Head-head relations in Problems of projection

2.1 Introduction

As noted above, PoP’s proposed solution to (b), i.e. to why there is a T-to-C relation but not a D-to-C relation, exploits the fact that in this case, the configuration [H [XP, YP]] arises when XP raises from a lower position after H is Merged. The solution therefore cannot be extended to many familiar cases of head-movement. We begin in §2.2 by illustrating the problem with respect to V-to-T movement across an external argument. §2.3 briefly considers whether it is relevant that in SVO languages, the subject raises out of vP: assuming the tail of a chain is invisible, this might yield a potential remedy in terms of the relative timing of EA and V-raising. §2.4 shows that V-to-T is possible in low-subject VSO languages and D-to-T is not an available alternative; hence EA raising is not the key to this asymmetry. §2.5 adds that V-raising across negation is also erroneously ruled out by Chomsky’s solution to (b).

2.2 V-raising across EA

Consider the structural relations between T and material in vP. By the logic of PoP, V-v and D, the immediate daughters to vP and DP respectively, are equally close to T. Therefore the puzzle PoP raised for T-to-C movement arises here as well. Why is it that in (6) there can be a V-v-to-T relation but not a D-to-T relation? In fact, D and V-v should be equally available for a relation to T, since both are equally close to it. Minimal search should obtain an ambiguous result. Yet V-v can raise to T in familiar languages, and D cannot. ^[10]

PoP’s argument that C-to-T is evidence of a C-T relation prior to EA-raising is weakened by the recurrence of the same phenomenon at this lower point in the clause. A parallel account of V-v-to-T movement would have to claim that a relation existed between T and v prior to Merge of the EA. However, in this case, there is no apparent lower position to shunt the external argument to in order to finesse the problem. Thus, if VPISH is correct, PoP’s proposed solution to (b) above is too narrow to account for the regularities of V-v-to-T movement. ^[11],[12]

2.3 A timing account?

Suppose the external argument raised to Spec, TP before V-v-to-T movement. Its unpronounced copy in vP would be invisible to T under PoP assumptions, hence solving the problem we presented in §2.2.

This timing provides a potential explanation for why it is true that only V-v, and not D, can raise to T in languages with verb raising. But it gives rise to other difficulties. For one thing, it appears to conflict with PoP’s account of T-to-C. Recall that PoP explains the fact that only T (and not D of spec T) raises to C in questions by assuming that the C-T relation is established BEFORE the EA raises from Spec, vP to spec TP. Crucially, PoP assumes that D of the subject cannot raise to C because the subject is “not there” (see (3), repeated below).

To maintain PoP’s timing solution to the original puzzle while at the same time adopting the timing solution we just suggested for (6), we would have to assume that T-to-C and EA-raising both precede V-v-to-T. This is highly stipulative, suspiciously counter-cyclic, and incompatible with the fact that V-v-to-T appears to feed T-to-C movement in many languages. We illustrate with Standard Norwegian in (8)a (from Taraldsen 1986: 8, who adopts a V-to-C analysis; the representation in (8)b is a slight update). ^[13]

Another problem arises for any appeal to timing in relation to these issues. Recall that PoP adopts both the VPISH and the Feature Inheritance hypothesis of Chomsky (2007, 2008) and Richards 2008. Under FI, only phase heads have edge and Agree features, and hence after operations in the vP phase are completed, there are no further operations until C is merged. Chomsky 2008: 151 writes, “It follows that the edge and Agree properties of P[hase] H[ead] apply in parallel: EF raises XP… to Spec, PH, while Agree values all uninterpretable features and may or may not raise XP to form an A-chain…. the edge and Agree features of the probe apply in either order, or simultaneously…”

This approach raises questions about whether it is even possible to claim that the C-T relation necessarily precedes EA raising, as PoP does; or to couch a potential account of V-v-to-T in terms of timing. Any timing solution is incompatible with simultaneity, since simultaneity expressly denies timing.

Summing up, we have explored the possibility of a timing solution to the puzzle of V-v-to-T across EA in Spec, vP. We have argued that such a solution is incompatible with PoP’s approach to T-to-C. Lastly we have argued that, in fact, any timing solution is incompatible with the simultaneity of application of operations internal to a phase proposed in Chomsky (2007, 2008).

We turn in §2.4 to a final piece of evidence against such a timing approach, namely that it is founded on the assumption that competition between EA and V-v-to-T are avoided only if EA raises to Spec, TP. §2.4 shows that V-v-to-T is the only option in various cases where EA does NOT raise to Spec, TP.

2.4 Low subject VSO languages

VSO clauses in languages such as Standard Arabic, Middle Egyptian, and Xhosa provide evidence that V-v-to-T is possible across subjects that remain low. ^[14] In such cases no timing approach could explain why D-raising from EA does not compete with V-v-to-T raising.

The morpho-syntax of VSO varies considerably across languages, and there are accordingly a number of approaches to it. In Standard Arabic, where only SVO is accompanied by full subject agreement, Mohammad (2000) argues that a sentence like (9) is the product of V-to-v-to-T across the in situ subject (see also Fassi Fehri 1993, Melebari and Seely 2012, and see Kramer 2009 for an analysis of Middle Egyptian along the same lines). By the logic of PoP’s approach to (1), in a VSO language with this derivation, D of the subject should be equidistant from T and the crucial ambiguity should arise. ^[15]

Carstens and Mletshe (2013) provide examples of VSO in Xhosa embedded clauses with overt complementizers. They argue that in a clause with default (= Class 17) subject agreement like (10)a, the subject remains very low in the structure. The verb raises across it and adjoins to a middle-field inflection below Tense, identified in the Bantu linguistics literature as Mood (see (10)b); (10) adapted from Carstens and Mletshe (2013). ^[16] The subject cannot surface higher as (10)c,d illustrate. The locations of C and T morphemes make it particularly unlikely that EA has raised to Spec, TP in the licit example (as it would have to for the proposal explored in §2.2 to be applicable). Had it so raised, the subject would be expected to surface between okokuba – ‘that’ and the future auxiliary be, contrary to fact.

We conclude that verb-raising is possible in Xhosa across an EA that remains low. As (11) illustrates, this means that prior to V+v-raising, D of the EA is at least as close to Mood as V+v is. But the contrast in (12) shows that D cannot raise instead of V. ^[17]

2.5 V-raising across intervening negation

A second class of problems for PoP (and the strategy sketched out in §2.3) lies in one of the standard diagnostics for V-v-to-T movement. Following Pollock 1989, we take the presence of negation between the surface position of V and its object to indicate that V has raised out of VP.

Recall PoP’s proposal that only hierarchical relations (not category) are involved in the calculation of closeness relevant to T-to-C movement. If we extend this idea to V-v-to-T movement, then it cannot be relevant whether the raised item is a verb, and Neg itself ought to be a candidate for movement to T, contrary to fact. If we assume that the subject does not pass through a Spec, NegP en route to Spec TP, then Neg should be the only candidate, since it is most local to T (see (14)a). Neg’s structural relation to T parallels that of T to C in (3), repeated below. If we assume instead that the subject occupies Spec, NegP at a point before verb-raising, then minimal search should yield an ambiguous result. Neg and D in the subject should compete for raising to T in (14)b, just as PoP argues would be true of T-to-C across a subject in Spec, TP in the hypothetical (2). In neither case is raising of the verb expected to cross Negation.

3.1 The problem

Consider now a second case of English T-to-C movement not addressed in PoP. T-to-C occurs in Wh-questions (WHQ) as well as Yes/No questions. Subject and non-subject Wh-questions display a well-known asymmetry: ^[18]

As (15) demonstrates, T-to-C is obligatory in all direct English WHQ except local subject questions like (15)a, where it is forbidden. (16) sketches out the relevant structure before Wh- movement for the three examples prior to C-to-T Feature Inheritance (Chomsky 2007, 2008). Shading indicates areas to which Transfer has applied; following Chomsky 2000 we assume the VP complement of v* Transfers before Wh moves to Spec, CP. This forces all but local subject WHs to move to the edge of the local vP phase edge, as is commonly assumed. Local subjects are externally Merged to the vP phase edge.

The problem for deriving (15)a-c should be evident: there is no difference between subject and non-subject WHQ in terms of the relation between T and C or in terms of distance between C and WH, and hence no clear basis on which to predict when T-to-C movement applies.

3.2 Pesetsky and Torrego’s (2001) solution

There is a simple way around this problem. Suppose with Pesetsky and Torrego 2001, and contra Chomsky 2007, 2008, that subject WH movement in English proceeds to Spec, CP via Spec TP, rather than directly from Spec, vP. ^[19] Representations of the two cases are sketched in (17) (relative locations of T and the Wh-phrase being crucial).

In (17)b, T is clearly closer to C than the Wh-phrase is, in marked contrast to (17)a. This assumption allows a potential account of the suppression of T-to-C movement in subject questions like (15)a on the basis of locality, much in the spirit of the PoP attempt to explain why C rather than D raises in (3). That is, in (17)b T raises to C since T is unambiguously close(est) unlike in (17)a. But in a departure from PoP assumptions, reference to features is crucial to ensure that a WH subject prevents T-to-C in (17) while a non-Wh-subject like that in (3) or (15)c does not, assuming subject and T are equidistant from C. Following Pesetsky & Torrego’s account, both have features relevant to C, but a Wh-subject has more of them, making a derivation that raises the WH-subject more economical than a derivation that raises both. ^[20]

This analysis is incompatible with PoP assumptions in several ways. First, as we saw in §2, equidistant candidates for raising should yield completely free alternatives; hence it would be expected that T and (subparts of) the Wh-subject would be equally able to raise to C(P) in a subject question. Second, PoP proposes that Wh-subjects never raise to surface in local Spec, CP, in contrast with Pesetsky & Torrego’s analysis. Hence either some PoP assumptions must be abandoned or some alternative account of this asymmetry must be found.

3.3 An affix-hopping alternative

Two anonymous reviewers suggest an alternative in terms of affix-hopping that we think it is instructive to explore. Consider again (16), a PoP-style derivation under which C merges before the subject raises to Spec, TP. Assume that T raises to C in all questions including the subject question in (16)a. After syntactic movements and transfer have applied, the results will be (18).

The reviewers suggest that affix-hopping, sensitive only to pronounced material, applies successfully in (18)a (see (19)a). In (18)b, the overt subject the boys blocks it, yielding the necessity for do-support in C, as shown in (19)b (see Chomsky 1957, Lasnik 1995, Lasnik et al. 2000, Halle and Marantz 1993 among many others).

While we see this as a viable alternative to Pesetsky & Torrego’s approach, its compatibility with PoP assumptions seems equally questionable. PoP’s claim regarding T to C is that “inversion depends on locality independent of category,” and it assumes that D of a subject is in principle as likely a candidate as T is for raising to C. The question that arises in connection with (19)b then is why affix hopping should not be equally a matter of pure locality, independent of category? But if this were the case, there would seem to be no reason for D of the subject in (19)b not to be a potential target for affix-hopping.

If, on the other hand, the affix-hopping operation is recognized to be categorically sensitive, the solution is viable. ^[21]

Summing up, the subject/non-subject asymmetry for T-to-C movement can be captured under the assumption that a wh-subject in Spec, TP suppresses it a la Pesetsky & Torrego, or by hypthesizing that affix-hopping applies across the subject trace but refuses to attach to D. Under either approach, reference to categorical and other features is critical.

3.4 A speculation re: the treatment of subject questions in PoP