Relativism and the De Se Interpretation of PRO*

Tamina Stephenson University of British Columbia

1.

Abstract

In this paper, I propose a view of attitude control predicates that uses a form of semantic relativism. Evidence for this view comes from two places: first, from an initially surprising restriction on the interpretation of words like tasty under attitude reports; and second, from a parallel between the behavior of PRO and embedded epistemic modals. 2.

Introduction: A Puzzle

I begin with a puzzle arising from two observations, one old and one new. 2.1

Predicates of Personal Taste

Consider (1)–(2). (1)

Sue thinks the cake is tasty.

(2)

Bill thinks the roller coaster is fun.

These sentences each have a natural and highly salient reading on which the subject of the attitude is understood as the experiencer whose taste or enjoyment is being reported. That is, they are most naturally understood as meaning, roughly, that Sue thinks the cake tastes good to her (1) and Bill thinks that the roller coaster is fun for him (2). This was observed by Lasersohn (2005), and is generally true for the class of predicates he dubs “predicates of personal taste,” which I take to include tasty, fun, entertaining, exciting, and boring, among others. *

Some of this material appeared in my Ph.D. thesis (Stephenson 2007, Ch. 4). For discussion and helpful suggestions, I would like to thank my thesis committee – Kai von Fintel, Irene Heim, Danny Fox, and Agustin Rayo – as well as Martina Wiltschko, Henry Davis, and audiences at UBC, NELS 38, and the 2008 LSA Annual Meeting. Any errors are my own.

Stephenson One might ask why I have not simply given paraphrases like ‘the cake tastes good to Sue’ and ‘the roller coaster is fun for Bill.’ I assume, following Lasersohn (2005), that these stronger meanings arise from the fact that people have privileged access to their own subjective experiences and thus typically do not have false beliefs about them. (This issue is discussed in greater detail in Lasersohn 2005, Stephenson to appear.) 2.2

Control Verbs

Here is a more familiar observation. It is well known that under want and other similar control verbs, the implicit subject (that is, PRO) must be the same as the attitude holder. For example, (3) must mean that Sue wants to go on the roller coaster herself. (3)

Sue wants PRO to go on the roller coaster.

2.3

An Expected Reading

Given these two observations, we would expect to find sentences of the form “X wants to be PRED” (where PRED is a predicate of personal taste), with a meaning equivalent to “X wants to be PRED to X themselves.” An obvious but ill-fated attempt at this is made in (4). (4)

# The cake wants to be tasty.

It is unsurprising that (4) is unacceptable, given that cakes generally have neither wants nor a sense of taste, and entities that do have these generally don’t eat themselves. As it turns out, however, the relevant kind of reading seems to be missing even in cases that are more pragmatically plausible. Consider the context in (5), for example. (5)

[Context: Sue hates spending time alone. When she’s with other people, they generally seem to enjoy her company, appreciating her stories and laughing at her jokes, but she just can’t ever seem to entertain herself, and wishes that she could.]

a. # Sue wants to be entertaining. b.

Sue wants to be entertaining to herself.

In (5a), a reading where Sue is the understood experiencer of entertaining does not seem to be possible. (5a) can only mean that Sue wants to be entertaining to others, making it odd in the given context (since she already knows that she is). In contrast, (5b), which has an overt reflexive, is perfectly acceptable and has this reading. This contrast can also be shown with reciprocals. For example, (6) can mean that Sam and Mary want others to find their company enjoyable (as a pair, perhaps); but it cannot mean that Sam and Mary wish to be able to entertain each other better.1

1

This was pointed out to me by Henry Davis (p.c.).

Relativism and the De Se Interpretation of PRO (6)

Sam and Mary want to be entertaining.

With some creative license, we can even create appropriate examples with tasty. Imagine a magical living cake (let’s call him Mr. Cake) who walks, talks, thinks, feels, eats, and so on. Imagine further that pieces can be cut off of Mr. Cake without causing him any pain, and he regenerates them easily; in fact, if he goes too long without having some of him removed, he gets stale and listless. If no one else is around to take slices, he’ll eat them himself (since he doesn’t like to see them lying around) – but oddly enough, they always taste terrible to him. Now consider (7). (7)

[Context: Mr. Cake was starting to feel stale. He called his friends to see if they would come by and have a slice, but none of them were home, so he cut off a few small pieces of himself and ate them. As usual, they tasted terrible to him.]

a. # This was very frustrating. Mr. Cake wanted to be tasty! b.

This was very frustrating. Mr. Cake wanted to taste good to himself!

Once again, (7a) cannot be understood to mean that Mr. Cake wants to taste good to himself, even though this can be expressed naturally with an overt reflexive as in (7b). It seems, then, that predicates of personal taste under attitude control verbs cannot be linked to the attitude holder – this despite the fact (seen in Section 2.1) that with noncontrol attitude verbs a linked reading is generally the default. 2.4

A Parallel

I assume (following work going back to Epstein 1984) that predicates of personal taste such as tasty and fun have some kind of covert “experiencer” argument. (I will be more specific about this below.) Now suppose moreover that this covert argument is a nonreflexive nominal that is coreferential with PRO. In this case, the examples in (5a), (6), and (7a) above would be parallel to those in (8)–(10) below. (8)

* Suei is entertaining to heri.

(9)

* [Sam and Mary]j⊕m are entertaining to themj⊕m.

(10) * Mr. Cakei tastes terrible to himi.

[compare to (5a)] [compare to (6)] [compare to (7a)]

The sentences in (8)–(10) are widely assumed to be ungrammatical on the indicated readings due to Principle B. I suggest, then, that the restriction described in Section 2.3 above is simply an instance of Principle B.2

2

This was suggested by Irene Heim (p.c.).

Stephenson 3.

Background on Predicates of Personal Taste

Before I can elaborate on the idea that the restriction from Section 2.3 involves Principle B, I need to give some background on predicates of personal taste. 3.1

Two Uses

The first thing to know about predicates of personal taste is that they have two distinct uses. First, they have a use where they just seem to take an implicit, salient experiencer argument. This use is illustrated in (11) (based on examples by Lasersohn 2005). (11)

[Context: Sam is watching his cat eagerly gobble down some cat food. Sam says:] Oh good, the new cat food is tasty.

Here the perspective of the cat is particularly salient, and tasty seems to be understood as “tastes good to the cat.” Crucially, though, predicates of personal taste have a second use in which they seem to be doing something more complicated. Consider the dialogue in (12). (I assume that Sue’s response involves the indicated ellipsis.) (12)

Mary: How’s the cake? Sam: It’s tasty. Sue: No it isn’t [tasty]!

In (12), Sam and Sue both seem to be expressing something about their own experience, and one might be tempted to think that the experiencer is simply the speaker in each case. This cannot be right, however, because then we would expect this discourse to be incoherent in the same way as one like (13) which contains overt first-person indexicals. (13)

Sam: I’m a doctor. Sue: # No, I’m not!

3.2

Judge-Dependent Approach to Predicates of Personal Taste

Now let me summarize the analysis of predicates of personal taste that I have developed in recent work (Stephenson 2006, to appear), based closely on ideas of Lasersohn (2005). Following Lasersohn, I treat propositions as sets of world-individual pairs rather than sets of worlds. (For present purposes, we can ignore the role of times.) In other words, the truth of sentences is relativized to an individual, which Lasersohn terms the “judge.” Formally, then, sentence meanings are of type . I will write [[α]]w,j for the denotation of an expression α at a world-individual pair .

Relativism and the De Se Interpretation of PRO I assume that on their basic meanings, predicates of personal taste are two-place predicates, as shown in (14). (Here and in some other parts of the analysis, I depart from Lasersohn for reasons that I explain in the works cited above.) (14)

[[tasty]]w,j = [λy . [λx . x tastes good to y in w] ] [[fun]]w,j = [λy . [λx . x is fun for y in w] ] [[entertaining]]w,j = [λy . [λx . x is entertaining to y in w] ]

The first argument of these predicates is the experiencer of taste, enjoyment, and so on. (Lasersohn links this argument directly to the judge parameter, but I connect it only indirectly, as we will see.) In other words, tasty (for example) expresses the relation that holds between those pairs of individuals such that x tastes good to y. In some cases, the first argument of a predicate of personal taste is overt, as in a sentence like The roller coaster is fun for Sam. In other cases, I assume, this argument is supplied covertly in one of two ways. The first possibility is for the context to provide a salient individual. This amounts to positing silent referential pronouns, which I will write as prox. So, for example, if the cat (let’s call her Whiskers) is particularly salient in the context, then tasty might take the silent item proWhiskers as its first argument. The resulting meaning of the predicate would be as in (15). (15)

[[tasty proWhiskers]]w,j = [λx . x tastes good to Whiskers in w]

This simply denotes the set of things that taste good to Whiskers. The second way that the first argument of a predicate of personal taste can be supplied covertly is in the form of a special silent nominal PROJ, which simply refers to the value of the judge parameter. That is, it has the lexical entry in (16). (16)

[[PROJ]]w,j = j

For example, if tasty takes PROJ as its first argument, then the result (evaluated at a world w and judge j) is the set of things that taste good to j in w. This is shown in (17). Thus the sentence the cake is tasty (where the implicit argument of tasty is taken to be PROJ) is true at a pair iff the cake tastes good to j in w, as shown in (18). (17)

[[tasty PROJ]]w,j = [λx . x tastes good to j in w]

(18)

[[The cake is [tasty PROJ] ]]w,j = 1 iff the cake tastes good to j in w

Here I will be focusing on sentences like (18) in the case where they are embedded under attitude reports; we will see that this has an effect akin to binding the judge parameter. I have looked in other work at how to analyze sentence meanings like (18) in matrix contexts (see, e.g., Stephenson 2006, to appear).

Stephenson 3.3

Attitude Reports

I assume a semantics for attitude reports based on the view of Lewis (1979), which has been adapted for natural language semantics by Chierchia (1989) and others. This view uses a notion of doxastic alternatives, defined in (19). (19)

Doxastic Alternatives: Doxw,x = { : it is compatible with what x believes in w that x (x’s self) is y in w′}

Given an individual x and world w, x’s doxastic alternatives in w are those pairs such that it is compatible with x’s beliefs in w that x (that is, x’s own self) is y in w′. (Again, I am ignoring the role of times.) As discussed by Lewis, Chierchia, and others, this concept of doxastic alternatives is needed to allow for self locating or de se attitudes. Recall that I am treating propositions as sets of world-individual pairs. This means that a proposition is the same kind of thing as a set of doxastic alternatives, and we can assume the meaning for think and believe given in (20).3 (20)

[[think]]w,j = [[believe]]w,j = [λp . [λze . ∀∈Doxw,z : p(w′)(y) = 1] ]

The important thing to notice in (20) is that the propositional argument is simply evaluated at the subject’s doxastic alternatives. This symmetry can be seen more clearly if the meaning is rewritten equivalently as in (21). For reasons of consistency, though, I will generally use the notation in (20). (21)

[[think]]w,j = [[believe]]w,j = [λp . [λze . Doxw,z ⊆ pS] ] where pS = the set characterized by p = { : p(w′)(x) = 1}

According to (20) or (21), a belief report of the form “z thinks that p” is true iff p is true at all of z’s doxastic alternatives – in other words, if z’s doxastic alternatives are a subset of p (where p is expressed as a set of world-individual pairs). I assume a parallel analysis for other attitudes. In particular, I assume the analogous notion of “want alternatives” given in (22) and the lexical entry for want in (23). (22)

Want Alternatives: Wantw,x = { : it is compatible with what x wants in w for x (x’s self) to be y in w′}

(23)

[[want]]w,j = [λp . [λze . ∀∈Wantw,z : p(w′)(y) = 1] ]

According to (22)–(23), a report of the form “z wants p” is true iff p is true at all of z’s want alternatives – that is, iff z’s want alternatives are a subset of p. 3

I assume a system where the basic type of sentences is t, and they combine with attitude predicates by a special rule of Intensional Functional Application.

Relativism and the De Se Interpretation of PRO A crucial property of this analysis is the way it interacts with the judge-dependent semantics for predicates of personal taste from Section 3.2 above. Consider sentences like (24)–(25), which contain a predicate of personal taste with implicit argument PROJ. (24)

Sue thinks the cake is tasty PROJ.

(25)

Sue wants the cake to be tasty PROJ.

The predicted meanings for (24)–(25) are given in (26)–(27). (26)

[[(24)]]w,j = 1 iff ∀∈Doxw,Sue : the cake tastes good to y in w′

(27)

[[(25)]]w,j = 1 iff ∀∈Wantw,Sue : the cake tastes good to y in w′

According to these meanings, (24) is true iff the cake is tasty at all of Sue’s doxastic alternatives – that is, if Sue believes that the cake tastes good to Sue herself. Similarly, (25) is true iff Sue wants the cake to taste good to Sue herself. This seems to be correct.4 4.

Proposal

Now I turn to a proposal for control that helps to explain the puzzle from Section 2. 4.1

Control Constructions

Let’s look again at the control constructions from Section 2. Some representative examples are repeated in (28)–(29). (28)

Sue wants PRO to go on the roller coaster.

(29)

[Context: Sue hates spending time alone. When she’s with other people, they generally seem to enjoy her company, appreciating her stories and laughing at her jokes, but she just can’t ever seem to entertain herself, and wishes that she could.]

a. # Sue wants PRO to be entertaining. b. 4.2

Sue wants PRO to be entertaining to herself. PRO as PROJ

My proposal has two parts. First, I propose that PRO in these sentences is simply PROJ. Second, I assume that PROJ has the right binding properties (such as being non-reflexive) to make it subject to Principle B. 4

(25) also has a salient reading on which Sue wants the cake to taste good to some other person or people (perhaps the people who are going to eat the cake). This reading arises when tasty takes a silent referential argument instead of PROJ.

Stephenson To see how PRO can be treated as PROJ, consider the case of (28). The embedded clause, (PRO) to go on the roller coaster has the meaning in (30). Putting this together with the meaning for want from Section 3.3, the predicted meaning for (28) is (31). (30)

[[PRO to go on the roller coaster]]w,j = [[PROJ to go on the roller coaster]]w,j = 1 iff j goes on the roller coaster in w

(31)

[[Sue wants [PROJ to go on the roller coaster] ]]w,j = 1 iff ∀∈Wantw,Sue : y goes on the roller coaster in w′

This says that (28) is true iff all of the situations compatible with Sue’s desires are such that Sue (herself) goes on the roller coaster, which is correct. 4.3

Principle B

Recall that (29a), Sue wants to be entertaining, does not have a reading where Sue wants to be entertaining to herself. This corresponds to the case where entertaining takes PROJ as its implicit argument, as in (32). Note that the analysis I have given assigns a perfectly coherent meaning for (32), namely the one in (33b). The meaning assigned to the embedded clause is given in (33a). (32) * Sue wants [ PROJ to be entertaining PROJ ]. (33)

Meaning assigned to (32):

a.

[[PROJ to be entertaining PROJ]]w,j = 1 iff j is entertaining to j in w

b.

[[(32)]]w,j = 1 iff ∀∈Wantw,Sue : y is entertaining to y in w′

However, the structure in (32) contains two instances of PROJ in just the kind of configuration that is standardly taken to violate Principle B. (One way to think of it is that they are co-arguments of the predicate entertaining.) Thus the structure is ruled out by a binding principle, and the reading represented in (33b) is impossible. Note that I am not trying to say anything new about Principle B here. When I say that (32) is ruled out by Principle B, I simply mean that it is ruled out by whatever it is that prevents Sue likes her from meaning the same thing as Sue likes herself. 5.

Predictions

Now I will turn to some additional predictions of my analysis. 5.1

De Se Interpretation of PRO

It is well known that controlled PRO must be interpreted de se in attitude reports (see, e.g., Morgan 1970, Chierchia 1989). Consider (34), for example.

Relativism and the De Se Interpretation of PRO (34)

Pavarotti wants to be famous.

(based on examples by Chierchia 1989)

If Pavarotti thinks to himself, “I want to be famous,” then this can be reported as (34). If, on the other hand, Pavarotti hears a recording of a talented singer on the radio, not realizing that it is a recording of his own voice, and thinks to himself, “I want the man on the radio to become famous,” this cannot be reported as in (34). Similar facts hold for the Italian verb credere, ‘believe’ when it appears with an infinitival complement, as in (35). For example, if Pavarotti thinks to himself, “I am a genius,” his belief can be reported as (35); but if he unwittingly hears himself on the radio and thinks, “That man on the radio is a genius,” his belief cannot be reported this way. (35)

Pavarotti crede di essere un genio. Pavarotti believes COMP be a genius ‘Pavarotti believes that he’s a genius.’ [Lit.: “Pavarotti believes to be a genius”] (Anand 2006, no. 1a, based on Chierchia 1989)

Chierchia’s (1989) solution, now widely adopted, is to treat the embedded clause in control constructions semantically as a property (type ) rather than a proposition (type ). Attitude control predicates such as want and credere, ‘think’ are then taken to be ambiguous between a proposition-taking meaning (used in non-control constructions) and a property-taking meaning (used in control constructions). The analysis I have proposed here immediately derives the de se interpretation of PRO in these attitude reports while keeping all propositions the same type (now ) and giving each attitude predicate only one meaning with both that-clause and infinitive complements. For example, the meaning predicted for (34) is given in (36). (36)

[[(34)]]w,j = [[Pavarotti wants [ PROJ to be famous] ]]w,j = [[want]]w,j ( [λw″ . [λj″ . [[PROJ to be famous]]w″,j″] ] ) ( Pavarotti ) = 1 iff ∀∈Wantw,Pavarotti : y is famous in w′

This says that Pavarotti wants to be famous is true iff all of the pairs such that it’s compatible with what Pavarotti wants for him to be y in w′, y is famous in w′ – in other words, iff Pavarotti wants to have the property of being famous. This correctly captures the de se interpretation. Of course, it is not an accident that my proposal has the same result as Chierchia’s property view. I treat propositions as sets of world-individual pairs, which is exactly what properties are on standard views. The difference here is that because of the introduction of the judge parameter, all sentence-like expressions are assigned this type – not just those embedded by control verbs. Moreover, this move is independently motivated by the behavior of predicates of personal taste and other expressions (as argued by Lasersohn 2005, Stephenson 2006, to appear).

Stephenson 5.2

A Locality Requirement

Locality is a well-known property of control. In particular, in an example like (37), which contains a control construction embedded within another attitude report, the implicit subject of the lowest clause (that is, PRO – or on my view, PROJ,) must be controlled by the closer attitude holder (in this case, Bill), and not the higher one (in this case, Sue). (37)

Sue wants Bill to want __ to go to the party.

This locality requirement follows directly from the compositional semantics on my view. For example, the meaning predicted for (37) is shown in (38). (38)

[[(37)]]w,j = [[Sue wants [Bill to want [PROJ to go to the party] ] ]]w,j = [[want]]w,j ( [λw′ . [λj′ . [[Bill to want PROJ to go to the party]]w′,j′ ) (Sue) = 1 iff ∀∈Wantw,Sue : ∀∈Wantw′,Bill : z goes to the party in w″

According to (38), Sue wants Bill to want to go to the party is true iff in all of Sue’s want alternatives, all of Bill’s want alternatives are such that z goes to the party in w″ – that is, iff Sue wants it to be the case that Bill wants to have the property of going to the party. The crucial thing here is that the most embedded clause, PROJ to go to the party, expresses the proposition that is true at a world-individual pair iff j goes to the party in w. When this is taken as an argument by the lower instance of want, the subject of go to the party is irreversibly linked to the subject of want. 5.3

Non-Principle B Violations

I began with the initially puzzling fact that a sentence like Sue wants to be entertaining cannot mean (for example) that Sue wants to be entertaining to herself. I suggested that the reason for this restriction is that the relevant reading would arise from a structure containing two instances of PROJ in a configuration that violates Principle B. This leads us to expect that similar readings should be possible as long as the two instances of PROJ are not in a Principle-B violating configuration. This prediction seems to be borne out, as we can see in examples such as (39). I assume that (39) has roughly the structure in (40). (39)

Sam wants to eat a cake that’s tasty.

(40)

Sam wants [S PROJ to eat [NP a cake [that’s tasty PROJ] ] ]

The sentence in (39) has a very natural reading on which Sam’s desire is to eat a cake that tastes good to Sam himself, which indicates that tasty can take PROJ as its implicit argument in this case. This is just as expected, since the second instance of PROJ in (40) is embedded within a relative clause, and thus the two instances of PROJ are not in a configuration that violates Principle B. In other words, (39) allows the relevant reading for the same reason that (41) allows a reading where him refers to Sam. (41)

Sami wants to eat a cake that tastes good to himi.

Relativism and the De Se Interpretation of PRO 6.

A Parallel with Epistemic Modals

Further evidence for my view comes from a parallel between controlled subjects and epistemic modals. As we will see, embedded epistemic modals share both the obligatory de se interpretation and the locality requirement seen in control. 6.1

The Judge-Dependent View of Epistemic Modals

In recent work (Stephenson 2006, to appear), I have argued that epistemic modals are judge-dependent items. In particular, I give the epistemic possibility modal might the meaning in (42). This uses a notion of epistemic alternatives, defined in (43), which is analogous to the doxastic alternatives and want alternatives which I have been using here. (42)

[[might]]w,j = [λp . ∃∈Epistw,j : p(w′)(x) = 1

(43)

Epistemic Alternatives: Epistw,x = { : it is compatible with x’s knowledge in w that x is y in w′}

This says that a sentence of the form “might p” is true at a world-individual pair iff p is true at some of j’s epistemic alternatives in w. For example, it’s raining is true at a pair iff it’s compatible with what j knows in w that it’s raining. The key difference between epistemic modals and predicates of personal taste on my account is that epistemic modals have judge dependency built right into their lexical meanings, whereas judge dependency arises with predicates of personal taste only when PROJ appears as an implicit argument. This has crucial implications for the interpretation of epistemic modals when they are embedded under attitude predicates. For example, the sentence in (44) is predicted to have only one possible reading, shown in (45). (44)

Sue thinks that it might be raining.

(45)

[[(44)]]w,j = 1 iff ∀∈ Doxw,Sue : [[it might be raining]]w′,x = 1 = 1 iff ∀∈ Doxw,Sue : ∃∈Epistw′,x : it’s raining in w″

This says that (44) is true iff Sue thinks that it’s compatible with her own knowledge that it’s raining. More generally, a sentence of the form “z thinks that might p” is true iff z thinks that it’s compatible with z’s own knowledge that p. I argue in my related work that this is correct (Stephenson 2006, to appear). It turns out that, like controlled PRO, epistemic modals embedded in attitude reports are both obligatorily de se and subject to a control-like locality requirement, as we will see below.

Stephenson 6.2

De Se Interpretation

We can see that embedded epistemic modals are obligatorily de se from examples like (46), which were brought to my attention by Pranav Anand (p.c.). (46)

[Context: Sam, a disgruntled NASA spokesperson, makes a false announcement on a radio show that there is new evidence of water on the moon. He is fired, goes home, and gets so drunk that when he hears a clip of his own interview on the radio, he doesn’t recognize it as himself. He says to himself, “That idiot doesn’t know that there’s no water on the moon!”] # Sam thinks there might be water on the moon.

In this context, Sam believes that it’s compatible with the knowledge of the man on the radio that there’s water on the moon, not realizing that the man on the radio is actually Sam himself. However, this cannot be reported as in (46). For (46) to be true, Sam must believe that it’s compatible with his own knowledge (understood de se) that it’s raining. This is correctly predicted by my analysis, which assigns to (46) the meaning in (47). (47)

[[(46)]]w,j = 1 iff ∀∈ Doxw,Sam : [[there might be water on the moon]]w′,x = 1 = 1 iff ∀∈ Doxw,Sam : ∃∈Epistw′,x : there’s water on the moon in w″

This says that (46) is true iff it’s compatible with the knowledge of all of Sam’s doxastic alternatives (so to speak) that there is water on the moon, which is de se. 6.3

A Locality Requirement

Epistemic modals also share a locality property of control. Consider (48), for example, where an epistemic modal is embedded in an attitude report that is itself embedded. (48)

Sue thinks that Bill thinks it might be raining.

In (48), what Sue believes is that Bill thinks that it’s compatible with his own knowledge that it’s raining. The sentence cannot mean that Sue thinks that Bill thinks that it’s compatible with her (Sue’s) knowledge that it’s raining. Again, this follows from my analysis. The meaning predicted for (48) is shown in (49). (49)

[[(48)]]w,j = [[think]]w,j ( [λw′ . [λj′ . [[Bill thinks it might be raining]]w′,j′ ] ] ) (Sue) = 1 iff ∀∈Doxw,Sue : [[Bill thinks it might be raining]]w′,x = 1 = 1 iff ∀∈Doxw,Sue : ∀∈Doxw′,Bill : ∃∈Epistw″,y : it’s raining in w′′′

This says that (48) is true iff, in all of Sue’s doxastic alternatives, it’s compatible with the knowledge of all of Bill’s doxastic alternatives (so to speak) that it’s raining. When the epistemic modal is embedded under think in the lower clause, it is irreversibly linked to

Relativism and the De Se Interpretation of PRO the subject of think (in this case, Bill), and further embedding can make no difference in its interpretation. 6.4

A Contrast: Shifting Indexicals

I conclude this section with a contrast. Shifting indexicals are items in certain languages (including Amharic and Zazaki) that seem to have indexical meanings such as I or now, but which can be shifted in the context of speech or attitude reports to refer to the speaker, time, etc. of the embedded context rather than the utterance context. It has been observed that shifting indexicals are obligatorily de se (see, e.g., Schlenker 2003, Anand 2006).5 However, in general they are not subject to the same kind of locality requirement as control. For example, consider the Zazaki case in (50). (50)

Ali Fatima-ra va kε Rojda Bill-ra va εz to-ra miradiša Ali Fatima-to said that Rojda Bill-to said I you-to angry.be.PRES [one reading] ‘Ali said to Fatima, “Rojda said to Bill that I am angry at you” ’ [Lit: Ali said to Fatima that Rojda said to Bill that I am angry at you] (Anand and Nevins 2004)

Sentence (50) has a number of possible readings (see Anand and Nevins 2004); on one of them, the indexicals ‘I’ and ‘you’ in the lowest clause are linked to the speaker and addressee of the higher embedding context (Ali and Fatima). I suggest that there are at least two separate kinds of de se items, some of which are derived via context shifting of the kind that has been proposed for shifting indexicals, and some of which are derived by relativist semantics.6 A locality requirement of the kind that applies to control is a diagnostic for the second kind. 7.

Conclusion

In this paper, I have proposed a new twist on a semantically based view of control. I treat control complements essentially as properties (like, e.g., Chierchia 1989), but bring in the extra individual argument by linking it to Lasersohn’s judge parameter. This makes the same key predictions as the property view with some additional advantages. First, all sentential expressions (including subjectless infinitives) share the same semantic type (namely, ). This in turn allows us to give a semantics for attitude predicates that fits seamlessly together with formal machinery independently needed for de se attitudes. Put together with the semantics for epistemic modals that I have proposed elsewhere, this view also captures two parallels in the behavior of controlled PRO and epistemic modals in attitude reports.

5

West African-type logophors pattern the same way (Schlenker 2003, citing work by Kusumoto and others). 6 Anand (2006) also argues for more than kind of de se item, for somewhat different reasons.

Stephenson References

Anand, Pranav, and Nevins, Andrew. 2004. Shifty Operators in Changing Contexts: Indexicals in Zazaki and Slave. Paper presented at Semantics and Linguistic Theory 14, Northwestern University, Evanston, Illinois. Anand, Pranav. 2006. De De Se. Doctoral dissertation, MIT. Chierchia, Gennaro. 1989. Anaphora and Attitudes De Se. In Language in Context, eds. R. Bartsch, J. van Benthem and P. van Emde Boas, 1–31. Dordrecht: Foris. Epstein, Samuel David. 1984. Quantifier-pro and the LF Representation of PROarb. Linguistic Inquiry 15:499–505. Lasersohn, Peter. 2005. Context Dependence, Disagreement, and Predicates of Personal Taste. Linguistics and Philosophy 28:643–686. Lewis, David. 1979. Attitudes De Dicto and De Se. Philosophical Review 88:513–543. Morgan, Jerry. 1970. On the Criterion of Identity for Noun Phrase Deletion. In Papers from the Sixth Regional Meeting of the Chicago Linguistic Society. Schlenker, Philippe. 2003. A Plea for Monsters. Linguistics and Philosophy 26:29–120. Stephenson, Tamina. 2006. A Parallel Account of Epistemic Modals and Predicates of Personal Taste. Paper presented at Sinn und Bedeutung 11, Universitat Pompeu Fabra, Barcelona. Stephenson, Tamina. 2007. Towards a Subjective Theory of Meaning. Doctoral dissertation, MIT, Cambridge, Massachusetts. Stephenson, Tamina. To appear. Judge Dependence, Epistemic Modals, and Predicates of Personal Taste. Linguistics and Philosophy. Department of Linguistics University of British Columbia Totem Field Studios Vancouver, B.C., Canada V6T 1Z4 [email protected]