context and logical form

JASON STANLEY CONTEXT AND LOGICAL FORM ABSTRACT. In this paper, I defend the thesis that all effects of extra-linguistic context on the truth-conditions of an assertion are traceable to elements in the actual syntactic structure of the sentence uttered. In the first section, I develop the thesis in detail, and discuss its implications for the relation between semantics and pragmatics. The next two sections are devoted to apparent counterexamples. In the second section, I argue that there are no convincing examples of true non-sentential assertions. In the third section, I ar- gue that there are no convincing examples of what John Perry has called ‘unarticulated constituents’. I conclude by drawing some consequences of my arguments for appeals to context-dependence in the resolution of problems in epistemology and philosophical logic. My purpose in this paper is to defend the thesis that all truth-conditional effects of extra-linguistic context can be traced to logical form.1 But before the import of this thesis can be understood, a few distinctions must be clarified, and its opponents introduced. By “logical form”, I mean a special sort of linguistic representation, rather than the form of a non-linguistic entity such as a proposition or a fact. However, even in its linguistic sense, there is certainly no one uniform use of the expression “logical form”. But there are two distinguishable senses underlying its many differing usages. It is only in the second of these two senses that the thesis I will defend is interesting and controver- sial. Perhaps the most prevalent tradition of usage of the expression “logical form” in philosophy is to express what one might call the revisionary conception of logical form. According to the revisionary conception, nat- ural language is defective in some fundamental way. Appeals to logical form are appeals to a kind of linguistic representation which is intended to replace natural language for the purposes of scientific or mathematical investigation. Different purposes may then give rise to different regiment- ations of natural language. For example, one might want to replace nat- ural language by a notation in which there is some kind of isomorphism between the true sentences in the notation and the facts they describe (e.g., Russell (1985)). Alternatively, one might want to replace natural language 1 By “context” in this paper, I will throughout mean extra-linguistic context. So, nothing I say bears on standard appeals to type-shifting principles, which involve the effects of linguistic context on interpretation. Linguistics and Philosophy 23: 391–434, 2000. © 2000 Kluwer Academic Publishers. Printed in the Netherlands. 392 JASON STANLEY by a notation which explicitly reveals the hidden contribution of logical expressions, such as the language of the predicate calculus. To say that all context-dependence is traceable to logical form in a revisionary sense of “logical form” might be taken to be the trivial claim that, for purposes of interpretation, one should replace natural language by a notation in which all context-dependence is made explicit in the favored notation. It is not in this sense that I intend the thesis. According to the second tradition of usage, which one might call the descriptive conception of logical form, the logical form of a sentence is something like the ‘real structure’ of that sentence (e.g., Harman (1972)). On this approach, we may discover that the ’real’ structure of a natural lan- guage sentence is in fact quite distinct from its surface grammatical form. Talk of logical form in this sense involves attributing hidden complexity to sentences of natural language, complexity which is ultimately revealed by empirical inquiry. It is in this sense that I intend the thesis that all context- dependence is traceable to logical form. What I shall defend is the claim that all truth-conditional context-dependence results from fixing the values of contextually sensitive elements in the real structure of natural language sentences. SECTION I In this paper, I focus, for clarity’s sake, on the speech act of assertion. My goal will be to defend the claim that all effects of extra-linguistic context on the truth-conditions of assertions are traceable to logical form. Though ordinary language philosophers (e.g., Austin (1962)) held such generaliza- tions to be illegitimate, I will nevertheless assume here that the arguments I advance for the case of assertion generalize to other speech acts. I will also assume that each successful assertion has a truth-condition. I will often call the truth-conditions of an assertion ‘what is expressed by that assertion’. This usage must be sharply distinguished from the usage found in authors such as Bach (1994), where ‘what is expressed’ is allowed to denote something that is not a truth-condition. At times in the course of this paper, I will speak of a truth-condition as a certain kind of thing, namely a structured proposition, an ordered sequence of objects and properties.2 There are two reasons for this. First, many 2 In his (1987), Scott Soames contrasts his structured proposition conception of se- mantics with a truth-conditional conception, and rejects the latter on the grounds that truth-conditions are too fine-grained to serve fundamental semantic purposes. However, these (important) issues are independent of the concerns of this paper. If one thinks of CONTEXT AND LOGICAL FORM 393 philosophers think of a semantic theory for a language as primarily in- volving an algorithm which assigns structured propositions to sentences re- lative to contexts, and so are more familiar with the issues when couched in these terms. Perhaps also because of the first reason, appeal to talk of struc- tured propositions makes the issues I discuss somewhat simpler to explain. For example, it allows us to speak of ‘constituents’ of what is expressed that correspond to constituents of sentences. It is slightly more difficult to avail ourselves of this useful metaphor on a straight truth-theoretic frame- work.3 For these two reasons, I will occasionally speak in these terms, though nothing substantial rests upon my uses of this framework. I will also assume, in this paper, that syntax associates with each oc- currence of a natural language expression a lexically and perhaps also structurally disambiguated structure which differs from its apparent struc- ture, and is the primary object of semantic interpretation. In accord with standard usage in syntax, I call such structures logical forms. In this paper, I will repeatedly be using the terms “semantic” and “prag- matic”. However, there are many different usages of these expressions. For example, according to one traditional use of the term “semantic”, se- mantics is the study of context-invariant aspects of meaning. On this ac- count, the semantic content of any two utterances of “I am tired” is the same, since their context-invariant meaning is the same. If one is using the term “semantic” in this sense, then there is a corresponding sense of the term “pragmatic”. In this sense, pragmatics is the study of those aspects of linguistic communication that depend on context. For example, the study of how the meaning of indexical expressions changes with their context of use is, on this way of using the terms, part of pragmatics (cf. Bar- Hillel (1954)). Before the work of Paul Grice, this use of “semantic” and “pragmatic” was standard. For example, it seems to be the best explication of the usages of the terms in the work of Richard Montague. This usage of the expressions “semantic” and “pragmatics” is very clear. However, it obscures important disanalogies. It is very natural to divide the process of linguistic interpretation into two phases. In the first phase, a hearer first assigns denotations to each element of the logical form pro- duced by the speaker, denotations that are determined by the meanings of the elements of the logical forms plus perhaps contextual factors. The structured propositions as more fundamental entities that determine the truth-conditions of an assertion, then one can take this paper to concern the proposition expressed by an assertion, and only derivatively the truth-conditions of an assertion. 3 One cannot, in a truth-theoretic semantics, speak of the ‘constituents’ of what is expressed that correspond to the sentence. Rather, one must speak, more awkwardly, of the properties and objects mentioned in the statement of the truth-conditions that are introduced by rules assigning them to expressions of the object-language. 394 JASON STANLEY hearer then combines these values in accordance with the structure of the logical form to derive the interpretation of the logical form, relative to that context. In many cases, e.g., words like “I”, “now”, “this”, and “she”, the context-invariant meaning of an element in the logical form does not exhaust its denotation, but rather serves as a guide for the interpreter in this process of denotation assignment. In the second phase, the hearer evaluates the result of the first phase with respect to general conversational maxims, such as relevance, quality, or quantity. This second stage of interpretation is not linguistic in nature. It does not involve the assignment of values to ele- ments of a structured representation produced by the speaker. Accordingly, the first stage of interpretation is “semantic”, the second, “pragmatic”.4 These two usages are very different. According to the first usage, what semantics interprets are expression types, simpliciter. On the first usage, there are no semantic differences between distinct uses of a sentence such as “I am tired”. According to the second usage, on the other hand, what semantics interprets are rather expressions relative to contexts. If Hannah is the speaker in context c, and John is the speaker in context c0, then there is a semantic difference between “I am tired”, relative to c, and “I am tired”, relative to c0. There is a third very standard usage of “semantic” and “pragmatic”. According to this third usage, semantics concerns truth-conditions, or pro- positions. There are many different usages of this familiar phrase (cf. Stal- naker (1970), for one such usage). However, the usage I have in mind is one according to which the phase of interpretation that is semantic is the one that results in truth-conditions (cf. Section II of Stanley and Szabó, forthcoming). It is this usage that underlies talk of “truth-conditional se- mantics”. Pragmatics is then the study of those aspects of interpretation that take as input the truth-conditions of a linguistic act, and yield other propositions implicated by that speech act. This is the usage that is most clearly suggested by the work of Grice.5 4 Bach (1999) gives a particularly clear explanation of this notion of “semantic”. 5 For example, Grice is very clear that his ‘favored use’ of “what is said” applies to utterances, or expressions in contexts, rather than expression types. Considering an utter- ance of “He is in the grip of a vice”, made about some person x, Grice writes “. . . for a full identification of what the speaker said, one would need to know (a) the identity of x, (b) the time of utterance, and (c) the meaning, on the particular occasion of utterance, of the phrase ‘in the grip of a vice’.” (1989, p. 25) As this passage makes clear, there is also no reason to think that Grice thought that every element of what is said must be the value of something in the logical form, since he claims that the time of utterance is a determinant of what is said, but never suggests that it is named by a constituent of the sentence. It is what is said in Grice’s favored sense that is, according to him, the input to pragmatics. CONTEXT AND LOGICAL FORM 395 These three distinct usages do not come close to exhausting the dif- ferent senses of “semantic” and “pragmatic” in the literature. To avoid debates that are at bottom terminological, it is important, in any discus- sion of issues involving context, to settle immediately upon one way of using these expressions. In this paper, I use the expressions “semantic” and “pragmatic” in the second of the above senses. That is, semantic inter- pretation involves the assignment of denotations to elements of a logical form relative to a context, and their combination. Extra-linguistic context enters in only when called upon by a linguistic rule governing an element. The result of semantic interpretation is some kind of non-linguistic entity, such as a proposition or a property, which is then the input to pragmatics. However, if my claim in this paper is correct, then the second and third usages of “semantic” and “pragmatic” coincide. That is, if all effects of extra-linguistic context on the truth-conditions of an assertion are traceable to logical form, then the result of semantic interpretation in the second of the above usages will be the truth-conditions of the assertion, and hence the result of semantic interpretation, in the third sense of “semantic”. One purpose of my paper is an attempt to bind together these two distinct us- ages of the term “semantic”, and thereby justify talk of “truth-conditional semantics”. Now that we are clear about my future use of the term “semantic”, I add a final assumption. The assumption is that composition rules do not vary as a function of extra-linguistic context. This assumption is entailed by every version of the principle of compositionality, which is a standard condition of adequacy on a semantic theory. According to one formulation of this principle, a semantic theory is compositional just in case, for each complex expression, there is exactly one way, determined solely by its structure, in which the meanings of its constituents are combined by the semantic theory to yield its meaning.6 It follows from this principle that, although the meaning of a non-complex word may vary with context, the way in which the interpretation of a complex expression is derived from the interpretations of its parts cannot vary with context. For if a semantic theory allowed the way in which the interpretation of a complex expression is built from the interpretation of its parts to vary with context, then it would not correlate with each complex expression, a unique way in which the interpretation of its constituents combine to yield its interpretation.7 6 On this characterization of compositionality, different syntactic constructions may be associated with different modes of semantic composition. For a useful discussion of different notions of compositionality and related principles, see Janssen (1997). 7 It is worth mentioning that most semantic accounts of variable-binding are in tension with compositionality as I have stated it (though of course are consistent with my assump- 396 JASON STANLEY There are certain authors who reject the principle of compositional- ity, since they hold that the meaning of a complex expression may de- pend upon its linguistic context (e.g., Higginbotham (1986), Hintikka and Sandu (1997)). However, this position is fully consistent with the assump- tion I have made, that composition rules do not vary as a function of extra-linguistic context. This latter assumption is far weaker than com- positionality. Since I am not aware of any author who is not an oppon- ent of systematic semantics who has denied it, the assumption should be uncontroversial, and I presuppose it in what follows. Suppose my principal claim is true, that all effects of extra-linguistic context on the truth-conditions of an assertion are traceable to logical form. Then, the effects of context on the truth-conditional interpretation of an assertion are restricted to assigning the values to elements in the expression uttered. Each such element brings with it rules governing what context can and cannot assign to it, of varying degrees of laxity. The effects of extra-linguistic context on truth-conditional interpretation are therefore highly constrained. If this picture of truth-conditional interpretation is cor- rect, then it is fundamentally different from other kinds of interpretation, like the kind involved in interpreting kicks under the table and taps on the shoulder.8 We do not interpret these latter sorts of acts by applying highly specific rules to structured representations. Nor is the role of extra- linguistic context in interpreting these acts in any way constrained, as it is in the case of linguistic interpretation. Thus, if the interpretation of asser- tions in fact functions in the way I have sketched, one should be suspicious of views that assimilate it too quickly to the ways in which we interpret non-linguistic acts. In recent years, there has been no shortage of philosophers of language, linguists, and cognitive scientists eager to reject the claim I have advanced. According to Kent Bach, Robyn Carston, François Recanati, Dan Sper- ber, Robert Stainton, Charles Travis, and Deirdre Wilson, among others, tion). An example is the ‘Predicate Abstraction Rule’ discussed in Heim and Kratzer (1998, pp. 186ff). Essentially, Heim and Kratzer assume a syntax that involves structures such as: T��xTSTNPTN John]] TVPTV offended] TNPTNxUUUUU. However, they assign no independent interpretation to ‘�x’. So, on their account, interpreting the node � does not amount to combining the value of ‘�x’ with the semantic value of the open sentence “John offended x”. Rather, they provide a non-compositional interpretation rule. This sort of violation of compositionality is fairly common, and should not raise any worries. Violations of compositionality only become worrisome from the standpoint of learnability when they involve an unlimited number of unrelated construction rules, as would be the case with context-dependent construction rules. 8 Where the latter are not governed by explicit meaning-granting stipulations. This proviso should be tacitly understood in future references to interpretation of non-linguistic acts. CONTEXT AND LOGICAL FORM 397 the truth-conditions of most assertions go well beyond what semantics can legitimately assign to the logical forms of the sentences uttered. In- stead of assigning propositions, entities that are truth-evaluable, to logical forms, semantic interpretation only involves “fragmentary representations of thought” (Sperber and Wilson (1986), p. 193) “partially articulated con- ceptual representations” (Carston (1991), p. 49), or “propositional rad- icals” (Bach (1994), p. 127; cf. also Bach (1982)).9 The examples motivat- ing these theorists all concern the effects of context on what is expressed in assertions. According to these theorists, there is no way to ‘constrain’ the effects of context on what is expressed within the domain of semantic in- terpretation. In most cases, what the semantic interpretation of a sentence’s logical form delivers is not what is expressed, but rather, in the words of Sperber and Wilson, “mental objects that never surface to consciousness”; these are then used in a pragmatic derivation of what is expressed. If these theorists are correct, then semantics is not about truth-conditions. It would then be more apt to replace, as does Recanati, talk of truth- conditional semantics with talk of truth-conditional pragmatics (cf. Re- canati (1993), Chapter 13). Underlying these arguments against the picture of interpretation I ad- vocate are two assumptions about semantic theory, both of which I ac- cept. The first assumption these theorists make about what is semantically legitimate is: First assumption: In semantic interpretation, one may never postulate hidden structure that is inconsistent with correct syntactic theory. According to some conceptions of semantics, the objects of semantic interpretation are not syntactic logical forms, but rather logical forms in some more revisiona