2. Collins and Postal (Reference Collins and Postal2014)

We assume there is a notion of syntactic negation, whose representative is NEG. Unlike most linguistic studies of negation, with their common appeal to ‘sentential negation’ (and now frequently a clausal NegP), we posit that NEG can syntactically modify a large range of categories including AdjPs, NPs, Ds, and in general any category with predicational semantics (see CP2014 and Collins Reference Collins2016). A category has predicational semantics if it denotes a function whose semantic type ends in t. For example, a DP like everybody has the type <<e,t>,t>, and so counts as a predicate. A fundamental assumption of the present framework is that while a NEG may raise from its position of origin, it is always and only interpreted as modifying its sister constituent in the original locus. The reader is referred to CP2014 (chapters 3 and 5) for further discussion of NEG interpretation.

According to CP2014, each NPI originates in a construction which obligatorily contains a syntactic negation, and some originate in a construction with two, yielding the partially contrasting structures in (3).

1. (3) Two Types of NPIs

   1. a. [NEG X] unary NEG NPIs

   2. b. [NEG [NEG X]] binary NEG NPIs

We can call the proposal that each NPI originates with at least one instance of NEG, the negative conception of NPIs, and the related proposal that there are the two types in (3a–b) the bifurcated (negative) view. Standard views of NPIs do not adopt either of these assumptions, as we review in section 3.

Consider from this point of view the examples in (4).

1. (4)

   1. a. I didn't advocate any proposal.

   2. b. No linguist accepted any proposal.

   3. c. Elissa doesn't do anything on Sunday.

   4. d. Elissa doesn't go anywhere on Sunday.

2. (5)

   1. a. Not many people advocated any proposal.

   2. b. If he accepted any proposal, he was mistaken.

   3. c. Everyone who did anything on Sunday was praised.

   4. d. Everyone who went anywhere on Sunday got rained on.

In the framework of CP2014, the NPIs in (4) have a structure and meaning identical to the structure and meaning of negative existential quantifier DPs, accounting for the truth-conditional equivalence of pairs like the one in (6).

1. (6)

   1. a. I advocated no proposal. ¬∃x[proposal(x) ∧ advocate(I, x)]

   2. b. I didn't advocate any proposal. ¬∃x[proposal(x) ∧ advocate(I, x)]

In particular, CP2014 analyzes both no proposal and any proposal in (6) as DPs of the form [[NEG SOME] proposal], where NEG modifies an existential quantifier SOME. The differences between (6a) and (6b) lie in the assumption that in (6a), SOME is realized as null and NEG is spelled out as no, while in (6b), NEG raises to the post-Aux position and SOME is spelled out as any (see rule (8) below).Footnote ²

In these terms, a fuller structure of (4a) is given in (7b).

1. (7)

   1. a. I didn't advocate any proposal.

   2. b. I did NEG₁ advocate [[<NEG₁> SOME] proposal]

      Raising of NEG₁

In (7b), NEG₁ originates in a position modifying SOME (internal to the NPI any proposal). While NEG₁ raises to the post-Aux position, it is nonetheless interpreted in its position of origin, modifying SOME. The angled brackets around the lower occurrence of NEG₁ in (7b) indicate a non-pronounced occurrence. This notation is used throughout when appropriate. The reader is referred to CP2014 (chapters 3 and 5) for further discussion.

Because the NPIs in (4) are analyzed as having the structure [[NEG SOME] NP], they are called unary NEG NPIs in CP2014. As seen below, they contrast with binary NEG NPIs, which contain two underlying NEGs. We will return to examples like (4b), which involve no post-Aux NEG, in the discussion of (13) below. These will also be analyzed as unary NEG NPIs.

CP2014 claim that any in (7a) is a suppletive form of SOME, determined by the rule in (8).

1. (8) SOME/any Mapping (CP2014: 21)

   1. a. SOME → any, in the context [<NEG> __] (NEG unpronounced)

   2. b. SOME → null, in the context [NEG __]  (NEG pronounced)

   3. c. SOME → some, otherwise

Since CP2014 analyze nobody as [[NEG SOME] body], the rule in (9) is also required for the realization of NEG.

1. (9) NEG Mapping

   1. a. NEG → no in the context [_D __ [_D <SOME>]]

   2. b. NEG → not, otherwise.

The statement in (9b) is a simplification, since there are other forms of NEG (such as un-, non-, n't-, n-) not discussed in this article.

The NPIs in (5) are analyzed in CP2014 as double-negation structures. Consider again (5b), repeated as (10):

1. (10) If he accepted any proposal, he was mistaken.

First, no overt NEG precedes the verb in (10), unlike the situation with unary NEG NPIs illustrated in (4a). Second, the interpretation of any proposal in (10) is equivalent to the existential quantifier DP some proposal. CP2014 argues that the NPI any proposal in (10) has the double-negation form shown in (11). In such a structure, the semantics of NEG₁ cancels that of NEG₂, so that the resulting interpretation is identical to that of some proposal.

1. (11) If you see [[<NEG₁> [<NEG₂> SOME]] proposal], tell me.

A binary NEG NPI (also called a reversal NPI) such as that in (11) contains two unpronounced NEGs. We sometimes refer to NEG₁ as the outer or reversal NEG. According to CP2014, the NEGs in (11) are unpronounced because they are deleted. NEG deletion involves a relation between individual NEGs and other phrases, their NEG deleters. Some relevant NEG deleters in the case of binary NEG NPIs are listed in (12).

1. (12) Some NEG Deleters in English Binary NEG Structures

   1. a. The conditional complementizer if

   2. b. The yes-no question complementizer (the Q morpheme)

   3. c. verbs such as surprise

   4. d. the quantifier every

   5. e. phrases of the form: [only DP]

In (11), the conditional complementizer if deletes the NEG₁ of the structure [NEG₁ [NEG₂ SOME]], while NEG₁ deletes NEG₂. Because NEG₂ is deleted, SOME is realized as any by rule (8a). See CP2014 (chapter 4) for further detail.

The distinction between unary and binary NEG NPIs corresponds to the traditional distinction drawn between strong and weak NPIs, and between strict and non-strict NPIs (see CP2014: section 9.4 for discussion). In general, weak NPIs are licensed in downward-entailing contexts, although there are some complications. For example, yes-no questions and the scope of only-phrases are not, by standard definitions, downward-entailing contexts (on only-phrases, see von Fintel Reference von Fintel1999). Addressing these questions goes beyond the scope of this article.

Return to example (4b), repeated in (13).

1. (13) No linguist accepted any proposal.

From the perspective of CP2014, a natural question is whether any proposal in this example is a unary or a binary NEG NPI. CP2014 advance several reasons to believe that any proposal in such cases is a unary NEG NPI; see the sections below on the distribution of jackshit _A and Horn clauses. On that assumption, (13) has the structure shown in (14).

1. (14) [[NEG SOME] linguist] accepted [[NEG SOME] proposal]

Whether the subject or object DP is taken to have higher scope in (14), the normal result, given the occurrence of two distinct NEGs, would be a so-called double-negation meaning. In the case of two negative existentials, this should give the interpretation that every linguist accepted some proposal. Since it does not, something in the posited overall structure must prevent this incorrect inference.

CP2014 (chapter 6) appeals to the semantic notion of polyadic quantification, while adding a novel syntactic basis for it. The latter involves the idea of the sharing of Ds, so that in such terms the subject and the object of (13) would have a representation like (15). Previous work invoking polyadic quantification for the description of negative concord (NC) includes de Swart (Reference De Swart1999) and de Swart and Sag (Reference De Swart and Sag2002).

1. (15) [_DP [NEG₁ SOME₂] NP]…… [_DP [NEG₁ SOME₂] NP]

In other words, while the subjects and objects involve distinct DPs with distinct NP components, these DPs share the identical D [NEG₁ SOME₂]. Critically, since the two distinct DPs share the D containing NEG₁, and since it is such structures which are semantically interpreted, the problem of the cancellation of the negative force of distinct NEGs simply does not arise.

A key element in the analysis just sketched is that the NEG occurring in any proposal in (13) is deleted. In the framework of CP2014, this deletion appeals to a second kind of NEG deletion (distinct from that seen above with binary NEG NPIs), one subject to a number of governing conditions discussed in detail in CP2014 (chapter 8). One of these conditions is that the first argument of the NEG-deletion relation must c-command the second argument.

A mild warning is in order here. While we use terms like ‘NPI’ and ‘licensing’, this is for descriptive convenience and to make contact with the standard literature. But such usages must be considered a rough disposable shorthand for the more complex structures and relations we take to be involved in this domain, including NEG raising and NEG deletion.

One important consequence of the bifurcated view is that every proposed analysis of any so-called NPI occurrence involves a decision as to whether to posit a unary or a binary NEG structure. Moreover, some morphologically defined NPIs might, in different contexts, allow either analysis. We take this to be the case with English any and ever. We return to this issue in section 6.

3. The Consensus View of NPIs

Although there are of course many different approaches to NPIs, it seems fair to state that current mainstream views of them share a core set of features. Focusing on NPIs such as English any forms, such expressions are almost always taken to be simply indefinites or existentials, as indicated in (16).

1. (16)

   1. a. “The three arguments presented here in support of the E-theory (in which polarity any is represented as an existential quantifier in the scope of NOT)…”

      Linebarger (Reference Linebarger1980: 217)

   2. b. “Any is an existential quantifier which must appear within the scope of negation.”

      Carlson (Reference Carlson1980: 800)

   3. c. “The NP with any has the usual semantic properties of an indefinite. We won't make a choice here concerning the proper way of treating indefinite NPs. If indefinite NPs are best regarded as existential quantifiers, then so is the any NP; if indefinite NPs are best treated as new variables (Heim Reference Heim1982), then the any NP too is a new variable. From either assumption the existential behavior of PS any follows without problems.”

      Kadmon and Landman (Reference Kadmon and Landman1993: 357)

   4. d. “Since Klima Reference Klima, Fodor and Katz1964, these NPIs have been termed as ‘indefinites’, and it is best to interpret this term in the sense of Heim 1982… Argument NPIs like any and ever can be treated as indefinites which are subject to twin licensing requirements, …”

      Ladusaw (Reference Ladusaw, van der Does and van Eijck1995: 211)

None of (16a–d) recognize that NPIs like any or ever forms involve any instance of NEG. In current views, syntactic negation plays a role only in the formation of other expressions which can serve as licensers of NPIs. Let us call such views the NEG-free conception (of NPIs).

The NEG-free conception is obviously motivated by the fact that in general, NPIs do not appear overtly with instances of NEG. But in any framework which recognizes NEG raising and deletion, the absence of some overt X internal to a constituent Y is anything but a definitive justification for the syntactic absence of X from Y.

Statements (16a–d) and many similar ones reveal their authors’ assumption that not only are the relevant NPIs nonnegative expressions, but also that any NPIs appearing in various contexts always have the same structure (namely, that of an indefinite/existential DP). None of these authors question whether there might be distinct types of NPI any (putting aside the issue of free choice any). Analogous remarks hold for ever. Let us call the assumption that there is a unique NPI type found in the range of environments occupied by English NPI any and ever the homogenous view. Thus currently common views of English any and ever forms assume a NEG-free, homogenous view of these expressions.

4. Resumptive NEGs

The view that the notion of NPI dissolves into a complex of analyses involving NEG raising, NEG deletion and morphophonemic modifications receives some support from simple facts about nonstandard English; see Blanchette (Reference Blanchette2015) for a much more detailed study. Consider the following triple, where (17a–b) represent standard English, and (17c) some variant of nonstandard English.

1. (17)

   1. a. I punched nobody.

   2. b. I didn't punch anybody.

   3. c. I ain't punched nobody.  (books.google.com/books?isbn = 1462819346)

For us, the ideal analysis gives all three object DPs the same underlying structure, shown in (18).

1. (18) [_DP [NEG₁ SOME] [_NP body]]

In standard English (17a), NEG₁ is neither raised nor deleted and SOME is then forced to be null, while NEG₁ is realized as no via the mappings in (8) and (9). In standard English (17b), NEG₁ raises and SOME is required to appear as any via mapping (8a).

Unanalyzed so far is the nonstandard case (17c). If the object DP nobody there receives the same analysis as the corresponding DP in standard English (17a), then the additional presence of the post-Aux NEG raises the obvious question of why the combination of that NEG and the NEG assumed to be part of nobody does not yield a double-negation reading. In other words, why doesn't (17c) mean ‘I punched somebody’? Commonly, approaches to cases like (17c) invoke some notion of negative concord, not recognized for standard English. But analyzing closely related variants of a language as differing in something as basic as the presence of negative concord is, we believe, suspect.Footnote ³

In the overall framework of C2014, these issues are addressed by appealing to one further concept relevant to the syntax of negation, that of a resumptive or copy NEG. While we will not attempt a formal characterization of that concept here, it can be thought of as the close analog of the notion of resumptive pronoun in work on extraction.

Before proceeding, it is worth clarifying the relation between copies and occurrences, which are distinguished in both CP2014 and the present article. ‘Occurrence’ is a general theoretical term needed in any system in which phrases occur in more than one position. When that is the case, statements about a phrase can be equivocal. Precision in such cases, then, must refer to occurrences of the phrases. For example, in a topicalization case like Pizzas, she doesn't eat, does the DP Pizzas precede or follow the verb? As such, the question makes no sense. We say that one occurrence follows the verb and another precedes it.

In minimalist syntax, copy is a term sometimes used as a synonym of occurrence. However, we draw a distinction between occurrences and copies both in CP2014 and here. There are at least two possible interpretations of our notion ‘copy’. In one, a copy of a syntactic object X is the overtly realized phonological spell-out of an occurrence of X, where X has more than one occurrence. On another interpretation, a copy of a syntactic object X is a distinct syntactic object associated with one of the occurrences of X. For the purposes of this paper, it does not matter which of these views is chosen.

The notion of a copy constituent is instantiated by the italicized pronoun in (19).

1. (19) Jane is the sort of person who I wonder whether she will ever finish her doctoral work.

One tradition allows that such pronominals can occur in extraction sites normally not filled with overt material, especially in cases where the extraction site is separated from the final locus of the extracted element – who, in (19) – by an island boundary.Footnote ⁴

Let us take the resumptive/copy NEG associated with any NEG_x to be represented as cNEG _x . The resumptive NEG concept is then relevant for (17c) as follows. Despite our claim that the object DPs in all three examples of (17) originate with the common structure in (18), these cases illustrate three distinct structural possibilities, two of which have already been gone over. The notion of a copy/resumptive NEG provides the third analysis, relevant for nonstandard (17c). There, NEG₁ raises out of its containing DP and moves to the post-Aux position, as in (17b). However, in (17c), NEG raising leaves a resumptive/copy NEG in the original position of NEG₁, resulting in the structure:

1. (20) I aux NEG₁ punch  [_DP [cNEG₁ SOME]] [_NP body]]

   |____________________|

   NEG raising + copying

Notably, (20) does not satisfy condition (8a) but does satisfy (8b). Hence it also satisfies (9a). Thus the object DP in (20), despite its partial structural contrast with the corresponding DP in (17a), is also spelled out as nobody.

Consider now the subjects in (21), which involve the same three DP types seen in (17).

1. (21)

   1. a. Nobody showed up.

   2. b. *Anybody didn't show up.

   3. c. And nobody ain't showed up here from FEMA…   (http://transcripts.cnn.com/TRANSCRIPTS/0509/12/asb.01.html)

Here, (21a–b) are representations of standard English, and (21c) an instance of nonstandard English.Footnote ⁵ For (21a), the subject form nobody is the expected result when NEG₁ neither raises nor is deleted. For (21b), the NEG has raised and the shape of the remnant subject is the same as that of the object in (17b). But for reasons so far unspecified, the overall result is ungrammatical.

What accounts for the contrast between the ungrammatical standard English (21b) and the grammatical nonstandard (21c)? Looking at the object cases in (17), one might have assumed that nonstandard no forms are just analogs of standard English NPI any. But if that were true in general, (21c) would be parallel to (21b) and thus predicted to be ungrammatical, contrary to fact.

A key question then is why in our terms the nonstandard English subject case (21c) is grammatical, while in standard English (21b) is not, although both object cases (17a–b) are grammatical. Two sorts of contrast manifest here. One is a contrast between standard and nonstandard English. The other is a contrast between standard English subjects and non-subjects with respect to co-occurrence with post-Aux NEG, which has raised from the relevant NPI DP.

Now, in the cases at hand, it is a banal observation that English subjects both precede and c-command the Aux and the NEG associated with it, while English objects are preceded and c-commanded by the Aux and NEG. Given the subject/object contrast between (17b) and (21b), some constraint must make reference to this distinction. A second observation is that the nonstandard English subject cases have been analyzed as involving overt copy NEGs in the extraction positions of raised NEGs, while the standard English cases involve gaps we represent as <NEG>. We can then suggest the grammatical constraint in (22).

1. (22) The Remnant Raising Condition

   If M = [_DP [_D <NEG_x> SOME] NP], then no occurrence of M c-commands an occurrence of NEG_x.

Condition (22) amounts to a ban on remnant raising of DPs containing a NEG gap.Footnote ⁶

In the terms we have developed, the contrasts between standard English and nonstandard English with respect to the distribution of any and no forms just discussed reduces to the relatively minor issue of whether NEG raising leaves a resumptive NEG or not.

5. Jackshit_A

A different kind of motivation for the unary/binary NEG distinction is derivable from the fact that there are two basic classes of environments as far as English NPIs are concerned: those like No linguist ___, and those like Not many people ___. This demarcation shows up in a way not yet touched on, specifically, in the behavior of a class of English NPIs referred to in CP2014 with the cover term JACK. This class includes dick, diddly, diddly squat, jack, jackshit, shit, squat, etc., of which we cite here only jackshit. These forms have in general two uses, illustrated in (23), which are annotated as jackshit _A and jackshit _Z in CP2014.Footnote ⁷

1. (23)

   1. a. Terry knows jackshit_Z about transponders.

   2. b. Terry doesn't know jackshit_A about transponders.

The form jackshit _Z (Z for ‘zero’) functions as an equivalent of nothing or zero and does not require a licenser in the sense of common approaches to NPIs. It is of no relevance to the present discussion and is ignored from this point on. The form jackshit _A (A for ‘any’) functions as an NPI and is the equivalent of one use of anything. It is a unary NEG NPI in the sense of CP2014.

The environments in (24) and (25) are differentiated by the distribution of jackshit _A :

1. (24)

   1. a. I didn't advocate anything/jackshit_A.

   2. b. No linguist accepted anything/jackshit_A.

   3. c. Elissa doesn't do anything/jackshit_A on Sunday.

2. (25)

   1. a. Not many people advocated anything/*jackshit_A.

   2. b. If he accepted anything/*jackshit_A, he was mistaken.

   3. c. Everyone who did anything/*jackshit_A on Sunday was praised.

A rough generalization is that jackshit _A can only occur in environments containing a post-Aux NEG or a preceding negative quantifier DP defining an antiadditive operator. We claim that those any forms that occur in environments which accept jackshit _A represent unary NEG NPI structures, while any forms in environments incompatible with jackshit _A are binary NEG NPI structures.

The analysis of jackshit _A is then that it represents a unary NEG structure, as in (26):

1. (26) [_DP [_D NEG₁ [_D SOME]] [_NP jackshit_A]]

That is, we propose that members of JACK can never have the reversal structure in (3b). Compared to unary NEG any forms, however, members of JACK_A are subject to additional constraints. First, NEG₁ in (26) must either be raised or deleted, determining that no NEG occurrence internal to the original DP is overt. So JACK_A forms never have overt no forms:

1. (27)

   1. a. They proved no theorem/nothing.

   2. b. * They proved no jackshit_A.

Second, according to rule (8a) above, raising or deletion of the NEG associated with SOME forces its realization as any. But a further peculiarity is that the resulting any must be null in a DP whose NP is a member of JACK_A. So instances of the DP structure in (26) always surface with only the NP jackshit _A overt (SOME is not realized overtly, and NEG raises out of the DP).

We can then account for the ungrammaticality of the bad jackshit _A cases like those in (25), in contrast to the well-formed ones in (24). Consider the structure of the JACK_A version of (25b):

1. (28) If he accepted [[NEG SOME] jackshit_A]…

In order for this structure to be realized as in (25b), NEG needs to raise to the post-Aux position and be deleted. To block this analysis, we appeal to a version of a principle originally due to Szabolcsi (Reference Szabolcsi2004), which we called the NEG Deletion Evenness Condition; see CP2014 (p. 32, p. 75) for initial and refined versions of this principle. For present purposes, we oversimplify and say simply that standard instances of general (nonlexical) NEG deletion must lead to the deletion of an even number of NEGs. Since jackshit _A cases like (28) have only a single NEG, the Evenness Condition cannot be satisfied. We ignore here, for simplicity, other reasons that the analysis in (28) is blocked, having to do with the types of NEG deleters that can delete unary NEGs; see CP2014 (chapter 10) for relevant discussion.

The anything case of (25b) does not involve a violation of the Evenness Condition, since anything can be analyzed as a binary NEG NPI, as shown below:

1. (29) If he accepted [[NEG₁ [NEG₂ SOME]] anything/jackshit_A]…

In (29), the conditional complementizer if is the NEG deleter. It deletes NEG₁, which in turn deletes NEG₂. Since two NEGs are deleted, there is no violation of the Evenness Condition.

This raises the obvious question of how the jackshit _A case in (24a) can be grammatical. The answer is that no NEG is deleted. Rather, the unary NEG of the original JACK_A DP is raised to post-Aux position, so no violation of the NEG Deletion Evenness Condition can arise. For cases like (24b), the situation is more complex, as we take these to be instances of polyadic quantifier formation based on D sharing between no linguist and jackshit _A. While there is NEG deletion in such cases, it is not subject to the Evenness Condition. The reader is referred to CP2014 – especially chapter 6 – for detailed discussion of these matters.

6. Horn Clauses

A second argument showing that English NPIs divide into the two classes represented in (3) above is based on Horn clauses; see CP2014, chapters 13 and 14, for extensive discussion. These are complement clauses manifesting the inverted subject–auxiliary order typical of the Negative Inversion construction. The key peculiarity of Horn clauses is that where a standard Negative Inversion clause involves a fronted overt negative or decreasing phrase, the fronted phrase in a Horn clause is an NPI. Thus one finds:

1. (30) Negative Inversion Clauses

   1. a. Under no conditions would she move to Alaska.

   2. b. They believed that under no conditions would she move to Alaska.

   3. c. None of the cats could they train to play ping pong.

   4. d. They expected that none of the cats could they train to play ping pong.

2. (31) Horn Clauses

   1. a. They didn't believe that under any conditions would she move to Alaska.

   2. b. They didn't expect that any of the cats could they train to play ping pong.

   3. c. I don't suppose that even a red cent would he contribute.

   4. d. It doesn't seem that even the slightest deviation from the plan would she accept.

One hypothesis might be that a Horn clause can be formed by the extraction of any NPI (or a PP containing such a NPI). But in fact only a subset of NPIs can be extracted in Horn clauses, and these must be of the unary NEG type. For instance, as already touched on in (5), NPIs can occur without negation in conditional clauses, restrictive relative clauses within universal (and negative existential) quantifier DPs, complements of predicates like astounding, in clauses with only DPs, etc., as in:

1. (32)

   1. a. If she rejects any application, the Dean will object.

   2. b. No professor who rejected any application was criticized.

   3. c. It is astounding that she rejected any application.

   4. d. Only Rachel bought any blouse.

But no NPI of the type in (32) can form a Horn clause:

1. (33)

   1. a. If you believe that she rejected any application, you are mistaken.

   2. b. *If you believe that any application did she reject, you are mistaken.

2. (34)

   1. a. No scientist who imagined that he taught any pigeons to text was rational.

   2. b. *No scientist who imagined that any pigeons did he teach to text was rational.

3. (35)

   1. a. It is astounding that they think she visited anywhere like that.

   2. b. *It is astounding that they think that anywhere like that did she visit.

4. (36)

   1. a. Only Carl thinks she bought any blouse.

   2. b. *Only Carl thinks that any blouse did she buy.

In the framework of CP2014, the above contrasts are not mysterious. Horn clauses are shown to be special cases of Negative Inversion clauses, and only negative phrases (and certain irrelevant exceptions) can be fronted in Negative Inversion clauses. The relevant NEGs are covert in the NPIs in Horn clauses because they have raised into the containing main clauses. Critically, the main verbs in such cases are triggers for Classical NEG Raising. So the conclusion is that the fronted phrases in Horn clauses manifest unary NEG NPIs because such fronted phrases must satisfy the conditions on Negative Inversion. The fact that NPIs can do that makes sense only under a view which can take them to represent underlying structures of the form [_DP [_D NEG₁ SOME] NP].

7. Overview of SC NPIs

If the distinction between unary and binary NEG NPIs drawn in the previous sections is correct, one would expect it to manifest in other languages. In English, common NPIs like any and ever do not morphologically mark that distinction, but there could well exist languages in which the two types posited in our framework are systematically morphologically distinct. And such cases are in fact already well described in the literature, although not in the theoretical terms we have developed.

The first documentation of this state of affairs that we know of is in the work of Progovac (Reference Progovac1990, Reference Progovac1991, Reference Progovac1993, Reference Progovac1994, Reference Progovac2005). Progovac (Reference Progovac1994) showed that SC has two classes of NPIs (see also Szabolsci 2004: 439, fn. 29, and see Zeijlstra Reference Zeijlstra2004 for a completely different approach to Slavic ni-NPIs). While the basic material in this subsection is from Progovac (Reference Progovac1994), we also cite data from other works of hers and from many extensive and extremely helpful personal communications from her and from Željko Bošković.

One class of SC NPIs, the ni-NPIs, are marked with the prefix ni- or morphological variants. The second class, the i-NPIs, are marked with the prefix i- or morphological variants. Some ni-NPIs are given in (37).

1. (37)

   1. a. ni(t)ko ‘no one’

   2. b. ništa ‘nothing’

   3. c. nikud ‘nowhere’

   4. d. nikad ‘never’  (Progovac Reference Progovac1994: 40)

These forms are compounds consisting of three elements which Progovac glosses from left to right as n- ‘NEG’, i- ‘any’ and a wh-form ‘who’, ‘what’, etc. In the framework of CP2014 it is tempting to think of i- as a morphological realization of SOME so that ‘no one’ would have the structure [[[_D n-] [_D i]] [_NP tko]].

Some i-NPIs are given in (38).

1. (38)

   1. a. i(t)ko ‘anyone’

   2. b. išta ‘anything’

   3. c. ikud ‘anywhere’

   4. d. ikad ‘ever’  (Progovac Reference Progovac1994: 41–42)

Both classes of forms qualify as NPIs in the usual sense. For example, neither can occur in a simple positive declarative sentence.

1. (39)

   1. a. Milan *(ne) vidi ništa

      Milan not see nothing

      ‘Milan cannot see anything.’  (Progovac Reference Progovac1994: 40)

   2. b. *Milan to i-kako odobrava

      Milan that any-how approves

      ‘Milan approves of that in any way.’  (Progovac Reference Progovac1994: 43)

On the basis of data like (39), one might conclude that SC ni-NPIs have the key property of only occurring in clauses which contain an overt NEG, ne, regardless of the position of the NPI as object, subject, or adjunct. However, a reviewer points out that there is a class of cases where ni-NPIs occur in certain small clauses or prepositional phrases in which no overt ne is present. The reviewer provided the example in (40).

1. (40) on place zbog nichega

   he cries because-of ni-npi

   ‘He cries for no reason.’

Similar SC data are given in Progovac (Reference Progovac2005: 190). These seem similar to English cases like (41) discussed in CP2014 (139–141).

1. (41) In no clothes, Bill might shock the audience.

The key similarity is that in both types of case the scope of the understood negation does not include the main clause but is limited to some subconstituent.

Fitzgibbons (Reference Fitzgibbons2010a–Reference Fitzgibbonsb) studied analogous constructions in Russian, positing what we would interpret as a covert NEG in these contexts. Reinterpreting Fitzgibbons' analysis in our framework, the class of contexts at issue could be said to involve the same obligatory NEG raising as in standard constructions, but with the additional obligatory deletion of the raised NEG.

Our initial hypothesis about SC NPIs is that while both types involve an existential component, SC ni-NPIs correspond to unary NEG structures like (42), while SC i-NPIs correspond to binary NEG structures like (43).

1. (42) Abstract Structure of SC ni-NPIs

   [_DP [_D NEG SOME] [_NP X]]

2. (43) Abstract Structure of SC i-NPIs

   [_DP [_D NEG [_D NEG SOME]] [_NP X]]

Under this view, there is an overt NEG in SC ni-NPIs but not in SC i-NPIs.Footnote ⁸

In (42) and (43) we posit DP structures for SC NPI nominals. However, Bošković (Reference Bošković, Compton, Goledzinowska and Savchenko2007) has argued that SC nominals lack a DP structure. We believe that adopting structures of the DP-free form he suggests would not fundamentally affect the arguments being made here.

The bases for the above conclusions are these. First, there is a parallel between the type of syntactic environments which sanction SC ni-NPIs and those that permit English unary NEG NPIs. Second, the properties of ni-NPIs in SC can be explained in terms of the unary NEG analysis. Third, there is a parallel between the syntactic environments that allow English binary NEG NPIs and those that allow SC i-NPIs.Footnote ⁹

A note on terminology is in order. We use the term ‘NPI’ to characterize both i-NPIs and ni-NPIs in SC. However, it is also common for NPIs in SC to be referred to as n-words (‘negative words’: Giannakidou Reference Giannakidou, Everaert and van Riemsdijk2006). The relationship between an n-word and a negative marker, or between two n-words, is called negative concord: ‘Roughly, we talk about negative concord in situations where negation is interpreted just once although it seems to be expressed more than once in the clause’ (Giannakidou Reference Giannakidou, Everaert and van Riemsdijk2006: 328–329). According to Giannakidou, a criterial property of n-words is that they are able to occur as fragment answers. By this criterion, the SC ni-NPIs are n-words (see section 8.2). For us, n-words are negative quantifier DPs, and have a similar structure to English unary NEG NPIs, which are also negative quantifier DPs. The syntactic difference between SC n-words and English unary NEG NPIs is that in SC, NEG raising must leave a resumptive copy, while standard English NEG raising cannot leave such a copy. Moreover, in our framework, the phenomenon of negative concord is analyzed in terms of NEG raising and determiner sharing (see below).

8. SC ni-NPIs

In this section, we discuss the distribution and analysis of SC ni-NPIs.

9. SC i-NPIs

In this section, we discuss the distribution and analysis of SC i-NPIs.

10. Sentential Negation

The traditional notion of sentential negation (and the related, more recent syntactic concept of a clausal NegP) have played no role in the preceding sections. In all the SC cases involving ni-NPIs we have discussed, the omnipresent clausal NEG was analyzed as arising via raising from a unary NEG NPI (leaving a copy NEG in situ). Such an analysis leaves open how to analyze the multitude of sentences containing clausal NEGs that do not appear to involve NPIs at all.

1. (81)

   1. a. Milan nije plesao.

      Milan neg.aux danced.sgm

      ‘Milan didn't dance’

   2. b. Milan ne pleše.

      Milan neg dances.

      ‘Milan does not dance.’

   3. c. Milan nije video Marij-u.

      Milan neg.aux seen Mary-acc

      ‘Milan did not see Mary.’  (Progovac, p.c.)

The open question is what is the correct syntactic analysis of ne in examples like (81). What is its structural origin? Since there is no overt NPI in such cases, ne appears not to have raised from any DP at all. Similar questions arise in every language for negative clauses not containing NPIs.

Specifically, let us consider whether dealing with this issue requires sentential negation of the kind that heads a clausal NegP syntactically and negates a proposition semantically. We suggest that it does not.

One approach to the problem posed by cases like (81) would be to assume that the NEG in so-called sentential negation examples modifies the relevant verb, VP, adjective, AdjP or other predicate in the clause. Such a NEG would then raise from its predicational position to a higher clausal position. This is the sort of analysis of simple negation (i.e., negation not accompanied by NPIs) given in CP2014.

Here, we will suggest a much more radical alternative, although we will not be able to argue for it in detail here; see Collins, Postal and Yevudey (Reference Collins, Postal and Yevudey2015) for a related treatment. See also Bošković (Reference Bošković, Schardl, Walkow and Abdurrahman2009), who arrives at a partially related conclusion about a ‘null operator (Op) with iNEG’. We propose that examples such as (81) involve a covert negated existential quantifier DP and that the so-called sentential NEG is actually raised from this quantifier DP. The semantics of this syntactic proposal fall in the general domain of so-called event semantics, introduced by Davidson (Reference Davidson and Rescher1967). But we believe the idea that the relevant quantifier has a syntactic (albeit covert) reality to be novel.

This covert existential quantifier DP will function just like other existential quantifier DPs, sharing with them the property that their scope occurrences are covert. Its unique feature is the additional covertness of its argument occurrence.

We do not assume that the relevant existential quantifier DP is limited to events, but would recognize a range of other possibilities (e.g. states), sometimes called eventualities. Maienborn (Reference Maienborn, Maienborn, von Heusinger and Portner2011: 809) cites the following relevant remark from Kim (Reference Kim and Rescher1969): ‘When we talk of explaining an event, we are not excluding what, in a narrower sense of the term, is not an event but rather a state or process.’ See Maienborn (Reference Maienborn, Maienborn, von Heusinger and Portner2011) for more recent discussion and alternative formulations of event semantics.

To illustrate these ideas, we begin with the NPI-free simple English sentence in (82).

1. (82) Claudia sang.

This sentence can be represented in predicate logic in terms of quantification over events, as in (83), omitting reference to time.

1. (83) ∃e.sing(e, Claudia)

   ‘There is an event e such that e is a singing event by Claudia.’

The syntactic version of this hypothesis simply posits a covert quantifier DP that quantifies over events, as in:

1. (84) [Claudia sang [_DP SOME EVENT]]

As with all quantifier DPs, this silent one will have two syntactic positions: a scope position and an in-situ position (interpreted as a variable); see CP2014, chapter 2. Structure (84) illustrates the in-situ position. But (84) is not intended to make any particular claims about the argument position of the quantifier DP; whether or not it is internal to the VP is not relevant to the points to be made here. We omit the representation of the scope position in (84).

Turning to the negative case, on the Davidsonian view, example (85a) would have the semantic representation in (85b).

1. (85)

   1. a. Claudia did not sing.

   2. b. ¬∃e.sing(e, Claudia)

      ‘There is no event e such that e is a singing event by Claudia.’

In other words, (85) represents a case of the negation of existential quantification. In the framework of CP2014, negated existential quantifiers have, as already indicated, a structure like (86):

1. (86) no chinchilla = [[NEG SOME] chinchilla]

In this representation, NEG is realized as no, while SOME is covert. So (85a) would be represented with a negated existential quantifier over events.

Given this background, the SC sentence in (81b) will have the structure in (87), ignoring the scope position occurrence of the event quantifier.Footnote ¹¹

1. (87) Milan NEG₁  pleše [[cNEG₁ SOME] EVENT]

   Milan   dances

In this structure, NEG₁ modifies SOME and raises to a pre-verbal position leaving a copy NEG. The whole expression [[cNEG₁ SOME] EVENT] is covert, the general case for event quantifier DPs. In other respects, the SC examples in (81) fall together with the ni-NPI cases.

While Herburger (Reference Herburger2001:302) also appeals to events in the analysis of negative sentences, her approach differs from ours in that we assume that there is a syntactically present event quantifier DP, and that negation syntactically modifies the D of this DP in the structure: [[NEG SOME] EVENT].

11. The Bagel Problem

Bifurcated systems of NPIs more or less parallel to that of SC are found in other languages already described in the literature, including Russian (Pereltsvaig Reference Pereltsvaig, Przepiorkowski and Brown2004), Hungarian (Tóth Reference Tóth1999) and Polish (Blaszczack Reference Blaszczak2003). Pereltsvaig suggests that the facts in Russian create what she calls the bagel problem. The basic fact is that Russian analogs of SC i-NPIs, like the SC examples, are incompatible with sentential negation under conditions predicted to be ideal for NPIs under semantic views of licensing (since the NPIs are then under the scope of an antimorphic operator).Footnote ¹²

The relevant SC pattern is illustrated in (88).

1. (88)

   1. a. *Marija ne poznaje i(t)ko-ga

      Mary not knows anyone-acc

      ‘Mary does not know anyone.’

   2. b. *I(t)ko ne vidi Milan-a

      anybody not sees Milan-acc

      ‘Anyone does not see Milan.’

   3. c. *Milan ikada ne vozi

      Milan ever not drives

      ‘Milan doesn't ever drive.’

   4. d. *Milan to i-kako ne odobrava

      Milan that any-how not approves

      ‘Milan does not approve of that in any way.’  (Progovac Reference Progovac1994: 42)

Similar ungrammatical examples can be given in which both a ni-NPI and an i-NPI occur in the same clause.

1. (89)

   1. a. Niko nije video niko-ga/ *iko-ga

      nobody neg.aux saw no-one-acc/ anybody-acc

      ‘Nobody saw anybody.’

   2. b. Milan nije nikada video niko-ga/ ?*iko-ga

      Milan neg.aux never saw no-one-acc/ anybody-acc

      ‘Milan never saw anybody.’  (Progovac, p.c.)

Recall that SC NEG raising from DPs is only relevant in the case of unary NEG NPIs. For binary NEG NPIs, which have the form [[NEG₂ [NEG₁ SOME]] NP], NEG₁ and NEG₂ are deleted as a consequence of some general NEG deleter. So, under these assumptions, the preverbal NEGs in (89) cannot have resulted from the raising of a NEG originating in the binary NEG i-NPIs. Rather, these preverbal NEGs must have originated as modifiers of the Davidson event-quantifier DP, [[NEG SOME] EVENT].

Evidently, some sort of clausemate condition is operative here to rule out such examples. There are at least three partially alternative ways one could formulate the relevant constraint in present terms. One could claim that (i) the NEG deleter of the outer NEG of the reversal NEG defining an i-NPI is not a surface clausemate of the i-NPI itself; or (ii) it is not a clausemate (at any point) of the argument position of the i-NPI or (iii) it is not a clausemate of the scope position of the i-NPI.

While we lack any strong SC basis for a decision about these three, (90) seems to us the most plausible theoretical choice:

1. (90) The Bagel Condition: If K = [_DP [_D NEG₁ SOME] NP ] is a general NEG deleter for a reversal NEG₂ of M = [_DP [_D NEG₂ [NEG₃ SOME]] NP], then K is not a clausemate of M's scope position.

   So consider a partial structure of (88a), shown in (91).

2. (91) [[_DP [NEG₁ SOME] EVENT]₅ [[_DP [NEG₂ [NEG₃ i]] (t)ko-[ga]]₇ [Marija poznaje DP₅ DP₇]]]

Here the coindexing of the full DPs in scope positions with the abbreviated DPs in argument (in-situ) positions represents the fact that these DPs have occurrences in two different positions.

In this case, the NEG deleter of NEG₂ is the negated event quantifier DP₅ = [[NEG SOME] EVENT]], but it is a clausemate of the scope position of DP₇, violating condition (90).

When pre-verbal negation and the i-NPI are not clausemates, no violation of (90) arises, as shown in (92).

1. (92) Milan ne tvrdi da Marija pozsnaje i(t)ko-ga

   Milan neg claims that Mary knows anyone-acc

   ‘Milan does not claim that Mary knows anyone.’  (Progovac Reference Progovac1994: 42)

In this case, [[NEG SOME] EVENT] is separated from the scope position of the i-NPI by the clause boundary before da, and there is no violation of the Bagel Condition.

There is a certain formal similarity between our condition (90) and the binding-theory approach to the relevant facts in Progovac (Reference Progovac1994). Both involve an antilocality condition on the allowable distribution of i-NPIs, for us (90), for Progovac a version of Principle B.

Suppose that (90), the Bagel Condition, is factually adequate; that is, it does not block any grammatical i-NPI SC cases. Nonetheless, it remains quite stipulative. This raises the question of whether (90) could be derived from some more general condition, either something about SC specifically, or preferably from a grammatical universal. Space precludes considering universality issues here. However, we briefly explore the first possibility, while acknowledging that our remarks are highly speculative.

Consider (44a), repeated here as (93).

1. (93) Marija ne voli ni(t)ko-ga

   Mary neg loves no-one-acc

   ‘Mary does not love anyone.’

   * ‘Mary does not love no one’ = ‘Mary loves someone’   (Progovac Reference Progovac1994: 80)

Example (93) has only a negative concord meaning; that is, cannot be interpreted as an instance of double negation.Footnote ¹³ In our terms, the question is why ne cannot be analysed as negating the existential quantifier expression in the Davidsonian event quantifier DP. We believe that such double-negation readings are in general absent from Serbo-Croatian clauses involving (what are for us) two NEG-containing phrases.

Actually, this statement is overly general. As Progovac (Reference Progovac, Horn and Kato2000) documents in some detail, in a variety of environments involving certain adverbials and instances of coordination some double-negation possibilities exist. We cannot attempt to take these data into account here. But we note that these ‘exceptions’ appear to correlate with those environmnents where ni-NPIs do not permit the obligatory NEG raising/resumptive NEG analysis we have posited.

From the standpoint of CP2014, the generalization seems to be that if two SC NEG-containing DPs appear in the same clause, they must have a polyadic quantification interpretation with a single underlying quantifier D = [NEG SOME]. This motivates a formulation like (94).

1. (94) The SC Double-Negation Constraint: If K = [_DP [_Da NEG₁ SOME] NP ] is a clausemate of M = [_DP [_Db NEG₂ …]…], then D_a = D_b.

In other words, a DP of the form [_DP [_D NEG₁ SOME] NP ] can only be a licit clausemate of another NEG-containing DP if they share a single determiner. That in turn would force the resulting interpretation to involve polyadic quantification.

As well as cases like (93), constraint (94) rightly blocks double-negation readings of SC clauses with multiple overt ni-NPIs like (56c), repeated here:

1. (95) Nijedan student nije pio nikakav viski.

   No student neg.aux drunk no whiskey

   ‘No student drank any whiskey.’

Here Nijedan student, which we analyze as having a D of the form [NEG SOME], is a clausemate of the NEG-containing DP nikakav viski, and thus falls under (94).

Notably then, (90), the Bagel Condition, falls out from the SC Double-Negation condition. This can be shown informally as follows. Consider a structure in which an i-NPI, which we have analyzed as a reversal structure [_DP [_D NEG [_D NEG SOME]] NP], cooccurs with either an overt ni-NPI (with clausal ne) or just with clausal ne (as in (88a)). Since we analyze examples like (88a) in terms of the negative event DP, both cases reduce to having a NEG-containing DP of the form of the first DP in (94). And any reversal structure DP, hence any i-NPI in our terms, will satisfy the second DP representation in (94). But if a DP instantiating that second DP representation is of the reversal form [_DP[_D NEG [_D NEG SOME]] NP], the identity condition on determiners cannot be met. The D of the ni-NPI or of a negative event quantifier DP is [_D NEG SOME], distinct from the D of an i-NPI, which is [_D NEG [_D NEG SOME]].

Our reduction of the Bagel Condition to the SC Double-Negation Constraint gives indirect support to the assumption that i-NPIs in SC involve a double-negation structure. If they did not involve a double-negation structure, they would not be subject to (94).

12. Conclusion

In this paper we have argued on the basis of evidence from SC for the claim in CP2014 that English NPIs instantiate two distinct structures: unary NEG NPIs and binary NEG NPIs. We claimed that SC i-NPIs correspond to English binary NEG NPIs, while SC ni-NPIs correspond to English unary NEG NPIs. We showed how the properties of SC ni-NPIs can be understood in terms of their analysis as unary NEG NPIs. We explained the syntactic differences between SC ni-NPIs and English unary NEG NPIs in terms of the assumption that ni-NPIs, like no forms in nonstandard English, involve copy raising. Further, we suggested that these ideas, combined with a syntacticized version of what is now called event semantics, could provide a novel approach to so-called sentential negation.

We have also indicated how a simple clausemate condition can account for the basic facts corresponding to the generalization Pereltsvaig (Reference Pereltsvaig, Przepiorkowski and Brown2004) referred to (for Russian) as the Bagel Problem. And we have suggested that this condition may in turn be a consequence of a more general constraint barring two SC negative DPs which fail to share a D.