The composition of meanings

Philosophical Review

Volumn 106, Issue 4, 1997, Pages 503-530

  Horwich, Paul ^a  

^a NONE

Author keywords

[No Author keywords available]

Indexed keywords

EID: 32844463968     PISSN: 00318108     EISSN: 15581470     Source Type: Journal    
DOI: 10.2307/2998510     Document Type: Article

References (22)

1. D. Davidson, "Truth and Meaning," in his Inquiries into Truth and Interpretation (Oxford: Oxford University Press, Clarendon Press, 1984).
2. Let me be more explicit about the route from my assumption about what it is to understand a complex expression to my conclusion about how its meaning is constituted. Suppose that understanding a complex expression of one's own language is (as I assume) identified with the state of understanding its parts and appreciating how they are put together. Then understanding 'dogs bark' (that is, knowing what it means) is constituted by knowing both what its parts mean and how they are combined. But (as we shall see in the response to Objection 8) these items of knowledge are implicit: someone knows (fully) what an expression means when what it means in his idiolect is the same as what it means in the community language. Therefore, the fact that 'dogs bark' means what it does (in language L) is constituted by the facts regarding how it is composed from primitives and what those primitives mean (in language L). Jim Higginbotham has argued (in conversation) that the mode of construction of a sentence may itself have a meaning that contributes to the meaning of the whole sentence, in which case our understanding the sentence would require more than simply understanding the words in it and seeing how they have been combined. It seems to me, however, that any so-called "method of combination" that intuitively has a meaning (for example, predication or conjunction or 'ns v) can be regarded as a schematic constituent of the sentence. Thus, we can make it a matter of stipulation that no meaning attaches to the procedures by which these and other constituents may be combined. Note, moreover, that even if Higgin- botham's suggestion were correct, this would not have a substantial effect on the deflationary position advanced here. Our constitution thesis would have to be revised slightly to say that the meaning of a sentence consists in its mode of construction having a certain meaning and its constituents having certain meanings. But the explanation of compositionality would then be no less trivial. And it would be equally clear that compositionality cannot constrain how the meaning properties of words are constituted.
3. As indicated, the general idea of "property S being constituted by property U" is a very familiar one. It obtains, roughly speaking, when S and U are co-extensional, and when facts about S are explained by this co-exten-sionality. I think it best not to equate the relation of property-constitution with that of property-identify', but nothing here hinges on preserving this distinction. For further discussion see the response to Objection 11.
4. Strictly speaking, meaning facts have the form, "S's expression k means E"-since one and the same sound type, k, may be meant differently by different people, depending on their language. However, since I am concerned with what is meant by the expressions of a single, arbitrarily selected speaker S, explicit reference to the speaker is suppressed for the sake of ease of exposition. Thus I write "'dog' means DOG' instead of "S's word 'dog' means DOG"; and I ask which property 'u(x)' constitutes the meaning property 'x means E', rather than asking which property 'u(S, x)' constitutes the meaning property 'S's expression x means E'.
5. One might suspect that dogs bark' means DOGS BARK cannot really be what we want to explain because, given our convention for naming meanings, it amounts to little more than 'dogs bark' means what 'dogs bark' means, which seems too obvious to be the item of knowledge that constitutes understanding. In order to assuage this concern, note the following points. First, although it is obvious, the fact at issue is nonetheless subject to substantive explanation; for we can show how 'dogs bark' comes to have the property that constitutes 'x means DOGS BARK'. Second, it is far from unusual to give substantive explanations of facts that exhibit the same appearance of triviality, for example, that the man elected was the man elected, that the color of blood is the color of blood, and so on Third, despite its obviousness, this may well be the fact about 'dogs bark' whose knowledge coincides with understanding that sentence. No doubt explicit knowledge of it is irrelevant-neither necessary nor sufficient for understanding; but we might suppose that when that fact holds of a given speaker (in virtue of his expression 'dogs bark' having the appropriate construction property), then it qualifies as implicitly known by that speaker, and it is such implicit knowledge that constitutes understanding. (See the response to Objection 8 for more on this point). And fourth, even the more familiar formulations of what needs to be explained-e.g. that 'dogs bark' expresses the proposition that dogs bark, or that 'dogs bark' is true if and only if dogs bark-focus on facts that are no less obvious.
6. This is because P(A and B / x) = P(A/x) + P(B/x) - P(A or B / x), that is, the probability of a conjunction (relative to any condition, x) depends not merely on the probabilities of its conjuncts but also on the probability of their disjunction. For example, let x = The car won't start, A = There is no petrol in the tank, and B = It has a broken starter; and suppose that P(A/x) = 0.8 and P(B/x) = 0.2. The problem is that these facts (which, on the proposed theory, help fix the meanings of 'A' and of 'B') don't determine the value of P(A and B / x) (and therefore don't enable us to determine the meaning of 'A and B'). "We would also need to know whether non-starting is, or is not, almost invariably the result either of A or of B. If it is, then P(A and B / x) is low; if not, then P(A and B / x) might well be high. But this further fact is not something provided by our knowledge of the meanings of 'A' and 'B'.
7. 'Jerry Fodor and Ernie Lepore have argued that the meaning of a term cannot be constituted from its stereotypical exemplars, because this would be inconsistent with the compositionality of meaning: for example, from knowledge of stereotypical pets and stereotypical fish, one cannot infer what would count as a stereotypical pet fish. But this line of thought is flawed, because it unjustifiably presumes that if the meanings of the primitive terms are given by stereotypes then so must the meanings of the complexes in which those terms appear. The alternative I am suggesting is that the meaning property of 'pet fish' might consist in some property of the form, 'x is the result of applying the schema 'a n to words whose stereotypes are SI and S2'. See their "The Pet Fish and The Red Herring: Why Concepts Can't Be Prototypes," Cognition 58 (1996): 243-276
8. For parallel arguments-questionable in the same way-against constituting word-meanings from conceptual roles or recognitional capacities, see their "Why Meaning (Probably) Isn't Conceptual Role," Mind and Language & (1991);
9. and see Fodor's "There Are No Recognitional Concepts; Not Even RED," in Philosophical Issues, vol.9, ed. E. Villanueva (Atascadero: Ridgeview Publishing, 1998).
10. A response to the last of these papers is given in my "Concept Constitution," in Villanueva, Philosophical Issues, vol.9.
11. Let me stress that we have no good reason to expect a uniform account. A superficially appealing rationale for one might derive from supposing that any reductive analysis of a meaning property whose form is 'x means F' must contain a reductive analysis of the constituent 'x means y', and concluding that there must be a general theory of meaning whose form is 'x means F = T(x, f)'-where T is independent of whether 'F is simple or complex. But this line of thought presupposes what I have elsewhere called the Constitution Fallacy: it wrongly assumes that reductive analysis preserves logical form-that is, that if a fact has a certain constituent, then any reductive analysis of that fact must contain either that constituent or something to which it reduces. This point is taken up again in the response to Objection 11. For more see my Meaning (Oxford: Oxford University Press, 1998) and my "Concept Constitution."
12. Thc difficulties to which I am responding in this section were impressed upon me by Stephen Schiffer. I would like to thank him for helping me to clarify and strengthen my position both here and in many other parts of this paper.
13. It might be felt that the meaning of a word can vary depending on the complex expression within which it occurs, and that consequently the meanings of these complexes are not explained via the meanings of their parts. For example, arguably, the meaning of 'sad' varies, depending on whether it is occurring in 'sad woman' or 'sad story'-and similarly for 'red' in 'red hair', 'red wine', 'red alert', etc., and for ' og' in 'big clog', 'wooden dog', etc. However, such examples do not really illustrate any failure of compositionality. Rather, they may be taken to reveal a holistic interdependence of the meanings of words-that is, that the meaning of 'sad' derives from a regularity of use that characterizes the various inferential roles of various complexes in which it appears. And this does not imply that this meaning depends on the meanings of those complexes. Alternatively, it may be conceded that the word 'sad' really is ambiguous. In that case we can also concede that the meaning of 'sad' in 'sad F may be inferred from the meaning of 'F on the basis of restrictions (either grammatical or pragmatic) on what 'sad' may sensibly be combined with. But again it would not follow that the meaning of 'sad' is explained by either the meaning of F or the meaning of 'sad F.
14. Davidson's position on the relationship between meanings and truth is not entirely clear. In some of his earlier writings he does appear to maintain that the truth condition of a sentence is what constitutes its meaning. This has the virtue of enabling one to see how Tarski-style derivations of the truth conditions of our sentences would provide an explanation of our capacity to understand them. However it turns out to be difficult to find a construal of "truth condition" in which that simple constitution thesis could be made plausible. (For, given a "material" reading of the conditional, the sentence 'snow is white' is true if and only if grass is green, but it does not mean GRASS IS GREEN). In later writings (and presumably in reaction to this difficulty) Davidson appears to retreat from this straightforward reduction of meaning properties to truth conditions. However, it is unclear (a) what exactly the weakened position is, and (b) whether anything less than the original, stronger position can help to explain, in the way Davidson wants, how our understanding of complex expressions depends upon our understanding of their parts.
15. These difficulties, briefly articulated in the previous footnote, are discussed by Martin Davies in his Meaning, Quantification, Necessity (London: Routledge and Kegan Paul, 1981), and also in chapter 3 of my Meaning.
16. In my Meaning, I argue (a) that the meaning-constituting property of a word is the characteristic that provides the explanatory basis of its overall use, and (b) that this characteristic is itself a regularity of use-something like "the fundamental law" governing the behavior of the word.
17. With respect to one's own idiolect, if an expression possesses a certain meaning then one qualifies as implicitly believing it has that meaning and (since error is impossible) as implicitly knowing it has that meaning. Thus meaning entails understanding.
18. In order to retain the possibility of finite axiomatization it is often allowed that the meaning theory deploy structural descriptions of sentences rather than their quote-names. Similarly one might drop the requirement that meaning properties be expressed in the form "x means E," and allow that they also be specified descriptively. Relative to these liberalizations it does appear to be possible to give a deflationary meaning theory that is finitely axiomatized. The central axiom would be The meaning property of the complex expression that results from applying the computational procedure P to the sequence of terms S = the property of "resulting from applying P to a sequence of terms whose meaning properties are those of the terms in S." From this we could deduce, for example, that The meaning property of the result of substituting 'dog' and 'bark' in 'tis v = the property of "resulting from the substitution of a word with the meaning property of'dog' and a word with the meaning property of'bark' into a schema with the meaning property of 'ns v." And given further premises specifying the meaning properties of the three constituents, this will entail The meaning property of the result of substituting 'dog' and 'bark' in 'tis v = the property of "resulting from the substitution of a word with the meaning property "x means DOG" and a word with the meaning property "x means BARK" into a schema witli the meaning property of "x means AS \"'," which in some sense specifies the meaning of 'dogs bark'. Thus a kind of first-order finitely axiomatized meaning theory is possible. But as I emphasize in the text, the possibility of providing such an account is of no great significance.
19. Though nothing in this paper depends on it, I argue in Meaning that meanings are concepts (the constituents of belief states), and that 'x means y' should be analyzed (roughly) as 'Occurrences of x provide reason to expect the presence of y'. Thus, "chien' means DOG" amounts to 'Occurrences of 'chien' provide reason to expect the presence (in the mind of the speaker) of the concept DOG'. This strategy goes against the popular Gricean idea that one must distinguish between the notion of "meaning" deployed within semantics and the one deployed elsewhere-for example, when we say, "Smoke means fire."
20. Note that a combinatorial procedure, P, acts on any sequence of functions and objects-it may say, for example, "Apply the 2nd member to the result of applying the 3rd member to the 1st member." Therefore P can perfectly well act both on a sequence of words (including schemata) and on the sequence of their meanings. The only requirement is that if a given expression can coherently be the argument of a given functional expression, then the meaning of the first' expression can coherently be an argument of the meaning of the functional expression.
21. Insofar as we are allowed to pick out sentences and meanings by means of descriptions, then the present approach yields a first-order finitely-axiom-atized meaning theory. For example, the Fregean axiom entails (The result of applying 'res v to 'dog' and 'bark') means (the result of applying the meaning of 'ns v to the meaning of 'dog' and the meaning of 'bark') which, together with the axioms specifying the meanings of '?is v, 'dog' and 'bark', entails (The result of applying 'hs v to 'dog' and 'bark') means (the result of applying AS I' to DOG and BARK) which would be the theorem specifying the meaning of 'dogs bark'.
22. S. Schiffer, "A Paradox of Meaning," Nous 28 (1994). In subsequent writings Schiffer argues (along lines that are similar to those developed here) that it is possible to give a trivial account of compositionality-what he calls a "pleonastic compositional meaning theory." See his "Meanings and Concepts" (forthcoming).