Attitudes and Changing Contexts Robert van Rooy Abstract: In the final chapter of his classic book about pragmatics, Gerald Gazdar (1979) wondered whether truth-conditional semantics can be autonomous with respect to pragmatics. Semantics is autonomous with respect to pragmatics, he said, when the truth conditions of natural language sentences can be determined without making reference to the presuppositions and other features of the contexts in which the sentences are used. At least since the rise of dynamic theories in the 1980s, wide agreement has been reached that this autonomy thesis should be given up: the truth conditions of an individual sentence cannot be determined without making reference to the facts about the conversational context in which this sentence is uttered. In this dissertation I will argue that what counts is not in the first place the preceding sentences of the discourse, but the circumstances in which these sentences are used, and the attitudes of the agents that use them. Although I believe that truth-conditional semantics can be fruitfully studied in abstraction from pragmatics, this does not mean that the beliefs and presuppositions of language users are irrelevant to the truth-conditional contents of natural language sentences. In this dissertation I try to motivate and use a conception of meaning most explicitly defended by Stalnaker (1984). According to this conception, the meaning and content of linguistic expressions should be explained in terms of the intentions, beliefs and conventions of language users. This is in particular the case for modal discourse and for referential expressions: modal discourse should be explained in terms of beliefs and activities of rational agents; and the key notions of model-theoretic semantics, the notions of reference and aboutness, should be explained in terms of what speakers do by their use of a term, not by properties of the term itself. This conception of reference and aboutness is not only defended by Stalnaker, but was also the received view of semantics until the late 1960s. According to the traditional view, the expression (or thought) E refers to, or is about, R because (i) the speaker intends to refer to (or thinks about) the object (or set of objects) that satisfies the definite description the speaker or thinker associates with E (or is presupposed to be associated with E); and (ii) R satisfies this description. However, Donnellan, Kaplan, Kripke, Putnam and others have shown that the received view leads to counterintuitive results. On the basis of these observations it has sometimes been suggested that the denotation relation between terms and objects should be explained in causal terms independent of the intentions and beliefs of language users. I will follow Stampe (1977) and Stalnaker (1984), however, and argue that the causal theory of reference itself should be grounded in a (partly) causal, information-theoretic account of intentionality, and that therefore reference can be explained again in terms of beliefs and intentions of language users. Crucial to this account is that the content of a representational state need not be explained in terms of what has actually caused the representational mechanism being in the particular state it is, but might also be explained in terms of what under normal circumstances causes the mechanism to be in that state. In this way we can not only analyse singular terms like indexicals and proper names by a causal theory of reference, but general terms as well. Also, we can in this way analyse not only knowledge and perceptual attitudes in (partly) causal terms, but the more dispositional attitude of belief too. The main goal of chapter | will be to defend this picture, and to defend what I take to be a consequence of this picture: that from the believer's point of view a belief state should be modelled by a proposition individuated by truth conditions. I will defend this picture, and the consequence of it by arguing for a three-way strategy to solve some problems that arise on such an approach. First, I argue that we should make use of diagonalisation to account for the intuition that it might be unclear what, under normal conditions, is causally responsible for a certain representation. Second, I use a counterpart theory that allows for the possibility that agents have different representations of the same real object. Third, I will argue that many of the problematic aspects of attitude attributions are pragmatic problems, due to the extreme context-dependence of attitude attributions. In chapter 1 I seek to reconstruct the Stalnakerian position with respect to content and belief attributions. It is partly built on the insight, due to Kaplan, Stalnaker and others, that it is good to make a conceptual distinction between two kinds of facts: (i) facts about the subject matter of thought or conversation, and (ii) facts about linguistic and speech conventions, and the conversational situation itself. This conceptual distinction will be used extensively in the following three chapters about anaphora and presuppositions. In chapter 2 I will account for anaphoric relations across sentential boundaries on the basis of the intuition that pronouns are normally used referentially, and the assumption motivated in chapter 1 that referring is something done by speakers with their use of a term, not by the term itself: Which object is referred to depends on the intention of the speaker. Kripke (1977) taught us that a distinction must be made between general and specific intention. I will argue that for pronouns it is normally the specific intention that counts. Speakers normally refer back with their use of a pronoun, or short description, to the speaker's referent associated with the indefinite that figures as its syntactic antecedent. In this chapter I will show that by means of diagonalisation such an analysis can be pushed further than many have supposed; and that in fact this analysis is close to, but not identical with, modern theories of anaphora like Discourse Representation Theory (Kamp, 1981), File Change Semantics (Heim, 1982), and the more recent Dynamic Semantics due mostly to Groenendijk & Stokhof (1991). The reason will be that participants in a conversation are normally not only unclear about the facts relating to the subject matter of conversation, but also of some relevant facts about the conversational situation itself. Of course, sometimes a singular pronoun that takes an indefinite as its syntactic antecedent can be appropriately used although it does not refer to the specific speaker's referent of the indefinite. Sometimes it is only the general intention that counts. I will argue that to account for many of those cases we need descriptive pronouns in addition to referential pronouns. The former are pronouns that go proxy for a description recoverable from the sentence in which its syntactic antecedent occurs. In chapter 2 I will be concerned mainly with motivating this division of labour, implementing this analysis of pronouns into a dynamic theory of meaning, and using this two-tiered approach to account for phenomena problematic for the above-mentioned popular theories of anaphora. In chapter 3 I discuss how to account for anaphoric relations across belief attributions, concentrating mainly on the problem of intentional identiry made famous by Geach's Hob-Nob sentences. I will discuss how much of the popular view, which takes so-called unbound pronouns either as abbreviations for the antecedent clause or as variables bound by a dynamic existential quantifier, can be maintained. I will suggest that this view cannot be maintained. If we should account for Edelberg's (1986) asymmetry problem in semantics, I will argue that it is useful to take the notion of speaker's reference senously in semantics. What proposition is expressed by a sentence depends partly on the speaker's presuppositions. This is in particular the case for quantified statements like Every German loves his Buick. Intuitively, this sentence is only about the set of Germans with Buicks that the speaker presupposes is conversationally salient. If he is not allowed in a particular conversational context to presuppose that there is such a salient set of Buick-owning Germans, the assertion will be counted as inappropriate because it is not clear what proposition is expressed by the sentence. I believe that this is the intuition we have to capture, and I argue in chapter 4 that we can account for this intuition in a two-dimensional analysis of presuppositions in the style of Karttunen & Peters (1979). Any two-dimensional analysis of presuppositions is based on the assumption that there is a semantic value of any sentence that can be determined independently of the context in which it is used. Traditionally, this semantic value was supposed to be its truth conditions. However, this assumption gave rise to the so-called binding problem. Karttunen & Peters suggested that it is the assumption that we should represent what is asserted and what is presupposed by separate propositions or logical forms that is responsible for the binding problem. Indeed, it is now commonly assumed that presuppositions cannot be handled in a two-dimensional theory of presuppositions. In this chapter I argue that this conclusion was drawn too quickly. Moreover, it is not in accordance with the conclusions we should draw from dynamic semantics. Traditionally, the semantic value of a sentence is its truth conditions, but according to dynamic theories it is its context change potential, a function from contexts to contexts. But once this insight is taken seriously in presupposition theory, it is again feasible to account for presuppositions in a two-dimensional way. In this chapter I argue that the binding problem can be solved in a two-dimensional theory by first determining the context change potentials of what is presupposed and what is asserted by a sentence independently of each other, and second, making the truth conditions of what is asserted dependent on what is presupposed. According to the Lewis/Stalnaker analysis of conditionals, the truth conditions of a conditional sentence depends crucially on the speaker's intentions. The speaker's intentions, together with the antecedent and (other) facts about the actual world, select the relevant world(s) with respect to which the truth value of the consequent, and thus of the whole conditional, is evaluated. Stalnaker tried to make a stronger claim: the formal properties of the function that does this selection should be explained in terms of the beliefs and presuppositions of language users. He proposed that the analysis of conditionals should be related to the analysis of belief revision. I will discuss this project in chapter 5 and give some attention to Lewis's triviality result, which showed that what is expressed by a conditional must be even more context-dependent than the original Lewis/Stalnaker analysis suggested, if conditionals are to be explained in terms of conditional beliefs. I will argue in chapter 5 that most conditionals express propositions, but that the proposition expressed by an indicative conditional depends more directly on what is believed and presupposed by the speaker than the proposition expressed by a subjunctive conditional. Finally I will argue that what is expressed by a subjunctive conditional depends in such a systematic way on the conversational context that we could analyse them as variable strict conditionals, where the relevant accessibility relation is defined in terms of the Lewis/Stalnaker notion of 'similarity’. It differs from the traditional analysis in that the notion of 'closeness’' can change its denotation through conversational means. In the final chapter, I make use of the analyses of conditionals, belief revision, and rational decision discussed in the foregoing chapter in order to account for the meaning of some attitude verbs other than 'believe'. I will give the most attention to verbs of desire, and to the analysis of permission sentences. For the analysis of desire attitudes, I will argue that just like beliefs, also desires are closed under logical implication. Also just like beliefs, they are only closed under logical implication with respect to the relevant alternatives, but that it is very context dependent what the relevant alternatives are. With respect to permission sentences I will investigate in how far they can be accounted for in terms of possible world semantics. For some of those verbs I will also propose a way to account for anaphoric relations across attitude verbs.