Propositional Attitudes and Indexicality: A Cross-Categorial Approach by Philippe Schlenker Agregation de Philosophie (1994), tcole Normale Sup&rieure DEA de Philosophie (1995), Universit6 de Paris 10 Submitted to the Department of Linguistics and Philosophy in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Linguistics at the Massachusetts Institute of Technology September 1999 @1999 Philippe Schlenker. All rights reserved. The author hereby grants to MIT permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole or in part. Signature of Author: DepartmnenytL9inguistics and Philosophy September 7, 1999 Certified by: Irene Heim Professor of Linguistics Thesis Supervisor Accepted by: Alec Marantz Professor of Linguistics Head, Department of Linguistics and Philosophy MAISNSSA"CITHUUTSEETTS OF TECHNOLOGY ARCHIVES SEP 2 1A99 9 LIBRARIES -- Propositional Attitudes and Indexicality: A Cross-Categorial Approach by Philippe Schlenker Submitted to the Department of Linguistics and Philosophy on September 10, 1999 in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ABSTRACT Universal Grammar uses the same distinctions (features) and the same interpretive procedures for reference to individuals, times, and possible worlds. We give a partial argument for this hypothesis: person, tense and (maybe) mood can be treated on a par when they occur in reported speech. We consider several generalizations that hold across sortal domains, and develop a theory of propositional attitudes and indexicality that captures these facts, and treats all three categories on a par. First, we extend the notion of 'Sequence' phenomena from tense to person. In Russian, the tense of a direct discourse can be preserved in reported speech, but in English tense agreement, i.e. 'Sequence of Tense', must generally hold. The same contrast exists between English and Amharic pronouns: in Amharic the indexical pronoun of a direct discourse can be retained in reported speech, while in English person agreement, i.e. 'Sequence of Person', must hold. Second, we extend the notion of 'Logophoricity' from person to tense. In Ewe, the indexical pronoun of a direct discourse can only be reported in indirect discourse if a special form is used, one that never occurs outside of attitude environments -a 'logophoric pronoun'. But logophoric tense/mood also exists, and is instantiated by one of the subjunctive forms that exist in modem German (the 'Konjunktiv I'). Third, we observe that both tense and person display the same idiosyncratic behavior in Free Indirect Discourse - an interesting fact given that other indexical elements pattern differently. Finally, we speculate that the notion of Obviation can be extended from person to tense, and suggest that English past tenses are the temporal counterpart of obviative person markers in Algonquian. Our main auxiliary assumption is that attitude operators are quantifiers over contexts of speech/thought, which allows an indexical expression to be evaluated with respect to the context of a reported speech act, and thus to be shifted. Every attitude operator is thus a Kaplanian 'monster', and shifted indexicals are analyzed as a morphological variant of De Se pronouns. Logical forms are assumed to be uniform across languages, with morphology as the only source of cross-linguistic variation. 2 Acknowledgments I cannot say how much this work owes to the help of others. For constant discussion and criticism, I wish to thank the members of my committee: Noam Chomsky, Sabine Iatridou, David Pesetsky, Bob Stalnaker, Kai von Fintel, and especially Irene Heim. I have also benefited from discussions with Karlos Arregi, Rajesh Bhatt, David Braun, Sylvain Bromberger, Marie-H6l.ne C8t^, Dave Embick, Danny Fox, Jean-Pierre Gendreau, Martin Hackl, Ken Hale, Morris Halle, Jim Higginbotham, Roumi Izvorski, Ed Keenan, Hilda Koopman, Idan Landau, Pascal Ludwig, Alec Marantz, Ora Matushansky, Jon Nissenbaum, Franqois R6canati, Dominique Sportiche, Tim Stowell, and Arnim von Stechow. Special thanks to Barry Schein for numerous conversations. I presented parts of this work in Berlin (GLOW), Leipzig (Sinn und Bedeutung), Los Angeles (USC and UCLA), Paris (Universitd de Paris 7), Thermi (CHRONOS), and Tilbingen - thanks to the audiences for their comments. I owe the native speaker judgments to Cornelia Krause (German), Martin Hackl (Upper Austrian & German), Sabine latridou (Greek), Idan Landau (Hebrew), Ora Matushansky (Russian), Lea Nash (Russian), and especially Degif Petros (Amharic & Chaha). The intellectual influence of Francis Wolff and Jean-Claude Pariente was indirect but important - merci. My first two years at MIT were sponsored by icole Normale Superieure and Fondation Thiers. Immeasurable support was provided by my family: Claire Schlenker & Michel Schlenker, Jean-Marc Schlenker, Martine Schlenker, Evi Schlenker - and Misha Levit. Michel Bar-Zeev, Frainzi Schlenker, Anna Schnaider, and Tamara Levit saw the beginning of my studies at MIT, but not their completion. I dedicate this work to their memory. 3 TABLE OF CONTENTS Overview 7 1. Semantic Uniformity 7 2. Empirical basis: cross-categorial generalizations in indirect discourse 8 2.1. Sequence Phenomena 9 2.2. Logophoricity 10 2.3. Free Indirect Discourse 11 3. Auxiliary assumption: attitude operators as quantifiers over contexts 11 4. A second plot: matrix contexts vs. embedded contexts 13 5. Language Variation 15 Chapter 1. Statement of the Cross-Categorial Generalizations 16 I. Generalization 1. Sequence Phenomena Across Domains 16 1.1. Sequence of Tense vs. Non-Sequence of Tense 17 1.2. Sequence-of-Person vs. Non-Sequence-of-Person 21 1.3. Replicating the Distinction within English: Temporal Adverbials 25 1.4. Where we are 28 II. Generalization 2. Logophoricity Across Domains 30 2.1. Author-denoting Logophoric Pronouns 32 2.2. Logophoricity with Tense and Mood 35 III. Generalization 3. Free Indirect Discourse Across Domains 37 IV.G eneralization 4. Obviation Across Domains (A Speculation) 39 4.1. Four Definitions of Obviation 39 4.2. Statement of the Hypothesis 40 4.3. Shared properties 43 4 Chapter 2. A Revised Semantics for Propositional Attitudes 47 I. Problems with a proposition-based system 47 1.1. All-purpose Indexicals 47 1.2. Attitudes De Se 49 II. Proposal 56 2.1. The system: Utterances and Propositional Attitudes 56 2.2. Temporal Adverbials 61 2.3. Comments 65 mI. Ross's Generalization and the Performative Analysis 67 3.1. Ross's Generalization 67 3.2. Comparison with Ross's 'Performative Analysis' 70 Appendix I. On Monsters 73 Appendix II. 'He' can be read De Se 75 Guide to the Fbrmal Systen 84 Chapter 3. Person 88 I. Indexical Pronouns as Bound Variables (English) 89 1.1. 'I' is sometimes a bound variable 89 1.2. '1' is always a bound variable 92 1.3. Implementation 94 II. Sequence of Person (English vs. Amharic) 95 2.1. English 1st person vs. Amharic 1st person 95 2.2. Agreement and De Se readings 98 2.3. Implementation 104 I. Logophoricity 104 3.1. Author-denoting logophoric pronouns 106 3.2. Hearer-denoting logophoric pronouns 110 IV. Free Indirect Discourse 111 V. Obviation 115 5.1. A Simple Analysis 115 5.2. A Refinement 117 5 Chapter 4. Present and Past Tenses 120 I. Setting Up the Correspondence between Person and Tense 121 1.1. The General Idea 121 1.2. Propositional Attitudes: Temporal De Re vs. De Se 123 1.3. The Present as a Bound Variable 125 1.4. Past tenses as obviatives (English) 127 II. Sequence of Tense (English vs. Russian) 132 2.1. The Null Hypothesis vs. the History 134 2.2. The Null Hypothesis: Sequence of Tense = Sequence of Person 137 2.3. Summary of the rules 154 II. Logophoric Tense/Mood 155 IV. Free Indirect Discourse 156 Appendix. Future tense and Mood 160 I. Feature transmission under 'will' 160 II. Upper Limit Constraint and Double-Access Readings 163 m. Speculations on Sequence of Mood 169 References 178 Notes 6 OVERVIEW 1. Semantic Uniformity We attempt to establish, within a limited domain, the plausibility of a general hypothesis - that a single interpretive system underlies reference to individuals, to times and to possible worlds; that, in other words, the grammar of person, tense and mood can and should be unified. We call this hypothesis 'Semantic Uniformity'; in its strongest form, it can be stated as follows: (1) Hypothesis of Semantic Uniformity Universal Grammar uses the same distinctions (features) and the same interpretive procedures for reference to individuals, times, and possible worlds. Specifically: a. Every feature that exists in one domain can (in principle at least) exist in every other domain as well b. The interpretive rules for those features are the same across sortal domains'. Although the name 'Semantic Uniformity' hasn't been used before, previous work has made this hypothesis plausible in two domains: natural language quantification and anaphora. Consider quantification first. It was an old insight of intensional logic that the same formal system could be used to handle both modality ('can', 'must', etc.) and tense - to handle, in other words, quantification over possible worlds and quantification over times. Originally it was thought that the system that was needed was strictly less powerful than what was required to formalize reference to individuals, i.e. a logic with full quantification over individuals. Interestingly, this hypothesis, which made quantification over times and worlds crucially different from quantification over individuals, was in the end disproved. Cresswell 1990 shows that temporal and modal discourse in natural language is rich enough to amount to full quantification over times and worlds (an idea first hinted at in van Benthem 1977). In the end, then, it seems that the same quantificational system underlies not just reference to times and to worlds, as was originally thought, but also reference to individuals. In that sense Cresswell's work can be seen as an argument for Semantic Uniformity from the study of natural language quantification. Next, consider anaphora. The notion was oiiginally applied solely to reference to individuals. Partee's discovery (in Partee 1973, further developed, and partly retracted, in Partee 1984) was that it could be extended to reference to time as well. Her argument was that tenses have exactly the same range of uses as pronouns do, and may thus be deictic ('I didn't turn off the stove'), anaphoric or bound. Partee's insights are extended to modality in Stone 1997, where it is claimed that Partee's temporal examples can be replicated with modality. This, in turn, suggests that anaphora applies not just to reference to individuals and to times, but also to reference to possible worlds (or scenarios, or whatever one uses to formalize modal discourse). Partee's discovery, together with Stone's extension of it, can thus be seen as an argument for Semantic Uniformity from the study of natural language anaphora2 7 Other cross-categorial generalizations are also relevant to the Hypothesis of Semantic Uniformity, but will not be discussed for reasons of space. To give one prominent example, it has been argued by Bach that the telic-atelic distinction is the verbal analogue of the count-mass contrast in the nominal domain. The idea is highly appealing from our perspective, but goes beyond the present work. Other attempts at unification in syntax or in semantics are less directly relevant. For the Hypothesis of Semantic Uniformity does not simply claim that different domains should be unified somehow. Rather, the suggestion is that a single mapping should yield at the same time a correspondence between the system of morpho-syntactic features that are used in different domains, and the semantic rules that interpret them . This provides a relatively simple heuristics for determining how the unification should be effected: take the system of features used, say, for reference to individuals; keep the same interpretive rules, but replace symbols ranging over individuals with symbols ranging over times and worlds3 - and you should have obtained the system used by Universal Grammar for reference to times and to worlds. The claim might be too strong, or plainly incorrect, but it is relatively easy to see how it could be fleshed out. Some attempts at unification do not satisfy it. To give but one example, Stowell has suggested in recent years that tenses should be analyzed by analogy with polarity items. His theory provides a simple mapping between the syntax of tense and that of polarity items, but it is not clear what the semantic correspondence should be (i.e. it is not clear that the semantic correspondence would in any way be natural). For an attempt at unification to fall under Semantic Uniformity, it must yield both a morphosyntactic and a semantic correspondence. 2. Empirical basis: cross-categorial generalizations in indirect discourse The general enterprise of arguing for Semantic Uniformity might be exciting, and it might also be wrong-headed; but what seems clear is that it goes far beyond the present work. We thus tackle a problem of more manageable dimensions - we attempt to show that tense, person and mood can be analyzed in uniform fashion when they occur in indirect discourse. The restriction is of some interest because a number of non-trivial phenomena occur precisely in these environments. The main line of argument will be as follows: (i) Three puzzling morpho-semantic phenomena affect person, tense and mood in indirect discourse and never in other environments. The facts under consideration are: (1) Sequence effects, (2) Logophoricity, and (3) Free Indirect Discourse. (ii) Each of these phenomena shows up both with person and with tense/mood', although not necessarily in the same languages. (iii) Therefore a uniform theory, i.e. a theory that treats in the same way tense, person and mood, should be posited to account for the generalizations. [We will also mention a fourth possible generalization across sortal domains. We will speculate that the tense system of English has a counterpart in the pronominal system of Algonquian. If this is correct, English tense will turn out to be an 'obviative' system. But since this generalization is not related to indirect discourse per se, we do not mention it any further in this Overview]. 8 2.1. Sequence Phenomena In Russian, the tense of a direct discourse must be retained in reported speech. Thus, if Peter said a week ago: 'It is raining', a Russian speaker has no choice but to report this by saying literally: 'Peter said a week ago that it IS raining'. This is of course impossible in English, where the embedded tense must be shifted so as to agree with the matrix past tense: 'Peter said a week ago that it WAS raining'. Languages like English are said to have a 'Sequence of Tense' rule, which Russian lacks. Our first task is empirical - we have to show that the same typological contrast holds for person (reference to individuals) and mood (reference to possible worlds). The point to establish is that in some languages the tense or mood of a direct discourse must be retained in reported speech, while in others it has to be shifted. The final generalization will look like this: (2) Sequence Generalization (1) a. Sequence-of-Tense vs. Non-Sequence-of-Tense Two weeks ago, Peter said: 'It is raining' i) Russian (lit.): Peter said that it IS raining ii) English: Peter said that it WAS raining b. Sequence-of-Person vs. Non-Sequence-of-Person John says: 'I am a hero' i) Amharic (lit.): Johni says that Ii am a hero ii) English: Johni says that hei is a hero c. Sequence-of-Mood vs. Non-Sequence-of-Mood i) M. Greek (lit.): If Mary <was in a situation where she> knew that Clinton IS dead, she would be devastated ii) English: If Mary <was in a situation where she> knew that Clinton WAS dead, she would be devastated As seen in b., just as Russian preserves in reported speech the tense of a direct utterance, Amharic can retain in reported speech the indexical pronoun of a direct utterance, so that 'I' in bi) can refer to John rather than to the actual speaker - an interpretation which is not open to the English speaker. Similarly, in c. if a person's speech or thoughts are reported in a counterfactual proposition in Modern Greek, the mood of the direct discourse can be retained, although this is impossible in English, at least on the intended interpretation [(2ci) is grammatical in English, but has a special reading which the Greek sentence lacks -this will be discussed later]. Tense, indexical pronouns and mood share an interpretive property - their denotation is defined with reference to the speech situation. Thus the present refers to an interval which contains the utterance time; a first person pronoun refers to the speaker; and the use of indicative mood suggests that the proposition expressed holds at worlds that are 'close enough' to the actual world, in a sense to be made precise. All three categories are thus indexical s. The typological distinctions introduced in (2) can thus be restated more briefly: 9 (3) Sequence Generalization (2) a. Constructions that lack a Sequence 'rule' allow an indexical element to be retained in reported speech in the same form as it would appear in direct discourse. b. Constructions that have a Sequence 'rule' do not allow this. The general idea of relating Sequence of Tense to parallel phenomena in the pronominal domain is not new. In Kratzer 1998, an attempt is made at showing that, within English, there exist locality constraints on certain readings of indexical pronouns which can be replicated in the temporal domain. Apart from the general idea of unifying tense and pronouns, there is very little in common between Kratzer's attempt and ours. In particular, we differ with her on (i) the data we consider (she concentrates on English, while our main argument is that the same cross-linguistic typologies are found in each domain), (ii) the theory we put forth, and (iii) the assessment of the data that underlie her own system (see the discussion below)6. 2.2. Logophoricity English present tense and English 'I' can only depend on the context of the actual speech act. The complement set of this class is made of elements that can only depend on the context of a reported speech act. In the person domain pronouns that have this property are called 'logophoric'. We suggest that Logophoricity applies across sortal domains. In Ewe and Gokana, a special pronoun or agreement marker is used whenever a discourse containing a 1st person pronoun is reported in the 3rd person. Thus if Kofi says: 'I am leaving', the English speaker will report this by saying: 'Kofi says that HE is leaving', while the Ewe speaker has to use a special form (call it 'he*') which occurs only in reported speech: 'Kofi says that HE* is leaving'. 'he*' can be used only when it occurs in the scope of an attitude operator -that is, in precisely those environments in which a 1st person pronoun in Amharic can be shifted. There is a difference between the two cases, however: Amharic 'I' can always denote the actual speaker; by contrast, 'he*' can only refer to the author of a reported speech act. But now consider tense/mood. In German there is a special form - call it 'be*' - which appears only in the scope of an attitude operator. If Kofi says: 'Peter is sick', the German speaker can' report this by saying: 'Kofi says that Peter BE* sick'. And, just as 'he*' can never be used in a direct discourse, so similarly 'be*' appears (almost) only when somebody's thoughts or words are reported. To summarize: (4) Logophoricity Generalization a. Logophoric pronouns vs. no logophoric pronouns Situation: Kofi says: 'I am leaving' i) Ewe / Gokana: Kofi says that he* is leaving ii) English: Kofi says that he is leaving 10