Double negatives, Negative Concord and metalinguistic negation Luis Alonso Ovalle and Elena Guerzoni UMass, Amherst

and MIT 1 Introduction This paper presents a new approach on Negative Concord (NC) in Italian and Spanish which explains some facts that do not find a unified account in any of the existing analyses of this phenomenon. In NC Languages of the Spanish and Italian variety, expressions with overt negative morphology like nadie / nada and nessuno / niente (also called nwords , after Laka 1990), exhibit unstable semantic and distributional behavior. Nwords sometimes behave as Negative Polarity Items (NPIs) like anybody / anything and sometimes as Negative Quantifiers (NQs) like nobody / nothing. For example, even if there is no negation besides the sentence initial n-word, the sentences in (1) are interpreted as negative statements. The n-word appears to contribute a Negative Quantifier with sentential scope. (1) a. Nessuno venuto. N-body is come b. Nadie ha venido. N-body has come ' Nobody came' (Italian) (Spanish)

On the other hand, n-words occurring within the scope of an overt negation, as in (2), do not contribute a negation, as the interpretation of the sentence involves only one negation. (2) a. *(Non) venuto nessuno. Not is come n-body b. *(No) ha venido nadie. Not has come n-body ' Nobody came' (Italian) (Spanish)

In addition, as illustrated in (2), non-initial n-words are dependent on the prese nce of the higher negation, an effect that is reminiscent of well-known Negative Polarity phenomena. In fact, the similarity with NPIs like any goes beyond the facts in (2), since non-initial n-words are acceptable in other contexts that typically license NPIs. Like any , n -words are licensed in the complement of without, and of before and in embedded clauses introduced by doubt. (3) a. Maria usc senza salutare nessuno. ' Mary left without greeting anyone' b. Maria usci' prima di aver salutato nessuno

'Mary left before greeting anybody' c. Dubito che Maria abbia visto nessuno. 'I doubt that Mary saw anyone' This pattern suggests the following preliminary descriptive generalization: (4) a. When n-words appear in sentence initial position there is no need for an overt negative licenser. The n-word itself appears to be a NQ ( ). b. In all other cases, a licenser is required and the n-word appears to be an existential quantifier, just like any. ()

We will first concentrate on the clause in (4b). A closer look at the distribution of n -words in non-initial positio immediately shows that (4b) is an oversimplified description of the facts. N-words are much more constrained in their postverbal occurrences than NPIs like any . Merely Downward Entailing (DE) environments and even antiadditive 1 ones (see Zwarts 1993) license NPIs, but they do not license n-words (as shown in (5a,b)). In addition, n -words are infelicitous in the antecedent of conditionals (5c) and completely ungrammatical in the fact ive environments in (5d-e), while any and minimizers like a red cent are fine in all these contexts. 2 (5) a. * Meno di tre studenti hanno mangiato niente. 3 Less than three students have eaten n-thing b. ?? Ogni passeggero che avesse bisogno di niente, fu assistito. Every passenger that needed n-thing, was assisted c. ?? Se Maria si accorgesse niente, sarebbe un problema. If Mary noticed n- thing, it would be a problem d. ?? Solo Maria ha visto nessuno degli studenti. Only Mary has seen n-one of the students e. * Mi spiacerebbe che tu vedessi nessuno. I would be sorry that you saw n-body

Non-initial n-words in the context of denials generate interesting unexpected interpretations, which also appear to escape the generalization in (4b). The dialog in (6) illustrates this phenomenon. 4 (6)
1

A: Maria stara'

morendo di fame, non ha mangiato niente oggi

Antiadditive operators are DE operators which resemble more closely negation in their logical properties, than merely DE ones, in that they also support the following entailment: Op (A B) Op (A) Op (B). 2 For an account of why this is so see Von Fintel (1999). 3 We include only the Italian examples, but Spanish exhibits exactly the same patterns. 4 The example (6) is reminiscent of data recently discussed in Herburger (2001) but see ft.11.

Mary will be starving , not has eaten n-thing today 'Mary is probably starving, she hasn't eaten anything today' B: Non ha mangiato NIENTE, ha mangiato un panino! Not has eaten N-THING, she ate a sandwich! 'It's not correct that she didn't eat anything: she ate a sandwich!' Within A's utterance, the negative word niente occurs under negation and is interpreted, as expected, as an existential quantifier, just like in (2). What is puzzling about (6) is that B uses exactly the same string A used ( non ha mangiato niente), modulo intonation, to actually express disagreement with what A said. B's remark is a denial of A's, even when the two sentences contain the same number of negations and negative words. Where does the additional negation expressing disagreement come from? What exactly is B disagreeing with? 5 A theory of NC for the two languages at stake should at least account for (i) the apparent negative interpretation of n-words in sentence initial position, (ii) their peculiar restrictions on their non-initial occurrences and (iii) their special interpretation in denials. Section 2 briefly illustrates that all previous analyses of NC fall short of coping with one or more of these aspects. In section 3 we propose that n -words are unambiguously non-negative existential quantifiers that come with a conventional implicature involving negation. In sections 3 and 4, we will argue that this semantic analysis of n -words can explain both the restrictions on their non-sentence initial occurrences and the special interpretations they generate in denials. In section 5, we envisage how to extend the proposal to the sentence initial cases. 2 Previous approaches and their problems The massive literature on NC in Romance falls into three categories. The first group proposes that n -words are unambiguously NPIs. The second entertain the hypothesis that n -words are unambiguously NQs. The third argues that n-words are lexically ambiguous: they are both NQs and NPIs. 6 The hypothesis that n-words are NPIs (Laka 1990, Ladusaw 1992, Giannakidou 1997, 2000) provides no straightforward account for why the distribution of n-words in non-initial position are more limited in their distribution than run of the mill NPIs like English any or minimizers like a finger in both English and Romance. 7 In addition, it fails to predict the difference between NPIs
5

Notice, in passing, that the English sentence with any under negation cannot be used as a denial of As remark, unless one negation is added, as in the English translation provided. 6 The advantages and disadvantages of each of these views are discussed at length in the relevant literature (see, for instance, Zanuttini (1991), Ladusaw (1992), Giannakidou (1997)-(2000), Herburger (2001)), to which we refer the reader for details. For reasons of space, in this section we will just provide a brief summary of those considerations, focusing especially on the problems these approaches have in dealing with the facts presented in section 1. 7 Any and minimizers like lift a finger and the Spanish and Italian counterparts muovere un dito and levantar un dedo are, indeed, fine in all the environments hosting non-initial n-words (e.g. the

and n-words in denials, since denials typically do not license NPIs while n-words can be used in these contexts, as we saw above. Finally, this view needs to postulate an abstract ne gation licensing sentence-initial n-words and contributing the negative interpretation of sentences like (1) above (see Ladusaw 1992). The hypothesis that n-words are always NQs is defended in Zanuttini (1991). Although it does not directly explain the non-negative readings of nwords and their peculiar NPI-like restrictions in postverbal position, at first sight it seems to have an advantage over the NPI-approach when it comes to explaining the behavior of n-words in denials. The meaning of Bs utterance in (6) could just be viewed as a double negation (DN) statement. If one negation is provided by non and the n-word niente contributed another one, as indeed assumed within the NQ approach, we would have a fully compositional analysis of (6B). Unfortunately, this option raises more questions than it can answer. First, one might wonder why the alleged DN readings are available in denials, but excluded in plain negative statements (e.g. in (2) above).8 The second problem becomes evident when (6B) is viewed in the context of As utterance. In fact, in As utterance the same n-word niente has its typical existential import under negation, rather than receiving an NQ interpretation. Given this, there seems to be no easy way to reconcile an NQ analysis of the n-word in (6B) with its reading in (6A), without assuming a lexical ambiguity for niente. This leads us directly to the third hypothesis, that Spanish and Italian n words are lexically ambiguous between an NQ and an NPI meaning, as defended in Herburger (2001). This hypothesis shares problems with both the above alternatives. For instance, the differences between NPIs and postverbal occurrences of n -words are obviously problematic for this approach as well. In addition, it has some drawbacks of its own. While the alleged NPI- interpretation is limited to negative contexts by its polarity sensitive nature, NQ readings should in principle be possible all over the place, contrary to the facts. An independently justified criterion governing the distribution of the two readings is needed. The only articulated solution to this problem we are aware of is presented in Herburger (2001). Herbuger assumes that postverbal n-words have scope
contexts in (3), but they are grammatical also in the environments in (5), see 3.2. for the relevant examples. The hypothesis that n-words are NPIs fails to account for this difference. Notice that Giannakkidou (2000) more fine grained classification of polarity items provides a more adequate generalization. According to her, n-words are acceptable only in cooccurrence with averidical expressions. Roughly, a propositional operator Op is averidical iff it support the following entailment: Op (p) ~ p, namely sentential negation or stronger functors. What needs to be explained is why N-words should be subject to the stronger requirement of being in the presence of averidical operators. This is in a sense an aspect of what Ladusaw (1979) refers to as the sensitivity problem, i.e. the problem of deriving the restrictions on polarity NPIs from independent properties. The proposal in this paper attempts to address this problem at least as far as n -words are concerned. 8 A possible response to this criticism (offered in Herburger 2001) is that Double Negatives are convoluted constructions, difficult to process and rare in every language, which can very well be governed by pragmatic restrictions confining them to cases like denials.

narrower than an existential quantifier ranging over events. In the case of object n-words expressing the theme of the event, one can read her account as relying on the distinction between predicates like eat, which semantically require the existence of a theme and those like look, which do not impose this requirement. The former class cannot tolerate NQ postverbal n-words. A sentence like Maria ate n-sandwich would be true if there was an event of eating by Mary where nothing was eaten. This is a contradiction as long as every eating event is an event where something is eaten. The latter can tolerate postverbal n-words, because, in fact, there are events of looking where nothing is looked at. Given this, the specific proposal Herburger puts forth fails to make sense of the facts in (6) as well. In fact, while the assumptions of a NPI-NQ ambiguity in principle, would allow different readings for the n-words in As and Bs utterance in (6), Herburgers account actually predicts that with predicates like eat an NQ reading of the object niente should be unavailable. 9 This is sufficient to exclude the possibility extending Herburgers analysis to the facts in (6). 10 On the basis of these considerations, we can conclude that at the moment there is no ambiguity-approach to n-words that provides a straightforward account of all the facts discus sed in the introduction. 3 The proposal. We said that a theory of n-words for the languages at stake should account (i) for their apparent negative interpretation in sentence initial position (as in (1)), (ii) for the peculiar restrictions on their non-initial occurrences ((shown in (2), (3) and (5)) and (iii) for the special interpretation they can trigger when used in denials (6). In this section we propose an analysis of n-words and show that it accounts for (ii). Section 4 shows that it naturally accounts for (iii) as well, and section 5 that it can be extended so as to account for (i). 3.1 N-words are existentials This is our proposal: we take all instances of n-words to be non-negative existential quantifiers that contribute a negative conventional implicature. As illustrated in (7), n-words are equivalent to non-negative existential quantifiers at the truth-conditional level, but contribute negative existentials by means of a conventional implicature.
9

We would like to thank Irene Heim for pointing out this prediction to us. Herburger discusses alleged DN readings of denials with n-words in object position of look at. These cases are reminiscent of our example in (6). However, they are responses to utterances (like i) where the object n-word is supposed to be interpreted as NQs. (i) A: Il bebe sta guardando nessuno (= the baby is looking at n-body) (ii) B: Il bebe non sta guardando nessuno. (= the baby not is looking at N-BODY) Besides the fact that the grammaticality status of (i) is at least very controversial, in our case in (6), As utterance contains a non-negative occurrence of niente in object position, which is licensed by the presence of the higher negation. It is the possibility of Bs utterance to be used as a denial of this particular structure what we aim to understand.
10

(7)

For every P of type <e,t> a. [[Nadie/nessuno]] (P) = x[person(x) & P(x)] and b. [[Nadie/nessuno ]] (P) (conv.) implicates: ~x[person(x) & P(x)]

The proposal predicts that n-words will be felicitous in the scope of negation or of some other averidical operator (see ft. 7 for a definition). Consider the sentences in (8). According to our analysis, they assert that somebody came and implicate that nobody came. The absence of sentential negation provokes a clash between the truth-conditions of the sentence and its implicature. The clash renders the sentences unusable. (8) a. * venuto nessuno. b. *Vino nadie Came n-body Assertion: Somebody came Conventional Implicature: Nobody came (Italian) (Spanish)

Consider, in contrast, the sentences in (9). According to our analysis, the sentences in (9) assert that nobody came, because the n-word is under the scope of sentential negation. They also implicate that nobody came. In this case, there is no clash between the truth conditional and the non truth-conditional components of meaning.11 (9) a. Non venuto nessuno. b. No ha venido nadie. Not has come n-body Assertion: Nobody came Conventional Implicature: Nobody came (Italian) (Spanish)

We want to propose that the distribution of non sentence-initial n-words derives from the fact that they render the sentence they appear in unusable, unless they are interpreted under the scope of negation or some other averidical expression. Something else, along these lines, will have to be said about their sentence-initial occurrences (see section 5). Now we want to show that our analysis predicts the full range of contexts that license non-initial n-words.

11

In most (if not all) cases where n-words are acceptable, the implicature introduced by the nword is either equivalent or entails the truth-conditions of the sentence. This is why we believe that this component of the meaning of n-words cannot be seen as a presupposition, in the sense of precondition on the utterance context, but more neutrally as an implicature of a conventional type. Like other instances of conversational implicatures, such an implicature exhibits all the projection properties of run of the mill presuppositions, as we will see below (Karttunen and Peters 1979).

3.2 The distribution of non-initial n-words Our analysis predicts that if the combination [[n-word]] (P) is negated in the semantics the truth conditions and implicatures of the sentence would be compatible. This turns out to be a welcome prediction, beyond the contrast between (8) and (9). Recall that n -words are not only licensed under the scope of sentential negation, but also in before clauses, without clauses and the complement of predicates like doubt: (10) a. Maria usc senza salutare nessuno. 'Mary left without greeting anyone' b. Maria usci' prima di aver salutato nessuno 'Mary left before greeting anybody' c. Dubito che Maria abbia visto nessuno. 'I doubt that Mary saw anyone' (Italian)

In fact all these contexts can be analyzed as involving in their semantics a negation. They are averidical, in Giannakidous sense (see Giannakidou 1997, 2002 and ft.7). Ta ke the sentences in (11). They assert that when Mary left at a certain time t and that, at t, she had not greeted anybody. According to our analysis, they implicate that Mary had not greeted anybody. There is no conflict between the truth-conditions of the sentences and its implicature. In fact, since p before q entails when p not yet q, the implicature that n-words contribute will not clash with the truth-conditional component and we correctly predict n-words to be licensed in the scope of before. 12 (11) a. Maria usc prima di aver salutato nessuno. b. Mara sali antes de haber saludado a nadie. 'Mary left before greeting anybody' (Italian) (Spanish)

We can make similar remarks for without clauses. Without p can be analyzed as and not p. Take the examples in (12). They assert that Mary left and at that point she didn't greet anybody. They implicate that Mary didnt greet anybody. There is no conflict between what is asserted and what is implicated. Consequently, n-words are licensed under the scope of without. (12) a. Maria usc senza salutare nessuno. b. Mara sali sin saludar a nadie. 'Mary left without greeting anybody' (Italian) (Spanish)

12

To predict n-words to be fine in before-clauses w e do not need a decompositional analysis of before as when ...not in the syntax. The presence of the implicature is compatible with anything that would yield the semantic effect that before Mary greeted n-body entails that Mary had greeted nobody at the evaluation time of the main clause, no matter how this result is achieved. This entailment is indeed generally supported by those before-clauses where n-words are felicitous.

Finally, doubt that p can be analyzed as think (or believe ) that not p. The sentences in (13) assert that Mary thinks that you can't help anybody and, given the projection properties of predicates like think and believe (see Heim 1992), they implicate that Mary thinks that you cannot help anybody. There is no clash between what is asserted and what is implicated and, consequently, n -words are licensed. (13) a. Maria dubita che tu possa aiutare nessuno. b. Mara duda de que t puedas ayudar a nadie. 'Mary doubts that you can help anybody' (Italian) (Spanish)

So far we have focused on the fact that the implicature associated with n words can clash with the truth-conditions of the sentence it occurs in. In fact, we predict that it can also clash with other non-truth conditional components of the meaning of the sentence they appear in. To the extent that it does, we will have further evidence of its presence. Recall that there are contexts where NPIs like any are licensed and n-words are not: (14) a. * Meno di tre studenti hanno mangiato niente. Less then three students have eaten n-thing b. ?? Ogni passeggero che avesse bisogno di niente, fu assistito. Every passenger that needed n-thing, was assisted c. ?? Se Maria si accorgesse niente, sarebbe un problema. If Mary noticed n- thing, it would be a problem d. ?? Solo Maria ha visto nessuno degli studenti. Only Mary has seen n-one of the students e. * Mi spiacerebbe che tu vedessi nessuno. I would be sorry that you saw n-body f. * Mi sono pentito di aver incontrato nessuno. I regrette d to have met n-body

Let us start by singling out the case of factives like surprise, regret and only. Von Fintel (1999) convincingly argues that, once we take into consideration their presuppositions, these predicates can be shown to ha ve a property very similar to Downward Entailment, Strawson-DE, which enable them to license NPIs, like un dito or un dedo (a finger) in (15). Nevertheless, as the contrast between (15) and (16) shows, they do not license n-words: (15) It: a. Mi sorprende che tu abbia alzato un dito per aiutarmi. Sp: b. Me sorprende que hayas movido un dedo por ayudarme. ' I am surprised that you lifted a finger to help me' a. *Mi sorprende che tu abbia/ha i visto nessuno. (It.)

(16)

b. *Me sorprende que te haya visto nadie. To me surprise:3s that you had:subj/ind seen n-body

(Sp.)

Our proposal provides an explanation for the contrast. Factive predicates like surprise and regret trigger the presupposition that its complement is believed to be true by the holder of the attitude (see Heim 1992). So, in this case, (16a) presupposes that the speaker believes that you saw somebody. The n -word in the complement of surprise projects the conventional implicature that the speaker believes that you saw nobody, thus conflicting with the factive presupposition of surprise. The sentence is rendered unusable and, consequently, the n-word is disallowed. The same line of reasoning applies to the contrast between (17) and (18). N-words are not licensed by regret because the implicature that they trigger systematically conflicts with the factive presupposition of regret. (17) a. A Maria dispiace di aver alzato anche solo un dito per aiutarmi. b. Mara lamenta haber levantado un slo dedo para ayudarme. Mary regrets having lifted a finger to help me a. * A Mary spiace che tu abbia/hai visto nessuno b. * Mary lamenta que t hayas visto a nadie Mary regrets that you have (ind./subj.) seen n-body (It.) (Sp.) (It.) (Sp.)

(18)

The reader will have already guessed how to explain the cases with only. Again, as the contrast between (19) and (20) illustrates, only licenses NPIs, but it does not license n-words. Only triggers the presupposition that Mary saw some student (Horn 1969), which conflicts with the implicature that the n-word contributes. i.e. that every individual in the domain (including Mary) saw no student. 13 (19) a. Solo Maria ha alzato un dito per aiutarmi. b. Slo Maria ha movido un dedo para ayudarme. Only Mary lifted a finger to help me a. * Solo Maria ha visto nessuno studente. b. * Slo Maria ha visto a ningn estudiante Only Mary saw n-one student (Italian) (Spanish) (Italian) (Spanish)

(20)

In sum, the implicature that n-words contribute can conflict not only with the assertion of the sentence, but also with its presuppositions, as expected.
13

In order to establish how the conventional implicature of the n-word projects in this environment we assumed, for simplicity, a generalized quantifier semantics of only+NP (see also von Fintel 1999) and then extended to this case Heims (1988) analysis on how presuppositions project in quantificational structures. However, that the imp licature of niente conflicts with the factive presupposition of only can be shown to follow from other analyses of only as well.

To end our discussion about the distribution of non-sentence initial nwords, consider the occurrences of n-words in the antecedent of conditionals and in the restrictor of universals. (21) illustrates that n-words in these contexts render the sentences they occur in somewhat infelicitous. (21) a. ?? Se Mar ia avesse visto nessuno, sarebbe un problema. If Mary had: subj. seen n -body, it would be a problem b. ?? Ogni passeggero che avesse bisogno di niente, verra' assistito. Every passenger that need:subj n-thing, will be assisted.

Our analysis can derive these facts. Assume any run-of-the mill analysis of conditionals where the antecedent helps determining the domain of quantification of the modal in the consequent. A sentence like (21a) is true in a world w, say, if all worlds most similar to w where the proposition expressed by the antecedent is true are worlds where the proposition expressed by the consequent is also true. Now, in (21a), the antecedent-worlds are restricted, by the implicature, to those where Mary sees n obody. The truth conditions require that all these worlds be worlds where Mary saw somebody. The set of worlds where Mary saw somebody and nobody is empty. The infelicity comes from expressing a conditional with a vacuous restrictor. A similar story can be told about (21b). The n-word triggers the implicature that every one of the passengers being quantified over needs nothing. The sentence asserts that every passenger who needs something will be assisted, thus requiring that the passengers being quantifie d over need something. Again, its strange, though not impossible, to express a universal generalization over an empty set. This, again, explains why the sentence is highly marginal, although not completely ungrammatical. We started the section by claiming that a theory of n-words for the languages at stake should account (i) for their apparent negative interpretation in sentence initial position, (ii) for the peculiar restrictions on their non-initial occurrences and (iii) for the special interpretation they can trigger when used in denials. We have shown that it is possible to give an account of (ii) by assuming that n-words are existential quantifiers associated with a negative implicature. Before saying something about (i), we will show in the next section how to account for (iii). 4 Denials as cases of metalinguistic negation Recall the dialog in (6), repeated below: (22) A: Maria stara' morendo di fame, non ha mangiato niente tutto oggi. Mary will be starving , not has eaten n-thing all day 'Mary is probably starving, she hasn't eaten anything all day' B:Non ha mangiato NIENTE, ha mangiato un panino! Not has eaten N-THING, she ate a sandwich!

'It's not correct that she didn't eat anything: she ate a sandwich!' Within A's utterance, the negative word niente is interpreted as an existential quantifier under the scope of sentential negation. However, Bs remark is a denial of the content of As utterance, even when both A and B use the same sentence, containing the same number of negations and negative words. We need to explain where the additional negation expressing disagreement come from and what exactly B is disagreeing with. Our analysis can handle this puzzle. We take the cases of denials with n-words, as in (22), to be cases where the speaker expresses disagreement with the implicature triggered by the n-word in a previous utterance. In the case at hand, B expresses disagreement with the implicature that Mary ate nothing, as her final remark illustrates. In other words, the negation in Bs utterance is an instance of external or metalinguistic negation. A speaker using negation externally in a sentence S needs not to deny the truth-conditional content of S, but denies the appropriateness or correctness of the utterance 'S'. Disagreement with a presupposition or with an implicature of S is one common reason why a speaker can say that it is not appropriate or correct to utter S. (Horn 1985). The sentences in (23) illustrate this phenomenon. Somebody can utter (23a) to express disagreement with the presupposition triggered by manage that solving the problem was somehow difficult for Mary. The sentence in (23b) can be uttered to express disagreement with the scalar implicature triggered by the expression warm, which conveys the information that the speaker thinks it is not hot. (23) a. Mary didn't MANAGE to solve the problem, it was easy for her. b. It is not WARM, its HOT.

At this point the reader might wonder whether there is any independent justification to assume that the meaning of negation in (22) and (23) is the same, i.e. metalinguistic. Horn (1985) provides the following diagnostic for metalinguistic negation. He observes that there are two uses of but, concessive and non-concessive, and that non-concessive but is excluded in external negation contexts. The distinction between concessive and non-concessive but is overtly realized in Spanish (see Horn (1985)). Spanish pero is non concessive but and sino (which conjoins NPs) is concessive. The distribution of pero and sino provides, indeed, one piece of evidence for the metalinguistic nature of negation in (22B). As expected if these are cases of metalinguistic negation, Spanish pero is excluded in (25B), where sino is perfectly fine: (24) (25) A: B: B': Mara no Mary not Mara no Mary not #Mara no est mirando nada. is looking at n-thing est mirando NADA, sino a nadie is looking at n-thing, but at n-body est mirando NADA, pero no est mirando a nadie

Mary

not

is looking

at n-thing,but not is looking at n-body

The mere existence of cases like (22) and (25) argues for the fact that n-words trigger a certain implicature that metalinguistic negation can target. 5 Preverbal n-words and abstract negation In this section we finally return to our question (i), concerning the case of sentence initial n-words like those in example (1), repeated below: (26) a. Nessuno venuto. b. Nadie ha venido. 'Nobody came' (Italian) (Spanish)

The problem these cases present to our analysis is twofold: if n-words are unambiguously existential quantifiers introducing the negative implicature described above what license their presence in sentences like (1) and what makes these sentences negative? In order to extend our proposal to cases like (1) an additional stipulation is required: that these sentences contain an abstract negative morpheme. This assumption was already proposed in Laka (1990) and Ladusaw (1992). However we will partly depart from Ladusaw and Lakas proposals for reasons that will become clear shortly. Laka and Ladusaw suggest that preverbal n-words are licensed by an abstract negation located in the syntax in the same position as the overt negative marker. We instead believe that the abstract negation licensing preverbal n-words is higher in the structure than the overt negation. Lets first illustrate the main features of our idea and then conclude by highlighting its possible advantages over Ladusaw and Lakas proposal. One possible implementation of our proposal can be formulated within Rizzis (1997) syntax of the left periphery. The idea is that one of the heads constituting the CP, perhaps Foc, can sometimes host a semantically active but phonologically unrealized negative feature (say [neg]) that requires an overt specifier with matching morphology. Although truth-conditionally inert, the negative morphology of negative words enables them to satisfy this requirement. The case can be seen as totally parallel to that of wh-questions, where a silent but semantically active morpheme in a question complementizer requires in many languages (including Italian and Spanish) an overt specifier with wh-morphology. Given this, the n-word in (1) moves overtly in Spec-FocP to legitimate an abstract negation, as shown in (27): (27) [ FopP Nessuno [ Foc [neg] [ IP t1 [ .

In the structure in (27), the formal requirements of the covert negation are satisfied, but the semantic requirements of the n -word are not. N-words, due to

their implicatures, are felicitous only when interpreted in the scope of negation. Therefore, at LF the n-word nessuno reconstructs to its base position, generating the observed ~ reading. One of the advantages of assuming that preverbal N-words moved in the overt syntax is that we can understand cases like (28a) and (28b) as cases of bare argument ellipsis, where ellipsis applies to structures similar to (27):14 (28) a. Chi hai visto? Nessuno. Who did you see? Nobody b. Voglio vedere te o nessuno. I want to see you or nobody (Italian)

The apparent negative meaning of n-words in elliptical answers like (28a) and in coordinated structures, like (28b) has often been held as an argument against the hypothesis that they are unambiguously existentials or indefinites (see Zanuttini 1990 and Herburger 2001). The negative meaning can, however, be accounted for by the presence of the [neg] feature if these two cases are derived as in (29): (29) a. [ FocP Nessuno1 [Foc [+neg] [IP ho visto t1 .]] b. Voglio vedere te o [ FocPnessuno1 [[Foc [+neg] [IP voglio vedere t1.].

Evidence that (29a) is the structure from which (28a) is given in (30): the non elliptical version of (29a) in (30a) is fine while (30b) is ungrammatical: (30) a. Chi hai visto? Nessuno1 ho visto t1. b. Chi hai visto? *Ho visto nessuno.

Similarly, evidence that (29b) is the most plausible analysis of (28b) comes from the fact that (31), where licensing by abstract negation is unavailable, is ungrammatical: (31) * Voglio ballare con nessuno o (al pi) con te. I want to dance with n-body or (at best) with you.

Finally, the assumption that the abstract negation is located in a different and higher position then the overt one accounts for the DN reading of examples like

14

See also Giannakidou (2000) for more arguments in favor of a bare-elliptical analysis of answers involving n-words in isolation and n-words in disjunctions.

(32), also unnecessarily taken to be evidence in favor of an inherent negative semantics of n-words in this position in Herbuger (2001). (32) Nessuno non ha chiamato. N-body didnt call. Nobody didnt call ( = Everybody called)

In fact, under the assumption made in this section, (32) contains two negations, one covert in FocP licensed by the presence of the n-word in its specifier and one overt negative marker non, hence the DN reading. (33) [ FocP Nessuno 1 [Foc [+neg] [IP t1 [non [ ha chiamato]]]]]

6 Conclusions We have proposed that n-words are non-negative existential quantifiers that contribute a negative conventional implicature. They behave like existentials at the truth conditional level, but also like NQs when we consider the implicatures of the sentences that host them. The main advantage of our proposal against previous ones is that it explains a wider range of facts. From the assumption that the interpretation of nwords involves a conventional implicature, we derive both their restricted distribution (distinguishing it from the apparently related distribution of NPIs) and also their uses in the contexts of denials. The negative conventional implicature that n-words contribute derives the restricted NPI-like distribution via interaction with the truth-conditions of the sentence. This is an advantage over the hypothesis that n -words are uniformly NQs, which has no straightfoward account to derive the restriction distribution. We have also shown that the distribution of n -words is more restricted than the distribution of run-of-the-mill NPIs in that Strawson downward entailigness is not a sufficient licensing condition. Strawson-DE factive environments do not license n-words, because the factive presupposition conflicts with their conventional implicature. This is a clear advantage over the NPI approach. Our approach shares with the NPI approach the need to postulate an abstract negation that licenses the sentence initial occurrences of n-words. However, our assumptions about the abstract licensor are free of some criticisms that the NPI approach raises: they provide an account for the fact that n-words are fine in the absence of an overt licensor when used as elliptical answers and in disjunction and that they generate DN readings when followed by negation. A major advantage of our proposal is that the negative conventional implicature explains the use of sentences containing n-words in the context of denials. Under our assumptions, uttering a sentence like Mary didnt see n-body implicates tha t Mary saw nobody. Metalinguistic negation expresses disagreement with the appropriateness of a certain utterance. Frequently, it expresses

disagreement with the implicatures that a certain utterance trigger. It is only natural that it can be used to express disagreement with the implicatures triggered by n -words. A speaker can use Mary didnt see n-body to express disagreement with the implicature that Mary saw nobody, hence their use in the context of denials. Our explanation does not take the uses of n -words in denials to be DN statements and so is free from the concerns that such approaches raise. Since we take these statements to be cases in which a speaker uses negation to target the negative implicature of the n -word, it does not come as a surprise that these readings are excluded in plain negative statements, where negation targets the truth conditions of the sentence. Acknowledgments We owe many thanks to Danny Fox, Jon Gajewski, Martin Hackl, Irene Heim, Angelika Kratzer and Utpal Lahiri for their generous help and extremely precious comments. For useful discussion, we would also like to thank all the participants in the Seminar on Negation in Amherst, where the project for this paper started. We also want to thank the participants of the syntax-semantics reading group at MIT, where we had a chance to discuss it further. Thanks to the organizers and the audience of CLS38. Last but not least the authors would like to thank each other for this pleasant coauthoring experience.
References von Fintel, Kai. 1999. NPI-licensing, Strawson-entailment and context dependency. Journal of Semantics 16:97-148. Giannakidou, Anastasia . 1997. The Landscape of Polarity Items. Ph.D. Dissertation, Groningen. Giannakidou, Anastasia. 2000. Negative...concord? NlLLT 18:457-532. Heim, Irene. 1988. On the projection problem for presuppositions. Proceedings of the Second WCCFL : 114-125, Stanford CA: Stanford University Press. Heim, Irene. 1992. Presupposition projection and the semantics of attitude verbs. Journal of Semantics 9:183-221. Herburger , Elena. 2001. The negative Concord puzzle revisited. NL S 9:289-333. Horn, Laurence, R. 1969. A presuppositional approach to only and even. Proceedings of the CLS 5:98-107 Horn, Laurence., R. 1985. Metalinguistic negation and pragmatic ambiguity. Language 61:121-175 Karttunen, Lauri and Peters, Stanley. 1979. Conventional implicature. Syntax and Semantics 11: Presupposition, 1-55. Ladusaw, William. 1979. Polarity Sensitivity as Inherent Scope Relation. Ph.D. thesis, Austin. Ladusaw, William. 1992. Expressing Negation. SALT 2:220 -229. Laka, Itziar, M. 1990. Negation in syntax: On the nature of functional categories and projections. Ph.D. Thesis, MIT. Rizzi, Luigi. 1997. The fine structure of the left periphery. Elements of Grammar, Lilijane Haegeman ed., Kluwer Publications, Dordrecht. Zanuttini, Raffaella. 1991. Syntactic properties of sentential negation: A comparative study of Romance languages. Ph.D. dissertation, University of Pennsylvania. Zwarts, Frans. 1993. Three types of polarity. Semantics , Hamm, F. and Hinrichs, E. eds.