spinster'. whenever (Human). the subset of the set ofuniversal semantic markers (as given in semantic theory) whose members represent concepts having the required degree of generality for language in general. so does (Physical Object). and.. this case is not an isolated one. 'man'. there will be an incredible amount of redundancy in the specification of the senses of lexical items unless we provide some way to eliminate such unnecessary . With a means of representing such regularities. viz. linguistic descriptions that are written in accord with semantic theory. etc. Thus. second. the occurrence of (Physical Object) in lexical readings that contain (Human) is actually redundant. are subject to a regularity governing their occurrence with respect to one another in the dictionary. (AnimaI) occurs whenever (MammaI) does. almost every lexical reading would exhibit a high degree of redundancy in the manner in which it specifies the semantic information in asense. occurs in a lexical reading. the semantic markers (Physical Object) and (Human) that appear in the lexical readings for the words 'bachelor'. as we have so far described the dictionary. we must provide some empirically motivated way for determining the semantic categories of a particular language and the semantic categories of natural language. To do this. For exam!1le. we have made no provision for the formulation of such generalizations. Formulated in this way. 'man'. (Anima\). the actual occurrence of the semantic marker (Physical Object) is dispensable because its occurrence in the lexical readings for 'bachelor'. 'spinster'. but other semantic markers besides (Physical Object) are redundant in the same way. Not only is there a broader regularity covering the occurrence of (Physical Object).116 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY semantic categories. Without a means for expressing such regularities. lexical readings contain a semantic marker for each independent conceptual component of the sense that they represent. can be correctly criticized for having missed an important generalization about their languages. and are similarly predictable from generalizations expressing the appropriate regularities.. as so far formulated. (Plant). Hence. we have no way to express this regularity in the formalism of semantic theory so that the redundancy of (Physical Object) for these cases is forced on us in order that we be able to fully represent the senses of these words. at present. However. Moreover. so does (Physical Object). the subset of the set of semantic markers appearing in the dictionary of a linguistic description of some particular language whose members represent concepts having the required degree of generality in that language.

making the dictionary a more economical formulation of the lexical information in the language. . Hefe. Thus. In the case discussed above.. We dejine a semantic category 0/ L to be any concept represented by a semantic marker that occurs on the right-hand side 0/ some rufe in the redundancy rufes in the dictionary 0/ the linguistic description 0/ L. (Artifact). (Mn) by dropping the occurrence of the semantic marker (Physical Object) from those lexical readings. The obvious way to make dictionary entries more economical and to provide a means of expressing these regularities is to extend the conception of the dictionary presented in Section 6 so that the dictionary indudes rules wbich state the appropriate generalizations and thereby enable us to exdude redundant semantic markers from lexical readings... is where the application of tbis formalism to the question of semantic categories comes in.... The redundancy rules not only simplify the statement of the dictionary and state significant lexical generalizations.)-+ (Physical Object) Adding tbis rule to the dictionary enables us to capitalize on the regularity noted above and economize lexical readings that contain one of the semantic markers (MI)' (Mz )' •. v (Human) v (Anima!) v (Artifact) v (Plant) v . . (Animal). Such rules will comprise a new component of the dictionary. we have an example of a rule of tbis type. For such rules can be interpreted as saying that the concepts represented by the semantic markers on the left-hand side of the arrow are induded in. or subsumed under. these rules will be of the form where (MJ is distinct from each (MX 1 ~ i ~ n and where v is the symbol for disjunction. but does not occur on the feft-hand side 0/ any rufe in that set 0/ redundancy rufes. they also represent indusion relations among the concepts represented by semantic markers. the concept represented by the semantic marker on the right-hand side.PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY 117 occurrences of semantic markers by finding some formalism to express these lexical generalizations. (Plant) . whose list of entries can now contain only lexical readings in maximally reduced form. we simply check over the list of redundancy rules in the linguistic description of L and pick each semantic marker for which there is a rule that says that that marker subsumes other markers. then. The significance of tbis definition is . v (M. to find the semantic categories of a particular language. (Human). In general.. Using the redundancy rules in the dictionary of a linguistic description for a language L. So much for the formalism. (MI) v (M z) v . These rules thus function to compress the readings in dictionary entries. we can formally determine wbich of the semantic markers in that linguistic description represents the semantic categories of L. and for wbich there is no rule that says that that marker is subsumed under other markers.

We define the semantic categories of language to be those concepts represented by the semantic markers belonging to the intersection of the sets of semantic categories for each particular natural language LI' L 2. First. • • • . This is the same sort of empirical justification appealed to in other branches of science when it is claimed that some theoretical ac count is best because it employs the simplest laws for describing the phenomena under study. That there are problems to which linguistic theory is irrelevant need not bequestioned. Ln in the manner just described. then it is quite natural to ask to what phIlosophical problems is linguistic theory irrelevant. finally. we can also formally determine the semantic markers that represent semantic categories of language in general. a semantic category of language is a concept represented by a semantic marker that is found in each and every set of semantic categories for particular natural languages.118 PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY two-fold. that the unexplicated notion of maximal generality on which the Aristotelian notion of categories is based is here explicated formally. it provides us with a clear-cut empirical basis for deciding what is a semantic category of language. First. i. That is. it provides us with a formal means to determine the categories of language.e. The significance of this definition is parallel to that of t. THE SCOPE OF PHILOSOPHICAL RELEVANCE If the considerations put forth in this paper on behalf of the thesis that linguistic theory is relevant to the solution of philosophical problems are convincing. Ln' as obtainedfrom the redundancy rules in the dictionaries ofthe linguistic descriptions for LI' L 2 . ••• . Such justification for a set of putative semantic categories is a matter of empirically establishing that no simpler formulation of the lexical readings in the dictionary of the language is provided by redundancy rules other than those which. it makes it possible to justify empirical claims to the etTect that such-and-such concepts are the semantic categories for a given language. in terms of membership in the set of semantic markers that comprises the intersection of the sets of semantic markers that are semantic categories for the natural languages. 9. where each of the semantic markers in these latter sets is obtained by the condition that it appears on the right-hand side of a redundancy rule but not on the left-hand side of any. Ln' Notice. The justification for a claim that some concept is a semantic category of language can be given on the basis of the same evidence that warrants the claims that that concept is a semantic category of LI' L 2' • • • . yield the set of putative semantic categories in question. On the basis of these considerations. Second. ._ it makes it possible for us to formally determine the semantic categories for a given language with respect to a linguistic description for it. by the above definition of semantic categories of L.he previous one. and second. the semantic categories for all natural languages as opposed to the semantic categories of some particular naturallanguage.

PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY

119

for the philosophy of mathematics and the philosophy of science provide abundant examples. Consequently, it would be highly desirable to have some handy criterion by which to dt;Cide whether a philosophical problem is essentially about the underIying conceptual structure of natural languages or about something else. But I find it hard to believe that we can have such a criterion, for not only do philosophical problems not come to us ear-marked as either linguistic or non-linguistic but, even given a fully developed linguistic theory, it would require considerable further inquiry to discover whether some portion of that theory is relevant to some particular philosophical problem and much further argument to establish that the relevance is such that linguistic theory provides an authentie solution for it. We have so far considered only one sense in which linguistic theory can be relevant to philosophie al questions. In condusion, I would like to consider another way in which it can have philosophical relevance, one that does not depend on the theory offering us the concepts that answer the philosophie al question. The problem of innate ideas, the crux of the controversy between empiricists and rationalists, is a case to which linguistic theory has a significant application in a somewhat different way than the one in which it applies to the problems discussed above. 16 This problem can be recast as the question of whether the acquisition of a natural language can be explained better on the basis of the empirieist hypothesis that the mind starts out as a tabula rasa, or on the basis of the rationalist hypothesis that the mind starts out with a rich stock of innately fixed principles which determine the general form of the rules for a natural language. Given that the child obtains his inner representation of the rules of a language from the linguistic data to which he is exposed during his formative years, we may consider the child's mind to be a black box whose input is such linguistic data and whose output is an internalization of the linguistic description of the language. Accordingly, we ask whether the empirieist hypothesis that this internalization is obtained by processing sensory data on the basis of principles of associative learning, or the rationalist hypothesis that this internalization is obtained by a specialization of the innate system of principles when they are activated by appropriate sensory stimulation, is the better account of how the black box converts its input into its output. We have a fairly dear idea ofthe associative principles with which the empirieist is willing to credit the child's mind prior to experience, but it is by no me ans dear what are the innately fixed principles concerning the general form of language on the rationalist's account. Here the relevance of linguistic theory is, then, that it provides a statement of the principles required to formulate the rationalist hypothesis in specific terms. The question is, therefore, whether we must assume as rich a conception of innate structure as is given by linguistic theory's account of the universals of language in order to explain language acquisition.
16

For a more complete discussion of the problem of innate ideas, cf. eh. 5 of my book, The Philosophy

0/ Language.

120

PHILOSOPHICAL RELEVANCE OF LINGUISTIC THEORY

Notice, however, that although this question, which reformulates the problem of innate ideas, can only be raised in an explicit form when linguistic theory supplies the conception of innate structure for the rationalist hypothesis, its answer is not given by linguistic theory. Linguistic theory does not validate the rationalist position in its controversy with the empiricist position, since it is outside the scope of linguistic theory to decide which of these two positions is best supported by the facts about the linguistic information available to the child and about how he copes with them. This case was introduced not only to show that the relevance of linguistic theory to philosophy goes beyond solving philosophical problems, but also to show that it can be relevant in the specific sense of providing the means by which a philosophical problem can be reformulated in a manner that makes it more susceptible to solution. Whether linguistic theory is relevant to philosophical investigation in still other ways must remain a matter for further philosophical and linguistic inquiry.

VII
SYMPOSIUM ON INNATE IDEAS
(a) RE CENT CONTRIBUTIONS TO THE THEOR Y OF INNATE IDEAS

Summary

0/ Oral Presentation

NOAM CHOMSKY

I THINK that it will be useful to separate two issues in the discussion of our present

topic---one is the issue of historical interpretation, namely, what in fact was the content of the classical doctrine of innate ideas, let us say, in Descartes and Leibniz; the second is the substantive issue, namely, in the light of the information presently available, what can we say about the prerequisites for the acquisition of knowledge -what can we postulate regarding the psychologically apriori principles that determine the character of learning and the nature of what is acquired. These are independent issues; each is interesting in its own right, and I will have a few things to say about each. What I would like to suggest is that contemporary research supports a theory of psychological apriori principles that bears a striking resemblance to the classical doctrine of innate ideas. The separateness of these issues must, nevertheless, be kept clearly in mind. The particular aspect of the substantive issue that I will be concemed with is the problem of acquisition of language. I think that a consideration of the nature of linguistic structure can shed some light on certain cIassical questions concerning the origin of ideas. To provide a framework for the discussion, let us consider the problem of designing a model of language-acquisition, an abstract 'Ianguage-acquisition device' that duplicates certain aspects ofthe achievement ofthe human who succeeds in acquiring linguistic competence. We can take this device to be an input-output system data
-+

ILA I -+ knowledge

To study the substantive issue, we first attempt to determine the nature of the output in many cases, and then to determine the character of the function relating input
From Boston Studies in the Philosophy 01 Science, vol. III (The Humanities Press, New York, 1968), pp. 81-107. Reprinted by permission of the authors, New York: Humanities Press Inc., and D. Reidel Publishing Company.

for example. thus extending the information bearing on this input-output relation. The only conditions we must meet in developing such a model of innate mental capacity are those provided by the diversity of language. In either case. the number of seconds in a lifetime is ridiculously small. The problem is quite analogous to the problem of studying the innate principles that make it possible for a bird to acquire the knowledge that expresses itself in nest-building or in song-production. we would try to determine from the behaviour of the mature animal just what is the nature of its competence. Compared with the number of sentences that a child can produce or interpret with ease. we can carry out the analogous study of language-acquisition under a variety of different input conditions. data. Thus we cannot suppose that the specific rules of English are innate to the device and these alone. and we would then try to construct a secondorder hypothesis as to the innate principles that provide this competence on the basis of presented data. The competence of an adult. vast differences in intelligence have . once we have developed some insight into the nature of the resulting competence. To study this question. there is no place for any dogmatic or arbitrary assumptions about the intrinsie. as indicated by actual performance. Similarly. there is no reason for any dogmatic assumptions about the nature of LA. Hence the data available as input are only aminute sampie ofthe linguistic material that has been thoroughly mastered. On the other hand. with data drawn from a variety of languages. Notice that the conditions of the problem provide an upper bound and a lower bound on the structure that we may suppose to be innate to the acquisition device. since this would be inconsistent with the observation that Chinese can be learned as readily as English. On apriori grounds.122 SYMPOSIUM ON INNA TE IDEAS to output. we must attribute to the device a sufficiently rich structure so that the output can be attained within the observed limits of time. we can turn to the investigation of the innate mental functions that provide for the acquisition of this competence. it becomes immediately apparent that it is no easy matter to formulate a hypothesis about innate structure that is rich enough to meet the condition of empirical adequacy. Notice that this is an entirely empirical matter. or even a young child. Furthermore. and by the necessity to provide empirically attested competence within the observed empirical conditions. Furthermore. innate structure of the device LA. We might deepen the investigation by manipulating input conditions. there is no way to determine the extent to which an instinctual component enters into these acts. great diversity of input conditions does not lead to a wide diversity in resulting competence. When we face the problem of developing such a model in a serious way. and access. is such that we must attribute to him a knowledge of language that extends far beyond anything that he has leamed. in the case of language-acquisition. so far as we can detect. We cannot impose so much structure on the ~evice that acquisition of some attested language is ruled out. To repeat. the diversity oflanguages. The upper bound is provided by the diversity of resulting competence--in our case.

The sentences used in everyday discourse are not 'familiar sentences' or 'generalizations of familiar sentences' in terms of any known process of generalization. I think real progress has been made in recent years in our understanding of the nature of grammatical rules and the manner in which they function to assign semantic interpretations to phonetically represented signals. as proposed in recent behaviourist speculations. I. and that it is precisely in this area that one can find results that have some bearing on the nature of a language-acquisition device. given the data to which the child is exposed. even to speak of 'familiar sentences' is an absurdity. Such observations as these lead one to suspect. Evidently. A person's competence can be represented by a grammar. The idea that sentences or sentence-forms are learned by association or conditioning or 'training'. as long as they consist of elementary units that he knows and are composed by the rules he has intemalized. and that this task is entirely beyond the capacities of an otherwise intelligent ape. There are several aspects of normal linguistic competence that are crucial to this discussion. . The normal use of language is 'creative' in this sense. of course). which is a system of rules for pairing semantic and phonetic interpretations.SYMPOSIUM ON INNATE IDEAS 123 only a small effect on resulting competence. It seems to me that this initial expectation is strongly supported by a deeper study of linguistic competence. it is important to realize that in no technical sense of these words can language use be regarded as a matter of 'habit' or can language be regarded as 'a complex of dispositions to respond'. external stimuli or independently identifiable intern al states. he is capable. Once a person has mastered the rules (unconsciously. of using them to assign semantic interpretations to signals quite independently of whether he has been exposed to them or their parts. one naturally turns to a deeper investigation of the nature of grammars. CREATIVE ASPECT OF LANGUAGE USE By this phrase I refer to the ability to produce and interpret new sentences in independence from 'stimulus control'-i. More generally. is entirely at variance with obvious fact. We observe further that the tremendous intellectual accomplishment of language-acquisition is carried out at aperiod of life when the child is capable of little else. as was widely noted in traditional rationalist linguistic theory. these rules operate over an infinite range. In fact. The central problem in designing a language acquisition device is to show how such a system of rules can emerge.e. in principle. In order to gain some insight into thi's question. from the start. that we are dealing with a species-specific capacity with a largely innate component.

consider the sentence 'The doctor examined John'. 'Noun-Phrase'. determined from the meanings of its minimal items by certain general rules that make use of the grammatical relations expressed by the labelIed bracketing. 'past tense'. must be the same in all of these examples as the relations in 'the doctor examined John'. in this case). It is transparent that grammatical relations (e. 'John was persuaded to be exarnined by the doctor'. namely. the categories 'Sentence'. and the categories to which the bracketed elements belong (that is. that deep structure is distinct from surface structure. as expressed by the deep structure. evidently. Object-of'. Thus consider the slightly more complex sentences: 'John was examined by the doctor'. 'Deterrniner'. universal set. by replacing 'the doctor to exarnine John' by 'John to be exarnined by the doctor' in the two cases. We can define the 'surface structure' of an utterance as its labelled bracketing. etc. In the example just given.g.124 SYMPOSIUM ON INNATE IDEAS 11. that the similarity of surface structure masks a significant difference in deep structure. 'someone persuaded the doctor to examine John'. The phonetic form of this sentence depends on the intrinsic phonological character of its minimal items ('The'. But it is obvious that these cannot in general be identified. 'examine'. 'Noun'. by application of these rules of semantic interpretation. It is c1ear. Concretely. 'Subject-of. we would not be far wrong ifwe took the deep structure to be identical with the surface structure. 'the doctor was persuaded to examine John'. So far. we can assert that there is very strong evidence that the phonetic form of a sentence is determined by its labelIed bracketing by phonological rules that operate in accordance with certain very abstract but quite universal principles of ordering and organization. where the brackets are assigned appropriate categorial labels from a fixed. Let us define the 'deep structure' of a sentence to be that labelIed bracketing that determines its intrinsic meaning.) can be defined in terms of such a labelIed bracketing. I have only made a negative point. ABSTRACTNESS OF PRINCIPLES OF SENTENCE INTERPRETATiON A grammar consists of syntactic rules that generate certain underlying abstract objects. the grammatical relations among doctor. 'doctor'. examine. consider the two sentences : someone expected the doctor to examine J ohn someone persuaded the doctor to examine John. With terms defined in this way. The meaning of the sentence 'the doctor exarnined John' is. and rules of semantic and phonological interpretation that assign an intrinsic meaning and an ideal phonetic representation to these abstract objects. in this case. the bracketing of the sentence (that is. as we can see. as [[[thel [doctorll [[examinedl [Johnll]). Much more important is the fact that there is very strong evidence for a particular solution to the problem of how deep and surface structures . Evidently. 'John'). 'Verb'. and John. But the surface structures will differ greatly. immediately. Furthermore.

to determine how the child determines that the structure of his language has the specific characteristics that empirical investigation of language leads us to postulate. There is no sense of 'simplicity' in which this design for language can be intelligibly described as 'most simple'. Furthermore. Let us now return to the problem of designing a language-acquisition device. of sentences that deviate in form from the idealized structures defined by the gramrnar that he develops. consequently.SYMPOSIUM ON INNATE IDEAS 125 are related. the underlying abstract structures and the rules that apply to them have highly restricted properties that seem to be uniform over languages and over different individuals speaking the same language. A theory of language acquisition must explain how tbis knowledge of abstract underlying forms and the principles that manipulate them comes to be acquired and freely used. given the meagre evidence available to hirn. 111. Notice. UNIVERSAL CHARACTER OF LlNGUISTlC STRUCTURE So far as evidence is available. There is no apriori necessity for a language to be organized in this highly specific and most peculiar way. Thus sentences may have very similar underlying structures despite the great diversity of physical form. but very degenerate in quality. it would be quite impossible to argue that this structure is simply an accidental consequence of 'common descent'. and that. Nor is there any content to the claim that tbis design is somehow 'logical'. they have nothing remotely like a point by point correspondence to signals. incidentally. that the evidence is not only meagre in scope. is that it involves formal manipulations of structures that are highly abstract. precisely. Deep structures seem to be very similar from language to language. it seems that very heavy conditions on the form of grammar are universal. and that seem to be largely invariant with respect to intelligence and specific experience. The available evidence shows that the output of tbis device is a system of recursive rules that provide the basis for the creative aspect of language use and that manipulate highly abstract structures. and one which seems inescapable. Furthermore. it is enough to point out that this structure must be rediscovered by each child who leams the language. The details ofthis theory need not concem us for the present. in the sense that their relation to signals is defined by a long sequence of formal rules. A crucial feature of it. and how deep and surface structures are formed by the syntactic component of the grammar. The problem is. and the rules that manipulate and interpret them also seem to be drawn from a very narrow class of conceivable formal operations. Quite apart from questions of historical accuracy. An engineer faced with the problem of designing a device meeting the given input-output conditions would naturally conclude that the basic properties of the output are a . and diverse underlying structures despite similarity of surface form. in large measure. Thus the child learns the principles of sentence formation and sentence interpretation on the basis of a corpus of data that consists.

a schema that defines the dass of possible grammars . Evidently. we are led by such evidence as I have mentioned to suppose that this device in some manner incorporates: a phonetic theory that defines the dass of possible phonetic representations . one can seek a hypothesis conceming language-acquisition that falls between the upper and lower bounds. the device might proceed to acquire knowledge of a language in the following way: the given schema for grammar specifies the dass of possible hypotheses . the evaluation measure selects the highest valued grammar compatible with the data. (2) abstract nature of deep structure. Once a hypothesis-a particular grammar-is selected. First. the leamer knows the language defined by this grammar. Proceeding in this way. discussed above. Consequently. More specifically. which has been elaborately discussed in many publications on transforrnational generative grammar. and that the linguistic theories that were developed within the framewark of this discussion are. for language leaming to take place the dasS' of possible hypotheses-the schema for grarnrnar-must be heavily restricted. a general method far interpreting grammars that assigns a semantic and phonetic interpretation to each sentence. the method of interpretation perrnits each hypothesis to be tested against the input data. a method of evaluation that assigns some measure of 'complexity' to grammars. We can give it content by specifying the language-acquisition system along the lines just outlined. theories of transformational grammar. but this is not the place to pursue this matter. This account is schematic an4 idealized. Nor is there any plausible alternative to this assumption. so far as I can see. I think that very plausible and concrete specifications can be given. in essence.126 SYMPOSIUM ON INNATE IDEAS consequence of the design of the device. the apriori principles that determine how and in what form such knowledge is acquired. a semantic theory that defines the dass of possible semantic representations. given a grarnrnar. Given such a specification. Let me now try to place this discussion in its historical context. It is interesting to observe that these three aspects of language are discussed in the rationalist philosophy of the 17th century and its aftermath. it would be historically accurate to describe the views regarding . that are set by the nature of the problem. (3) apparent universality of the extremely special system of mechanisms formalized now as transforrnational grammar. in particular. I mentioned three crucial aspects of linguistic competence: (1) creative aspect of language use. along these lines. Thus his knowledge extends far beyond his experience and is not a 'generalization' from his experience in any significant sense of 'generalization' (except. I have so far been discussing only the substantive issue of the prerequisites for acquisition of knowledge of language. the sense defined by the intrinsic structure of the language-acquisition device). trivially. he is capable of pairing semantic and phonetic interpretations over an indefinite range of sentences to which he has never been exposed.

if we take the essence of this view to be that the general character of knowledge. Thus only the procedures and mechanisms for acquisition of knowledge consititute an innate property. which has attempted to elaborate a system of inductive analytic procedures of segmentation and classification that can be applied to data to determine a grammar. simply does not have the properties of an associative net or ahabit family. In the case of language-acquisition. but of an entirely ditTerent approach to the question. This stimulation provides the occasion for the mind to . I employed it. it is difficult to imagine how the vague suggestions about conditioning and associative nets that one finds in philosophical and psychological speculations of an empiricist cast might be refined or elaborated so as to provide for attested competence. and (2) certain analytical mechanisms or inductive principles or mechanisms of association. the schematism assigned as an innate property to the language-acquisition device determines the form of knowledge (in one of the many traditional senses of 'form'). It is conceivable that these methods might be somehow refined to the point where they can provide the surface structures of many utterances. the categories in which it is expressed or internally represented. again. In the traditional view a condition for these innate mechanisms to become activated is that appropriate stimulation must be presented. are determined by the nature of the mind. The role of experience is only to cause the innate schematism to be activated. but the only relatively clear attempt to work out some specific ac count of them is in modern structural linguistics. A system of rules for gene rating deep structures and relating them to surface structures. there has been much empiricist speculation about what these mechanisms may be. Furthermore. It is quite inconceivable that they can be developed to the point where they can provide deep structures or the abstract principles that generate deep structures and relate them to surface structures.SYMPOSIUM ON INNATE IDEAS 127 language structure just outlined as a rationalist conception of the nature of language. hence no elaboration of principles for developing such structures can be appropriate to the problem of designing a language-acquisition device. and the basic principles that underlie it. are acquired by application of the innate inductive principles to this initially analysed experience. beyond this. and then to be ditTerentiated and specified in a particular mann er. In our case. What is assumed is that a preliminary analysis of experience is provided by the peripheral processing mechanisms and that one's concepts and knowledge. in the manner characteristic of naturallanguage. In sharp contrast to the rationalist view. we have the classical empiricist assumption that what is innate is (1) certain elementary mechanisms of peripheral processing (a receptor system). to support what might fairly be called a rationalist conception of acquisition of knowledge. in the classical fashion. This is not a matter of further refinement. Similarly. I have said nothing explicit so far about the doctrine that there are innate ideas and innate principles of various kinds that determine the character of what can be known in what may be a rather restricted and highly organized way.

There is. association. by . therefore.128 SYMPOSIUM ON INNATE IDEAS apply certain innate interpretive principles. but rather the authentic triangle' (Haldane and Ross. and Cartesian Linguistics (1966). In this sense. For Leibniz what is innate is certain principles (in general. and avoiding what would be the normal assumption. or philosophy. nothing known in psychology or physiology that suggests that the empiricist approach is weil motivated. conditioning. in the way in which geometricians consider it. Similar views are elaborated at length in rationalist speculative psychology. apprehend not it itself. that 'enter into our thoughts. Methodology and Philosophy o[ Science. there are no other grounds for pursuing these empiricist speculations. when we see the composite figure. Further empirieist speculations contribute nothing that even faintly suggests a way of overcoming the intrinsic limitations of the methods that have so far been proposed and elaborated. vol. 227). the idea of triangle is innate. such a proposal raises non-trivial questions of historical interpretation.. 'Ideas and truths are for us innate as inclinations. unprejudiced by doctrine. certain concepts that proceed from 'the power of understanding' itself. 'analogy'.' Experience serves to elicit. is essentially this. we. What does seem to me fairly clear is that the present situation with regard to the study of language learning. But because we already possess within us the idea of a true triangle. see also Chomsky. 11. just as the statue of Mercury is contained in a rough block of wood. or that gives any grounds for scepticism concerning the rationalist alternative sketched above. V): ' . 'Explanatory Models in Linguistics'. and so on) of any sort that have been developed or discussed within linguistics. so understood. see Chomsky. or natural potentialities. substitution procedures. of which they form the soul and the connection'. in particular. unconscious). because the true triangle is contained in this figure. To take a typical example from Descartes (Reply to Objections. This evidence shows clearly that knowledge of language cannot arise by application of step-by-step inductive operations (segmentation. from the faculty of thinking rather than from external objects. habits. are fully in accord with the doctrine of innate ideas. Ch. Of course. Furthermore. When first in infancy we see a triangular figure depicted on paper. in Logic. and it can be more easily conceived by our mind than the more complex figure ofthe tri angle drawn on paper. and other aspects of human intellectual achievement of comparable intricacy. dispositions. not to form. these innate structures. 1. that one would formulate if confronted with empirical evidence of the sort sketched above. It seems to me that the conclusions regarding the nature of language-acquisition. classification. ed. We have a certain amount of evidence about the gramm ars that must be the output of an acquisition model. For further discussion of the question of historical interpretation.. discussed above. psychology. For further discussion of matters touched on here. and can be regarded as providing a kind of substantiation and further development of this doctrine. p. this figure cannot show us how areal tri angle ought to be conceived. Aspects of the Theory of Syntax (1965).

The Philosophy 0/ Language (1966). J. Review of A. ed. . Elements 0/ General Linguistics (1966). Katz. Martinet.SYMPOSIUM ON INNATE IDEAS 129 E. and the selections in section VI of The Structure 0/ Language. Nagel. Suppes. by J. M. P. Katz (1964). Readings in the Philosophy 0/ Language. Postal. Fodor and J. Tarski (1962). P. and A.

and Martians would.H. and I beg his pardon in advance if I misstate the I.. as opposed to a fact about what has been leamed. the I.~. (Possible I What 'built in' means is highly unc1ear in this context. I have heard Chomsky maintain. or at least a theory. is the hypothesis that the human brain is 'programmed' at birth in some quite specijic and structured aspects of human natural language. I have also relied upon Chomsky's paper 'Explanatory Models in Linguistics'. in which the I. then. rather all the members of 1: have some quite strong similarities. and if the 'Martians' speak a language whose grammar does not belong to the subdass 1: of the dass of all transformational grammars. plays a considerable role. then.H. If intelligent non-terrestrial life-say. To beg in. It should be noticed that Chomsky has never offered even a schematic account of the sort of device that is supposed to be present in the brain.H. or misrepresent any of the arguments for it. a human child brought up by Martians would fail to acquire language. 2.H.H. not the explanation being sought. I owe a debt of gratitude to Chomsky for having repeatedly exposed me to the I. as characteristics of all human natural languages. in any detail. In addition to relying upon oral communications from Chomsky. THE INNATENESS HYPOTHESIS THE 'innateness hypothesis' (henceforth. .(b) THE 'INNATENESS HYPOTHESIS' AND EXPLANATORY MODELS IN LINGUISTICS HILARY PUTNAM I. it may be meaningless. and that is supposed to do the job of selecting the highest weighted grarnmar compatible with the data. The weighting function by itself determines only the relative ease with which various grammars can be leamed by a human being.') is a daring-or apparently daring. experience similar difficulties with human tongues. Martians--exists. The details of this programming are spelled out in some detail in 'Explanaiory Models in Linguistics'. But only a description.e. We should assurne that the speaker has 'built in' 1 a function which assigns weights to the grammars GI' G 2' G J' • • • in a certain dass 1.' is not dass of all possible transformation al grammars . humans (except possibly for a few geniuses or linguistic experts) would be unable to leam Martian. then doubtless this is 'innate' in the sense of being a fact about human learning potential. But this sort of fact is what leaming theory tries to account for. the 'I. These similarities appear as 'Iinguistic universals' --i. If a grammar GI can be leamed more easily than a grammar G. conversely.: of transformational grammars. of such a device could properly be caIled an innateness hypothesis at all. I have relied heavily in what follows on oral communications from hirn. in which case it is not daring-hypothesis proposed by Noam Chomsky.

pp. the performance of the human child in learning his native language may be understood as folIows. or conjecture.SYMPOSIUM ON INNATE IDEAS 131 difficulties in pronunciation are not at issue here. and some information concerning which utterances count as repetitions of earlier utterances. and may be assumed not to exist for the purposes of this argument. on this model. in L. . to 'intuit' (or assert. the existence of a non-phrase-structure portion of the grammar. and the 'plausibility ordering' of grammars compatible with given data of the kinds mentioned is likewise 'innate'. according to Chomsky. a hypothesis concerning the innate intellectual equipment of the child'. Presumably. vague at many points. On this scheme. a list of corrections. but is 'innate'. not more 'complicated' in any absolute sense than grammars in the dass.) that a sentence is grammatical is to 'intuit' (or assert.H. such that correct and perspicuous grammars of all human languages can and should be written in that normal form. etc.H. M. Simplifying slightly. the child is supplied with a list of grammatical sentence types and a list of ungrammatical sentence types. in effect.) As examples ofthe similarities that all grammars of the subdass L are thought to possess (above the level of phonetics). which enable him to dassify the input utterances as grammatical or ungrammatical. we may mention the active-passive distinction. Conversely. etc. 2 Given such a highly restricted dass L of grammars (highly restricted in the sense that grammars not in the dass are perfectly conceivable. the general form of grammar is not learned from experience.H. according to Chomsky. But what is an intuition of 'grarnmaticality' an intuition of? According to Chomsky. 'E. verb taking an abstract subject. we may say that. any non-trivial normal form for grammars. containing both grammatical and ungrammatical sentences . He then 'selects' the grammar in L compatible with this information to which his weighting function assigns the highest weight. 550.'. or conjecture. He may be thought of as operating on the following 'inputs' 3: a list of utterances. So muchfor a statement of the l. p. is supposed to play in linguistics. the sort of theory-construction programmatically outlined above is what is needed to give this question the only answer it can have or deserves to have. lf I have left the IR. etc. if such there be). the presence of such major categories as concrete noun. then. A couple of remarks may suffice to give some idea ofthe role that I. Linguistics relies heavily. 'constitutes. I believe that this is no accident-for the I. seems to me to be essentially and irreparably vague-but this much of a statement may serve to indicate what belief it is that I stigmatize as irreparably vague. The project of delimiting the dass L may also be described as the project of defining a nonnal form for grammars. and may weH be employed by nonhuman speakers. according to Chomsky. 530-531. upon 'intuitions' of grammaticality.) that the sentence is generated by the highest-valued Gi in the dass L which is such that it generates 2 3 'Explanatory Models in Linguistics'.

Chomsky believes that the I. M. In Syntactic Structures Chomsky. the I. even if language is wholly learned.H. it is the idea that these abilities and these intuitions are close to the human essen ce. in L. M. is supposed to justify the claim that what the linguist provides is 'a hypothesis about the innate intellectual equipment that a child brings to bear in language learning'. M. whenever it assigns two or more structural descriptions to the same sentencc. • 'E. "E. the fact that a grarnmar which agrees with a sufficient amount of 'input' must be in the cIass . then. 533). in Chomsky's formalism.'. Clearly. 531. one could with equal truth say that a driver's manual 'characterizes the car-driving abilities of the mature driver' and that a calculus text provides 'an explanatory model of the calculus-intuitions of the mathematician'. unfortunateIy. 'Many of these---e. 'E. 530. so to speak. I am grateful to Paul Ziff and Stephen Leeds for calling this point to my attention. p. for Chomsky at least. and to define the weighting function. 533. 9 and that a grammar 'could properly be called an explanatory model of the linguistic intuition of the native speaker'. 530. Apparently. M. M. At most it is told which are del'iant-but it may not be told which are deviant for syntactical and which for semantical reasons. and he conjectures 6 that a child needs semantical information only to 'provide motivation for language learning'.lo However. in L'. in L'. Lastly. 10 'E. in L'.8 Of course. gives this as an objective for linguistic theory: to give an effective procedure for choosing between rival grammars. in L'. In addition to making cIear what it is to be grammatical. 4 I doubt that the c1uld really is told which sentences it hears or utters are ungrammatical. indeed. that gives linguistics its 'sex appeal'. p. n. the aUeged 'ambiguity' in 'the shooting of the elephants was heard'-require coaching to detect. 5 At the same time. p. and not to arrive at theformal grammar of its language.g. 5. The claim that grammar 'explains the ability to recognize ambiguities' thus lacks the impressiveness that Chomsky believes it to have. confronts the linguist with the following tasks: To define the normal form for grammars described above. but fail to correctly 'predict'·7 ambiguities (cf. he Iists no semantical information in the input.132 SYMPOSIUM ON INNATE IDEAS a1l the grammatical sentence types with which we have been supplied by the 'input' and none of the ungrammatical sentence types listed in the 'input'. it must account for certain ambiguities. p. "E.4 Chomsky also says that the Gi which receives the highest value must do more than agree with 'intuitions' of grammaticality.E to be 'seIected' by the child is what mIes out grammars that generate all and only the grammatical sentences of a given natural language.H. . p. for example. 7 A grammar 'predicts' an ambiguity. it is still true that linguistics 'characterizes the linguistic abilities of the mature speaker'.

the agony has somewhat abated' (to a lady who had spilled hot tea on him). 12 and then displayed this ability at a relatively late age to startled adults who had given them up for mutes. which runs. a complicated physical theory. 529. it is said. there is the 'argument' that runs 'what else could account for language learning?' The task is so incredibly complex (analogous to learning. What could we expect to find in their language? If the Martians' brains are not vastly richer than ours in complexity. seems to be all that the normal child requires to develop the competence of the native speaker. 'What else could account for the fact that commodities have different value except the fact that the labour-content is different?') III.l and without any explicit instruction.'.' 11 (b) The fact that reinforcement. CRITICISM OF THE ALLEGED EVIDENCE A. The Irrelevance of Linguistic Universals 1. in essence. IH. 'E. THE SUPPOSED EVIDENCE FOR THE I. (c) Tbe ability to 'develop the competence ofthe native speaker' has been said not to depend on the intelligence level. that it would he miraculous if even one tenth of the human race accomplished it without 'innate' assistance.SYMPOSIUM ON INNATE IDEAS 11.'s 'internalize' the grammar of their native language. Mere exposure to the language. at least implicitly.Q. were: 'Thank you. Even quite low I. Let us assume for the purpose a community of Martians whose 'innate intellectual equipment' may he supposed to he as different from the human as is compatible with their heing able to speak a language at all. 'in any interesting sense'. 11 12 . Since limitations of space make it impossible to describe all of them here. it is said). p. and for a remarkably short period. a few examples will have to suffice.H. Macaulay's jirst words. in L. (a) the ease of the child's originallanguage learning. Not surprising on any theory Let us consider just how surprising the 'Iinguistic universals' cited above really are.H.H. seems to he unnecessary for language learning. then they. 'A young child is able to gain perfect mastery of a language with incomparably greater ease [than an adult-Ho P. Madam. (This is like Marx's 'proof of the Labour Tbeory of Value in Capital. (e) Lastly. (d) Tbe 'Iinguistic universals' mentioned in the previous section are allegedly accounted for by the I. M. vol. Some children have apparently even learned to speak without talking. of course. 133 A number of empirical facts and alleged empirical facts have heen advanced to support the I.

interests. that a particular mighty arbitrary set L of grammars is 'built in' to the brain of both Martians and Humans is not a hypothesis we would have to invoke to account for these basic similarities. would the LH. not less surprising. But (a) the presence of this. But the sentence 'That is IX (X is a lady and I saw you with X last night)' is generated by a phrase-structure rule in the grammar of mathematical logic. The sentence 'John and Jim came horne quickly' is not generated by a phrase-structure rule. but would not this be more. on any reasonable view. But trus too is not surprising. have to be invoked save in the trivial sense that memory capacity. Are the foregoing reflections arguments for or against the LH. is not surprising-why shouId not there be a . then how could the I. Is it really surprising. And again the idiomatic English sentence can be obtained from its phrase-structure counterpart by a simple rule of abbreviation. for example.. are all relevant to language learning. find rote learning difficult. the sentence 'That was the lady I saw you witli last night' is not generated by a phrase-structure rule in English. their language will contain no concrete nouns. If belief in 'innate intellectual equipment' is just that. and all depend in part. on the biological make-up of the organism? If Martians are such strange creatures that they have no interest in physical objects. Chomsky has pointed out that no natural language has a phrase-structure grammar. be false? How could something with no innate intellectual equipment leam anything? To be sure. intelligence. that these operations which break the bounds of phrase-structure grammar appear in every natural language? 13 13 Another example of a transformation is the 'active-passive' transformation (cf. if this means that such parameters as memory span and memory capacity playa crucial role. Again. it will not be surprising if they too have short lists of phonemes in their languages. does it really point to anything more interesting than general intelligence. But what rank Behaviourist is supposed to have ever denied t his? On the other hand. will find it possible to employ a practically infinite set of expressions only if those expressions possess a 'grammar' ---i. needs.e. if they are built up by recursive rules from a limited stock of basic forms. Those basic forms need not be built up out of a short list of phonemes---the Martians might have vastly greater memory capacity than we do--but if Martians. human 'innate intellectual equipment' is relevant to language learning.f.H. or at least not in Chomsky's description of English. than their having an interest in physical objects? (Would it be surprising if Martian contained devices for forming truth-functions and for quantification?) Two more detailed points are relevant here. etc. ? I find it difficult to tell. S)7ltactic Structures). where constitutional factors playa large role for obvious reasons. like humans.134 SYMPOSIUM ON IN NA TE IDEAS like us. and Chomsky's famous 'and-transformation' is just an abbreviation rule. if it is apart of the grammar. But for wh at similarities above the level of phonetics. But the sente'nce 'John came horne quickly and Jim came horne quickly' is generated by a phrase-structure rule in the grammar of mathematical logic. in Chomsky's formalization of English gramm3.

SYMPOSIUM ON INNATE IDEAS 135 Again. and space required for the computation--lead to the same result here.. If it contains noun phrases it contains verb phrases--phrases which when combined with a noun phrase by a suitable construction yield sentences. human beings have a natural interest in one another. it may appear startling at first blush that such categories as noun. quite independently of the detailed structure of the computing machine employed. it contains adverb phrase~­ phrases which when combined with a verb phrase yield a verb phrase. as Curry has pointed out.. Thus the existence of nouns is all that has to be explained. phrases which occupy the environments of nouns. it is too easy to multiply 'facts' here. . verb. and ifthe need for identifying persons rests on something innate-but what Behaviourist would or should be surprised? Human brains are computing systems and subject to some of the constraints that atTect all computing systems. but this is just not so. and the difficulty of always finding adefinite description which will suffice instead. But. The fact is that all the natural measures of complexity of an algorithm---size of the machine table. Perhaps Chomsky would reply that 'simplicity' is subjective here. Once again. (2) The fact that all naturallanguages contain proper names. adverb. If it contains verb phrases. that is. in spite of the effect ofwhat Chomsky calls 'transformations'. and is not inadequate on that account. But why must evcry relation be mirrored by syntax? Every 'speaker' of the canonical languages of mathematical logic is aware that each sentence (X) (FX::O GX) is related to a sentence (X) (o'r::o FX). have 'universal' application. etc. And this reduces to explaining two facts: (I) The fact that all naturallanguages have a large phrase-structure portion in their grammar. If that is 'innateness'. adjective phrases. time. length of computations. If a language contains nouns-that is. But (1) is not surprising in view of the fact that phrasestructure rules are extremely simple algorithms. can be defined in terms ofthe two basic categories 'noun' and 'sentence'. etc. yet the definition of'well formed formula' fails to mirror 'what every speaker knows' in this respect. Similarly. 'innate' factors are relevant to be sure-----if choosing simple algorithms as the basis of the grammar is 'innate'. a phrase-structure category which contains the proper names-it contains noun phrases. It is contended that a gramm ar which 'defines' active and passive forms separately (this can be done by even a phrase-structure grammar) fails to represent something that every spcaker knows. weIl and good! systematic way of expressing the conrerse of a relation ?-and (b) the argument for the existence of such a 'transformation' at all is extremely slim. Is it surprising that algorithms which are 'simplest' for virtually any computing system we can conceive of are also simplest for naturally evolved 'computing systems'? And (2)---the fact that all natural languages contain proper names -----is not surprising in view of the utility of such names. in the sense just illustrated. viz_ that active and passive forms are related.

if Chomsky were right. even if surprising. and normal children of each type should fail to learn the languages spoken by the other types. in fact. The 'Ease' of Langua/?e Learning is not Clear Let us consider somewhat c10sely the 'ease' with which children do learn their native language. Random variation may. and that the existence today of what are called 'unrelated' languages is accounted far by the great lapse of time and by countless historical changes. etc. or quantifiers. there should be two or more types of human beings descended from the two or more original populations. A typical 'mature' college student seriously studying a foreign language spends three hours a week in lectures.. I repeat.H.H. is false. In fourteen weeks of term he is thus exposed to forty-two hours of the language. is not so surprising as to require the I. In any case. since there is only one c1ass I: built into all human brains. indeed. indeed. this hypothesis--a single origin for human language---:is certainly required by the I. proper names.H. But this is not 10 say that a dass I and a funclionfare 'built in'. In four years he may pick up over 300 hours of the language. or even if only some general dispositions in the direction of language using are 'built in'. direct-method teachers estimate that 300 hours of direct-method teaching will enable 14 It is very difficult to account for such phenomena as the sponlaneous babbling of infants without Ihis much 'innateness'. and then spread this through conquest or imitation to the rest of the human population. is true) that language-using is an evolutionary 'leap' that occurred only once. since the human race itse1f is now generally believed to have resulted from a single evolutionary 'leap'. Then. without invoking the /. likely even if the I. is. or common nouns. Linguistic uni versals could be accounted for. By contrast. it is overwhelmingly likely that all human languages are descended from a single originallanguage. and verbs.. very little of which is actuallistening to native informants. we have to conc1ude (if the I.H. alter many things. This is.H. But just this consequence of the I. nouns. lndeed. enough to account for 'linguistic universals' ! For. but much weaker than the I. and only gradually spread from Asia to other continents. then just such highly useful features of the common parent as the presence of some kind of quantifiers. But in that case. Thus.H. as required by the I. 14 it is likely that some one group of humans first developed language as we know it.136 SYMPOSIUM ON INNATE IDEAS 2. and since the human population was extremely small and concentrated for millennia. would be expected to survive. Suppose that language-using human beings evolved independently in two or more places. B. if all human languages are descended from a common parent.H. but that it should fail to strip language of proper names. even if language-using was learned or invented rather than 'built in'. Since we do not observe this. .H. we do know that this is just how alphabetic writing spread.

Indeed. What does this mean? If it means that children do not rnake serious grarnmatical blunders. 'doggie') in the presence of babies.H. What is more serious is what 'grammar' means here. What this 'argument' reduces to is 'W ow! How cornplicated . the evidence is today slirn that any learning requires reinforcement 'in any interesting sense'. Every normal adult learns what every adult leams. but nine or ten years is enough time to becorne pretty darn good at anything. even by the standards of descriptive as opposed to prescriptive grammar. in which rnany adults are also deficient. The child indeed has a better accent. Moreover even if 'reinforcernent' is not necessary.H. or not allowed to play. By nine or ten years of age this has ceased to happen. the child's gramrnatical mistakes. It does not include mastery of vocabulary. learned to play chess by sirnply watching adults play. C. the constant repetition of simple one-ward sentences ('cup'. It means pure1y and simply the ability to leam what every normal adult leams. Reinforcement another Issue As Chornsky is aware. Now the child by the time it is four or five years old has been exposed to I'astly more than 600 hours of direct rnethod instruction. What can equally weil be accounted for by learning theory should not be cited as evidence for the I. Also. N on-geniuses normally do require practice both to speak correctly and to play chess. Capablanca. 300 hours of direct-method teaching and 300 hours of reading-will enable any adult to speak and read a foreign language with ease. and has a huge set of habits to unleam. But it seems to rne that this 'evidence' for the I. nor ability to understand complex constructions.SYMPOSIUM ON INNATE IDEAS 137 one to converse fiuently in a foreign language. It will be objected that the adult does not acquire a perfect accent. This is comparable to Macaulay's achievement in leaming language without speaking. in which even many adults are deficient. most children are consciously and repeatedly reinforced by adults in a host of ways--e. So what? The adult has been speaking one way all of his life. in the second case. arise not from carrying over previous language habits. and to use an incomparably larger vocabulary than a young child. in the first case. with sufficiently prolonged observation.g. Yet probably anyone could leam to speak or to play chess without practice if rnuffied. which are numerous. Certainly 600 hours---say. for example. but from not having fully acquired the first set. any foreign adult living with the child for those years would have an incomparably better grasp of the language than the child does. perhaps (I speak as a parent). this is just not true for the young child. stands the facts on their head. D. Independence of Intelligence Level an Artifact Every child leams to speak the native language.

. A not so clever person can get a 'near-solution'two pegs Ieft--in the same time. But the I. Let a complete seventeenth-century Oxford Unil'ersity education be innate if you like. is utterly unfounded. to my knowledge. perhaps we should be impressed. still the solution to 'jurnp' was not innate. But that is no help. aIl the holes but the centre one are filled. cannot on any theory be indil'idually 'innate'. if true. A clever person can get the solution in perhaps eight or ten hours of experimentation. automobile driving. I will be told. of course. The theorems of mathematics. cannot. by its very nature. i. But what of the 'What Else?' argument? Just how impressed should we be by the failure of current learning theories to account for complex learning processes such as those involved in the learning of language? If Innateness were a general solution. No programme exists. the solutions to puzzles.' Like the preceding argument. In conversation Chomsky has repeatedly used precisely such skiIls as these to support the idea that humans have an 'innate conceptuaI space'. Consider the following puzzle (calIed 'jump'): * * * * * * * * * * * * * * * * * * * To begin with. etc. the assertion that such strategies (which absolutely must exist and be employed by all humans) cannot account for this or that leaming process. even though the machine can spend the equivaIent of many human lifetimes in experiment at ion. etc. it reduces to the 'What Else' argument.. that would enable a computer to solve even the 'near solution' problem without running out of both time and space.. WeIl and good. What else could it be but innate. In the absence of any knowledge of what general multipurpose learning strategies might even look like. The object of the game is to remove all the pegs but one by 'jumping' (as in checkers) and to end with the one remaining peg in the centre. that the answer or an answer schema must be 'innate'.H. When we come to the discovery of even the simplest mathematical theorem the situation is even more striking. what must be 'innate' are heuristics. the Prime Number Theorem was not innate. and everyone in fact can solve many problems that no computer can solve. and that not everyone solves puzzles or proves theorems. But everyone does learn pattern recognition.e. be generaIized to handle aIl complex learning processes.138 SYMPOSIUM ON INNATE IDEAS a skill every normal adult learns. that el'eryone learns his native language (as weIl as everyone does). learning strategies.

Until we understand the strategies which make general learning possibleand vague talk of 'classes of hypotheses' and 'weighting functions' is utterly useless here-no discussion of the limits of learning can even begin. Invoking 'Innateness' only postpones the problem o/leaming. . it does not solve it.SYMPOSIUM ON INNATE IDEAS 139 and so on.

have certain remarkable properties in common. JASON: After some centuries. But does this call for elaborate explanation? J: The properties in question are more interesting properties of grammatical form and of meaning. A: I can imagine having a good deal of trouble mastering a language with an alphabet of a million letters and no word less than a million letters long. A: What facts? J: In the first place. but we do not take them as posing a problem. Onee one language is available and can be used for giving explanation and instruction. This seems to me not only remarkable but incredible. A: That answers my objection. and enthroned as the only adequate explanation for some striking facts conceming human linguistic proficiency. for the human mind strikes me as agile enough to leam. the theory of Innate Ideas has been disinterred. almost any transformation or distortion of an already familiar language. A: But is it remarkable that the elements of any collection have some remarkable properties in common? Surely we can find throughout the random deals of a pack of cards during an evening some very special uniformities. I suppose these nomads have constructed languages lacking the properties in question.) ANTI C U s: Tell me about the resurrection.(c) THE EPISTEMOLOGICAL ARGUMENT NELSON GOODMAN (Jason has brought back from the nomads of Outer Cantabridgia something that Anticus suspects is more fleece than golden. that all natural languages. J: The claim is of course much stronger: that any language a human being can acquire has the properties in question. however diverse in origin and in superficials. and found that eamest efforts to teach them to human beings fail. A: Then the claim is indeed material and testable. the limitations are transcended. but now I am puzzled as to how they propose to . J: I have done them an injustice. with appropriate instruction and explanation. They hold only that no language lacking the properties in question can be acquired by a human being as an initiallanguage.

but I am not moved to try in this case. What you have said does not convince me that the claim ought to be rejected. One case that might carry some weight with you. Can they really take an infant at birth. but I know you would say that each of these has been tailored to fit the known natural languages and derives rather from the conceptual apparatus we impose upon these languages than from any remarkable affinities among them. the claim is that certain statements about the properties of languages that can be initially acquired are plausible. A: Though. he has had an abundance of practice in developing and using rudimentary prelinguistic symbolic . isolate it from all the influences of our language-bound culture. The claim is that while a user of ordinary English might be taught to use Grubleen. Still. we ought rather to help examine and deve10p it. But I have another worry. J: Nevertheless. They regard their claim as a hypothesis not subject to such direct experimental test. and a certain explanation illuminating. though. no human being could acquire Grubleen as an initial language. Don't you think. that before anyone acquires a language. experimental support for this can hardly be expected. important conjectures often cannot in the early stages be either precisely stated or actually tested. So far we seem to have concluded first that the claim we are discussing cannot be experimentally tested even when we have an acknowledged example of a 'bad' language. But so far we have been speaking vaguely of 'certain properties' or 'the properties in question'. and attempt to inculcate it with one of the 'bad' artificial languages? J: No. They cite some general grammatical properties now and then. A: You are right in principle. concerns a concocted language called Grubleen. J: Such as? A: What we call a language is a fairly elaborate and sophisticated symbolic system. It differs from ordinary English only in that it contains the predicates 'grue' (for 'examined before t and green or not so examined and blue') and 'bleen' (for 'examined before t and blue or not so examined and green') instead of the predicates 'green' and 'blue'. what in general is the difference between Grubleen-like and English-like languages? I see by your gesture that you are painfully aware of the difficulties of answering that question. as you say. If it is suggestive or promising enough. Let us assurne that we now have before us an example of a language that cannot be so acquired. and second that the claim has not even been formulated to the extent of citation of a single general property of 'bad' languages. J: My informants are not always very explicit about this. A: Very weil. we must surely have a clearer formulation or illustrations of what these properties are. but supported by ancillary considerations and evidence. Jason.SYMPOSIUM ON INNATE IDEAS 141 examine the claim experimentally. They readily admit this cannot be done. If we are to judge plausibility. since the claim seems to me discredited by antecedent considerations.

142 SYMPOSIUM ON INNATE IDEAS systems in which gest ures and sensory and perceptual occurrences of all sorts function as signs? J: Yes. that initiallanguage acquisition is astonishingly fast. and obviously we cannot argue back from uniformity of language to uniformity of preiinguistic system.you say undermines also a second claim. but surely you do not call those rudimentary systems languages. but language-acquisition is what is at issue. A: Yes. If the language were the first symbolic system acquired. All sophistry aside. J: True. but I submit that our facility in going from one symbolic system to another is not much atTected by whether each or either or neither is called a language. On the other hand. Rather than facts crying for a theory. since once some language is available. that the real issue is over initial acquisition of languages. A: We'd certainly have an even harder time doing it. A: You remember. And experimentation under all these difficulties would have to begin with symbol-using from the moment of birth. is . then it must begin virtually at birth and takes a long time. if the process of acquiring the first language is thought of as beginning with the first use of symbols. and that as we find no interesting limitations upon wh at we can acquire as a secondary language. In other words. J: Your objections are more telling against my inadequate presentation than against the spirit and substance of what I am trying to present. we have no ground for believing that there are such limitations upon what we can acquire as a secondary symbolic system. A: No. J: Does not all this just move the question back from the nature of languages that can be initially acquired to the nature of symbolic systems that can be so acquired? I suspect we would find remarkable uniformities and astonishing speed of acquisition here. that acquisition of an initial language is acquisition of a secondary symbolic system. We'd have to examine symbols that are not overt and articulate but rather inaccessible and ill-defined. acquisition of others is relativeiy easy. And since the prelingui~tic systems are likely to be fragmentary as weil as rudimentary. when initiallanguage acquisition is seen as secondary-symbolic-system acquisition. we'd have trouble deciding when a system is acquired. I suppose we rnight manage to work up some astonishment. and the process of acquisition considered to begin with the first overt use of words. though. Little of the unimpressive evidence adduced with respect to languages would be pertinent here. But I hardly have to refute your suspicions. J: I am afraid that what . the claim that there are rigid limitations upon initial-language acquisition is deprived of plausibility by the fact that there are no such limitations upon secondary-language acquisition. the theory is crying for the facts. But if acquisition of the first language is mereiy passage from a symbolic system already acquired to another that we are taught. that is a much easier step.

adjust. I may have no direct access to them at all. obey still more special laws that require reference to innate ideas. J: Roughly that. no matter where dropped. transform . and moreover to fall. The mind does not merely kick when tapped. and capable of being brought into full consciousness? J: Not even that. Of course. InabiJity to explain a fact does not condemn me to accept an intrinsically repugnant and incomprehensible theory. constancy amid instability. the seeking and finding. J: And for this remarkable fact we do need a theory---the theory of gravitation. it gropes. And human beings. J: You Berkeleyans always overstress the groping. they are in consciousness? J: No. in their cognitive behaviour. Now I gather that the theory here proposed is that certain ideas are implanted in the mind as original equipment. and now name-calling! But seriously. A: And you Leibnizians overstress the predetermination. seem to me more characteristic than any mere programme-reading. Are you saying that human cognition is explicable only by supposing that the mind is supplied at the start with the interpretation of certain symbols? If that means only that it responds in a fixed way to certain stimuli. not necessarily. A: And being ideas. this suggests a view of mind we would both reject. My only way of . its ability to adapt. A: A set of laws subsuming this behaviour under a very general description. operating upon cognitive processes. I leamed instantly to fall when dropped. Let us now assume that for certain remarkable facts I have no alternative explanation. A: Then they are in the subconscious mind. J: We go from peuifogging analysis to loose metaphor. The groping and grasping. J: For example? A: Well. What seems to me notable is not the fixity but rather the flexibility of the mind. not even usually. that alone does not dictate acceptance of whatever theory may be offered. for that theory might be worse than none. Let us take it more slowly. its way of achieving unity in variety. A: Your speed there is remarkable enough. of inventing rather than obeying. I think it is just those capacities of the mind that you praise that can be accounted for only by the instrumentation of innate ideas. common to animate and inanimate objects alike. A: We have been paying much less attention to what the theory is than to what it is supposed to explain. Living things obey more special laws framed in terms of other notions. precisely toward the centre of the earth. but I am not inclined to attribute knowledge of these laws to the falling objects. J: But this is mechanical behaviour.SYMPOSIUM ON INNATE IDEAS 143 there nothing in human behaviour you find striking enough to demand special explanation? A: I can think of some remarkable behavioural facts that call for no such explanation as a theory of innate ideas.

A: Then I am puzzled. but I bow before the subtlety of that last statement. in anything like its normal use. Once we apply the term. or pictures. it is the theory that counts. what is advocated is the rather trivial truth that the mind has certain capacities. A: But I thought the ideas were posited to explain the capacities. it becomes false or meaningless. You seem to be saying that these innate ideas are neither innate nor ideas. A: A few minutes aga you accused me of sophistry. or natural potentialities'.. A: In that case. images. but unfortunately. habits. limitations.144 SYMPOSIUM ON INNATE IDEAS discovering them in my own mind may be by the same methods that someone else might use to infer that I have them. reason: that until the term 'innate idea' is applied. dispositions. if you like. If aB that is claimed is that the mind has certain inclinations and capacities. I probably should have said that these ideas exist as or are 'innate as' such capacities. Jason. the thesis becomes far from obvious. John Locke made aB this acutely clear. do these people go on using it? F or a very compelling. Go again. we can talk again about unsubstantiated conjectures that cry for explanation by implausible and untestable hypotheses that hypostatize ideas that are innate in the mind as non-ideas. etc. not the term 'innate idea'. after admitting the term is controversial and claiming it is unnecessary. Then. . but rather 'inclinations. formulae. J: What is innate are not concepts. J: Again I am afraid I have not been careful enough. why all the effort at historical justification? And why. Rather than identify the innate ideas with capacities. how can you justify caBing these ideas? J: The justification is historical. Descartes and Leibniz used the term 'innate idea' in just this sense. or I to infer that he does. tendencies. But after all. and bring back to me aB the mysteries of ideas being innate as capacities. but not very good.

Individuals (1959). He is the editor of Philosophical Logic in the present series. A. and his book Speech Acts was published in 1969. formerly a FeUow of St. of the Massachusetts Institute ofTechnology. Harvard University. and Forecast (1955). Current lssues in Linguistic Theory (1964). He edited. Among his publications are Syntactic Structures (1957). H. SEARLE (the editor of the present volume) is a member ofthe Department ofPhilosophy of the University of Califomia at BerkeIey. Fodor. Oxford. was John Locke Lecturer in Oxford in 1962. KATZ. NELSON GOODMAN. HILARY PUTNAM. The last -named is a version. has written many valuable papers on the philosophy of science and mathematics. Fiction.NOTES ON THE CONTRIBUTORS J. STRAWSON is Waynflete Professor of Metaphysical Philosophy at Oxford. . of Brandeis University. is the author of The Problem o/Induction and its Solution (1962). J. JERROLD J. R. He was William James Lecturer at Harvard in 1966. F. not fuUy completed by Ausrin. he is the author of The Structure 0/ Appearance (J 951). The Philosophy 0/ Language (J 966). with J. GRICE. NOAM CHOMSKY. with Paul Benacerraf. of the Massachusetts Institute ofTechnology. is now a member of the Department of Philosophy at Berkeley. The Structure 0/ Language (1964). and Languages 0/ Art (1968). as weil as on linguistic theory. John's College. He edited. and Cartesian Linguistics (1966). and The Bounds 0/ Sense (J 966). Philosophy 0/ Mathematics: Selected Readings (1964). of the Department of Philosophy. Besides many important papers. of his William James Lectures at Harvard in 1955. Aspects 0/ the Theory 0/ Syntax (1965). He has contributed many valuable articIes to philosophical periodicals. Among his publications are An Introduction to Logical Theory (1952). Sense and Sensi bilia in 1962. His Philosophical Papers were published in 1961. He was John Locke Lecturer in Oxford in 1969. L. and of many articIes. AUSTlN was White's Professor of Moral Philosophy at Oxford from 1953 until his early death in 1960. P. and How To Do Things with Words also in 1962. Fact. is perhaps the most influential of contemporary writers on linguistic theory. P.

'Mr.BIBLIOGRAPHY (not including material in this volume) I. P. (2) Some recent books ALSTON. GEACH. T. P. New Jersey. Mind (1950). London. R. J. P. Mind (1957). 1953).. S. AUSTIN. 1959). F.: 'Referenee and definite descriptions'. New York. 1960). RUSSELL. 1956). ed. Flew (Maernillan. Feigl and W. Oxford. J. London and New York. B. ed. J. H. V. F. 1949). Sellars (Appleton-Century-Crofts Ine. Blaek (BlackweU. C. Ithaea. R. J. London. STRAWSON.: An inquiry into meaning and truth (Allen & Unwin.: Individuals (Methuen. B. KATZ.: Speech acts (Cambridge University Press. London.: The philosophy o/language (Prentice Hall. T. A. Marsh (AUen & Unwin. P. R USSELL. Geaeh and M. 1969). BOOKS (I) Modern classics in the philosphy o/language FREGE. G. reprinted in RusselI. American philosophical quarterly (1965). 1966). 'Identifying reference and truth values'.: Lectures on the philosophy o/logical atomism. G. 1952).: Tractatus logico-philosophicus. L.: 'Performative utteranees'. ARTICLES ON SELECTED TOPICS (I) Re/erence and the theory 0/ descriptions DONELLAN. F. 11. ZIFF. Cambridge. Oxford. 1962). Oxford. 1956). C HOMSKY. . Press. New York and London. trans. Blaek (Blaekwell. D. W. Phi/osophical review (1966). 1961). N. 1962). f. K. Philosophical papers (Clarendon Press.: Reference and generality (Cornell University Press. Mass. McGuinness (Routledge & Kegan Paul. L. Strawson on referring'. G. T. FREGE. Geaeh and M.: 'On Denoting'. trans.Oxford. L. London.: 'Linguistie aets'.: Philosophical writings. 0. London.: How to do things with words (Clarendon Press.: Aspects 0/ the theory 0/ syntax (M. 1964). R USSELL. P. Philosophical writings. WITIGENSTEIN.: The phi/osophy o/language (Harper & Row. P. B. Ithaea.. (2) Speech acts and propositions ALSTON. Logic and knowledge. 1961). reprinted Readings in philosophical analysis. Philosophical im'estigations (Blackwell.: 'Sense and referenee'. Theoria (1964). AUSTIN. ed.: Semantic analysis (CorneU University Press. J. W. STRAWSON. QUINE.: Word and object (Teehnology Press and John Wiley & Son. 1962). New York. trans.I. W. Pears and B. N. 1965). 1948). Oxford. Mind (1905). reprinted Essays in conceptual analysis. 1960).: 'On referring'. SEARLE.T.

1962). London. R. London. 1969. 1964).: 'Propositions'. H. 1964).: 'Meaning'. S. Analytical philosophy. J. G. (5) Philosophical aspects 0/ generative grammar BOYD. (1950). N. J.: 'The semantic conception of truth'. Fodor (Prentice Hall.. The struclure 0/ language. reprinted Philosophicallogic.: 'Propositions'. V. Pitcher (Prentice Hall. P. KATZ. W. (6) Analytic propositions BENNETT.: 'Truth and meaning'.B1BLlOGRAPHY 147 CARTWRIGHT. ed. G. TEICHMAN. STRAWSON. Philosophical quarterly(1965). reprinted H. ed. M. Strawson (Oxford University Press. Parkinson (Oxford University Press. C. 1967). London. ed. (3) Meaning CAVELL.: '00 iIlocutionary forces exist?'. Philosophical review (1968).: 'Austin on locutionary and iUocutionary acts'.. F. B. Oxford. (1950). vol. C. Proceedings 0/ the Aristotelian Society (1958-9). Philosophy and phenomelogical research (1944). British philosophy in mid-century. Proceedings 0/ the Aristotelian Society. SHWAYDER. ed. Journal 0/ linguistic studies.: 'Analyticity and contradiction in natural languages'. The journal 0/ philosophy (1967). A.. J. Englewood Cliffs. The Hague. Montefiore (Routledge & Kegan Paul. 1964).: 'Uses of language and uses of words'. 1967). L: 'Truth'. D. T ARSKI. SEARLE. N. R.J. A. J. P.. N. N. New York.: 'The semantics of modal verbs'. Supp. N. Philosophical rel'iew (1965). London. 1968). Katz and J. 1966). Englewood Cliffs. KIPARSKY. reprinted Philosophical logic. Pitcher (Prentice Hall. R. A. Philosophical review (1961).: 'Analytic-synthetic'. Theoria (1960). (4) Truth AUSTIN. J. Englewood Cliffs.. Englewood Cliffs. GRlCE. Mace (Macmillan. Recent advances in Linguistics. Bierwisch and Heidolph (Mouton. British analytical philosophy. COHEN. N. reprinted Ordinary language. LEMMON. J. c. F.: 'Must we mean what we say?'. J. P. ed. G. G. Philosophical quarterly (1964). 1949). A. Synthese (1967). J.: 'Truth'. Strawson (Oxford University Press. STAMPE. CHOMSKY. Chappell (Prentice Hall. J. New York.: 'Sentences. F. ed.: 'Assertion'.. reprinted The theory 0/ meaning. P. ed.: 'Toward a grammar of meaning'. Proceedings 0/ the Aristotelian Society (1958-9). W. Katz and J. Sellars. L J. J.J. A. 1964). J..J.: 'Fact'. The structure 0/ language. Inquiry (1958). J. Readings in philosophical analysis (Appleton-Century-Crofts Inc. ed. reprinted in Truth.: 'The theory of meaning'. D. ed. P. 1964). Philosophical review (1968). HARMAN. D. J. statements and propositions'.: 'Truth'. Butler (Blackwell. Fodor (Prentice Hall. O. R. eds. Williams and A. Supp. W. DAVIDSON. ed.: 'Current issues in linguistic theory'. GEACH. 1969). J. Philosophical rel'iew (1957). and KIPARSKY. E. G. P. H. DUMMETT. Englewood Cliffs. P. ed. reprinted Truth. J.J. Proceedings 0/ the Aristotelian Society. Feigl and W. 'Truth: a reconsideration of Austin's views'. and THORNE. R YLE. . 1957). vol.: 'Psychological aspects of the theory of syntax'.

: 'The analytic and the synthetic'. Minnesota studies in the philosophy 0/ science. 1961). vol. M. 'Two dogmas of empiricism'. STRAWSON.: 'Metaphor'. W. P.: 'On describing'. and GRlCE. P. New York.: 'Subject and predicate'. Minneapolis. F. (7) Miscellaneous BLACK. Mind (1950). Mind (1952). . Models and metaphors (Cornell University Press.: 'In defense of a dogma'. H. T. III (University of Minn. Philosophical quarterly (1957).. S. Cambridge. 1953). 1962). QUINE.148 BIBLIOGRAPHY PUTNAM. F. Philosophical review (1956). 0. H. TOULMIN. V. From a logical point 0/ view (Harvard University Press. K.. GEACH. Press. and BAIER. P.: 'Propositions. concepts. P. and logical truth'. Mass. STRAWSON.

1. P.. B.. Fillmore. 102n. M. W. 106... 43 Harris. 51 Halle. 8-10. R. M. 129m. 86. Rosenbaum.. I IOn... H.... I IOn. 105n.. P. R.. Nagel.. I.. 95n... 73n. 43 Reichling. S. 106 Polanyi. 84n. 39 Black. 132n.. 105n.. 86-8. Plato.... 0. H. 95. 129n. 129n. J. G. 73n. 72n.. 90.1. 85n. 72.12. 129n. S. 43 Locke. 7. 74 Kant. Goodman. 87 .103. H. 41n.. 96 Searle. R. 11. C. 78 Ilouseholder. 28-30. 71. 35. 71. 97. J. S. 102n. I. 30-2. I IOn. 75. 6-8. 23-7.. 144 Luce.. 87.. 110n. 12... Sheffer. A.INDEX OF NAMES (not including authors mentioned only in the Bibliography) Aristotle.. 8n. II On.. Tarski. Frege.114. 129n. 114. 80n. 35-7. 87. 33n. 80.. 75.. D.. 72n.. R. 46. IlOn. 91. 10-12.. 3n. R. S. Rawls. L... 96 McGuinness. M.. 106. H. 37... 30n.. RusselI. 90-3. 3. 108n. 108 Ziff. F. 71.. 106. P. 12. P. 4. H. I I. 83.. 87 Hare... M. 85n. C. 128-32. V... 71-3... R. 91n. 134-8 Curry. C. R. 75. H. 72n. 135 Descartes. 108n. 3--6. MiII.. P. 72 Martinet.. A. J. Illn. 129n. E. 78. 44. N. F. 108 Ryle. A. 89-91. 128. R. 87 Lunt.. 80n. Klima. 82.. 1. R. 72 MacLean. 144 Fant. Bush. Matthews.. 115n. 1.. A. 82.98-100. 73n.... . G.. E. 8.... 95n. 11. 129n. J. Leibniz. F. P. 1 Miliar. Lees. F. G. 12 Quine. I. 97. 0. 78-80. M.43 Galanter. W. J. Chomsky. 108.. II. 108. Putnam.. StockweIl. 73n.. 106 Jakobson. 33. Z. N. 90 Lccds.. 83 Hume. B. 71. C. I1I n.. Wittgenstein... L... 86-91. A. 144 Lewis. 92 Fodor. 80. 80. 105n. 72.. S. 87 Ferguson.. H. Katz. 80n. Il3n.. L. 46. G. II5n. Lukoff.. 11 Suppes. E. H. 2-7.Icspersen.94-6. F. 115n. I I Postal. 72n. 41n.11. C. 88 Hart.. Hockett. 8. 96 Strawson. 109. 83. Il3n.. 43 Smith. 11. 89-93. J. 72n... 97. A... 3. 7-9. Carnap. 12 Grice. 132n. Reichenbach. 42n. J. I. I 15 Austin... G. P. 108 Schachter...