1 Quantification in the Clause
There’s a family of proposals now suggesting that the quantificational force for noun phrases is associated with heads situated higher in the extended projection of the clause (Beghelli and Stowell 1997, Szabolcsi 1997, Hallman 2000, Sportiche 2005). While these proposals differ in their implementation,1 all converge on the expectation that the absence of such projections limits the scope options for quantified nouns. Sportiche (2005) points out that this view may offer a way to understand why the small clause subject (SCS) in (1a) fails to take scope under the raising verb seem, in contrast to the subject raised from the infinitive in (1b).
(Williams 1983:293, (40a))
These data were used by Williams (1983) to argue that the SCS in (1a) does not raise from a Stowellian small clause (Stowell 1983). On the view that nominal quantification is dependent on some amount of clause structure, an SC analysis remains tenable: SCs are simply too small to house the heads that license nominal quantification. What has gone unnoticed is that SCSs can have lowered scope interpretations and that this depends on the choice of predicate in the SC. When the SC predicate is a modal adjective like necessary, the SCS can be interpreted in the scope of that adjective, and as a result within the scope of the embedding verb. No such lowering is possible with a nonmodal adjective, like available or sick.
A new solution seems necessary. But none presently exists.
A new solution seems available. #But none presently exists.
On all viable theories of scope reconstruction—whether syntactic (Fox 1999, Sauerland and Elbourne 2002) or semantic (Cresti 1995)—the lowered interpretation in (2a) is tied to movement.2 This means that SCs are raising constructions, and as a consequence the lack of narrow scope in (1a) and (2b) cannot be attributed to lack of raising.
The difference is conditioned, instead, by the nature of the positions to which the SCS can reconstruct (see Hallman 2004 on the diversity of such positions). Adjectives like available and sick denote garden-variety properties of individuals. Modal adjectives (MAs) such as necessary, I argue, are like intensional transitive verbs, particularly in embedding covert clausal material (McCawley 1974, Larson, Den Dikken, and Ludlow 1997). This covert clause is large enough to contain at least some of the functional structure that licenses quantification. The SCS in (2a) is interpreted in this covert clause and raises from it. SCs, as always, reveal the functional scaffolding of the clause, even when those clauses go silent.
2 Scope Reconstruction in Small Clauses
Williams’s observations extend to a range of scope-taking SCSs (Heycock 1995). These include numeral indefinites, which give rise to scope interactions that are more easily truth-conditionally distinguishable.3
Scope reconstruction is possible, however, when the SC contains an MA, like necessary, likely, or required. A range of indefinites, including existentially interpreted some and at least, fall in the scope of the MA in the SC.
In virtue of being interpreted narrowly with respect to the embedded predicate, these SCSs are interpreted narrowly with respect to the embedding predicate.4
Scope reconstruction is also shown by ‘‘split’’ readings of negative quantifiers. Negative quantifiers appear to be composed of two parts: sentential negation and existential quantification (Klima 1964, Ladusaw 1992, Larson, Den Dikken, and Ludlow 1997). That these two components can obtain independent scopes is diagnosed by the intervening modal in (5). SCs that contain an MA allow a split reading (6).5
≈ It isn’t required that a book about Nixon be written
(Iatridou and Sichel 2011:606)
More evidence that SCSs undergo scope reconstruction—when there is an MA in the SC—is to be found in negative polarity item (NPI) licensing. An NPI can be licensed by non- (surface) c-commanding negation when its containing phrase takes scope below this negation.
A doctor who knew anything about acupuncture wasn’t available.
(Linebarger 1980:227, (21a))
When an SCS takes inverse scope below a negated MA, an NPI within it is licensed.
A doctor who knows anything about acupuncture seems unnecessary.
*?A doctor who knows anything about acupuncture seems not available.
If scope reconstruction is necessary for NPI-licensing here, SCSs must undergo scope reconstruction in (8a).
Scope reconstruction can feed the evaluation of Condition C (Lebeaux 1988, Romero 1998, Fox 2000). In (9), a disjoint reference effect obtains if the quantified expression is interpreted in the scope of the raising verb (an interpretation promoted by the choice of a creation verb).
(9) For these issues to be clarified,
many more/new papers about his1 philosophy seem to Quine1 to be needed.
#many more/new papers about Quine1’s philosophy seem to him1 to be needed.
(Fox 2000:9, (18))
Likewise, scope reconstruction into SCs feeds Condition C. A modal adjective promotes scope reconstruction, and this gives rise to a disjoint reference effect. A nonmodal adjective (outdated, capable), however, forces wide scope, and no Condition C effect obtains.
#Papers about Quine1’s philosophy seem necessary to him1.
Papers about Quine1’s philosophy seem outdated to him1.
#A new student of David1’s seems necessary to him1.
A new student of David1’s seems very capable to him1.
If Condition C effects diagnose movement, then at least the NP portions of these SCSs must originate within the SC. What is the difference between MAs and other adjectives such that the former allow SCSs to reconstruct?
3 Modal Adjectives and Covert Clauses
3.1 Evidence for a Covert Clause
MAs, unlike garden-variety intersective adjectives, are intensional operators that interact scopally with DPs.6 In this section, I show that MAs are like intensional transitive verbs (ITVs) such as want. Like ITVs (12a), MAs have semantically opaque object positions (12b). Familiar tests, like failure of extensional substitution and lack of existential import, diagnose that the objects in (12) fall in the ‘‘scope’’7 of the ITV and the MA (Quine 1960, Montague 1974).
My advisor wanted a hat like mine, thinking I owned a fedora.
A hat like mine is necessary.
One analysis of ITVs posits a covert, clause-sized complement equivalent to that expressed with have in (13b) (McCawley 1974, Ross 1976, Larson, Den Dikken, and Ludlow 1997).8 The covert clause analysis is supported—for at least some ITVs—by the ambiguity of temporal modifiers.
A week ago Bill wanted your car tomorrow.
A week ago Bill wanted to have your car tomorrow.
*A week ago Bill painted your car tomorrow.
(McCawley 1974:2, (4))
In (13a), there are two temporal modifiers, the first modifying the verb want, the second an implicit clause, overtly expressed in (13b). A regular transitive verb, like paint in (13c), allows no such possibility.
Similar conclusions hold for MAs. The examples in (14)–(16) allow a variety of MAs to be modified by one temporal modifier and allow another to modify covert content. They behave just as their paraphrases with overt propositional complements do.
Until the funding cut, two more seminars were possible next term.
Until the funding cut, it was possible to have two more seminars next term.
Yesterday, a new grill was necessary tonight. We no longer need one.
Yesterday, it was necessary for there to be a new grill tonight. We no longer need one.9
Right now a storm is likely tomorrow.
Right now a storm is likely to happen tomorrow.
Propositional anaphora provide further evidence that MAs take a covert clausal complement. McCawley (1974) reports that it in (17a) refers to having horses, just as it does when such a clause is overt (17b). MAs also support propositional anaphora.
Joe wants some horses but his mother won’t allow it. (McCawley 1974:6, (10b))
John wants to have some horses but his mother won’t allow it.
Some paperclips are necessary, but the budget won’t allow it.
It’s necessary that we have/there be some paperclips, but the budget won’t allow it.
3.2 Scope of Negation
Countenancing a clausal complement under MAs means, then, that they are raising constructions.10 The fact that an MA can outscope its subject follows from a raising analysis: the subject takes its scope within an embedded clause. This in turn explains restrictions on the kinds of quantificational expressions that can be interpreted low—restrictions that hold generally of raising constructions (Lasnik 1999, Iatridou and Sichel 2011).
Certain ITVs, like want and active need, can put the negative component of a NegDP in their scope.
I’m trying to finish my paper this weekend so I need no visitors.
(attributed to Irene Heim and Kai von Fintel; see Schwarz 2006:266)
MAs do not permit NegDP subjects to be interpreted in their scope (19a)/(20a). This is true even when there is an overt raising complement to reconstruct into (19b)/(20b).11 Compare these with (19c)/(20c), where a NegDP in the finite complement takes narrow scope.
(19) I’m trying to finish my paper this weekend . . .
#So no visitors are required/necessary.
#So no visitors are required/necessary to be here.
So it is required/necessary that no visitor be here.
(20) Don’t worry about having any food . . .
#No one is certain/guaranteed.
#No one is certain/guaranteed to come to your party.
c It is certain/guaranteed that no one will come to your party.
This is expected if MAs embed a covert clausal complement.12
We began with the hypothesis that SCSs don’t reconstruct because SCs don’t contain the clausal structure necessary to license nominal quantification. This led to a prediction: when quantifiers are interpreted in SCs—as documented here with MAs—that option must be provided by clause-level material introduced by the adjective, not by the SC per se. And that prediction was borne out: as the argument in section 3.2 shows, MAs embed covert clausal material. Whatever the exact implementation (see footnote 1), these observations recommend a theory in which the heads that license quantificational noun phrases (like the Qs in (21)) are situated in the clausal spine. SCs lack such heads, and noun phrases must associate with a Q in the matrix clause (21a). MAs, on the other hand, embed covert clauses that are large enough to house Q heads (21b).13
[ . . . Q . . . seem [AP (*Q) at least one student sick]]
[seem [AP necessary [XP . . . Q . . . at least one fridge]]]
Here, then, is an account for both Williams’s (1983) case in (1a)—itself a long-standing obstacle for SC analyses—and those cases where scope reconstruction is possible. The resolution hinges, however, on taking seriously the idea that the quantificational force for nouns is licensed by clausal material. SCs remove that material, letting us uncover another source: surprisingly, silent clauses are larger than small but overt ones.14
4 Strong Quantifiers and the Content of Covert Complements
There is one respect in which the MAs discussed above do not pattern either like their counterparts that take overt clauses, or like certain ITVs such as need and want. These can put a strong quantifier like most in their scope (Schwarz 2006), as in (22a). MAs that take covert clauses, however, cannot put most in their scope (22b). Compare with the overt finite complement in (22c).15
This is not due to any restriction against interpreting most below raising adjectives. This is possible when likely takes an overt raising complement (23a) (after von Fintel and Iatridou 2003). But when likely takes a covert clausal complement (23b), most cannot scope low.
The infelicity of (23b) may, however, reveal a restriction on the content of the covert clausal complement. While ITVs like need and want appear to select for clauses with have or get (McCawley 1974, Larson, Den Dikken, and Ludlow 1997) or the abstract primitive HAVE (Harley 2004), the content of the covert clause selected by some MAs appears to be restricted to an abstract BE. That is, when the context merely requires the complement to express an existential commitment, a covert clause is possible.
(24) What do you expect to happen?
A coup is likely/guaranteed/certain.
A coup is likely/guaranteed/certain to happen.
It’s likely/guaranteed/certain there will be a coup.
If the content of the covert clause needs to have richer content—say, topple—the result is odd, even with an indefinite subject.
(25) What will topple the government?
#A coup is likely/guaranteed/certain.
A coup is likely/guaranteed/certain to topple the government.
If the covert content of MAs is that of an existential statement, as in (26a), this may explain why (25a) is not a very good answer to (25). Moreover, the restriction against strong quantifiers like most taking narrow scope is reduced to that found in overt existential statements (26b) (e.g., Milsark 1974).
likely [THERE BE a coup/*most students]
There was a coup/*most coups.
A covert existential clause resolves the tension between the facts that support a property analysis (wide scope for most) and the evidence in favor of a covert clause (neg-split, temporal modification, propositional anaphora). Whether this is merely a syntactic variant of the property analysis (see footnote 11) I leave to further investigation.16
If a covert clause analysis can be maintained, we’ve uncovered a surprising thing: there are many more raising adjectives in English than previously recognized. The puzzle is why possible and necessary and perhaps others admit a raising complement only when that complement goes silent.
I thank Bernhard Schwarz, Junko Shimoyama, Kyle Johnson, Tim Stowell, Dominique Sportiche, Orin Percus, Winnie Lechner, two LI reviewers, and audiences at NELS 41, GLOW, and UCLA.
1 The proposals differ in whether the quantificational force itself is introduced higher (Sportiche, Hallman) or whether quantificational force is merely licensed in various high positions (Beghelli and Stowell). I will use the term license without committing to either implementation. Indefinites have long been treated this way (Heim 1982). I leave decisions about the compositional implementation of such approaches—alternative semantics, (un)selective binding, choice functions—for another occasion.
2 A nonmovement account could compose the matrix verb with the embedded complement via function composition (Jacobson 1992). With certain assumptions about modal adjectives, this could give the SCS a lowered interpretation (Moulton, to appear), but at the expense of introducing compositional mechanisms into the grammar.
3 It is hard to tell whether universals take narrow scope because the embedding predicate is a universal (modal) quantifier. Nonetheless, speakers often perceive a difference between Every student seems sick and Every student seems to be sick.
4 While one may be tempted to see (ia) as a case of reconstruction into the SC, this likely involves a generic operator outside the SC (Heycock 1995). This is corroborated by the fact that DPs that resist being understood as generics— like at least–NPs—only have a wide scope, existential interpretation as SCSs (ib).
A BMW seems expensive.
At least one BMW seems expensive.
5 The neg-raising properties of appear/seem obscure their relative syntactic scope with respect to negation (Iatridou and Sichel 2011).
6 These adjectives are exceptional in this respect. A reviewer points out a possible connection to ‘‘unaccusative’’ adjectives (Cinque 1990).
7 The terms de dicto/de re are often used in this regard. Here, we are just looking at quantifier scope, to be distinguished from the transparent/opaque distinction (Fodor 1970).
9 Of course, the necessity statement as a whole will be true at the time at which its prejacent may hold.
10 Not all MAs take an overt raising complement ( possible); some speakers find necessary acceptable with an overt (tough-construction) complement: A new fridge is necessary to have.
11 A property analysis (Zimmermann 1993) also accounts for the lack of low negation, as it does for those ITVs like look for for which a covert clausal analysis is less accepted (Partee 1974). MAs could also be given a propertytype analysis.
〚necessary〛 = λ P〈e,st〉.λs. ∀s′ ∈ Nec(s) [∃x [P(x)(s′)]]
〚A fridge is necessary〛 = ∀s′ ∈ Nec(s0) [∃x [ fridge(x)(s′)]]
This analysis would have to say more to account for the temporal modification facts and complex quantifiers like at least and more than n (see Van Geenhoven and McNally 2005). The neg-split cases would be a problem, too: putting the existential portion of a NegDP into the denotation of an adjective is not compatible with the morphological composition of NegDPs (Zeijlstra 2004).
12 If, as Iatridou and Sichel (2011) argue, the (un)availability of NegDPs taking narrow scope is correlated with neg-raising, then these data provide no evidence for covert complements under MAs, none of which are neg-raisers.
13 A reviewer points out a further prediction: SCSs cannot take intermediate scope, below the matrix predicate but above the MA (I believe two Green senators necessary). My impression is that this prediction is borne out, but more systematic testing is needed.
14 PPs must be large enough to house quantificational heads, since their objects can take scope within SCs (Williams 1983).
15 In this respect, MAs resemble ITVs like look for (Zimmermann 1993). The scope facts and the temporal modification facts, which usually pattern together, do not for MAs. I thank a reviewer for the contexts illustrating scope of most.
16 Of course, the question arises why strong DPs raised from this position that take wide scope (as in John’s tools/Most of his theorems are necessary) can leave traces compatible with the requirement of existential sentences.