The Status of Head Movement
The easiest way to solve this puzzle is to proceed as we did for feature movement, banning it from syntax. This is what Chomsky (2001) does: he assumes that head movement is not a syntactic operation but part of the ‘‘phonological branch of the computation.’’ If this is true, then the minimal amount of material a feature can pied-pipe becomes a phrase, and the phrasal properties of (standard) wh-movement end up being derived for free.
This move is untenable, however, for several reasons: it is based on wrong assumptions, both theoretical and empirical; it is incompatible with other aspects of Chomsky’s theory (e.g., phases); and it brings un desirable consequences.
First, this proposal is based on a wrong theoretical assumption, namely, that head movement violates crucial syntactic requirements—and in particular, that head movement does not extend the target (hence being countercyclic). This is true for the standard head adjunction configuration given in (1).
(1)
The traditional reason for this configuration is related to the standard cases of head movement usually considered in the literature: in V-to-T movement, or in V-to-C movement, the two heads conflate and behave like a single constituent, hence the assumption that they form a sort of ‘‘derived lexical item’’ represented in the head adjunction configuration in (1).
But this is not the only configuration head movement can in principle produce. Suppose we have a head X endowed with an uninterpretable feature a that needs to be deleted. In principle, nothing prevents a copy of a head Y endowed with the matching feature b from merging with the root of the structure, as in (2).
(2)
The configuration in (2) is obtained by merging a new item to the root of the tree, hence complying with the Extension Condition. Given (2), the head conflation effect correlated in many cases with this configuration can be the result of an independent process, perhaps phonological, which in principle has nothing to do with head movement: this process—call it ‘‘affixation’’—is something that can happen to two adjacent heads independently of how and why they ended up being adjacent.
Separating head movement from affixation is a good idea in light of several empirical facts: first, there are cases of affixation that do not correlate with head movement, but with phrasal movement (see below); second, there are cases where head movement does not yield any effect of this kind; third, there are cases where affixation does not in volve moved heads, but merged ones.
As an illustration of the first cases, consider (3) as a possible derivation for a clitic construction in Italian.
(3)
Clitic pronouns in Romance have always been problematic given standard assumptions about phrasal and head movement: clitics are clearly maximal projections in the position where they are generated, but they end up being affixed to an inflectional head. This tension is at the root of standard and influential analyses such as Kayne’s (1989), where the clitic starts out as a phrase but moves and adjoins as a head, changing its phrase structure status in the course of the derivation (violating a condition like that on the uniformity of chains; Chomsky 1995). The derivation in (3) is by far more minimal: here the phrase structure status of the clitic never changes during the derivation, and the clitic moves as a phrase. The D-to-T incorporation is an independent process akin to phonology.
Summarizing so far, nothing about head movement prevents it from being cyclic, and assuming the existence of cyclic head movement appears to be empirically convenient.
But also inherent in banning head movement from syntax is a troublesome implication, which appears to be too strong: being phonological, head movement should never feed LF, hence never have any interpretive effect. This may well be true if we consider movement of predicative heads, which need to reconstruct for interpretability (Heim and Kratzer 1998). But even in these cases, the issue depends crucially on the status we assign to the notion of extended projection (Grimshaw 1991): the extended projections of V, or of N, are typically defined as head movement domains, and they do look like interpretable objects as a whole.4 On the other hand, the LF effects of head movement become simply unquestionable once we consider movement of quantificational heads, which do acquire their scope through this mechanism.
Returning once again to Chomsky’s proposal, there is also a technical problem with its implementation if it is combined with other aspects of the more recent framework: the idea that head movement is phonological clashes with the details of the derivation-by-phases approach to computation. Suppose the system is building a simple structure like (4).
(4)
Suppose moreover that V0 must move to T0 to check some uninterpretable feature. At the level of the first strong phase, vP, the sister of v (VP) is sent to Spell-Out. But the edge of the current phase, v0, is not yet accessible to phonology. In other words, the launching site and the landing site of this head movement operation are not accessible to phonology in the same phase. A solution could be that V-to-T movement happens when the next phase (CP) is sent to PF; but then the derivation must deal with two phases at once, hence raising the computational burden, contrary to the whole spirit of phase theory (Matushansky 2006).
Finally, pushing head movement onto a phonological branch of the computation has a broader undesirable consequence for the theory: it complicates the grammar by positing a ‘‘second syntax’’ in the PF branch. It also raises at least two questions. For one thing, if both phrasal movement and head movement are driven by features, it seems at least strange to locate them on two different computations. For another, consider that Chomsky (2000) assumes that PF also involves some kind of phrasal movement (e.g., extraposition; see also Lebeaux 1990), so that the problem of the duplicity of movement is reduplicated within this second syntax.
The conclusion is that there is no principled way to exclude head movement from narrow syntax, and a simple stipulation opens a number of problems. Head movement is indeed an operation available to narrow syntax; moreover, it is the default movement option, being the minimal pied-piping strategy. The next step, then, is to find some property of head movement that makes it unavailable in standard wh-movement con texts, hence triggering the less minimal, more costly operation: phrasal movement.
