Confounding Cases: Strict NPIs and Scope
Confounding Cases: Strict NPIs and Scope
Abstract and Keywords
This chapter focuses on the scope of strict negative polarity items (NPIs). It first considers scope issues concerning nonfinite complement clauses, with particular emphasis on similar pairs with analogous judgment markings, the relevance of stress contrasts to the ambiguities of NPI any forms as well as to those of other nominal NPIs, and the differential scope of the determiner phrase represented by anything (“Vaughn didn't accept to write anything about radiation”). It then turns to cases of infinitival complements containing strict NPIs, along with cases where the issue of high-scope confounds involve finite complement clauses. Finally, it suggests that any attempt to diagnose the presence or absence of Classical NEG Raising (NR) must always take into account the possibility of NEG raising out of main clause scope positions.
9.1 Scope Issues and Nonfinite Complement Clauses
Consider examples like (1a,b) and (2a,b):
a. Andrea doesn't believe/think that Carl said jackshitA about compilers.
b. *Andrea doesn't accept/grasp that Carl said jackshitA about compilers.
a. No expert believes/thinks that Carl said jackshitA about compilers.
b. *No expert accepts/states that Carl said jackshitA about compilers.
Similar pairs with analogous judgment markings play a key role throughout this monograph. By analogous judgment markings, we mean that cases like (1a) and (2a), with CNRP main verbs, are taken to be grammatical, while those like (1b) and (2b), with non-CNRP main verbs, are taken to be ungrammatical. By similar pairs, we mean that both (a)-and (b)-type examples have a strict NPI in the complement clause.
However, such examples manifest certain complexities and raise various issues. These require clarification in order to justify the heavy weight we place on pairs like (1a,b) and (2a,b). These complexities will require discussion of quantifier scope as well as the behavior of a range of strict NPIs distinct from the type in the examples above.
To begin, consider the following observation:
(3) (Kempson 1985:236)
“The final datum about any and its construal is that any in negative polarity environments is ambiguous, any in a non-negative polarity environment is not. That is, any case where an ‘existential’ interpretation of any is available can always be over-ridden by a focusing device such as contrastive stress, giving rise to ambiguity.”
(p.80) Kempson's invocation here of contrastive stress has pointed us to the fact that stress contrasts are relevant to the ambiguities of NPI any forms as well as to those of other nominal NPIs. Consider:
(4) Vaughn didn't accept to write anything about radiation.
Example (4) has two readings involving the differential scope of the DP represented by anything. On one reading, associated with weak stress on this form, its scope is internal to the complement. On this reading, (4) paraphrases (5):
(5) Vaughn didn't accept to write something about radiation.
In our terms, (4) is a reversal case; the initiator of the NEG deletion chain involved in the covertness of the two NEGs underlying any is taken to be the negated main verb [NEG accept]. A scenario where such an interpretation is natural is one where the chief editor of the paper is handing out topics for people to write on. She asks Vaughn to write something about radiation, but he refuses. Later on, she reports his reaction using (4).
A second reading of (4) is associated with strong stress on anything; the scope of the quantifier is in the main clause. On this reading, (4) is equivalent to (6):
(6) There is nothing about radiation that Vaughn accepted to write.
In terms of the assumptions and notations adopted in this monograph, the relevant aspects of this reading of (4) would be represented most fully as follows:
(7) Vaughn did NEG1 [<[[<NEG1> SOME] thing]5> [accept to write DP5]]
The key observations about (4) on the reading represented in (7) are the following: (i) accept is not a CNRP, (ii) no reversal structure is involved, and (iii) the surface position of the DP with main clause scope is in the complement clause. Despite these properties, a structure like (7) nonetheless yields an output with the NEG raised to a position immediately adjacent to the matrix clause Aux from the higher occurrence of DP5. On its reading equivalent to (6), the structure of (4) thus involves no raising of a NEG out of the complement clause. This is possible because under our assumptions, the main clause NEG in (7) starts out in the DP in a main clause scope position. The NEG can then raise out of the DP in the high-scope position internal to the main clause. No Classical NR is present in (7) or required to account for its properties.
In examples like (8), analogous to (4) except that the main verb is replaced by a CNRP, the situation is slightly more complex:
(8) Vaughn didn't want to write anything about radiation.
(9) Vaughn1's wish was that he1 write nothing about radiation.
Now we turn to cases of infinitival complements containing strict NPIs:
a. *Virginia forced/ordered/told Carmen to say jackshitA about compilers.
b. Virginia didn't force/order/tell Carmen to say jackshitA about compilers.
At first glance, perfectly grammatical examples like (10b) might appear to reduce our claim that strict NPIs like jackshitA require local licensers to non-sense.1 In these examples, the putatively strict NPI occurs in a complement clause, while the NEG, shown by (10a) to be the critical licensing element, is in the main clause. And, independently of NPIs, these verbs of course show no equivalences of the sort typical of Classical NR cases:
a. Virginia didn't force/order/tell Carmen to sing.
b. Virginia forced/ordered/told Carmen not to/to not sing.
Since (11a,b) represent entirely distinct propositions, there is no basis for invoking Classical NR in (11a) or (10b).
Despite these uncontroversial facts, it nonetheless makes sense to speak of forms like jackshitA as requiring a local licenser, although the predictions of that claim need to be adjusted to take account of recent discussion of nonstrict NPI cases like those in (4). That is, since the main verbs in (11) are not CNRPs, a unary NEG associated with the NPI quantifier could only appear in the main clause if the scope position of that quantifier were in the main clause. Reversals are, in this case, irrelevant since we argued in chapter 4 that strict NPIs like jackshitA do not permit reversal structures. So if jackshitA is a strict NPI, the claim is derived that in cases like (10b), the scope of the quantifier DP must be in the main clause, as it is. In other words, to the extent that (10b) is acceptable, it means that there is nothing about compilers that Virginia ordered Carmen to say.
But in analogs of (10b) involving main clause CNRPs, the prediction is that both main clause and complement clause scope is possible. Consider then:
(12) Virginia didn't want Carmen to say jackshitA about compilers.
(p.82) As predicted, we believe that this has not only the high-scope reading parallel to that of (10b), that is, one paraphrasable as ‘There was nothing Virginia wanted Carmen to say about compilers’, but also a low-scope reading paraphrasable as ‘Virginia wanted there to be nothing that Carmen said about compilers’. On the latter reading, the NEG can achieve its main clause position by raising out of the DP in a complement clause scope position into the main clause. NEG raising from the complement clause poses no problem, since the matrix verb is a CNRP.
9.2 The Moral
For present purposes, the moral from this discussion of scope and NEG raising is the following. Any attempt to diagnose the presence or absence of Classical NR must always take into account the possibility of NEG raising out of main clause scope positions. More precisely, suppose one finds cases of the schematic form [NEG Verbx … [A … [strict NPI]a …]], where there is no NEG in the complement clause A and Verbx is not a CNRP. Such a case does not falsify our claims that the strict NPI requires a local (clausemate) licenser, because there is an analysis not invoking Classical NR consistent with the locality requirement of the strict NPI. But that analysis generates a testable consequence: namely, the scope of the quantifier corresponding to NPIa must be in the main clause, so that the NEG originating in the NPI structure can raise out of the high-scope position.2 That alone allows the situation to be described without any raising of NEG out of the complement of a non-CNRP.3
9.3 Applying the Moral to Finite Complements
The discussion in this chapter so far has involved nonfinite complement clauses. However, many cases where the issue of high-scope confounds arise in this monograph involve finite complements. We stress, then, that the same considerations yield the same conclusions as those reached so far for nonfinite complement cases. There is one important proviso, though. In some (perhaps many) cases for some (perhaps many) speakers, the high-scope reading of a negative quantifier DP in a finite complement clause is difficult to accept.
Consider first an example involving a nonstrict NPI:
(13) Rodney did not claim that Evelyn owned any cheetahs.
a. Rodney [NEG6 claim] [that [<[[<NEG5> [<NEG4> SOME]] cheetahs]1> [Evelyn owned DP1]]]
b. Rodney did NEG4 <[[<NEG4> SOME] cheetahs]1> [claim that [Evelyn owned DP1]]
Structure (14a) is the now familiar reversal case in which the surface NEG in the main clause originates on the matrix V/VP, and both NEGs of the reversal structure are deleted, with the main clause negative verb being the initiator of the relevant NEG deletion chain. Structure (14a) specifies the reading of (13) on which it is equivalent to (15), where the scope of some is internal to the complement:
(15) Rodney did not claim that Evelyn owned some cheetahs.
A natural context for (13) on interpretation (15) is as follows. One speaker says, “Rodney claimed that Evelyn owned some cheetahs.” The addressed speaker denies this by saying, “No, Rodney didn't claim that Evelyn owned any cheetahs, and Frank didn't claim that either.” Here, the antecedent of that appears to be Evelyn owned any cheetahs, where any cheetahs is a reversal.
Structure (14b) gives the high-scope reading on which (13) is equivalent to (16):
(16) There were no cheetahs which Rodney claimed that Evelyn owned.
Such an interpretation is brought out by putting stress on any in (13). It is consistent with a scenario where one speaker says, “Rodney claimed that Evelyn owns the cheetahs Simba and Nala.” The addressed speaker denies the claim by uttering (13). On such an interpretation, it is awkward to continue with “… and Frank didn't claim that either.” On the awkwardness of the continuation, see the paragraph following (18).
The interpretations in (15) and (16) are not logically equivalent. For example, (16) is true if there are no instances of a cheetah x, where Rodney claims that Evelyn owns x. However, even if (16) is true, it may also be the case that Rodney has steadily maintained that Evelyn does in fact own cheetahs, but there is no specific cheetah that he claims that she owns. Since (16) can be true, while (15) is false, they are not logically equivalent.
Return then to the real focus of the issue being addressed, represented by pairs like (1a,b), repeated here:
a. Andrea doesn't believe/think that Carl said jackshitA about compilers.
b. *Andrea doesn't accept/realize that Carl said jackshitA about compilers.
(p.84) Given the preceding discussion, we can be precise about our view of purported contrasts like (17a,b). Since the occurring NPI does not permit a reversal analysis, we recognize only two a priori possibilities in such cases: a low-scope analysis of the [[NEG SOME] jackshitA] quantifier DP, or a high-scope one. But the former analysis is only possible in (17a) given the main clause locations of the NEGs, since that locus requires that a NEG raise out of the complement clause. And that requires the main verb to be a CNRP, which leaves only a high-scope reading as a possibility for (17b). We must then ask, what is the status of that reading, and how does it relate to our placement of a star on (17b)?
Our partially historical answer is this. We now believe that the high-scope reading is in general grammatical for cases like (17b), although at one time we were of a different opinion and affixed stars to them without explication. When the high-scope reading is excluded, as we have just observed, the result is indeed ungrammatical. The confounding factor then is that grasping high-scope readings is often not clear in finite clause cases, especially with NPIs like jackshitA, a fact for which we have no account. But to evaluate the status of examples like (17a,b) and of putative contrasts between them, the scope factor can never be ignored.
Since, as just indicated, scope intuitions relevant to the present issues are often rather subtle, showing that a particular example has only main clause scope for a complement quantifier DP is not straightforward. That difficulty is, we suggest, the key reason why judgments of grammaticality for cases like (17b) have proven to be wavering and difficult over time, and not only for us.
In support of the wide scope reading of the quantifier in (17b) (to the extent that it is acceptable), consider the following contrast:
a. Andrea didn't claim that Carl said jackshitA about compilers, *but Mary did claim that.
b. Andrea didn't claim that Carl said anything about compilers, but Mary did claim that.
In (18a), jackshitA needs to have scope in the main clause, as discussed above. But then, the embedded clause (whose semantics contains a free variable) cannot serve as the antecedent of that. However, in (18b), anything as a reversal takes scope in the embedded clause, and the embedded clause is a legitimate antecedent of that.
We observe, though, that there are strict NPIs for which the considerations discussed above are irrelevant, because the relevant NPIs do not scope out of (p.85) finite clauses. For such NPIs, we can safely affix stars without complications. The NPI series in ages/days/weeks/months/years … is of this type. Consider the following:
a. Anthony hasn't visited his mother in ages.
b. Not in ages has Anthony visited his mother.
c. Gloria doesn't believe/think that Anthony has visited his mother in ages.
d. *Gloria doesn't know/realize that Anthony has visited his mother in ages.
In cases like (19c,d), for whatever reasons, there is no issue of high scope for in ages. Therefore, the contrast between the grammatical CNRP main clause cases in (19c) and the non-CNRP main clause cases in (19d) argues directly that the NEG associated with and required by in ages, seen in the main clauses in (19a,b), has raised out of the complement clause in (19c). It follows that Classical NR is licit in (19c) but not in (19d), because the main clause is based on a CNRP only in the former.
A second example of a strict NPI for which confounding scope factors are irrelevant is provided by the expression so good/great/hot:
a. Virginia is *(not) feeling so hot.
b. Luke doesn't believe/think that Virginia is feeling so hot.
c. *Luke doesn't know/realize that Virginia is feeling so hot.
Again, since a high-scope analysis is impossible, the clausal separation of the NEG and the NPI in (20b) must depend on Classical NR. Since that is also impossible in (20c) owing to the lack of a main clause CNRP, the relevant examples are predictably ungrammatical.
A third and last example of a strict NPI not subject to the confounding scope issue is the adjectival modifier all that:
a. Arnold is *(not) all that intelligent.
b. Lucinda doesn't believe/think that Arnold is all that intelligent.
c. *Lucinda doesn't know/realize that Arnold is all that intelligent.
So far, we have discussed confounding high-scope issues only for cases in which a strict NPI quantifier DP is associated semantically with an overt NEG located in the post-Aux position in the main clause. But (2), which we repeat here, presented the distinct case where the forms permitting complement clause strict NPIs are main clause quantifier DPs:
a. No expert believes/thinks that Carl said jackshitA about compilers.
b. *No expert accepts/states that Carl said jackshitA about compilers.
(p.86) We believe that exactly the same considerations just discussed pertain here, except that ascertaining the scopes is complicated by the possibility of forming a polyadic quantifier combination of the main clause quantifier DP and complement clause NPI DP. But abstracting away from difficulties in perceiving such readings, characteristic for at least some speakers, other speakers find that (22a) has both a low-scope, Classical NR reading and a high-scope one for jackshitA. The former is equivalent to ‘Every expert thinks Carl said nothing about compilers’ (this reading is discussed at length in chapter 16). The latter is equivalent to ‘There is no pair <x, y>, x an expert and y a thing, such that x thinks Carl said y about compilers’.
But allowing for the hedges just above, (22b) has only the analog of the high-scope, polyadic reading, since Classical NR is unavailable with the main clause non-CNRPs. So a star like that on (22b) should, strictly, only mark the low-scope reading for those who accept the high-scope possibility.4
9.4 Classifications of NPIs
It is appropriate at this point to reconsider the terminology for NPIs we have adopted and its relation to other terminologies found in the massive NPI literature.
In the framework of this monograph, NPIs are classified according to whether they represent unary-or binary-NEG structures. For example, the NPI variant of jackshitA can only represent a unary-NEG structure. The NPIs anybody and ever can, in distinct environments, represent either type of structure. There are certain contexts where anybody can only represent a unary-NEG structure (e.g., Horn clauses; see chapter 13) and others where it represents only a binary-NEG structure.
There are subdivisions in the class of unary-NEG NPIs. For example, we claimed earlier in this chapter that certain unary-NEG NPIs can take matrix scope out of a finite clause, and others (such as in ages, until, so good/great/ hot, all that, and half bad) resist taking scope out of a finite clause. There are also divisions in the class of binary-NEG DPs, which we have not discussed in this monograph for reasons of space.
In chapter 1, we introduced the term strict NPI for those NPIs that need a local licenser. This induces a classification of NPIs into strict and nonstrict NPIs. An obvious problem with this definition of strict NPIs is that the present framework does not actually incorporate a notion licenser. Instead, it maintains the view that NPIs originate with at least one NEG, that NEGs often raise away from particular NPIs, and that many NEGs originating in NPIs are deleted by NEG deleters. In other views of NPIs, the cases we differentiate in (p.87) these terms are normally conflated. Specifically, we know of no other view of NPIs in which a NEG can raise from its associated NPI, since in these views no NEG originates in an NPI. Rather, nominal NPIs are standardly treated as indefinites, under the scope of a decreasing or at least nonincreasing operator.
The question then arises of how the notions unary/binary-NEG NPI map onto the notions strict/nonstrict NPI. There are three cases to consider:
a. NEG raising from a unary-NEG structure,
b. deletion of the NEG of a unary-NEG structure, and
c. deletion of the two NEGs of a binary-NEG structure.
The possibility in (23a), the raising of NEG from a unary-NEG structure, is highly restricted. NEG raising cannot cross either an island boundary or a non-CNRP. In most cases, such NEG raising is clause-bounded; CNRPs are the exception since these permit Classical NR and hence the raising of NEGs out of clauses. Apparent counterexamples of raising over a non-CNRP discussed in earlier sections of this chapter were analyzed such that the NPI takes matrix scope. Hence, the relevant NEGs can raise from the matrix scope position of the NPI. We need not repeat justification for those conclusions here.
Case (23b) involves three subcases: lexical NEG deletion, NEG deletion in polyadic quantification, and a last possibility discussed at length in chapter 16. Such examples are narrowly constrained by the fact that systematically, the NEG deleter and the NEG it deletes must be clausemates.
Finally, consider (23c). The deletion of the NEGs of a binary-NEG structure is fundamentally different from the other cases. Such deletion is in a sense nonlocal. We described two alternative views of this in chapter 8. One possibility is that the NEGs of a reversal structure never undergo raising, but are simply deleted long-distance. Another possibility is that the outer NEG of a reversal structure does undergo raising, but that such raising is not subject to all island constraints and is not subject to condition (4) of chapter 1, that limiting NEG raising to dominating clauses based on CNRPs. Under either analysis, it is clear that the deletion of the NEGs of a reversal structure is nonlocal in a sense that the deletion of other kinds of NEGs is not.
Given this asymmetry between unary-and binary -NEG structures, we make the following identification, linking terms previously used in the literature on NPIs with our present theoretical framework:
(24) Strict NPIs = Unary-NEG structures
Nonstrict NPIs = Binary-NEG structures
(p.88) Of course, much more needs to be said here, especially concerning the classification of binary-NEG structures. But to a first approximation, (24) appears to us to be correct. NPIs that require a local licenser (in the traditional conception) are all unary-NEG NPIs.
We can also relate our views to the very influential division of NPIs into weak, strong, and superstrong, given by Zwarts (1998):
(25) (Zwarts 1998:233)
“Laws of Negative Polarity
a. Only sentences in which a monotone decreasing expression occurs can contain a negative polarity item of the weak type.
b. Only sentences in which an anti-additive expression occurs can contain a negative polarity item of the strong type.
c. Only sentences in which an antimorphic expression occurs can contain a negative polarity item of the superstrong type.”
Putting aside (25c), this classification maps neatly onto our classification of NPIs into unary-NEG structures and binary-NEG structures. Consider the following examples:
a. At most half of the students know anything/*jackshitA about physics.
b. No student knows anything/ jackshitA about physics.
Given such data, Zwarts's conceptual framework would classify jackshitA as a strong NPI, since it is only possible with an antiadditive such as no student, and hence is not possible with the decreasing but not antiadditive at most half of the students. In the present framework, jackshitA is a unary-NEG NPI analyzed underlyingly as [[NEG SOME] jackshit]. The NEG cannot be deleted in (26a) because of the evenness requirement represented by condition (18) of chapter 8 and the Antiadditive NEG Deletion Condition of note 6 of chapter 16. Furthermore, no polyadic quantification structure is possible in (26a), since no determiner can be shared. In (26b), the NEG is deleted because no student and jackshitA form a polyadic quantification structure, as described in chapter 6.
Given these considerations, we postulate the following:
(27) Weak NPIs = Binary-NEG structures
Strong NPIs = Unary-NEG structures
The only category left out is superstrong NPIs. We are not aware of a clear case of such an NPI in English. But any existing superstrong NPI in any language would be characterized in present terms as a unary-NEG structure that (p.89) prohibits determiner sharing (and is hence unable to enter into polyadic quantifier formation). The NEG of any such unary-NEG structure would raise away from the NPI, giving the appearance of an NPI licensed only by NEG (and hence giving the appearance of an NPI licensed only by an antimorphic expression).
Putting (24) and (27) together and glossing over many issues yields the following linking of our unary/binary distinction with terms that have been used by others in the literature (strict/nonstrict and weak/strong):
(28) Unary-NEG NPIs = Strict NPIs = Strong NPIs
Binary-NEG NPIs = Nonstrict NPIs = Weak NPIs
9.5 Unexpected Contradictions
In this section, we briefly consider certain contradictions at first sight unexpected under the high-scope analyses of various strict NPIs we have appealed to in the first sections of this chapter.
Consider the following:
(29) She didn't ask me to do anything.
In the present framework, (29) can have two structures, neither of which can invoke Classical NR since ask is not a CNRP:
a. She did NEG1 [<[[<NEG1> SOME] thing]5> [ask me to do DP5]]
b. She did NEG1 [<NEG1> ask] me [<[[<NEG2> [<NEG3> SOME]] thing]4> to do DP4]
These structures have the following rough paraphrases:
a. There is nothing that she asked me to do.
b. She didn't ask me to do something.
These interpretations are not logically equivalent. For instance, in a workplace scenario, (31a) is true if there is no specific task (among those in my job description) that I have been assigned on day X. But even so, I may have been asked to keep busy on day X by choosing some task on my own. So (31a) can be true, while (31b) is false. The relevant interpretation of (29) is brought out by adding in particular to anything: She asked me to keep busy, but she did not ask me to do anything in particular.
Our assumptions predict, though, that if the nonstrict NPI anything in (29) is replaced by a strict NPI, there can only be one interpretation:
(32) She didn't ask us to do jackshitA.
(p.90) Under our view, only a structure parallel to (30a), with an interpretation like (31a), should be possible in (32). That follows from the claim that strict NPIs like jackshitA are incompatible with reversal structures, which is what are found in (30b). Recall from section 4.8 the independent evidence for the conclusion that strict NPIs like jackshitA cannot represent binary-NEG structures:
a. *If you do jackshitA, you will be in trouble.
b. *Everybody who knows jackshitA will pass the exam.
c. *Less than half of the class did jackshitA.
Since only binary-NEG NPIs are possible in contexts like (33a–c) and since jackshitA is disallowed in all of them, we concluded that jackshitA can never manifest a binary-NEG structure.
Consider next the following example, illustrating what we will call the contradiction test (suggested to us by Dylan Bumford):
(34) Although he did suggest we do something, he didn't ask us to do jackshitA.
Limiting attention to the first-clause scope order suggest > something, this sentence seems to express a contradiction. The problem is that our views do not predict that fact. If the main clause in (34) is assigned a structure parallel to (30a), hence the interpretation in (31a), no logical contradiction should exist.
This result seems to suggest that the only structure for the main clause of (34) is parallel to (30b). Then the contradiction in (34) would follow straight-forwardly (since the whole would reduce to an instance of P contradicting ¬P). But assigning the main clause of (34) a reading parallel to (30b) is evidently inconsistent with our conclusion that NPIs like jackshitA are limited to unary-NEG structures.
Furthermore, even if (contrary to our analysis) jackshitA could have a binary-NEG structure, nothing should block (34) from also having a structure parallel to (30a), where jackshitA has a unary-NEG structure, with wide scope. On that structure, (34) should not be a contradiction.
Although we cannot give a full account of these facts, we suggest that there is a strengthening (perhaps an implicature) operating in cases such as (34) that can be characterized somewhat as follows: (i) [[NEG SOME] jackshitA] in (34) has wide scope over ask; (ii) [[NEG SOME] jackshitA] is interpreted as a quantifier with a maximally inclusive restriction (the set of all individuals in any model), yielding a strong statement; (iii) that statement is taken to be strengthened even further to ‘He made no request that we do something’. (p.91) The pragmatic basis of the extralogical strengthening in (iii) might be found in the fact that while on strictly logical grounds, cases like (34) are consistent, the circumstances that would permit a true model of them are quite rare.
There is evidence for the sort of pragmatic strengthening just alluded to outside the domain of NPIs. Consider:
(35) Although there is nothing he suggested we drink, he did suggest we drink something.
In this case, the there construction guarantees the high scope of nothing. On purely logical grounds, given that something in the second clause can have low scope, this too should be consistent. But a consistent reading does not seem like a normal use of the expression; making it consistent again requires assuming an extremely unlikely model. Genuine acceptability, or at least real naturalness, for a consistent version of something like (35) seems to demand expanding nothing with forms such as in particular or specific.
While (34) seems worse than (35), perhaps this is related to the fact that restrictive phrases like in particular cannot be appended to jackshitA. This could suggest that (35) is more easily taken as consistent than (34) because it involves a marginal covert version of one of the restricting expressions, a possibility not existing for the strict NPI under a hypothesis that such phrases reject both overt and covert versions of restrictive phrases. (p.92)
(1) . A parallel remark could be made about the negative force verb cases like (i) of note 1 of chapter 8.
On the assumption that in at least n years is a strict NPI, (ia,b) provide two challenges for views of strict NPIs. First, why is (ia) grammatical, since the context between the main clause NEG and the NPI contains no CNRP(s)? Second, why does whatever permits (ia) not equally permit (ib)?
The obvious suggestion in current terms is that (ia) represents a high-scope structure with the combination of the NEG and in at least two years scoping over everything to the right of the main clause Aux. This would yield a meaning for (ia) representable as something like (ii):
((ii)) ‘There is no point in time t where t is a point in time less than two years before the present moment such that it is permissible for an applicant to have left the country at t.’
This seems to us like a plausible candidate for the meaning of (ia). If so, the view that (ia) is not a Classical NR case can be justified.
Then the question is why a parallel analysis is not available for (ib). We note first that the strict NPI until is possible in both contexts:
This suggests that what is wrong with (ib) has specifically to do with the meaning of in at least n years.
(p.230) Support for that view and for the claim that the (ia,b) contrast shows nothing general about the nature of strict NPIs is provided by the grammaticality of both pairs of strict NPI cases like those in (iv):
We have nothing further of substance to say about the (ia,b) contrast. Gajewski proposes a semantic account based on the fact that in (ia) the main clause context yields an antiadditive operator while that in (ib) does not. We advance further problems for a pure antiadditivity approach to strict NPIs in chapter 10.
Finally, if the cases cited in this note really involve high-scope analyses, then they do not directly bear on any analysis of Classical NR.
(3) . Vincent Homer (personal communication) observes contrasts like the following:
The issue he raises is why Classical NR, which is fine with the strict NPI a thing in (ia), fails in (ib). Homer suggests (as part of a general view of NPI licensing) that the ungrammaticality is due to the fact that the meaning of the external NEG in (ib) combines with the meaning of the prefixed negation un- to create an increasing context. We are unconvinced that (ia) represents Classical NR, but we cannot pursue this issue here.
(4) . The discussion in the text of the strict NPIs in ages, so hot, and all that raises the question of whether these NPIs fail to occur in scope positions at all and so are scope-less, or whether they do appear in scope positions but those scope positions are limited to the smallest finite clause containing the overt occurrence of the strict NPI. Our belief is that all these NPIs do occur in scope positions.
The only candidates we know of for NPIs not occurring in scope positions are verbal ones such as that involving modal need:
(ia) is a unary-NEG structure of the form [NEG need] (where on one analysis both NEG and need undergo various raisings to yield the surface structure in (ia)). However, we believe that (ib) is a binary-NEG structure, for the following reason. If need in (ib) were a unary-NEG structure, the mechanisms for polyadic quantifier formation in chapter 6 would be inapplicable, since need does not take scope. But under conditions where both need was a unary-NEG structure in (ib) and polyadic quantifier formation was impossible, there would be no way for the unary NEG in (ib) to be deleted. Therefore, the only remaining analysis is that need in (ib) is a binary-NEG structure.
Even though need does not appear in many typical reversal contexts, as shown in (ii), it appears in other contexts unavailable to strict NPIs, as shown in (iii):
Therefore, to maintain that need can appear in a binary-NEG structure (to account for (ib) and (iiia)), further constraints are necessary to rule out (iia–c). We cannot pursue this issue here.
This note highlights the issue of scopeless NPIs, which so far remains uninvestigated. It also highlights the issue of the classifi cation of NPIs based on the various tests given in this monograph. In some cases, as with need, the classifi cation of an NPI as either a unary-NEG structure or a binary-NEG structure is not as straightforward as might have been expected given our framework.