No title

JOURNAL OF SEMANTICS Volume 23 Number 2 CONTENTS ANDREA BONOMI Truth and Reference in Context 107 ANASTASIA GIANNAKID...

Author: Oxford University Press

6 downloads 282 Views 673KB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

JOURNAL OF SEMANTICS Volume 23 Number 2

CONTENTS ANDREA BONOMI Truth and Reference in Context

107

ANASTASIA GIANNAKIDOU AND LISA LAI-SHEN CHENG (In)Definiteness, Polarity, and the Role of wh-morphology in Free Choice

135

ATSUKO NISHIYAMA The Meaning and Interpretations of the Japanese Aspect Marker -te-i-

185

Please visit the journal’s web site at www.jos.oxfordjournals.org

Journal of Semantics 23: 107–134 doi:10.1093/jos/ffh034 Advance Access publication November 2, 2005

Truth and Reference in Context ANDREA BONOMI Dipartimento di Filosofia, Universita` degli Studi di Milano

Abstract

1 TRUTH IN CONTEXT A context X, conceived of as a body of information that is supposed to be shared by the participants in a communicative exchange, plays two crucial roles in the semantics of natural languages. First of all, X serves to fix the content of an utterance U, since it represents the background against which U must be interpreted. Secondly, it provides the basis on which it is possible to build up the updated context, i.e. the context which results from adding this content to X itself. A fruitful formalization of these concepts is based on Stalnaker’s idea that a set X of shared propositions (the ‘common ground’2) is represented as a context set C, i.e. the set of possible worlds that are compatible with the propositions in X, and that the proposition p expressed by uttering a sentence S with respect to X is a subset of C, so that the update of C is obtained by eliminating from C the worlds that are not in p. 1

A number of people commented on earlier versions of the paper. In particular, I am grateful to F. Del Prete, A. Zucchi, P. Schlenker and the anonymous reviewers for this journal. 2 See Stalnaker (1999). In Stalnaker (2002: 701) the relationship between presuppositions and common ground is informally defined as follows: ‘To presuppose something is to take it for granted, or at least to act as if one takes it for granted, as background information—as common ground among the participants in the conversation.’ See also the definition of speaker presupposition in Soames (1982: 485). The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In communicative exchanges one of the most familiar phenomena is accommodation, which enables the addressee to incorporate a missing piece of information into her own view of the common ground. A less familiar, but equally important, phenomenon is what I call discommodation, whose main feature consists in the fact that the missing piece of information, although essential to the comprehension of the utterance, cannot be shared by the addressee because it sounds problematic or even false to her. In such cases it is possible to open a ‘presuppositional slot’ to take into account the assumptions that serve to select the reference of the noun phrase, but that are not incorporated into the revised context. One of the main purposes of the paper is to propose a definition of truth (with respect to a presuppositional apparatus) that does not ignore the role of discommodation when different views of the common ground are involved.1

108 Truth and Reference in Context If X is the presumed common ground and C is the corresponding context set (that is, C ¼ fw 2 W: w 2 p for every proposition p in Xg), we say that a sentence / is true with respect to X if and only if the proposition expressed by / with respect to C is compatible with the actual world @. More exactly:3 (T)

Given a context X, represented by the context set C, an utterance of the sentence / in X is true iff @ 2 [[/]](C).

ðTRANS1 Þ The P is Q 0 kCkw½CðwÞ ^ QðwÞðthe PðwÞÞ: Combined with this schema of translation, (T) tells us that ‘The P is Q’ is true, with respect to a context set C, iff the actual world @ belongs to C and in @ u is Q, where u is the only individual (if any) that satisfies the property P in @. Note that if X contains any wrong assumption, then @ ; C and, for every proposition p expressed in X, @ ; p (because p must be a subset of C): which means that no true proposition can be expressed in the context X. This is a minor problem, of course, if the false piece of information contained by the context at issue can be eliminated without serious consequences: which is possible if it has no significant role in determining the expected truth conditions (if, in particular, it has no role in picking out the intended reference of the definite description).6 The obvious solution, in this case, is to replace the ‘deviant’ context with a new one, where the false proposition is simply eliminated. In 3 Alternative definitions are possible in a dynamic framework. In (T) the role of the common ground is made explicit, as desired. 4 The meaning of / with respect to an assignment g. I will ignore the reference to an assignment, unless required by the context. 5 That the proposition expressed by / with respect to C be a subset of C is a natural requirement in the theoretical framework under discussion: ‘What the context provides is the domain of possible worlds that propositions distinguish between. Semantics in general gives us rules for picking a subset of situations from such a domain. On this kind of account, context is not just information that mediates between utterance and proposition; it is the material out of which propositions are constructed.’ (Stalnaker, 1999: 156.) 6 For example, imagine a situation where it is correctly presupposed that yesterday, at a party, only one person (i.e. Leo) wore a fez. Yet, because of a false piece of information provided during the conversation, it is also presupposed that fezzes were formerly worn by Afghan men. In this case we have no difficulty to admit that, since Leo really is a poet, an utterance of the sentence ‘The man wearing a fez is a poet’ is intuitively true, even if it is false that fezzes were formerly worn by Afghan men. A similar problem is discussed in Heim (1982: 337–338).


Here, [[/]] is the ‘meaning’4 of /, which is a function from context sets (i.e. sets of possible worlds) into propositions (i.e. once again, sets of possible worlds), such that [[/]](C) 4 C.5 So, a (simplified) translation schema for a sentence like ‘The P is Q’ is given by the rule:

Andrea Bonomi 109

other terms, C must be replaced by CX,@ ¼ fw 2 W: w 2 p for every proposition p in X such that @ 2 pg. Thus, if X is any context and C the corresponding context set, CX,@ can be seen as the result we obtain after eliminating the false proposition(s) contained in X. As a consequence, definition (T) must be replaced with the following definition: (T@) Given a context X, represented by the context set C, an utterance of the sentence / in X is true iff @ 2 [[/]](CX,@).

2 IMPORTED PRESUPPOSITIONS As we have just seen, the context shift on which (T@) is based is made possible by a simple contraction of the common ground. But there are situations in which a false presupposition cannot simply be eliminated, because, unlike the wrong assumptions mentioned in the previous section, it plays an essential role in determining the reference of the definite description and the content of the utterance. Thus, it must be replaced by a new piece of information. This point can be illustrated by the following example. Case A. The false presupposition that Tom is the only man wearing a fez is part of the common ground among the participants in a conversation during a party. (Actually, the man with the fez is Leo, a distinguished poet.) At a certain point, Lea joins in the conversation and utters the sentence (1)

The man wearing a fez is a poet.

From a number of hints, it is clear that she is referring to Leo. And since everybody knows that Lea is very well informed about the situation, the old presupposition is cancelled and the assumption that Leo is the man wearing a fez is tacitly incorporated into the common ground. So, Lea succeeds in communicating the true statement that he is a poet. In general, in communicative exchanges, the common ground is affected not only by what the participants assert, but also by what they show to presuppose. Different views of the common ground may


If @ 2 C (which happens when X contains only true presuppositions), then CX,@ ¼ C: in this case, as desired, (T@) is equivalent to (T), and there is a vacuous context shift. To sum up, with this adjustment of truth conditions it would be possible to take care of the presence of false but inessential assumptions.

110 Truth and Reference in Context

(i) S is totally ordered by 4 (ii) W 2 S (as a consequence, W is the largest element in S) (iii) C 2 S and, for any B in S, C 4 B (i.e. C is the 4-minimum of S) (iv) For any proposition p: if there is any sphere B in S such that B \ p 6¼ ˘, then there is a smallest sphere B# such that B# \ p 6¼ ˘. (This is the limit assumption discussed by Lewis).8 In virtue of (i)–(iv), a system of spheres S centred on C can be associated with a function fC : PðWÞ/PðWÞ which is defined as follows: for any p 2 PðWÞ: fC(p) ¼ W if B \ p ¼ ˘ for every B in S fC(p) ¼ E \ p, where E is the smallest sphere in S such that E \ p 6¼ ˘.

7

See Grove (1988) for the modified version of Lewis’s system that is adopted here. To avoid unnecessary complications, I will consider only those contexts that are logically consistent, even though they may contain false propositions. Removing this assumption would entail a complication of the semantics that is not relevant here. See Arlo´-Costa (2002) for a treatment of propositions that would allow us to deal with the case of inconsistent contexts. 8 Lewis (1973) points out some problems for this assumption. Yet, his counterexamples are based on the fact that his system is centred on the real world @, so that whenever we assume that a given sphere B is the smallest sphere such that B \ p 6¼ ˘, it is always possible to find worlds that are closer to @ than those in B. As a consequence B cannot be considered as the smallest sphere with those characteristics any longer. It should be noticed, however, that in the present framework a system of spheres is centred, in general, on a context set A that represents a set X of presupposed propositions. So it seems reasonable, in this case, to assume that worlds which are too finely individuated to be discernible with respect to the context X of presuppositions count as equally ‘close’ to C. See, on this point, Bonomi and Zucchi (2003).


influence each other: what is tacitly presupposed by a speaker may be incorporated into the presuppositional apparatus of another speaker, and the new information may be incompatible with the old one. Thus, a more complex type of context shift is needed. To account for such situations I will assume that our truth conditions, where contexts are treated as set of possible worlds, are associated with a suitable ordering on worlds. There are several ways of formalizing such an ordering. Here, I will adopt a ‘system of spheres’ which is a modification7 of the one introduced by Lewis (1973) and which is defined as follows. Let C be any subset of W (the set of possible situations). A collection S of subsets of W is a system of spheres centred on C if it satisfies the following conditions:

Andrea Bonomi 111

(TINC) Given a context X, represented by the context set C, an utterance of the sentence / in X, with additional presuppositions represented by p, is true iff @ 2 [[/]]( fC(p)). In our example, C represents the given context and p Lea’s presuppositions. Since the worlds in fC(p) are p-worlds the identity ‘Leo ¼ the man wearing a fez’ is true in all these worlds: which means that these presuppositions are satisfied and that Leo is the intended referent of the definite description, as desired. It should be noticed that the process of incorporation defined in (TINC) is general enough to cover other cases of context revision: in particular, the familiar phenomenon of accommodation. The idea is that in this case p contains information not present in C but compatible with it (which in general is not true, as illustrated by case A), so that fC(p) ¼ C \ p. To sum up, (TINC) accounts for a kind of context shift where some essential assumptions must be added to the basic common ground because they are missing (accommodation) or replace false presuppositions (like in case A). 3 POINTS OF VIEW So far, we have endorsed a very naı¨ve notion of common ground or context, for we have tacitly characterized it as the set X of the beliefs or attitudes that are really shared by the participants in a communicative exchange. Yet, even if these beliefs or attitudes are limited to those that


Intuitive meaning. The smallest sphere C (on which the system is centred) can be seen as a context set, representing a certain background of assumptions, i.e. a set X of propositions. Given two worlds w and w#, if there is a sphere B such that B contains w but not w#, we can say that w is ‘closer’ to C than w#, i.e. closer to the idea of world associated with X. Thus, fC(p) is the set of p-worlds which are ‘maximally’ close to the background X and, in the situations I am going to discuss, it can be interpreted as the result of the minimal revision of X under the assumption p. The idea is that the only admissible modifications, in X, are those strictly necessary to incorporate p. In the case of a context shift like the one illustrated by example A, where false presuppositions are replaced by true propositions, I shall speak of truth through incorporation, and the relevant definition might have this form (where p represents a set of veridical presuppositions associated to an external source of information):


(UPINC) (i) Given b’s relevant presuppositions, represented by the context set C, and b’s conjectures about a’s relevant 9 In Beaver (2001: 236–249) an agent’s uncertain knowledge of the common ground is formally treated as an ‘information set’ (i.e. a set of information states: the idea is that ‘each state in the set corresponds to a possibly correct model of [the agent’s] assumptions about the common ground’). 10 See Stalnaker (1978: 85) for a definition of defective contexts: ‘Each participant in a conversation has his own context set, but it is part of the concept of presupposition that a speaker assumes that the members of his audience presuppose everything he presupposes. We may define a NONDEFECTIVE CONTEXT as one in which the presupposition of the various participants in the conversation are all the same. A DEFECTIVE CONTEXT will have a kind of instability, and will tend to adjust to the equilibrium position of a non defective context [. . .] So it is not unreasonable, I think, to assume that in the normal case contexts are nondefective, or at least close enough to being nondefective.’ Unlike Stalnaker, I consider defective contexts not as a pathology to be ignored, but as an intriguing phenomenon characterizing the dynamics of communicative exchanges. As emphasized in Donnellan (1966), this is evident in the so-called referential use of a definite description, because ‘it is perfectly possible for our audience to know to whom we refer, even though they do not share our presupposition’. I will address this problem in section 7.


are strictly relevant to the current communicative exchange (a restriction that will be taken for granted in the present discussion), such a local characterization of the common ground is still empirically inadequate for the following reasons. Whilst it is quite natural to think that X is not empty (for, in any communicative exchange, there are, at the worst, shared beliefs about some public facts connected to the exchange itself: e.g. the fact that someone is speaking, or the fact that someone is listening, and so on), it is also plausible to admit that, in general, the agents have only a partial and hypothetical view of X.9 This is so, of course, because nobody has a direct access to the beliefs of the other participants and, as a consequence, each of them can only make conjectures about the background of shared beliefs, without knowing what exactly it is in many circumstances. As we shall see, in some cases there can be a conflict between different views of this background.10 In the light of these considerations, (TINC) can be reformulated in order to account for a kind of context revision that enables an addressee to modify her own view of the common ground by making room for the speaker’s presuppositions and to accept (or reject) the content determined by these presuppositions. So, we might call this process update through incorporation (UPINC), because the speaker’s presuppositions become an integral part of the revised context to which the content of the utterance must be added, and if we concentrate on the role of definite descriptions, which is the main topic of the present paper, we get the following definition (where a is the author of the utterance and b the addressee):

Andrea Bonomi 113

presuppositions, conjectures represented by the context set p, a proposition q is taken for granted by b, with respect to C and p, iff, for every w 2 fC(p), w 2 q; (ii) by virtue of definition (TRANS1), the update of the context fC(p) with the utterance of the sentence ‘The P is Q’ is the set of worlds w such that w 2 fC(p) and in w u is Q, where u is the only individual (if any) that satisfies the property P in w.


According to this picture, b has her own presuppositions (represented by the context set C) and her own view of what is presupposed by the speaker a (represented by the context set p), and, in order to grasp what a says (and to accept it as a true statement), she assimilates the relevant presuppositions she attributes to him through a minimal revision of her own presuppositions. In other terms, (UPINC) expresses two crucial steps in b’s interaction with a: (i) revision of her own presuppositions to make them consistent with the presuppositions that she attributes to a: the result is the new context fC(p), which includes these presuppositions and which is used to determine the proposition expressed by a’s utterance; (ii) update of the context fC(p) by eliminating all the possible situations where this proposition turns out to be false. Of course, the complexity of this revision process depends on the compatibility of a ’s presuppositions with b’s presuppositions. For example, as in case A, imagine that b is convinced that the man wearing a fez is not Leo but Tom. Once she has realized that a presupposes that the man with the fez is Leo, she may want to revise her own presuppositions (provided that she thinks that a is a reliable source of information on this subject). In this case fC(p), which represents the result of a revision made possible by incorporating a’s presupposition, is not a subset of C, which represents b’s original presuppositions, since the assumption that the man wearing a fez is Leo is not compatible with C. A simpler illustration of (UPINC) is provided by accommodation proper. In this case we have a ‘smoother’ type of context revision based on incorporation, because the proposition presupposed by a is consistent with b’s presuppositions, so that the revised context fC(p) is simply a subset of C. Since it is this revised context that is updated by adding the content of the utterance, I will speak, in such cases, of update through accommodation. In other words, accommodation is conceived of as a particular (and simple) case of context revision based on incorporation, whose general definition is given in (UPINC).

114 Truth and Reference in Context 4 DISCOMMODATION

Case B. Imagine that a and b are speaking of Proust’s Recherche and that, to emphasize the fictional nature of this work, a utters the following sentence: (2)

No famous novelist has ever expired while visiting an exhibition of Swann’s favourite painter.

In this scenario, the truth of the statement depends on two facts: (i) on the one hand, the reference of the definite description is fixed with respect to the world(s) of the Recherche (where Swann’s favourite painter is Vermeer), not with respect to the world of evaluation (where Swann does not exist); (ii) on the other hand, it is in this world that no famous novelist has ever expired while visiting the exhibition at issue, not in the world(s) of the Recherche (where (2) turns out to be false because Bergotte, a famous novelist, expires while visiting the exhibition at issue).12 But these two requirements cannot be satisfied by (TINC), where the world of evaluation must belong to the set of worlds that are 11 In Bonomi (1998) an example similar to (2) is analysed in a formal framework that is a little different from the one adopted here. See Fauconnier (1986) and Recanati (1996) for other examples that seem to require what I call a ‘local’ context shift. Predelli (2000) contains an interesting discussion of the sentence ‘The person who guided Dante through Hell and Purgatory is one of my favourite poets’. 12 This means that a possible paraphrase of (2) is (2a): ‘No famous novelist has ever expired while visiting an exhibition of the artist who in the Recherche is Swann’s favourite painter’, but not (2b): ‘In the Recherche no famous novelist has ever expired while visiting an exhibition of Swann’s favourite painter’. The semantics of statements such as (2b) is discussed in Bonomi and Zucchi (2003).


As we have just seen, this kind of strategy is based on the fact that the ‘external’ presuppositions represented by p are incorporated into the revised context fC(p). In other words, these presuppositions become an integral part of the new context that is used to identify and evaluate the content of an utterance. Thus, at the end of this process of incorporation, a single context is relevant here. To use a metaphor, one might say that truth and reference share the same destiny, in the sense that fC(p) is the context set to which the world of evaluation must belong and, at the same time, the context set which serves to fix the intended reference of a part of the sentence (the definite description). Witness example A, in some cases such a strategy (based on the assimilation of an external presuppositional apparatus) is quite appropriate. Yet, apart from the cases of misdescription that will be discussed in a moment, the need for an alternative strategy seems to be motivated by independent reasons. To see this, consider the following example:11

Andrea Bonomi 115

relevant for fixing the reference of the definite description (i.e. the set of worlds compatible with what is narrated in the novel). A similar inadequacy can be ascertained in the case of (UPINC): since it is based on the idea that the speaker’s presuppositions are incorporated into the addressee’s view of the common ground, it cannot account for situations where the addressee is not willing to share these presuppositions. This is what happens, for instance, in the case of misdescriptions, witness this new example:

(3) a: The man wearing a fez is a poet. b: I am happy to learn that he [¼ Leo] is a poet. But his hat is not a fez. What is peculiar to this situation is the fact that a and b have different views of the common ground, for what is presupposed by a is questioned by b. But this fact does not prevent b from referring to Leo in the first sentence of her answer, for the simple reason that b takes into account the (false) assumption that Leo is the man wearing the fez as a tool to pick out the intended reference of the definite description. Thus, under the assumption that a (in spite of his confusion about Leo’s hat) is a reliable source of information about Leo’s literary activity, b can update her own view of the common ground by adding the information that Leo is a poet, without sharing the false presupposition that his hat is a fez (witness the second sentence in her answer). For reasons that will be clear in a moment, I will call such a process update through discommodation. 13 This kind of example is discussed in Kripke (1979), who judges a dialogue like (3) perfectly acceptable. According the analysis developed in Donnellan (1978) the pronoun ‘he’, in (3b), is anaphoric: which would prove, according Donnellan, the semantic relevance of the speaker’s reference. This thesis is rejected by Soames (1994). Pronominal contradiction is extensively discussed in van Rooy (2001), who accounts for Donnellan’s intuition in the framework of dynamic semantics.


Case C. Sentence (1) (i.e. ‘The man wearing a fez is a poet’) is uttered by a, who is sure that Leo is the only person wearing a fez. Unfortunately, this presupposition turns out to be false (whilst it is true, as before, that Leo is a poet). b, who is provided with independent information, knows that the only person wearing a fez is not Leo but Theo, and she is also convinced that everybody at the party shares this presupposition. This is why a’s use of the definite description surprises her. In such a situation this is a possible dialogue between a and b:13

116 Truth and Reference in Context I have assumed for simplicity that, unlike a, b is provided with true information. But this is not crucial in such situations. We can even imagine a situation where a uses a definite description (e.g. ‘The man wearing an elegant hat’) with a vague predicate like ‘elegant’, with respect to which a and b adopt different standards. In these cases, it would be misleading to speak of ‘false’ or ‘true’ presuppositions, although a dialogue similar to (3) is still possible: (4)

a: The man wearing an elegant hat is a poet. b: I am happy to learn that he [¼ Leo] is a poet. But his hat is not elegant.

ACCOMMODATION: Permanent14 expansion of b ’s view of the common ground, in order to determine the content of the utterance; this expansion is obtained by adding some new piece of information which b judges to be presupposed by the speaker a and which b is willing to take for granted without objection.15 DISCOMMODATION: Temporary suspension of b ’s view of the common ground, in order to determine the content of the utterance; this suspension is required to take into account some new piece of

14

At least in the context of the communicative exchange. See Stalnaker’s discussion of the example ‘I can’t come to the meeting. I have to pick up my sister at the airport’ uttered in a context where the addressee does not know that the speaker has a sister. In Stalnaker’s words: ‘The phenomenon of accommodation, in general, is the process by which something becomes common ground in virtue of one party recognizing that the other takes it to be common ground.’ (Stalnaker, 2002: 711.) It should be noticed that what Stalnaker describes here is the effect on the real common ground of the parallel processes of revision (assuming that these processes are successful) that the different agents achieve with respect to their own views of the common ground. These processes are one of the main topics of the present paper. 15


In general, such situations are characterized by the presence of a disagreement between a and b, and, at the same time, by the willingness of b to suspend her assumptions temporarily in order to grasp what a wants to say. Thus, in dialogue (3), b is perfectly aware that the man wearing a fez is not Leo, but she takes into account a’s point of view to determine the proposition he intends to express and to update her own view of the common ground with this proposition (always admitting that she judges a well informed about Leo’s literary activity, although misinformed about Leo’s hat). As I have already suggested, this phenomenon is, in a sense, symmetrical with respect to accommodation. For this reason I will speak of discommodation, which can be characterized as follows (with respect to a speaker a and an addressee b):

Andrea Bonomi 117

information which b judges to be presupposed by the speaker a and which b is not willing to take for granted.

5 PRESUPPOSITIONAL SLOTS In the present theoretical framework there is no problem in accounting for accommodation, which is a very simple illustration of the process of context revision described in (UPINC ), a process based on the incorporation of the relevant presuppositions. The result of this process is represented by the new context fC(p), which is simply a subset of the given context C. Discommodation is more intriguing, because in this case p, the proposition which b judges to be presupposed by a, must not be


In these intuitive definitions, what is crucial is not the common ground as such (i.e. as a set of shared beliefs or attitudes), but, on the one hand, the views of the common ground that different agents have, and, on the other hand, their willingness to modify these views, if necessary, in order to determine the content of an utterance. This can be done in two ways: by adding information (if this information does not sound problematic), or by a temporary suspension of pre-existing information in order to make room for information which sounds problematic, but which is required to make the continuation of the exchange possible. The former situation can be described as a case of merging between different (but compatible) views, whilst in the latter situation these views are kept separate. The term ‘temporary’ is necessarily vague here, for the addressee may suspend her own presuppositions just for an instant (as in the examples (3b) and (4b), where b takes into account a’s point of view only to grasp the proposition he wants to express: a point of view which is immediately questioned in the second sentence of the answer), or for a longer period. Imagine, for instance, that, in the situation described in case C, b is not concerned with the pedantries about Leo’s hat and that, to be cooperative, she refrains from uttering the second sentence, i.e. the sentence which expresses her objection to the way a uses the definite description. In this case a’s (false) assumption can remain in force until the end of the exchange. Needless to say, such a co-operative attitude is highly motivated in all situations in which there is no way to settle which of the competing views is the correct one, as in example (4). To sum up, both cases (accommodation and discommodation) can be described in terms of a context shift, which involves different views of the common ground. What changes is the nature of this operation, as we shall see in the next section.



incorporated into the revised context. Unfortunately, such a requirement cannot be satisfied by (UPINC ), which does not allow us to keep the worlds relevant for truth separate from the worlds relevant for reference. But such a distinction is exactly what is necessary to account for discommodation, because in this case p, the presupposition which determines the intended reference, is not compatible with b’s assumptions, so that the worlds in which the content of a ’s utterance should be evaluated as true or false cannot be p-worlds. This is a general problem that does not concern only the cases of misdescription, witness the Proustian example discussed in section 3. Intuitively speaking, what we do, when we judge (3a) or (2) to be true, is to resort to the ‘counterfactual’ context of the speaker’s presuppositions or to the ‘counterfactual’ context of the Recherche in order to pick out the intended reference of the definite description, whilst we resort to the real world (or what we assume to be the real world) in order to evaluate the whole sentence. The problem is that in a global context shift the worlds of the revised context are used to determine both the reference of the definite description and the truth-value of the sentence. This is perfectly appropriate to account for a situation where new assumptions are incorporated into the relevant presuppositional apparatus. But if one reflects on the cases of ‘misdescription’ (in a broad sense, so as to include (2b), where the reference of the definite description is determined with respect to counterfactual worlds) it is easy to see that a different analysis is required. Take for instance (4a). When accepting this statement as true, b is perfectly aware of the fact that she and a have different opinions about the reference of the definite description ‘the man wearing an elegant hat’. Thus, on the one hand, she has no reason to give up her own suppositions about the person who satisfies the definite description. On the other hand, she takes into account a’s beliefs on the matter in order to grasp the content of his utterance and to update her own view of the common ground with that content, for she thinks that a is a reliable source of information about Leo’s literary activity. It is as if, in processing the utterance made by a, b should open a temporary slot within her own presuppositional apparatus in order to take into account a’s presuppositions. This effect can be described as a local context shift, which involves only the definite description at issue (or, in general, the NP at issue), and a little modification of the translation schema proposed at the outset is sufficient to account for it. The idea is that what we need is not only a contextual variable associated with the whole sentence, but also a specific one, associated with the definite description. Thus, since the embedding context

Andrea Bonomi 119

(relevant for the evaluation of the whole sentence) may not coincide with the embedded context (relevant for the reference of the NP), the new translation schema must be something like: ðTRANS2 Þ The P is Q 0kCkw½CðwÞ ^ QðwÞðthe PðKÞðwÞÞ

(DESC)The meaning of the definite description ‘The P’ is the function [[The P]] from contexts to identifying functions (i.e. functions from worlds to individuals) which is defined as follows: (i) if there is an individual u that in every world in C is the only individual satisfying P, then [[The P]](C) is the constant function f such that, for any w 2 W, f(w) ¼ u; (ii) otherwise, [[The P]](C) is the partial function g such that, for any w 2 W, g(w) is the only individual that satisfies P in w (g(w) is indefinite when, in w, P is satisfied by no individual or by several individuals). We are now in a position to sketch a global solution to the problems raised by discommodation phenomena. The case of update through discommodation, illustrated by dialogue (3), corresponds to the second option made possible by (TRANS2): the value of the free variable is the context set representing b’s conjectures about a’s relevant 16

See, on this point, the discussion on the referential/attributive distinction in Stalnaker (1970). I will come back to (DESC) in section 8. It should be noticed that adopting this kind of definition is not essential for the general strategy I have sketched to deal with discommodation phenomena. Alternative treatments of definite descriptions, in particular a Russellian one, might be adopted. As for other singular terms (namely, proper names and demonstratives), defining their meaning in the theoretical framework under discussion is an issue that cannot be addressed in the present paper.


where K is a free variable whose value depends on the assignment p. In the default cases, p(K) ¼ R (where R is the evaluation context). Thus, the embedding context and the embedded context coincide, and we have something equivalent to (TRANS1). But if K is given a different value (namely the speaker’s context S), then the embedding context and the embedded context can be different. Of course, a third option is theoretically possible: the variable might be bound by a suitable quantifier, e.g. an adverb of quantification. Thus, the question is: is such an option empirically justified? If it is (as I think) the presence of a free variable in the relevant logical form would be justified by reasons that are independent of the phenomena of accommodation analysed here. I will address this problem in section 9. To be more specific about the main topic of this paper, suppose that the meaning of a definite description is defined by the following (simplified) rule of interpretation:16

120 Truth and Reference in Context presuppositions. This means that the embedding context is distinct from the embedded one. Intuitively speaking, the idea is that, as shown by this particular use of the pronoun ‘he’, b wants to update her own view of the common ground with the proposition that Leo (i.e. the reference of the definite description ‘The man wearing a fez’ according to the presuppositions that b attributes to a) is a poet. So, if R stands for b’s presuppositions and S for the relevant presuppositions she attributes to a, the proposition by which b wants to update R is given by: (5)

In other words, the reference of the definite description is fixed with respect to S, not with respect to R. Since [[The_man_wearing_a_fez]](S) is the constant function f such that, for any w, f(w) ¼ Leo, (5) denotes the proposition that b is willing to accept in order to update her own view of the common ground, i.e. the proposition that Leo is a poet. This attitude is made explicit in (3b), where b expresses her willingness to accept that proposition, while questioning the way in which one of its constituents (i.e. Leo) is identified. But consider this other dialogue between a and b:17 (6)

a: The man wearing a fez is a poet b: No, he is not a poet. The man you are referring to does not wear a fez.

This time, witness the second sentence in (6b), b sticks to her own presuppositions, so that the individual she refers to by using the pronoun ‘he’ is not Leo, but Theo. In other words, the referent of the definite description is picked out with respect to R itself, not to S. To account for this situation we can resort to the first option made possible by (TRANS2), the default case, and the value of the variable is the evaluation context. The idea is that the embedding context and the embedded context coincide, and what we obtain is the following proposition: (7)

17

kCkw[C(w) ^ Poet(w)(The_man_wearing_a_fez(K)(w))](R) ¼ kw[R(w) ^ Poet(w)(The_man_wearing_a_fez (R/K)(w))] where R/K means that R is the value of K for the given assignment p.

This is a second type of dialogue that Kripke (1979) judges acceptable in a case of misdescription. Unlike (3), based on what he calls the speaker’s referent, (6) is based on the semantic referent of the definite description.


kCkw[C(w) ^ Poet(w)(The_man_wearing_a_fez(K)(w))](R) ¼ kw[R(w) ^ Poet(w)(The_man_wearing_a_fez (S/K)(w))] where S/K means that S is the value of K for the given assignment p.

Andrea Bonomi 121

18 See Stalnaker (1999: 156). In the present theoretical framework, the translation of a sentence such as ‘x believes that /’ would be something like

kCkw½CðwÞ ^ "w#ðBx;w ðw#Þ//ðBx;w Þðw#ÞÞ where Bx,w, i.e. the set of worlds compatible with x’s beliefs in w, is the derived context (relative to w) for the embedded sentence /. Note that this derived context is not included in the basic context C if x holds beliefs that are not compatible with w.


As desired, this time the reference of the definite description is not Leo, because the relevant context is R (i.e. b’s presuppositions) and [[The_ man_wearing_a_fez]](R) is the constant function f such that, for any w, f(w) ¼ Theo. This is exactly the situation described by dialogue (6), where b is not willing to update her own view of the common ground with the proposition at issue, i.e. the proposition that Theo is a poet. A methodological remark is in order at this point. Usually the need for a context shift is justified by the presence, in a given sentence, of an (implicit or explicit) operator that makes the ‘basic’ context inappropriate to evaluate the embedded sentence. This is the case, for instance, of a sentence such as ‘x believes that /’, uttered in the context X, where the proposition denoted by ‘that /’, which is relevant to evaluate the whole sentence, cannot be defined as a subset of the basic context X. This is why, to obtain this proposition, we need a derived context.18 So, the natural question, at this point, might be the following: what is the difference between these familiar cases of context shift and the kind of context shift that should account for the discommodation phenomena analyzed here? A first difference is that in the former case the context change is triggered by the occurrence, in the sentence at issue, of an implicit or explicit operator. On the contrary, in the latter case the need for a context shift is simply determined by the existence of a tension between different views of the common ground: a tension that might undermine the exchange of information. It is in this sense that one might speak of a spontaneous context shift, motivated not by the presence of a suitable operator, but by the simple intention of keeping the possible discrepancies under control. Thus, what is essential here is not the internal articulation of a complex sentence, but the complexity of the utterance context, which involves a plurality of agents (with different views of the common ground, in the relevant situations). A second, and crucial, difference is that in the familiar cases the derived context is relevant for evaluating the sentence as a whole (no discrimination between truth and reference), whilst to account for discommodation phenomena the context relevant for truth and the context relevant for reference may be different.

122 Truth and Reference in Context 6 TRUTH IN CONTEXT (REVISED)

(TC)

Given a context X, represented by the context set C such that @ 2 C, and given a presuppositional saturation p, an utterance of the sentence / in X is true with respect to p iff @ 2 [[/]]p(C).


We have just seen how the translation schema (TRANS2) can help to account for the notion of update through discommodation. The same happens with the parallel notion of truth through discommodation, which is the problem discussed at the outset. The distinction between the embedding context, which is relevant for evaluating the sentence as true or false, and the embedded context, which is relevant for fixing the reference of the definite description, allows us to solve the problems raised by misdescriptions (in the broad sense of the term). Consider again dialogue (6), where, according to Kripke’s analysis, the first sentence in (6b) is about the semantic reference of the definite description occurring in (6a), i.e. about Theo. In the theoretical framework adopted here, this means that the embedding context must coincide with the embedded one, and that it must be compatible with the real world. In particular, in the case of (6), it must contain the correct information that there is only one person wearing a fez and that this person is Theo. Let R be such a context. Thus, the utterance of the sentence ‘The man wearing a fez is a poet’ in (6a) is simply false, with respect to R, because the reference of the definite description is not Leo, but another person who is not a poet. On the contrary, in the case of dialogue (3), which is about the speaker’s reference, the utterance of the sentence ‘The man wearing a fez is a poet’, in (3a), turns out to be true, because this time the reference of the definite description (i.e. Leo) is fixed with respect to the embedded context S, representing the speaker’s assumptions, and distinct from the embedding context R. To sum up, the two options made available by (TRANS2) correspond to two different ways of determining the reference of the definite description, according to whether the embedding context and the embedded one coincide or not. When the value of the context variable associated with the definite description is R itself (where R has the characteristics specified above) we get the interpretation based on the semantic reference; when the value of the variable is the background of the speaker’s presuppositions, we get the interpretation based on the speaker’s reference. Neglecting the details, this notion of truth in context can be generalized in the following way:

Andrea Bonomi 123

(TC@) Given a context X, represented by the context set C, and given a presuppositional saturation p, an utterance of the sentence / in X is true with respect to p iff @ 2 [[/]]p(CX,@).

7 DONNELLAN’S DILEMMA In his 1966 paper, Donnellan distinguishes two uses of a definite description like ‘the so-and-so’. In the attributive use, the speaker ‘states something about whoever or whatever is the so-and-so’, whilst in the referential use the speaker ‘uses the description to enable his audience to pick out whom or what he is talking about and states something about that person or thing. In the first case, the definite description might be said to occur essentially [. . .]; but in the referential use the definite description is merely one tool [. . .] for calling attention to a person or thing’ (Donnellan 1966: 285). This is why, in this second use, we may


Here, p is the given assignment of contexts to context variables and [[/]]p is the ‘meaning’ of / with respect to the parameter p. The presence of a free variable makes the final interpretation depend on the value of this variable. When p(K) is the evaluation context C (the default case), C is relevant both for truth and reference and no context shift is required. By contrast, if p(K) is not the evaluation context, then the embedding context and the embedded context do not coincide, and this means that the reference of the description is not fixed with respect to C. This time C is relevant for truth but not for reference. One last remark. In (TC) it is explicitly assumed that the basic context C is a ‘true’ context, i.e. @ 2 C. This means that there are no false presuppositions. Such an assumption is partially justified by the fact that the cases of misdescription can now be treated by resorting to the distinction between the context that is relevant for truth and the context that is relevant for reference. Yet, we should not forget that, independently of the presence of misdescriptions, C might contain false, although irrelevant, assumptions. I have ignored this problem because misdescriptions are the main topic of the present paper, and (TC) seems to offer an adequate solution in this connection. In any case, as shown at the outset, the problem of ‘irrelevant’ mistakes might be solved by adopting in (TC) a ‘harmless’ context shift like the one adopted in (T@). Accordingly, in the definition proposed in (TC) [[/]]p should be applied not to C but to CX,@, as defined in connection with (T@) and we would obtain the following definition:


19 He says that in these cases ‘the speaker’s reference determines the semantic reference’ or that ‘the speaker reference appears necessary to provide semantic reference.’ (Donnellan, 1978: 55, 61.) Note that this use of the expression ‘semantic reference’ to speak of the relevance of the intended individual for the truth conditions of the sentence does not coincide with Kripke’s use of the term ‘semantic referent’, which applies only to the thing, if any, that does satisfy the definite description. Kripke’s ‘semantic reference’ corresponds to Donnellan’s ‘denotation’ (of a definite description), even though there is some fluctuation in the use of these terms in Donnellan’s paper. 20 In Soames (1994) several arguments are presented to reject the idea that, in such cases, the pronoun ‘he’ is an anaphoric element. This issue deserves a separate discussion that cannot be developed here. 21 ‘Let us say that a definite description is uttered in a referential context when speaker reference exists relative to it [. . .] A definite description will be uttered in an attributive context when speaker reference relative to it is absent.’ (Donnellan, 1978: 53.)


succeed in picking out this person or thing ‘even though he or it does not really fit the description’. Some years later, Donnellan addresses a different, but related, problem. Consider, once again, case C, where the referential use of the definite description ‘the man wearing a fez’ is intended to denote Leo, whose hat is not a fez (even though everybody is convinced that it is). Actually, the only person wearing a fez is Theo. Since, according to Donnellan, we are willing to admit that, by uttering sentence (1) (i.e. the sentence ‘The man wearing the fez is a poet’) a has said something true of Leo, he wonders whether this is just a pragmatic phenomenon or, on the contrary, the speaker’s reference (Leo, in this case) has a semantic role to play. If it has such a role, the truth conditions of (1), in these circumstances, do not depend on Theo’s properties, but on Leo’s properties (namely, the property of being a poet). And if we adopt a Russellian characterization of propositions, this idea might be formulated by saying that Leo, the speaker’s reference, is a constituent of the proposition expressed by (1) in the context at issue. It is in this sense that, according to Donnellan, the speaker’s reference has a ‘semantic significance’,19 and that the truth-value is determined by the properties of the speaker’s reference. Indeed, he mentions some examples where, as in our dialogue (3), the reference of the anaphoric20 pronoun ‘he’ is the speaker’s reference, as we should expect, he thinks, if we assume that the speaker reference has a semantic relevance, that is a role to play in the truth conditions of the sentence. Let us sum up. On the one hand, there is the attributive/referential distinction, which many theorists are reluctant to treat as a semantic distinction (for they are reluctant to admit that the definite article is ambiguous). On the other hand, from Donnellan’s point of view, the presence of the speaker’s reference is what characterizes the referential use of a definite description, not its attributive use.21 And since, according to him, the speaker’s reference has a semantic role to play, as

Andrea Bonomi 125

8 ESCAPING THE DILEMMA There is a fair amount of discussion about the semantic relevance of the speaker’s reference and its relationship with the attributive/referential distinction. In particular, a number of authors have endorsed Donnellan’s dilemma, by rejecting22 the thesis of the semantic relevance as entailing the ambiguity of the definite article or, on the contrary, by accepting that thesis and the semantic ambiguity that it is believed to entail. In what follows, I will refrain from reviewing the relevant literature and I will confine myself to considering the attributive/referential distinction in connection with the phenomena of discommodation analysed in the present paper. Part of the problem, of course, might depend on terminological issues. Yet, if by ‘semantic ambiguity’ one means that the semantics of the definite article should include two distinct rules of interpretation, then the theoretical framework adopted here suggests a possible way 22 As far as I can judge, this is officially Kripke’s position. According to his analysis, there is no need to admit the semantic relevance of the speaker’s reference, because the cases of misdescription like those illustrated by dialogue (3) can be accounted for by appealing to a pragmatic explanation based on Gricean rules. Yet, in some passages in his paper he expresses some doubts about the real nature of these phenomena: ‘I myself feel that such a sentence [the sentence ‘‘Her husband is kind to her’’] expresses a falsehood, even when ‘‘her husband’’ is used referentially to refer to a kind man; but the popularity of Donnellan’s view has made me uncertain that this intuition should be pressed very far [. . .]. In Naming and Necessity [. . .] I suggested tentatively that Donnellan’s remarks about reference have little to do with semantics or truth conditions. The point would be put more exactly if I had said that Donnellan’s distinction is not itself a semantical one, though it is relevant to semantics through pronominalization.’ (Kripke, 1977: 269, 275.) As I have already pointed out, the idea that the notion of speaker’s reference is strictly pragmatic and has no semantic significance is clearly stated in Soames (1994), where Donnellan’s remarks about pronominalization are discussed at length.


shown by its relevance for the truth conditions of the sentence in some cases of misdescription, this means that the truth conditions determined by the referential use are different from the truth conditions determined by the attributive use (where no speaker’s reference is involved). But such a conclusion seems to entail a semantic characterization of the attributive/referential distinction, which is exactly what many theorists are reluctant to accept. Thus, we have to face the following dilemma: if the thesis of the semantic relevance of the speaker’s reference were correct, ‘it might be thought that [. . .] an ambiguity in the definite article would at least be suggested and that it is intuitively very implausible to suppose such an ambiguity. So that, until the question of ambiguity is resolved, a real doubt remains about whether that position can be correct.’ (Donnellan, 1978: 66.)


(DESC) The meaning of the definite description ‘The P’ is the function [[The P]] from contexts to identifying functions (i.e. functions from worlds to individuals) which is defined as follows: (i) if there is an individual u that in every world in C is the only individual satisfying P, then [[The P]](C) is the constant function f such that, for any w 2 W, f(w) ¼ u; (ii) otherwise, [[The P]](C) is the partial function g such that, for any w 2 W, g(w) is the only individual that satisfies P in w (g(w) is indefinite when, in w, P is satisfied by no individual or by several individuals). The idea is that, in case (i), the body of information associated with the relevant context C is strong enough to select an individual u as the only individual satisfying P in C, so that u is the intended referent of the description ‘the P’ with respect to any evaluation world w (even though u might not satisfy this description in w). On the contrary, in case (ii), as suggested by Donnellan’s characterization of the attributive use (according to which the speaker ‘states something about whoever or whatever is the so-and-so’), no particular individual is selected by C,24 and the denotation of the description, with respect to any evaluation world w, depends on which individual, in w, satisfies the description. 23 As I have already specified, adopting alternative treatments of definite descriptions, e.g. a Russellian one, would not interfere with the remarks I am about to develop and would be equally appropriate. 24 In most circumstances this is so because the information available to the speaker a is not rich enough to select a particular individual as the reference of the description ‘the P’ (even though a presupposes that there is only one individual that satisfies the description). Yet, as suggested by Donnellan, there are situations in which the speaker is provided with the necessary information about the identity of the individual at issue, but she suspends it in order to state something, in general, about whoever or whatever is the P. In this case, the relevant context C, in definition (DESC), does not coincide with the full information available to a.


out by allowing us to recognize the semantic relevance of the speaker’s reference without assuming any ambiguity of the definite article, because definite descriptions are associated with a single rule of interpretation. So, what I would like to show, in this last part of the paper, is not that the ambiguity hypothesis is false (because such a discussion is beyond the purpose of the present paper, although I find unitary theories preferable), but only that it is possible to account for the semantic role of the speaker’s reference and its relevance for the determination of the truth conditions without assuming the ambiguity of the definite descriptions. As a starting point,23 let us reflect on the rule of interpretation (DESC), repeated here:

Andrea Bonomi 127

Let us return to sentence (1), which in the context of a dialogue such as (6) receives an interpretation based on the semantic reference of the definite description. (1) is evaluated as false in this case. (Do not forget that, in Kripke’s sense, the semantic reference of a designator is given by the formula: meaning of the designator (as determined by linguistic conventions) + relevant facts). But we have also seen that, according to Donnellan, there is a sense in which, in the interpretation based on the speaker’s reference, as in the case of dialogue (3), (1) is evaluated as true. As we saw in section 5, these different results can be accounted for by the fact that the world @ belongs to proposition (5), repeated here:

But @ does not belong to proposition (7): ð7Þ kw½RðwÞ ^ PoetðwÞðThe man wearing a fez ðR=KÞðwÞÞ: Note that we get these results by applying rule (DESC) in both cases: what changes, when passing from (5) to (7) is only the value of the variable associated with the NP, which in (7), but not in (5), coincides with the evaluation context. To illustrate this point, suppose that the evaluation context is the set R representing a context of true assumptions where Theo (the semantic reference) is the man wearing a fez. In the default cases, the reference of the description is fixed with respect to this context. This is what happens in (7). But in (5) the reference is fixed with respect to S, representing the speaker’s presuppositions, so that, this time, the relevant individual is Leo. Intuitively speaking, (5) corresponds to an interpretation in which a presuppositional slot is tentatively opened in order to take into account the speaker’s assumptions with respect to a local constituent of the sentence (the definite description). In the theoretical framework discussed above, we might say that the embedding context and the embedded context do not coincide. But no slot is opened in (7), where the relevant assumptions are those represented by the basic context R. Here, the embedding context and the embedded one coincide. Given this kind of analysis, there is a further question to be addressed: for what reason is the notion of truth based on the speaker’s reference often associated, from an intuitive point of view, with a feeling of uneasiness25 that is absent from the notion of truth based 25

As remarked by Kripke (1977: 262): ‘It seems hard for us to say that when he [the speaker] uttered, ‘‘Her husband is kind to her’’, it expressed a truth, if we believe that her husband is unkind to her.’


ð5Þ kw½RðwÞ ^ PoetðwÞðThe man wearing a fez ðS=KÞðwÞÞ:


9 CONCLUSIONS AND OPEN PROBLEMS In communicative exchanges context shifts are often induced by the presence, in a given sentence, of operators such as ‘If . . . , then . . .’, ‘x believes that . . .’, ‘In fiction F . . .’, and so on. A natural point of reference, from this standpoint, is Stalnaker’s analysis in terms of context sets and operations on context sets. Yet, in other situations, the exchange of information is made possible by a sort of spontaneous context shift, i.e. a revision of the relevant beliefs and attitudes that enables the agents to communicate the intended information even if their presuppositional backgrounds diverge on some specific point. In these cases, the need for a revised context is motivated not by the presence of an (explicit or implicit) operator, but by the simple intention of taking into account a different view of the common ground.


on the semantic reference? A possible answer to this question is that when a speaker a uses a misdescription to identify a particular individual u and to state that u is Q, it is not correct, strictly speaking, to say that a stated something true, of u, for the simple fact that u is Q. It would be much more appropriate to say that, in this case, a stated something true, of u, with respect to the criteria of identification available to a. In general, the notion of truth based on the speaker’s reference is defined with respect to a presuppositional apparatus which is the vehicle of potentially false information. This is what happens in the case of (5), because @ ; fw 2 W: in w Leo is the man wearing a fezg (even if @ 2 fw 2 W: in w Leo is a poetg). Finally, it is noteworthy that, if such a reconstruction is correct, to fix the speaker’s reference there is no need to appeal to ‘Gricean’ pragmatic rules, as suggested by Kripke. The procedure is the same that is used to fix the semantic reference and established by the semantic rule (DESC): take the meaning of the definite description and apply it to a context (i.e. a set of worlds representing an information state). In the default cases, what is relevant is context C (which, as specified by (TC), represents a body of true assumptions), so that the semantic reference is picked out. On the contrary, the speaker’s reference is picked out with respect to the context that corresponds to the speaker’s assumptions. No ambiguity of the definite description is presupposed here. In both cases the reference is given by Kripke’s formula: meaning (determined by linguistic conventions) + relevant facts. As expected, different representations of the facts may yield different results.

Andrea Bonomi 129


In the present paper I have focused on this second type of context shift. I started from the most familiar cases, where the relevant information is added to the basic common ground because it is missing, as required by accommodation, or because it must replace some false presuppositions, as required by the intention of sharing only true assumptions. Such revision strategies, based on the incorporation of presuppositions originated by an external source of information, are accounted for by definitions (TINC), centred on the notion of truth, and (UPINC), centred on the notion of update. Both of them involve a global context shift, since there is no reason for keeping considerations about reference distinct from considerations about truth: the worlds that are relevant for evaluating the sentence as true or false coincide with the worlds that are relevant for fixing the intended reference of a term, for example a definite description. But a different strategy, based on local context shifts, must be adopted to deal with the phenomenon of discommodation, which is the main topic of the paper. This phenomenon is characterized by the presence, in the context presupposed by a speaker a, of an assumption p that an addressee b must take into account even though b cannot accept it because it is in contrast with her own assumptions. To account for the existence of such ‘presuppositional slots’ that an agent can temporarily open up when interacting with other agents, in the formalism sketched in the paper a definite description (or, in general, a noun phrase) is associated with a free variable for contexts, whose value depends on the assignment p. In the default cases, p(K) ¼ R (where R is the evaluation context), so that the context that is relevant for evaluating the statement as true or false coincides with the context that is relevant for fixing the reference of the NP. But if K is given a different value (namely the speaker’s context S), then those contexts can be different. In this case, the context that is relevant for fixing the reference represents an alternative source of presuppositions. Such a local context shift, which concerns only the NP, is made possible by the truth conditions proposed in definition (TC), and this definition is used in the last section of the paper to explain why, in the cases of misdescription like those made popular by Donnellan, the speaker’s reference can have a semantic relevance in determining the truth conditions of the utterance. Interestingly enough, without assuming any ambiguity of the definite article, the approach discussed in the paper allows us to account for this role of the speaker’s reference as a simple case of discommodation, where different views of the common ground are involved. From a formal point of view, the problem is that the nature and the exact position of the variable that has a crucial role in (TC) should be

130 Truth and Reference in Context made more explicit. Several proposals26 are available in the literature, but one of the major problems with them, as far as I can see, is to make them compatible with cases of discommodation like those discussed here (e.g. (2): ‘No famous novelist has ever expired while visiting an exhibition of Swann’s favourite painter’), where the worlds that are relevant for reference are not relevant for truth. First of all one should specify the precise domain of such variables. Treating them as variables for worlds may not be sufficient, as shown by the fact that a sentence such as: (8a)

? Leo is always an erudite person

(8b)

Faust is always a very erudite person

and (9)

The person who makes a deal with Faust is always a very erudite person

are not problematic. For example, (9) is perfectly acceptable in the interpretation that can be paraphrased by the conjunction: In Marlowe’s tragedy the person (in that tragedy) who makes a deal with Faust is a very erudite person and in Goethe’s tragedy the person (in that tragedy) who makes a deal with Faust is a very erudite person and . . . To explain this fact, the most natural solution is to assume, as I have done in this paper, that in (9) the variable bound by the adverb of quantification ‘always’ is a variable not for worlds, but for contexts, that is, a variable for sets of worlds, under the hypothesis that contexts are represented by context sets. In fact, if the domain of quantification is the set of relevant contexts (i.e. Marlowe’s tragedy, Goethe’s tragedy, etc.), a possible (rough) paraphrase of (9) should be something like: (9#)

For every (relevant) context C and for every x such that x is the person who makes a deal with Faust in C, x is a very erudite person in C.

26 To account for the role of context in the interpretation of the Definites, Chierchia (1995: 220– 222) uses relational variables that represent the contextually supplied information. In Stanley and Szabo´ (2000: 251) the variables for domain restriction are associated with the common noun occurring in quantified expressions. In Percus (2000: 194–195) definite descriptions contain a situation pronoun that is bound by a lambda operator. Finally, contextual variables stand for subsets of the domain in Westersta˚hl (1985) or for submodels, with respect to a given model, in Bonomi (1998). In all these cases it is not immediate to see how the proposed analysis can deal with examples like (2) and in general with phenomena of discommodation.


is hardly acceptable in normal circumstances, whilst the sentences:

Andrea Bonomi 131

In this case, the context variable bound by the adverb of quantification is the variable for the evaluation context, which coincides with the context variable that is relevant for fixing the reference of the definite description. So, there would be no need here for a specific variable associated with the definite description. Yet consider this other example: (10)

The number of nuclear weapons is always significantly smaller than the number Rumsfeld has suggested during the last press conference.

(10#)

There is a context C and there is a number y such that C is the context of Rumsfeld’s utterance and y is the number of nuclear weapons provided in C, and for every (relevant) context C# 6¼ C and for every x such that x is the number of nuclear weapons available to Iraq in C#, x is smaller than y.

As predicted by the kind of analysis developed in this paper, the quantificational structure obtained in the case of (10) is considerably different from the one obtained in the case of (9), because in (10) the variable bound by the adverb of quantification is not the variable for the evaluation context, but the one associated with the definite description. Unfortunately, a systematic treatment of these binding phenomena is far beyond the rough formalization suggested in the paper. A further problem concerns the exact role of the free variable in a more explicit logical form. Consider Strawson’s conjecture that the presuppositions usually associated with a definite description are relevant only when the description has a ‘topical’ occurrence. Supposing that some partition of propositions into a focal and a topical part is available, one possible way of interpreting this conjecture in terms of Donnellan’s distinction is to say that the referential interpretation of a definite


This sentence is appropriate in a situation where a reporter, in order to prove that Rumsfeld is lying about Iraq’s nuclear armament, utters it after consulting different sources of information (as, for example, the United Nations, the European Community, and so on). What he means is that, whenever we compare the figure provided by Rumsfeld with the figure provided by any of those sources, the former is much bigger than the latter. In this case, if the domain of quantification is the set of relevant contexts (i.e. the information available to the United Nations, the information available to the European Community, etc.) a possible paraphrase of (12) should be something like:


(11) ‘Only [the man wearing a fez]F is a poet’ the interpretation based on the speaker’s reference, i.e. on the referential use of the definite description, should not be compatible with the ‘focal’ occurrence of the description. But it is not difficult to find situations where that kind of reading is the appropriate one for (11). Think of the following comment: (12) Yes, he is a poet, but his hat is not a fez. Once more, the role of the articulation topic/focus in selecting the relevant reading is an open problem that should be addressed in a more refined theoretical framework. Finally, there is a problem concerning the ontological commitments of the treatment outlined here for discommodation phenomena. As the reader remembers, this treatment is based on the idea that, in a sentence such as (2), the reference of the definite description ‘Swann’s favourite painter’ is fixed by the presence of a variable that is locally anchored to a ‘counterfactual’ context such as the Recherche. It is this reference which is used to evaluate the sentence as ‘factually’ true, that is true in the real world @, which does not belong to the ‘counterfactual’ context set associated with the Recherche. No problem arises here, 27 In Dekker (1998: 319–322) this hypothesis is developed in the framework of dynamic semantics. 28 I am grateful to an anonymous referee for this suggestion. 29 For example, the link is questioned in von Fintel (2004).


description is possible only when the description is part of the topical material.27 From this point of view, the dependence on the relevant assignment p would characterize not only the occurrences of definite descriptions in this position but, in general, any topical material, so that the solution proposed here for definite descriptions would be independently motivated. In connection with the analysis developed in the paper, one might suggest28 that the material in focus is always interpreted with respect to the evaluation context C, whilst topical material is evaluated with respect to a context K selected by p. This is an intriguing problem that deserves further investigation. Anyway, there are at least two reasons for being cautious on this point. (i) In general, the link between ‘topicality’ and existential presuppositions is a controversial issue.29 (ii) A first problem that seems to arise in connection with the proposal under discussion is that if the partition into a focal and a topical part is taken at its face value, in a statement such as:

Andrea Bonomi 133

because the intended reference (Vermeer) exists not only in the Recherche but also in @. Yet, consider a sentence such as: (13) Vinteuil, the talented composer that Swann likes best, is a very unpretentious man. Unfortunately, most of the so-called superstars that have signed a contract with La Scala this year are much more arrogant and fastidious than Swann’s favourite musician.

ANDREA BONOMI Dipartimento di Filosofia Universita` degli Studi Milano Via Festa del Perdono, 7 20122 Milano Italia e-mail: [email protected]

Received: 06.07.05 Final version received: 26.09.05 Advance Access publication: 02.11.05

REFERENCES Arlo´-Costa, H. (2002) ‘Epistemological foundations for the representation of discourse context’. MS. Beaver, D. (2001) Presuppositions and Assertion in Dynamic Semantics. CSLI Publications. Stanford. 30

Bonomi, A. (1998) ‘Indices and contexts of discourse’, Lingua e Stile 33: 471–486. Bonomi, A. & Zucchi, S. (2003) ‘A pragmatic framework for truth in fiction’. Dialectica 57:103–120.

It should be noticed that this is exactly what happens in the intensional semantics proposed in Kripke (1963), where the extension of a predicate P, in a world w, may contain individuals that do not exist in w (whilst the interpretation of the quantifiers is restricted to the domain of w).


Suppose that the second sentence, in (13), is factually true, exactly as (2). In both cases the referent of the definite description is fixed with respect to the Recherche. The difference is that the referent of the definite description ‘Swann’s favourite musician’, that is Vinteuil, is a person who exists in the Recherche but not in the real world. To be sure, the presence of the contextual variable, which selects the ‘local’ context required by the definite description, allows us to account for the peculiarity of such a situation, where the factual truth of the sentence is based on a relation between real persons on the one hand and a fictional one on the other. Formally speaking, this means that in some cases the extension of a relational predicate in the real world @ might involve individuals that do not exist in @.30 Thus, it is natural to wonder if such a requirement can be motivated by independent reasons.

134 Truth and Reference in Context Percus, O. (2000) ‘Constraints on some other variables in syntax’. Natural Language Semantics 8:173–229. Perrault, C. R. (1990) ‘An application of default logic to speech act theory’. In P. Cohen et al. (eds). Intentions in Communication. MIT Press. Cambridge, MA. Predelli, S. (2000) ‘The teetotaler and his Martini’. Mind & Language 15:511–527. Recanati, F. (1996) ‘Domains of discourse’. Linguistics and Philosophy 19:445–475. Rooy, R. van (2001) ‘Exhaustivity in dynamic semantics; referential and descriptive pronouns’. Linguistics and Philosophy 24:621–657. Soames, S. (1982) ‘How presuppositions are inherited: A solution to the projection problem’. Linguistic Inquiry 13:483–545. Soames, S. (1994) ‘Donnellan’s referential/attributive distinction’. Philosophical Studies 73:149–168. Stalnaker, R. (1970) ‘Pragmatics’. Synthese 22:272–289. Stalnaker, R. (1978) ‘Assertion’. In P. Cole (ed.). Syntax and Semantics, volume IX. Academic Press. New York. Reprinted in Stalnaker (1999: 78–95). Stalnaker, R. (1999) Context and Content. Oxford University Press. Oxford. Stalnaker, R. (2002) ‘Common ground’. Linguistics and Philosophy 25:701–721. Stanley, J. & Szabo´, Z. (2000) ‘On quantifier domain restriction’. Mind & Language 15:219–261. Westersta˚hl, D. (1985) ‘Determiners and context sets’. In J. van Benthem & A. ter Meulen (eds). Generalized Quantifiers in Natural Languages. Foris. Dordrecht.


Chierchia, G. (1995) Dynamics of Meaning. The University of Chicago Press. Chicago and London. Dekker, P. (1998) ‘Speaker’s reference, descriptions and information structure’. Journal of Semantics 15:305–334. Donnellan, K. (1966) ‘Reference and definite descriptions’. Philosophical Review 75:281–334. Donnellan, K. (1978) ‘Speaker reference, descriptions and anaphora’. In P. Cole (ed.). Syntax and Semantics, vol. IX. Academic Press. New York. Fauconnier, G. (1985) Mental Spaces. MIT Press. Cambridge, MA. Grove, A. (1988) ‘Two modellings for theory change’. Journal of Philosophical Logic 17:157–170. von Fintel, K. (2004) ‘Would you believe it? The King of France is back! (Presuppositions and truth-value intuitions)’. In M. Reimer & A. Bezuidenhout (eds.). Descriptions and Beyond. Oxford University Press. Oxford. Heim, I. (1982) The Semantics of Definite and Indefinite Noun Phrases. Department of Linguistics, University of Massachusetts. Amherst, MA. Kripke, S. (1963) ‘Semantical considerations on modal logic’. Acta Philosophical Fennica, 16:83–94. Kripke, S. (1979) ‘Speaker’s reference and semantic reference’. In P. A. French et al. (eds). Contemporary Perspectives in the Philosophy of Language. University of Minnesota Press. Minneapolis, MN. Lewis, D. (1973) Counterfactuals. Blackwell. Oxford. Ludlow, P. & Neale, S. (1991) ‘Indefinite descriptions: In defense of Russell’. Linguistic and Philosophy 14:171–202.

Journal of Semantics 23: 135–183 doi:10.1093/jos/ffl001 Advance Access publication April 4, 2006

(In)Definiteness, Polarity, and the Role of wh-morphology in Free Choice ANASTASIA GIANNAKIDOU University of Chicago

Abstract In this paper we reconsider the issue of free choice and the role of the whmorphology employed in it. We show that the property of being an interrogative whword alone is not sufficient for free choice, and that semantic and sometimes even morphological definiteness is a pre-requisite for some free choice items (FCIs) in certain languages, e.g. in Greek and Mandarin Chinese. We propose a theory that explains the polarity behaviour of FCIs cross-linguistically, and allows indefinite (Giannakidou 2001) as well as definite-like FCIs. The difference is manifested as a lexical distinction in English between any (indefinite) and wh-ever (definite); in Greek it appears as a choice between a FCI nominal modifier (taking an NP argument), which illustrates the indefinite option, and a FC free relative illustrating the definite one. We provide a compositional analysis of Greek FCIs in both incarnations, and derive in a parallel manner the Chinese FCIs. Here the definite versus indefinite alternation is manifested in the presence or absence of d ou, which we take to express the maximality operator. It is thus shown that what we see in the morphology of FCIs in Greek is reflected in syntax in Chinese. Our analysis has important consequences for the class of so-called whindeterminates. In the context of current proposals, free choiceness is taken to come routinely from interrogative semantics, and wh-indeterminates are treated as question words which can freely become FCIs (Kratzer and Shimoyama 2002). Our results from Mandarin and Greek emphasize that wh-indeterminates do not form a uniform class in this respect, and that interrogative semantics alone cannot predict either sensitivity of free choice to definiteness, or the polarity behaviour of FCIs.

1 INTRODUCTION It is often observed that languages exploit wh-morphology for the formation of free choice paradigms. For example, wh-phrases, augmented by some kind of modal marking or focus additive particle The Author 2006. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]


LISA LAI-SHEN CHENG Leiden University

136 (In)Definiteness, Polarity, and the Role of wh-morphology (meaning too, and, even or) appear as free choice items (FCIs) in languages as typologically diverse as Greek, Japanese, and Hindi:

English is part of this paradigm too, as can be seen in the use of whever, which has a reading parallel to the so-called free choice any: (2) a. I will order whatever is recommended by the chef. b. I will order anything that is recommended by the chef. Horn (2000a) analyses wh-ever items as FCIs. Syntactically, any and whever differ in that wh-ever requires a clausal complement and always forms a free relative, whereas any is used as a DP constituent, a difference that will prove central to the analysis we will propose in this paper. Given the paradigm above, the first question we are going to ask is whether the property of being an interrogative wh-word alone is a sufficient condition for free choiceness. There are two sets of facts suggesting that the answer to this question must be negative. The first observation has to do with the formation of the Greek FCI paradigm that we see below: (3) [o-pjos]-dhipote the-who/which-modal marker (Greek) [o-ti]-dhipote the-what-modal marker [o-pote]-dhipote the-when-modal marker [o-pu]-dhipote the-where-modal marker (4) pjos-dhipote; ti-dhipote; pote-dhipote; pu-dhipote The FCI form consists, on the one hand, of a wh-phrase (or determiner) preceded, crucially, by what appears to be the definite article o, and the modal free choice marker dhipote ‘ever’, on the other.


(1) a. Greek opjos-dhipote, lit. who-modal marker (Giannakidou 1998, 2001) b. Catalan qual-sevol, lit. who-modal marker (Quer 1998) c. Spanish qual-quiera, lit. who-modal marker (Quer 1999) d. Dutch wie dan ook, lit. who-then-too (Rullmann 1996) e. Korean nwukwu-na, lit. Who-or nwukwu-to, lit. who-and (Lee 1997; Gill et al. 2002) f. Japanese dare-demo lit. who-even (Nishigauchi 1986) g. Hindi jo-bhii lit. lit. which-even (Dayal 1995; Lahiri 1998)

Anastasia Giannakidou and Lisa Lai-Shen Cheng 137

(5) a. Sheí d ou jı`nlaí-le. who d ou enter-PERF ‘Everyone entered.’ b. Na˘-ge xue´sheng d ou jı`nlaí-le. d ou enter-PERF which-CL student ‘Every/any student entered.’ c. Na˘-ge xue´sheng d ou ke˘yı˘ jı`nlaí. which-CL student d ou can enter ‘Any student can enter.’ Na˘ ‘which’ is morphologically identical to the demonstrative na` ‘that’, suggesting a parallel with the Greek definite marker o we just noted.1 Example (5) shows that a bare wh-phrase such as sheí ‘who’ is less restricted in its distribution, in comparison to na˘-ge xue´sheng ‘which student’, which exhibits the limited distribution observed for polarity FCIs in Greek, Spanish, and Catalan (Giannakidou 1997, 1998, 2001; Quer 1998, 1999). Specifically, na˘-CL NP is unacceptable in episodic contexts (5b) (with the verb being marked by -le, the perfective marker). This observation is further supported by the cases below,

1 Na˘ ‘which’ and na` ‘that’ in Mandarin Chinese differ in tone; the former has the third tone (falling-rising).


The sequence [o + interrogative wh-word] is used in forming free relatives in Greek, and is distinct from the interrogative wh-word, which is unable to combine on its own with the free choice marker to form FCIs as we see in (4). The determiner o in the free-relative-wh appears as a bound morpheme and remains invariant in terms of gender number and case (contrary to what is normally the case with Greek determiners, which inflect fully for / features and case), suggesting that what is used for FCI formation is actually not the definite determiner itself, but the definiteness core. From the contrast between (3) and (4), it is clear that the wh-component in the FCI, e.g. opjos, oti, opote, etc., is a definite wh-form, and not just a mere question word. This runs counter to the idea that plain interrogative wh-words are a sufficient source for free choice (Kratzer and Shimoyama 2002). The second piece of evidence supporting the role of some form of definiteness comes from Mandarin Chinese, a language with so-called wh-indeterminates. Mandarin employs only a single wh-paradigm (Cheng 1991). However, bare wh-phrases do not exhibit FCIinterpretations routinely, or polarity FCI behaviour; it is only the D-linked wh-phrase na˘-CL ‘which’ that does.

138 (In)Definiteness, Polarity, and the Role of wh-morphology where the FC reading of na˘-CL NP is unacceptable in episodic questions and with negation: (6) a. Ta ma˘i-le na˘-be˘n sh u ma? (Mandarin Chinese) he buy-PERF which-CL book Y/N Intended: ‘Did he buy any book?’ b. Ta meíyo˘u ma˘i na˘-be˘n sh u? he not-have buy which-CL book ‘Which book did he not buy?’ Not: ‘He didn’t buy any book.’ Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In the episodic questions in (6), na˘-be˘n-sh u cannot be interpreted as any book and contributes only the interrogative wh-meaning (which is incompatible with the yes/no question particle ma in (6a)). Hence na˘CL NPs seem to be true polarity FCIs, amenable to the intensional analysis of Greek, Spanish and Catalan FCIs, that we will propose in section 5. Again, the fact that there is a split in the wh-paradigm—only the which phrase becomes an FCI—is not compatible with the idea that the source of free choiceness is simply interrogative semantics; instead, we must say something specific about the D-linked na˘ ‘which’. It will turn out, in section 5, that na˘-CL-FCIs and the Greek FCI have very similar compositions, which explains their polarity behaviour. Our primary goal in this paper is to address the question of what makes wh-phrases suitable candidates for FCIs, and what exactly the conditions are on the candidacy. We will propose a variable binding semantics in the Heim-Kamp sense, augmented with the widely accepted implementation of definiteness for free relatives ( Jacobson 1995). We propose that there are two kinds of FCIs: indefinite as well as definite ones, depending on whether or not there is an active definiteness function, i.e. an expression that contributes maximality operating on top of the wh-core of the FCI. We show that, typically, the difference is reflected syntactically as one between FCI-nominals, i.e. FCIs like any that take NP arguments, which are indefinite, and FCI free relatives (i.e. FCIs taking CP arguments like English wh-ever), for which a definite analysis is plausible. For the Greek item opjosdhipote both analyses will be appropriate depending on whether the FCI modifies an NP or a CP. For Mandarin na˘-CL NP, the presence of d ou, which we analyse as the iota operator equivalent to Greek o, renders na˘-CL semantically a definite regardless of the type of argument it selects. However, the definite versus indefinite distinction is still found in Chinese with na˘-CL NP since it is possible to have na˘-CL NP ou. The former is akin to a with d ou and also na˘-CL NP without d semantic definite and the latter to an indefinite, as can be seen from


2 CORE PROPERTIES OF FREE CHOICE: ANTI-EPISODITY, POLARITY AND Q-VARIABILITY There are two fundamental properties of FCIs cross-linguistically. The first has to do with their distribution: many FCIs exhibit limited distribution characteristic of polarity items (PIs; Giannakidou 1998, 2001; Quer 1998, 1999). The second property is that FCIs exhibit the quantificational variability characteristic of indefinites (Horn 2000a; Giannakidou 2001; Dayal 1995). At the same time FCIs retain a universal-like, scalar dimension in their meaning which has been characterized in various works as widening (Kadmon and Landman 1993), concessiveness (Lee 1997), or arbitrariness (Tovena and Jayez 1997). The same intuition of domain extension is present also in Lee


the examples which show that the presence of d ou leads to ruling out the empty set (section 5). The structure of the paper is as follows. In section 2, we present the basic properties of free choice that we consider in this paper: antiepisodicity, polarity, and variation. In section 3 we outline the background analysis that we are assuming (Giannakidou 2001), and offer a novel compositional analysis of the class of FCIs that operate on NP arguments and are indefinite. In section 4 we discuss the precise nature of the definiteness of FCIs with clausal arguments. We address the contrast between plain and -ever structures, as well as any versus whever in English, and present evidence supporting the definiteness analysis of free relatives and wh-ever forms. In this connection, we also consider certain arguments that appear to question it (from Grosu 1996, Larson 1999, and Horn 2000a). It is illustrated that these arguments do not really undermine the idea that some form of definiteness is involved in the free relative, but rather make clear the need to distinguish between morphological definiteness and semantic definiteness which we take to be essentially maximality. In section 5 we focus on the Mandarin data. It will be proposed that d ou expresses the iota-as-maximality operator (rather than a universal or distributive operator as was previously assumed), and speculate that the use of additive particles in other languages (Dutch, Japanese, Korean, Hindi) serves the same purpose (see Tancredi and Yamashina 2004 for an analysis of Japanese mo consistent with this idea). We further argue that na˘-CL NP is intensionalized by means of a (possibly covert) wu´lu`n, which counts as the Mandarin equivalent to the Greek dhipote. We close with the implications and predictions of our analysis for whindeterminates, as well as the general theory of polarity.

140 (In)Definiteness, Polarity, and the Role of wh-morphology

2.1 Anti-episodicity and polarity The polarity property of FCIs is shown below: FCIs are unacceptable in episodic contexts, positive and negative alike; FCIs in Spanish, Catalan and Greek are typical examples: (7) a. Idha opjondhipote. (Greek; Giannakidou 2001) saw.perf.1sg FCI person ‘I saw anybody.’ b. Dhen idha opjondhipote. not saw.perf.1sg FCI person Intended: ‘I didn’t see anybody.’ (8) (Non) Expulsaron del partido a cualquier disidente. not expel.3pl from-the party ACC FCI dissident Intended: ‘They expelled any dissident from the party.’ Intended: ‘They didn’t expel any dissident from the party.’ (Spanish; Quer 1999) (9) (No) Li va comprar qualsevol ram not her/him aux.3sg to.buy FCI bouquet Intended: ‘ S/he bought him/her any bouquet.’ Intended: ‘S/he didn’t buy him/her any bouquet.’ (Catalan; Quer 1998)


and Horn’s (1994) analysis of any as containing a silent even (but see Lahiri 1998 for a criticism of this idea). This dimension of domain extension is often responsible for the concessive flavour of statements with FCIs (see especially Lee 1997), and sometimes materializes in the form of actual concessive markers such as the particles ke na and ke an in Greek, which may appear with FCI-free relatives (see discussion in section 4.3). Scalarity alone, however, does not seem to affect the grammatical characteristics of FC that we identify here, i.e. it cannot explain the polarity behaviour of FCIs (for an attempt, see Krifka 1995), as it cannot rule out FCI nominals in the core ungrammatical cases of positive sentences (a point argued for in detail in Giannakidou 2001). But it certainly is a component of FC meaning that we want to capture in the intensional analysis we will put forth. To avoid repeating discussions that are made in the literature already, we describe the polarity and quantificational variability of FCIs only briefly below (for more details see Giannakidou 1998, 2001; Quer 1999; Horn 2000a, 2000b). The properties we are describing apply to what we called FCI-nominals, i.e. FCIs that take NP arguments, or appear as independent QP constituents themselves, which are mainly the FCIs studied in the works mentioned above.


2 The observation about anti-episodicity is consistent with Jayez and Tovena’s (J&T 2005) nonindividuation constraint which requires that ‘the information conveyed by the sentence [with the FCI, addition G&C] should not be reducible to a ‘‘referential’’ situation, i.e. a situation in which particular individuals in the current world satisfy the sentence.’ (J&T 2005: 2) This constraint is stipulated composition-externally as a filter for ruling out improper cases of any and FCIs, but sensitivity to episodicity in Giannakidou (2001) is derived as a result of a lexical semantic property of FCIs. It is this latter line of reasoning we will build on here, as it presents a simpler account where (limited) distribution is derived without stipulating composition external principles. 3 Very occasionally, episodicity can be overridden with negation in combination with deverbal predicates, like (i):

(i)

?O proedros dhen proxorise se opjadhipote the president not proceeded.perf.3sg in FC ‘The president didn’t proceed with any negotiations.’

diapragmatefsi. negotiation

Such sentences, however, are never impeccable. Crucially, the complement NP of the FC determiner must be deverbal (see Vlachou in prep. for more extensive corpus data supporting this). An obvious hypothesis for the improvement may be that the deverbal NP contains a clausal-like structure thus rendering these cases subtrigging (the term from Le Grand 1975); we return to this phenomenon in sections 4.3 and 4.4. However, we must note that improvement is impossible with de-verbal predicates in the absence of negation. The positive version of the sentence above, (ii), remains bad: (ii) O proedros proxorise se opjadhipote diapragmatefsi. the president proceeded.perf.3sg in FC negotiation ??The president proceeded with any negotiations. So the question as to what is going on with examples like (i) remains open, and the fact that without negation anti-episodicity wins suggests that invoking clausal structure for (i) is probably not the only factor responsible for improvement.


The sentences above are episodic. A sentence is episodic when it is about exactly one event that happens at a particular time (Giannakidou 1997, 1998, 2001; Krifka et al 1995): (10) a. d!e /(e) (Giannakidou 2001: 662, (5)) b. Episodicity understood here: d!e d!t [/(e) ^ e 4 t] Episodic sentences involve (in a particular world) just one event that happens at a particular point in time, an interpretation signalled by perfective aspect in the Greek examples (recall also the perfective Le in the Chinese).2 They are in this sense event-specific. FCIs are, apparently, incompatible with this kind of specificity (and require variation for reasons that will be made precise later). Further, the fact that FCIs are excluded from positive and negative episodic sentences alike suggests that it is not non-veridicality per se that FCIs are sensitive to but antiepisodicity (Giannakidou 1997, 1998, 2001).3 Finally, let us also note that the anti-episodicity observation holds for FCI-nominals but not for FCIs with clausal structure, which are fine in episodic sentences and have non-polarity status (a point to which we return in section 4): (11) a.Last night at the party, Bill talked to any woman. b. Last night at the party, Bill talked to any woman who seemed interested.


b. Ta meíyo˘u ma˘i na˘-beˇn sh u he not-have buy which-CL book. ‘Which book did he not buy?’ Not: ‘He didn’t buy any book.’ Na˘-beˇn-sh u ‘which book’ is unacceptable in an episodic question containing ma, the question marker in (15a); not even the question 4 The sentence in (15a) becomes better if we use -guo ‘experiential aspect’ instead of -le, which indicates the perfective aspect. Note that with -guo, there is no limitation to one event, as in (i):

(i)

Ta qu` guo` rı`beˇn he go EXP Japan ‘He has been to Japan.’

This sentence is compatible with the fact that he has been to Japan many times—recall our earlier discussion of episodicity.


For FCI nominals, anti-episodicity seems to be at work also in episodic questions (again, with perfective aspect): (12) Su sistisan opjondhipote thavmasti? (Greek) you introduced.perf.3pl FC admirer Intended: ‘Did they introduce any admirer to you?’ (13) Et van presentar qualsevol admirador? to-you aux.3pl introduced FC admirer Intended: ‘Did they introduce any admirer to you?’ (Catalan; Quer 1998: 220) (14) Te presento´ a cualquier admirador? (Spanish) admirer you introduced.3pl ACC FC ‘Did they introduce any admirer to you?’ Notice also that Did they introduce almost any admirer to you? is unacceptable, suggesting that the FCI interpretation is not available in an episodic question for any either since almost is typically taken as a diagnosic for FC any (Davison 1980). (Being aware of the problems with the precise workings of almost, noted especially by Lee and Horn 1994, we use the appearance of almost only as a descriptive diagnostic here.) The FCI na˘-CL NP in Mandarin that we identified earlier has the same sensitivity to anti-episodicity, as we noted already. We repeat the relevant examples below:4 (15) a. Ta ma˘i-le na˘-beˇn sh u ma? (Mandarin Chinese) he buy-PERF which-CL book Y/N Q-marker Intended: ‘Did he buy any book?’


meaning ‘which book’ is available, since this is a yes/no question. Likewise in (15b), and the example below: (16) Tameń yo˘u-meí-yo˘u jie`shaò na˘-ge chońgbaìzheˇ geˇi nıˇ. give you they have-not-have introduce which-CL admirer Intended: ‘Did they introduce any admirer to you?’

Table 1 (Partial) contrastive distribution of any, FCIs and APIs. Environments 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.

Episodic negation Episodic questions Conditionals Restriction of universal Future particle/will Modal verbs Directive intensional verbs Imperatives Habituals Stative verbs Generics Only Emotive factive verbs Affirmative episodic sentences Epistemic intensional verbs

Any

FCIs

APIs

OK OK OK OK OK OK OK OK OK OK OK OK OK

OK OK OK OK OK OK OK OK OK

OK OK OK OK OK OK OK OK OK


This sentence is a neutral yes-no question with the A-not-A form. Again, just as in (15a), since the non-interrogative (FC) reading of na˘-CL NP ‘which NP’ is not available (due to episodicity), the sentence becomes ungrammatical because the wh-interrogative cannot appear within a yes-no question. Hence, na˘-CL NP behaves like a polarity FCI. FCIs typically appear in sentences that involve quantificational (Q-) structures, i.e. with modal, generic, habitual, and intensional Qoperators, in subjunctive complements of volitional and other directive propositional attitudes, in imperatives, and with Q-adverbs of various kinds (for an extensive illustration see Giannakidou 2001, Quer 1998, 1999 and Table 1; earlier discussion also in Bosque 1996). We present here the summary of the relevant data in Table 1 by comparing FCIs to any, and the class labelled in Table 1 as APIs. API stands for ‘affective PI’, following Giannakidou (1998, 2001): APIs correspond to what is traditionally, but quite inaccurately, known as ‘negative’ PIs of the weaker kind, i.e. NPIs that appear with negation as well as a wide variety of non-veridical environments (including downward entailing, and nonmonotone or upward entailing ones such as questions, non-monotone


Protasis of conditionals (17) An kimithis me fopjondhipote/kanenang tha se skotoso. if sleep.2sg with FCI-person/API-person FUT you kill.1sg ‘If you sleep with anybody, I’ll kill you.’ Directive intensional verbs (selecting subjunctive) (18) I Ariadne epemine na afisoume the Ariadne insisted.3sg subj let.1pl fopjondhipote/kanenang na perasi mesa. FCI-person/API-person subj come.3sg in ‘Ariadne insisted that we allow anyone in.’ With kanenan: ‘Ariadne insisted that we allow someone or other to come in.’ Modal verbs (19) Bori na anapse fopjosdhipote/kanenasg to fos can.3sg subj lit.3sg FCI-person/API-person the light ‘Anyone may have turned on the light.’ With kanenas: ‘Someone or other must have turned on the light.’ (Notice the need to change to epistemic must in the case of someone or other.) (20) Boris na dhanistis fopjodhipote/kanenag vivlio. can.2sg subj borrow.2sg FCI/ API book


quantifiers, modal contexts and propositional attitudes; see Giannakidou 1998, 1999 for extensive discussion). The generalization seems to be that FCIs are unacceptable in veridical and episodic contexts (bottom four rows), but fine in contexts involving implicit or explicit quantification over alternatives (modal, generic, habitual, individual level predicates, and the like). Any, on the other hand, exhibits a distribution freer than both FCIs and APIs, and therefore is fine in episodic contexts, and even in some veridical ones such as, for example, with emotive factive verbs and only. Some illustrating data are given below. As an example of API we use the Greek form kanenan, which is interpreted as an existential indefinite quantifier without a free choice, scalar, or universal-like flavour (unlike FCIs). The API existential is always a non-specific, non-referential indefinite, leading Giannakidou (1998) to define it as dependent in that it cannot introduce a discourse referent in the actual world (or current file in Heim’s 1982 sense), like regular indefinites can. To indicate the difference between the FCI and the non-FCI existential API, in certain clear cases, we translate ‘kanenan’ as someone or other:


‘You may borrow any book.’ With kanena vivlio: ‘You may borrow some book or other.’ Imperatives (21) Dhialekse fopjodhipote/kanenag vivlio. choose.2sg FCI/ API book ‘Choose any book.’ With kanena vivlio: ‘Choose some book or other.’ Habituals Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(22) Stis sigentrosis, i Ariadne sinithos milouse me at-the meetings, the Ariadne usually talked.imperf.3sg with fopjondhipote/kanenag fititi. FCI/ API student ‘At the meetings, Ariadne usually talked to any student’. With kanena: ‘At the meetings, Ariadne usually talked to some student or other.’ An interesting question arising with this preliminary contrastive data is whether the API can indeed be represented as containing a covert marker like or other, an idea that would be consistent with its nonspecificity and narrow scope behaviour. A related task would be to identify the precise contribution of English or other in this structure. Crucially, someone or other has a flavour of epistemic non-specificity (or indifference) distinct from free choice: the statements with some or other convey no scalarity, or the universal-like reading characteristic of FCIs. We will not consider this contrast further, since our focus is on the FCI, but the fact that the use of a disjunctive expression or other does not yield free choiceness is worth noting, especially when we try to link free choiceness to disjunction (Zimmerman 2000; Aloni 2003). As noted in Table 1, FCIs are unacceptable in certain cases where any is OK, e.g. with negation and episodic questions (recall the examples in the introduction). Here we add factive verbs and only: (23) Monon i Theodora idhe fopjondhipote/kanenang fititi. only the Thedora saw.3sg FCI/ API student ‘Only Theodora saw any students.’ (24) a. Ekplisome pu exi fopjondhipote/kanenang filo. be-surprised.1sg that has FCI/ API friend ‘I’m surprised she has any friends.’ b. Xerome pu exi fopjondhipote/kanenang filo. be-glad.1sg that has FCI/ API friend ‘I’m glad she has any friends.’


2.2 Q-structures and quantificational variability The second typical property of FCIs is their universal-like interpretation—a property that prompted analyses of FCIs as universal quantifiers (FCI any in Quine 1960; Horn 1972: chapter 3; Dayal 1998, 2004; Sæbø 2001): (25) Opjosdhipote fititis bori na lisi afto to provlima. FCIstudent can SUBJ solve.3sg this the problem ‘Any student can solve this problem.’ However, Vendler (1962) already noted that we need a subtler account of the alleged universal nature of free choice any; and recent cross-linguistic research specifically on wh-based FCIs (lexically distinct from APIs and morphological universals, like in Greek and Chinese) presents compelling evidence that these are not universal quantifiers— FCIs are interpreted existentially, e.g. with modal verbs, in conditionals, and imperatives (see Horn 2000b; Giannakidou 2001: 685–703): (26) I epitropi the committee opjondhipote ipopsifio. FCI candidate ‘The committee can offer ‘The committee can offer

bori na dosi ti thesi se can subj offer.3sg the position to this job to any candidate.’ 6¼ this job to every candidate.’

(27) If you sleep with anybody I’ll kill you. (Uttered by a jealous husband, for whom just sleeping with one guy will be reason enough to proceed with the cruel act.)


FCIs and APIs thus differ from any which appears with only, and in the complement of a factive emotive verb, as we see. Again, what drives the particular distributional differences between any and the respective Greek items is worth discussing, but doing so here would lead us far afield (see Giannakidou 1998, 2001, section 5 for specific suggestions; also Giannakidou 2006 for an updated view). We simply note here that Greek FCIs behave on a par with APIs in this respect, thus supporting their polarity status. Regarding FCIs, in order to explain their distribution, Giannakidou proposes an analysis of FCIs as intensional expressions containing a world variable that must be bound by some Q-operator; in an extensional context (and episodic contexts are such) there is no such operator, the world variable remains unbound, and the FCIs become ungrammatical. We come back to this discussion in section 3.


(28) Press any key to continue. (As a command to resume function.)


Furthermore, FCIs occur with Q-adverbs yielding statements that are reminiscent of construals with indefinites whose variable is bound by the Q-adverb (see Dayal 1995; Giannakidou 2001: 701–703, 715–717 from which the examples below are taken): (29) a. Sinithos dhiavaze i Ariadne opjodhipote vivlio me usually read.imperf. the Ariadne FCI book with megali prosoxi. great attention ‘Ariadne usually read any book very carefully.’ b. USUALLYs,x [book (x,s) ^ read (Ariadne, x,s); read-carefully (Ariadne,x,s)] (30) a. Spania dhiavaze i Ariadne opjodhipohte vivlio me rarely read.imperf. the Ariadne FCI book with megali prosoxi. great attention ‘Rarely did Ariadne read any book very carefully.’ b. RARELYs,x [book (x,s) ^ read (Ariadne, x,s); read-carefully (Ariadne,x,s)] These are habitual sentences with imperfective aspect in Greek. In (29), Ariadne read most books with great attention, but in (30) she read only few books with such attention, depending on the choice of adverb. Giannakidou (2001: 715–717) offers extensive discussion, that we will not repeat here, of Q-adverbs and FCIs contrasts in this context with universal quantifiers. The observed variability of FCIs and their striking similarity with indefinites prompted analyses of FCIs as indefinites (in the sense of Heim 1982; Kamp 1981), bound by universal and other quantifiers (for variants of this idea see Bolinger 1977 and more recently: Haspelmath 1997; Kadmon and Landman 1993; Lee and Horn 1994; Dayal 1995; Quer 1998; Giannakidou 1998, 2001; Horn 2000a, 2000b, 2005). A welcome consequence of such an analysis is that it provides a way to capture the fact that FCIs occur only if a Q-operator is present, and that their quantificational force is parasitic on that of the Q-operator. It is not necessary for our purposes here to review the arguments for indefiniteness in more detail since most of the points are made at length elsewhere (see especially Horn 2000b, 2005 and Giannakidou 2001). So we take the indefinite analysis of FCIs as our starting point: FCIs are indefinites. But, unlike regular indefinites, FCIs have limited distribution, which means that they must be different somehow. In what sense are they different?

148 (In)Definiteness, Polarity, and the Role of wh-morphology 3 FCIs AS INDEFINITES: INTENSIONALITY AND EXHAUSTIVE VARIATION In this section we consider the analysis of FCI nominals, i.e. FCIs when they are used independently as QPs, or when they are used as determiners and take an NP argument. Not all FCIs can take up this function, e.g. wh-ever cannot: (31) a. fAnybody/Any studentg can solve this problem. b. fWhoever/Whichever studentg can solve this problem.

3.1 Intensionality Giannakidou (2001) proposes that FCI nominals are intensional indefinites: they contain a world variable that is dependent and must be bound by some operator external to the FCI but indeed available in the sentence of occurrence. This accounts immediately for the need to be in a quantified sentence and the anti-episodicity effect. Free choice intensionality in Greek is designated by -dhipote. In this analysis, the FC determiner is treated as a property modifier which, when applied to the NP denotation, returns an intensionalized property as its output. (32) ½½DETFC ¼ kP<s,

et>.

kwkx [P(x)(w)]

So, the FC-determiner takes a property P as input and gives back its Montagovian intension, i.e. a function from worlds w to sets of individuals (^P). Note that the NP argument contains already an open s position (in line with recent observations in the literature, e.g. about temporal and situational arguments of NPs, see Encx 1991), hence the FC determiner works as an identity function and preserves the intensional type of the NP. In the ordinary case, i.e. with a regular (non-FC) determiner, the s index of the NP argument will be interpreted as a constant, i.e. as referring to the actual world and can


Wh-ever requires a CP argument thus always forming a free relative (FR)—though there are some limited exceptions to this generalization that we discuss more in footnotes 5 and 8. We will call this construal free choice free relative (FC-FR). The analysis we use as background here (Giannakidou 2001) did not consider FC-FRs and offered an account of FCI nominals only. We start by fleshing out, and modifying, that account. In section 4 we extend this analysis to FC-FRs, and discuss how the intensionality of FC interacts with the syntactic structure of FRs in producing FCIs which end up definite, and not polarity sensitive. From this analysis, a simpler account of sub-trigging will emerge in section 4.4.


thus be ignored; the NP will therefore denote a set of actual individuals. This is not an option for the FCI, whose w variable remains dependent in that it is in need of binding, as we elaborate on next. It is in this way that the FC-determiner contrasts with regular determiners, and ‘intensionalizes’ the NP. In the end, we have a phrase that denotes an NP with two variables instead of one: the regular individual variable, and an additional world variable: (33) ½½opjosdhipote fititis ¼ kwkx.student(x)(w) Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This w variable cannot remain free, but must be bound by an operator that has the ability to bind such variables (a Q-operator, i.e. a generic, habitual, modal, intensional operator). It is in this sense dependent (Giannakidou 2001: 704–705). The presence of a dependent world variable is the defining feature of FCIs—and we propose in section 5 that the Mandarin FCIs contain it too. A different form of dependency will be shown to arise when the w variable remains bound by the koperator, as is the case with definite FCIs in Mandarin and Greek. Before we proceed, let us note that admitting a dependent variable in need of Q-closure is not at all unconventional. It is widely accepted in the literature that not all variables are alike. This underlies, for example, the very common idea that some variables must be used only if their presuppositions are satisfied, e.g. the variables of definites are distinct in this way, or the variables of ‘specific’ indefinites (e.g. Farkas 2002; see also discussion of positive polarity items in Giannakidou 2002); recall also Heim’s familiarity and novelty conditions on the interpretation of definite and indefinite variables. Likewise, anaphors contain variables dependent on an antecedent for well-formedness. The dependent world variable of FCI nominals of the Greek type can be viewed as anaphoric in this broader sense, and we can posit the dependency as a presupposition if it fits better the more general assumptions that we have in mind. Because the w variable cannot remain free, FCIs will be wellformed only if there is some Q-operator in the sentence that can bind it. In episodic sentences, FCIs are out because no such operator is present, and w remains unbound. This analysis explains the polarity status and anti-episodicity property of FCI nominals that have the characteristic distribution of the Greek type. To make the composition explicit we will use another observation showing that an FC-phrase can co-occur with an indefinite determiner (Giannakidou 2001: example (85) repeated below as (34)), in Greek as well as in other languages (e.g. Spanish, see Quer 1999).

150 (In)Definiteness, Polarity, and the Role of wh-morphology (34) Dhen ime enas opjoshipote ego ja na mou not be.1sg a FC-person I for subj me ferese etsi! treat.2sg so (Ime o aderfos su!) (am the brother yours) ‘I am not just anybody to be treated this way. (I am your brother!)’

(35)

Free choice QP, 5 Q-det, 4

enas/

NP, 3

FC-det, 2

opjosdhipote

NP, 1

fititis

1. ½½fitititis ¼ kwkx.student(x)(w) 2. ½½opjosdhipote ¼ kP<s, et> kwky.P(y)(w) 3. ½½opjosdhipote (½½fititis) ¼ kP<s, et> kwky.P(y)(w) (kwkx. student(x)(w)) ¼ kwkx.student(x)(w). This is the intension of the predicate ‘student’. 4. ½½enas. This is the indefinite determiner in Greek, and in the case of the FCI nominal it can be overt or covert. We take it that it works like a Heimian indefinite function, i.e. as a function from properties to propositions: it takes the NP property as its input and gives back a sentence with open variable positions—this time with two variables x and w since the input to enas is the intension of the property P. ½½enas ¼ kP<s, et> P(x)(w) After application of the indefinite determiner at the top of the FC phrase, we end up with an open world variable, which is dependent and cannot receive the value of w0, the actual world: 5. ½½Free choice QP ¼ student(x)(w)


This provides sufficient grounds for assuming that in fact the FCI always contains an indefinite determiner on top of the FCI layer, as indicated in the structure below. The FCI derivation of (enas) opjosdhipote fititis ‘any student’ then proceeds as indicated:


Finally, a comment on the contrast between opjos and opjosdhipote, which will also be helpful in the discussion of the FR paradigm later. As we see in (36b), the Greek bare wh-phrase (opjos i.e., without -dhipote) cannot take an NP argument; it also cannot serve as an independent QP constituent by itself (36c):

The fact that the bare wh-form is not allowed as a determiner or quantifier, and, crucially, the fact that the FCI nominal is itself the input to an indefinite determiner, as we have seen, suggest that opjosdhipote, as an NP modifier, is a lexical unit without being semantically or morphologically decomposed in a strict compositional way from all its parts. The effect of this is that the morphological definiteness of o- remains inactive, or, to put it differently, does not contribute definiteness in the semantics, and the unit as a whole behaves as an indefinite. In our discussion of Chinese na˘ in section 5, a parallel situation will arise, where the D-linked na˘-phrase ‘whichphrase’ will behave like an indefinite in the absence of d ou. In this case too we will have to say that, though present morphologically, definiteness is actually bleached semantically.5 The situation with nominal FCI contrasts with the FC-FR, where opjos as well as opjosdhipote are available, as we see: (37) a. Opjosdhipote (fititis) irthe sto party efxaristithike. wh+FC-det student came.3sg to-the party was.happy.3sg ‘fWhoever/Whichever studentg came to the party had a great time.’ b. Opjos (fititis) irthe sto party efxaristithike. ‘fWho/Which studentg came to the party had a great time.’ 5 Though this may sound like a weakening of our claim for the role of definiteness in deriving FC meanings from wh-morphology, it is worth pointing out that in essence it amounts to saying that, given the presence of morphological definiteness, it is then up to the syntactic structure to determine whether definiteness will be allowed to play a role in the semantics or not. In the case of the free relative, the structure is closed by a definiteness node (following Jacobson), and the result will be a definite FC phrase (as we will see in more detail in section 4); but in the case of the FCI nominal, we have closure under an indefinite determiner, which blocks, or cancels out definiteness, in either case preventing it from making a contribution in the semantics.


(36) a. Opjosdhipote fititis bori na lisi afto to provlima. FCI student can subj solve this the problem ‘Whichever student can solve this problem.’ b.Opjos fititis bori na lisi afto to provlima. ‘Any student can solve this problem.’ c. fOpjos/Opjosdhipoteg bori na lisi afto to provlima. ‘Anyone can solve this problem.’

152 (In)Definiteness, Polarity, and the Role of wh-morphology Here it can indeed be argued that the FC-wh phrase is derived compositionally by application of -dhipote to the meaning of the bare wh-phrase. Notice, also the contrast with the English FR: only the wh-ever paradigm is allowed.6 We come back to this fact, and the consequences it has for the English FC-FR in section 4, after the analysis of the Greek FR is made clear.

3.2 Exhaustive variation

(38) i-alternatives A world w1 is an i-alternative wrt a iff there exists some w2 such that ½½a w1 6¼ ½½a w2 and for all b 6¼ a: ½½b w1¼ ½½b w2 Two i-alternatives are worlds w1 and w2 agreeing on everything but the value assigned to the FCI a, and we will incorporate this in our definition below (adopted from Giannakidou 2001). Importantly, what is significant for capturing the variation of FCIs is not so much the characterization of worlds as i-alternatives, but rather the availability of possible worlds in the first place. Such worlds will be, naturally, the worlds that the Q operator quantifies over, and some of these worlds 6

Plain what FRs present an exception to this : I ate what he cooked and I ate whatever he cooked. We may also find occasional cases of plain who and where, as a reviewer points out, such as I like who(ever) she likes, and I’ll go where(ever) you go. However, such cases seem to form a distinct class in at least two ways: they involve identical VPs in the main clause and the FR, and are sensitive to anti-episodicity. If these conditions do not hold, bare who becomes ill-formed: (i) a. /?? (Last night at the party) I talked to who you met at the store. b. /?? (Last night at the party) I talked to who you talked to. Additionally, bare wh-FR other than what is not allowed in the subject position: (ii) Who you talked to yesterday was a great guy. It makes sense, then, to treat the occasional instance of bare wh- in FR as exceptional, on a par with what, which is known to be different for other reasons too, and we will not take them to threaten our generalization here. Moreover, the fact that the attested examples of bare wh-phrases mostly concern object positions seem to suggest that these wh-clauses may be genuine clausal objects rather than FRs, an ambiguity created by the fact that there is no morphological marking specific to the FR in English (unlike in Greek).


Apart from intensionality, another important lexical semantic feature of FCIs is exhaustive variation: the FCI variable must be assigned distinct values in each world or situation we consider, and we consider all (relevant) possible worlds. This property derives the quasi-universal effect of FCIs, and Giannakidou (2001) proposes to capture it in the notion of an i(dentity)-alternative (following Dayal 1997):


can be much less stereotypical, i.e. less similar, to the actual one, an assumption necessary to capture the intuition of domain extension and scalarity that we get with free choice. It is also important to emphasize that the existence of possible worlds for variation is a condition on the context of the FCI (i.e. a presupposition), and that the FCI nominal itself cannot introduce these alternatives; if it could, it should be able to do so also in an episodic sentence with the result of licensing itself contrary to fact. (38) gives the criterion for what counts as a valid ialternative: only worlds with differing values for the FCI count as ialternatives. Summarizing, then, we have the following semantics for FCIs (Giannakidou 2001: (129)):

We see that the requirement of exhaustive variation is a universal clause posited as a presupposition of FCIs: the context must provide alternative worlds (i.e. the existential commitment statement at the basis of all universals in (39a)), we must consider all alternatives, and since in each alternative we have a different value, we exhaust the possible values for the FCI.7 In the assertive component, the FC indefinite is bound by the relevant operator and obtains the quantificational force contributed by that operator. This accounts for the quantificational variability effect, i.e. the shift from existential to universal readings, and the intermediate forces observed with Q-adverbs. We will not go into more detail here—for extensive

7 At this point we may wonder whether the FC definition we are using here may be too intensional (thanks to a reviewer for bringing this up). For instance, a sentence like John can beat any of his students at chess is it about actual students of John, or do we also want to include possible but not actual students of John, as would be implied by (39)? The same question would also arise in the absence of the partitive: John can beat any student at chess. The intuition, we believe, is that such sentences are not only about actual students but also non-actual ones (something close to if some x in a world w is a student (of John), then John can beat x at chess in w), thus supporting the definition we give here.


(39) Free choice item (¼ FCI nominal) Let Wi be a non-empty, non-singleton set of possible worlds. A sentence with a free choice item ½½OP DETFC (P, Q) is true in W0 with respect to Wi iff: (where OP is a nonveridical operator; P is the descriptive content of the FC-phrase; Q is the nucleus of the tripartite structure; W0 is the actual world): a. Presupposition: "w1, w2 2 Wi: ½½a w16¼½½a w2, where a is the free choice phrase. b. Assertion: ½½OPw, x [P(x, w); Q(x, w)] ¼ 1 where x, w are the variables contributed by a.

154 (In)Definiteness, Polarity, and the Role of wh-morphology discussion of particular examples and subcases see (Giannakidou 2001). This analysis succeeds in handling a substantial amount of data concerning the distribution and interpretation of FCI nominals. Moreover, it can be made fully compositional, as we illustrated in the previous subsection. However, it does not offer a way to address the difference between FCI-nominals and FC-FRs, and the parallel distinction between any and the wh-ever form in English; nor does it address the related issue of definiteness that we emphasized at the beginning. It is these issues that we take up next.

In this section we will present an analysis of the class of FCIs that we identify as definite. We will take it that FCIs that occur in free relative structures (FC-FRs) exhibit typically this case —although being an FR is not a necessary condition for being a definite FCI (e.g., the Mandarin na˘-items that we discuss in section 5 are definite even though they are not clausal). In order to account for the semantic difference between FCI-nominals (any, opjsodhipote modifying an NP), which we just analysed as indefinite, and FC-FRs (wh-ever, opjosdhipote selecting a CP complement), which we intend to analyse as definite in a way to be made precise shortly, we need to think a little bit more about their empirical differences. It will be helpful to look at these differences as they are manifested in the contrast between any and wh-ever in subsection 4.1. After this is done and we establish that a form of semantic definiteness is involved in the FC-FR construal, we turn to the derivation of the plain Greek FR, adopting the analysis of Jacobson 1995 (section 4.2). Then we discuss the impact of FC-dhipote on the FR (section 4.3). Finally, we revisit wh-ever and explain its non-polarity status in the light of the analysis of the Greek data, and the phenomenon of subtrigging (4.4). It will turn out that the wh-ever FR is actually ambiguous between FC and the plain construal, and it is this that allows it to appear in the purely extensional and veridical context of ignorance readings, where the Greek FC-FR, being only free choice, is disallowed.

4.1 The nature of definiteness in FC free relatives First, the defining syntactic difference between a FCI-nominal (any) and a FC-FR with wh-ever is that the latter is always part of an FR


4 FREE CHOICE FREE RELATIVES


structure and, unlike any, cannot occur independently or as a nominal modifier:8 (40) a. Whoever saw a fly in his soup complained to the manager. b. fWhoever/whichever customerg complained to the manager. c. fAnybody/any customerg can complain to the manager. (with a modal added for well-formedness)

8 A reviewer points out that there are indeed some cases where a wh-ever form occurs as an independent nominal in a non-subject position, e.g.:

(i)

a. Q: What should I do? A: fWhatever / ??Anythingg b. You can leave whenever. c. She complains to the manager about fwhatever/anythingg.

However, there are reasons not to want to treat these cases of self-standing whatever as part of the regular FC wh-ever cases we are talking about here. Note, first, that they have an idiomatic-like interpretation quite different from that of the regular FC-FR. Specifically, self-standing whatever receives only the indiscriminative reading of Horn (2000a), which tends to express only negative evaluation, and lacks the actual free choice reading of the typical FCI which contains also a neutral evaluation. In support of this, notice the contrast in (i)a, where anything is odd as an self-standing answer exactly because it cannot receive in isolation the indiscriminative meaning (which appears typically with just anything). Second, self-standing wh-ever involves specific wh-forms, e.g. whatever, and to some extent whoever and whenever but not, for example, the wh-ever form as a modifier: (ii)

Give the flowers to f?whoever/??whichever studentg.

Finally, note that, unlike regular wh-ever, which is not triggered and can occur in episodic contexts (i.e. subtrigging that we discuss next in the text), these apparent independent forms are triggered, and cannot be used in an episodic context. Notice the contrast between the good cases above, and the bad episodic ones below, as they differ from regular FC-FRs: (iii) (iv) (v) (vi)

Q: Who did he talk to at the meeting? A: Whoever. OK. fWhoever/Anyone whog was willing to talk to him. She left whenever. OK: She left whenever she was told to. At the party, Laura gave the flowers to whoever/OK: whoever she liked. She complained to the manager about whoever/anything (as opposed to the non-episodic (ic)).

In view of this contrast with respect to episodicity, it makes more sense to treat the occasional independent wh-ever not as part of the typical wh-ever that is used in the FR paradigm, but rather as a distinct lexicalization that is an PI itself. Unlike typical FC-FRs, independent PI wh-ever follows the triggering pattern of polarity sensitive FCI nominals, and is ruled out by episodicity. Whether this analysis would necessarily render the independent wh-ever form an indefinite is a question that we will not speculate on here.


The sentences above show that wh-ever is different from any and the nominal FCI we have been discussing so far, which select NP arguments, or can be used as such. If FCIs selecting NPs are intensional indefinites with the structure we described in (35), then the impossibility of wh-ever to select an NP argument suggests that this analysis cannot be extended to this item. In other words, wh-ever

156 (In)Definiteness, Polarity, and the Role of wh-morphology cannot be an intensional indefinite of the Greek type described here. This, of course, squares with the fact that wh-ever does not seem to exhibit polarity behaviour at all—it is good, for example, in positive episodic contexts:

The sentences above are all positive episodic past, hence veridical: anyone and nominal opjosdhipote in (41b,c) are out, as expected because they are PIs (for the reasons we explained in 3.1). Whoever, in contrast, is good in (41a) because it is accompanied by a clausal complement, which enables a subtrigging structure (the term from LeGrand 1975) which is known to ‘rescue’ FCIs in the otherwise hostile episodic environments; and it is subtrigging that rescues anyone and opjosdhipote in (41d,e) too. Hence whoever selects a sentential complement, giving rise to subtrigging; such an item consequently will not be subject to licensing at all, for reasons that we discuss in section 4.4. The presence of sentential structure in auto-licensed FCIs (Korean, Japanese) seems to be the reason for autolicensing and the absence of polarity behaviour in those paradigms (as suggested also in Gill et al. 2002), but we will not discuss the cross-linguistic extension further in this paper. So, there seems to be a split between FCIs selecting NP arguments (the ones we have been calling FCI-nominals), on the one hand, and FC-FRs, on the other, crucially in terms of polarity item status. The analysis we have given in section 3 applies to the former but it cannot extend to the later. Furthermore, the fact that wh-ever heads FRs makes the indefinite analysis less plausible, as FRs are more appropriately treated as definites ( Jacobson 1995), a position that we adopt in this paper. A second difference that supports precisely this split between FCI-nominals (indefinites) and FC-FRs (definites) is observed in the


(41) a. Whoever saw a fly in his soup complained to the manager. b. Anyone complained to the manager. c. Opjosdhipote paraponethike ston diefthindi. (Greek) FCI-person complained.3sg to-the manager d. Anyone who saw a fly in his soup complained to the manager. e. [Opjosdhipote idhe miga sti soupa tu] [wh-ever person saw.3sg fly in-the soup his] paraponethike sto diefthindi (Greek) complained.3sg to-the manager


sentence below, in terms of what appears to be an expectation of existence: (42) a. If any student calls, I am not here. b. Whichever student calls, I am not here.

(43) She may never marry, but a. whoever she does marry will be Jewish. b. anyone she does marry will be Jewish. c.#the person she does marry will be Jewish.

(Horn 2000a)

(44) a. Any beer there is in the fridge is mine. b. Whatever beer there is in the fridge is mine. c. The beer in the fridge is mine. [no existence-neutral reading possible] These contrasts agree with our position that whatever the existential import of the FC-FR is, it must not be deemed identical to the


The sentence (42a) with any is a neutral statement expressing my desire not to talk to anybody, and there is no expectation that somebody will actually call. The one with whichever student (42b), on the other hand, seems to favour (but not require) a context where there is indeed an expectation of call; in fact it can (but does not have to) be an instruction to avoid talking to somebody undesirable. This expectation, which seems to not be as strong as a presupposition, makes sense only in the definite analysis of FRs because we tend to exclude the empty set from the plural FR collection we are forming, as we will suggest below. With an indefinite, there is no such inclination, hence the unmarked use of the FCI indefinite in a neutral context. It is important to emphasize that the expectation of existence that comes with FRs and wh-ever is not as strong as the existential commitment that comes with morphological plural definites. With morphological definites, in English as well as in Greek, the existential commitment appears to be a presupposition (though this must be taken with a grain of salt, see especially the closing discussion in our footnote 9, and our example (45) below), but with FC-FRs, as we noted, it is not. The contrast between morphological definites and FRs is illustrated in the examples below (thanks to an anonymous reviewer); the discrepancy between FRs and morphological definites was noted already in Horn (2000a: 101–102), and earlier in Grosu (1996: 271) where the examples in (44) are taken from:


(45) a. This golden coin belongs to the sailor that sights the White Whale. b. This golden coin belongs to whichever sailor sights the White Whale. (Both consistent with a context where White Whale is not sighted.) Importantly, in this case the definite and the FR are used attributively (Donnellan 1966) rather than referentially, and this appears to have equal impact on both FRs and morphological definites thus supporting their parallel. The fact that FRs express a somewhat looser commitment to existence may suggest that FRs, unlike morphological definites, are used more often attributively than referentially, an expected pattern given the absence of definite morphology (English), or degenerate morphological definiteness (Greek; recall that the 9

Notice, crucially, that universal quantifiers also come with an existence presupposition, or ‘commitment’ (to use Horn’s 1997 terminoloogy), which becomes evident in cases like (i), parallel to the ones discussed in the text: (i)

fEvery/Eachg bottle of beer in the fridge is mine (#but there is no beer in the fridge).

The requirement for a non-empty restriction is particularly strong for the strongly distributive each (which is characterized, for this reason, as veridical in Giannakidou 1998, 1999), and can occassionally be waived in the case of every (which is characterized as non-veridical), just like it can be waived in the attributive uses of definites mentioned next in the text: (ii)

A golden coin will be given to every sailor that helps capture the White Whale.

Similarly, as we shall see, definites are stripped off existence in such cases. FRs, then, seem to be distinct from morphological definites and universals alike with respect to existence. As regards to the differences between universal quantifiers and FRs, we refer to Jacobson (1995: section 1) for a quite extensive discussion.


presupposition of a referential definite description.9 Jacobson herself also notes (1995: 472) that the FR plural is not identical to the morphological plurals in languages with number marking like English (thus also Greek), as in these languages the morphological plural excludes atoms, and probably also the empty set from its domain (though this may not be, strictly speaking, part of its truth conditional meaning). Our intuition that wh-ever seems to favor, but not require, a context (in example (42b)) where there were indeed callers may therefore be taken to indicate that the empty set tends to also be excluded in the case of FR (though atoms, contrary to morphological plurals, are still allowed). In this connection, notice that not all instances of morphological definites are committed to existence. Morphological definites seem to be stripped off existence in certain usages, as in the example below, where the FR and the morphological definite behave alike:


(46) a. Max gave Alice almost f?whatever/anythingg she asked for. b. Max gave Alice almost the things she asked for. (from Larson 1999: fn. 10) Here the wh-FR seems to align with any in being compatible with almost whereas the definite the things is pretty bad (see also Horn 2000a for various Googled examples that illustrate the same point). If we assume, following Lee and Horn (1994) among others, that almost modifies high scalar and universal values, the real question posed by these data is why morphological definites are incompatible with it. It is not decisive for our discussion of FC to address this question here, but a plausible hypothesis, simplifying somewhat, would be to say that the high scalar reading needed for almost modification requires a collective reading which the definite cannot get unless, for example, all is used, in which case the definite becomes fine: (47) Max gave Alice almost all the things she asked for. This suggests that the morphological plural definite must be ‘hopelessly’ distributive in the absence of all. Distributivity is generally at odds with almost: it is this property, for instance, that rules out the


definite marker o that appears with the Greek FR is not, as we mentioned at the beginning, the fully inflected definite determiner, but an uninflected form of it.). Overall, then, our data here indicate that it is important to distinguish morphological definiteness from semantic definiteness. Given that existence can be dispensed with even with morphological definites, as we just noted, the core of semantic definiteness appears to be the formation of a maximal plural entity, and wh-FRs fully meet this condition. In our more detailed analysis below, we identify o as Jacobson’s iota type-shifter, and in the context of the more refined observations we make here it is appropriate to understand iota as the maximality operator, rather than the morphological definite article (which in the referential use contributes both maximality and a presupposition of existence). We will not elaborate on the formal differences at this point, and follow Jacobson (1995) in using iota; and our treatment of iota shift should be understood as simply yielding maximality, and not necessarily presupposing existence, just like in Jacobson’s analysis. In the same context, certain other apparent differences between morphological definite descriptions and FC-FRs become better understood (again, thanks to a reviewer for bringing these up). For instance, the latter but not the former are modifiable by almost:

160 (In)Definiteness, Polarity, and the Role of wh-morphology strongly distributive universal each, and renders it different vis-a`-vis almost from other universal quantifiers like all or every: (48) Max gave Alice almost feach book/every bookg she asked for.

(49) a.

Opjosdhipote (fititis) irthe sto party efxaristithike. wh+FC-det student came.3sg to-the party was.happy.3sg ‘fWhoever/Whichever studentg came to the party had a great time.’ b. Opjos (fititis) irthe sto party efxaristithike. ‘fWho/Which studentg came to the party had a great time.’

We see that the FR wh-phrase in Greek can occur with the FCI modifier -dhipote (in which case it is homophonous with the nominal FCI), or without it, in which case it is a plain FR. The difference is meaningful in Greek in a way that we will try to make precise in the next section. 10 A reviewer suggests an additional difference between definites and FC-FRs in that wh-ever is much better at licensing NPIs than definites, which according to the reviewer, ‘aren’t great’ at this job: (i) I’ll read fwhatever /any/ theg books you’ve ever asked me to read.

Though in this example the definite is indeed odd with ever, this case is far from illustrating the norm. It has been shown in the literature that definites (mainly plural) are in fact good licensers of NPIs (Giannakidou 1998, 1999: 396–404, where the example below is taken from): (ii)

The students who know anything about the case should speak now.

According to Giannakidou the crucial factor is whether the definite presupposes existence of a nonempty domain (in which case it is used referentially), or not (in which case it is used attribitively, as we just saw). When it is unclear whether the domain is non-empty, i.e. in the attributive use, NPIs will be OK, as in (ii). If the FC-FR turns out to be freer in licensing NPIs, this supports our earlier suggestion that FRs tend to be used attributively more often than morphological definites. And, again, the fact that definites can come with empty domains (as is possible with (ii)), contradicts the view that definites always contribute existence, a fact that actually brings them closer to FRs, hence supporting the parallel we are trying to establish.


More discussion at this point would lead us too far afield; suffice it to say that almost and the contrast between plural definites and FC-FRs seem to indicate that FC-FRs are not strongly distributive the way morphological plural definites are. But this is not necessarily an argument against them denoting maximal pluralities in the sense we assume here; it is rather another indication that it would be wrong to collapse morphological with semantic definiteness.10 Finally, an additional decisive syntactic fact about wh-ever is that, generally, it cannot occur without -ever, in contrast to its Greek counterpart. We repeat below the contrast from 3.1.


4.2 The derivation of the plain FR in Greek The use of the definite marker o and the FR source of the whcomponent of the Greek FCI support the idea that FRs denote (contextually specified) maximal plural entities; see Jacobson (1995), Rullmann (1996), also Dayal (1995) for a similar analysis of Hindi correlatives, including those containing the FC additive marker bhii. In what follows we will use the analysis of the FR as involving an iota function (which was also developed against an earlier universal analysis of FRs) to propose a derivation of the Greek FC-FR, and its Chinese counterpart na˘-CL NP in section 5. The idea that the Jacobson analysis is appropriate for Greek FRs is found already in Alexiadou and Giannakidou 1998 (pace Iatridou and Varlokosta 1998 who claim that Greek FRs are universal quantifiers); and primary support for it comes from the use of the definite o that appears in the Greek FR. Jacobson argues that FR wh-words in a language without morphological distinctions like English are not born exactly with the same meaning as the similar wh-relative ( Jacobson 1995: 467). Following the classical thesis of Cooper (1983), all wh-phrases start out as properties (the wh-core; type et), but according to Jacobson, the FR undergoes type-shifting to a definite meaning. Here is Jacobson’s derivation (using upper case variables; Jacobson 1995: 473): (50) a. ½½whatFR ¼ kPkX[P(X) ^ "Y (P(Y) / Y < X)]


Our proposal in a nutshell is the following. Free choice variables can be definite (denoting maximal plural individuals) or indefinite, depending on whether the FCIs take nominal or CP arguments. The possibility of a definite variable is fully consistent with, and indeed predicted by, the variable approach to FCIs that we adopt here, and our proposal should be seen as a refinement in this direction. Nominal any exhibits the case of an indefinite variable, and wh-ever illustrates the case of a definite FC variable. In a language like Greek, lacking the phonological distinction between a non-wh and wh-FCI, the wh-form employed encompasses both readings, which are however derived in distinct ways. The FCI-nominal is an indefinite, but the FC-FR contains a definiteness function; the difference is shown to have repercussions for their polarity status, as is the case in English (any being a PI, but wh-ever not). We now start with the derivation of the plain Greek FR adopting the analysis of Jacobson (1995).

162 (In)Definiteness, Polarity, and the Role of wh-morphology b. ½½what John ordered ¼ ½½what (½½ John ordered) ¼ kPkX[P(X) ^ "Y (P(Y) / Y < X)] (kx ordered( j, x)) ¼ kX[ordered( j, X) ^ "Y(ordered( j, Y) / Y < X)]

(51) ½½what John ordered ¼ iX[ordered( j, X) ^ "Y(ordered( j, Y) / Y < X)] The last type-shifting step is invisible in English, but it is visible in Greek which uses the definite article o. Hence Greek offers direct empirical evidence for the maximality type shifter posited by Jacobson, and the FR is interpreted as a maximal plural definite (with the caveat about existence that we noted in the previous subsection.) We illustrate now the derivation for the simple FR wh-opjos ‘who’, building on the analysis of Jacobson’s which, as Jacobson suggests (2005: section 2) can be seen as a variant of the Comp analysis (see originally Groos and van Riemsdijk 1981). Recall that in Greek we have the option of having FC-FR with or without -dhipote, an option that does not, under normal circumstances, exist in English (whoever came to the party versus who came to the party). Naturally, the wh-ever in English will encompass both Greek paradigms. The derivation of the regular FR will closely follow Jacobson’s analysis with the wh-phrase denoting a set of individuals, Greek o being the iota, and no intensionalization; the -dhipote wh-FR will be derived compositionally from it. In our notation below we use lower case individual variables instead of Jacobson’s set variables, assuming that x may be a plural entity (atomic entities being a subcase thereof, again following Jacobson). We will also take it that the contribution of the relative wh-pronoun is to trigger the k-abstraction rule of Heim and Kratzer’s (1998: 96, rule (15)), and that C itself is semantically vacuous, unlike, for example, the interrogative C. We will see cases, however, in the next subsection where C actually hosts free choice complementizers.


The FR wh-expression characterizes the set of maximal plural entities that John ordered. Plural entity is assumed in the broadest sense, and includes both atomic individuals as well as the null set ( Jacobson 1995: 472). So the wh-expression characterizes some subset of e, the set of plural entities (including atoms and the empty set). This means that the FR-wh expression in effect denotes a set which is guaranteed to be a singleton. It then iota type-shifts, quite expectedly, to denote the single individual characterized by the predicate, as shown below (again, from Jacobson 1995: (57)):


(52) FR without -dhipote DP, 7 D, 6

CP, 5

| o

whP,4 pjos

C', 3 C,2

IP,1

½½IP ¼ came.to.party(t1) ½½C ¼ kp.p (identity function) ½½C# ¼ came.to.party(t1) ½½pjos ¼ no denotation of its own; it triggers k-abstraction by predicate abstraction (Heim and Kratzer 1998: 96) 5. ½½CP ¼ kx.came.to.party(x); or kx[P(x) ^ "y(P(y) / y < x)]. The second formulation makes explicit the fact that x may be a plural individual. Keeping this in mind, in our derivation we will henceforth adhere to the simpler clause for simplicity (see also Rullmann 1995). 6. ½½o ¼ kP i(kx.P(x)) Here o is treated as the maximality operator on the set characterized by the IP of the free relative (and not as a variable binder, as in Jacobson’s formula that we gave earlier (51). 7. ½½DP ¼ ½½o (½½CP) ¼ kP i(kx.P(x)) (kx.came.to.party(x)) ¼ i(kx.came.to.party(x)) 1. 2. 3. 4.

In clarification of the final steps 6, 7, given Jacobson’s assumptions, kx.came.to.party(x) will be a singleton set consisting of a plural entity comprising all individuals that have the property that they came to the party. When i applies to this set, the result is to create the plural individual that corresponds to this set. The i(kx. came.to.party(x)) in the actual world w0, will give us the maximal set of individuals that came to the party in w0. This indeed seems to be the meaning of the plain FR in Greek.

4.3 Adding free choice to the free relative We take it here that the composition of wh+ dhipote happens in the morphology (since syntactically the units are indistinguishable), and that


t1 erthi sto parti

164 (In)Definiteness, Polarity, and the Role of wh-morphology it is compositional application of dhipote to the CP derived after application of predicate abstraction with the rule triggered by pjos. This is illustrated in the steps 6 and 7 below, with the addition of the FC-CP node; pjos further moves into the FC node and we get the right linear order: (53) Free choice FR DP, 9 D, 8

FC, 6 -dhipote

CP, 5 whP,4 pjos

C', 3 C,2

IP,1 t1 erthi sto parti

½½IP ¼ came.to.party(t1) ½½C ¼ kp.p (identity function) ½½C# ¼ came.to.party(t1) ½½pjos ¼ no denotation of its own; it triggers k-abstraction by predicate abstraction (Heim and Kratzer 1998: 96) 5. ½½CP ¼ kx.came.to.party(x) Thus far the derivation proceeds exactly as in the case of the plain FR we just discussed. Now, we add FC: 6. ½½-dhipote ¼ kP<s, et> kw kz.P(z)(w) This would require lifting first the CP property kx.came.to.party(x) to its intension kw.kx. came.to.party(x)(w). After this is done (and we may assume an additional node for it, though we do not illustrate here), we get: 7. ½½FC-CP ¼ ½½-dhipote (½½CP) ¼ kP<s, et> kwkz.P(z)(w) (kw.kx.came.to.party(x)(w))¼ kwkx.came.to.party(x)(w). 1. 2. 3. 4.

This is the intensional FC-FR set, namely the intension of the set characterized by the FR predicate. It can thus apply to any possible world w yielding the individuals that have the CP property in that world. If applied to the actual world the function will naturally yield the set of individuals that came to the party in the actual world. The input to iota will thus now be of type <s, et>.


o

FC-CP, 7


8. ½½o ¼ kP<s, et> i(kwkx.P(x)(w)) 9. ½½DP ¼ ½½o (½½FC-CP) ¼ kP<s, et> i(kwkx.P(x)(w)) (kwkx.came.to.party(x)(w)) ¼ i(kw.kx.came.to.party(x)(w))


Hence the wh-FR opjosdhipote IP denotes a maximal sum, and, let us again repeat, in an extensional (episodic) sentence, i(kwkx.came. to.party(x)(w)) will apply to the actual world w0, giving us the maximal set of persons that came to the party in w0, just like with the plain FR. However, because the input of i is a function from worlds to sets of individuals (characterized by a given property P in those worlds), inside the FC-set there will be actual as well as possible individuals (more accurately, world-individual pairs). It is this that gives the distinctive domain extension FC flavour to the FR, since we may consider individuals also in less stereotypical or accessible alternatives. At the same time, as we said, the FC-FR can be used in an episodic context and be quantified over. This is the case of subtrigging that we discuss next. In the derivation of the FC-FR the w variable remains dependent because it is bound by k, and not because it remains unbound as the w variable of the indefinite FCI in an episodic context. This explains why the FR will generally be licensed in episodic contexts (unlike the FCI indefinite), and we believe it captures accurately the way the FR, via subtrigging, actually rescues such structures. Let us comment briefly on our final result. If our analysis of FC-FR is correct, then our account is consistent with Jacobson’s suggestion that the FC-FR is really not so different from the plain one, certainly not in terms of quantificational force ( Jacobson 1995: 479–481). According to Jacobson, what FC adds is broadening of the atomic domain; it is this broadening that we tried to capture in the formation of opjosdhipote that we proposed here. As we are suggesting, the creation of this extended modalized domain of the FR carries with it a scalar inference. This can be viewed as a bottom of scale inference in that it allows us to include in the domain less likely or less prototypical individuals (cf. widening in the sense of Kadmon and Landman 1993). It can thus be seen as containing an even meaning which would be consistent with the observation that expressions corresponding to even, often with concessive meaning (see especially Lee 1997), and additive particles are used with FCI in many languages. Crucially, d ou in Mandarin is also used as even, a fact that fits nicely in this picture. The low ranking inference, as we said, carries with it a concessive meaning roughly equivalent to no matter what/who which becomes particularly clear in cases of, for example, French FCIs like n’importe qui,

166 (In)Definiteness, Polarity, and the Role of wh-morphology and connects to the indiscriminative reading identified in Horn (2000a) that we mentioned earlier. In Greek, the concessive flavour arises also when -dhipote is absent but some other concessive marker is used, e.g. the concessive ke an or modal ke na particles that can be added to the plain FR: (54) Opjos fke an/ke nag irthe sto parti, who.FR and-if/and-subjunctive came.3sg to-the party, efxaristithike. was.happy.3sg ‘Whoever came to the party had a good time.’

(55) Opjoshipote (f?ke an/?ke nag) irthe sto parti, who.FR and-if/and-subjunctive came.3sg to-the party, efxaristithike. was.happy.3sg ‘Whoever came to the party had a good time.’ Given that ke an/ke na are complementizers in Greek (ke an introducing concessive sentences par excellence, i.e. ‘even if ’, and na being the subjunctive complementizer), it can be argued in the light of the examples here that the FR complementizer is not semantically vacuous but quite the contrary: it can host FC modalization. This idea would further substantiate the intuition that FRs are really a wh-species distinct from the relative or interrogative, hence also in the function of their C. We will not pursue the details in this paper. We will just comment that, obviously, in the event FC happens at C, the analysis we have proposed in (53) will have to be modified so as to capture the fact that C, instead of FC-CP is the locus of FC application. Crucially, the fact that free choice can happen at C follows only under the Comp analysis of FRs that we adopt (and for which Jacobson 1995 argued extensively), and remains problematic for the competing Head analysis of FRs (Bresnan and Grimshaw 1974) which expects C to be vacuous. Hence the facts here can ultimately be used as an argument in favour of the Comp analysis, a welcome result from this perspective.

4.4 Wh-ever and a reconsideration of sub-trigging The two things that we noted about wh-ever were first that it occurs mainly as an FR creating subtrigging in episodic contexts; hence it is not subject to licensing. Second, there is typically no option of using


The FC flavour of this sentence is equivalent to its counterpart with -dhipote that we give below. It is probably this equivalence that explains the redundancy with -dhipote and fke an/ke nag:


bare wh in FR in English: who came to the party had a good time (with the exceptions that we noted earlier). Subtrigging refers to a structure where the PI is followed by a relative clause in an episodic context, and generally has a rescuing effect, as we noted. For example, it allows any to appear in an episodic context that would otherwise rule it out, e.g. episodic perfective past. (56) a. Anyone brought a present. b. Anyone who came to the party brought a present.

(57) a. I ate whatever he cooked. b. I ate anything he cooked. c. Efaga otidhipote majirepse. ate.1sg FCI-thing cooked.3sg The FCI-nominal, i.e. anything in (57b), will be translated as a mere indefinite bound by the conditional universal quantifier: (58) "w,x [cooked (he, x, w) / ate (I, x, w)] This LF, augmented with the presupposition of exhaustive variation that we described in section 3, captures the universal-like meaning of any without analyzing it as a universal (unlike Dayal). When we think of the sentences with the FC-FRs, however, two possible analyses of subtrigging emerge. One will be to stick to the conditional structure of Giannakidou and Quer. In this analysis we will have the following LFs: (59) Greek (57c) with FC-FR "w, x[x ¼ i(kw#kz.cooked(he)(z)(w#)) / ate(I, x, w)], where w# is a world accessible from w, including w itself


The fact that wh-ever, as opposed to opjosdhipote which can also take a nominal argument, typically forms an FR renders wh-ever structures exclusively subtrigging structures, thus making wh-ever a non-PI. Additionally, the absence of bare wh-FR in English renders wh-ever ambiguous between the FC and the non-FC FR we just discussed. Dayal (1995), in her discussion of Hindi correlatives, and later Dayal (1998) in a discussion of FC-any analyses subtrigging as involving universal quantification over individuals. Quer (1998) and Giannakidou (2001), on the other hand, argue that subtrigging has a conditional structure and allows FCIs by satisfying non-veridicality. This analysis renders subtrigging parallel to the indefinite FC cases where a Q-operator binds a world variable. Subtrigged sentences will then have the following analysis given what we have said so far:

168 (In)Definiteness, Polarity, and the Role of wh-morphology (60) English (57a) with wh-ever FR: ambiguous: a. "w, x [x ¼ i(kz cooked(he)(z)(w)) / ate(I, x, w)] ¼ the meaning of the plain Greek FR, derived as in (52) b. "w, x [x ¼ i(kw# kz. cooked (he)(z)(w#)) / ate (I, x, w)], where w# is a world accessible from w, including w itself ¼ the meaning of the FC-FR in Greek, derived as in (53)

(61) a. Whichever woman is the editor of this magazine got a prize last night.


The other possible analysis will be to derive subtrigging directly from the FR meanings of plain and FC-FRs we have described in this section. The extensional FR, as we said, is expected to be fine in all episodic contexts, since when applied to the actual world, the iota function will give us the maximal plural entity that satisfies the predicate denoted by the FR CP in the actual world. Interestingly, the intensional FC-FR should also be fine in an extensional/episodic sentence, but making us think of possible individuals—what we have described as the broadening scalar effect. This explains easily why FC with clausal structure defies episodicity without appealing to the conditional rule and makes the analysis simpler. Hence we distinguish FC-FRs, which are not really licensed, from indefinite FCI-nominals, which are (because they need their dependent w variable to be bound by an operator), by appealing to their compositions directly. With the latter, we expect the conditional to create the appropriate context for binding the w variable, hence the need for anti-episodicity; but in the former, there is no such need. Interestingly, the difference between FCI nominal any and its subtrigged incarnation renders this item ambiguous in the way Greek opjosdhipote is, between a derivation yielding an indefinite (as a FCI nominal) and one yielding a definite FCI, which is not subject to licensing and which would be parallel to the FC-FR. Would that mean that FC any is lexically ambiguous between a definite and an indefinite? Not really. Just as in the Greek case, definiteness or indefiniteness is not a case of lexical ambiguity, but of ambiguous syntax: it relies on whether the FC-phrase is closed under an indefinite or a definite function (derived as in (35) versus (53) respectively). The free choice core is intensionality, which is derived identically in both cases. Now, when we further consider the difference between plain versus FC-FRs with wh-ever, we must note that it is the possible interpretation of wh-ever as a non FC-FR that explains why wh-ever can be used in episodic sentences receiving the so-called ignorance readings:


b. Whichever movie Avon is playing right now is boring. (from Jacobson 1995) The above sentences are about one particular woman and one particular movie respectively, and they express ignorance in that the speaker does not know exactly what the movie is or who the woman is (see Tredinnick 1994 for a discussion of similar cases in English that she labels don’t know readings; see also von Fintel 2000). Dayal (1995) presents similar examples with Hindi correlatives which routinely receive such readings:

These ignorance contexts are episodic, and in this particular case also deictic. The fact that wh-ever is fine in such cases is consistent with its being interpreted as an extensional individual, i.e. under the expected bare FR interpetation we have illustrated in (52). Hindi correlatives and English -ever FRs can thus be directly interpreted this way—though English whichever more marginally so in the deictic context, as we see above.11 Yet the Greek FRs with FCIs are hopelessly odd in this context: (63) a. ?#Opjadhipote jineka ine i arxisindaktria aftou FCI woman be.3sg the editor this.gen to periodikou, the.gen magazine.gen pire ena vravio xthes vradi. got.3sg a prize last night ‘Whichever woman is the editor of this magazine got a prize last night.’ b. ?#Opjodhipote koritsi stekete eki ine fili FCI girl stand.3sg there be.3sg friend tou Jani. the.gen John.gen. ‘f?Whichever girl is standing there/The girl standing thereg is Jani’s friend.’ 11 A reviewer suggests that what makes whichever girl is standing there slightly odd is that it is unlikely that one is in a position to know that a girl is standing there without being able to identify her. Compare with the following which are fine exactly because identification is possible:

(i)

a. b.

Whichever one of my roommates had a late snack (here) left a mess. Whichever girl is standing outside my window is yelling too loudly.


(62) Jo bhii laRkii vahaa khaRii hai vo Ravi kii dost hai. which ever girl there standing is she Ravi’s friend is ‘f?Whichever girl is standing there/The girl standing thereg is Ravi’s friend.’

170 (In)Definiteness, Polarity, and the Role of wh-morphology The oddity of the Greek examples is due to a conflict between the deictic reference imposed by the context and the FC result of having an intensionalized property as the input to iota. Consider what the sentence (63b) actually says: (64) friend.of.John’s (i(kwkz.woman-standing.there (z)(w)))

(65) Opjadhipote jineka ine i arxisindaktria aftou FCI woman be.3sg the editor this.gen to periodikou, perni sinithos pola vravia. the.gen magazine.gen, get.3sg usually many awards. ‘Whichever woman is the editor of this magazine usually gets a lot of prizes.’ (66) Opjodhipote koritsi stekete eki ine sinithos fili FCI girl stand.3sg there be.3sg usually friend tou Jani. the.gen John.gen ‘Whichever girl stands there is usually Jani’s friend.’ (‘there’ is a meeting point for Jani’s friends). This exclusive compatibility with only generic readings is expected if Greek FC-FRs are intensional FCIs only, as we are arguing here, and must be quantified over. These sentences would have structures parallel to the ones in (59) but with the respective Q-operators: (67) USUALLYw, x[x ¼ i(kw# kz.woman(z)(w#) ^ editor.of.this. magazine(z)(w#)); gets.many.prizes(x, w)], where w’ is a world accessible from w, including w


The property kwkz.woman-standing.there(z)(w) will include actual as well as possible individuals in it, as we said earlier. So iota will create the complex individual that contains these individuals (or world-individual pairs). But this is too big an individual in the deictic context employed, and certainly not the one intended, hence the oddity. Note, crucially, that it is indeed a sort of deviance that we are talking about here, and not crude ungrammaticality. On the other hand, with generic present tense or with the addition of a Q-adverb, the sentences become fine; and then, of course, we no longer talk about a particular woman or a particular girl (in fact the case of girl is neutral with respect to number, i.e. it can be more than one girl at a time standing there, as expected since there is no uniqueness condition in ‘standing there’, unlike with being the editor of a magazine):


(68) USUALLYw, x[x ¼ i(kw# kz.girl(z)(w#) ^ standing.there (z)(w#)); friend.of.Johns(x, w)], where w’ is a world accessible from w, including w

5 Na˘-CL NPs AND CHINESE FCIs What we have seen in Greek FCIs is a distinction between indefinite FCIs (FCI-nominal) and definite FCIs (FC-FRs). The definiteness distinction is thus reflected in a distinction in the type of arguments that the FC takes. In Mandarin Chinese, as we will show, the definiteness distinction is also found, but it does not correspond to a difference in the type of arguments. Rather, it is the presence or absence of d ou that makes it visible. We propose that d ou in Chinese expresses the iota operator (corresponding to o- in Greek), and the intensionality in FCIs in Chinese comes from a possibly covert wu´lu`n ‘no matter’. As we noted in section 1 already, the licensing of na˘-CL NP ‘which NP’ in Chinese is not the same as the licensing of typical


And these, of course, are not cases of subtrigging (which characterizes episodic contexts) but regular quantified structures. Again, it is worth emphasizing that Greek does not allow ignorance readings of opjoshipote FRs, a fact supporting the idea that we are dealing with a real ambiguity in English and the other languages that allow both readings. To sum up, in section 4 we have shown that we need to separate a definite-like (maximal) FC meaning, and, in the cases we looked at, this meaning is a result of FCIs occuring in FR structures. The FC morpheme intensionalizes the wh-phrase by adding a w variable to it, and this variable remains bound by the k operator. Unlike with nominal FCIs, the intensionalized wh-property in the FC-FR is the input to an iota type shifter, contributed by the definiteness marker o, and the FC-FR denotes a maximal intensional plural. Our analysis stayed close to Jacobson’s at every step, spelling out precisely the contribution of FC, and offered a straightforward way of capturing the difference between Greek FC-FRs and wh-ever in English, as well as a core similarity, i.e. that they are not polarity sensitive the way indefinite FCIs are. As an additional welcome result, the account enabled a novel look at subtrigging by doing away with the conditional analysis, since it derived the non-polarity status of FC-FRs in episodic contexts by appealing directly to the FR denotation. We are now in position to examine the Chinese na˘-CL FCI which provided part of the motivation for initiating a definite analysis for FCIs.


Table 2 Chinese FCIs. Environments 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19.

Episodic negation Episodic questions Conditionals Restriction of universal Future particle/will Modal verbs Directive intensional verbs Imperatives Habituals Stative verbs Generics NP-comparatives only/zhi Negative factives Affirmative episodic sentences Existential constructions Epistemic intensional verbs Progressives Non-negative factive verbs

bare wh

na˘-CL

re`nhe´-NP

OK OK OK OK^ OK^ OK OK^

OK OK^ OK OK^

OK OK OK OK OK OK OK OK^ OK OK

OK^: indicates OK in cases which there is fronting plus d ou.


wh-indeterminates (such as bare wh); nor is it the same as the non-whFCI re`nhe´-NP ‘any NP’ (see Cheng and Giannakidou 2005). Bare wh’s behave like PIs in Chinese (see Cheng 1991, Li 1992, Lin 1998) and they can also be interpreted as FCIs, which contrasts with re`nhe´ NPs, which are always interpreted with free choice meaning. We will not repeat the data here, but see Cheng and Giannakidou (2005) for an extensive comparison between na˘-CL NP and the other paradigms. We do want to add the observation that na˘-CL NP as FCI exhibits a more limited distribution than that of the Greek FCI: it is only licensed in modal contexts, as can be seen in Table 2. We also see in Table 2 that FC readings are available for other whparadigms in Chinese, which end up behaving like PIs as a class. Clearly, we want to explain why the definite form na˘-CL NP exhibits the limited distribution it does; for now, however, we will leave this question open. We will concentrate here on the behaviour and properties of na˘-CL NPs and what they can tell us about the ingredients of FCIs in Chinese (in comparison with Greek).


5.1 Definite vs. indefinite FCIs in Chinese

The sentences in (69b-c) show that d ou is not obligatory. D ou appears to be obligatory with the modal ke˘yı˘ ‘can’, as we can see also from (70): (70) a. Ta d ou ke˘yı˘ jie` na˘-beˇn sh u. he d ou can borrow which-CL book Intended: ‘He can borrow any book.’ (d b. Ta na˘-beˇn sh u ou) ke˘yı˘ jie`. he which-CL book d ou can borrow ‘He can borrow any book.’ Na˘-beˇn sh u ‘which book’ cannot stay in its canonical object position when it appears under the modal ke˘yı˘ ‘can’, but must front to a pred ou position.12 In (70b), d ou is obligatory. One may ask whether or not d ou can appear in sentences such as (69b) and (69c). The answer is yes. The question which arises then is ou and na˘-CL whether there is any difference between na˘-CL NP with d NP without d ou. To answer this question, consider first the sentences in (71), which are comparable to the pair we discussed earlier about English and Greek: (71) a.

12

(i)

Ru´guo˘ if

(yo˘u) na˘-ge reń have which-CL person

da˘-dia`nhua` telephone

laí come,

Na˘-beˇn sh u ‘which book’ in (70b) can also appear preceding the subject t a ‘he’, as in (i):

Na˘-beˇn sh u ta d ou which-CL book he d ou ‘He can borrow any book.’

keˇyıˇ can

jie` borrow


Though na˘-CL NP is more restricted in its distribution, it is similar to other FCIs in that it does not necessarily require d ou in order to have a free choice reading (see Cheng and Giannakidou 2005 for more details). Consider the following sentences: (69) a. Na˘-ge xue´sheng (d ou) ke˘yı˘ jı`n-laí. which-CL student d ou can enter ‘Any student can enter.’ b. Ta bu` xia˘ng ma˘i na˘-be˘n sh u. he not want buy which-CL book ‘He doesn’t want to buy any book.’ c. Ru´guo˘ na˘-ge reń qifu` nı˘, . . . if which-CL person bully you ‘If anyone bullies you, . . ..’


In (72a), the bare wh na˘r ‘where’ appears under negation. This sentence can be used in cases where there isn’t anywhere in particular or special that he wants to go to (though he may indeed want to go some place or other). In contrast, (72b) means that there is absolutely no place, of a contextually determined set of places, that he wants to go. Again, we have the flavor of wanting to exclude the empty set that comes with definite-like expressions. The same contrast can be shown with (69b), repeated in (73a), in contrast with (73b): (73) a. Ta bu` xia˘ng ma˘i na˘-beˇn sh u. he not want buy which-CL book ‘He doesn’t want to buy any book (in particular).’ b. Ta na˘-beˇn sh u d ou bu` xia˘ng ma˘i. not want buy he which-CL book all ‘He does not want to buy any book at all.’ (73a), without d ou, can be interpreted as ‘he does not want to buy any particular book’; but (73b) can only be interpreted as ‘there is absolutely no book what-so-ever (from a contextually determined set) that he wants to buy. In (73a) there is no contextually determined set of books that we are talking about.


jiu` su o wo˘ bu` zaì. then say I not be ‘If anyone calls, say that I’m not here.’ b. (wu´lu`n) na˘-ge reń da˘-dia`nhua` laí, wo˘ no-matter which-CL person telephone come I. d ou bu` zaì all not be ‘Whoever calls, I’m not here.’ Though both sentences are grammatical, (71a) cannot be used in situations in which the phone is ringing. It seems then that d ou contributes the tendency observed for the definite plurals and FR, namely to exclude the empty set. In fact, there is no necessary expectation of a call in (71a) while in (71b) there is. (72a, b), with bare wh-forms, further illustrate this difference: (72) a. Ta bu` xia˘ng qu` na˘r. he not want go where ‘He does not want to go anywhere (in particular).’ b. Ta na˘r d ou bu` xia˘ng qu`. he where d ou not want go ‘He does not want to go anywhere what-so-ever/at all.’


(74) a. (Wu´lu`n) sheí d ou ke˘yı˘ laí. no-matter who d ou can come ‘No matter who can come.’ b. (Wu´lu`n) nı˘ zuo` sheˇnme, wo˘ d ou meí yı`jia`n no-matter you do what I d ou not opinion ‘No matter what you do, I won’t have an opinion.’ Note that wu´lu`n ‘no matter/regardless’ is optionally present; note also that the above examples contain a bare wh-word. According to Lin (1996), ‘‘. . . the function of wulun is to form the generalized union over the set of propositions, i.e., the set of sets of situations, denoted by the wh-clause following it.’’ (p. 76) We propose that wu´lu`n (overt or covert) is actually the element that provides the intensionalization, along with the presupposition of exhaustive variation (which as we said in section 3.2 corresponds to a universal statement about i-alternatives, bringing our analysis close to Lin’s). So we take the meaning of wu´lu`n ‘no matter/regardless’ to be equivalent to that of the Greek intensionalizer –dhipote: (75) ½½wulun ¼ kP<s,

et>

kwkx.P(x)(w)

D ou, on the other hand, is a generalized distributive operator in Lin’s analysis. In the case of wu´lu`n . . . d ou, it distributes over the set of situations in the generalized union corresponding to the denotation of the wu´lu`n-clause. In our analysis, this gets translated into a claim 13

Though we are not entirely convinced that ALL d ou sentences are elliptical wu´lu`n . . . d ou sentences, we believe that the ones expressing FCIs must be; see Cheng and Giannakidou (2005) for more data and arguments for this.


If this characterization of the difference between FCIs with d ou and FCIs without d ou is correct, it appears that d ou contributes to the definiteness and exhausitivity of the FCIs. We can then ask how that is done. From Greek definite FCIs, we see that there are two necessary ingredients: intensionality and the maximal sum formation (iota). The question then is where these ingredients are to be located in the Chinese definite FCIs. To answer the first question, we follow Lin (1996) who analyses all d ou sentences as wu´lu`n. . .d ou sentences (wulun ¼ regardless). In other words, typical d ou sentences are just elliptical wu´lu`n. . .d ou sentences.13 There are thus two kinds of wu´lu`n . . . d ou sentences. One is clausal and one is nominal (examples and translations from Lin 1996) (note that regardless of whether wu´lu`n takes a nominal or a clause, d ou is present):

176 (In)Definiteness, Polarity, and the Role of wh-morphology that d ou is the iota operator (like the definite article o in Greek). Here we define it as taking an intensionalized property as its input, since it is such a property that it combines with after application of wu´lu`n: (76) ½½d ou ¼ ½½o ¼ kP<s,

et>

i(kwkx.P(x)(w))

(77) a. ½½na-ge ren ¼ kw.ky.person(y)(w) b. ½½wulun (½½na-ge ren) ¼ kP<s, et> kwkx.P(x)(w) kw.ky.person (y)(w)) ¼ kwky.person(y)(w) This intension of the person property is then the input to d ou: (78) ½½d ou (½½wulun na-ge ren) ¼ kP<s, et> i(kwkx.P(x)(w)) (kwky. person (y)(w)) ¼ i(kwky.person(y)(w)) The w variable of this intensional quantifier will be dependent, just like in the case of the Greek FCI-FR because it will be bound by the k-operator under iota. This analysis entails that what we see in the Greek morphology, we see in the syntax in Chinese. In Greek, the morphological composition of FCIs has a definiteness marker as well as the FC determiner -dhipote, providing the intensionality. Chinese on the other hand, does not have these ingredients in the morphological composition. Rather, they are


D ou is made parallel to Greek o, though it might be more accurate to attribute to it both maximality and exhaustivity as we said earlier. We have not discussed the differences between the two in this paper, and we will therefore continue employing iota as a neutral term to cover both maximality and exhaustivity (see Cheng and Giannakidou (2005) for some more discussion). Extending our account, we claim that the strategy to employ focus particles (mo, bhii) in FCIs crosslinguistically reflects precisely this need in languages with whindeterminates: to create definite FCIs. At this stage, however, our claim remains at the level of speculation, since clearly more work needs to be done. Now, wu´lu`n composes with na˘-CL, or a bare wh-, and the product is always an FC meaning. This means that wu´lu`n is always present with na˘-CL, as we suggested in footnote 13 already, but it is optionally present with bare wh. This, again, is how we relativize Lin’s claim that all d ou sentences contain wu´lu`n. Given what we said so far, the denotation of na˘-CL+noun+d ou is in ou. This suggests the following fact always wu´lu`n+na˘-CL+noun+d composition parallel to opjosdhipote (without iota):


(79) ½½which mani ¼ the (ky[man(y)(w) ^ y ¼ xi]) The output of this function will be the unique individual that is a man in a given context. At the same time, the world variable w can either be bound by the question operator in C0—in which case it acquires a dependent or intensional value—or it can be free in which case it is assigned the value of the actual world. Beck and Rullmann (1999) actually argue for this choice, and the difference is used to account for 14

Chinese differs from Greek however, in that the Greek definite-o is present in all the FCIs, definite or indefinite alike; d ou, in Chinese, however, is only present in the definite FCIs. Yet recall that in our derivation of the indefinite nominal FCI in section 3.1 we treated opjosdhipote as a lexical unit, thus blocking the contribution of o in this case. In Chinese, the addition of d ou is the mechanism of switching from an indefinite to a definite FCI, suggesting that Chinese wh-elements are indeed indefinites inherently.


present in the syntactic composition of FC, with wu´lu`n. . .d ou, 14 providing the intensionalization as well as iota. There are two final things one may wonder about. First, why would a D-linked wh-phrase such as na˘-CL NP ‘which NP’ require an additional element, i.e. d ou to express iota (or maximality). To answer this question, we note that the use of the D-linked wh-paradigm for definite FCIs does not exclude the use of d ou; indeed, d ou provides the maximality needed, which the wh-phrase alone does not have. Moreover, when na˘-CL NP appears without d ou (such as in cases under negation and in conditionals), we will then be dealing with an indefinite FCI. This is reminiscent of the Greek FCI nominal such as opjosdhipote, which may also be an indefinite FCI though o- is also present. Second, how come which-phrases in English are not FCIs? This question is particularly pressing in the context of Den Dikken and Giannakidou’s (2002: 42) claim that which-phrases are presuppositional and cannot be used as PIs. In favour of this claim they illustrate, e.g. that the PI the hell cannot modify which-phrases: which student the hell, or self-standing partitive which as in There are a lot of books on the table – which (the hell) do you want? This observation, which correctly captures the fact that which-phrases in English are not PIs, raises a question about the consequences of our analysis: if which provides prime material for polarity FC status, then why is it that which-phrases in Chinese but not in English are FCIs? To answer this question, the recent analysis of English which as a definite description in Rullmann and Beck (1998), and Beck and Rullmann (1999), will be relevant. These authors explicitly analyse which-phrases as definites, and propose the following denotation for which man:


6 CONCLUSION One of the main conclusions to be drawn from this paper is that the link between wh-morphology and free choice supports the variable analysis of free choice phenomena (Giannakidou 1998, 2001; Horn 2000a, 2000b, 2005). The novel refinement that we proposed is to


de re and de dicto readings in intensional contexts. What matters for our purposes is that assimilation of a which-phrase to a definite is independently motivated. If we compare Rullman and Beck’s formula to our denotation of wu´lu`n na˘-ge reń ‘wu´lu`n which person’ in (77), we see that the w variable in wu´lu`n na˘-ge reń ‘wu´lu`n which person’ is bound by the k and can thus not be free. This explains the fact that na˘-CL nominal in Chinese is dependent but which man in English is not. And it is all due to the previous application of wu´lu`n. Hence we predict the FCI polarity behavior of na˘-CL nominal (and its similarity to the Greek opjosdhipote), as well as the fact that English wh-phrases are not FCIs: their world variable is independent and can be assigned the actual world as its value, as argued in Beck and Rullmann. But there is one remaining detail: na˘-CL also has a regular interrogative meaning, and in this case the na˘-set or individual can indeed be a set or an individual in the actual world. This means that, in order to derive the interrogative meaning of na˘-CL, we need to admit that its w variable can also be independent. But then, how is it different from which in terms of polarity? The key lies in intensionalization: our crucial premise is that the intensionalization that derives free choice PI-status is done by wu´lu`n. With FCI na˘-CL, wu´lu`n, we must argue, is always present, covertly or overtly, and restricts the w variable to dependent values only. In the interrogative use, there is no wu´lu`n, thus no restriction on the values, just like with which. To sum up, what we saw in this section was that our analysis, which postulates free choice intensionalization together with a core split between definite and indefinite FCIs in Greek, affords an accurate description of the distribution and interpretation of the novel FCI ou as a maximality operator na˘-CL ‘which’ in Mandarin. We analysed d (iota), and illustrated that depending on whether or not it appears with na˘-CL, the item is interpreted as definite or indefinite. We also argued for an intensionalization process in the Mandarin FCI, akin to the function of -dhipote in Greek, and we held this intensionalization accountable for the polarity status of na˘-CL.



acknowledge, next to the class of indefinite FCIs, a class of FCIs that behave like definites in denoting maximal plural entities. We showed that the contrast is syntactic, i.e. it correlates with whether or not the FCI contains an expression that contributes iota. Typically, closure under iota is a given in an FC free relative, hence FC free relatives present the standard case of what we take to be definite FCIs. In Mandarin, on the other hand, d ou modifies FCIs both with nominal or clausal complements, and in both cases it contributes maximality when it occurs. Alternatively, any, and the Greek opjosdhipote taking a nominal argument, exhibit the case of indefinite FCIs familiar from the earlier literature (Giannakidou 2001, Horn 2001, 2005). The optionality of d ou in Mandarin was shown to reflect exactly the definite versus indefinite contrast we identified for any and Greek FCIs, and likewise, we suggested, the same distinction is expressed lexically in English between FCI nominal any versus wh-ever. In this context, wh-phrases turn out to be good candidates for FCIs because they provide predicates upon which certain operations apply: intensionalization, and additionally either maximalization, or embedding under an indefinite determiner, as we proposed here for FCI nominals. We have given an explicit syntax-semantics for both definite and indefinite FCIs and illustrated that in both cases FCIs contribute a world variable that is dependent—either because it cannot remain free (indefinite FCIs), or because it remains bound by the k-operator under iota (definite FCIs). This dependency is consistent with the view of PIs as lexically deficient expressions advocated in Giannakidou (1998, 2001) and constitutes an attempt to capture formally this deficiency. Crucially, it is the presence of a dependent variable that renders a wh-phrase a polarity sensitive FCI and restricts its distribution in non-episodic contexts; our account thus captures easily the polarity behaviour of Greek and Chinese FCIs. At the same time, we also acknowledged the fact that clausal structure, as in the case of FC relatives and wh-ever, creates FCIs that are not subject to licensing, and derived their freer distribution by appealing directly to the meaning of the free relative. This enabled a simpler analysis of subtrigging, i.e. the occurrence of FC free relatives in episodic contexts, by using directly the meaning of the free relative. Our initial observation was that the morphological make-up of wh-words that function as FCIs in Mandarin and Greek indicate that interrogative semantics alone cannot serve as the basis of free choice. Our findings, then, question Kratzer and Shimoyama (2002) which attempt to do precisely this. In the context of the facts


Acknowledgements We would like to thank the anonymous reviewers of the Journal of Semantics for their very helpful insights and suggestions which led to considerable improvements in terms of both content and presentation. We are also grateful to the editor, Bart Geurts, for his observations and comments at every stage, which also proved instrumental to improving the paper. For discussions and additional suggestions, we would also like to thank Larry Horn, Chris Tancredi, Jason Merchant, Georges Tsoulas, and Jerry Sadock, Eri Vlachou and Jim Huang; Boya Li, Rint Sybesma, Dylan Tsai and Yang Shen for discussions on Chinese as well as help with Chinese data, as well as KookHee Gill and Eun-Hae Park for discussions on Korean that unfortunately didn’t make it into the paper. Finally, we would like to thank the audience of the Workshop on Quantification at the University of York, UK (July 2004), where a preliminary form of this paper was presented.


that motivated our analysis, it is also hard to see how interrogative semantics alone can predict the polarity behavior of FCIs, or the observed difference between polarity and non-polarity FCIs. In fact, the non-polarity cases of FCIs that we discussed, which would fall indeed in the class described by Kratzer and Shimoyama, were shown to be free relatives, hence substantially different from interrogative structures. We have not claimed to have provided answers to all questions that arise with free choice. In fact, we left some important questions open: for instance, the very restricted distribution of the definite na˘-CL NP in modal contexts, and the contrast in this respect with its Greek definite counterpart—the FC free relative—which exhibits the freer distribution of wh-ever, and can occur even in episodic sentences via subtrigging. Obviously, this contrast must be taken to suggest that definiteness in itself is neither a necessary nor a sufficient condition for a freer distribution of this class of FCIs. Apparently, the fact that we have free relative structures in the cases of Greek and wh-ever in English but not in Chinese plays a role, but we will not speculate further predictions beyond exactly this at the present stage. We also have not discussed how maximality and exhaustivity are to be linked in the analysis of d ou, or how our analysis of d ou in Chinese can be extended to capture the contribution of related additive particles in other languages with wh-indeterminates, e.g. Japanese (dare) mo, Korean to and na (but see Park 2005 for a recent analysis of na as a definitenes marker in the spirit of our analysis here). Our hope has been to offer a plausible and flexible enough framework where such questions can indeed be addressed, and in future research, answered in a satisfactory way.

Anastasia Giannakidou and Lisa Lai-Shen Cheng 181 ANASTASIA GIANNAKIDOU Department of Linguistics University of Chicago 1010 E. 59th Street Chicago, IL 60637, USA [email protected]

First version received: Second version received: Accepted: Advance Access publication:

14.04.05 20.10.05 13.12.05 04.04.06

REFERENCES Alexiadou, A. & Giannakidou, A. (1998) ‘Specificational pseudoclefts as lists’. In K. Shahin et al. (eds), Proceedings of WCCFL 17: 1–15. Aloni, M. (2003) ‘On choice offering imperatives’. In P. Dekker and R. van Rooy (eds), Proceedings of the Fourteenth Amsterdam Colloquium 14: 1–6. Beck, S. & Rullmann, H. (1999) ‘A flexible approach to exhausitivity in questions’. Natural Language Semantics 7.3: 249–298. Bolinger, D. (1977) Meaning and Form. Longman. London (Chapter 2, Any and some). Bosque, I. (1996) La polaridad modal. Unpublished MS. University of Madrid. Bresnan, J. & Grimshaw, J. (1978) ‘The syntax of free relatives in English’. Linguistic Inquiry 9: 331–391. Cheng, L. L.-S. (1991) On the typology of Wh-questions. Ph.D. dissertation, Department of Philosophy and Linguistics, MIT (published by Garland, New York, 1997). Cheng, L. L.-S. & Giannakidou, A. (2005) ‘The non-uniformity of whindeterminates with free choice in Chinese. To appear in G. Tsoulas and K.-H. Gill (eds), Strategies of Quantification. Selected papers from

the workshop on Quantification and Crosslinguistic Variation, University of York. Oxford University Press. Cooper, R. (1983) Quantification and Syntactic Theory. Reidel. Dordrecht. Davison, A. (1980) ‘Any as a universal or existential’. In J. van der Auwera (ed.) The Semantics of Determiners. Croom Helm. London., 11–34. Dayal, V. (1995) ‘Quanitification in correlatives’. In E. Bach et al. (eds.), Quantification in Natural Language. Kluwer. Dordrecht, 179–205. Dayal, V. (1997) ‘Free choice and -ever: identity and free choice readings’. Proceedings of Semantics and Linguistic Theory (SALT) VII: 99–116. Dayal, V. (1998) ‘Any as inherently modal’. Linguistics and Philosophy 21: 433–476. Dayal, V. (2004) ‘The universal force of free choice any’. In J. Rooryck (ed.) The Linguistic Variation Yearbook 4. John Benjamins. Amsterdam, 5–40. Dikken, M. den & Giannakidou, A. (2002) ‘From Hell to Polarity: ‘‘Aggressively non-D-linked Wh-phrases as polarity items’’ ’. Linguistic Inquiry 33: 31–61. Donnellan, K. (1966) ‘Reference and definite descriptions’. The Philosophical Review LXXV: 281–304.


LISA L.-S. CHENG Department of Linguistics/TW Leiden University P.O. Box 9515 2300 RA Leiden, The Netherlands [email protected]

182 (In)Definiteness, Polarity, and the Role of wh-morphology Heim, I. (1982) The semantics of definite and indefinite noun phrases. Unpublished Ph.D. dissertation, University of Massachusetts. Amherst, MA. Heim, I. (1992) ‘Presupposition projection and the semantics of attitude verbs’. Journal of Semantics 9: 183–221. Heim, I. & Kratzer, A. (1998) Semantics in Generative Grammar. Blackwell. Oxford. Horn, L. R. (1972) On the semantic properties of logical operators in English. Unpublished Ph.D. dissertation. University of California at Los Angeles. Horn, L. R. (1997) ‘All John’s Children are as Bald as the King of France: existential import and the geometry of opposition’. Chicago Linguistics Society 33: 155–179. Horn, L. R. (2000a) ‘Any and ever: Free choice and free relatives’. Proceedings of Israeli Association for Theoretical Linguistics 15: 71–111. Horn, L. R. (2000b) ‘Pick a theory: not just any theory’. In L. Horn & Y. Kato (eds), Negation and Polarity: Syntactic and Semantic Perspectives. Oxford University Press. Oxford, 147–192. Horn, L. R. (2005) ‘Airport ’68 Revisited: Toward a unified indefinite any’. In G. Carlson & F. J. Pelletier (eds), The Partee Effect. CSLI. Stanford, 179–205. Iatridou, S. & Varlokosta, S. (1998) Pseudoclefts crosslinguistically. Natural Language Semantics 6.1: 3–28. Jacobson, P. (1995), ‘The quantificational force of English free relatives’. In Bach et al. (eds) Quantification in Natural Language. Kluwer. Dordrecht, 451–486. Jayez, J. & Tovena, L. (2005) ‘Free choiceness and non-individuation’. Linguistics and Philosophy 28: 1–71. Kadmon, N. & Landman, F. (1993) ‘Any’. Linguistics and Philosophy 16: 353–422. Kamp, H. (1981) ‘A theory of truth and semantic interpretation’. In


Enc, M. (1991) ‘The semantics of specificity’. Linguistic Inquiry 22: 1–25. Farkas, D. (2002) ‘Varieties of indefinites’. Proceedings of Semantics and Linguistic Theory (SALT) 12: 59–83. von Fintel, K. (2000) ‘Whatever’. Proceedings of SALT 10: 27–40. Giannakidou, A. (1997) ‘Linking sensitivity to limited distribution: the case of free choice’. In P. Dekker et al. (eds), Proceedings of the 11th Amsterdam Colloquium. ILLC. University of Amsterdam, 139–145. Giannakidou, A. (1998) Polarity Sensitivity as (Non)veridical Dependency. John Benjamins. Amsterdam. Giannakidou, A. (1999) ‘Affective dependencies’. Linguistics and Philosophy 22: 367–421. Giannakidou, A. (2001) ‘The meaning of free choice’. Linguistics and Philosophy 24: 659–735. Giannakidou, A. (2002) ‘Licensing and Sensitivity in Polarity Items: from Downward Entailment to (Non)veridicality’. In M. Andronis, A. Pycha & K. Yoshimura (eds), CLS 38: Papers from the 38th Annual Meeting of the Chicago Linguistic Society, The Panels, 29–53. Giannakidou, A. (2006) ‘Only, emotive factive verbs, and the dual nature of polarity dependency’. To appear in Language 82. Gill, K.-H., Harlow, S. & Tsoulas, G. (2002) Disjunction, quantification, and free choice. Unpublished MS. University of York. Groos, A., and van Riemsdijk, H. (1981) ‘Matching effects in free relatives: a parameter in core grammar’. Proceedings of the Pisa Colloquium on Markedness, Annali dela Scuola Normale Superiora, Pisa, 171–216. Grosu, A. (1996) ‘The proper analysis of ‘‘Missing P’’ free relative constructions’. Linguistic Inquiry 27: 257–293. Haspelmath, M. (1997) Indefinite Pronouns. Oxford University Press. Oxford.

Anastasia Giannakidou and Lisa Lai-Shen Cheng 183 Nishigauchi, T. (1986) Quantification in syntax. Ph.D. dissertation. University of Massachusetts, Amherst, MA. Quer, J. (1998) Mood at the Interface. Unpublished Ph.D. dissertation, University of Utrecht. Quer, J. (1999) The quantificational force of free choice items. Unpublished MS. University of Amsterdam. Quine, W. van O. (1960) Word and Object. MIT Press. Cambridge, MA. Park, E.-H. (2005) Wh-indeterminates and free choice in Korean. Unpublished MS. University of Chicago. Rullmann, H. (1996) Two types of negative polarity items. Proceedings of Northeastern Linguistics Society (NELS) 26: 335–350. Rullmann, H. & Beck, S. (1998) ‘Presupposition projection and the interpretation of which-questions’. In D. Stgrolovitch and A. Lawson (eds), SALT VIII: 215–232. Sæbø, K.-J. (2001) ‘The semantics of Scandinavian free choice items’. Linguistic and Philolosophy 24: 737–787. Tovena, L. & Jayez, J. (1997) The modal arbitrariness of any. Unpublished MS. University of Geneva and EHESS, Paris. Tancredi, C. & Yamashina, M. (2004) Interpretation of indefinites in the Japanese wh-mo Construction. Paper presented in the Workshop on Quantification, York, July 2004. Tredinnick, V. (1994) ‘On the interpretation and distribution of -ever in English free relatives’. In R. Eckardt & V. van Geenhoven (eds), Proceedings of CONSOLE 2: 253–268. Vendler, Z. (1962) ‘Each and every, any and all’. Mind 71: 145–160. Vlachou, E. (in preparation) Varieties of free choice. Ph.D. thesis, University of Utrecht. Zimmerman, T. E. (2000) ‘Free choice disjunction and epistemic possibility’. Natural Language Semantics 8: 255–290.


J. Groenendijk et al. (eds), Formal Methods in the Study of Language: Proceedings of the Third Amsterdam Colloquium, Part I. Mathematical Centre Tracts. Amsterdam. Kratzer, A. & Shimoyama, J. (2002) Indeterminate pronouns, the view from Japanese. Unpublished MS. University of Massachusetts, Amherst, MA. Krifka, M. (1995) ‘The semantics and pragmatics of polarity items in assertion’. Linguistic Analysis 15: 209–257. Krifka, M., Pelletier, F. J., Carlson, G., ter Meulen, A., Chierchia, G. & Link, G. (1995), ‘Genericity: an introduction’. In G. Carlson & F. J. Pelletier (eds), The Generic Book. University of Chicago Press. Chicago, 1–124. Lahiri, U. (1998) ‘Focus and negative polarity in Hindi’. Natural Language Semantics 6: 57–23. Larson, R. (1999) Free relative clauses and missing P’s. Reply to Grosu. Unpublished MS. Lee, C. (1997) ‘Negative polarity and free choice: where do they come from?’ In P. Dekker et al. (eds), Proceedings of the 11th Amsterdam Colloquium. ILLC, University of Amsterdam, 217–222. Lee, Y.-S. & Horn, L. R. (1994) Any as an indefinite plus even. Unpublished MS. Yale University. LeGrand, J. E. (1975) Or and Any: The semantics and syntax of two logical operators. Unpublished Ph.D. dissertation. University of Chicago. Li, Y.-H. A. (1992) ‘Indefinite Wh in Mandarin Chinese’. Journal of East Asian Linguistics 1.2: 125–155. Lin, Jo-Wang (1996), Polarity Licensing and Wh-phrase Quantification in Chinese. Unpublished Ph.D. dissertation. University of Massachusetts, Amherst, MA. Lin, J.-W. (1998) ‘On existential polarity wh-phrases in Chinese’. Journal of East Asian Linguistics 7: 219–255.

Journal of Semantics 23: 185–216 doi:10.1093/jos/ffh036 Advance Access publication February 10, 2006

The Meaning and Interpretations of the Japanese Aspect Marker -te-iATSUKO NISHIYAMA University at Buffalo, the State University of New York

The Japanese marker -te-i- can have progressive, resultative, and existential perfect readings and has often been regarded as ambiguous. This paper shows that there is no clear evidence that -te-i- is ambiguous. It proposes a monosemous analysis of -te-i- that unifies its multiple readings and shows how progressives and perfects can form a natural semantic class. Within the context of a Discourse Representation Theory (Kamp and Reyle 1993, de Swart 1998), I propose that -te-i- consists of an imperfective operator -te- and a stativizer -i-. The imperfective operator -te- takes an eventuality as its argument and outputs a (non-necessarily proper) subpart of the eventuality, which precedes a reference time interval. Secondly, a stativizer -i- maps the subpart of the eventuality, i.e. -te-’s output, onto a state which overlaps with reference time and whose category is semantically underspecified and is determined via pragmatic inferences. The vague output of the imperfective operator, i.e. whether it is a proper subpart or nonproper subpart of an eventuality, leads to the contrast between progressive readings and perfect readings of -te-i-.

1 INTRODUCTION Progressives and perfects differ in that the denoted event is incomplete and ongoing in one case, but typically complete and not ongoing in the other. Progressives are often associated with an imperfective marker which encodes the incompleteness of the described event, while perfects are often associated with a perfective marker which encodes the completeness of the event or some type of aspectual marker which can entail the termination of the event if not the completeness of it (Smith 1997). Many languages such as English, French, and Chinese have distinct aspectual markers or verbal forms for progressives and perfects. It is therefore puzzling that the Japanese aspect marker -te-i- can have either progressive or perfect interpretations as seen in (1) - (4). (1) Ken-ga hashi- -tte-i- -ru. -TE-I- NONPAST Ken-NOM run The Author 2006. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]


Abstract

186 The meaning and interpretations of the Japanese aspect marker -te-ia. ‘Ken is running.’ b. ‘Ken has (already) run.’ (2) Ki-ga taore- -te-i- -ru. Ki-NOM fall -TE-I- NONPAST ‘A tree has fallen down (and it is lying on the ground).’ (3) Yoko-wa kono natsu san kai umi ni i- -tte-i- -ru. Yoko-TOP this summer three times sea LOC go -TE-I- NONPAST ‘Yoko has been to the sea three times this summer.’

One interpretation for -te-i- in (1) corresponds to a present progressive in English, as seen in (1a). (2) corresponds to a present resultative perfect, that is, ‘the direct result of a past event still continues (‘‘stative perfect’’)’ (McCawley 1971:104). (1b) or (3) corresponds to a present existential perfect which expresses ‘the existence of past events’ (McCawley 1971:104).1 (4) corresponds to a present progressive perfect or a continuative perfect. In addition to progressive and perfect readings, habitual readings are also possible with -te-i-, as seen in (5). (5) Ken-ga (mainichi) hashi- -tte-i- -ru. Ken-NOM (every day) run -TE-I- NONPAST ‘Ken runs (everyday).’ It should be noted that Japanese tense is either marked by the past tense -ta or the non-past tense -ru. Therefore, each distinct reading of -te-imarked by a non-past tense -ru in (1)–(5) can have corresponding future readings, depending on contexts.2 The Japanese data raise questions: Why and how can those seemingly different interpretations, especially progressive and perfect interpretations, be expressed by the same marker -te-i-? To answer these questions, this paper gives a compositional and monosemous analysis of -te-i-, assuming Krifka’s (1998) event structure along with others (Koenig and Muansuwan 2000; Egg 2005) and modifying Kamp and Reyle (1993), van Eijck and Kamp (1997), and de Swart’s (1998) approach to aspect in Discourse Representation Theory. It also shows 1 The interpretation in (1b) or (3) is commonly called keiken/kiroku (experience/record) (Fujii 1966) or ‘experiential perfect’. I call it an existential perfect in this paper. 2 Future readings of -te-i-non-past are ignored in this paper unless relevant so as to focus on the distinct readings of -te-i-. See Ogihara (1996) for more detailed discussion of the Japanese tense system.


(4) Ken-ga ichi-jikan-mae-kara hashi- -tte-i- -ru. Ken-NOM one-hour-ago-since run -TE-I- NONPAST ‘Ken has been running since one hour before.’

Atsuko Nishiyama 187

2 PROBLEMS WITH AMBIGUITY APPROACHES TO -te-iThis section first discusses problems that plague approaches treating -te-i- as ambiguous. Despite extensive discussion of the behavior of the different interpretations of -te-i- in past research, much less work has been done on the semantics of -te-i- per se. According to Kindaichi’s (1950) four-way verbal classification, -te-i- has a progressive interpretation when it is attached to durative verbs (keizoku doshi), e.g. hashi(‘run’) or kak- (‘draw’) in en-o kak- (‘draw a circle’), which correspond to activity and accomplishment verbs in Vendler’s classification, respectively (Vendler 1957). In contrast, it has a resultative reading when it is attached to punctual verbs (shunkan doshi),3 e.g. shin- (‘die’), which correspond to Vender’s achievement verbs (Kindaichi 1950).4 Following Kindaichi’s classification, uses of -te-i- (-te-i-2 in Table 1) in which it receives an existential perfect interpretation (keiken/kiroku (‘experience/ record’) in traditional terminology) have been considered to exemplify 3 The two verbal categories go back to Matsushita (1928: 409–11). Shunkan doshi is also translated as ‘instantaneous verbs,’ and keizoku doshi as ‘continuative verbs’ ( Jacobsen 1991; Ogihara 1998). 4 There are two other Japanese verbal categories in Kindaichi’s classification.

i. Stative verbs: verbs which have stative meanings without -te-i-, e.g. a- (‘exist’), i- (‘exist’), mie- (‘be visible’), etc. ii. The verbs of fourth category: verbs which always occur with -te-i-, e.g. sobie- (‘tower’), sugure- (‘become superior’), etc. Kindaichi (1950) did not give any mnemonic name to the last category. Ogihara (1998) regards it as a defective category of punctual verbs or instantaneous verbs since they do not usually occur without -te-i- in the main clause.


how those different interpretations can be derived from an unambiguous meaning of -te-i- built out of an imperfective marker -te- and a stativizer -i-. This paper, thus, provides a unified analysis of -te-iwhich covers an unusual cluster of aspectual interpretations. There are two consequences of this analysis. First, my monosemous analysis refutes the prevailing view among many Japanese linguists that -te-i- is ambiguous or polysemous. Secondly, this analysis challenges the view that resultative and existential perfects constitute different meanings (McCawley 1971; Michaelis 1998). I argue that the difference between these two readings of -te-i- is pragmatic in nature. This paper is organized as follows: Section 2 discusses problems with previous studies and shows that -te-i- is not semantically ambiguous but vague. Section 3 gives a unified analysis of the meaning of -te-i-. Section 4 shows how the multiple readings of -te-i- can be obtained from the single meaning of -te-i- and provides the discourse representation structures for the different interpretations of -te-i-. Section 5 concludes the paper.

188 The meaning and interpretations of the Japanese aspect marker -te-iTable 1 Traditional view: two entries of -te-i-. -te-i-1 Verb class

Durative Progressive Punctual Resultative perfect

-te-i-2 +ima (now) YES

+ima (now)

Existential perfect

NO

YES

(6) Genzai ano hito-wa tegami-o kai- -te-i- -ru. Now that person-TOP letter-ACC write -TE-I- NONPAST ‘That person is writing a letter/letters now.’ (7) a. Ano hito-wa takusan-no shousetsu-o kai- -te-i- -ru. That person-TOP many-GEN novels-ACC write -TE-I- NONPAST ‘That person has written a lot of novels.’ (Kindaichi 1955) takusan-no shousetsu-o kaib. () Genzai ano hito-wa () Now that person-TOP many-GEN novels-ACC write -te-i-ru. NONPAST -TE-I‘That person has written a lot of novels now.’ (Fujii 1966) 5

Some other studies only discuss the contrast between progressive readings with durative verbs and resultative perfect readings with punctual verbs ( Jacobsen 1991; McClure 1995). They do not discuss cases in which durative verbs (activity) can receive perfect readings with -te-i-. 6 The verbal classification in Table 1 follows Kindaichi (1950). More recent researchers have replaced the terms ‘durative verbs’ with hi-genkai dooshi (literally ‘non-limit-verb’) and ‘punctual verbs’ with genkai dooshi (‘limit-verb’) (see Okuda 1978; Kudo 1995; Kinsui 1995).


a distinct morpheme from the other two uses (-te-i-1), because -te-i-2 can receive an existential interpretation whether it occurs with durative verbs or punctual verbs (Kudo 1995; Igarashi and Gunji 1998; Ogihara 1998; Shirai 2000).5 The interpretations of -te-i-1 and -te-i-2 have been distinguished by the co-occurrence restrictions of present-time adverbials, as seen in Table 1.6 This co-occurrence restriction has been one major piece of evidence for the hypothesis that -te-i- is ambiguous between -te-i-1 and -te-i-2. According to the ambiguity hypothesis, if a sentence with -te-icannot co-occur with a present time adverb genzai or ima (‘now’) (see (7b)), it is considered to have an existential perfect reading (‘experiential perfect’ in Fujii (1966) or ‘action perfect’ in Kudo (1995)) (Fujii 1966; Kudo 1995; Ogihara 1998; Shirai 2000). Present progressive and resultative perfect interpretations, on the other hand, can co-occur with a present-time adverb genzai (‘now’) as shown in (6) and (8).


(8) Genzai ano hito-wa kekkon-shi- -te-i- -ru. Now that person-TOP marriage-do -TE-I- NONPAST ‘That person is married now.’ According to Fujii (1966), because one cannot add genzai to (7a), (7a) and (8) should be distinguished.7 Two difficulties plague this hypothesis. First, it is empirically wrong. In some contexts, (7a) can co-occur with genzai (‘now’). For example, imagine a situation where the speaker of (7b) is talking about someone’s achievements and continues (7b) by listing titles of novels as follows:

In such a context, (7b) is felicitous.8 Therefore, it does not show the ambiguity between resultative and existential perfect readings. The following example also shows that present-time adverbs are compatible with existential readings: (10) Genzai Now -te-i-TE-I‘He has

kare-wa san-kai taitoru-o bouei-shihe-TOP three-times title-ACC defend masu. NONPAST(POLITE). defended his (championship) title three times now.’

Secondly, even if the present time adverbs’ constraint were true, it is not clear why present-time adverbs would not be able to co-occur with existential perfect readings of -te-i-. Michaelis (1998) considers presenttime adverbs to be a test of stativity. She argues that the present-time 7 Following Matsushita (1928), Kindaichi (1955) considers (7a) a resultative perfect use and assumes that the verb kak- (‘write’) is ambiguous between a punctual and durative verb. However, there is no independent evidence to support the ambiguity of kak- (‘write’), as I discuss shortly. 8 Some may compare the use of genzai (‘now’) in (7b) with non-temporal use of now in English (‘the argumentative use’ in Nyan 1998 or its use as a ‘discourse marker’ in Schiffrin 1987) as seen in (i). In these uses, now metalinguistically marks transitions of topic or argument in discourse and often neutralizes in agreement with the temporal use of now. If genzai (‘now’) had the use, that would explain the felicity of (7b) or (10) within the ambiguity view of -te-i-. However, the Japanese presenttime adverb genzai does not have non-temporal uses, as seen in (ii):

i. Now where else did we go. Um . . . (Schiffrin 1987: 237) ii. f?? Genzai /Sateg, hokani doko-ni itta Now /Well else where go PAST ‘Now, where else did we go? (intended)’

kashira .. wonder

(ii) sounds odd with genzai (‘now’) following a list of the name of the participants’ favorite restaurants. The present-time adverb ima (‘now’) would also sound odd in the above context, although it has some, but not common, argumentative uses.


(9) Tatoeba, Kiseki, Ki-no-kuni, ... for example, miracle, Ki-GEN-country, . . . ‘For example, Kiseki, Ki-no-kuni (the titles of books), . . .’

190 The meaning and interpretations of the Japanese aspect marker -te-iadverb now is compatible with all present perfects including resultative and existential perfects in English because of their stativity. If Japanese resultative and existential perfects with -te-i- correspond to English resultative and existential perfects, it would be odd for present-time adverbs to distinguish present resultative perfects and existential perfects in Japanese, but not in English. Furthermore, not all states welcome present-time adverbs. The ‘now’ test rules out eternal states (generic states) (Ismail 2001), as (11) illustrates: (11)

Now, whales are fish.

(12) Kare-wa kyonen Kyoto-ni i- -tte-i- -ru. He-TOP last year Kyoto-to go -TE-I- NONPAST a. ‘%JHe’s been to Kyoto last year. (He is not in Kyoto now).’ b. ‘()He’s gone to Kyoto last year. (He is in Kyoto now.)’ (13) Fuirumu-wa senshuu genzo-ni dashi- -te-i- masu. Film-TOP last week development-to submit -TE-I- NONPAST ‘%JI’ve taken the film to the store for development last week.’ The symbol %J marks literal English translations of Japanese sentences which are unacceptable in English. Although the English translations of (12a) and (13) are ungrammatical, the corresponding Japanese sentences are not. The most natural reading of (13) is the resultative reading that the film is in the store. The last piece of evidence typically adduced in favor of the distinction between resultative and existential -te-i- is the fact that only existential -te-i- is compatible with location adverbs which denote the location of the event (Kudo 1995). An existential -te-i- can co-occur with an event-location adverb marked by the postposition -de as in (14a), while a resultative -te-i- cannot as shown in (14b) (Kudo 1995).


We may therefore at most conclude that the compatibility with presenttime adverbs indicates the stativity of a sentence, but it is not a necessary condition of stativity. Without clarifying what type of states presenttime adverbs can or cannot really co-occur with, one cannot use their co-occurrence properties to decide that progressive/resultative uses of -te-i- and existential use of -te-i- correspond to two distinct meanings. Another piece of evidence which has been adduced to argue for the existence of two meanings for -te-i- is the fact that only existential perfects can occur with past-time adverbial phrases in the non-past tense as shown in (12). However, this is not always the case, either. See (13).


(14)

a. Kanojo-wa suisu-no kyookai-de kekkon-shi- -te-i- ru. She-TOP Swiss church-at marriage-do -TE-I- NONPAST ‘She has gotten married at church in Switzerland.’ b.

Genzai kanojo-wa suisu-no kyookai-de kekkon-shichurch-at marriage-do Now she-TOP Swiss -te-i-ru. NONPAST -TE-I‘She is married at church in Switzerland now.’

(15) Genzai kanojo-wa suisu-de kekkon-shi- -te-i- -ru. Now she-TOP Switzerland-at marriage-do -TE-I- NONPAST ‘She has gotten married in Switzerland (and is still married there) now.’ When an adverbial marked by -de only describes the location (or institution) where the wedding ceremony is held, it is harder for -te-ito receive the entailed resultative perfect reading that a resultant state entailed by a main verb and its arguments holds at present, i.e. ‘she is married,’ because the information about the location of the ceremony is irrelevant to whether the state of her being married holds at present. On the other hand, (15) is fine, because suisu-de (‘in Switzerland’) can be interpreted as related to the entailed resultant state holding at present. In some contexts, though, (14a) can receive an entailed resultative reading too, for example, when people are wondering if she is really married. What (14a) cannot receive is the reading that the entailed resultant state is holding ‘at church,’ since kyokai (‘church’) is a location of the wedding ceremony, not of a state of being married in normal contexts. Thus far, I have shown that there are counterexamples to all three pieces of evidence typically adduced for the ambiguity approach. I


In (14a), suisu-no kyokai-de (‘at church in Switzerland’) denotes the location of the event of getting married, not the location of the resultant state of being married. According to the ambiguity hypothesis, if the sentence has a resultative reading, it should be able to co-occur with present-time adverbs. Since (14a) cannot co-occur with genzai (‘now’) as seen in (14b), it is said not to have a resultative reading. However, as I discussed, the incompatibility of present-time adverbials with a certain use of -te-i- cannot be used as evidence for the ambiguity of -te-i-. Secondly, adverbials marked by -de are fine with a resultative reading when it does not pick up the location of the wedding ceremony as in (15).

192 The meaning and interpretations of the Japanese aspect marker -te-iconclude that there is no clear evidence to support the distinction between two meanings of -te-i-, -te-i-1 and -te-i-2,9 and that the different readings of -te-i- may correspond to a single meaning. Before turning to the analysis of the meaning of -te-i-, one more observation must be noted. There is one clear difference between perfect uses of -te-i- and English perfects, that is, the perfect uses of -te-ican co-occur with both ‘definite past-time adverbs’ and ‘deictic pasttime adverbs’ in present (non-past) tense as seen in (16)–(18), while English existential or resultative present perfects cannot co-occur with any past-time adverb (Michaelis 1998: 164–166).10

(17) 1972-nen-ni Jon Lenon-ga kono hoteru-ni taizai-shi1972-year-in John Lennon-NOM this hotel-LOC stay-DO -te-i- -ru. -TE-I- NONPAST ‘%J John Lennon has stayed in this hotel in 1972.’ (‘This hotel is well-known for the fact that John Lennon has stayed.’) (18) Kare-wa kesa hashi- -tte-i- -ru. He-TOP this morning run -TE-I- NONPAST. ‘%JHe has run this morning.’ (‘He is still tired.’) As seen in (16)–(18) and in (13), all present perfect readings of -te-i-, including entailed resultative perfect readings, can co-occur with a past time adverbial phrase. The fact that -te-i- can co-occur with past time adverbials may suggest that this use of -te-i-+non-past is simply a past tense. However, it has often been noted that the present perfect can co-occur with past-time adverbials in many languages such as German (the German Perfekt) (Klein 2000; Lo¨bner 2002), Dutch (Boogaart 1999), and French (passe´ compose´) (Borillo et al. 2004). Therefore, the 9 Certain Japanese dialects in Uwajima and some western parts of Japan have two forms -yo-ru and -to-ru to express present progressive and perfect interpretations (see Kudo 1995; Kudo 2001 for detailed studies). They may be compared with the multiple interpretations of -te-i- (Shirai 1998). However, as Kinsui (1995) shows, the different meanings of those two forms do not correspond to the progressive/resultative or existential interpretations of -te-i- and therefore the existence of those two forms cannot be regarded as evidence for the ambiguity of the progressive/resultative or existential uses of -te-i-. For example, while -yo-ru only has a progressive interpretation, -to-ru has both progressive and perfect interpretations in many western parts of Japan (Inoue 2001; Kishie 2001; Murakami 2001). 10 There are some exceptions in English too (McCoard 1978).


(16) 1972-nen-ni /kyonen kanojo-wa kekkon-shi- -te-i- -ru. 1972-year-in /last year she-TOP marriage-do -TE-I- NONPAST ‘%JShe has got married in 1972/last year. (She is married.)’


(19) #Kinoo-wa takusan-no gakusei-ga kurasu-de shitsumon-shi Yesterday-TOP a lot of students-NOM class-LOC question-do te-i- ru. Watashi-wa minna-ni wakaru-youni shinsetsu-ni TE-I- NONPAST. I-TOP everybody understand-to kindly oshie te-i- ru. Tanaka-kun-ga kenkyuu-shitsu-ni ki teach TE-I- NONPAST. Tanaka-Mr.-NOM office-to come te-i- ru. TE-I- NONPAST. Yesterday a lot of students asked questions in class. I taught them kindly so that everyone could understand. Mr. Tanaka came to my office. (Intended) If (19) is a journal’s description of the day’s incidents and is intended to describe a sequence of events, -te-i-+non-past sounds odd and should be replaced by the simple past tense -ta. If -te-i-+nonpast tense was a past tense when co-occurring with past-time adverbials, (19) should be able to express a sequence of events (Partee 1984b).11 11 There is another criterion to test whether a present perfect form has a narrative use: If the present perfect form can occur in a when-clause that modifies a past event in a main clause, the form has a narrative use (Boogaart 1999). However, because the present tense interpretation is the one that is incompatible with the function of when-clauses as Boogaart (1999) argues, while the criterion is valid to test the narrative uses of languages which have absolute tense systems, it cannot be used in languages which allow relative tenses in subordinate clauses such as Japanese. That is, even though the perfect use of -te-i-non-past can occur in subordinate when-clauses modifying a past event in the main clause, it does not mean that -te-i-ru has a narrative use in main clauses, because the former use is due to the relative tense interpretation of the non-past tense in Japanese. See Ogihara (1996) for more details on Japanese relative tenses and also Comrie (1985) for absolute and relative tense systems.


compatibility with past time adverbials does not mean that -te-i-+nonpast is a past tense, rather than a present perfect. Another observation which has been made to distinguish a past tense from a present perfect is that in some languages such as English and Dutch a simple past tense occurs in a narrative context and can be used to describe a sequence of events, i.e. a temporally ordered set of events, while a present perfect resists a narrative context and does not easily describe such events (Partee 1984b; Boogaart 1999). Briefly, in a narrative context, after the first sentence of a narrative sequence establishes the initial temporal set up, the temporal information of each subsequent sentence can be interpreted relative to that of the preceding sentence independently of speech time (‘bracketing of the speech point’ in Sandstro¨m 1993. See also Nakhimovsky 1988; Boogaart 1999). Now, the Japanese -te-i-+non-past is not appropriate to describe a sequence of events in narrative contexts as seen in (19).

194 The meaning and interpretations of the Japanese aspect marker -te-iThe oddity of the present perfect use of -te-i- to describe a sequence of events as seen in (19) can be explained if present perfects express states introduced by the perfect that hold at present (e.g. at the time of writing a journal). States holding at present are irrelevant to the description of a sequence of events. Moreover, such states do not move reference time forward and the states introduced in (19) all share the same reference time (Partee 1984b). Hence, the inability of the sequence of perfect sentences to describe a series of events in (19).12 3 A UNIFIED ANALYSIS OF -te-i-

3.1 The function of -te-i-: Preliminaries I begin with a rough sketch of my analysis. Given the discussion in the previous section, the single meaning of -te-i- needs to provide two types of outputs, an incomplete event for progressive (and progressive perfect) readings and a complete event for perfect readings. This is achieved by making use of a part-whole relation between events (Krifka 1989; Krifka 1998). I propose that -te-i- consists of two morphemes: the imperfective operator -te- and the stativizer -i-. The imperfective operator -te- maps a class of eventualities which satisfy an eventuality description / onto another class of eventualities /#, those eventualities that are (not necessarily proper) subparts of the eventualities which satisfy / (i.e. e# < e when /(e) and /#(e#). < is a subpart relation.). The stativizer -i- maps /# onto a stative description /$, which is related to /# and whose temporal trace overlaps with a reference time interval (i.e. /$(s), and the state s overlaps a reference time). An eventuality description is a predicate together with its arguments (hereafter, a sentence radical), i.e. a sentence without tense, aspect, and other inflectional components (Herweg 1991a; Herweg 1991b; de Swart 1998; de Swart and Verkuyl 1999). 12 This explanation for the inability of the present perfect to describe a sequence of events may in turn favor the view that the French passe´ compose´ and the German Perfekt have a past tense meaning as one of their meanings since they have narrative uses and that this past tense meaning is responsible for their narrative uses (Waugh 1987; Lo¨bner 2002). However, the semantics of the perfect forms in French and German is beyond the scope of this paper.


This section presents a monosemous analysis of the meaning of -te-i-. Section 3.1 gives a brief sketch of my proposal. Section 3.2 describes the meaning of -te- as an imperfective morpheme with a MAX operator, assuming Portner’s (1998) notion of inertia worlds, and section 3.3 analyses the meaning of -i- as a stativizer.


Take (20): (20) Ken-ga ie-o tate- -te-i- ru Ken-NOM house-ACC build -TE-I- NONPAST ‘Ken is building a house.’/‘Ken has built a house.’ (21) Ken-ga ie-o tate- (Ken-NOM house-ACC build-)

(22) Benkyoo-shi- -te- -wa -i- -ru. Study-DO- TE- TOP -I- NONPAST ‘I am studying.’ Furthermore, the eventuality denoted by the sentence radical which is followed by -te- and other aspectual verbs can be interpreted either as 13

Atelic and telic distinction does not coincide with the distinction between events and states. Atelic events, i.e. activities, typically require a minimal subinterval constraint on their subinterval property while states do not (Dowty 1979). 14 Te- is often said to derive historically from the perfective marker tsu. However, most linguists agree that this hypothesis cannot explain modern uses of -te- (Hasegawa 1995; Igarashi and Gunji 1998; Ogihara 1998).


(21) is the sentence radical for (20). Aspectual classes (situation types) such as accomplishments, achievements, activities and states are assumed to be associated with an eventuality description, i.e. a sentence radical in this paper. (A)telicity is a property of eventuality descriptions characterized in sentence radicals. While atelic eventuality descriptions, activities and states, do not encode any endpoint; telic descriptions, accomplishments and achievements, encode a natural or intended endpoint (Depraetere 1995).13 I restrict the use of those terms at the level of the eventuality description unless specified. Because -te-’s output does not always have to be a proper subpart of the input eventuality, there are two types of possible outputs of -te-, which lead to the contrast between progressive and perfect readings of -te-i- ’s final output. First, when the subpart of the described eventuality is proper and therefore does not include the final part of the described event, the event of Ken building a house is not completed and (20) receives a progressive reading. Secondly, when the subpart of the event is non-proper and equivalent to the whole event, Ken building a house is completed and (20) receives perfect readings. Analysing -te- as an imperfective marker is not consistent with the widely accepted view that -te- historically derives from a perfective marker14 and it might not be obvious why -te-i- should be analysed as two separate morphemes rather than one morpheme -te-i-. One evidence for separating -te- from -i- is that a topic marker -wa can be inserted between -te- and -i- (Ishikawa 1985; Hasegawa 1995), as seen in (22).

196 The meaning and interpretations of the Japanese aspect marker -te-icomplete or incomplete depending on the context, as seen in (23) and (24). This is easily explained if we analyse -te- as an imperfective operator. (23) I-kkai-bun-no kusuri-o juusu-ni tokashi- -te-mi- -ta. 1-CL-dose-GEN medicine-ACC juice-LOC melt- -TE-see PAST. ‘(I) tried melting one dose of the medicine in some juice. (But it didn’t melt completely./And it melted completely.)’

(25) Terebi-ga koware- te-ki-ta. TV set-NOM break TE-come PAST. ‘The TV set is halfway broken.’ (Something is wrong with it but not completely broken.) Finally, -te- can also be followed by other types of verbs such as -temora/kure (‘TE-receive’) and -te-age/-ya (‘TE-give’) (‘benefactive verbs’) or -te-hoshi- (‘TE-want’). They do not seem to have vague readings at 15 The sequence of morphemes -te-a- (-de-a- in (24)) is often claimed to constitute a resultative construction because of its non-aspectual function, e.g. its detransitivizing function (valency reduction) or the selectional restriction it imposes on its argument, etc. (Hasegawa 1995). There are also uses of -te-a- which preserve the valency of a main verb (Soga 1983). 16 All V1-te-V2 forms can be followed by -te-i- except for the ones whose outputs are stative, e.g. V1-te-a- (‘exist’) and V1-te-hoshi- (‘want’). I do not discuss the semantics or syntax of those aspectual verbs or other verbs appearing as V2 in the V1-te-V2 in this paper. See Matsumoto (1990); Hasegawa (1995); Nightingale (1999) for more detailed discussion.


(24) Reizooko-no gyuunyuu-ga non-de-a-ru. Refrigerator-GEN milk-NOM drink- -TE-exist NONPAST ‘The milk in the refrigerator has been drunken. (Some of it is left./It’s gone.)’ The events of drinking the milk in the refrigerator and melting one dose of the medicine do not have to be interpreted as complete.15 The vagueness of sentences (23) and (24) shows that -te- can function as an imperfective operator when followed by other aspectual verbs as well. Other aspectual verbs which can follow a main verb+-te- include shima(‘complete, end, put away’) (-te-shima- ‘finish -ing’), ok- (‘leave, put’) (-te-ok- ‘leave V’), ik-/k- (‘go/come’) (-te-ik-/-te-ku- ‘is going/getting to’).16 The aspectual interpretation of -te- ’s output may not be vague in some of those aspectual verbs because their meanings may require an exclusively incomplete or exclusively complete interpretation of -te’s output. For example, -te-shima- (‘finish-ing’) expresses the completion of the event described in the main verb (V1) and its arguments, and therefore, V1+-te-shima does not show any vagueness in terms of the aspectual interpretation of V1. V1+-te-ku- (‘come’) also blocks the completion reading when V1 is a change of state verb as seen in (25).


first glance, but (26) and (27) show that -te-’s output event does not have to be complete. Here, the event of buying a house cannot be complete until Mr. Tanaka’s daughter borrows money from a bank. (27) shows another example where the eventuality followed by -te+ a verb of giving is not interpreted as complete.

(27) Kare-ga kinoo-no shukudai-o ya- -tte-kure- -ta kedo, He-NOM yesterday-GEN homework-ACC do TE-give PAST but zenbu-wa ya- -te-i- nai. all-TOP do TE-I -NEG ‘He did (some of ) yesterday’s homework for me, but he hasn’t done it all. (I will have to do the rest.)’ Thus, many of -te- + other verbs show some aspectual vagueness of the eventuality described by the verb + -te and its arguments, this providing ample motivation to separate -te- in -te-i- and analyse it as an imperfective operator whose output is vague.

3.2 The function of -teMy preliminary proposals for the meanings of -te- and -i- do not constrain their input eventuality descriptions. They can take either atelic or telic eventuality descriptions as input. (28) Ken-ga hashiKen-NOM run ‘Ken is running.’

-tte-i- -ru. -TE-I- NONPAST

(29) Fuji-san-ga mie-te-iMt.Fuji-NOM be-visible -TE-I‘Mt. Fuji is being visible.’

-ru. NONPAST

(1), repeated as (28), and (29) show that -te-i- can modify atelic eventualities, e.g. activities or states, eventualities without any explicit endpoint. However, when -te-i- takes a state as its input, there is a presupposition that the state is episodic. The input state cannot be


(26) Tanaka-san-wa musume-san-ni ie-o ka- -tte-ageTanaka-Mr.-TOP daughter-Ms.-DAT house-ACC buy TE-give -ru sou-desu. NONPAST hear ‘I hear that Mr. Tanaka will buy his daughter a house.’ (Mr. Tanaka will pay only a down payment, say, one-fourth or fifth of the cost, and his daughter and her husband will pay the loan for the rest of the money next twenty years.)

198 The meaning and interpretations of the Japanese aspect marker -te-iinterpreted as a generic statement,17 or, said differently, the output of -te- cannot be a subpart of a generic state when the input eventuality description is a state.18 The following contrast shows that the described stative eventuality is understood as episodic when it occurs with -te-i-:

b. Sono jiinzu-wa oheso-ga mie-ru. navel-NOM visible NONPAST That jeans-TOP ‘The navel is visible with that pair of jeans.’ (A speaker is describing the pair of jeans at the store.) The contrast is subtle, but in (30a) the speaker has to see the exposed navel when uttering the sentence, i.e. there must be an actual occurrence of the state; a generic state that does not require somebody’s navel being actually visible is not enough. (30a) is impossible without the speaker seeing the navel. On the other hand, (30b) can be a description of that pair of jeans without anyone trying it on and does not require the occurrence of the state of the navel being visible as long as the pair of jeans have the property of showing the navel.19 This phenomenon is not unique to the progressive use of the Japanese -te-i-. English progressives also take stative input (Dowty 1979; Binnick 1991; Egg 2005), as seen in the following. (31) Mike is being silly. (31) does not mean that Mike is a silly person in general but that Mike is acting silly (Egg 2005). I propose to account for the contrast between (30a) and (30b) as follows. When -te-i- takes atelic eventuality description as input, 17 Generic statements such as habitual sentences or characterizing sentences are often considered to be stative (Krifka et al. 1995). I assume the view and avoid the detailed ontological discussion on generics for the purpose of this paper. 18 As described in Section 1, -te-i- can have habitual readings, which are considered to be generic and therefore unbounded. It may seem a counterexample against the observation that the input state of -te-i- cannot be unbounded. However, the input eventuality of -te-i- is not a state but an event when it has a habitual reading. Habitual interpretations with -te-i- are typically related to episodic event predicates such as smoke (Krifka et al. 1995). In addition, habitual readings of -te-i- assume the existence of actual occurrences of the event. 19 Similar observations have been made by Yamagata (1998).


(30) a. Sono jiinzu-wa oheso-ga mie-te-i- -ru. That jeans-TOP navel-NOM visible -TE-I- NONPAST ‘(Your) navel is visible with that pair of jeans.’ (A speaker is looking at an addressee’s navel.) ‘The navel is visible with that pair of jeans.’ (A speaker is describing a pair of jeans at a store.)


a bounding operator is applied and the observed episodicity of (30a) is the effect of that bounding operator. I propose that the semantics of -teincludes an operator to quantize its input eventuality if the input description is unbounded. Because the input state is bounded when followed by -te-, even though -te-i-’s final output state is unbounded, it cannot be interpreted as generic. The quantizing effect of -te- can be captured through the use of a maximality operator (MAX) (Koenig and Muansuwan 2000; Egg 2005), which is based on Krifka’s (1998) notion of telicity, or more precisely, his definition of the initial and final parts of events. The MAX operator is defined as follows:

In (32) ‘

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents