8
Truth and Meaning
It is conceded by most philosophers of language, and recently by some linguists, that a satisfactory theory of meaning must give an account of how the meanings of sentences depend upon the meanings of words. Unless such an account could be supplied for a particular language, it is argued, there would be no explaining the fact that we can learn the language: no explaining the fact that, on mastering a finite vocabulary and a finitely stated set of rules, we are prepared to produce and to understand any of a potential infinitude of sentences. I do not dispute these vague claims, in which I sense more than a kernel of truth.1 Instead I want to ask what it is for a theory to give an account of the kind adumbrated.
One proposal is to begin by assigning some entity as meaning to each word (or other significant syntactical feature) of the sentence; thus we might assign Theaetetus to 'Theaetetus' and the property of flying to 'flies' in the sentence 'Theaetetus flies'. The problem then arises how the meaning of the sentence is generated from these meanings. Viewing concatenation as a significant piece of syntax, we may assign to it the relation of participating in or instantiating; however, it is obvious that we have here the start of an infinite regress. Frege sought to avoid the regress by saying that the entities corresponding to predicates (for example) are 'unsaturated' or 'incomplete' in contrast to the entities that correspond to names, but this doctrine seems to label a difficulty rather than solve it.
The point will emerge if we think for a moment of complex singular terms, to which Frege's theory applies along with sentences. Consider the expression 'the rather of Annette'; how does the meaning of the whole depend on the meaning of the parts? The answer would seem to be that the meaning of'the father of is such that when this expression is prefixed to a singular term the result refers to the rather of the person to whom the singular term refers. What part is played, in this account, by the unsaturated or incomplete entity for which 'the father of stands? All we can think to say is that this entity 'yields' or 'gives' the father of at as value when the argument is x, or perhaps that this entity maps people on to their fathers. It may not be clear whether the entity for which 'the father of is said to stand performs any genuine explanatory function as long as we stick to individual expressions; so think instead of the infinite class of expressions formed by writing
1 See Donald Davidson (1965). Theories of Meaning and Learnable Languages'.
156
The Essential Davidson
'the father of zero or more times in front of 'Annette'. It is easy to supply a theory that tells, for an arbitrary one of these singular terms, what it refers to: if the term is 'Annette' it refers to Annette, while if the term is complex, consisting of 'the father of prefixed to a singular term t, then it refers to the father of the person to whom ŕ refers. It is obvious that no entity corresponding to 'the father of is, or needs to be, mentioned in stadng this theory.
It would be inappropriate to complain that this litde theory uses the words 'the father of in giving the reference of expressions containing those words. For the task was to give the meaning of all expressions in a certain infinite set on the basis of the meaning of the parts; it was not in the bargain also to give the meanings of the atomic parts. On the other hand, it is now evident that a satisfactory theory of the meanings of complex expressions may not require entities as meanings of all the parts. It behoves us then to rephrase our demand on a satisfactory theory of meaning so as not to suggest that individual words must have meanings at all, in any sense that transcends the fact that they have a systematic effect on the meanings of the sentences in which they occur. Actually, for the case at hand we can do better still in stating the criterion of success: what we wanted, and what we got, is a theory that entails every sentence of the form lt refers to x' where V is replaced by a structural description2 of a singular term, and V is replaced by that term itself. Further, our theory accomplishes this without appeal to any semantical concepts beyond the basic 'refers to'. Finally, the theory clearly suggests an effective procedure for determining, for any singular term in its universe, what that term refers to.
A theory with such evident merits deserves wider application. The device proposed by Frege to this end has a brilliant simplicity: count predicates as a special case of functional expressions, and sentences as a special case of complex singular terms. Now, however, a difficulty looms if we want to continue in our present (implicit) course of identifying the meaning of a singular term with its reference. The difficulty follows upon making two reasonable assumptions: that logically equivalent singular terms have die same reference, and that a singular term does not change its reference if a contained singular term is replaced by another with the same reference. But now suppose that'/?' and 'S' abbreviate any two sentences alike in truth value. "Then the following four sentences have the same reference:
(1)   R
(2)  x(x =x.R)~ x(x = x)
(3)  x(x = x. S) = x(x = x)
(4)   S
2 A 'structural description' of an expression describes the expression as a concatenation of elements drawn from a fixed finite list (for example of words or letters).
Truth and Meaning                                      157
For (1) and (2) are logically equivalent, as are (3) and (4), while (3) differs from (2) only in containing the singular term 'x(x = x.S)' where (2) contains ix(x=x, RY and these refer to the same thing if S and R are alike in truth value. Hence any two sentences have the same reference if they have the same truth value.3 And if the meaning of a sentence is what it refers to, all sentences alike in truth value must be synonymous—an intolerable result.
Apparendy we must abandon the present approach as leading to a theory of meaning. This is the natural point at which to turn for help to the distinction between meaning and reference. The trouble, we are told, is that questions of reference are, in general, setded by extra-linguistic facts, questions of meaning not, and the facts can conflate the references of expressions that are not synonymous. If we want a theory that gives the meaning (as distinct from reference) of each sentence, we must start with the meaning (as distinct from reference) of the parts.
Up to here we have been following in Frege's footsteps; thanks to him, the path is well known and even well worn. But now, I would like to suggest, we have reached an impasse: the switch from reference to meaning leads to no useful account of how the meanings of sentences depend upon the meanings of the words (or other structural features) that compose them. Ask, for example, for the meaning of 'Theaetetus flies'. A Fregean answer might go something like this: given the meaning of 'Theaetetus' as argument, the meaning of 'flies' yields the meaning of 'Theaetetus flies' as value. The vacuity of this answer is obvious. We wanted to know what the meaning of 'Theaetetus flies' is; it is no progress to be told that it is the meaning of 'Theaetetus flies'. This much we knew before any theory was in sight. In the bogus account just given, talk of the structure of the sentence and of the meanings of words was idle, for it played no role in producing the given description of the meaning of the sentence.
The contrast here between a real and pretended account will be plainer still if we ask for a theory, analogous to the miniature theory of reference of singular terms just sketched, but different in dealing with meanings in place of references. What analogy demands is a theory that has as consequences all sentences of the form 's means ní where V is replaced by a structural description of a sentence and lrri is replaced by a singular term that refers to the meaning ofthat sentence; a theory, moreover, that provides an effective method for arriving at the meaning of an arbitrary sentence structurally described. Clearly some more articulate way of referring to meanings than any we have seen is essential if these criteria are to be met.4 Meanings as entities, or the related concept of synonymy, allow us to
3  The argument derives from Frege. See A. Church, Introduction to Mathematical Logic, 24—5. It is perhaps worth mentioning that the argument does not depend on any particular identification of the entities to which sentences are supposed to refer.
4  It may be thought that Church, in 'A Formulation of die Logic of Sense and Denotation', has given a theory of meaning that makes essential use of meanings as cntides. But this is not the case: Church's logics of sense and denotation are interpreted as being about meanings, but they do not
15 8                                     The Essential Davidson
formulate the following rule relating sentences and their parts: sentences are synonymous whose corresponding parts are synonymous ('corresponding' here needs spelling out of course). And meanings as entities may, in theories such as Frege's, do duty, on occasion, as references, thus losing their status as entities distinct from references. Paradoxically, the one thing meanings do not seem to do is oil the wheels of a theory of meaning—at least as long as we require of such a theory that it non-trivially give the meaning of every sentence in the language. My objection to meanings in the theory of meaning is not that they are abstract or that their identity conditions are obscure, but that they have no demonstrated use.
This is the place to scotch another hopeful thought. Suppose we have a satisfactory theory of syntax for our language, consisting of an effective method of telling, for an arbitrary expression, whether or not it is independendy meaningful (i.e., a sentence), and assume as usual that this involves viewing each sentence as composed, in allowable ways, out of elements drawn from a fixed finite stock of atomic syntactical elements (roughly, words). The hopeful thought is that syntax, so conceived, will yield semantics when a dictionary giving the meaning of each syntactic atom is added. Hopes will be dashed, however, if semantics is to comprise a theory of meaning in our sense, for knowledge of the structural characteristics that make for meaningfulness in a sentence, plus knowledge of the meanings of the ultimate parts, does not add up to knowledge of what a sentence means. The point is easily illustrated by belief sentences. Their syntax is relatively unproblematic. Yet, adding a dictionary does not touch the standard semantic problem, which is that we cannot account for even as much as the truth conditions of such sentences on the basis of what we know of the meanings of the words in them. The situation is not radically altered by refining the dictionary to indicate which meaning or meanings an ambiguous expression bears in each of its possible contexts; the problem of belief sentences persists after ambiguities are resolved.
The fact that recursive syntax with dictionary added is not necessarily recursive semantics has been obscured in some recent writing on linguistics by the intrusion of semantic criteria into the discussion of purportedly syntactic theories. The matter would boil down to a harmless difference over terminology if the semantic criteria were clear; but they are not. While there is agreement that it is the central task of semantics to give the semantic interpretation (the meaning) of every sentence in the language, nowhere in the linguistic literature will one find, so far as I know, a straightforward account of how a theory performs this task, or how to tell when it has been accomplished. The contrast with syntax is striking. The main job of a modest syntax is to characterize meamngßtlness (or sentencehood). We may have as much confidence in the correctness of such a
mention expressions and so cannot of course be theories of meaning in the sense now under discussion.
Truth and Meaning
159
characterization as we have in the representativeness of our sample and our ability to say when particular expressions are meaningful (sentences). What dear and analogous task and test exist for semantics?5
We decided a while back not to assume that parts of sentences have meanings except in the ontologically neutral sense of making a systematic contribution to the meaning of the sentences in which they occur. Since postulating meanings has netted nothing, let us return to that insight. One direction in which it points is a certain holistic view of meaning. If sentences depend for their meaning on their structure, and we understand the meaning of each item in the structure only as an abstraction from the totality of sentences in which it features, then we can give the meaning of any sentence (or word) only by giving the meaning of every sentence (and word) in the language. Frege said that only in the context of a sentence does a word have meaning; in the same vein he might have added that only in the context of the language does a sentence (and therefore a word) have meaning.
This degree of holism was already implicit in the suggestion that an adequate theory of meaning must entail all sentences of the form ls means m. But now, having found no more help in meanings of sentences than in meanings of words, let us ask whether we can get rid of the troublesome singular terms supposed to replace 'ni and to refer to meanings. In a way, nothing could be easier: just write cs means that/»', and imagine '/>' replaced by a sentence. Sentences, as we have seen, cannot name meanings, and sentences with 'that' prefixed are not names at all, unless we decide so. It looks as though we are in trouble on another count, however, for it is reasonable to expect that in wrestling with the logic of the apparendy non-extensional 'means that' we will encounter problems as hard as, or perhaps identical with, the problems our theory is out to solve.
The only way I know to deal with this difficulty is simple, and radical. Anxiety that we are enmeshed in the intensional springs from using the words 'means that' as filling between description of sentence and sentence, but it may be that the success of our venture depends not on the filling but on what it fills. The theory will have done its work if it provides, for every sentence s in the language under study, a matching sentence (to replace '/>') that, in some way yet to be made clear, 'gives the meaning' of s. One obvious candidate for matching sentence is just s itself, if the object language is contained in the metalanguage; otherwise a translation of s in the metalanguage. As a final bold step, let us try treating the position occupied by '/>' extensionally: to implement this, sweep away the obscure 'means that', provide the sentence that replaces '/»' with a
5 For a recent statement of the role of semandcs in linguisdcs, see Noam Chomsky, 'Topics in the Theory of Generative Grammar'. In this article, Chomsky (1) emphasizes the central importance of semandcs in linguistic theory, (2) argues for the superiority of transformational grammars over phrase-structure grammars largely on the grounds that, although phrase-structure grammars may be adequate to define sentencehood for (at least) some natural languages, they are inadequate as a foundation for semantics, and (3) comments repeatedly on the 'rather primitive state' of the concepts of semantics and remarks that the notion of semantic interpretation 'still resists any deep analysis'.
160
The Essential Davidson
proper sentential connective, and supply the description that replaces V with its own predicate. The plausible result is
(T) s is T if and only if p.
What we require of a theory of meaning for a language L is that without appeal to any (further) semantical notions it place enough restrictions on the predicate 'is 7" to entail all sentences got from schema T when V is replaced by a structural description of a sentence of L and 'p' by that sentence.
Any two predicates satisfying this condition have the same extension,6 so if the metalanguage is rich enough, nothing stands in the way of putting what I am calling a theory of meaning into the form of an explicit definition of a predicate 'is T\ But whether expliddy defined or recursively characterized, it is clear that the sentences to which the predicate 'is 7" applies will be just the true sentences of L, for the condition we have placed on satisfactory theories of meaning is in essence Tarski's Convention T that tests the adequacy of a formal semantical definition of truth.7
The path to this point has been tortuous, but the conclusion may be stated simply: a theory of meaning for a language L shows 'how the meanings of sentences depend upon the meanings of words' if it contains a (recursive) definition of truth-in-L. And, so far at least, we have no other idea how to turn the trick. It is worth emphasizing that the concept of truth played no ostensible role in stating our original problem. That problem, upon refinement, led to the view that an adequate theory of meaning must characterize a predicate meeting certain conditions. It was in the nature of a discovery that such a predicate would apply exacdy to the true sentences. I hope that what I am saying may be described in part as defending the philosophical importance of Tarski's semantical concept of truth. But my defence is only distandy related, if at all, to the quesdon whether the concept Tarski has shown how to define is the (or a) philosophically interesting conception of truth, or the question whether Tarski has cast any light on the ordinary use of such words as 'true' and 'truth'. It is a misfortune that dust from futile and confused battles over these questions has prevented those with a theoretical interest in language—philosophers, logicians, psychologists, and linguists alike—from seeing in the semantical concept of truth (under whatever name) the sophisticated and powerful foundation of a competent theory of meaning.
There is no need to suppress, of course, the obvious connection between a definition of truth of the kind Tarski has shown how to construct, and the concept of meaning. It is this: the definition works by giving necessary and sufficient conditions for the truth of every sentence, and to give truth conditions is a way of giving the meaning of a sentence. To know the semandc concept of
6  Assuming, of course, that the extension of these predicates is limited to the sentences of L.
7  A. Tarsld, 'The Concept of Truth in Formalized Languages'.
Truth and Meaning
161
truth for a language is to know what it is for a sentence—any sentence—to be true, and this amounts, in one good sense we can give to the phrase, to understanding the language. This at any rate is my excuse for a feature of the present discussion that is apt to shock old hands; my freewheeling use of the word 'meaning', for what I call a theory of meaning has after all turned out to make no use of meanings, whether of sentences or of words. Indeed, since a Tarski-type truth definition supplies all we have asked so far of a theory of meaning, it is clear that such a theory falls comfortably within what Quine terms the 'theory of reference' as distinguished from what he terms the 'theory of meaning'. So much to the good for what I call a theory of meaning, and so much, perhaps, against my so calling it.8
A theory of meaning (in my mildly perverse sense) is an empirical theory, and its ambition is to account for the workings of a natural language. Like any theory, it may be tested by comparing some of its consequences with the facts. In the present case this is easy, for the theory has been characterized as issuing in an infinite flood of sentences each giving the truth conditions of a sentence; we only need to ask, in sample cases, whether what the theory avers to be the truth conditions for a sentence really are. A typical test case might involve deciding whether the sentence 'Snow is white' is true if and only if snow is white. Not all cases will be so simple (for reasons to be sketched), but it is evident that this sort of test does not invite counting noses. A sharp conception of what constitutes a theory in this domain furnishes an exciting context for raising deep questions about when a theory of language is correct and how it is to be tried. But the difficulties are theoretical, not practical. In application, the trouble is to get a theory that comes close to working; anyone can tell whether it is right.9 One can see why this is so. The theory reveals nothing new about the conditions under which an individual sentence is true; it does not make those conditions any clearer than the sentence itself does. The work of the theory is in relating the known truth conditions of each sentence to those aspects ('words') of the sentence that recur in other sentences, and can be assigned identical roles in other sentences. Empirical power in such a theory depends on success in recovering the structure of a very complicated ability—the ability to speak and understand a language. We can tell easily enough when particular pronouncements of the theory comport with our understanding of the language; this is consistent with a feeble insight into the design of the machinery of our linguistic accomplishments.
8  But Quine may be quoted in support of my usage:"... in point of meaning ■.. a word may be said to be determined to whatever extent the truth or falsehood of its contexts is determined.' {'Truth by Convention', 82.) Since a truth definition determines die truth value of every sentence in the object language (relative to a sentence in the metalanguage), it determines the meaning of every word and sentence. This would seem to justify die tide Theory of Meaning.
9  To give a single example: it is clearly a count in favour of a theory that it entails ' "Snow is white" is true if and only if snow is white'. But to contrive a theory that entails diis (and works for all related sentences) is not trivial. I do not know a wholly satisfactory theory that succeeds with this very case (the problem of 'mass terms').
162
The Essential Davidson
The remarks of the last paragraph apply direcdy only to the special case where it is assumed that the language for which truth is being characterized is part of the language used and understood by the characterizer. Under these circumstances, the framer of a theory will as a matter of course avail himself when he can of the built-in convenience of a metalanguage with a sentence guaranteed equivalent to each sentence in the object language. Still, this fact ought not to con us into thinking a theory any more correct that entails ' "Snow is white" is true if and only if snow is white' than one that entails instead:
(S) 'Snow is white' is true if and only if grass is green,
provided, of course, we are as sure of the truth of (S) as we are ofthat of its more celebrated predecessor. Yet (S) may not encourage the same confidence that a theory that entails it deserves to be called a theory of meaning.
The threatened failure of nerve may be counteracted as follows. The gro-tesqueness of (S) is in itself nothing against a theory of which it is a consequence, provided the theory gives the correct results for every sentence (on the basis of its structure, there being no other way). It is not easy to see how (S) could be party to such an enterprise, but if it were—if, that is, (S) followed from a characterization of the predicate 'is true' that led to the invariable pairing of truths with truths and falsehoods with falsehoods—then there would not, I think, be anything essential to the idea of meaning that remained to be captured.10
What appears to the right of the biconditional in sentences of the form 's is true if and only if/»' when such sentences are consequences of a theory of truth plays its role in determining the meaning of s not by pretending synonymy but by adding one more brush-stroke to the picture which, taken as a whole, tells what diere is to know of the meaning of í; this stroke is added by virtue of the fact that the sentence that replaces lp' is true if and only if s is.
It may help to reflect that (S) is acceptable, if it is, because we are indepen-dendy sure of the truth of'Snow is white' and 'Grass is green'; but in cases where we are unsure of the truth of a sentence, we can have confidence in a characterization of the truth predicate only if it pairs that sentence with one we have good reason to believe equivalent. It would be ill advised for someone who had any doubts about the colour of snow.or grass to accept a theory that yielded (S), even if his doubts were of equal degree, unless he thought the colour of the one was tied to the colour of the other.'' Omniscience can obviously afford more
10 Critics have often railed to notice die essential proviso mentioned in this paragraph. The point is that (S) could not belong to any reasonably simple theory that also gave the right truth conditions for 'That is snow' and 'This is white'. (See the discussion of indexical expressions below.) [Footnote added in 1982.]
1' This paragraph is confused. What it should say is that sentences of the theory are empirical generalizations about speakers, and so must not only be true but also lawlike. (S) presumably is not a law, since it does not support appropriate countcr-factuals. It's also important that the evidence for accepting the (time and speaker relativized) truth conditions for 'That is snow' is based on the causal connection between a speaker's assent to the sentence and the demonstrative presentation of snow.
Truth and Meaning
163
bizzare theories of meaning than ignorance; but then, omniscience has less need of communication.
It must be possible, of course, for the speaker of one language to construct a theory of meaning for the speaker of another, though in this case the empirical test of the correctness of the theory will no longer be trivial. As before, the aim of theory will be an infinite correlation of sentences alike in truth. But this dme the theory-builder must not be assumed to have direct insight into likely equivalences between his own tongue and the alien. What he must do is find out, however he can, what sentences the alien holds true in his own tongue (or better, to what degree he holds them true). The linguist then will attempt to construct a characterization of truth-for-the-alien which yields, so far as possible, a mapping of sentences held true (or false) by the alien on to sentences held true (or raise) by the linguist. Supposing no perfect fit is found, the residue of sentences held true translated by sentences held false (and vice versa) is the margin for error (foreign or domestic). Charity in interpreting the words and thoughts of others is unavoidable in another direction as well: just as we must maximize agreement, or risk not making sense of what the alien is talking about, so we must maximize the self-consistency we attribute to him, on pain of not understanding him. No single principle of optimum charity emerges; the constraints therefore determine no single theory. In a theory of radical translation (as Quine calls it) there is no completely disentangling questions of what the alien means from questions of what he believes. We do not know what someone means unless we know what he believes; we do not know what someone believes unless we know what he means. In radical interpretation we are able to break into this circle, if only incompletely, because we can sometimes tell that a person accedes to a sentence we do not understand.l2
In the past few pages I have been asking how a theory of meaning that takes the form of a truth definition can be empirically tested, and have blithely ignored the prior question whether there is any serious chance such a theory can be given for a natural language. What are the prospects for a formal semantical theory of a natural language? Very poor, according to Tarski; and I believe most logicians, philosophers of language, and linguists agree.13 Let me do what I can to dispel the pessimism. What I can in a general and programmatic way, of course, for here the proof of the pudding will certainly be in the proof of the right theorems.
12  This sketch of how a theory of meaning for an alien tongue can be tested obviously owes its inspiration to Quine's account of radical translation in Chapter II ofWordand Object. In suggesting that an acceptable theory of radical translation take the form of a recursive characterization of truth, I go beyond Quine. Toward the end of this paper, in the discussion of demonstratives, another strong point of agreement will turn up.
13  So far as I am aware, there has been very little discussion of whether a formal truth definition can be given for a natural language. But in a more general vein, several people have urged that the concepts of formal semantics be applied to natural language. See, for example, the contributions of Yehoshua Bar-Hillel and Evert Beth to The Philosophy of Rudolph Carnap, and Bar-Hillel's 'Logical Syntax and Semantics'.
164
The Essential Davidson
Tarski concludes the first section of his classic essay on the concept of truth in formalized languages with the following remarks, which he italicizes:
... The very possibility of a consistent use of the expression 'true sentence'which is in harmony with the laws of logic and the spirit of everyday language seems to be very questionable, and consequently the same doubt attaches to the possibility of constructing a correct definition of this expression. (165)
Late in the same essay, he returns to the subject:
... the concept of truth (as well as other semantical concepts) when applied to colloquial language in conjunction with the normal laws of logic leads inevitably to confusions and contradictions. Whoever wishes, in spite of all difficulties, to pursue the semantics of colloquial language with the help of exact methods will be driven first to undertake the thankless task of a reform of this language. He will find it necessary to define its structure, to overcome die ambiguity of the terms which occur in it, and finally to split the language into a series of languages of greater and greater extent, each of which stands in the same relation to the next in which a formalized language stands to its metalanguage. It may, however be doubted whether the language of everyday life, after being 'rationalized' in this way, would still preserve its naturalness and whether it would not rather take on the characteristic features of the formalized languages. (267)
Two themes emerge: that the universal character of natural languages leads to contradiction (the semantic paradoxes), and that natural languages are too confused and amorphous to permit the direct application of formal methods. The first point deserves a serious answer, and I wish I had one. As it is, I will say only why I think we are justified in carrying on without having disinfected this particular source of conceptual anxiety. The semantic paradoxes arise when the range of the quantifiers in the object language is too generous in certain ways. But it is not really clear how unfair to Urdu or to Wendish it would be to view the range of their quantifiers as insufficient to yield an explicit definition of 'true-in-Urdu' or 'true-in-Wendish'. Or, to put the matter in another, if not more serious way, there may in the nature of the case always be something we grasp in understanding the language of another (the concept of truth) that we cannot communicate to him. In any case, most of the problems of general philosophical interest arise within a fragment of the relevant natural language that may be conceived as containing very little set theory. Of course these comments do not meet the claim that natural languages are universal. But it seems to me that this claim, now that we know such universality leads to paradox, is suspect.
Tarski's second point is that we would have to reform a natural language out of all recognition before we could apply formal semantical methods. If this is true, it is fatal to my project, for the task of a theory of meaning as I conceive it is not to change, improve, or reform a language, but to describe and understand it. Let us look at the positive side. Tarski has shown the way to giving a theory for interpreted formal languages of various kinds; pick one as much like English as possible. Since this new language has been explained in English and contains
Truth and Meaning
165
much English we not only may, but I think must, view it as part of English for those who understand it. For this fragment of English we have, ex hypothesis a theory of the required sort. Not only that, but in interpreting this adjunct of English in old English we necessarily gave hints connecting old and new. Wherever there are sentences of old English with the same truth conditions as sentences in the adjunct we may extend the theory to cover them. Much of what is called for is to mechanize as far as possible what we now do by art when we put ordinary English into one or another canonical notation. The point is not that canonical notation is better than the rough original idiom, but rather that if we know what idiom the canonical notation is canonical for, we have as good a theory for the idiom as for its kept companion.
Philosophers have long been at the hard work of applying theory to ordinary language by the device of matching sentences in the vernacular with sentences for which they have a theory. Frege's massive contribution was to show how 'all*, 'some', 'every', 'each', 'none', and associated pronouns, in some of their uses, could be tamed; for the first time, it was possible to dream of a formal semantics for a significant part of a natural language. This dream came true in a sharp way with the work of Tarski. It would be a shame to miss the fact that as a result of these two magnificent achievements, Frege's and Tarski's, we have gained a deep insight into the structure of our mother tongues. Philosophers of a logical bent have tended to start where the theory was and work out towards the complications of natural language. Contemporary linguists, with an aim that cannot easily be seen to be different, start with the ordinary and work toward a general theory. If either party is successful, there must be a meeting. Recent work by Chomsky and others is doing much to bring the complexities of natural languages within the scope of serious theory. To give an example: suppose success in giving the truth conditions for some significant range of sentences in the active voice. Then with a formal procedure for transforming each such sentence into a corresponding sentence in the passive voice, the theory of truth could be extended in an obvious way to this new set of sentences.14
One problem touched on in passing by Tarski does not, at least in all its manifestations, have to be solved to get ahead with theory: the existence in natural languages of 'ambiguous terms'. As long as ambiguity does not affect
14 The rapprochement I prospectively imagine between transformational grammar and a sound theory of meaning has been much advanced by a recent change in the conception of transformational grammar described by Chomsky in the article referred to above (footnote 5). The structures generated by the phrase-structure part of the grammar, it has been realized for some time, arc those suited to semantic interpretation; but this view is inconsistent with the idea, held by Chomsky until recently, that recursive operations are introduced only by the transformation rules. Chomsky now believes the phrase-structure rules are recursive. Since languages to which formal semantic methods direcdy and naturally apply are ones for which a (recursive) phrase-structure grammar is appropriate, it is clear that Chomsky's present picture of the relation between the structures generated by the phrase-structure part of the grammar, and the sentences of the language, is very much like the picture many logicians and philosophers have had of the relation between (he richer formalized languages and ordinary language. (In these remarks I am indebted to Bruce Vermazen.)
166
The Essential Davidson
grammatical form, and can be translated, ambiguity for ambiguity, into the metalanguage, a truth definition will not tell us any lies. The chief trouble, for systematic semantics, with the phrase 'believes that' in English lies not in its vagueness, ambiguity, or unsuitability for incorporation in a serious science: let our metalanguage be English, and all these problems will be carried without loss or gain into the metalanguage. But the central problem of the logical grammar of 'believes that' will remain to haunt us.
The example is suited to illustrating another, and related, point, for the discussion of belief sentences has been plagued by failure to observe a fundamental distinction between tasks: uncovering the logical grammar or form of sentences (which is in the province of a theory of meaning as I construe it), and the analysis of individual words or expressions (which are treated as primitive by the theory). Thus Carnap, in the first edition of Meaning and Necessity, suggested we render 'John believes that the earth is round' as 'John responds affirmatively to "the earth is round" as an English sentence'. He gave this up when Mates pointed out that John might respond affirmatively to one sentence and not to another no matter how close in meaning.15 But there is a confusion here from the start. The semantic structure of a belief sentence, according to this idea of Carnap's, is given by a three-place predicate with places reserved for expressions referring to a person, a sentence, and a language. It is a different sort of problem entirely to attempt an analysis of this predicate, perhaps along behaviouristic lines. Not least among the merits of Tarski's conception of a theory of truth is that the purity of method it demands of us follows from the formulation of the problem itself, not from the self-imposed restraint of some adventitious philosophical puritanism.
I think it is hard to exaggerate the advantages to philosophy of language of bearing in mind this distinction between questions of logical form or grammar, and the analysis of individual concepts. Another example may help advertise the point.
If we suppose questions of logical grammar settled, sentences like 'Bardot is good' raise no special problems for a truth definition. The deep differences between descriptive and evaluative (emotive, expressive, etc.) terms do not show here. Even if we hold there is some important sense in which moral or evaluative sentences do not have a truth value-.(for example, because they cannot be verified), we ought not to boggle at' "Bardot is good" is true if and only if Bardot is good'; in a theory of truth, this consequence should, follow with the rest, keeping track, as must be done, of the semantic location of such sentences in the language as a whole—of their relation to generalizations, their role in such compound sentences as 'Bardot is good and Bardot is foolish', and so on. What is special to evaluative words is simply not touched: the mystery is transferred from the word 'good' in the object language to its translation in the metalanguage.
15 B. Mates, 'Synonymity'.
Truth and Meaning
167
But 'good' as it features in 'Bardot is a good actress' is another matter. The problem is not that the translation of this sentence is not in the metalanguage— let us suppose it is. The problem is to frame a truth definition such that' "Bardot is a good actress" is true if and only if Bardot is a good actress' —and all other sentences like it—are consequences. Obviously 'good actress' does not mean 'good and an actress'. We might think of taking 'is a good actress' as an unanalysed predicate. This would obliterate all connection between 'is a good actress' and 'is a good mother', and it would give us no excuse to think of'good', in these uses, as a word or semantic element. But worse, it would bar us from framing a truth definition at all, for there is no end to the predicates we would have to treat as logically simple (and hence accommodate in separate clauses in the definition of satisfaction): 'is a good companion to dogs', 'is a good 28-years old conversationalist', and so forth. The problem is not peculiar w the case: it is the problem of attributive adjectives generally.
It is consistent with the attitude taken here to deem it usually a strategic error to undertake philosophical analysis of words or expressions which is not preceded by or at any rate accompanied by the attempt to get the logical grammar straight. For how can we have any confidence in our analyses of words like 'right', 'ought', 'can', and 'obliged', or the phrases we use to talk of actions, events, and causes, when we do not know what (logical, semantical) parts of speech we have to deal with? I would say much the same about studies of the 'logic' of these and other words, and the sentences containing them. Whether the effort and ingenuity that have gone into the study of deontic logics, modal logics, imperative and erotetic logics have been largely futile or not cannot be known until we have acceptable semantic analyses of the sentences such systems purport to treat. Philosophers and logicians sometimes talk or work as if they were free to choose between, say, the truth-functional conditional and others, or free to introduce non-truth-functional sentential operators like 'Let it be the case that' or 'It ought to be the case that'. But in fact the decision is crucial. When we depart from idioms we can accommodate in a truth definition, we lapse into (or create) language for which we have no coherent semantical account—that is, no account at all of how such talk can be integrated into the language as a whole.
To return to our main theme: we have recognized that a theory of the kind proposed leaves the whole matter of what individual words mean exacdy where it was. Even when the metalanguage is different from the object language, the theory exerts no pressure for improvement, clarification, or analysis of individual words, except when, by accident of vocabulary, straightforward translation fails. Just as synonomy, as between expressions, goes generally untreated, so also synonomy of sentences, and analyticity. Even such sentences as 'A vixen is a female fox' bear no special tag unless it is our pleasure to provide it. A truth definition does not distinguish between analytic sentences and others, except for sentences that owe their truth to the presence alone of the constants that give the theory its grip on structure: the theory entails not only that these sentences are
168                                     The Essential Davidson
true but that they will remain true under all significant rewritings of their non-logical parts. A notion of logical truth thus given limited application, related notions of logical equivalence and entailment will tag along. It is hard to imagine how a theory of meaning could fail to read a logic into its object language to this degree; and to the extent that it does, our intuitions of logical truth, equivalence, and entailment may be called upon in constructing and testing the theory.
I turn now to one more, and very large, fly in the ointment: the fact that the same sentence may at one time or in one mouth be true and at another time or in another mouth be false. Both logicians and those critical of formal methods here seem largely (though by no means universally) agreed that formal semantics and logic are incompetent to deal with the disturbances caused by demonstratives. Logicians have often reacted by downgrading natural language and trying to show how to get along without demonstratives; their critics react by downgrading logic and formal semantics. None of this can make me happy: clearly demonstratives cannot be eliminated from a natural language without loss or radical change, so there is no choice but to accommodate theory to them.
No logical errors result if we simply treat demonstratives as constants;16 neither do any problems arise for giving a semantic truth definition.' "I am wise" is true if and only if I am wise', with its bland ignoring of the demonstrative element in T comes off the assembly line along with ' "Socrates is wise" is true if and only if Socrates is wise' with its bland indifference to the demonstrative element in 'is wise' (the tense).
What suffers in this treatment of demonstratives is not the definition of a truth predicate, but the plausibility of the claim that what has been defined is truth. For this claim is acceptable only if the speaker and circumstances of utterance of each sentence mentioned in the definition is matched by the speaker and circumstances of utterance of the truth definition itself. It could also be fairly pointed out that part of understanding demonstratives is knowing the rules by which they adjust their reference to circumstance; assimilating demonstratives to constant terms obliterates this feature. These complaints can be met, I think, though only by a fairly far-reaching revision in the theory of truth. I shall barely suggest how this could be done, but bare suggestion is all that is needed: the idea is technically trivial, and in line with work being done on the logic of the tenses.17
We could take truth to be a property, not of sentences, but of utterances, or speech acts, or ordered triples of sentences, times, and persons; but it is simplest just to view truth as a relation between a sentence, a person, and a time. Under such treatment, ordinary logic as now read applies as usual, but only to sets of sentences relativized to the same speaker and time; further logical relations between sentences spoken at different times and by different speakers may be
16  See W. V. Quine, Methods of Logic, 8.
17  This daim has turned out to be naively optimistic. For some serious work on the subject, see S. Weinstein, 'Truth and Demonstratives'. [Note added in 1982.]
Truth and Meaning                                      169
articulated by new axioms. Such is not my concern. The theory of meaning undergoes a systematic but not puzzling change; corresponding to each expression with a demonstrative element there must in the theory be a phrase that relates the truth conditions of sentences in which the expression occurs to changing times and speakers. Thus the theory will entail sentences like the following:
'I am tired' is true as (potentially) spoken by p at t if and only if/» is tired at t. 'That book was stolen' is true as (potentially) spoken by p at t if and only if the book demonstrated by/» at ŕ is stolen prior to ŕ.18
Plainly, this course does not show how to eliminate demonstratives; for example, there is no suggestion that 'the book demonstrated by the speaker' can be substituted ubiquitously for 'that book' salva veritate. The fact that demonstratives are amenable to formal treatment ought gready to improve hopes for a serious semantics of natural language, for it is likely that many outstanding puzzles, such as the analysis of quotations or sentences about propositional attitudes, can be solved if we recognize a concealed demonstrative construction.
Now that we have relativized truth to times and speakers, it is appropriate to glance back at the problem of empirically testing a theory of meaning for an alien tongue. The essence of the method was, it will be remembered, to correlate held-true sentences with held-true sentences by way of a truth definition, and within the bounds of intelligible error. Now the picture must be elaborated to allow for the fact that sentences are true, and held true, only relative to a speaker and a time. Sentences with demonstratives obviously yield a very sensitive test of the correctness of a theory of meaning, and constitute the most direct link between language and the recurrent macroscopic objects of human interest and attention.19
In this paper I have assumed that the speakers of a language can effectively determine the meaning or meanings of an arbitrary expression (if it has a meaning), and that it is the central task of a theory of meaning to show how this is possible. I have argued that a characterization of a truth predicate describes the required kind of structure, and provides a clear and testable criterion of an adequate semantics for a natural language. No doubt there are other reasonable demands that may be put on a theory of meaning. But a theory that does no more than define truth for a language comes far closer to constituting a complete theory of meaning than superficial analysis might suggest; so, at least, I have urged.
Since I think there is no alternative, I have taken an optimistic and programmatic view of the possibilities for a formal characterization of a truth
la There is more than an intimation of this approach to demonstratives and truth inj. L. Austín, 'Truth'.
19 These remarks derive from Quine's idea that 'occasion sentences' (those with a demonstrative element) must play a central role in constructing a translation manual.
170
The Essential Davidson
predicate for a natural language. But it must be allowed that a staggering list of difficulties and conundrums remains. To name a few: we do not know the logical form of counterfactual or subjunctive sentences; nor of sentences about probabilities and about causal relations; we have no good idea what the logical role of adverbs is, nor the role of attributive adjectives; we have no theory for mass terms like 'fire', 'water', and 'snow', nor for sentences about belief, perception, and intention, nor for verbs of action that imply purpose. And finally, there are all the sentences that seem not to have truth values at all: the imperatives, optatives, interrogatives, and a host more. A comprehensive theory of meaning for a natural language must cope successfully with each of these problems.20
20 For attempted solutions to some of these problems see Essays 9 and 10 in this book, with special reference to section 1 of Essay 10. See also Essays 6-10 of Davidson, Essays on Actions and Events {1980, 2nd edn. 2001), and Essays 6, 8, and 10 in Davidson, Inquiries into. Truth and Interpretation (1984, 2nd edn. 2001).
9
On Saying That
'I wish I had said that', said Oscar Wilde in applauding one of Whisder's witticisms. Whisder, who took a dim view of Wilde's originality, retorted, 'You will, Oscar; you will'.1 This tale reminds us that an expression like 'Whisder said that' may on occasion serve as a grammatically complete sentence. Here we have, I suggest, the key to a correct analysis of indirect discourse, an analysis that opens a lead to an analysis of psychological sentences generally (sentences about prepositional attitudes, so-called), and even, though this looks beyond anything to be discussed in the present paper, a clue to what distinguishes psychological concepts from others.
But let us begin with sentences usually deemed more representative of oratio obliqua, for example 'Galileo said that the earth moves' or 'Scott said that Venus is an inferior planet'. One trouble with such sentences is that we do not know their logical form. And to admit this is to admit that, whatever else we may know about them, we do not know the first thing. If we accept surface grammar as guide to logical form, we will see 'Galileo said that the earth moves' as containing the sentence 'the earth moves', and this sentence in turn as consisting of the singular term 'the earth', and a predicate, 'moves'. But if 'the earth' is, in this context, a singular term, it can be replaced, so far as the truth or falsity of the containing sentence is concerned, by any other singular term that refers to the same thing. Yet what seem like appropriate replacements can alter the truth of the original sentence.
The notorious apparent invalidity of this move can only be apparent, for the rule on which it is based no more than spells out what is involved in the idea of a (logically) singular term. Only two lines of explanation, then, are open: we are wrong about the logical form, or we are wrong about the reference of the singular term.
What seems anomalous behaviour on the part of what seem singular terms dramatizes the problem of giving an orderly account of indirect discourse, but the problem is more pervasive. For what touches singular terms touches what they touch, and that is everything: quantifiers, variables, predicates, connectives. Singular terms refer, or pretend to refer, to the entities over which the variables of
From H. Jackson, The Eighteen-Nineties, 73.