2 Bipartite negation: Basic facts

Before discussing NPIs and their analysis, we give an overview of the basic facts of Ewe negation in order to help the reader parse the relevant sentences which follow. As shown in (3), both NEG $_{1}$ and NEG $_{2}$ are necessary in bipartite negation:

NEG $_{2}$ immediately follows the verb phrase. In the following example, NEG $_{2}$ must follow the direct object; (4c) shows that there cannot be two occurrences of NEG $_{2}$ :

The following examples show that NEG $_{2}$ must follow various kinds of VP-internal PPs (see also Aboh Reference Aboh, Aboh and Essegbey2010: 122–123):

VP adjuncts must also appear to the left of NEG $_{2}$ .

The facts in (4)–(8) suggest that NEG $_{2}$ is in a position following the VP.

However, certain sentential elements can follow NEG $_{2}$ , for example, question particles (Ameka Reference Ameka1991: 64–65):

Furthermore ‘because’-adjuncts either precede or follow NEG $_{2}$ with contrasting interpretations:

Compare (10b) to (8b), where NEG $_{2}$ may only follow the adjunct. The data suggest that ‘before’-clauses and ‘because’-clauses occur in different syntactic positions. Arguably, ‘before’-clauses are VP-internal, and hence never follow NEG $_{2}$ , while ‘because’-clauses may either be VP-internal or VP-external.

We turn to the placement of NEG $_{1}$ , which precedes the verb and also preverbal auxiliary elements, for example, the future marker:

Furthermore, in the negative imperative, NEG $_{1}$ precedes the negative imperative particle ga-:

In summary, the following generalizations about NEG $_{1}$ and NEG $_{2}$ hold for finite clauses which manifest bipartite negation:

While finite clauses include NEG $_{1}$ and NEG $_{2}$ , gerundive phrases do not in general manifest bipartite negation. As the examples in (14) show, gerundive phrases are formed by verbal reduplication. When the verb is negated, the combination of the negation marker and the verb is reduplicated, yielding the sequence: NEG–V–NEG–V.

In these examples, there is a preverbal NEG ma-, but no post-VP NEG in the gerundive phrase. While we cannot pursue the issue of the distribution of NPIs in gerundive phrases, the following sentence shows that when an NPI is present, NEG $_{2}$ appears:

In some conditional constructions, there is a post-VP $o$ , but no preverbal negation:

The analysis of NEG $_{2}$ given in Section 9 below does not cover the first post-VP $o$ in cases like (16). A reviewer asks whether the first post-VP $o$ may be a conjunction. We have not investigated the issue.

3 Collins & Postal Reference Collins and Postal2014

CP2014 (Chapter 3 and passim) argues that Universal Grammar defines two fundamental types of NPIs. Type 1 are illustrated in (17), and Type 2 NPIs are illustrated in (18):

Roughly, Type 1 NPIs require a negation somewhere in the sentence (not necessarily overt, see CP2014: Chapter 3). In (17a), there is a negation following the finite auxiliary. In (17b), the negation is part of the subject quantifier DP nobody. Type 2 NPIs do not require any negation. For example, in (18a), the NPI anything occurs, but there is no overt negation and no motivation for positing a covert one.

CP2014 (Chapter 3 and passim) argues that Type 1 and Type 2 NPIs have partially different syntactic structures. This represents a sharply distinct position from mainstream views of NPIs, where NPIs, including those in (17) and (18), are usually analyzed as indefinites. Therefore, in mainstream views, there is no difference between the structure of the NPIs in (17) and (18).

Type 1 NPIs have a structure and meaning identical to the structure and meaning of negative quantifiers, accounting for the truth conditional equivalence of pairs like the following:

In particular, CP2014 (Chapter 3) analyzes both the nobody and anybody of (19) as DPs of the form [[NEG SOME] body], where NEG modifies an existential quantifier expression SOME. The differences between (19a, b) lie in the fact that in the (19a), SOME is realized as null, while NEG is spelled out as no. In (19b), NEG raises to the post-Aux position, and SOME is spelled out as any (see rule (21)).

In these terms, a more precise structure of (17a) is given in (20b):

In (20b), NEG $_{1}$ originates in a position modifying SOME (internal to the NPI anything). NEG $_{1}$ then raises to the post-Aux position, but is interpreted in its position of origin, modifying SOME. The angled brackets around the lower occurrence of NEG $_{1}$ in (20b) indicate a non-pronounced occurrence. The reader is referred to CP2014 (Chapters 3 and 5) for further discussion.

CP2014 (19, 21) claim that any in (20a) is a form of SOME, determined by the rule in (21):

Type 2 NPIs are analyzed in CP2014 (Chapter 3 and passim) as double negation structures. Consider again (18a), repeated here as (22):

First, there is no overt NEG preceding the verb in (22), unlike the situation with Type 1 NPIs illustrated in (17a). Second, the interpretation of anything in (22) is equivalent to the existential quantifier something. It is argued that the NPI anything in (22) has the double negation structure in (23). In such a structure, the semantics of NEG $_{1}$ cancels that of NEG $_{2}$ , so that the resulting interpretation is equivalent to that of something.

A binary NEG structure such as (23) contains two unpronounced NEGs. According to CP2014 (especially Chapters 7 and 8), the NEGs in (23) are unpronounced because they are deleted. NEG deletion involves a relation between individual NEGs and other phrases, their NEG deleters. The relevant NEG deleters in the case of Type 2 NPIs include the following:

So in (23), the conditional complementizer if deletes the NEG $_{1}$ of the structure [NEG $_{1}$ [NEG $_{2}$ SOME]] (while NEG $_{1}$ deletes NEG $_{2}$ ). Because NEG $_{2}$ is deleted, SOME is realized as any by rule (21a). See CP2014 (Chapter 4) for further detail.

CP2014 (Chapter 3 and passim) refer to Type 1 NPIs as unary NEG NPIs, since only one NEG modifies SOME. Type 2 NPIs are referred to as binary NEG NPIs, since there are two NEGs present. The distinction between unary and binary NEG NPIs corresponds to the traditional distinction drawn between strong versus weak NPIs, and strict versus non-strict NPIs (see CP2014: Section 9.4 for discussion). The analysis of Type 2 NPIs plays only a very marginal role in this paper.

4 Analysis of ke-NPIs

The following examples illustrate a range of Ewe expressions that contain áɖéké (built with the morpheme -ké; see (29) below for an analysis breaking down áɖéké morpheme by morpheme):

In (25c), nánéké is a suppletive form of nú áɖéké, which can be used in careful speech. In the remainder of the paper, we just use nánéké.

As the following examples illustrate, when a ke-NPI is present, NEG $_{1}$ and NEG $_{2}$ are obligatory (just as when no ke-NPI is present, as shown in (3) above):

Ewe does not have negative expressions like English nobody, nothing and nowhere, which can stand alone without an additional negative particle (like n’t or not). English sentences with such negative quantifiers are translated with ke-NPIs.

The form áɖéké is composed of áɖé, an indefinite marker (illustrated in (27)) and -ké (see Westerman Reference Westermann1930: 70; Agbedor Reference Agbedor1994: 57):

In spoken Ewe, the indefinite is expressed either as avú áɖé or avú ɖé ‘a dog’. Crucially, the NPI is also expressed either as avú áɖéké or avú ɖéké. We do not pursue these alternative forms here.

Indefinites can be expressed in Ewe in two different ways. The form avú áɖé in (27) is a specific indefinite, and can be paraphrased as ‘a certain dog’. But there is also a bare indefinite illustrated in (28). We do not investigate the difference between these two types of indefinite DPs here.

Now consider the internal structure of ke-NPIs. Agbedor (Reference Agbedor1994: 57) calls -ké a ‘negative quantifier marker’ and assumes that ke-NPIs involve a ‘negative particle in the negative quantifier’. Rongier (Reference Rongier1988: 76) calls -ké a ‘suffixe de négation’, and uses the expression ‘négation du nom’ for ke-NPIs. He notes the relationship between ke-NPIs and verbal negation as well: ‘La négation du nom entraine celle du verbe’ [Negation of the noun requires that of the verb].Footnote ^[2]

We assume that -ké is just NEG, which modifies áɖé. On these assumptions, the structure of Ewe ke-NPIs is as in (29):

This structure could be refined in various ways not directly relevant to the current paper. For example, the structure in (29) violates Kayne’s (Reference Kayne1994) LCA (Linear Correspondence Axiom), since NP precedes D (instead of following it). Furthermore, in Principles and Parameters/Minimalist syntax, áɖé and -ké would head separate projections (see CP2014: 27 for discussion).

While we have used the term ‘NPI’ to characterize ke-NPIs, it would have been equally in line with current usage to refer to them as n-words. There are a variety of different approaches to n-words conflicting in various ways including those in Haegeman & Zanuttini (Reference Haegeman and Zanuttini1991, Reference Haegeman, Zanuttini, Belletti and Rizzi1996), Ladusaw (Reference Ladusaw, Barker and Dowty1992), Giannakidou (Reference Giannakidou1998, Reference Giannakidou2000, Reference Giannakidou, Everaert and van Riemsdijk2006), Haegeman (Reference Haegeman1995, Reference Haegeman, Forget, Hirschbühler, Martineau and Luisa Rivero1997), Zanuttini (Reference Zanuttini1997), de Swart & Sag (Reference De Swart and Sag2002), Watanabe (Reference Watanabe2004), and Zeijlstra (Reference Zeijlstra2004, Reference Zeijlstra2007), among many others.

Our analysis of Ewe ke-NPIs implicitly takes a position on the treatment of n-words. Contrary to various current views, we take them to involve a syntactic NEG, as well as a form representing an existential quantifier. This is parallel to the CP2014 view mentioned in Section 3 that English negative quantifier DPs such as nobody and unary NEG NPIs such as anybody both have the underlying structure [[NEG SOME] body]. In Ewe, the parallel structure has a morphological manifestation, in that both NEG -ké and SOME áɖé are realized overtly. Therefore, the Ewe structure directly supports the claim that UG admits the possibility of NEG modifying SOME.

Now consider the relation between the NEG -ké and the preverbal NEG marker mé-. Consider again (1b) and (2b) above, repeated as (30a, b):

As discussed in Section 3, we assume that the post-Aux NEG in (30a) (that is, -n’t) originates in a position modifying SOME internal to the NPI, but raises to the post-Aux position. We propose that (30a) and (30b) have parallel structures. Just as (30a) involves NEG raising, so does (30b). But a key difference between Ewe and English is that Ewe NEG raising in cases like (30b) leaves a copy NEG in the origin position.

The DP internal NEG is resumptive element, arguably similar to the highlighted resumptive pronoun in English cases like He is the kind of guy who I wonder if he will ever get married. In this example, the resumptive pronoun occupies the position that in Principles and Parameters syntax would normally be occupied by a trace of the movement of the relative pronoun who.

Crucially, we are assuming that while a copy NEG can have a phonological shape identical to the raised element (see Bell Reference Bell2004 for a discussion of this situation in Afrikaans), this need not be the case (just as in the resumptive pronoun case, the raised wh-phrase and its associated resumptive pronoun are not identical phonologically). In (30b), the preverbal NEG $_{1}$ is mé- and the DP internal copy is -ké.

Given the assumption in (31), the analysis of (30b) is given below. The notation cNEG $_{1}$ is used to represent the fact that -ké is a copy NEG, associated with the raising of NEG $_{1}$ .

Leaving out the postverbal NEG $_{2}$ for the moment, the structure of (32) is as follows:

The T element in (33) can be filled by the future marker in some sentences, or left empty (in past and present tense sentences). The structure in (33) could be refined in various ways, but suffices for our purposes.

Given this analysis of ke-NPIs, two parameters arguably distinguish Ewe from English. First, when a NEG raises from a negated existential DP, it leaves a copy in Ewe, but not in standard English. Specifying standard English in this generalization is important, since NEG raising can leave a copy in varieties of non-standard English, as in the non-standard English sentence I didn’t see nobody (see Blanchette Reference Blanchette2015 and Collins & Postal Reference Collins and Postal2017). This parameter is given below:

The second parameter is that NEG raising is optional in English, but obligatory in Ewe:

This parameter accounts for the fact that English allows both (19a, b) while Ewe only has the analog of (19b). Example (19a) does not involve NEG raising, while (19b) does.

5 Non-negative contexts

In this section, we show that Ewe ke-NPIs cannot appear in conditionals, in yes–no questions, in the complement of a verb meaning ‘surprise’, in the restriction of universal quantifiers or in the scope of ‘only’-DPs. In this way, ke-NPIs differ from any-NPIs in English which appear in all those environments. The generalization is that ke-NPIs only appear if a preverbal negation is present. We will show that this generalization follows from the analysis of ke-NPIs presented in Section 4.

We will illustrate each context with two NPIs, nánéké ‘anything’ and avú áɖéké-wó ‘any dogs’. The translations will illustrate that English any-NPIs are available in the corresponding contexts.

If the NPI is replaced by an indefinite, the resulting examples are grammatical, as shown below:

The following examples show that ke-NPIs do not occur in yes–no questions:

The following examples show that ke-NPIs are not licensed in the clausal complement of a verb meaning ‘surprise’:

The following examples show that ke-NPIs are not licensed in the restriction of a universal quantifier (the gloss tp stands for ‘terminal particle’ (see Ameka Reference Ameka1991), which should not be confused with TP ‘tense phrase’ used in the syntactic structures in this paper):

Lastly, ‘only’-DPs do not license Ewe ke-NPIs:

Evidence for the claim that ke-NPIs only appear if a preverbal negation is present is provided by the novel Ku le Xɔme (Akafia Reference Akafia1970). A search revealed 27 instances of nánéké ‘anything’, all of them in contexts containing a preverbal negation mé-.

These facts about ke-NPIs follow from our analysis of ke-NPIs as negative DPs. For example, consider (36a), repeated below:

Under our assumption in (31) that -ké is a copy left by NEG raising, (46) is ungrammatical because the only way -ké can be introduced into the structure is as a copy of a raised NEG. But there is no raised NEG in (46). A similar explanation holds for (38), (40), (42) and (44).

The English translation of (46) with anything is grammatical because anything does not have to represent a unary NEG NPI. So there is no reason for it to be accompanied by a raised NEG. Rather, in the translation of (46), anything is a Type 2 NPI, which, in the framework of CP2014, is a binary NEG NPI with the conditional complementizer if as the NEG deleter.

There are unary NEG NPIs in English that are not homophonous with binary NEG NPIs. As discussed in CP2014 (Section 4.8), jackshit has both an NPI and a non-NPI usage:

We gloss the occurrence in (47a) as  because it is equivalent to ‘zero’. See Postal (Reference Postal2004: Chapter 6). We gloss the occurrence in (47b) as  because it is equivalent semantically to anything.

The  usage cannot appear in non-negative contexts:

Example (48) is ruled out on the relevant interpretation, illustrating that  cannot occur in non-negative contexts.

A striking generalization about the environments where Ewe ke-NPIs occur is that they correspond to those environments where  appears in English. This parallel distribution strongly suggests that ke-NPIs and  should be analyzed in the same way. In present terms, both are analyzed as unary NEG NPIs.

Another generalization apparent from the data in (36)–(45) is that Ewe lacks Type 2 NPIs in these contexts. For example, consider (37a), repeated below:

There is no Type 2 NPI in Ewe corresponding to something that appears in a conditional clause. Rather, Ewe simply uses the indefinite. Similarly, there is no Type 2 NPI used in yes–no questions, in the complement of ‘surprise’, in the restriction of a universal quantifier or in the scope of an ‘only’-DP.

A simple way to state this difference between Ewe and English is the following:

We return in Section 12 to facts which ground our caution in limiting this statement about Ewe to nominal NPIs rather than generalizing to all Ewe NPIs.

6 The Remnant Raising Condition

One difference between English any-NPIs and Ewe ke-NPIs is that only ke-NPIs can appear in subject position when the NPI ‘licenser’ is in the same clause (see Agbedor Reference Agbedor1994: 56):

In the CP2014 framework, (51a) would have the following structure:

In cases like (52), NEG $_{1}$ raises to the post-Aux position, while the remnant DP $_{2}$ raises to subject position.

We suggest that these cases are ungrammatical because such remnant raising is barred universally:

In (52), the higher occurrence of DP $_{2}$ c-commands NEG $_{1}$ , violating condition (53). This constraint could be thought of as a version of the well-known c-command constraint on NPIs, stated in terms of the framework of CP2014. As will be seen in Section 9, the constraint in (53) also plays a role in accounting for a difference between the behavior of fragment answers in Ewe and English.

The structure of the Ewe sentence in (51b) is given below:

Structure (54) does not violate (53) since a copy NEG, cNEG $_{1}$ rather than ${<}$ NEG $_{1}>$ fills the original position of NEG $_{1}$ in DP $_{2}$ . In effect, the copy NEG allows the structure to avoid a violation of (53), just as resumptive pronouns in certain English cases allow a structure to avoid a violation of island constraints.

7 Determiner sharing

As in English, multiple NPIs can appear in a single Ewe clause:

While (56a) above has two ke-NPIs, and (56b) has three, both are interpreted as having one semantic negation. If -ké is analyzed as NEG, how can multipleke-NPIs yield only a single semantic negation? CP2014 (Chapter 6) propose that in such cases there is determiner sharing. So in (56a) there is a single underlying determiner [NEG SOME] which is shared by two DPs. Determiner sharing is indicated in (58) below by co-indexation of the two quantifiers and the two NEGs. The single determiner gives rise to two copies of áɖéké.

Since there is only one underlying syntactic determiner [NEG SOME], it follows that there is only one semantic negation. CP2014 propose that such determiner sharing is interpreted in terms of polyadic quantification, where a single quantifier quantifies over n-ary sequences. In the case of (56a), where only two DPs share a D, the result is interpreted as: there is no ${<}$ x,y ${>}$ x a student and y a quantity of whiskey such that x drank y.

Given this background, (55) and (56) raise the question of how NEG $_{1}$ can be related to the -ké of each ke-NPI. Recall that in (31) we assumed that -ké is always a copy of a NEG that has raised, repeated below:

We propose that in such cases of multiple ke-NPIs the NEG $_{1}$ raises to T from the shared determiner [SOME NEG]. Since the shared determiner has two occurrences (two places in the structure), the result can be represented as follows (ignoring the VP final NEG $_{2}$ for the moment):

Representation (58) represents the underlying structure of (56a) before the subject raises to Spec TP (subject position). In (58), NEG $_{1}$ undergoes copy raising from the underlying shared D which has two occurrences.

Such raising recalls the phenomenon of Across-the-Board (ATB) extraction illustrated in the sentence Which plan did Bob buy and Luke sell? In both ATB wh-movement and NEG raising in Ewe, a single raised syntactic object has two underlying occurrences (positions).

8 Long distance licensing

Ewe ke-NPIs need not occur with a clausemate preverbal negation. Example (59b) is a response to the assertion in (59a). Example (59c) with clausemate negation is given for comparison. The sentences in (60) provide an additional example.

Since we are analyzing ke-NPIs as Type 1 NPIs, which are the unary NEG NPIs, one might expect the relation between negation and ke-NPIs to be clause bounded (see CP2014: Chapter 9 on the clause boundedness of unary NEG NPIs). However, (59) shows that the ke-NPI may be separated from its preverbal NEG by a clause boundary.

We propose that such sentences involve high scope of the ke-NPI. Before presenting our analysis, we briefly discuss our assumptions about scope. We assume the scope of quantifiers is represented syntactically by the presence of DPs in clausal scope positions. We follow May (Reference May1985, Reference May1989) and assume that one such scope position for a quantificational DP $_{\text{i}}$ is of the form [ $_{\text{S}}$ DP $_{\text{i}}$ S] (and there are other scope positions lower in the clause as well). In these cases, the clause S contains a DP bound by DP $_{\text{i}}=$ [ $_{\text{DP}}$ D $_{\text{i}}$ NP] so that S is in effect the syntactic representation of an open sentence containing a variable bound by the quantificational DP $_{\text{i}}$ . NP $_{\text{i}}$ then denotes the restriction of the quantifier represented by D $_{\text{i}}$ . So a DP in scope position will always have at least two distinct occurrences, a higher one in a scope position and a lower one in a non-scope position (an ‘argument’ position in some approaches).

Given these assumptions, we propose that the structure of (59b) is as in (61a) (ignoring NEG $_{2}$ and the adjunct for simplicity), and its interpretation would be as in (61b):

In (61a), the higher occurrence of DP $_{2}$ in scope position is not pronounced, as indicated by the angled brackets. Since NEG $_{1}$ raises to the matrix T from the scope position of DP $_{2}$ , there is no clause boundary separating the scope occurrence of the ke-NPI from the raised NEG $_{1}$ .

Cross-linguistically, unary NEG NPIs are commonly clause-bounded. For example, Serbo-Croatian ni-NPIs (which are in many ways similar to Eweke-NPIs), must in general have a clausemate negation (Progovac Reference Progovac1994: 41). The difference between Ewe and Serbo-Croatian, on our view, is that Serbo-Croatian does not allow its unary NEG NPIs to take matrix scope as in (61).

Our analysis of (59) tracks closely the analysis that CP2014 (Chapter 9) gave of English cases where strict NPIs seem to be separated from their associated NEG by a clause boundary, illustrated in (62):

Example (62) shows that although  is a strict NPI, if stressed, it can link to a non-clausemate negation. CP2014 propose that the DP  in (62) has matrix scope and that NEG raising takes place from the scope occurrence of  in the matrix clause. That analysis is entirely parallel to the one just given for the Ewe data in (59).

9 Bipartite negation: Analysis

The goal of this section is to explain the syntactic relationship between NEG $_{1}$ and NEG $_{2}$ in cases like (63).

First, we argue that in any clause containing both negation markers, the post-VP $o$ is structurally higher than the preverbal mé-. The argument is based on ellipsis involving Ewe NPIs.

In (64), the response phrase is an object. (64b) is a fragment answer to the question in (64a), while (64c) is the non-elliptical form.

In the following examples, the response phrase is again an object, but with the question phrase nú ka ‘what’:

In the following examples, the response phrase is a subject:

In the following examples, the response phrase is a locative:

In all the elliptical examples above, the presence of NEG $_{2}$ o is obligatory. We propose a deletion analysis of these facts. In particular, we follow the treatment of sluicing proposed in Ross (Reference Ross, Binnick, Davison, Green and Morgan1969) and defended in Merchant (Reference Merchant2001) (see also Merchant Reference Merchant2004). Consider the following English sluicing example:

According to Ross (Reference Ross, Binnick, Davison, Green and Morgan1969) and Merchant (Reference Merchant2001), the second clause of (68) has the following analysis:

In this sentence, what raises to the left periphery (Spec CP) and the TP is deleted (as indicated by the angled brackets). In the analysis of Merchant (Reference Merchant2001), the deletion of the TP happens under semantic identity with the TP in the first clause in (68). See Merchant (Reference Merchant2001) for the exact definition of semantic identity.

Transposing Ross’ and Merchant’s analyses to the relevant Ewe facts yields the following structure for (64b):

In (70), the DP [ame áɖéké] $_{1}$ raises to the left periphery and the TP remnant is deleted. Crucially, NEG $_{1}$ is deleted, but NEG $_{2}$ is not. This supports the claim that NEG $_{2}$ is higher than NEG $_{1}$ . If, on the contrary, NEG $_{1}$ were higher than NEG $_{2}$ , it would be possible for NEG $_{2}$ to be deleted, leaving NEG $_{1}$ . See Aboh (Reference Aboh, Aboh and Essegbey2010: 131) for a different argument reaching a similar conclusion about the relative height of NEG $_{1}$ and NEG $_{2}$ .

An analysis of fragment answers parallel to that in (70) can also explain why an NPI cannot serve as an answer to an English wh-question, although an overtly negative DP can:

On the analysis of CP2014, the structure of the NPI DP in (71b) would be the following:

In this structure, NEG $_{1}$ raises to the post-Aux position, then remnant [[ ${<}\text{NEG}_{1}>$ SOME] body] $_{1}$ raises to the left periphery. Finally, TP is deleted. This structure violates the Remnant Raising Condition in (53), since [[ ${<}\text{NEG}_{1}>$ SOME] body] $_{2}$ c-commands NEG $_{1}$ .

Haspelmath (Reference Haspelmath2000: 194–196, citing earlier work by Bernini & Ramat Reference Bernini and Ramat1996), Watanabe (Reference Watanabe2004: 562) and Giannakidou (Reference Giannakidou, Everaert and van Riemsdijk2006: 328) take fragment answers to distinguish n-words (or negative concord items) from negative polarity items. N-words but not negative polarity items can be used as fragment answers. This distinction is clearly illustrated in (71), where, for us, nobody is analyzed as [[NEG SOME] body], with the NEG remaining in situ, precluding any violation of the Remnant Raising Condition.

Consider what such a diagnostic says about Ewe, where ke-NPIs must be accompanied by NEG $_{2}$ . In this respect, ke-NPIs are unlike n-words in other languages (such as Italian), where no such negative particle is needed in addition to the n-word itself in fragment answers.

However, Ewe ke-NPIs pattern like n-words with respect to other criteria For example, they can, unlike English NPIs, appear in subject position (see Watanabe Reference Watanabe2004: 562 on the use of n-words in subject position).

For purposes of language classification, Ewe ke-NPIs are NPIs since they are sensitive to negation but they also have some properties of n-words (such as being able to be used in subject position).

In the framework of CP2014, the distinction between NPIs and n-words can be captured as follows. Unary NEG NPIs are unary NEG structures from which the NEG raises and leaves a gap. N-words are unary NEG structures where either the NEG does not raise or where it raises and leaves a copy (instead of a gap).

So here are the assumptions made so far about bi-partite negation in Ewe sentences like (63):

We propose that NEG $_{1}$ is also a copy, left by movement of NEG $_{2}$ to a right peripheral position in the clause. In particular, we will assume that there is a rightward complementizer position COMP and that NEG $_{2}$ raises and adjoins to this position. The resulting analysis of (63) is sketched below:

On this analysis, there is only one underlying NEG, which originates in a position where it modifies SOME. NEG raises to T and leaves a copy in D. Further, NEG raises again to COMP, leaving a copy in T. So NEG raises twice, leaving copies in two distinct positions. But the NEG is only interpreted in its underlying (SOME modification) position.

This treatment explains why there are three surface occurrences of NEG, but only one semantic negation. A structure illustrating this analysis is given below:

What remains to explain is the presence of bipartite negation in sentences that do not involve ke-NPIs, such as the examples in (3). Discussion of that requires a bit of background about so-called event semantics.

10 Event syntax

Consider the following simple English sentence involving no NPIs:

Example (76) can be represented in predicate logic in terms of quantification over an event (we leave out reference to time). See Davidson (Reference Davidson and Rescher1967) and Maienborn (Reference Maienborn2011) for more recent discussion:

Critically, we adopt a syntactic version of this hypothesis, containing a covert quantifier DP, which ranges over events (see Beghelli & Stowell (Reference Beghelli, Stowell and Szabolcsi1997: 93) for a related proposal).

On this view, a silent quantificational occurrence of DP $_{1}$ binds an occurrence of DP $_{1}$ which is interpreted as a variable. The noun EVENT as well as the quantifier SOME are silent. We do not take any stand on the exact location of the event variable DP $_{1}$ . We only assume that the scope position of the quantificational DP $_{1}$ must c-command its variable occurrence.

On the Davidsonian view, the negative sentence (79a) would have the semantic representation in (79b).

In other words, (79) represents negation of an existential quantification. In the framework of CP2014, negated existential quantifiers have syntactic representations like that of (80):

Here NEG is realized as no and SOME is covert. So (79a) would be represented with a negated existential quantifier over events, as follows:

An issue that (81) brings up is that NEG has raised to Aux from the clause initial [ $_{\text{DP}}$ [ ${<}$ NEG ${>}$ SOME] EVENT] violating (53), the Remnant Raising Condition. Given this consideration, a more adequate representation of (79a) would be (82):

Here [ $_{\text{DP}}$ [ ${<}\text{NEG}_{1}>$ SOME] EVENT] is in a low scope position (perhaps adjoined to VP), and NEG $_{1}$ raises to Aux, which c-commands the low scope position; see CP2014 on this use of low scope positions, and the relation between NEG raising and scope. In other words, the event quantifier DP has scope lower than the overt occurrence of NEG in the post-Aux position.

11 Bipartite negation and event syntax

We can now analyze Ewe sentences not containing any ke-NPIs which manifest bipartite negation such as (3a), repeated below:

If in (83) NEG $_{1}$ occurs in the preverbal position as the result of NEG raising, what is the source of NEG $_{1}$ ? Analogizing from the account of English in Section 10, we propose that (83) contains a silent event quantifier DP, and that the NEG raises from this quantifier DP to T, as shown below (ignoring the post-VP NEG for the moment):

But the diagram is misleading in one important way. In a completely filled out analysis, NEG $_{1}$ would raise from the scope position of the event quantifier, not the in-situ position (see CP2014: Chapter 5 for discussion). We leave out the representation of scope positions here for the sake of readability. A diagram for this analysis is as follows:

In (85) NEG raises from D to T, and then from T to C. See Aboh (Reference Aboh, Aboh and Essegbey2010) who discusses C positions in the Gbe languages.

12 Non-ke-NPIs in Ewe

So far we have focused on Ewe ke-NPIs. In this section, we investigate three other NPIs, haɖé ‘yet’, kúrá ‘at all’, and gbeɖé ‘ever’, none of which involve the morpheme -ké.

Example (86a) shows that haɖé ‘yet’ can appear with negation, while (86b) illustrates that it cannot appear in a positive declarative clause:

However, unlike ke-NPIs, haɖé ‘yet’ does occur in the non-negative context of yes–no questions:

But even though haɖé ‘yet’ appears in yes–no questions, it does not appear in conditionals, with the verb meaning ‘surprise’, in the restriction of a universal quantifier or with ‘only’-DPs, as shown below. Note that English yet is also unacceptable in the corresponding contexts.

Like ke-NPIs, haɖé ‘yet’ can appear as a fragment answer to a yes–no question:

Consider now the NPI kúrá ‘at all’. (90a) shows that kúrá can appear with negation, while (90b) indicates that it cannot appear in a positive declarative clause following the object. But example (90c) shows that kúrá can be used following the subject, where it translates as ‘even’. We have not investigated this usage.

Given further linguistic context, kúrá can follow the direct object. We have not investigated this usage either:

Furthermore kúrá ‘at all’ cannot occur in other NPI environments, such as in conditional clauses, the complement of ‘surprise’, the restriction of ‘every’ and in the scope of ‘only’-DPs:

Like the ke-NPIs, kúrá can be used in a fragment answer to a question:

Lastly, example (96a) shows that the NPI gbeɖé ‘ever’ can appear with negation, while (96b) illustrates an often used reduplicated form gbeɖé gbeɖé. (96c) shows that this NPI cannot appear in a positive declarative clause. Examples (97a, b) provide additional illustration:

Just like the previous two NPIs, gbeɖé occurs in yes–no questions. For reasons unclear to us, it only occurs in this context in the presence of  ‘one time’ which is itself not an NPI; see Rongier (Reference Rongier1989: 212), and example (28) above:

Like the other NPIs discussed in this section gbeɖé ‘ever’ does not appear in other NPI contexts, such as conditionals, the complement of ‘surprise’, the restriction of ‘every’ or the scope of ‘only’-DPs:

But like ke-NPIs, gbeɖé ‘ever’ can be used as a fragment answer to a question. However, gbeɖé ‘ever’ is different from all other Ewe NPIs in that NEG $_{2}$ is optional in the fragment answer.

Ameka (Reference Ameka1991: 691) proposes that gbeɖé can be used either adverbially or as a ‘completive signal’, which is used to express disagreement or rejection of a proposition. Without going into syntactic detail, we propose that when NEG $_{2}$ is present in (100b), gbeɖé is an adverbial NPI. When NEG $_{2}$ is absent, it is a completive signal.

The data in (86)–(100) support the following generalization:

The data concerning Ewe non-ke-NPIs are difficult to account for because their distribution characteristics differ both from those of ke-NPIs and from those of English Type 2 any-NPIs. Ke-NPIs only occur in sentences having a preverbal negation. But non-ke-NPIs are not subject to such a stringent requirement. However, they are far more restricted than English Type 2 any-NPIs.

If Ewe non-ke-NPIs were unary NEG NPIs, then that would explain why they do not appear in conditionals, with ‘surprise’, ‘every’ or ‘only’-DPs. However, it would leave unexplained the fact that they can appear in yes–no questions lacking an overt NEG.

If non-ke-NPIs were Type 2 NPIs (binary NEG NPIs in the framework of CP2014), then that would explain why they occur in yes–no questions (with no overt NEG present), but would leave unexplained the fact that they do not occur in conditionals or with ‘surprise’, ‘every’ or ‘only’-DPs.

Unfortunately, our limited research on Ewe NPIs does not permit us to offer a justified hypothesis as to whether Ewe non-ke-NPIs are property analyzed as Type 1 or Type 2 NPIs or perhaps even as some third category not posited in the framework of CP2014.