cls.giannakidou

Transcript

1 To appear in Chicago Linguistic Society 39, 2002 Licensing and Sensitivity in Polarity Items: from Downward Entailment to (Non)veridicality Anastasia Giannakidou University of Chicago 1. Introduction Polarity phenomena in language are pervasive and quite diverse. A quite familiar any . Any a PI because it exhibits limited distribution: it is polarity item (PI) is ungrammatical in positive sentences, but becomes fine with negation, in questions, : few with modal verbs, and in the scope of downward entailing quantifiers like (1) a * John saw any students. John didn’t see any students. b c Did John see any students? John may talk to any students. d Few professors invited any students. e For a given paradigm of PIs, hence also for any , two questions are central. the licensing First, what is the property that restricts distribution? question: Second, the sensitivity question: why is the distribution of the PI restricted by this property? In the traditional account, since Ladusaw 1980 and through the mid-90's, the research agenda focused almost exclusively on the first question. It was proposed that downward entailment (DE) is the property that licenses PIs: Ladusaw’s (1980) licensing condition (2) α is a trigger for negative polarity items in its scope iff α is downward (where entailing. is the expression that licenses α ) trigger This licensing condition talks about ‘negative polarity items’ (NPIs), relying on the contrast between negation and affirmation. But this is not exactly the central contrast we observe in (1), hence it is more accurate to talk about PIs, and employ the term ‘NPIs’ for those PIs that are indeed identifiable as being licensed only by negation and the like, as I proposed in Giannakidou 1998. Licensing conditions like (2) proved quite fruitful and inspired a number of significant contributions in the area (Hoeksema 1983, Zwarts 1986, 1993, van der shared enthusiasm has Wouden 1994, Dowty 1994, among many others). The been that we are finally in position to characterize semantically the class of PI- licensers, a major advance compared to the previous alternative which was postulating merely the syntactic feature affective (Klima 1964). Yet the advance

2 on the licensing front brought with it a persistent limitation: conditions like (2) stipulated global, composition external filters on the as that were structures semantic filtering : contained PIs. This can be seen clearly in Ladusaw’s (1986) (3) Semantic Filtering ( φ ) = ∧ Syn ( φ ) grammatical Sem ( φ ); def where is syntactic well-formedness, Sem is semantic-well-formedness Syn is semantic conjunction of syntactic and Grammaticality well-formedness. a Syntactic rules determine which structures will be grammatical and which not, and so do semantic rules. In the case of polarity, a syntactically well-formed structure will be filtered out semantically: (1a) with the PI any is subject to the condition in (2), and (2) is not satisfied in (1a). Why PIs are subject to such semantic rules remains a mystery: there is no obvious link between DE and the lexical semantic content of PIs, and it seems hard to establish one-- which is why essentially no attempts were made in this tradition. Filtering conditions of this kind are posited at the polarity module, and are in no way derivable from the meaning of the PIs. emergence of a massive amount of Recently, however, with the cross- linguistic work, the sensitivity question has surfaced as an equally important field of inquiry (see, for example, Israel 1996, Giannakidou 1998, 2001, Tovena 1998, and Lahiri 1998). We no longer want to treat PIs as items subject to mysterious constraints. Rather, we want to understand why PIs are subject to the kinds of constraints they are, and derive these from other more familiar operations that are motivated independently in the grammar. In this context, the goal is to derive the limited distribution of a given PI-paradigm from its lexical semantics, and dispense entirely with global filters and the polarity module. The discussion in this paper must be seen in this light. Polarity indeed raises the issue of semantic well-formedness, but of viewing semantic- instead well-formedness in terms of filtering, I will view it in terms of interpretability, as I did in Giannakidou 1998, 2001. I explain below what exactly I mean by this. 1.1 Polarity: interpretability, licensing, and presupposition failure PIs are sensitive expressions. An intuitive way of capturing sensitivity is to say that PIs have a semantic ‘deficit’. For a sentence containing a PI to be interpreted the deficit must be nullified somehow, and this is what drives the need to be licensed. Without a licenser the deficit surfaces, and the sentence containing the PI cannot be interpreted, i.e. it cannot be assigned a truth value in any model, and will therefore be ruled out as ungrammatical. The idea here is consistent with Heim and Kratzer’s 1998 view of uninterpretability as ungrammaticality, and the need to ill- keep a meaningful distinction between uninterpretability and pragmatic 2

3 formedness, e.g. presupposition failure. The latter and can be yields infelicity fixed in contexts satisfying the presupposition; but polarity failures are cases of any uninterpretability, and cannot be repaired by any context. Sentence (1a) with is ungrammatical, no matter how we manipulate the context. This fact is worth attempts to ‘pragmaticize’ polarity emphasizing, especially when we consider . (for more discussion see Giannakidou 2001) At this initial stage it will be useful to single out some borderline cases which indicate a kind of sensitivity distinct from that involved in polarity proper. Consider, e.g. the so-called positive polarity items (PPIs) like some student : (4) a John didn’t see some student. ∃ x [ student (x) ∧ ¬ saw (John, x)] b ∧ # ∃ x student (x) [ saw (John, x)] ¬ c some has always been that it must scope The descriptive generalization about PPI over negation, as in (4b), and never below it, as in (4c). Hence, the term ‘polarity’ is used here to indicate not limited distribution but limited scope; and in this sense contrasts with a ‘regular’ indefinite like a student which has some unrestricted scope, and can be interpreted in both ways in (4). To the extent that we treat limited scope as an instance of polarity, however, we must be careful to keep a clear distinction between limited scope and limited distribution: crucially, there is never are ungrammatical; no PI-licensing in limited scope. Limited scope items they are just fine, they must only be interpreted in a particular way. In a context blocking their required interpretation, e.g. in a context forcing the reading in (4c) some  but there is no licensing above, the choice to use would be infelicitous failure yielding ungrammaticality, as is the case with any in (1a). some is not uninterpetable, but partially interpretable In other words, is : it interpretable in the models compatible with its required scope. This means that for some , the requirement that it be interpreted with a particular scope must have the status of a presupposition. So, we may argue that the presupposition for some is that it can be interpreted only with wide scope (this is an oversimplification, but for our purposes it suffices; see Farkas 2002 and Szabolsci tendency for more details)  a hypothesis supported by its 2001 to receive specific readings and inverse scope in construals with universal quantifiers, and by the fact that it cannot be bound by Q-operators (Langacker 1991), unlike regular indefinites. If the scoping of some is constrained by the presupposition that it takes wide scope, then we predict that if some occurs with negation, it will only scope above it. The ‘failed’ reading, then, just doesn’t surface in the ordinary case. , however (and the other items we will discuss in this paper), In the case of any 3

4 there is no reading in the absence of a licenser: these items are not available some but plainly uninterpretable. partially interpretable like even- superlatives Consider in this connection items quantificational like and minimizers, as in the examples below: He didn’t say a word . (5) a (minimizer: He didn’t say any word) He said a word. b (6) a He didn’t solve even the easiest problem. (= He didn’t solve any problem) He solved the even the easiest problem. b of a In these cases too, ‘polarity’ seems to be associated with the availability reading, not with rendering an expression grammatical. The minimizer certain reading arises with negation in (5a) and does not surface in the absence of it in (5b). is not ungrammatical without negation, it just lacks the polarity Say a word any- interpretation. The same holds for the quantificational superlative in (6a): the any -reading is lost, equivalent reading arises with negation; in the absence of it, the some but the superlative is still fine. Hence, just like any, minimizers , but unlike and even superlatives exhibit only limited interpretation. In all cases, the need for a particular interpretation is arguably encoded as a presupposition. If it is not satisfied, as with minimizers and even superlatives without negation, the items are good, but without their presupposition-dependent interpretation. for reasons of brief but the The discussion will have to remain space, bottomline is obvious: although we may relax the term polarity to include both distribution and limited interpretation phenomena, it is a mistake to limited collapse the two. In particular, it is a mistake to try to explain limited distribution in PIs like any by using tools appropriate for limited interpretation. There is, as we noted, a crucial difference between limited distribution and limited a interpretation: later case, ‘licensing’ is just the case of imposing certain in interpretations, and excluding others; but with limited distribution PIs licensing enables the very occurrence of an expression. For such an expression, there is no clearly interpretation available without a hence the problem is licenser; lexical semantic, and cannot be merely reduced to a case of pragmatic failure. 1.2 Outline of the paper With the above background in mind, the discussion proceeds as follows. I first show that in order to establish empirically adequate licensing conditions we need to employ a notion broader than DE—nonveridicality. The general definition of what constitutes a PI will be the following: 4

5 (7) DEFINITION 1 (Giannakidou 2001) Polarity item α is a polarity item iff: A linguistic expression (i) The distribution of α is limited by sensitivity to some semantic of the context of appearance; property β and β is (non)veridicality, or a subproperty thereof: β ∈ {veridicality, (ii) nonveridicality, antiveridicality, modality, intensionality, extensio- nality, episodicity, downward entailingness}. This definition presents a general heuristic format from which various conditions can be derived, predicting of course distinct distributions, as is the case with PIs. Simplifying somewhat, the set of properties subsumed under (non)veridicality is presented as a closed set in (ii) in order to cover the PI-paradigms that have thus far been identified in the literature. Multiple sensitivities are also allowed in this system (e.g. with free choice items, as we shall see later on). Veridicality and nonveridicality are defined in terms of truth as in (8); see also Zwarts 1995. (For first mention of veridicality see Montague 1969.). (8)  (Non)veridicality for propositional operators DEFINITION 2 F is veridical iff Fp entails p: Fp i. A propositional operator → p ; otherwise F is nonveridical. → F is anti veridical iff Fp entails not p : Fp . p ii. A nonveridical operator A propositional operator is a proposition embedding function: a sentence modifier ; (type a sentence-level adverb, modal operators, tense, temporal/aspectual adverbs, connectives), or an expression taking a proposition as its first argument, e. g. a propositional attitude verb (type > ), or the question operator > ). F (type Fp is true, p is also true; if this does is veridical iff whenever not hold, F is nonveridical. A nonveridical F is antiveridical iff whenever Fp is true p is not true. Antiveridical operators are a proper subset of the nonveridical. Positive operators like past tense adverbials are veridical and exclude PIs. Modal verbs, intensional operators, and questions are nonveridical and license PIs, as we noted in (1). Antiveridical operators, finally, are negation and without ; these are the prototypical licensers of negative PIs. (9) Yesterday, Paul saw a snake. → Paul saw a snake. (10) a ? Did Paul see a snake -/ → Paul saw a snake. b -/ → Paul saw a snake. Paul may have seen a snake. It is not the case that P. left. a Paul didn’t leave. → (11) It is not the case that P. left. b without Paul leaving. → 5

6 Truth is not absolute but relative to at least times and individuals, hence the core definition must be relativized accordingly to account for the veridicality properties of temporal and aspectual operators, as we see in the discussion in section 2. affective PIs (APIs, Greek I will concentrate on three PI-paradigms: any, dhipote paradigm), and free choice items (FCIs; Greek - kanenan- series). Once we establish licensing conditions for all three based on nonveridicality, I show how we can actually get rid of these conditions by making them derivable from the lexical semantics of the items. The discussion will necessary be brief, given space for extensive details, see Giannakidou 1998, 2001. limitations  2 Why do we need nonveridicality? The answer is: because the previous semantic alternative, DE, does not predict the PIs. I illustrate correct distribution of eleven cases where DE wrongly below predicts PIs to be bad. Two further problematic cases, and negative factives, only will be discussed in section 3. Recall also the conceptual problem with DE: it is hard to see what the connection is between DE and the lexical semantics of PIs; and it is precisely for this reason that polarity filters have been quite popular. PIs will be As a starting point, DE predicts that in the scope of fine negation, DE quantifiers like few N , at most n N , no N , and the restriction of every : (12) a No students saw anything. John didn’t see anything. b Few children saw anything. c Every student who heard anything should report to the police. d But there are many non-DE environments where PIs are fine. Observations about the individual cases can be found scattered in the literature, but here I list them in full, to give a complete picture of the extent of the empirical problem: 2.1 Non-monotone quantifiers Quantifiers like exactly three students , neither student , nobody but John , and almost nobody any . are non-monotone but they nevertheless admit (13) a % Exactly three students saw anything. (Linebarger 1980) b Neither student saw anything. c Nobody but John saw anything. d Almost nobody saw anything. 2.2 Hardly/barely Hardly and barely allow for any , yet they do not exhibit the DE pattern as we see below (for more discussion of this point see Atlas 1996 and Horn 2001). 6

7 (14) John {hardly/barely} talked to anybody. a John barely studied linguistics → John barely studied syntax (15) -/ -/ J. hardly talked to his mother John hardly talked to anybody b → 2.3 Questions PIs are quite frequent in questions: (16) Heb je ook maar iets gezien? (Dutch) a have.2sg you anything seen ‘Did you see anything?’ Idhes { b }? (Greek) tipota/*otidhipote perf. saw. 2sg API/ FCI ‘Did you see anything?’ a kanenas (kanas), tipota, etc. (Greek APIs) (17) opjosdhipote, otidhipote, etc (Greek FCIs) b Two things are noteworthy here. First, questions are not monotone (especially if they denote the proposition corresponding to their true answer; see Groenendijk and Stokhof 1997). Ladusaw 1980 admits the problem of questions and appeals to the availability of a biased negative answer. Though this may be justifiable for minimizers like give a damn, it obviously cannot provide the basis for a general account of PIs in questions, since PIs like any and its Greek/Dutch counterparts above are perfectly neutral, and do not exhibit bias towards a negative answer. Second, presumably “strong” PIs, like the Dutch and the ook maar iets Greek tipota, expected to be licensed by antiadditive or antimorphic triggers only (Zwarts 1993, 1998, van der Wouden 1994), are grammatical in questions (and other non-DE structures, as we shall see below). The point is illustrated in (18) for : it is good with niemand ‘nobody’, but bad with weinig mensen ook maar iets ‘few’; likewise in (19) we observe exactly the same pattern for the Greek tipota : it requires negation (there are no negative quantifiers in Greek; Giannakidou 2000); a merely DE-quantifier like ‘few’ and to poli pende ‘at most five’ will not suffice: liji (18) a * Weinig mensen hebben ook maar iets gezien. (Dutch) few people have.3pl anything seen ‘Few people saw anything.’ b Niemand heeft ook maar iets gezien. ‘Nobody saw anything.’ 7

8 (19) a Liji anthropi idhan tipota. (Greek) * few people saw.3pl anything ‘Few people saw anything.’ To poli 5 anthropi idhan tipota . b * At most five people saw anything. Ta pedhia c tipota . idhan dhen The children didn’t see anything. and tipota are thus strong PIs; yet they are good in questions, as we Ook maar iets saw in (16). The observed pattern is common crosslinguistically, and provides one of the strongest obstacles in trying to make questions fit the DE-hierarchies. Nonveridicality, on the other hand, predicts PIs should be fine in that questions because questions are nonveridical. I will briefly sketch here an account. In a semantics which treats questions as the set of their true answers (Karttunen answers, 1977) the denotation of a polar question would contain two possible represented here as a disjunction: a. [[ Did you see John ]] (w) = λ p [p(w) ∧ (20) λ w’ saw (you, John, w’)] = p = b. Q = {you saw John ∨ you didn’t see John} Disjucntion is nonveridical: from p q it does not follow that p is true and that q is true-- unlike conjunction, which is veridical (for p p and q must q to be true, both Being nonveridical, disjunction be true; for more discussion see Zwarts 1995). allows for PIs, as illustrated in the example below, from Giannakidou 1997: I bike mesa kanenas i afisame to fos anameno. (21) Either somebody came in, or we left the light on. with on a par seen disjunctions; the The licensing of PIs in questions can be 1 good in questions. parallel can serve as the basis for explaining why PIs are of Additional PIs and questions are, of course, also relevant for properties determining what interpretations will further be possible or not, e.g. whether there is a bias for a negative answer or not, and where it comes from. But crucially, bias 1 In English, any is not licensed in disjunctions but it is in questions: * Either anybody came in is bad. This indicates that disjunctions and questions are not identical, which is of course consistent with what I am saying here. It supports furthermore the view that any is subject to a weaker dependency to nonveridicality, as I suggest in section 2.13. At any rate, the fact that questions and disjunctions pattern together in terms of PI-licensing in the general sense suffices to establish the parallelism and use disjunction as the basis for nonveridicality in questions. 8

9 alone cannot predict grammaticality of PIs in questions, as we noted, for not all PIs come with bias, certainly not . any 2.4 The future PIs appear in future sentences. The future is a modal context, and depending on the analysis one adopts for modals, it will turn out to be either non-monotone or upward monotone (Giannakidou 1995, Giannakidou and Zwarts to appear). PIs, let alone alleged strong ones, are not expected to be good there, but they are. The data here from Greek, Dutch and English; see also Vikner 1999 for Danish. tha kanena bukali krasi. a agorasi (22) O Janis John will buy a/any bottle of wine. b ook maar iets ontdekken. De kinderen zullen vertrekken zodra zij The children will leave as soon as they discover anything. PIs are fine in the scope of the future expression; but the We see here that definition of (non)veridicality we have thus far does not suffice to explain why this is so. What we want to do is relativize truth with respect to times: DEFINITION 3 (23)  (Non)veridicality for temporal/aspectual operators Let be a temporal/aspectual operator; t an instant or an interval. F must F veridical iff for Fp to be true at a time t i. p is be true at a , (contextually relevant) time t’ t . Otherwise Op is nonveridical . ii. A nonveridical operator F is antiveridical iff for Fp to be true at a time t , ¬ p t’ t . must be true at a (contextually relevant) time F t , then F is veridical iff for all (contextually iii. If is true of an interval t’ t, p relevant) true at t’. Otherwise, F is nonveridical. If for all is (contextually relevant) t’ t , ¬ p is true at t’ , then F is antiveridical. The future is nonveridical by this definition, but the past tense and adverbials are nonveridical, and PI-licensing follows. t’, t

10 2. 5 The habitual This is another non-monotone context (Krifka 1995). Giannakidou 1995, et al Giannakidou and Zwarts (to appear) observe that PIs are good in habituals: (26) dhiavaze opjodhipote vivlio me megali prosoxi. Sinithos usually read.3sg FC book with great attention ‘S/He usually read any book very carefully.’ a O Janis me idhopiouse molis evlepe kamia agelia . (27) warned.impf. saw.impf.3sg any ad as-soon-as John me ‘John used to call me as soon as he saw any ads. BAD: * At 8 pm, J. called me as soon as he saw any job ads. De kinderen vertrokken zodra zij ook maar iets ontdekten. b the children left.3pl as soon as they anything discovered.3spl ‘*The children left as soon as they discovered anything.’ OK, as: ‘The children used to leave as soon as they saw anything.’ The habitual is nonveridical by definition 3. t’ [[ HAB p ]] ]. = 1 iff MOST t’ [ t’ ∈ C ∧ t’ t , [[ p]] = 1 at (28) t In other words, it is not the case that is true at all the contextually relevant t’ t . p always. The definition actually predicts PIs to be bad with a Q-adverb meaning In Giannakidou 1995 it is shown that this prediction is borne out. *Otan pijene o Pavlos ja ipno, ksefilize panda (29) . kanena periodhiko when went.3sg the P. for sleep, browsed.3sg always any magazine ‘When Paul went to bed, he always browsed through a magazine.’ The occurrence of PIs in habitual domains is considerably more complex, of course, but even this basic picture suffices to make the point clear that DE cannot when - predict the core facts. Notice that PIs are licensed in the restriction (the clause) as well as in the scope of HAB (the main clause), which makes it impossible to invoke DE of a potentially universal restriction for licensing. 2.6 Generic sentences Likewise, non-monotone generic sentences accept PIs. Any cat hunts mice. (30) 10

11 considered That GEN is not monotone has come be operator as generic the standard, see Krifka et al. 1995 and references therein for discussion. 2.7 Modal verbs Modal verbs create generally good environments for PIs, despite the fact that they are non-monotone, or worse, upward entailing. a John may talk to anybody. (31) Any minors must be accompanied by their parents. b The search committee can give the job to any candidate. c a. John {may/can/ must} dance -/ → John {may/can/ must} tango (32) → John {may/can/ must} dance b. John {may/can/ must} tango Notice that trying to void this argument by saying that in this case we have ‘free ’ is besides the point, since FCIs any are choice PIs. 2.8 Imperatives Here we have a parallel with modal verbs and intensional contexts in general: (33) kanena/opjodhipote } milo. Pare { take.2sg AP- FC apple ‘Take any apple.’ PIs in imperatives are quite problematic for DE. Note again that appealing to FC does not help. In Greek, as we see, FCIs and APIs occur in imperatives. any 2.9 Protasis of conditionals Heim 1984 shows that the protasis of conditionals is not strictly speaking DE; but it is one of the most common environments for PIs. An kimithis me { kanenan/opjondhipote }, tha se skotoso. (34) if sleep.2sg with AP- FC-person fut you kill.1sg If you sleep with anybody, I’ll kill you. (35) a. If you go to Spain you will have a great time -/ → b. If you go to Spain and get sick you will have a great time. 2.10 Directive intensional verbs PIs are licensed with directive propositional attitudes but not with epistemic ones. John would like to invite any student. (36) a 11

12 b John asked us to invite any student. d * John believes that we invited any student. * John dreamt that we invited any student. e na afiso { (37) na perasi mesa. epemine I Ariadne opjondhipote/kanenan} the Ariadne insisted.3sg subj let.1sg FC-person subj come.3sg in ‘Ariadne insisted that I allow anyone in.’ pistevi oti akuse { kanenan/opjondhipote } thorivo. * O Pavlos (38) the Paul believe.3sg tha. indicative heard.3sg API /FCI noise * Paul believes that he heard any noise. The facts were introduced and discussed in Giannakidou (1998, 1999), and seem generalize in many other languages: Spanish, Catalan (Quer 1998, 2000), to Russian (Pereltsvaig 2000; see Haspelmath 1997 for more). In Giannakidou 1998, 1999 I proposed an analysis of epistemic attitudes as veridical, and directive ones follows; I will not as nonveridical, from which PI-licensing (and mood choice) repeat the analysis— suffice it to note that DE cannot explain these facts. Worse, recent attempts to “monotonize” attitudes (von Fintel 1999) predict no PIs with directive attitudes because they are upward entailing, contrary to fact. 2.11 Problematic restrictions of universals Finally, consider the contrast between the restriction of every , which allows for PIs, and that of each and both , which don’t (see Horn 1972, Seuren 1984 already). (39) { / the students } who saw anything should report to the Every student police. (But I doubt that there will be any such students). (40) * Each student who saw anything should report to the police. a. (But I doubt that there will be any such students). *Both students who saw anything should report to the police. b. (But I doubt that there will be any such students). The contrast doesn’t seem to follow from DE—and again, it is predicted not to obtain in analyses like von Fintel’s 1999 (to be discussed in section 3), since both and each are Strawson-DE by his definition. Rather, the contrast follows if we assume that and both , but not every , are presuppositional, as evidenced by each the contradictory continuations in (40): their restriction must be non-empty. (41) DEFINITION 4 — Presuppositionality of determiners and quantifiers A determiner/quantifier δ is presuppositional iff for all A, B ⊆ D, if A = ∅ ). (based on Heim and Kratzer 1998: 163) then ∉ Dom ( δ 12

13 Presuppositional determiners and quantifiers are veridical, is clear from the as definition below, from Giannakidou 1998: (75) (Non)veridicality of determiners and quantifiers DEFINITION 5 — is veridical w.r.t. its NP argument iff it holds δ A determiner/quantifier NP VP]] = 1 → [[ NP]] δ that: [[ ∅ ; otherwise, δ is nonveridical. ≠ the domain In determiners, then, truth meets existence. For extension of of (non)veridicality for determiners in this spirit see Bernardi 2002. 2.12 Hierarchy of DE triggers can’t offer much help most seen so far To sum up, we have PI-environments are actually not that accounted for by DE. Would appealing to DE-hierarchies (Zwarts 1993, van der Wouden 1994) offer any help? Obviously not: we noted already that presumably strong PIs (expected to be licensed roughly by negative triggers), are grammatical in non-DE sentences: e.g. questions, habituals, modals, etc. On top of that, consider that there are PIs that don’t fit the DE-hierarchy at all. FCIs are typical such cases: they are grammatical in the non-DE contexts previously, but are unacceptable in the scope of DE mentioned negative and operators, if episodic. Episodic structures are structures which involve a single event as we see; such structures are perfective in Greek. (42) a Dhen idha kanenan. not saw.perf.1sg API-person ‘I didn’t see anybody.’ b *Dhen idha opjondhipote. not saw.perf.1sg FC-person (‘I didn’t see anybody.’) (43) ∃ !e φ (episodicity) (e) The decisive factor is that perfective positive and negative sentences are episodic: they involve existential closure of an event variable. FCIs, but not any or APIs, as we see, are apparently sensitive to this property. Anti-episodicity systematically blocks FCIs from other expected contexts, e.g. questions, and is also observed in Spanish and Catalan FCIs (see data in Quer 1998, 2000, and Giannakidou 2001): (44) * Su sistisan opjondhipote thavmasti? you introduced.perf.3pl FC admirer ‘Did they introduce any admirer to you?’ 13

14 So, for FCIs is not just nonveridicality that matters, but episodicity. We also revisit the issue in 4.2, where we derive anti-episodicity from the lexical semantics of FCIs, following the analysis I proposed in Giannakidou 2001. What we need to keep in mind here is that FC sensitivity is clearly not of DE nature. 2. 13 Licensing conditions and nonveridicality The discussion thus far provided enough motivation for the need to move from DE to nonveridicality in order to account for both DE- non-DE licensers. and Since DE functions are a proper subset of the nonveridical (for a proof see Zwarts 1995), the theoretical move from DE to nonveridicality is not at all in conflict with the previous alternative, but rather a (conservative) extension of it. Based on the distributions we have seen, we can formulate the following licensing conditions for APIs and FCIs: Licensing condition for FCIs ( -dhipote items) (45) α A FCI is grammatical in a sentence S iff: α β in S; and (i) is in the scope of a nonveridical operator (ii) S is not episodic. Licensing condition for APIs (46) kanenas items) ( A FCI α is grammatical in a sentence S iff: α is in the scope of a nonveridical operator β in S. These conditions have the status of global filters, albeit empirically more adequate ones than the DE-based (2). But as I said at the beginning, we want to derive these conditions from the lexical semantic content of the PIs. In section 4, I show that there is a direct link between nonveridicality and the lexical semantics of these PIs. any , we have to postulate a weaker condition, because (a) there are For structures where any nonveridical not accepted, and (b) there are veridical is structures where any occurs if a nonveridical inference is licensed pragmatically, e.g. as a negative implicature. The latter is illustrated by and negative factives, only and discussed in section 3 next. The former becomes evident when we consider any in propositional attitudes. Though any is fine with directive verbs, as we saw in 2.10, its occurrences are more limited than those of its Greek counterparts, which are freely licensed. The contrast is illustrated in the examples below with the verb efxome ‘wish’. (47) Efxome na me voithuse kanenas. * I wish anybody helped me. 14

15 Wish is nonveridical and licenses ; any however is bad. Likewise, any is kanenas perhaps : bad with disjunctions, as we note in fn 1, and with adverbs like (48) Isos na bike mesa kanenas. * Perhaps anybody came in. the relation of any to veridicality Obviously, not the must relation then, is must not relation that indicates antilicesing imposed by licensing, but the weaker (Giannakidou 1998). In Giannakidou 1999, I proposed the following condition: Anti-licensing condition regulating the distribution of any (49) Any any i. is interpreted in the will not be grammatical in a sentence S if n S. scope of a veridical expression i ii. In certain cases, (i) can be voided if S gives rise to a negative implicature. Items subject to this condition are expected, of course, to be good in nonveridical but this expectation does not have structures, status of a necessary and the sufficient condition, as in the case of licensing (which is an attraction relation between PIs and context). Furthermore, unlike licensed-by-nonveridicality items, any can be good as long as veridicality is voided somehow (since anti-licensing is merely an avoidance relation). For APIs and FCIs, the requirement that they be in the of a nonveridical expression is a must condition and can never be scope violated: this is why Greek PIs are bad, but any only and negative is good, with factives. 3. Pragmatic licensing: and emotive factives only is known to be good with only Any and negative factive verbs: (50) Only Larry talked to anybody. (51) a Bill {is surprised/regrets} that I have any friends. b *Bill is glad that I have any friends. These data are problematic if we want to maintain a purely semantic account for the distribution of , since only and negative factives are veridical and not DE. any Linebarger 1980 pointed out this problem for Ladusaw’s theory, and proposed to any in these contexts in terms of a negative implicature: only and negative handle factives license negative implicatures and it is by virtue of this that they allow for it . (Linebarger actually proposed this as a general theory for any , but any obviously cannot be right, given the data we just discussed in section 2). 15

16 Note first the problem is not general. PIs that are licensed by that grammatical in these cases, as I illustrated nonveridicality, e.g. Greek APIs, are not in earlier work (Giannakidou 1999, 2001), because the sentences are veridical: if , then somebody ate and negative, are veridical only Mary ate . Factives, positive because they presuppose the truth of their complement. * Monon i Theodora idhe { opjondhipote/kanenan } fititi. (52) student only the Thedora saw.3sg FC/ API ‘Only Theodora saw any students.’ a * Ekplisome pu exi { opjondhipote/kanenan } filo. (53) be-surprised.1sg that has FC / API friend ‘I’m surprised she has any friends.’ b * Xerome { opjondhipote/kanenan } filo. pu exi ‘*I’m glad she has any friends.’ So, the Greek items do not violate their licensing condition which requires a nonveridical trigger. We see in that other PIs licensed by nonveridicality (54) cualquier confianza behave exactly the same way, e.g. the Spanish : * Solo el professor mostro cualquier confianza . (data from Josep Quer) (54) Only the teacher showed any confidence. Hence the behavior of seems to be exceptional (see Haspelmath 1997 for more any only data in support of this). Still, the fact that some PIs are admitted with and factives poses a problem, if we want to maintain a purely semantic approach. To this end, there has been a recent attempt by von Fintel 1999 to render only and factives DE, despite the fact that they seem not to be: (55) Only Larry ate a vegetable → Only Larry are kale. -/ Larry may have eaten spinach, for instance. Larry regrets that I bought a car. -/ → Larry regrets that I bought a Honda. (56) Because, in fact, I bought a Ferrari, and Larry might not regret this at all. Von Fintel argues that, although only and negative factives do not exhibit the classical DE-pattern, they can be shown to be DE in a special way. He defines a notion of Strawson-DE for expressions that come with presuppositions. The idea is that when we evaluate the monotonicity properties of such expressions we must check for DE only in contexts where the presuppositions are satisfied. 16

17 (57) Strawson DE (von Fintel 1999: 14) σ , > is Strawson-DE iff A function f of type < τ such that x y, and f is defined: f(y) → f(x). σ for all x, y of type → Strawson DE is clearly not entailment in the classical sense that whenever f(y) is true, f(x) is true too. Srawson-DE comes with the following reasoning scheme: a. P Q (58) → b. [[ only John ]] (P) is defined. c. [[ only John ]] (Q) is true. d. Therefore, [[ only John ]] (P) is true. The crucial premise is (58b): the intended presupposition. In a context satisfying this premise, premises (58a) and (58c), the sentence whose monotonic inference we want to derive, allow the conclusion in (58d). Here is an example: (59) a. Kale is a vegetable. b. Somebody ate kale for breakfast. c. Only John ate a vegetable for breakfast. d. Therefore, only John ate kale for breakfast. Suppose that b is given in the context, i.e. that somebody ate kale for breakfast. c is true, namely that John only ate a vegetable for breakfast, we can Then if d, conclude only is Strawson DE and that only John ate kale for breakfast. Hence it therefore licenses . any The argument for negative factives takes a much more complicated route, but it goes against many of the basic assumptions about the implications we get with propositional attitudes, and how the dynamics of context change work; I will not discuss it here for reasons of space. Von Fintel proposes an analysis of directive attitudes as upward entailing; but if this is correct, we can’t explain why PIs, included, are grammatical with verbs like insist, be willing to, and ask , as any verbs It more we noted in 2.10. to analyze these seems as non- appropriate monotone, as in the more classical approaches (Heim 1992). Likewise, the argument for only turns out to be quite problematic. To begin with, it is not uncontroversial that the relevant component of only is a presupposition. Atlas (1993, 1996) argues that it is an entailment, and likewise Horn 2001 is willing to relax the presupposition analysis: the presupposition part is asserted though it remains assertorically inert . But even if we accept the presuppositional analysis, the argument still suffers from two serious problems. The first one has to do with what is taken to 17

18 count as a presupposition. What von Fintel characterizes as a presupposition of only in premise b is not the presupposition extracted from the sentence we which which premise c, but an arbitrary proposition actually evaluate, we is decide to include as part of the common ground. Such arbitrary propositions are allowed to influence reasoning in an unconstraint way, and this leads to the second problem—generating Strawson-DE and PIs in contexts where they do not occur. To see the first problem, consider what the presupposition of the sentence Only John ate a vegetable for breakfast is. It is (60b): that somebody ate (60c) a vegetable for breakfast, and not kale as in von Fintel’s (59b). (59b) is just some independent of the proposition that we agree to have in as background knowledge, presupposition of the sentence. If we replace (59b) with the actual presup- only- position, we see that Strawson-DE doesn’t go through: (60) a. Kale is a vegetable. b. Somebody ate a vegetable for breakfast . c. Only John ate a vegetable for breakfast. d. DOES NOT FOLLOW: Only John ate kale for breakfast. So, for Strawson-DE to work, we allow not the presupposition of the evaluated sentence but just any arbitrary proposition to count as relevant. This has quite undesirable consequences, the obvious one being that it overgeneralizes. Wrong two cases: generalized focus, and predictions are made in (at least) . both/each Consider focus first. Von Fintel’s scheme predicts that other focus exhaustive structures, e.g. preposed focus and clefts will also make any licit, since it is not just only , but exhaustive focus in general that validates his Strawson DE: (61) a. Kale is a vegetable. b. Somebody ate kale for breakfast. c. It was John who ate a vegetable for breakfast. d. Therefore, it was John who ate kale for breakfast. c’ : JOHN ate a vegetable for breakfast. d’. Therefore, JOHN ate kale for breakfast. With (61a,b) as the background, the cleft sentence in (61c) and the preposed focus in (61c’) end up Strawson-DE; yet, unlike only, clefts and preposed exhaustive focus, do not license any. (62) a * It was John who talked to anybody. * JOHN talked to anybody. b 18

19 Strawson-DE doesn’t get things right here. Now, consider presuppositional both each . In 2.11 we noted that these exclude from their restriction. and any *Each student who saw anything reported to the Dean. a (63) * Both students who saw anything reported to the Dean. b and Yet, the restriction of is Strawson-DE by von Fintel’s criterion. each both (64) a. A linguistics student is a student. b. There is a set of exactly two linguistics students that we are talking about. Both students got an A. c. d. Therefore, both linguistics students got an A. (65) a. A linguistics student is a student. b. There is a set of linguistics students that we are talking about. c. Each student got an A. d. Therefore, each linguistics student got an A. Again, Strawson-DE makes the wrong predictions. So, if we allow Srawson-DE we end up overgeneralizing ( both, each, clefts, and preposed focus ), or under- would generalizing And, of course, we attitudes). still have to (propositional account for the considerable number of non-DE environments we mentioned in section 2. We must conclude, then, that Strawson-DE is not a very useful notion. This conclusion seems to threaten Horn’s DE- assertion (versus entailment ) too: the problem of overgeneralization that we noted for Strawson-DE seems to carry over to DE-assertion, at least in the exhaustive focus structures. (And it is hard to see how negative factives can be handled.). For only , then, the correct analysis seems to be the non-monotone position of Atlas 1993, 1996, and our best bet for an explanation of why some PIs occur with it must still make use of a pragmatic condition invoking the availability of a negative implicature, as I suggested in (49). 4 Deriving licensing for lexical semantics: dependent variables In this last section, I consider the issue of sensitivity. We have to show that there is a connection between nonveridicality and the meaning of PIs. I will concentrate here on APIs and FCIs. For reasons of space, I will not discuss any in great detail (but see Giannakidou 2001 for a more detailed discussion). 19

20 4.1 Dependent indefinites PIs have restricted distribution because they are lexically ‘deficient’ expressions. They can only be interpreted in structures that are consistent with their deficit; if a PI occurs in a structure that imposes an interpretation that the PI cannot have because of its lexical deficit, the structure cannot receive an interpretation, and will be ruled out as ungrammatical. The basic observation is that some variables cannot be existentially closed ordinary way. Compare in sentences below, the ungrammatical one the the containing PIs, and the grammatical one containing a ‘regular’ singular indefinite: * Idha { opjondhipote/kanena } fititi. (66) FC/ AP student saw.1sg ‘*I saw any student.’ (67) Idha enan fititi. saw.1sg a student ‘I saw a student.’ enan fititi ‘a student’ contributes a restricted variable as in The non-PI indefinite (68a) which undergoes text level existential closure, as in Heim 1982. The formula receives the usual truth conditions of existential statements. (68) a [[ a student ]] = student (x) (x) ∃ x [ Existential closure (Heim 1982) student ∧ b saw (I,x)] Text-level closure assigns the ‘specific’, wide scope reading existential of indefinites (as opposed to nuclear scope existential closure which gives the narrow scope). In Giannakidou 1998, 2001 I argued that we must identify certain PIs as indefinites introducing variables which cannot undergo these standard procedures . I proposed a new type of variable which is dependent ; APIs and FCIs introduce dependent variables. In the former case the dependent variable is of the individual type e , in the latter of the world type s ; it is the presence of such variables that restricts the distribution of APIs and FCIs. Ontology of variables (69) Type Independent Dependent e x x d s w w d Examples: a student (x) kanenas fititis (x ) d ) opjosdhipote fititis (x, w d 20

21 Before we let me just note that admitting different kinds of proceed, not at peculiar. It has been a commonplace observation in the variables is all variables This alike. all underlies, for example, the very literature that not are 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 extensions of this idea imposing constraints on how the variables of ‘specific’ indefinites must be interpreted (e.g. Farkas 2002; recall also our discussion in 1.1). Admitting that PIs are dependent variables must be understood in certain context. Only this instead of positing dependency as a presupposition, I propose to encode it in the type of the expression itself. This way, we explain why the failure of a dependent variable in the wrong context yields ungrammaticality, and is not merely a case of presupposition failure. APIs are indefinites that cannot assert existence in the default world of — though they may do so in a hypothetical, non-actual context. In evaluation, w 0 x is dependent because the case of APIs, then, the variable x cannot introduce a d d discourse referent in the actual world. DEFINITION 6 — Dependent Indefinites (70) (cf. Giannakidou 1998: 140) An is dependent iff the variable x indefinite it contributes cannot d introduce a discourse referent in the actual world w . 0 This means that x closed by the default existential quantifier (either cannot be ∃ - d text-level in the nuclear scope) in a veridical context. Notice that it is or an not that matters: x extensionality veridicality and can still be ∃ - closed in d not ∃ closure happens under negation, since x be extensional context if - would d forced to introduce a discourse referent under negation. ∃ -closure will generally be fine in the scope of a nonveridical operator, because the nonveridical operator ensures that x will not be forced to introduce a discourse referent in the actual d scope general, are good world. This explains why APIs, and PIs the in of in 2 negation and in intensional, modal, and conditional contexts. 2 The notion of an unidentifiable variable (Farkas 2002) may be relevant here. A variable x in φ is unidentifiable iff for any , if f satisfies φ relative to w there are several values assigned to x relative to f. The idea is that there is no context in which an unidentifiable variable receives a single value only (roughly, the property of having wide scope). This makes such variables always take narrow scope, which is an obvious property of APIs in the definition of dependency I pursue here. But from my view of dependency, it also follows that APIs will not be good in a veridical context, even if they will be able to receive narrow scope, e.g. if a higher universal quantifier is present: * Kathe aghori diavase kanena vivlio *‘ Every boy read any book’. It is not clear to me how unidentifiability could rule out this case. Unidentifiability may be a more useful notion for the PIs egy-egy in Hungarian, which, contrary to any and the Greek items, seem to have just the property of narrow scope, and are fine in the case just mentioned. 21

22 If we want to cast the idea in Kamp and Reyle’s DRT, we must say that dependent existentials cannot introduce a variable into a main DRS; but they can do so in embedded DRSs. In Giannakidou (1998: 138-141), it is further claimed a introduce the set of worlds referent that the dependent indefinite cannot in compatible with what an individual believes. This was needed to account for APIs in propositional attitudes— see that work for more details. Crucially, the variable dependency of APIs derives their narrow scope and the need to be c-commanded by a nonveridical licenser from the semantics, and nothing special needs to be said about the syntax. Without a nonveridical licenser, ∃ -closed and the structure is uninterpretable, thus x ungrammatical. cannot be d Dependency has another welcome result: it can help us explain the recalcitrant any under problem of s-structure c-command which seems to hold for APIs and negation; space prevents me from elaborating on this here, but see Giannakidou and Merchant (2002) for more details. kanenas any, as I argued in The Greek items are dependent; and so is Giannakidou (2001); den Dikken and Giannakidou 2002 further characterize wh- phrases as PIs which belong to this class. The analysis can also extend to the-hell and English ever . ooit include Dutch 4.2 Free choice items as intensional indefinites FCIs contain a dependent world variable. Recall that nonveridicality alone was not enough to predict the correct distribution; our licensing condition had to include a clause further restricting FCIs to contexts which are not episodic. In Giannakidou 2001, I explained anti-episodicity by arguing that FCIs are intensional indefinites. They contain a dependent variable w that must be bound d by an operator that can bind such a variable— a Q, modal, or intensional operator. In an episodic context (veridical or not) there is no such operator, the variable remains unbound, and the FCI is uninterpretable. (71) *Idha opjondipote ston kipo. not saw.1sg anybody in-the garden (72) *Dhen idha opjondipote ston kipo. not saw.1sg anybody in-the garden (73) # ∃ !e ∃ x [ person (x, w (e)] ) ∧ saw (I, x, e) ∧ in-the-garden d ∧ # ¬∃ !e ∃ x [ person (x, w (e)] ) (74) saw (I, x, e) ∧ in-the-garden d Because the word variable w is dependent it cannot be assigned the default value d which would have to be the case in an extensional context. of the actual world w 0 22

23 Since w , and since there is no world-binder in is not a possible value for w 0 d episodic sentences, FCIs are uninterpretable and the structures are ruled out. type is as a , then, shifter of type The free choice determiner - dhipote , which, when applied to a property, the common noun <,>> denotation, returns an intensionalized property as its output. (75) λ P . λ ]] = [[ -dhipote . λ x[P(x)(w )] w FC <, >> d d [[ opjosdhipote fititis ]] = student (x)(w (76) ) d If the property is intensionalized already, then the free choice determiner will act as a filter passing intensionality to the higher node, as in the case of opjosdhipote ‘any alleged criminal’. ipotithemenos eglimatias This account explains the distribution of FCIs straightforwardly, rendering the licensing condition in (45) superfluous. Out goal is thus achieved. The residual issue of what yields quasi-universal readings with FCIs calls for an explanation-- and for this we need to posit a presupposition of exhaustive variation in identity Giannakidou 2001 for details). Crucially, alternatives (see variation exhaustive (and possibly scalarity, or widening in the sense of Kadmon and Landman 1993, if one wants to include these in the semantics of FC), are not properties that restrict the distribution of FCIs; they only further describe their interpretation. For this reason, they cannot be used successfully to predict where PIs will be grammatical or not. This can only be done by the type combinatorics I postulated here. 5 Conclusions The main conclusions of the discussion can be summarized as follows. 1. We need a conservative extension from DE to nonveridicality in order to capture correctly PI-distribution within and across languages. It is hard to find a PI-series that is licensed just by DE (such a PI would have to be admitted by DE quantifiers, but not in questions, for instance). DE-approach. It 2. Strawson-DE does not help rescuing the the leaves occurrence of PIs in modal, intensional, future, habitual, generic and other non- monotone contexts unexplained. And, it overgeneralizes with respect to focus and presuppositional determiners by predicting PIs where in fact they do not occur. 3. Within the nonveridicality hypothesis it is possible to derive limited distribution from lexical semantics: we just need to identify a new type of variable dependent . Free choice items come with a dependent world variable that that is must be bound by a Q-operator; if there is no such operator, as is the case in structure remains unbound, rendering the episodic contexts, the world variable ungrammatical. Affective PIs, on the other hand, come with a dependent individual 23

24 variable which cannot introduce a discourse referent in the actual world. Such a variable can obviously not be interpreted in veridical contexts. In sum, by deriving the restricted distribution of some PIs from the lexical semantics, we showed that at least some polarity cases are not that special, i.e. they need not be encoded in a specific polarity module in the grammar. Ultimately we want to derive the distribution of a given PI from mechanisms that are needed in the grammar anyway, and dispense entirely with polarity-specific filters of the traditional kind, which have been so popular in the DE-tradition. Acknowledgements I am grateful to all the people who commented on versions of my previous works on the issues discussed; I cannot possibly repeat the all the names here, but I wish to thank again Larry Horn, Jason Merchant, Josep Quer and Frans Zwarts. Many thanks also to Jay Atlas, and Donka Farkas for discussion of this paper, and to the audience of CLS 38 for their feedback. References Atlas, Jay David. 1993. The importance of being Only: Testing the Neo-gricean versus Neo- entailment paradigms. Journal of Semantics 10:301-318. Atlas, Jay David. 1996. Only Noun Phrases, Pseudo-negative quantifiers, negative polarity items, and monotonicity. Journal of Semantics 13: 265-328. Raffaella. 2002. Reasoning Bernardi, Polarity in Categorial Type Logic. PhD Thesis, with University of Utrecht. Aggressively phrases as den Dikken, Marcel, and Anastasia Giannakidou. 2002. non-D-linked Linguistic Inquiry 33: polarity items. 31-61. Dowty, David. 1994. The role of negative polarity and concord marking in natural language reasoning. Semantics and Linguistic Theory (SALT) IV. Cornell University, Ithaca, NY. Farkas, Donka. 2002. Varieties of indefinites. In the Proceedings of Semantics and Linguistic (SALT) 12. Theory Journal of Von Fintel, Kai. 1999. NPI-Licensing, Strawson-Entailment, and Context-Dependency. Semantics 16: 97-148. Giannakidou, Anastasia. 1995. Subjunctive, habituality and negative polarity. In SALT V: 132- 150. Cornell University, Ithaca, NY. Anastasia. Giannakidou, Polarity Sensitivity as (Non)veridical Dependency . John 1998. Benjamins, Amsterdam and Philadelphia. Giannakidou, Anastasia. 1999. Affective dependencies. Linguistics and Philosophy 22: 367-421. Giannakidou, Anastasia. 2000. Negative ...Concord? Natural Language and Linguistic Theory 18: 457-523. Giannakidou, Anastasia. 2001. The meaning of free choice. Linguistics and Philosophy 24: 659- 735. Giannakidou, Anastasia, and Jason Merchant. 2002. Modularity and the minimalist program. Manuscript. University of Chicago. Giannakidou, Anastasia and Frans Zwarts. To appear. Temporal, aspectual operators and (Non)Veridicality. To appear in The Syntax and Semantics of Tense and Mood Selection , edited by A. Giorgi, J. Higginbotham, and F. Pianesi, Oxford University Press. 24

25 Groenendijk, Jeroen and Martin Stokhof. 1997 Questions. Handbook of Logic and Language, ed. by J. van Benthem & A. ter Meulen. Elsevier Science Publishers, Amsterdam: 1055-1124. Indefinite Pronouns. Oxford: Clarendon. Haspelmath, Martin. 1997. Heim, Irene. 1982. PhD thesis, University The Semantics of Definite and Indefinite Noun Phrases. of Massachusetts, Amherst. Heim, Irene. 1984. A note on negative polarity and downward entailingness. NELS 14: 98-107. Heim, Irene. 1992. Presupposition projection and the semantics Journal of of attitude verbs. Semantics 9: 183-221. Blackwell. Heim, Irene, and Angelika Kratzer. 1998. Semantics in Generative Grammar. Hoeksema, Jacob. 1983. Negative polarity and the comparative. Natural Language and Linguistic Theory 1: 403-434. Horn, L. R. 1996. Exclusive company: and the dynamics of vertical inference. Journal of only Semantics 13: 1-40. Horn, L. R. 2001. Assertoric inertia and scalar inference. Formal Pragmatics Workshop, Berlin. Linguistics and Philosophy Israel, M. 1996. Polarity sensitivity as lexical semantics. 19:619-666. 16: 353-422. Kadmon, Nirit and Fred Landman. 1993. Any. Linguistics and Philosophy From Discourse to Logic. Kluwer, Dordrecht. Kamp, Hans, and Uwe Reyle. 1993. 1:3-44. Karttunen, L. 1977. Syntax and semantics of questions. Linguistics and Philosophy Klima, E. 1964. Negation in English. In Katz and Fodor (eds), 246-323. Krifka, Manfred, F. J. Pelletier, G. Carlson, A. ter Meulen, G. Chierchia and G. Link (1995) The Generic Book, 1- Genericity: an introduction. In G. Carlson & F. J. Pelletier (eds.), Chicago: U. of Chicago Press. 124. Ladusaw, William A. 1980. Polarity Sensitivity as Inherent Scope Relations. Garland, NY. Ladusaw, William. A. 1986. Principles of semantic filtering. In the Proceedings of WCCFL 5. Natural Language Semantics 6-57-23. Lahiri, Utpal. 1998. Focus and negative polarity in Hindi. Foundations of Cognitive Grammar. Stanford Univ. Press Langacker, R. 1991. . Linebarger, Marcia. 1980. The Grammar of Negative Polarity. PhD thesis. MIT. Montague, Richard. 1969. On the nature of certain philosophical entities. The Monist 53: 159-94. Formal Philosophy. Selected papers of Richard Montague Reprinted in 1974 in . ed. by R.H.Thomason. Yale University Press, New Haven, 148-87. Quer, Josep. 1998. Mood at the Interface. PhD thesis, University of Utrecht. Quer, Josep. 2000. Licensing free choice items in hostile environments: the role of aspect and mood. Ms. University of Amsterdam. Asya. 2000. Monotonicity-based versus Pereltsvaig, approaches to negative veridicality-based polarity: evidence from Russian. In Formal Approaches to Slavic Linguistics, 328-346. Szabolcsi, Anna. 2001. Positive polarity-negative polarity. Manuscript. New York University. Tovena, Lucia. 1998. The Fine Structure of Polarity Items. Garland, NY. Vikner, Carl. 1999. Episodic and habitual temporal connectives in Danish, Norwegian and Swedish. NORDSEM Report, University of Copenhagen. Negative Contexts. PhD thesis, University of Groningen. van der Wouden, Ton. 1994. Zwarts, Frans. 1986. Categoriale Grammatica en Algebraische Semantiek. Een studie naar negatie en polariteit in het Nederlands. PhD thesis. University of Groningen. Zwarts, Frans. 1993. Three types of polarity. Ms. Also appeared as Zwarts 1998 in Plural Quantification , ed. F. Hamn and E. Hinrichs, Kluwer, Dordrecht. 25, 3-4: 286-312. Zwarts, Frans. 1995. Nonveridical contexts. Linguistic Analysis 25

Related documents

itemlist2009

itemlist2009

OKLAHOMA DEPARTMENT OF TRANSPORTATION DATE : 04/22/19 LISTING OF SELECTED ITEMS PAGE : 1 SP ITEM ITEM ITEM ITEM YR NO. DESC UNIT TYPE CLS --------------------------------------------------------------...

More info »
Volume 14 tab 3

Volume 14 tab 3

Current Limiting Fuses Product Overview 3.1 Eaton Current Limiting Fuses 3 V14-T3-2 Product Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3.2 Fuse Selecti...

More info »
OnlineSOCfall

OnlineSOCfall

Schedul e of Onl i ne Cl asses Report ID: Online Class Schedule For The Term Fall 2019 5/8/2019 8:30:18 AM Report Run Date: University of Georgia Offi ce of Regi strar SUBJECT COURSE NO TITLE ONLINE B...

More info »
00050 98541

00050 98541

The Incidence of Contact Lens–Related Microbial Keratitis in Australia 1,2,3 2,3 2,3 Keay, PhD, Stapleton, PhD, Katie Edwards, BAppSc (Optom), Lisa Fiona 1,2,3 4,5 3,6 1,2,3,6 Brian, Dart, DM, Naduvil...

More info »
Microsoft Word   the joy of micro c.doc

Microsoft Word the joy of micro c.doc

The Joy of Micro -­‐ C This document explains how to write code for the Micro Engines or “ME”s on 6xxx chip. We will use Netronome’s Programmer Studio SDK. - the NFP There are over a hundred MEs on th...

More info »
Law & Justice2.pmd

Law & Justice2.pmd

THE CONSTITUTION OF INDIA Preamble. WE, THE PEOPLE OF INDIA, having solemnly 1 resolved to constitute India into a [SOVEREIGN SOCIALIST SECULAR DEMOCRATIC REPUBLIC] and to secure to all its citizens: ...

More info »
Have you ever wondered why we say fiddle faddle and not faddle fiddle

Have you ever wondered why we say fiddle faddle and not faddle fiddle

Order Preference for Reduplicated Words with Differing Vowels M ELDON ARIA S. B ou ever wondered why fiddle-faddle and not faddle-fiddle ? Why it ping-pong Have y we say is rather than pong-ping and p...

More info »
Faster R CNN: Towards Real Time Object Detection with Region Proposal Networks

Faster R CNN: Towards Real Time Object Detection with Region Proposal Networks

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks ∗ Kaiming He Ross Girshick Jian Sun Shaoqing Ren Microsoft Research v-shren, kahe, rbg, jiansun } @microsoft.com { Abstra...

More info »
MVS Diagnosis: Tools and Service Aids

MVS Diagnosis: Tools and Service Aids

z/OS Version 2 Release 3 MVS Diagnosis: Tools and Service Aids IBM GA32-0905-30

More info »
A Design Rationale for C++/CLI

A Design Rationale for C++/CLI

A Design Rationale for C++/CLI Version 1.1 — February 24, 2006 (later updated with minor editorial fixes) Herb Sutter ([email protected]) 2 1 Overview Key Goals 2 1.1 3 1.2 Basic Design Forces Pre...

More info »
os guide v9 march 2014

os guide v9 march 2014

OPERATING AND SUPPORT ESTIMATING GUIDE COST- OFFICE OF THE SECRETARY OF DEFENSE COST ASSESSMENT AND PROGRAM EVALUATION MARCH 2014

More info »
u blox ZED F9P InterfaceDescription (UBX 18010854)

u blox ZED F9P InterfaceDescription (UBX 18010854)

u-blox ZED-F9P Interface Description Abstract The Interface Description describes the UBX (version 27. 10), NMEA and RTCM protocols and serves as a reference manual for the u-blox ZED-F9P high precisi...

More info »
A Financial System That Creates Economic Opportunities: Capital Markets

A Financial System That Creates Economic Opportunities: Capital Markets

U.S. DEPARTMENT OF THE TREASURY A Financial System That Creates Economic Opportunities Capital Markets OCTOBER 2017

More info »
National Dairy Products Sales Report

National Dairy Products Sales Report

Announcement of Class and Component Prices United States Department of Agriculture Agricultural Marketing Service Market Information Branch Dairy Program CLS - 0419 May 1, 2019 April 2019 Highlights w...

More info »
bookdown: Authoring Books and Technical Documents with R Markdown

bookdown: Authoring Books and Technical Documents with R Markdown

Yihui Xie bookdown: Authoring Books and Technical Documents with R Markdown

More info »
Comparison of Differences In Insurance Costs; For Passenger Cars, Station Wagons/Passenger Vans, Pickups, and Utility Vehicles on the Basis of Damage Susceptibility; Calendar Years 2015   2017

Comparison of Differences In Insurance Costs; For Passenger Cars, Station Wagons/Passenger Vans, Pickups, and Utility Vehicles on the Basis of Damage Susceptibility; Calendar Years 2015 2017

DOT HS 812 September 2018 538 Comparison of Differences In Insurance Costs For Passenger Cars, Station Wagons/Passenger Vans, Pickups, and Utility Vehicles on the Basis of Damage Susceptibility; Calen...

More info »
Microsoft Word   June 2017 1225 Report

Microsoft Word June 2017 1225 Report

ENHANCING SECURITY AND STABILITY IN JUNE 2017

More info »
Minimum Operating Standards for MI Health Link March 2017 557197 7

Minimum Operating Standards for MI Health Link March 2017 557197 7

Michigan Departm ent of Health and Human Services Minimum Operating Standards for MI Health Link Program and MI Health Link HCBS Waiver Minimum Operating Standards For MI Health Link Program and MI He...

More info »