PAUL HORWICH
A DEFENSE OF MINIMALISM
My aim in this paper is to clarify and defend a certain ‘minimalist’ thesis about truth: roughly, that the meaning of the truth predicate is fixed by the schema, ’The proposition that p is true if and only if p’.1 The several criticisms of this idea to which I wish to respond are to be found in the recent work of Davidson, Field, Gupta, Richard, and Soames, and in a classic paper of Dummett’s. But before addressing these criticisms let me begin by saying something more about the thesis itself. Consider biconditionals like hsnow is whitei is true ↔ snow is white and hlying is wrongi is true ↔ lying is wrong2 – that is, instances of the equivalence schema hpi is true ↔ p It can be argued that such biconditionals are epistemologically fundamental: – we do not arrive at them, or seek to justify our acceptance of them, on the basis of anything more obvious or more immediately known. It can be argued, in addition, that our underived inclination to accept these biconditionals is the source of everything else we do with the truth predicate. For example, from the premises What he said is that he was abducted and What he said is true we are prepared to infer He was abducted Synthese 126: 149–165, 2001. © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
150
PAUL HORWICH
This particular use of the word “true” is explained by supposing that we first employ Leibniz’ Law to get from our pair of premises to hHe was abductedi is true and then invoke the relevant instance of the equivalence schema. And, more generally, it can be made plausible that no further fact about the truth predicate – nothing beyond our allegiance to the equivalence schema – is needed to explain any of our ways of using it. It is for this reason that we are entitled to conclude that the meaning of “true” is determined by that schema. For, plausibly, the property of a word that constitutes its having the particular meaning that it has should be identified with the property that explains the symptoms of its possessing that meaning – and these symptoms are the various characteristic ways in which it is used.3 Thus my minimalist thesis is the product of two prior claims: first, that our underived endorsement of the equivalence schema is explanatorily fundamental with respect to the overall use of the truth predicate; and second, that the meaning of any word is engendered by the fact about it that explains its overall use.4 This line of thought can be challenged at various points and no doubt stands in need of considerable further justification.5 But my main aim here is not to defend my route to the minimalist conclusion, but rather to defend that conclusion itself: namely, that the meaning of “true” stems from the equivalence schema. For most of the recent objections to this thesis do not target any particular rationale for it, but purport to demonstrate that the thesis itself cannot be correct. However, before addressing these objections, let me help to prepare the ground for my replies to them by saying a little more to clarify just what the proposal is, and is not, intended to encompass. Several different kinds of theory, with very different explanatory objectives, might appropriately be labeled “theories of truth”. So it is important to be clear about what sort of theoretical work the minimalist proposal is not meant to do and should not be blamed for failing to do. In the first place, it is not intended to provide an explicit definition of the word “true”, neither descriptive nor stipulative. Therefore it does not offer a way of rearticulating the contents of sentences containing the word. Indeed, it implies that no such reformulations are possible.6 In the second place, the proposal does not amount to a substantive reductive theory of the property of being true – something in the style of ’water is H2 O’ – which would tell us how truth is constituted at some underlying level. Again, it suggests that the search for such a theory would be misguided.7 And in the third place, it is not a ‘theory of truth’ in the sense of a set of fundamental theoretical
A DEFENSE OF MINIMALISM
151
postulates on the basis of which all other facts about truth can be explained. Its immediate concern is with the word “true” rather than with truth itself. It purports to specify which of the non-semantic facts about that word is responsible for its meaning what it does; and the fact it so specifies is our underived allegiance to the equivalence schema. Now let me turn to an array of objections. I will look at two difficulties raised by Donald Davidson, one posed by Hartry Field, three devised by Anil Gupta, one due to Mark Richard, one that I put to myself, and one old, but still influential, objection of Michael Dummett’s. My discussion of each of these problems will be fairly brief – merely indicating the lines along which I think the response should be given, rather than giving it in full. Objection 1: The minimalist proposal implies that the meaning of “true” is established on the basis of the meaning of “the proposition that . . . ”. For someone’s acceptance of, for example, “The proposition that snow is white is true if and only if snow is white” manifests a standard understanding of the truth predicate only to the extent that the component expression, “The proposition that . . . ”, is being understood in the standard way. Thus minimalism implies that one must already understand that-clauses – i.e., one must understand sentences of the form “u expresses (i.e., means) the proposition that p” – in order to be in a position to acquire the concept of truth. But this surely gets things the wrong way round! Surely the intimately related notions of meaning and proposition must be analysed in terms of truth.– Otherwise we wouldn’t be able to account for the compositionality of meaning. More specifically, we must (for the sake of compositionality) suppose that ‘u expresses the proposition (i.e., means) that p’ consists in the fact that ‘u is true (i.e., expresses a truth) if and only if p’ Thus truth is conceptually prior to meaning, contrary to what is required by the minimalist proposal. (Davidson8 ) The compositionality of meaning is the fact that the meanings of sentences depend on the meanings of their component words and on how those words are put together. Or, in other words, it is the fact that our interpretations of the complex expressions of someone’s language derive from our interpretations of that person’s primitive terms. Now, despite the considerations
152
PAUL HORWICH
advanced by Davidson in his influential essay, “Truth and Meaning”, and elsewhere, this fact does not really call for an analysis of meaning in terms of truth conditions.9 A viable alternative to the Davidsonian strategy is to suppose, with Frege, that whenever a complex expression is formed by applying a function-expression (e.g., a predicate) to a sequence of argument-expressions (e.g., names) the meaning of the complex is the result of applying the meaning of the function-expression to the meanings of its arguments. For example, the meaning of the result of applying the predicate “rotates” to the name “Hesperus” – i.e., the meaning of that sentence – is the meaning of “rotates” applied to the meaning of “Hesperus”. On the basis of this principle, and given specifications of the meanings of the words in a language, it is possible to deduce characterizations of the meanings of every sentence (structurally described) and hence to interpret the entire language.10 Davidson’s objection to the Fregean view of compositionality is that it is “vacuous” – that it implies merely that “Hesperus rotates” means whatever it means, that it fails to show how that meaning depends on the structure of the sentence, and that it does not help us to give interpretations of complex expressions. But these objections are wrong. The Fregean principle can be no emptier than the compositionality it expresses; it says that “Hesperus rotates” means the result of applying the meaning of “rotates” to the meaning of “Hesperus” – which is not a tautology; it does offer a structure-dependent characterization of what the sentence means; and for the sake of interpretation such characterizations are all that are needed. Moreover, this account of compositionality puts no constraint at all on how the meanings of words are in turn constituted. In particular, it squares perfectly well with supposing that the meanings of words are engendered by non-truth-theoretic aspects of their use. Indeed, I believe such a view of word meaning can be made highly plausible.11 In that case, our understanding of (and ability to deploy) “u means that Hesperus rotates” will emerge from our knowledge of the uses (hence meanings) of “Hesperus” and “rotates” and from our appreciation of how the sentence results from applying one of them to the other. Thus, contrary to this initial Davidsonian objection, it is quite reasonable to allow that we could first possess the concepts of meaning and proposition and then, on that basis, fix the meaning of the word “true” (and acquire the concept of truth) by accepting instances of the equivalence schema.12 Objection 2: Sentences like “The proposition that Hesperus rotates is true”, insofar as they are construed as predicating truth of the propositions to which that-clauses refer, are in fact unintelligible, since that-clauses
A DEFENSE OF MINIMALISM
153
cannot be regarded as referring terms. And this is so because there is no way of seeing how their referents would be determined by the referents of their component words. But if such truth ascriptions (so construed) are unintelligible, then the minimalist proposal cannot be correct. (Davidson13 ) Davidson’s basic reason for maintaining that alleged referents of thatclauses would not be determinable by the referents of the parts of these clauses is – following Frege – that substitution of co-referential terms (e.g., putting “Phosphorus” in place of “Hesperus”) within a that-clause occuring in some sentence (e.g., “John believes that Hesperus rotates”) may not preserve the truth value of that sentence. But why does he not continue along Fregean lines, and conclude that an expression within a that-clause does not have its standard referent, but instead refers to the meaning of that expression? Why not identify the referent of “the proposition that Hesperus rotates” with the meaning of “Hesperus rotates” and identify the referents of the contained words “Hesperus” and “rotates” with their meanings? And why not suppose, as suggested in the response to Objection 1, that the meaning of “Hesperus rotates” is the result of applying (in the sense of applying a function to an argument) the predicative meaning of “rotates”, to the nominal meaning of “Hesperus”? Davidson has made two objections to this Fregean proposal. His first objection, given in “Truth and Meaning”, is that it is unilluminating. But, as we have just seen, this criticism is overstated. No more substantive account of compositionality than is contained in the Fregean principle is required to derive interpretations of complex expressions from interpretations of their parts. Admittedly, on this account, compositionality is strikingly easy to explain. But that shows merely that there is much less to the so-called ‘problem’ of compositionality than is often supposed.14 More recently (in his “The Folly of Trying to Define Truth”) Davidson has advanced a somehwat different reason for rejecting the Fregean picture (and hence for concluding that the referents of the parts of a that-clause could not determine a referent for the whole). He observes that the meanings of words in that-clauses are just their normal meanings. – After all, we understand “The proposition that Hesperus rotates” only if we understand the isolated sentence “Hesperus rotates”. And in the biconditional, “The proposition that Hesperus rotates is true ↔ Hesperus rotates”, the two occurences of “Hesperus rotates” are clearly supposed to be understood in the same way. But in that case – since meaning determines reference – how could words in that-clauses fail to have their standard referents. And if they do have their standard referents then that-clauses cannot refer, since
154
PAUL HORWICH
what would be determined by those standard referents would be the wrong thing (e.g., “that Hesperus rotates” would acquire the same referent as “that Phosphorus rotates”). To this line of thought it seems to me that two reasonable responses are available. In the first place we might well deny that meaning determines reference. We might suppose, on the contrary, that the referent of a term is fixed in part by the context in which it occurs. More specifically, we might say that the single meaning of “Hesperus” yields one referent (the planet) for standard (non-opaque) occurences of the word, and that it yields a different referent (the meaning of “Hesperus”) for occurences of the word within that-clauses. An alternative response would be to deny that the referent of a complex expression is determined by the referents of its grammatical parts. We might prefer to say instead that it is only for logically articulated expressions that their referents are determined by the referents of their parts. For we might suppose that although the words “Hesperus” and “rotates” are indeed used in the superficial grammatical form of “The proposition that Hesperus rotates”, the underlying logical form might be something in which those words are not used, but are merely mentioned (e.g., “The proposition expressed by the sentence ‘Hesperus rotates’, as I currently understand it”). Admittedly, this is no longer Fregean. Nonetheless, it would treat that-clauses as singular terms, and it would conform to Davidson’s requirements on such a treatment: namely, that their referents be determined by the referents of their logical parts, and that these parts have the same meanings inside that-clauses as they do outside.15 Objection 3: The minimalist proposal would leave it mysterious how we are able to attribute content to sentences that predicate truth of foreign statements we can’t understand. For if an utterance u is known to mean (say) that dogs bark, then (according to the proposal) the sentence “u is true” (or “u expresses a truth”) might be interpreted as saying roughly that dogs bark; whereas if u is not understood, then the proposal enables us to attach no meaning at all to “u is true”. But we surely do think that it is meaningful to predicate truth of statements we cannot understand. So the minimalist proposal is defective. (Field16 ) But, as we have already seen, the proposal is not intended as a definition of “true”: it does not purport to provide a way of reformulating or rearticulating the content of each sentence containing the word “true”. On the contrary, it implies that such a thing is not possible. So one cannot reasonably complain that the minimalist proposal fails to yield a concep-
A DEFENSE OF MINIMALISM
155
tual analysis of the sentence “u is true”. One certainly could complain if the proposal implied that this sentence lacked content.– But it has no such implication. It aims to specify the underlying use property in virtue of which the truth predicate means what it does. To that end, it identifies certain tokens of that predicate as explanatorily fundamental and hence meaning-constituting – namely, those that appear in instances of the equivalence schema. But other tokens may perfectly well have the very same meaning, as long as their deployment stems from the fundamental ones. Thus someone who reasons inductively to “u expresses something true” on the basis of the fact that the other assertions of the speaker – those that can be translated – have turned out to be true, uses the truth predicate with a constant meaning, one that is engendered by the equivalence schema. Objection 4: The equivalence schema is not enough to fix the meaning of the truth predicate, because exactly parallel schemata are satisfied by predicates that do not mean the same as “true”. For example, instances of the schema, ‘hpi is true and not red ↔ p’, are just as obviously correct as instances of the equivalence schema. But “is true and not red” is not a synonym of the truth predicate. More generally, the schema, ‘hpi is f ↔ p’, will be endorsed relative to a variety of predicates, “f ”, with different meanings from one another; therefore, for any given “f ”, it cannot be that our acceptance of instances of ‘hpi is f ↔ p’ is what fixes “f ”’s meaning. (Gupta17 ) Indeed. However, according to minimalism, what fixes the meaning of the truth predicate is not merely our allegiance to the equivalence schema but, in addition, the fact that this allegiance is epistemologically basic – i.e. the fact that our endorsement of the equivalence schema is that use of “true” which is not derived from any more fundamental assumptions formulated by means of the truth predicate. Certainly there are parallel schemata, constructed with other predicates in place of “true”, that are equally acceptable. However, in every such case our commitment to the schema is derived. For example, it is as a consequence of our endorsing the equivalence schema for truth, and of our accepting “No proposition is red”, that we accept instances of ‘hpi is true and not red ↔ p’. Only in the case of the truth predicate does the corresponding schema capture what is epistemologically basic in our usage of the predicate. And it is this property that engenders its meaning. Objection 5: The minimalist proposal implies either that the word “true” cannot be fully understood or that the meaning of each person’s truth pre-
156
PAUL HORWICH
dicate depends on, and varies with, whatever else is in his vocabulary. For the proposal is tantamount to the definition: x is true ≡ [x = hdogs barki & dogs bark; or x = hpigs flyi & pigs fly; or . . . and so on] Therefore, if the “and so on” is intended to cover all propositions, then – since some of them must involve concepts that no-one possesses – the meaning of “true” cannot be fully known to anyone. And if, alternatively, the definition of each person’s truth predicate is supposed to cover only those propositions he can grasp, then, as new concepts are deployed and new terms coined, his definition of “true” will change. But neither of these alternative implications of minimalism is acceptable. Surely our understanding of the truth predicate is both complete and constant. (Gupta18 ) Agreed. But my proposal is perfectly consonant with such intuitions, because it is not at all equivalent to the above definition. As already emphasized, the minimalist thesis does not offer any explicit definition. Rather it purports to specify the fact of usage that provides the truth predicate with its meaning. That fact of usage, it claims, is our underived inclination to accept instances of the equivalence schema – a fact that remains the same as the rest of our language evolves. So, for example, at the moment that the term, “tachyon”, enters our language, we become inclined to accept htachyons go backwards in timei is true ↔ tachyons go backwards in time But this is merely one more application of a single and invariable regularity – our inclination to accept any instance of the schema that we understand. That inclination preceded the introduction of the term “tachyon” and was in no way altered by it. Thus the minimalist thesis does not imply that the meaning of the word “true” can’t be fully grasped, or that it changes with the expansion of our vocabulary. Objection 6: Our reliance on the equivalence schema will not suffice to explain our knowledge of general facts about truth. Consider, for example, “All propositions of the form, hp → pi, are true”. No doubt our particular logical convictions together with our commitment to the equivalence schema can explain, for any single proposition, why we take it to be true that this proposition implies itself. Thus we can explain, given our logical commitment to “dogs bark → dogs bark”, why we also accept “The proposition that dogs bark → dogs bark is true”. But we have not thereby
A DEFENSE OF MINIMALISM
157
explained how the above generalization is reached. Thus our allegiance to the equivalence schema does not really suffice to account for all uses of the truth predicate. Therefore that practice does not fix the meaning of “true”, contrary to what the minimalist maintains. (Gupta, Soames19 ) Granted, some further explanatory premise is needed. But this concession provides an objection to the proposal only if the needed additional premise specifies properties of the word “true”. For only then will it emerge that our commitment to the equivalence schema, together with facts that have nothing specifically to do with the truth predicate, are insufficient to explain its overall use. But actually it is far from obvious that the premise we should add will explicitly concern the truth predicate. Suppose, for example, it were a fact that whenever someone can establish, for every F , that it is G, then he comes, on that basis, to believe that every F is G. Combined with such a fact (which does not explicitly concern the truth predicate) our disposition to accept, for each proposition of a certain form, that it is true would suffice to explain our acceptance of the generalization, “Every proposition of that form is true”. Of course this response to the objection will not do as it stands, because the proposed extra explanatory premise is clearly incorrect. It is not always the case that the ability to establish, regarding each F , that it is G, engenders the belief that all F s are G. For example, suppose someone mistakenly suspects that there are planets within the orbit of Mercury. In that case he might nevertheless be able to show, of every planet, that its distance from the sun is not less than Mercury’s; but he does not believe the generalization that all the planets have this property. Obviously, he is not tempted to draw the generalization because, although in fact he can establish for each F that it is G, he does not appreciate this fact – indeed he would deny it. This suggests that a more plausible version of our extra premise would run along the following lines: Whenever someone can establish, for any F , that it is G, and recognizes that he can do this, then he will conclude that every F is G. It seems to me that this is more-or-less what we need to explain our acceptance of the generalization about truth. We can establish, for any proposition of the form , hp → pi, that it is true. Moreover, we can become aware of this ability. For we can see that any of the propositions in question may be designated by some expression of a certain form. And we can also come to see that there is a general strategy which, given any such expression, will establish the sentence that results from writing the expression before “is true”.20
158
PAUL HORWICH
Thus we have a plausible explanatory premise which, in conjunction with our endorsement of the equivalence schema, will enable us to explain the acceptance of generalizations about truth. And since that premise does not explicitly concern the truth predicate, the need for it does nothing to suggest that the basic regularity governing the truth predicate has to go beyond our underived commitment to the equivalence schema.21 Objection 7: Certain people (mostly philosophers) do not have a completely general inclination to endorse the equivalence schema. They might hold, for example that ethical pronouncements, contingent statements about the future, applications of vague predicates to borderline cases, or sentences containing empty names, fail to yield acceptable instances. But these people nevertheless mean the same thing as we do by the word “true”. After all, we might disagree with them on the question, “Can ethical pronouncements can be true?” – yet each of us expresses the issue in just that way. Consequently, it cannot be that to understand the English truth predicate one must have an inclination to accept all instances of the schema. (Richard22 ) Notice that the present objection is not based on the claim that certain instances of the equivalence schema are incorrect, or that they are unhesitatingly rejected by everyone.23 The point rather is that if someone has no inclination to accept a certain class of instances (despite understanding them) he might nonetheless understand the truth predicate as we do. And so we have to conclude, it would seem, that a general endorsement of the schema is not what provides that word with its meaning. But this conclusion is unwarranted. For we can invoke social externalism in order to accommodate our minimalist thesis to the facts under consideration. We might suppose that the tendency of some people to restrict the equivalence schema stems from confusion about the meaning of the truth predicate. We might suppose that their basic use of it does not quite match its meaning-constituting use; but that they are nonetheless correctly ascribed the standard concept of truth because they belong to a linguistic community in which that concept is the ‘right’ one to have – the one deployed by the appropriate ‘experts’. Moreover, a good case can be made that it is indeed those who restrict the schema, rather than those of us who do not, who are confused and mistaken. For they tend to be in the grip of the idea that truth is a substantive property, analysable in terms of ‘correspondence with facts’. They do not appreciate the real role of the truth predicate as a device of generalization – a role which requires that it be applicable to any proposition (i.e., to what is expressed by any
A DEFENSE OF MINIMALISM
159
declarative sentence). Thus one can say that the meaning-constituting fact about the truth predicate is the fact that explains the overall use of it by those who are not confused. And this fact is our allegiance to the fully general equivalence schema. Objection 8: But couldn’t we come across someone who denies that there is such a thing as truth – someone who is not prepared to accept any instances of the equivalence schema? Couldn’t such a person nonetheless understand our talk of truth, and nonetheless mean what we do by the word “true” when he says, for example, “Even though dogs bark, it is not true that they do”? I believe this is correct. Moreover, the social externalism of meaning does not suffice to reconcile minimalism with such a scenario. For our imagined sceptic’s use of the word “true” is not even approximately a matter of endorsing the equivalence schema. The moral would seem to be, rather, that there must be some use of “true” that (a) is implicit in, but weaker than, an endorsement of the equivalence schema, (b) is displayed by the sceptics and by ourselves, and (c) constitutes what we both mean by that word. This conclusion is also suggested by the plausibility of allowing – contrary to what we have been supposing so far – that our endorsement of the equivalence schema is not epistemologically fundamental. For it would seem to be the product of two deeper factors: first, there being some term, “f ”, such that all instances of an equivalence schema involving that term are accepted as basic (i.e., instances of ’hpi is f ↔ p’); and second, the conditional decision that, if there is such a term, it shall be the word “true”. Given this way of factorizing our endorsement of the equivalence schema, then it is plausible that although that endorsement is sufficient to fix the meaning of “true”, it is not necessary. Rather, all that is necessary is the second of these factors – the conditional commitment to express the concept which satisfies the schema, if any concept does, using the word “true”. In that case, someone might reject the antecedent of the conditional – he might deny that there is any concept satisfying the schema – and yet agree with us about what the truth predicate means.24 Thus our initial minimalist proposal must be revised. Meaning what we do by the truth-predicate is not constituted by an inclination to accept instances of the equivalence schema; but rather by the commitment to have that inclination, on condition that one is inclined, for some “f”, to endorse ‘hpi is f ↔ p’.
160
PAUL HORWICH
Objection 9: Truth is valuable: we ought to pursue it and we ought to avoid false belief. But these sentiments are not contained in (nor can they be extracted from) instances of ‘hpi is true ↔ p’, which are entirely nonnormative. Consequently, our concept of truth is not fully captured by the equivalence schema: so the minimalist proposal is false. (Dummett25 ) On the contrary, the equivalence schema does suffice to explain the normative force of truth. To see this, consider specific norms of belief such as One should believe that wombats fly ↔ wombats fly Clearly our commitments to norms like this one have nothing to do with the concept of truth; for that concept is completely absent from their articulation. Not of course that there is no call to explain why we adopt such commitments. The point is merely that one should not expect the concept of truth to be doing any of the explanatory work. Let us then imagine that all such specific norms of belief are somehow explained.26 Suppose, that is, we can account for our attachment to all norms of the form One should believe that p ↔ p Given the equivalence schema, we will then be able to explain our attachment to every norm of the form One should believe hpi ↔ hpi is true that is, to every norm of the form One should believe x ↔ x is true But this engenders (via the mechanism discussed in the response to Objection 6) a commitment to the generalization (x)(One should believe x ↔ x is true) – or, in English, to the principle One should believe what is true and only what is true27 Thus the value we attach to true belief is explained by the role of truth as a device of generalization – which is itself explained perfectly by the equivalence schema.
A DEFENSE OF MINIMALISM
161
The full minimalist picture of truth includes considerably more than the thesis I have been defending in this essay. It involves, besides the present claim about how the meaning of “true” is constituted, an affiliated view about the function of the truth predicate (namely – as just illustrated – that it is merely a device of generalization), an affiliated view about the underlying nature of truth (namely, that there is no such thing), and an affiliated view about the general shape of the basic theory that will best explain all the facts about truth (namely, that its postulates are instances of the equivalence schema). I have not attempted to elaborate or establish these further minimalist doctrines. However, since what I have been concerned with here is the central component of minimalism, my defense of that thesis, if successful, provides important support for the view as a whole.
ACKNOWLEDGEMENTS
I would like to thank Hartry Field and Michael Lynch for their comments on a draft of this paper.
NOTES 1 A note on the relationship between deflationism and minimalism about truth. Deflationism is the somewhat vague idea that truth is not a ‘substantive’ property, that no reductive theory of it should be anticipated, and that our grasp of the truth predicate comes from our appreciation that each statement specifies its own condition for being true. But philosophers who sympathize with this general point of view disagree amongst themselves about how best to elaborate it. Minimalism is one such strategy – the one defended here and previously articulated in my Truth (2nd edition, Oxford University Press, 1998). Besides minimalism, the main alternative forms of deflationism about truth are: (1) disquotationalism, according to which sentences (rather than propositions) should be regarded as the bearers of truth and the schema, “p” is true ↔p, will be what defines the truth predicate; (2) prosententialism, which stresses the use of “That is true” to save having to repeat what has just been asserted and which denies that “true” should be logically formalized as a predicate; (3) the redundancy theory, whereby “The proposition that p” means exactly the same as “p”; (4) the quantificational theory, which analyses truth-talk in terms of substitutional quantification into sentence positions: – ‘x is true’ means ‘(∃p)(x = hpi & p)’; and (5) Tarski’s theory, which explains the truth of each sentence of a language in terms of the referential properties of its components and the logical structure in which they are embedded. The relative advantages of the minimalist version of deflationism cannot be spelled out here. But, in a nutshell, its merits are (a) that it deals with our actual concept of truth, rather than some allegedly superior one; (b) that it does not attempt to explain truth in terms of notions that should themselves be explained in terms of truth (e.g. substitutional quantification); (c) it recognizes that there is no call for an explicit definition; (d) it can
162
PAUL HORWICH
countenance the attribution of truth to propositions whose logical forms we do not know; (e) by acknowledging truth as a property, it squares perfectly with its role as a device of generalization. 2 “hpi” abbreviates “the proposition that p”; and “↔” is the material biconditional. 3 This view of how meaning-constituting properties are to be identified is an instance of the general idea that an underlying property U constitutes a relatively superficial property S when U ’s being co-extensive with S explains why possession of S has the symptoms that it does. 4 I am inclined to say that a word’s meaning-property (e.g., ‘w means TRUE’) is constituted by (or reduces to, or may be analysed as) a certain way of using the word. But, in light of the relational character of meaning-properties, this thesis is in tension with the principle that any analysis of a complex property must result from the analysis of (at least one of) its constituents. One strategy here is to criticize that principle. Thus the FregeRussell reduction of ‘the number of dogs owned by x = 0’ to ‘−(∃y)(y is a dog owned by x)’ is a plausible counterexample. Another strategy is to retreat to a slightly weaker thesis. Rather than speaking of constitution (or reduction, or analysis), we might say that the use-property of a word engenders (or determines, or is the explanatory basis for) its corresponding meaning-property. 5 For justification of the claim that the equivalence schema is explanatorily fundamental, see my Truth (op. cit.). For justification of the use theory of meaning, see my Meaning, Oxford University Press, 1998. 6 To claim that “x is true ↔ x is P ” is the explicit definition of the truth predicate (where “P ” might be replaced with “in correspondance with reality”, “verifiable”, “useful”, etc.) is to claim that our acceptance of this principle is explanatorily fundamental with respect to our overall use of that predicate. But this claim is incompatible with the minimalist thesis according to which it is our endorsement the equivalence schema that is explanatorily basic. 7 I would argue that the axioms of the fundamental theory of truth – those that will provides the best explanation (i.e., simplest derivation) of all facts about truth – are instances of ‘hpi is true ↔ p’. For (a) such axioms would suffice (in conjunction with theories of other matters) to explain every other fact about truth; and (b) it is hard to imagine a simpler body of principles on the basis of which those instances could themselves be explained. For further discussion see Truth, pp. 25–31, 50–51 (op. cit.). 8 Davidson, D. “The Folly of Trying to Define Truth”, Journal of Philosophy 87 (1966) 267–78. Objection 1, with its focus on compositionality, seems to me to present Davidson’s strongest reason for concluding, contrary to minimalism, that truth is conceptually prior to meaning. However, he does not elaborate this point in any detail, but instead stresses various other arguments for that conclusion. First, he infers it from the fact that there are sentences (such as “That is red”) whose meaning-constituting, assertibility conditions are to accept those sentences only when they are true. But this reasoning presupposes (wrongly, in my view) that meaning-constituting regularities of use are explicitly known by the speakers of a language. Second, he infers it from the fact that someone’s acceptance of a sentence may be justified only if he believes that the sentence is true. But that fact is better explained by minimalism, together with the fact that one should only assert what one believes. And third, he urges a direction of explanation which goes from (a) facts about the circumstances that cause the acceptance of sentences, to (b) facts about their truth conditions, to (c) facts about their meanings. But there is a long-standing, notorious difficulty with that line of thought: namely, to articulate a conception of ‘truth condition’ that is strong enough. For, given a material construal of the conditional, “p” may very well
A DEFENSE OF MINIMALISM
163
be true iff q, without meaning that q. And stronger construals of “iff” merely make such counterexamples slightly harder to construct. 9 Davidson, D. ‘Truth and Meaning’, Synthese 17 (1967) 304–23. 10 For further discussion of this approach, see my “The Composition of Meanings”, Philosophical Review 106 (1997) 503–31, reprinted as chapter 7 of Meaning (op. cit.). In that work I defend the idea that the meaning-property of a complex expression is constituted by its being constructed in a certain way from words with certain meanings. This proposal is stronger than the Fregean principle deployed in the present essay; for it implies that two expressions have the same meaning, not only if, but also only if, they are similarly constructed. 11 See Meaning (op. cit.). A rationale for the use theory of meaning is sketched at the beginning of this paper, where I motivate the minimalist view of how the meaning of the truth predicate is constituted. 12 One might be tempted by a more holistic theory whereby truth and meaning (and proposition) are on the same conceptual level and are jointly explained in terms of one another and in terms of other matters. Such a theory would be less simple and less explanatory – hence, less attractive – than a more atomistic view such as the one proposed here, and should be embraced only as a last resort. 13 Davidson, D. “Truth and Meaning” and “The Folly of Trying to Define Truth”. 14 Strictly speaking, the Fregean approach that I am recommending does not explain the fact that meaning is compositional, but takes this fact, in its Fregean form, to be explanatorily fundamental. 15 Davidson is careful to emphasize that his critique of minimalism does not amount to the claim that there are no such things as propositions. Presumably he thinks that, if they exist, they must be designated by expressions of the form, “The proposition expressed by u” (where “u” refers to a sentence-token), rather than by expressions of the form “The proposition that p”. But even if this point were correct – and I have been arguing that it is not – the essence of the minimalist proposal would not be affected. For it can be reformulated in terms of the schema, “The proposition expressed by the following sentencetoken is true ↔ p”. We can suppose that this is what fixes the meaning of “true”. In addition to Davidson’s critique, which concerns the way that propositions are designated, there are several objections to the very existence of propositions – objections that a minimalist must be able to rebut. The main ones are (1) that propositions lack satisfactory identity conditions; (2) that false propositions do not exist (because any actual combination of objects and properties would amount to a fact; (3) that propositions are ontologically weird and explanatorily unnecessary. For discussion of these issues, see Truth (op. cit.). 16 Field, H.: 1992, ‘Critical Notice: Paul Horwich’s Truth’, Philosophy of Science 59, 321– 30. 17 Gupta, A.: 1993, ‘A Critique of Deflationism’, Philosophical Topics 21, 57–81; and his ‘Minimalism’, Philosophical Perspectives 7 (1993) 359–69. 18 Gupta, A. ‘A Critique of Deflationism’ (op. cit.) and ‘Minimalism’ (op. cit.) 19 Gupta, A. ‘A Critique of Deflationism’ (op. cit.) and ‘Minimalism’ (op. cit.). This objection has also been forcefully articulated by Scott Soames in his “The Truth About Deflationism”, E. Villanueva (ed.) Philosophical Issues 8, Atascadero, Cal., Ridgeview Publishing Company, 1997. See also “Some Remarks on Deflationism” (unpublished) by Paolo Casalegno (University of Milan). 20 One might well wonder whether all general facts about truth can be explained in this way; but I think that we have some reason to think that they can be. For it would seem
164
PAUL HORWICH
that any such fact could be put into the form: All propositions of type K have property J . For example: (1) Given any conjunction, if it is true then so are its conjuncts; (2) Given any proposition of the form hp → qi, if it and its antecedent are both true, then so is its consequent; (3) Given any atomic proposition, it is true if and only if its predicate is true of the referent of its subject; etc. Now, for any such generalization, if we can show, with the help of the equivalence schema, that it holds of an arbitrary proposition, we can then invoke the proposed additional premise to explain our acceptance of that generalization. Hartry Field (in his “Deflationist Views of Meaning and Content”, Mind 103, 1994, 249– 85) offers a solution to the ‘generalization’ problem, which supposes, roughly speaking, (1) that we can reason schematically – e.g., we can prove and assert “(p & (p → q)) → q” – and we then have the right to substitute sentences for schematic variables; (2) that we can introduce a rule allowing us to go from any schematic theorem, ‘hpi is K’, to the conclusion ‘All propositions are K’; and (3) that we assert the schema, “hpi is true ↔ p” (and not merely its instances). My proposal differs from this only in that, instead of clause (3), I have a claim (in effect) about the circumstances in which our disposition to believe all propositions of a certain form entitles us to assert the corresponding schema. This enables me to preserve the thesis of this paper – namely, that the meaning of “true” is fixed by our inclination to accept instances of the equivalence schema – rather than having to move to the claim, that it is our inclination to accept the schema itself that fixes the meaning of the truth predicate. ‘Assertion of a schema’ strikes me as too sophisticated an activity to be plausibly attributed to ordinary people. 21 A further objection of Anil Gupta’s – one that I do think is correct – is that our underived endorsement of the equivalence schema will not explain our confident acceptance of sentences like “Julius Caesar was not true”. To accommodate this point we can suppose that the explanatorily-basic, meaning-constituting facts about “true” include, not merely our underived allegiance to the equivalence schema, but also our underived acceptance of the principle, “Only propositions are true”. 22 Richard, M. “Deflating Truth”, E. Villanueva (ed.), Philosophical Issues 8, op. cit. 23 I would argue that the moral of the ‘liar’ paradoxes is that not all instances of the equivalence schema are correct. But I don’t believe that those who come to accept this moral, and who come to balk at certain instances, are thereby altering what they mean by the truth predicate. This is my motivation for supposing that the meaning-constituting fact about “true” is a mere inclination to accept any instance of the schema, rather than a disposition to accept any instance. In problematic cases the inclination will be over-ridden. But its continued existence is what sustains the sense of paradox. 24 N.B. the factoring of a scientific theory, T (f 1, . . . , f n), into its Ramsey sentence, (∃x1) . . . (∃xn)T (x1, . . . , xn) (which says that there exist properties that relate to one another and to observable facts just as the theory says) and the conditional, (∃x1) . . . (∃xn)T (x1, . . . , xn) ⇒ T (f 1, . . . , f n) (which says that if such there are such properties they are f 1, f 2, . . . , f n). It is plausibly just the second of these commitments that is needed to fix the meanings of the theoretical terms. For further discussion see my “Implicit Definition, Analytic Truth and Apriori Knowledge”, Nous, 1997, reprinted (slightly revised) as chapter 6 of Meaning (op. cit.); and my “Stipulation, Meaning, and Apriority” in New Essays on the A Priori, edited by P. Boghossian and C. Peacocke, Oxford University Press, 2000. 25 Dummett, M. “Truth”, Proceedings of the Aristotelian Society 59, (1958–59), 141–62.
A DEFENSE OF MINIMALISM
165
26 I would argue that the basis for our commitment to these specific norms is pragmatic:
we are more likely to get what we want if we abide by them. See chapter 8 of Meaning (op. cit.) and my “Norms of Truth and Meaning”, in What Is Truth, edited by Richard Schantz, Gruyter: Berlin and New York, 2001. 27 Clearly this is merely a first approximation of the proper norm. For one thing, there is nothing wrong with not bothering to investigate, and hence failing to believe, certain extremely trivial facts. Department of Philosophy University College London Gower Street London WC 1E 6BT E-mail:
[email protected]