In work on the tough-construction (TC), Hicks (2003, 2009) has advocated an analysis in which the TC matrix subject originates within the embedded infinitival clause and then moves to the matrix clause, recasting in Minimalist terms the early transformational analyses of Rosenbaum (1967) and Postal (1971) and rejecting that of Chomsky (1977), in which the TC subject is base-generated in the matrix clause and coindexed with an embedded null operator. In particular, Hicks proposes that the TC matrix subject undergoes A-movement, which he explicitly likens to raising, as its final step en route from the embedded clause to matrix subject position.
I show here that this proposed A-movement of TC matrix subjects parts ways with raising (and passive) with respect to scope reconstruction. It has long been observed that the TC matrix subject is unable to take scope below the tough-predicate ( Postal 1974, Epstein 1989), a fact that Hicks (2009:553) attributes to a ban on reconstruction of D heads in A-chains, citing Boeckx 2001. Boeckx’s ban has exceptions, however, and in cases where scope reconstruction in raising and passive A-chains is in fact possible, it remains unavailable for TC matrix subjects. These facts can be explained if we analyze the TC matrix subject as base-generated in the matrix clause, as on the copy-raising analysis of Rezac (2004, 2006). Alternatively, the unwanted readings might be ruled out as weak island violations on a Hicks-style raising analysis, an option that Hicks’s phrase structure makes available but that he does not discuss.
1 A-Movement and Scope of Tough-Construction Subjects
The TC derivation proposed by Hicks (2009) is sketched in (1) for the example John is tough to love. I have simplified certain details that are irrelevant for present purposes; the key points are that (a) the matrix subject is base-generated as the complement of a null operator within a complex DP that originates in embedded object position, (b) this complex DP undergoes Ā-movement to the embedded Spec,CP, and (c) the null operator’s DP complement then undergoes A-movement to matrix subject position.
(1) [TP Johni is [AP tough [CP[DP Op ti]j to love tj]]]
Hicks overcomes the central syntactic dilemma of the TC—namely, how to account simultaneously for the fact that there is evidence of Ā-movement in the embedded clause and for the fact that the understood filler of the embedded gap ends up in a Case position—by making the Ā- and A-chains in the TC distinct syntactic objects.
By proposing a movement derivation of the TC subject, however, Hicks opens the door to the possibility that TC matrix subjects might exhibit scope reconstruction, given the grammatical availability of scope reconstruction in A-chains (see, e.g., Fox 2000, Boeckx 2001, Lebeaux 2009, Iatridou and Sichel 2011). But TC matrix subjects have long been known not to take scope below the tough-predicate, as shown in (2).
Few girls would be difficult for Jim to talk to.
≠ It would be difficult for Jim to talk to few girls.
( Postal 1974:224)
Many people are easy to talk to.
≠ It is easy to talk to many people.
Nothing is hard for Melvin to lift.
≠ It is hard for Melvin to lift nothing.
( Postal 1974:356)
How many students are easy to teach?
≠ What number n is such that it is easy to teach n-many students?
In order to avoid making the erroneous prediction that TC matrix subjects should exhibit scope reconstruction, Hicks (2009:553) proposes that ‘‘an NP constituent of an A-moved DP may optionally reconstruct, whereas the D head cannot,’’ adding that ‘‘[t]he D head is what determines scope relations.’’ In this, Hicks invokes Boeckx (2001), who argues that D heads in A-chains must be interpreted in the position where they check Case. Boeckx’s proposal is designed to account for raising and passive A-chains, and Hicks (2009:554, 560) is elsewhere at pains to emphasize the similarities between his proposed A-movement of TC subjects and ordinary raising and passive.
Boeckx, however, recognizes several exceptions to the generalization that D heads in A-chains do not reconstruct. Indeed, a major focus of Boeckx 2001 is to provide an account of precisely which A-moved DPs exhibit scope reconstruction. Hicks (2009:553n11) cursorily acknowledges that indefinites exhibit narrow scope readings in A-chains—a phenomenon he attributes to ‘‘the meaning of indefinites’’ rather than to reconstruction, citing Lasnik 1999—but he overlooks Boeckx’s broader conclusion that the class of DPs exhibiting scope reconstruction in A-chains includes not just indefinites but all weak DPs.1 Witness the weak DPs from (2), which take only wide scope in the TC but can undergo scope reconstruction in raising and passive, as in (3). The scope reconstruction disparity between the TC subjects in (2) and the raising and passive subjects in (3) is problematic on its face for Hicks, who suggests that, for principled reasons, the former should mirror the latter.
Few girls are likely/expected to talk to Jim.
( few > likely/expected, likely/expected > few)
Many people are likely/expected to attend.
(many > likely/expected, likely/expected > many)
Nothing is likely/expected to please Melvin.
(neg > ∃ > likely/expected, neg > likely/expected > ∃)
How many people are likely/expected to win the contest?
(n-many > likely/expected, likely/expected > n-many)
Binding tests confirm that the narrow scope readings of the weak DPs in (3) are derived via A-reconstruction (or interpretation of a lower copy; I treat these as equivalent for present purposes). I illustrate with few in (4), modeled on examples with negative DPs from Iatridou and Sichel 2011:608.
Few books about himi are expected by Johni to be written next year.
*Few books about Johni are expected by himi to be written next year.
Few books about Johni are expected by himi to go out of print.
The creation verb rules out wide scope for the A-moved DP in (4a) and (4b): we cannot be talking about books that presently exist. The unacceptability of (4b), in turn, suggests that the narrow scope reading is derived via reconstruction, which induces a Condition C violation in this example. Absent the creation verb, wide scope is available, as in (4c).
A few comments on the examples with negative DPs and how many–DPs are in order. I follow Iatridou and Sichel (2011) in decomposing negative DPs like nothing in (3c) into a negative operator and an existential quantifier. As Iatridou and Sichel show, the negative operator’s scope is fixed, but the existential quantifier may undergo scope reconstruction in A-chains, yielding a ‘‘neg-split’’ reading. (See Iatridou and Sichel’s article for detailed diagnostics and argumentation.) For present purposes, it is important to note that even the existential portion of the negative DP is unable to undergo scope reconstruction in the TC; there is no neg-split with TC subjects, as shown in (5) and (6).
The scopal behavior of how many–DPs has been extensively discussed for cases of Ā-movement (Cresti 1995, Heycock 1995, Rullmann 1995, Romero 1998, Fox 2000). I follow Romero (1998:chap. 2) and Fox (2000:151ff.) in decomposing how many into an interrogative (how) and a quantificational determiner (n-many); the latter can interact with other scopal operators to yield ambiguities (see also Hackl 2000).2 Examples like (3d) show that scope ambiguities with how many are present in A-chains, as well. I sketch the two readings in (7); imagine that the contest in question is one in which there are five winning tickets among thousands made available to contestants.
When how many heads a TC subject, as in (2d) and the (a) examples in (8)–(10), only the wide scope reading is available. This contrasts sharply with what we find in the TC’s impersonal counterpart in the (b) examples (where no A-movement is implicated).3
The TC (8a) can only be interpreted as a question about particular students, not as a question about ideal class size;4 the impersonal (8b) has both interpretations. In (9), the semantic content makes the wide scope reading of how many pragmatically awkward; when the narrow scope reading is also unavailable, as in the TC (9a), the sentence is unacceptable.
Everything above points to a single conclusion: where scope reconstruction is available in canonical A-chains, it remains unavailable in the TC. Hicks’s (2009:553) attempt to explain obligatory wide scope for TC subjects in terms of Boeckx’s (2001) account of A-movement thus misses the mark, as it makes A-reconstruction of D heads out to be far more restricted than it actually is.
2 Excluding Scope Reconstruction: Copy Raising versus Weak Islands
The scope reconstruction disparity between TC matrix subjects and their raising and passive counterparts calls for an explanation. I consider two possibilities here, both of which are consistent with the view that there is an A-chain linking the TC matrix subject with the embedded clause. Independent evidence supporting the presence of an A-chain—and thus against proposals in the classic mold of Chomsky 1977—comes from experiencer-PP intervention effects noted by Hartman (2009), shown in (11).5
2.1 Tough-Construction A-Chains as Copy Raising
One possibility is to adopt the proposal of Rezac (2004, 2006) that the TC is a variant of copy raising (Rogers 1971, Potsdam and Runner 2001, Asudeh 2012:chap. 12). In copy raising, a nonthematic matrix subject is linked to a thematic embedded pronominal subject via an A-chain but not via A-movement: the matrix subject is base-generated in the matrix clause, not moved or copied from the embedded subject position. Among other things, this accounts for the matrix subject’s inability to undergo scope reconstruction in copy raising, as shown in (12) (Potsdam and Runner 2001:463).
Rezac (2004, 2006) proposes that DPs base-generated in nonthematic positions are interpreted via predication: their sisters are λ-abstracts with which they compose via function application. With the further assumption that the TC embedded-clause operator is a null pronoun (Browning 1987), Rezac (2006:302) concludes that ‘‘the nonthematic DP in both tough-movement and copy-raising λ-binds a pronoun, not a copy. Copy-raising wears this on its sleeve; for toughmovement it follows if the Op/gap must be pronominal.’’
Rezac’s copy-raising treatment of the TC correctly predicts that TC subjects should lack narrow scope readings—indeed, deriving the absence of scope reconstruction for TC subjects is highlighted by Rezac as a core virtue of his analysis—and that TC subjects should behave differently from ordinary raising and passive subjects in this respect. Moreover, if the TC matrix subject and the embedded operator are connected via an A-chain produced by φ-Agree, as Rezac proposes, then his account correctly predicts Hartman’s experiencer-PP intervention effects noted in (11), unlike the classic account given by Chomsky (1977), which posits no A-chain connecting the TC subject and embedded null operator.
2.2 Weak Islands
A second possible explanation for the absence of scope reconstruction in the TC is suggested by the complex null operator structure employed by Hicks (2009): this constituent may constitute a weak (or selective) island for extraction. As an anonymous reviewer points out, the complex DP from which the TC matrix subject is extracted on Hicks’s proposal, shown in (13a), is similar in relevant respects to the structure proposed by Boeckx (2003:28ff.) for resumptive pronouns, shown in (13b).
On Boeckx’s analysis, the complement of D in (13b) raises and serves as the antecedent of the DP remnant (which is spelled out as a resumptive pronoun), much as the DP complement of Hicks’s null operator in (13a) raises to become the TC matrix subject. Boeckx notes that the complex DP in (13b) can undergo Ā -movement (like Hicks’s complex DP) and constitutes a weak island (see also Rullmann and Beck 1998), with only D-linked phrases able to extract, that is, to serve as antecedents for resumptives. Boeckx (2003:31) shows this for Hebrew in (14), attributing the observation to Doron (1982).
Eyze student nifgašta ito?
which student you.met him
‘Which student did you meet?’
*Mi nifgašta ito?
who you.met with.him
‘Who did you meet with?’
Evidence that scope reconstruction is sensitive to weak islands comes from several sources; I illustrate with wh-islands. Consider first topicalization, sketched in (15).6 Binding tests confirm that reconstruction is the relevant mechanism for scope diminishment, with narrow scope unavailable where it would induce a Condition C violation, as in (16).
Heycock (1995:562) likewise observes that the narrow scope reading of Ā -moved how many is blocked by weak islands: (17a) has only a wide scope reading for how many, while the creation verb in (17b) renders this lone reading infelicitous.
How many books did she wonder whether to publish this year?
#How many lies do you wonder whether he may come up with?
Neg-split is also sensitive to weak islands in Ā -chains, as can be seen in cases of negative inversion. Neg-split is available under negative inversion, as in (18), but is blocked when it would involve reconstruction into a weak island, as in (19).7
If the complex null operator DP from which Hicks (2009) proposes that the TC subject extracts is a weak island—as its similarities with Boeckx’s (2003) complex DP structure for resumptives suggest— then the absence of scope reconstruction in TC-subject A-chains might receive a principled explanation on his analysis, given the independently attested weak island sensitivity of scope reconstruction in Ā-chains.8 That is, the cases that earlier made Hicks’s reliance on Boeckx 2001 appear problematic—weak DPs that undergo scope reconstruction in raising and passive but not in the TC—could be explained independently, with scope reconstruction ruled out in the TC as a weak island violation. Hicks’s complex null operator structure thus may indeed successfully predict the total absence of scope reconstruction for TC subjects, albeit for reasons he did not discuss.9
We can maintain the view that TC subjects are connected via an A-chain to the embedded clause, while explaining the scope reconstruction disparity between TC subjects and raising and passive subjects, in at least two ways. We might posit that the TC matrix subject is part of an A-chain not formed by movement or copying, as on Rezac’s (2004, 2006) copy-raising analysis; alternatively, we might appeal to a restriction on scope reconstruction—sensitivity to weak islands—not discussed by Hicks (2009) but compatible with his analysis. Space precludes a fuller consideration of the relative merits of these options, and I will not attempt to choose between them here. I note that the approaches make divergent predictions about the possibility of variable binding into the TC subject (copy raising against, Amovement for), though the judgments are disputed (Rezac 2006:301, Hicks 2009:552). In either case, we can conclude from the scope reconstruction facts discussed above that the strong parallel Hicks draws between TC-subject A-chains and raising A-chains is in need of qualification, if not outright abandonment.
The work presented here has benefited greatly from the detailed comments of two anonymous LI reviewers. I thank Jason Merchant, Line Mikkelsen, and Ljiljana Progovac for their feedback on earlier drafts, as well as audiences at Michigan State University and the University of Chicago, where earlier versions of this work were presented.
1 As noted above, Boeckx ties the scope possibilities for D heads to the position where they check Case. Weak DPs can serve as associates in existential there-sentences, checking Case from their base position. Boeckx (2001:528) argues that these DPs’ scope possibilities mirror their Case-checking properties, with scope reconstruction derived via the insertion of a silent expletive subject and concomitant lowering of the weak DP into its base position.
3 Note that I have used positively oriented scalar adjectives (easy and safe) here in order to avoid independently ruling out the narrow scope reading of how many because of an undefined maximum (Rullmann 1995). Compare #How many doughnuts is it hard to eat?, where if it is hard to eat five doughnuts, then it is also hard to eat six, seven, and so on, ad infinitum.
4 Certain speakers sometimes accept the narrow scope reading. Modal would and certain adverbials may be enabling factors. Crucially, TC subjects also receive a narrow scope interpretation in answers to how many–questions that are interpreted this way. Whitman (2011) calls this the ‘‘state-of-affairs’’ (SOA) reading. Consider the following examples:
Q: How many questions would be easy to answer in five minutes?
A: Three questions would be easy to answer in five minutes. = ‘It would be easy to answer three questions in five minutes.’
Q: How many turnips would be easy to eat?
A: No turnips would be easy to eat. = ‘It would be easy to eat no turnips.’
It seems that the availability of narrow scope here is due to whatever licenses the SOA reading, that is, to some interpretive mechanism other than reconstruction. Beyond these examples’ flouting of the general ban on scope reconstruction of TC subjects, note in particular that the answer in (ii) involves narrow scope negation, which is otherwise impossible in A-chains (Iatridou and Sichel 2011). I set such cases aside for the remainder of the discussion. I thank Ljiljana Progovac (pers. comm.) for bringing examples like (i) and (ii) to my attention.
5 Note that for-PPs in the TC, as in John is tough for me to love, do not give rise to intervention effects. This leads Hartman (2009) to analyze for as a complementizer, despite the well-known thematic restrictions on the DP following TC for (see Hicks 2009:555–556). Hicks (2009:557) treats TC forphrases as matrix PPs and likens their noninterventionary nature to that of experiencer-PPs in raising, as in John seems to me to be tired.
6 An anonymous reviewer notes that not all topicalized DPs can undergo scope reconstruction. For present purposes, what is important is that in cases where scope reconstruction of topicalized DPs is indeed available—as in (15a), where the narrow scope reading of the topicalized DP can be seen by considering the felicity of continuations like But I’m not sure which ones or But John doesn’t have any friends—it is blocked by weak islands, as in (15b). Note further that some speakers find even the wide scope reading of (15b) marginal, presumably because of the difficulty of getting a D-linked reading for two friends. If you prefer, substitute the more readily D-linkable two friends of his; the weak wh-island still blocks scope reconstruction (despite the lack of any Condition C violation; compare (16b)).
7 In (18), I have endeavored to avoid the independent scopal confound of neg-raising.
8 Given the traditional landscape of A-movement constructions, the matter of their island sensitivity is not much discussed in the literature. The null hypothesis appears to be that A-chains are sensitive to islands: for example, Hornstein (1999:91), in his A-movement approach to obligatory control, observes that sentential subjects allow only nonobligatory control and attributes this fact to their status as islands.
9 Another possibility is suggested by an anonymous reviewer, who asks whether TC subjects differ from raising and passive subjects with respect to the A-chain antireconstruction criteria proposed by Bobaljik and Wurmbrand (2005). Bobaljik and Wurmbrand observe that A-moved DPs in restructuring infinitives do not exhibit scope reconstruction, and they propose that this is because the VP complement of a lexical verb is a distinct agreement domain, blocking Agree and thus requiring movement (and concomitant scope freezing at LF) to check Case. While the matrix tough-predicate might count as a lexical verb for these purposes (i.e., as something whose complement constitutes its own agreement domain), the availability of unbounded Ā-movement within TC infinitival clauses is standardly taken to suggest that they are full CPs. This, in turn, makes them incompatible with restructuring, which requires an infinitival clause no larger than VP on Bobaljik and Wurmbrand’s assumptions.