Horwich Contra Davidson:
A Critical Study
In Truth (1990), Horwich argued for deflationism, the view that, inter alia, truth is not apt to play a core role in a theory of meaning. Horwich had commended such a divorce without delivering a decree absolute. Now in Meaning (1998), Horwich attempts to show that most of the constraints which have guided recent philosophical theorising about meaning are thoroughly misplaced precisely because of their commitment to a core notion of truth. Horwich's principal and natural target is Donald Davidson's (1984) seminal idea that a theory of meaning for a natural language should take the form of a Tarski-style theory of truth. My brief in this critical study is not so much to rebut Horwich's arguments directly, but, for the most part, to defend Davidson against them. If my considerations are sound, then Horwich’s positive theory should not stir the Davidsonian from her ‘prejudices’, for the theory fails to impact against what she in fact believes.
2: Semantic Deflationism
Horwich (1998) proposes a use theory, in which truth does not play a central role. An explication of the relationship between meaning and use, of course, is a traditional constraint upon any candidate theory. Horwich's tack, however, is to take the relationship as explanatorily basic and seek to show that all the other constraints are either easily satisfied by the use theory, or otherwise spurious.
Horwich thinks that a theory of meaning should account for facts such as
(1) "dog" means DOG[]
That is, Horwich wants a theory of meaning-constituting properties; in our example, the property of the English word-type dog that constitutes the fact that it means DOG.[] Now Horwich argues that both traditional and contemporary theories make a "basic error" about the shape such reductive analyses must have (Horwich, 1998, p. 22-7). This constitution fallacy is the assumption that if a fact has a certain component, then any reductive analysis of the fact must either feature the component, or a further component which constitutes it. In terms of meaning facts, the fallacy takes the form of assuming that any reductive analysis must preserve the 2-place relational logical form exhibited in (1). In other words, it is assumed that a meaning analysis will have the general form (CF):
(CF) e means F = T(e, f ),
where ‘T ‘ is a non-semantic relation which holds between expression-types ‘e’ and concepts or properties ‘f ‘ (the 'meaning entity'). Satisfaction of the schema (CF) ensures that a meaning theory for a set expressions will be uniform in virtue of the common component ‘T’. Let us refer to any theory which is so uniform as inflationist. The fallacy of inflationism, according to Horwich, is the very assumption that there is some uniform property which underlies all meaning facts.
Horwich proposes that facts of language usage are optimally explained by laws governing linguistic usage, and such laws express the underlying properties that constitute meanings. The laws have the general form (L) (Horwich, 1998, p. 45):
(L) All uses of word w stem from its possession of acceptance property A(x).
It is the acceptance property A(x) that constitutes the meaning of w. Such properties are dispositions to accept a specifiable set of sentences (in specifiable circumstances) that feature w, and this disposition explains the overall pattern of a subject's use of sentences featuring w. In other words, the sentences specified by A(w) are explanatorily basic for the use of w. Now it is easy to see that this theory does not cleave to the general form stipulated by (CF). The meaning-constituting property of "dog", for instance, is some acceptance property u("dog"):
(2) "dog" means DOG = [is constituted by] u("dog").
The underlying fact - the fact that English speakers have a certain acceptance disposition toward a set of sentences featuring "dog" - does not include a component corresponding to the 2-place relation "x means y". There is therefore no entailment from the form of the theory to the presence of a uniform property underlying meaning facts; and the nature of use is such that there can be no such property. In particular, the theory makes no appeal to truth conditions. Indeed, far from depending upon a notion of truth, the theory grounds the deflationary analysis of truth offered in Horwich 1990. What is the constituting property of the fact that "true" means TRUE? Horwich's answer is given by (3) (Horwich, 1998, p. 45):
(3) All uses of "true" stem from the disposition of the community of English speakers C to accept instances of the schema (E) "the proposition that p is true iff p".
(3) also provides a nice example of what Horwich means by his claim that a set of sentences are explanatorily basic. The deflationist thesis, recall, is that the truth predicate is a device for disquotation. Alternatively put, the overall use of "true" is to be explained in terms of its disquotational function as enshrined in (E). A disposition to accept instances of (E), therefore, is explanatorily basic of "true" being a disquotational device, and this in turn explains the overall use of "true" in the community. (3) may be viewed as simply an encapsulation of this line of reasoning.
Horwich concludes that “truth is captured by the equivalence schema [(E)] and that the meaning of a word is engendered by its use”; a position he refers to as “semantic deflationism” (Horwich, 1998, p. 42).
3: Horwich on Davidson
In the way of Horwich's semantic deflationism is Davidson's (1984, 1990) seminal conception of a truth-theoretic account of meaning. Horwich, I shall argue, mistakes Davidson on a number of key points. In particular, these mistakes allow Horwich to present Davidson as a simpleminded target for deflation. Once we have a rectified picture of the relationship between truth, meaning and compositionality that Davidson proposes, we shall be in a position to sanguinely accept much of what Horwich takes to be radically deflationary.
from our deflationary perspective on truth, it is necessary to reject the Davidsonian analysis of sentence meanings as truth conditions (whereby, for example, 's means that dogs bark' is explicated as 's is true if and only if dogs bark'). Consequently, we must find an alternative to Davidson's widely accepted explanation, based on that analysis, of the compositionality of meaning (Horwich, 1998, p. 9) [my emphasis].
What I want to highlight here is Horwich’s reading Davidson as offering an analysis of meaning in terms of truth conditions. Horwich is assuming that Davidson seeks to reduce the property 'x having the meaning y' to a truth-theoretic property, i.e., Davidson is claiming that "the meaning of "snow is white" consists in the property of 'being true if and only if snow is white'" (Ibid, p. 71); that is, "the truth condition of a sentence constitutes its meaning" (Ibid, p. 73). From this assumption, Horwich takes Davidson's account of compositionality as equally strict, i.e., it is in some sense necessary that compositionality be explained in terms of truth and reference. On Davidson's view, Horwich avers, "compositionality dictates an explication of meaning properties in terms of reference and truth conditions" (Ibid, p.158). Or again: the "Davidsonian thesis [is] that compositionality can be explained only by explicating meanings as truth conditions" (Ibid, p.170) (my emphasis). And again, this time in explicit opposition: "Compositionality can be explained... without analysing [my emphasis] meaning in terms of reference and truth... one cannot insist [therefore] that the Davidsonian research programme defines the problem of compositionality" (Ibid, p.169).
To be fair, Horwich does claim that "Davidson appears to retreat from this straightforward reduction of meaning properties to truth conditions" (Ibid, p.169, fn. 11). However, where Davidson has retreated to is "not entirely clear" (op cit), it being "a more hedged and elusive position" (Ibid, p. 74, fn. 18).
Horwich, then, insofar as he understands Davidson, attributes to him two key theses. First, meanings are metaphysically constituted from truth-theoretic properties. On this view, Davidson's claims are meant to be taken as metaphysical or conceptual truths: for x to mean that p just is for x to be true iff p. That is to say, Davidson cannot be taken as offering a truth conditional theory as simply a fruitful approach which may ultimately prove inadequate. On the contrary, every non-truth conditional theory of meaning is somehow metaphysically mistaken about meaning properties. Second, following on from this analysis of meaning, as Horwich would have it, the very notion of compositionality "dictates" a truth conditional analysis; for since sentence-meanings just are truth conditions, if meanings are compositional, then they must be composed of reference conditions. Again, this is not a fruitful option among possible competitor explanations: the analysis of meanings "defines" the problem of compositionality as the problem of finding reference properties which compose to constitute the meaning properties of sentences.
Now I think that this reading of Davidson is mistaken in every detail. Indeed, the proposed interpretation appears not to be based upon anything Davidson has ever said about meaning, truth and compositionality. Before we read Davidson aright, let us first see how the metaphysical reading of Davidson makes him an easy target for Horwich's general position against meaning inflationism. I shall argue below that Davidson is essentially set-up to be refuted; it is important to see how the sting is effected. Two features of Horwich's account are here relevant, bearing on meaning and compositionality respectively.
Firstly, recall that Horwich, in part, diagnoses inflationism regarding meaning as based upon the constitution fallacy. Theories satisfying (CF) make for inflationism in that every meaning-fact will have a uniform underlying fact qua the presence of a uniform component T. Now since Davidson, according to Horwich, analyses meaning we should expect him to fall victim to the fallacy. And indeed he does, or so claims Horwich. He takes Davidson to be committed to (P):
(P) Predicate x means F = ("y)(x is true of y ↔ y exemplifies f-ness),[]
where, presumably, the property of '...being true of all and only those things that exemplify...' is what constitutes means (as it holds of predicate expressions), and so constitutes the uniform underlying nature of predicate meaning.
Horwich's Davidson, then, is already on the back foot. His analysis of meaning commits the common fallacy, as Horwich would have it, of attempting to find a substantive relation to constitute meaning relations.
The second key effect of Horwich's reading of Davidson is that it makes the compositionality constraint appear far too strong. Horwich assumes that a Davidsonian explanation of compositionality is bottom-up; that is, the ability of subjects to understand complex expressions is explained by "their attaching meanings to the (relatively few) basic terms of the language" (Ibid, p. 32). And such explanation "severely constrains what sought of thing a meaning property might reduce to" (op cit). As we saw, Horwich understands Davidson to be, in some sense, metaphysically committed to an explanation of compositionality in terms of truth-theoretic notions. Compositionality, therefore, dictates that meaning properties of words are truth-theoretic.
Now setting up Davidson in this way means that the dialectical situation is such that if Horwich can show that compositionality does not dictate the metaphysical status of meanings, then the Davidsonian project in toto is in ruins. Horwich has an account of compositionality whose brief is to do just that. Thus, “Davidson and his followers would appear to be quite mistaken in supposing that compositionality provides a compelling rationale for embracing the truth conditional account of meaning” (Ibid, pp. 35-6). Horwich's argument for this conclusion is beyond the scope of my brief.[] The point I am presently making is just that the metaphysical reading of Davidson makes him especially liable to the kind of refutation Horwich has in mind.
Horwich's reading of Davidson, then, is not innocent. Understood as a theory of the metaphysics of meaning, Davidson's account is arch-inflationist and is thus susceptible to the general complaints Horwich marshals against all such analyses. The dialectical convenience of this understanding is that what is genuinely novel in Davidson's approach can be glided over on the wings of a general renunciation of inflationism without pausing to observe just what is inflationist about the landscape below.
4: Davidson Aright
As already made patent, I think that Horwich's reading of Davidson is quite mistaken. What I shall do now is unpack this reading and demonstrate where it goes wrong. We shall also see why Davidson is not susceptible in the manner Horwich imagines to the two general deflationist complaints rehearsed just above. I shall organise my comments around a number of headings for ease of exposition.
(i) Reduction and the T-schema. Horwich understands a Davidsonian theory as a "straightforward reduction of meaning properties to truth conditions" (Horwich, 1998, p 169, fn., 11). Or rather, he does not know how else to understand it. But Davidson clearly does not understand the relationship between truth and meaning in this way. He writes: "Truth conditions are not to be equated with meanings; at best we can say that by giving the truth conditions of a sentence we give its meaning. But this claim too needs clarification" (1970, p. 56, fn. 3), the clarifications being that T-sentences can be said to give the meaning of their object-sentences only if viewed in the context of a truth theory as a whole which satisfies a host of constraints (see subsection (ii) below). Indeed, Horwich inconsistently appears to acknowledge this when he contrasts reductionist theories with Davidsonian ones, which "[remain] at the semantic level, aiming at a systematisation of familiar meaning facts in terms of theoretical semantic notions" (Horwich, 1998, p. 52 (also see p. 72)). This second thought of Horwich's must be correct, for, as we just saw, Davidson does not think that there are any such things as meanings, and so there is nothing to reduce. Moreover, the reduction of any intentional concept is anathema to Davidson: he rejects both behaviourism and physicalism (1973a; 1980). Of course, truth and reference are intentional concepts, not behavioural or physical ones; nevertheless, the relation Davidson proposes between these concepts and meaning is not one of constitution, analysis, or any kind of reduction. Suffice it to say that nowhere does Davidson speak of reducing meaning, or meaning being constituted from some other property.
Perhaps the reason Horwich imposes a reductionist reading on Davidson, in spite of Davidson explicitly rejecting meaning properties and nowhere talking about reduction, is because of Davidson's commitment to the T-schema as the theorem-form of a theory of meaning. Horwich appears to reason as follows. 'Davidson explicitly rejects deflationism, yet he is committed to the disquotational T-schema. Davidson, therefore, must understand the T-schema in some inflationist way, namely, he construes instances of (T) as specifications of the meaning constituting properties of the particular sentences described to the left of the biconditional'. As we saw above, Horwich certainly attributes such a construal to Davidson: he is explicit that particular instances of (T) give the meaning constituting property of the particular sentences they govern:
Davidson's approach requires that the truth condition of a sentence constitutes its meaning; and this is problematic since it is not the case that
[(TM)] u is true if and only if snow is white → u means that snow is white (Horwich, 1998, p. 73)
Horwich's reasoning here is wholly mistaken. I shall return to the significance of (TM) below. Horwich's first claim is our first concern. We have already seen that Davidson does not think that there is a relation of constitution between truth and meaning. We can also now see Horwich's crucial error: he assumes, because the Davidsonian T-schema cannot have a deflationary construal, that particular instances of (T) specify the meaning of the sentences mentioned.
The closest Davidson comes to endorsing the kind of account Horwich has in mind for him is his claim that since "the words 'is true if and only if' are invariant [in each instance of (T)], we may interpret them if we please as meaning 'means that'. So construed, a sample might then read '"Socrates is wise" means that Socrates is wise'". (Davidson (1970, p. 60)). Davidson appears to be identifying truth as an invariant underlying property that constitutes meaning-facts, hence the permission to interpret truth-facts as meaning-facts. Note, however, that Davidson is not even so much as inviting us to make such an interpretation, let alone obliging us. But surely, if we are here being presented with an analysis of meaning, we cannot "please" ourselves how we read the biconditionals? The reason Davidson does not commend the interpretation on offer is that "it encourage[s] certain errors" (Ibid, p. 61), precisely the errors Horwich falls foul of. What is wrong with the interpretation is that it naturally leads to the idea that all the theory can tell us "about the meaning of a particular sentence is contained in the biconditional[s] demanded by Convention T" (op cit). This error is compounded in the reading "wrongly [suggesting] that testing a theory of truth calls for direct insight into what each sentence means" (op cit). What Davidson is explicitly rejecting is the Horwich interpretation that each instance of (T) records the constitution of the meaning of the described sentence. This interpretation demands the kind of "direct insight" Davidson renounces, for what a sentence "s" means on this reading consists in the property of 'being true if and only if s', i.e., each instance of (T) effectively gives the meaning of its particular object sentence. Yet as Davidson says, "on reflection it is clear that a T-sentence does not give the meaning of the sentence it concerns" (1973a, p. 138).
(ii) Truth and Meaning. We saw above that Horwich thinks that the Davidsonian program faces a basic problem. If truth conditions constitute sentence meaning, then T-sentences give the meaning of their object sentences.
The problem the truth theorist putatively faces can be illustrated by considering the conditional schema (TM):
(TM) u is true if and only if p → u means that p
Horwich appears to assume that a truth conditional theory of meaning is committed to the truth of each instance of (TM). However, the antecedent of (TM) is the T-schema, and if we do not stipulate that instances of "p" will translate instances of "u", then meaning will not be preserved across the biconditional. This is because the equivalence relation is material: ‘p↔ q’ is true just if ‘p’ and ‘q’ share the same truth value. But we cannot stipulate a translation across the equivalence without forgoing the notion that a truth theory explicates meaning. Thus, as it stands, there is no constraint against false instances of (TM): e.g.,
(4) Snow is white is true iff grass is green → Snow is white means that grass is green
Horwich suggests that it is precisely this problem which led Davidson to move away from a "straightforward reduction" of meaning to truth towards a position Horwich finds "elusive" (Horwich, 1998, p. 74/168).
Now Horwich is certainly correct in thinking that a theory which seeks to show that meaning is constituted from truth conditions is prima facie refuted by the considerations rehearsed above. But, not only was Davidson aware of this problem from the beginning, the problem, contra Horwich's suggested diagnosis, is also in part why Davidson rejects the simpleminded view of T-sentences as recording the meaning constituting properties of sentences. Let us now see how this is so.
In "Truth and Meaning", where the truth theoretic approach to meaning is introduced, Davidson acknowledges the problem posed by (TM), but he does not accept (TM) at any point (1967, pp. 25-6). Rather, Davidson seeks to constrain truth theories - without appeal to translation or its cognates - in such a way that their output will be interpretative, and (TM) patently provides no hint how this might be achieved. A meaning theory, then, does not need to satisfy (TM), but instead something along the lines of (CTM):
(CTM) (C(T) ├ x is true ↔ p) → (p interprets x),
where "C(T)" denotes a truth theory that satisfies a collection of formal and empirical constraints and ‘interprets’ means that ‘p’ may express the content of utterances of ‘x’. In "Truth and Meaning" and other papers of the period Davidson thought that simply placing the demand that T-sentences be canonically derived from the referential axioms of the theory was sufficient to rule out non-interpretative cases. Such derived T-sentences determine the meaning of the described sentence "not by pretending synonymy [viz. something like (TM)] but by adding one more brush-stroke to the picture which, taken as a whole, tells what there is to know of the meaning of" the described sentence (Davidson, 1967, p. 26). The idea is that an adequate theory must entail every T-sentence - the whole picture, as it were - from a finite set of axioms, with each derivation - brush-stroke - thus revealing the compositional structure of the described sentences in terms of the semantic values of their constituent words. Each T-sentence thus places its object-sentence in a semantic location relative to every other sentence, with each location determined by the compositional pattern of words occurring throughout the language. This constraint does not in any way stipulate against non-interpretive theorems; the idea rather, is that a theory that entailed such theorems would have non-interpretative axioms and so would be one which entailed many more non-interpretative T-sentences. It would, therefore, actually deliver false T-sentences if we were thinking of it as a theory of English.
Davidson did indeed change his mind about this constraint, but not for the reasons Horwich claims. We have just seen that Davidson never thought that T-sentences alone give meaning, and so it was not dissatisfaction with any such constitution thesis that led to the shift of position. Thus, in "Radical Interpretation" Davidson claims that if we constrain the totality of T-sentences entailed by a theory to "optimally fit the evidence" about the sentences speakers regard as true, then the T-sentences will be interpretative (1973a, p. 139). Again, it is the totality of derived T-sentences and now the attendant evidence for them which makes a particular T-sentence interpretative, it is not any metaphysical relation of constitution. In more recent years Davidson has appealed to the notion that T-sentences should be construed as laws, and so be supportive of counterfactuals relating sentence-utterances with environmental conditions of utterances (such pairs of utterances and conditions of utterance constitutes the evidential core of the theory); the idea is that non-interpretative T-sentences will not support the correct counterfactuals which govern a subject's speech behaviour (Davidson, 1984, Introduction, p. xiv; 1990, p. 313).
Now I am not here concerned with the effectiveness of any of these options, or others which have been forwarded. The crucial point which counts against Horwich is that Davidson has never been concerned with the truth of the instances of (TM), because he has never claimed that T-sentences give the meaning of the object-sentences, still less specify meaning-constituting properties. Where Davidson and others have changed their minds is on the appropriate constraints, C, required to satisfy (CTM), i.e., the issue is about what kind of truth theory would produce interpretative T-sentences.
5: Horwich's Sting
In this section I shall discuss Horwich's sting, viz. the accusation that Davidson commits the constitution fallacy, and the attribution to Davidson of the thesis that compositionality is, in some sense, necessarily explained in truth theoretic terms. Horwich attempts to use these claims to refute the Davidsonian program, but Davidson is neither guilty of the fallacy nor committed to compositionality in the strong sense imagined by Horwich. Given the material already amassed, it will be relatively straightforward to demonstrate this.
(i) The Constitution Fallacy. Horwich, it will be recalled, considers a prime motive for inflationism to be a pseudo-constraint upon the kind of facts which could constitute meaning-facts. Horwich labels this constraint the constitution fallacy. In regard to meaning, the fallacy is to assume that facts of the form 'x means y' must be constituted from facts made up of the same or corresponding components. Since the means relation is a constant in all meaning-facts, every constituting fact must share some uniform corresponding component; it is such uniformity which makes for inflationism. As we saw, Horwich's use theory eschews uniformity: there are still underlying facts, but there is no essence uniting all such facts, so it is not the job of a theory of meaning to unearth some such common component.
Theories which commit the fallacy are said to instantiate the general schema (CF):
(CF) x means F = T(x, f )
Here T is the constant component corresponding to 'means' and f is the meaning property expressed by F. Does a Davidsonian truth theory instantiate (CF)? Horwich (1998, p. 23) thinks it does. In regard to predicates, he takes Davidson's general claim to be (P):
(P) Predicate x means F = ("y)(x is true of y ↔ y exemplifies f-ness),
where the underlying uniform property instantiating T is '...being true of all and only those things that exemplify...'. We have already seen much that is wrong with this attribution. Davidson rejects any constitution relation (=); thus, the very relational form of (P) is inaccurate. The relation between meaning and a truth theory is one of rational re-description that cannot be expressed in the simple terms of (P), even if it were otherwise correct. Further, Davidson is not committed to meanings, and so the left-hand side of the relation plays no part in a truth conditional theory. As such, 'f-ness' is an ontology which occurs nowhere in a Davidsonian theory; nor does exemplification. Indeed, in (P), exemplifies operates as an analysis of true of, but for Davidson such truth theoretic concepts are primitives, and so receive no analysis, and certainly not in terms of property exemplification. I shall return to this issue directly. Note first that from a Davidsonian perspective, (P) is simply false. The instances of (P) would determine the meaning of each predicate independently of other predicates, and independently of their semantic role in the sentences of the language. But Davidsonian truth theories are holistic not building block accounts, i.e., the theory interprets a word in terms of its systematic contribution to the truth conditions of the sentences in which it features; and sentences are interpreted not individually but as a totality related one to another in terms of the co-occurrence of semantically significant constituents. In other words, evidence for the interpretation of a given utterance u is based upon discerning the structure exhibited in u such that the occurrence of components of that structure in other utterances brings to bear greater evidence to sharpen the given interpretation. (P) is precisely what Davidson rejects!
Other serious problems beset the attribution of (P) to Davidson. As noted, the right-hand side of (P) is wrongly stated in that it includes an ontology of properties - f-ness - and the relation of exemplification required to relate objects to the properties they instantiate. Thus, (5) plays no role in a truth theory:
(5) ("y)(x is true of y ↔ y exemplifies f-ness)
The general form of predicate axioms Davidson commends, following Tarski, is:
(AP) ("y)(F is true of y ↔ Fy)
(where 'F' is a schematic place-holder for descriptions of predicates and 'F' is a place-holder for the predicates described). An instance of (AP) is:
(6) ("y)('is white' is true of y ↔ y is white)
We see then, that the axioms Davidson favours are the very same disquotational axioms Horwich (1990, Chp. 7; 1998, Chp. 5) appeals to in his account of reference. Whither (P)?
Horwich's appeal to the grotesque (P) is perhaps explained by his thinking that Davidson must have something like it in mind because his theory is intended to explicate meaning, not rely upon it. Recall that I suggested such a reading regarding Horwich's thoughts on Davidson's use of (T). The present thought is equally wrong-headed as the earlier one. All a Davidsonian theory (for English) states about 'is white' is contained in (6), but the meaning of the predicate is not captured therein; rather, it is the role of the axiom in exhibiting the compositional structure of sentences containing 'is white', and the evidential support for the T-sentences governing those sentences, that makes the axiom fit to interpret the predicate. Again, this shows that a simple equation along the lines of (P) has nothing whatsoever to do with the Davidsonian program. How, then, does the real Davidson fall prey to the constitution fallacy?
The answer is that he does not. As well as (AP), a truth theory also contains singular term axioms (again adopting Horwich's notation):
(AST) ("y)(n refers to y ↔ y = n)
And the theorems are the familiar instances of (T):
(T) X is true ↔ p
(where 'X' is substituted for sentence descriptions, and 'p' is substituted for the sentences described). All these schemata are non-uniform in that there is no component which occurs on the right of the biconditional in each instance of the given schema. Indeed, they are non-uniform in precisely the manner commended by Horwich in his accounts of truth and reference. Of course, it is only homophonic theories that exhibit disquotational axioms and theorems, but heterophonic theories are no more uniform. Their resistance to disquotation is based on the metalanguage not containing the objectlanguage, to use Tarski's parlance, not on their harbouring any uniform component absent from their homophonic siblings. If, then, we drop the spurious idea that a truth theory is committed to analysing or reducing meaning in a form answerable to (CF), a Davidsonian theory will be victim to the constitution fallacy to the exact extent that Horwich's deflationary theory is. This conclusion should not be surprising. The constitution fallacy can only be committed by those who analyse meaning; since Davidson does not, he is free from the fallacy.
This conclusion is of the utmost importance to understand where Horwich goes wrong. It is the assumption that Davidson attempts to reduce meaning to some more basic property that makes him liable to Horwich's general impugnation against the assumptions which drive such reductions into uniformity. Yet to accept this general disavowal of uniform analysis is in part to be Davidsonian about meaning! Horwich sets-up an exclusive differentiation between inflationism and deflationism in terms of whether a theory is committed to uniform or non-uniform analysis respectively. On this basis he is obliged to find Davidson committed to uniformity - inflationism - because Davidson is not a deflationist, i.e., he does not think that the significance of truth is exhausted by disquotation or some similar function: a theory of truth can explicate meaning, not the reverse as Horwich claims. Yet the accusation cannot seriously be premised upon anything Davidson says; everything he says points to the contrary. Davidson, it seems, appears to Horwich as an inflationist just because he is not a deflationist. The resolution of this confusion is simple. The deflationism-inflationism differentiation is not exclusive. We may assume that it cleaves in two the set of potential analytical theories, but the distinction makes very little sense with reference to Davidson's account qua non-analytical. On the one hand, the theory is not deflationist because it seeks to explicate meaning via truth. On the other hand, the theory is not inflationist because its axioms and theorems are non-uniform and, in general, it seeks no analysis or reduction of meaning, still less of truth and reference. It is unsurprising that Davidson's program should cause Horwich the problems it does, for it resides in a place Horwich believes not to exist.
(ii) Compositionality and Analysis. Insofar as I can tell, Horwich attributes to Davidson the following reasoning. 'Whatever meaning is, it is compositional, i.e., the meanings of primitive expressions must compose into meaningful complexes (given syntax). The only way of explaining compositionality is via a Tarski-style truth theory. Meaning, therefore, must be explained in terms of the composition of reference conditions (for singular terms) and satisfaction conditions (for predicates and recursive constants) into truth conditions'. Horwich, though, thinks that "compositionality per se provides absolutely no constraint upon... the underlying nature of meaning" (1998, p. 158). Horwich thinks this because he has an account of compositionality under which nigh any account of meaning is consistent with the idea that meaning complexes are composed out of their primitive constituents, i.e., whatever one's view about word-meaning, the composition of words into meaningful complexes is trivial. What I want to do here is question Horwich's attribution to Davidson of a strong conceptual connection between compositionality and a truth theoretic account of meaning.
Horwich's claim that about the Davidsonian attitude to compositionality can be read in one of two ways. A strong reading says that the very notion of compositionality necessitates (in some sense) a truth theoretic account; a weaker reading says that if one favours a truth conditional approach to meaning, then compositionality is defined via a Tarski-style truth theory. Horwich, I think, must have the strong reading in mind, for, as we shall see, the reasoning supporting the weaker reading is acceptable to Horwich. Either way, Davidson accepts neither claim, or at least not in any sense required by Horwich. Let us first consider the strong reading under which compositionality necessitates the Davidsonian approach.
We have seen that Davidson is not in the business of analysing or reducing meaning. We should not expect, therefore, Davidson to make any claims that compositionality can "only" be explained truth theoretically, or that such a theory "defines the problem of compositionality"; and indeed, no such claims are made by Davidson.
Davidson continually emphasises that the theories of meaning he proposes are empirical theories which are to be verified by evidence of speakers situated linguistic behaviour (e.g., see Davidson (1967, p. 24)). The theories will be verified if they enable a possessor of the theory to interpret the speakers, i.e., to make sense of their actions and goals through the attribution of propositional attitudes whose content is provided by the theory. In this sense, then, it is a hypothesis that a Tarski-style truth theory satisfying the constraints already detailed can serve as a theory of meaning. For example, a theory that delivers the T-sentence
(7) Der Schnee is weiss is true iff snow is white
must be tested against a speaker's actual linguistic behaviour in real situations. The theory does not simply claim or stipulate that "Der Schnee is weiss" has the same meaning as "Snow is white". We can, speaking outside of the theory, make this claim only if the T-sentence proves to be explanatory of a speaker's actions and utterances. Now as explained earlier, it is only sentential utterances that are tested, because only such complete utterances are 'observable' to an interpreter; words are observable only in the context of utterances. The axioms detailing the referential conditions for singular terms and predicates are theoretical constructions posited to account for the semantic pattern found at the sentential level. Such an account is afforded by the theory (via its deductive structure) depicting each utterance as a structure formed from a finite stock of words that make a uniform semantic contribution throughout the totality of potential sentences used by a speaker. Thus, the axioms count as correct just if they "yield a satisfactory account of the use of sentences" (Davidson, 1990, p. 300). In other words, "the theory gives up reference... as part of the cost of going empirical" (Davidson, 1977, p. 223).
Davidson's theory, then, qua empirical, appears to have no room for the kind of strong conceptual connection between compositionality and truth Horwich imagines for it. This becomes clear when we consider the direction of explanation Davidson has in mind. What we learn from a verified theory is the correctness of its T-sentences, and this provides us with an interpretation of the object language. But the T-sentences themselves carry no implication to referential axioms or compositionality.
This last point may be clarified by reference to Tarski (1956). A theory of meaning counts as correct if it delivers an interpretative T-sentence for each potential utterance, just as Tarski's definition of truth is materially adequate if it entails an instance of (T) for each sentence of the objectlanguage (Tarski's Convention T). The respective consequence classes are criterial of adequacy, but neither theory makes any demand as to how their respective criteria are to be met. How the respective criteria are satisfied is determined by independent constraints on what would count, respectively, as an adequate theory of meaning and an adequate definition of a truth predicate. For Davidson, this means that T-sentences should be entailed in such a way as to reveal the compositional structure of the language. For Tarski (1956), the demand is that the definition is consistent and finite (lest the metatheory contains an infinitely long formula). Davidson's key idea, of course, is that these constraints segue: the meaning theorist can exploit Tarski's recursive machinery of satisfaction to cleave to compositionality. Such a happy outcome, though, is in no sense entailed by the criterion of successful interpretation embodied in the form of Convention T Davidson adopts: one can accept Convention T but neglect compositionality. As such, contrary to Horwich's claims, by Davidson's lights, to accept that meaning is compositional does not in itself involve a commitment to a truth theoretical explication of meaning. Davidson is clear on this: "a theory of truth" offers an explanation of "how 'the meaning of each sentence depends on the meaning of the words'"; yet "there is no reason not to welcome alternative readings if they are equally clear". Even so, "a theory of truth does without [meanings]; [and] this should be counted in its favour, at least until someone gives a coherent and satisfactory theory of meaning that employs meanings" (Davidson, 1970, p. 61).
What Davidson insists upon is that a theory of meaning, whether in terms of truth conditions or not, must be compositional, but this insistence does not issue from the very nature of the theory's core concepts; the idea that a theory of meaning should produce an interpretative T-sentence for each potential utterance in no way constrains just how this flood is to be revelatory of compositional structure.
In sum, then, the idea that a truth theory can serve as a theory of meaning is an empirical hypothesis, and the constraint that a theory of meaning be compositional is a general one that Tarski showed how a truth theory may satisfy. This greatly counts in the theory's favour, but the constraint is in no sense constitutive of the demand for interpretative T-sentences, and that it is satisfied is what counts, not precisely how it is.
Davidson, therefore, explicitly contradicts the idea that compositionality demands any particular explanation. A truth theory offers a good one, but "alternative" ones are acceptable, if coherent; compositionality does not "dictate an explanation of meaning properties in terms of reference and truth conditions" (Horwich, 1998, p. 158); nor, as such, does compositionality constrain an explanation of meaning in terms of a truth theory. There is of course nothing surprising here. We saw above that Davidson thinks of compositionality as a general constraint upon a theory of meaning, and this is simply because for all the world linguistic understanding appears to be compositional. No theory can lay claim to compositionality as such unless it presents itself as the correct analysis of meaning. But this is not Davidson's ambition. His ambition is to give an empirically supported account of the content and structure of linguistic understanding (at the intentional level). It would, therefore, be absurd conceit of Davidson to claim that his theory "defines" compositionality.
We turn now to the second, weaker reading of Horwich's claims about Davidson on compositionality. On this reading, Davidson's claim is conditional: if meaning is accounted for in truth conditional terms, then a Tarski-style theory will define the problem of compositionality - "only" such a theory will do. I earlier averred that Horwich does not read Davidson in this way; his remarks indicate the stronger reading. Moreover, if we read defines loosely, the above reasoning falls under a quite uncontentious dialectic. Ask a philosopher: 'What is your account of meaning?' The philosopher replies: 'I account for meaning in terms of X'. Assuming, as most philosophers do, including Horwich, that meaning is compositional, the philosopher is beholden to give a coherent answer to the next question: 'How do Xs compose?' Davidson answers the two questions with 'truth conditions' and 'a Tarski-style truth theory' respectively. Horwich's answers are 'laws governing use' and 'via procedures to compose uses'. Jonathan Bennett's (1976) answers are 'speaker's intentions' and 'conventions governing the expressiveness of words'. And so on for Fodor (1987), Peacocke (1992), Barwise and Perry (1982), Smolensky (1988), et al. What all such pairs of answers reflect is that compositionality is a general phenomenon that an account of meaning must explain. Furthermore, any of these philosophers (and linguists and psychologists) are entitled to construe compositionality as defined in terms of their core explanatory concepts. Yet such a construal will trivially be modulo their theoretical commitments. Bennett can think of ‘compositionality’ just as the problem of segmenting the intentions behind an utterance so that the constituent words are associated with some convention governing their significance. Likewise, Davidson can construe ‘compositionality’ just as the problem of providing an axiomatic Tarski-style semantics for natural languages. It does not follow that either Bennett or Davidson are committed to the view that compositionality per se dictates the respective explanations they offer.
Since, I presume, Horwich accepts the basic demand behind the dialectic, he must have something different in mind for Davidson. As said, he appears to think that Davidson lays claim to compositionality in the stronger sense, as if any non-truth theoretic explanation would be somehow conceptually confused. Nevertheless, keeping with the present conditional reading, we must understand defines in a much stricter sense than that just rehearsed if Horwich's complaints against Davidson are not to collapse. The problem for Horwich, however, is that Davidson's work offers no stronger reading of the connection between truth and compositionality than that expressed in the uncontentious dialectic. Indeed, following the letter of Davidson's remarks, the connection is even weaker than that proposed by many on our list of philosophers.
We saw above that even given an acceptance of a truth conditional account of sentence meaning, Davidson does not insist upon a Tarski-style axiomatic structure to entail T-sentences. The truth theory, of course, must present its T-sentences in such a way that compositional structure is revealed, but this in itself carries no implication as to how such revelation should be achieved. The reason for this is that the relations of singular term reference and predicate satisfaction are theoretical constructions that are not open to empirical confirmation; their job will be successfully done if they give rise to interpretative T-sentences. Thus, in diametric opposition to Horwich's reading, Davidson's account of compositionality imposes no constraint on the actual content - the 'meaning properties', as it were - of the axiomatic structure of the theory; compositionality merely demands that there be some such structure.
Now this view of the connection between compositionality and the truth theoretic concepts is very weak indeed. Fodor (1987; 1998), for example, has a much stronger conception under which, it seems, only a denotational semantics (essentially an extensional referential semantics) is consistent with compositionality. Fodor argues at length that the kind of conceptual role or use theories favoured by many (including Horwich) are constitutively incapable of explaining the compositionality of complex representations because conceptual roles or uses just do not compose. Davidson's position enables him to agree with Fodor that conceptual roles do not compose. Equally, however, he is free to think that conceptual roles do compose, but that such theories are untenable on independent grounds. For example, Davidson (1973a, 1982) rejects the intention based semantics advanced by Grice (1989), Bennett (1976) and others. His principal reason for doing so, though, is not that intentions do not compose; the complaint, rather, is that the contents of intentions are not identifiable independently of linguistic content, so intentions are not explanatory of linguistic meaning. There may or may not be fundamental problems with the very idea of compositional intentions, but Davidson is not committed to the existence of such problems simply because he commends a truth theoretic semantics.
What we see here is that any argument that compositionality may be explained in non-truth theoretic terms is not a refutation per se of the Davidsonian program; it cannot be, for compositionality is a general constraint any theory must respect. The acceptance, therefore, that other theories can reflect compositionality too merely signals the thought that the truth theory is not the only possible game in town. If this is so, as it no doubt is, then our decision on what the best theory is will be a matter of the usual constraints pertaining to any theory evaluation: evidential corroboration, economy, simplicity, consistency with other commitments (in, say, linguistics, cognitive theory, etc.), and so on. This, of course, was always the situation we were faced with, for Davidson has never claimed that the theory's satisfaction of some one constraint, be it finiteness, completeness, or compositionality, is such as to show the theory to be metaphysically correct.
Notwithstanding my catholic construal of Davidson, we should not loose sight of his commitment to a Tarski-style account of compositionality. He clearly does believe, contra Horwich, that compositionality is a substantive constraint on candidate theories of meaning, and that a truth theory is a viable candidate because, in part, of its pellucid treatment of compositionality. Moreover, Davidson no doubt does think his theory offers the best account, but such an attitude is not only trivial - if he did not think that, then he would not proffer the theory he does - it is also perfectly consistent with compositionality being a general constraint, and so not defined in terms of his truth conditional conception.
6: Concluding Remarks
I have only been concerned with correcting Horwich's reading of Davidson. Still, even if Horwich is as mistaken as I claim, Davidson is still not off the deflationist hook; for Horwich contends that compositionality offers no constraint at all: nigh any account of meaning cleaves to compositionality. If this is so, then a major reason for favouring a truth conditional account is removed. It would be wrong to think, though, as Horwich does, that the Davidsonian program would be in ruins. Since it is no part of the program to argue that compositionality can only be explained in truth theoretic terms, it is open to a Davidsonian, in the face of the deflation of compositionality, to claim that a truth theory still offers the best all round account of meaning. As explained above, even if compositionality is not uniquely explained truth theoretically (or is trivial), there might still be good reason to think that a truth theory is the best account on offer. For instance, if one holds that a theory of meaning is firstly a theory of the semantic interpretation of syntactic structure, then it might be that a compositional truth theory furnishes the best interpretation of that structure.[] We need not, however, concede compositionality to the deflationist. But that is a subject for another time.[]
Horwich's deflation of compositionality (1998, Chp. 7) is based upon the idea of a construction property. In general, the meaning analysis of a complex expression is this: e means E = e results from the application of procedure P to <W1, W2,..., Wn>. Where 'e' is a complex expression, 'P' is a syntactic schema, and the 'W's are word meanings (e.g., acceptance properties): the use properties of the constituent words of 'e'. Since this construction property can be stated in general terms, Horwich's idea is that it leaves entirely open the analysis of word meaning, i.e., what one takes the 'W's to be. Thus, if construction properties explain the meaning of complexes in terms of the meaning of their parts, and they leave open the nature of meaning entities, then compositionality imposes no constraint upon the nature of meaning entities. Nothing I say below hangs on the falsity of this account. Nevertheless, I think that it is problematic (see Author 1).
Barwise, J., and Perry, J. (1983). Situations and Attitudes.
Bennett, J. (1976). Linguistic Behaviour.
Davidson, D. (1967). 'Truth and Meaning.' Synthése, 17: 304-23. All page references to Davidson (1984): 17-36.
Davidson, D. (1970). 'Semantics for Natural
Languages.' In B. Visentini et al.
(ed.), Linguaggi nella Societá e nella Tecnica.
Davidson, D. (1973a). 'Radical Interpretation.' Dialectica, 27: 313-28. All page references to Davidson (1884): 125-39.
Davidson, D. (1973b). 'In Defence of Convention
T.' In H. Leblanc (ed.) Syntax, Truth
and Modality (pp.
Davidson, D. (1980). Essays on Actions and Events.
Davidson, D. (1982). "Communication and Convention." In D. Davidson (1984): 265-280.
Davidson, D. (1984). Inquiries into Truth and Interpretation.
Davidson, D. (1990). 'The Structure and Content of Truth.' Journal of Philosophy, 87: 279-328.
Fodor, J. (1987). Psychosemantics: The Problem of Meaning in the Philosophy of Mind.
Fodor, J. (1998). Concepts: Where Cognitive Science has Gone Wrong.
Grice, P. (1989). Studies in the Way of Words.
Higginbotham, J. (1985). 'On Semantics'. Linguistic Inquiry, 16: 547-593.
Higginbotham, J. (1986). 'Linguistic Theory and
Davidson's Program in Semantics.' In E. Lepore (ed.), Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson (pp. 29-48).
Horwich, P. (1990). Truth.
Horwich, P. (1998). Meaning.
Peacocke, C. (1992). A Study of Concepts.
Smolensky, P. (1988). 'On the Proper Treatment of Connectionism.' Brain and Behavioural Sciences, 11: 1-23