SCOTT SOAMES
THE NECESSITY
ARGUMENT
Jerrold Katz and Paul Postal (this issue) present three positive arguments for their 'Realist', non-psychological conception of linguistics. One of these, the Necessity Argument, is based on considerations that are special to semantics. Unfortunately, the version of the argument given in the text is flawed, and can be quite misleading. Although there is a modified version of the argument for which a qualified measure of success can be claimed, its significance for advancing 'Realism' and undermining 'Conceptualism' seems to me to be highly restricted. Moreover, getting clear about what is, and what is not, established by this argument is useful in illustrating that there are defensible conceptions of linguistics which do not fit neatly into either the category of Chomskian Conceptualism or that of Katz-Postal Realism. The Necessity Argument given in the text by Katz and Postal is based on the observation that a semantic theory for a natural language must issue in claims of the sort illustrated by (1) and (2). (1)
'S' is analytic (i.e., true in virtue of meaning) in L.
(2)
'P' entails 'Q' in L.
The important thing about these claims is that they involve necessity. Since, for Katz and Postal, analytic truths are necessary, truths of the form (1) establish that certain natural language sentences are necessary. 1 Similarly, truths of the form (2) establish necessary connections between natural language sentences. But if an adequate semantic theory must issue in necessary truths, then the claims made by such a theory cannot be contingent. By contrast, psychological claims about the structure, organization, and operation of the human mind, or any subsystem thereof, are thoroughly empirical, and hence contingent. Thus, semantic theories for natural languages cannot be psychological in nature. Since linguistic theories (grammars) of natural languages must contain semantic components, 1 When indexicals are taken into consideration, I believe, following David Kaplan, that analytic sentences are those whose meaning guarantees that they express truths - i.e., they are sentences that express truths in every (proper) context of utterance. On this conception a sentence can be analytic without being necessary. (Similar remarks hold for entailment.) However, in this paper I will ignore indexicals, and follow Katz and Postal in taking analytic sentences to be necessary.
Linguistics and Philosophy 14: 575-580, 1991. © 1991 Kluwer Academic Publishers. Printed in the Netherlands.
576
SCOTT
SOAMES
it follows that Chomsky's 'conceptualist' view of linguistic theories is incorrect, and must be replaced by one which recognizes the abstract and necessary character of some linguistically significant facts. It is important to note that this argument cannot be accepted as it stands. The reason it can't is that it depends on a crucial assumption that has not been justified. The argument assumes that if a sentence s expresses a necessary truth, then the claim that s is necessary is itself necessary. But this is by no means obvious. For example, it is a commonplace in modal semantics to emphasize the compatibility of the following two claims: (i) s is necessarily true - i.e., s expresses something which is true with respect to every possible world; (ii) s could have expressed something other than what it in fact expresses - i.e., in some possible world s expresses something other than what it expresses in the actual world. For example, in the literature on rigidity, it is common to observe that some term a, as used by us in this world, refers to a certain object o with respect to every world, even though a could have referred to something else - i.e., even though there are possible worlds in which a, as used in those worlds, refers to some other object o'. Given a modest essentialism, one can then show that there are sentences, a is F, or if a exists then a is F, that are necessary (i.e., true with respect to every world) even though they could have failed to express necessary truths. On this widely accepted view, it is a contingent fact that these sentences express necessary truths. Moreover, the point has nothing essential to do with rigidity. Presumably, the same sort of reasoning could be applied to analytic sentences. If it is a contingent fact that words have the meanings that they do, then a sentence can be true in virtue of meaning, and hence necessary, even though it is a contingent fact that it has the meaning that it does. This is enough to undermine the Necessity Argument given above. For according to that argument, the non-empirical, and hence non-psychological, nature of semantic theories is supposed to result from the fact that they are required to make claims of the form (1) and (2). But if these claims are empirical and contingent, then we have been given no reason for thinking that semantics itself is not empirical and contingent, or even psychological in nature. One way to resist this objection would be to claim that since sentences, and other expressions, of natural languages are abstract objects, they are not located in space and time, and so bear their linguistically significant properties essentially. On this view the fact that any expression has the meaning it does - or indeed, the syntax or phonology that it does - is never contingent. What is contingent is which language - considered as a system of abstract objects - a given population speaks. Such questions
THE NECESSITY ARGUMENT
577
about populations are distinct from the subject matter of linguistic theories (grammars), which is the structure of languages, considered as abstract semantic, syntactic, and phonological systems. Although this conception of linguistics seems to be the one favored by Katz and Postal, it cannot be invoked as a way of saving the Necessity Argument. The reason it can't is that the argument is supposed to show that something specia! about semantic truths of the form (1) and (2) forces the conclusion that linguistic theories are not contingent, and therefore are not psychological. But if in the end the whole argument depends on characterizing natural language sentences as abstract objects that bear their properties essentially, then it doesn't depend centrally on analytic truths, necessary truths, or even on facts about meaning. Rather, it depends on facts like the following being necessary and not contingent: the fact that 'Snow is white' is true in English iff snow is white, the fact that 'Snow is white' is a sentence of English, the fact that 'white' is an adjective and 'snow' is a noun in English. But then any of these claims will serve in the Necessity Argument equally as well - or as poorly - as claims about analytic truths. Katz and Postal recognize this, and have chosen not to defend the Necessity Argument in this way. Their reason seems to be given in footnote 13. There they maintain that properties like analyticity and necessity apply in the first instance to senses, and only derivatively to sentences. Thus, a sentence is analytic (necessary, etc.) iff it expresses a sense that is analytic (or necessary). They maintain that once this is recognized "one could allow that the connection between words and their senses is contingent and that it is a contingent question whether a sentence is analytic. This would have no effect on the argument in the text because 'analytic' applies to senses and not to the sentences which express them. Therefore, one could concede for the sake o f argument that a necessarily true sentence could have expressed something other than what it in fact expresses without undermining the argument in the text". How so? Presumably the idea is this: A semantic theory of the sort favored by Katz and Postal will characterize sentences as analytic as a result of first characterizing which senses they express, and then characterizing which senses are analytic. Thus claims of the form (1) are seen as derived from claims of the form (3) and (4). (3)
'S' expresses sense M.
(4)
M is analytic.
Although Katz and Postal are prepared to concede (for the sake of argu-
578
SCOTT
SOAMES
ment) that (3) is contingent, they insist that (4) is necessary. Since claims of this form are essential parts of semantic theories of natural languages, these theories are not entirely contingent, and therefore cannot be identified with purely empirical theories of any sort, let alone psychological theories. Although this version of the argument does achieve a measure of success, its significance seems to me to be somewhat restricted. It is consistent with the argument that every claim attributing semantic properties to sentences of a natural language is contingent, or even psychological. The only non-contingent claims are those that are entirely about senses, independent of the sentences that express them. Moreover, it might be argued that the only reason that a linguist should be interested in these claims is as intermediaries needed to derive the contingent claims about language that make up the proper subject matter of linguistics. Finally, it is not obvious that a theory ceases to be empirical simply by virtue of containing subparts that issue in necessary truths about abstract objects. Physical theories employ mathematical subparts without ceasing to be about the natural world. Thus, if all that the reformulated Necessity Argument tells us is that parts of semantics are non-empirical, it doesn't rule out the possibility that semantics as a whole is an empirical psychological theory whose subject matter is to account for the contingent semantic properties of natural language sentences. It should be remembered in this connection that Katz and Postal intend the Necessity Argument to apply to all semantic theories that recognize analyticity, entailment, and other necessary connections between sentences. In particular, the argument is intended to be largely independent of controversies about the form that a semantic theory should take. It is therefore instructive to consider the import of the modified version of the Necessity Argument when applied to a theory that does not take senses as basic, but rather defines the semantic properties of sentences model theoretically. Such a theory would consist in a characterization of truth in a model for the language plus a definition of analytic truths as (something like) sentences true in all admissible models, together with similar definitions of entailment and related notions. If it is conceded that it is a contingent fact that a sentence has the meaning that it does, then it would also have to be conceded that the characterization of truth in a model for the language is contingent. Although the model theoretic definitions of logical properties and relations could be regarded as necessary, it is hard to imagine that the inclusion of definitions of this kind could deprive a theory of its overall empirical status. Since theories that posit senses might be put in this form - i.e. a contingent assignment of senses to sentences,
THE NECESSITY
ARGUMENT
579
plus definitions of what it is for senses to have the relevant properties the same might be said for them. For these reasons, I don't think that the Necessity Argument has the force that Katz and Postal attribute to it. Nevertheless, I think that they are right in supposing that considerations involving semantics provide special reasons for taking the subject matter of linguistics to be languages, abstractly cons!dered, rather than the psychology of language users. These reasons have to do with two of the fundamental goals of semantics. The first is to assign content, including truth conditions, to sentences (relative to contexts of utterance). The second is to provide truth-theoretic characterizations of semantic properties and relations such as analyticity and entailment. An important point to notice regarding the first of these goals is that, in at least some cases, the assignment of content to expressions involves considerations external to the cognitive systems of individual speakers. Thus, as I have argued in Soames (1985, 1986), even if facts specifying the content of expressions are empirical, at least some of them depend on empirical considerations that extend well beyond the domain of theories that could properly be regarded as psychological. The important point regarding the second goal - that of providing a truth-theoretic characterization of logical properties and relations - is to distinguish it from the psychological task of specifying the internal computational structure that allows speakers to recognize such logical properties and relations. It is plausible to suppose that speakers have internal computational procedures of some sort that allow them to draw inferences, and to form judgements about the logical relations that exist between sentences. However, semantic theories do not provide such procedures, and there is little reason to expect to find a close relationship between speakers' internalized "proof procedures" and the truth-theoretic characterizations of logical properties and relations needed for semantics. 2 In short, the questions about meaning asked in doing semantics are different from the questions asked in psychological theories of linguistic competence, or understanding. Thus, it should not be surprising that semantic theories cannot properly be regarded as psychological theories in Chomsky's sense. 3 I agree with Katz and Postal about this negative conclusion. The point that I want to emphasize is that this conclusion does
2 This is argued in detail in Soames (1985). 3 I discuss the relationship between the two types of theory in Soames (1989).
580
SCOTT SOAMES
not depend on characterizing languages as abstract Platonic objects 4, or on taking semantic claims about the logical properties and relations of natural language sentences to be necessary.
REFERENCES Katz, J. and P. Postal: 1991, 'Realism vs. Conceptualism in Linguistics', Linguistics and Philosophy 14, 515-554 (this issue). Soames, S.: 1985, 'Semantics and Psychology', in J. Katz (ed.), The Philosophy of Linguistics, Oxford University Press, Oxford, pp 204-226. Soames, S.: 1986, 'Peacock on Explanation in Psychology', Mind and Langauage 1, 372387. Soames, S.: 1989, 'Semantics and Semantic Competence', Philosophical Perspectives 3, 575596.
Dept. of Philosophy, 1879 Hall, Princeton University, Princeton, NJ 08544-1006, U.S.A.
4 It is not that I am unwilling to accept the Platonic characterization of languages as systems of abstract objects. On the contrary, I prefer it. However, I don't think the refutation of Chomsky's 'Conceptualism' depends on such a characterization; nor do I think that characterizing languages in that way requires one to view linguistics as non-empirical.