ÍNDICE DE TABLAS
ANEXO 1: GLOSARIO DE TÉRMINOS
The proposals made by Rizzi, Boeckx, and Richards code the fact that movement is (upward) bounded. Of course, we have known
1 The results of Mueller and Sternefeld’s (1993) Principle of Unambiguous Binding, as well as Bosˇkovic´’s (2005b, forthcoming) treatment of freezing eVect will also be encompassed by the analysis developed in this chapter. Thanks to Sam Epstein and Terje Lohndal for pointing this out to me.
2 Richards (2001), unlike Boeckx (2003a), allows for chains that are too strong to be formed in the syntax, provided they are repaired at PHON. Though not insigniWcant, I ignore this diVerence in what follows.
this since Ross unambiguously established the fact that there are domains out of which movement cannot take place.3 This is the very idea behind islands. But to this Rizzi, Boeckx, and Richards added the fact that movement is subject to Last Resort. It is not just that an element may end up being ‘‘stuck’’ in virtue of being in a certain domain (say, an adjunct clause), an element may also end up being
‘‘stuck’’ once it has reached a certain checking site (a strong pos-ition/occurrence for Boeckx/Richards; a criterial site for Rizzi).
When Chomsky introduced the notion of Last Resort in Chomsky (1986a), he did so to rule out instances of movement from a Case-checking position to another Case-Case-checking position, as in (1)
(1) *John seems [t’ is [t happy]]
" "
Case Case
But the idea of Last Resort generalized to virtually all features, to rule out instances of multiple checking of the same feature by the same element (ban on iterated wh-checking, ban on iterated topi-calization, ban on multiple theta-checking, etc.) (I return to a few controversial cases below). Thus we can state (2).
(2) Last Resort condition on movement
An element E can only check a feature F of type T once
The literature on Relativized Minimality has independently exam-ined the issue of feature type. At present, it seems relatively uncon-troversial to say that there are three types of features: ‘‘Theta,’’
‘‘Case,’’ and ‘‘A-bar’’—corresponding roughly to the three layers of the clause: V, T, and C. But, as we will see in a moment, the distribution of these features is quite complex, and will require close scrutiny.
For now let me reiterate an observation I made in Boeckx (2003a) (see also Gallego 2007). As soon as we combine the notion of upper bound of movement (island) with Last Resort (feature checking), the possibility arises to characterize islands in featural terms, as
3 For now, I ignore the fact that for Ross only some instances of (movement) rules are subject to islands. I’ll return to this important caveat below.
opposed to the more traditional geometrical/conWgurational terms.
If we take seriously the idea that (some) checking sites constitute an upper bound for movement, characterizations of island eVects like
‘‘movement out of left-branches is banned’’ won’t suYce. They will have to be replaced with statements that focus more on relations than on conWgurations, like ‘‘movement out of Case-checking sites is banned.’’ The fact that all Case-checking sites are left-branches may thus be a mere side-eVect. The geometry of the tree alone may not constrain movement. This in turn gives rise to the possibility (exploited in Boeckx 2003a) that no domain is an absolute island, since all islands will have to be relativized to the type of features in chain formation. Ross’s idea that islands only constrain some move-ment rules can thus be entertained again, in a fresh, minimalist light.
The possibility of revival of the featural view on islands may be good news, given what we have learned about islands since Ross’s original study. Although island eVects are found in all languages, there is some variation in the patterns of extraction that may be hard to capture on a purely conWgurational view of locality. The latter appears too rigid and universal to allow for variation. By contrast, given that features are standardly considered to be the locus of variation, we may expect that diVerent feature combinations will result in diVerent extraction patterns.
Needless to say, these are just a priori considerations in favor of a featural view of islands. As always, the proof of the pudding remains in the eating, but a priori considerations may help us choose which pudding to start eating Wrst.
In this chapter, I would like to establish (in part building on Boeckx 2003a) the correctness of the statement in (3).4
(3) An element can only move to a single feature-checking site
Like (2), I intend the condition in (3) to act as a boundary condition on movement. The feature-checking site to which the element
4 Bosˇkovic´ (forthcoming) makes a similar claim, although his interpretation of the ban on multiple feature-checking under movement and his implementation of it diVers from mine in many respects.
moves is intended to mark the upper bound of that element’s chain, the edge deWning the chain’s maximal extension. Since the central contention of this work is that syntactic objects can only have two edges, (3) would follow from QED. Taking the site of External Merge to mark one edge of the chain, the feature-checking site occupied by Internal Merge would count as the other edge. Perhaps the most challenging situation for (3) is the alleged existence of mixed A+A’-chains (movement for Case checking followed by A-bar movement, as in subject questions like who does John think t’ may have been arrested t?). Mixed chains appear to contain two checking sites. I will thus devote special attention to situations where they may arise below.
To even begin to explore the path deWned by the considerations above, it is important to stress what the hypothesis in (3) does not say. Here two remarks are in order. First, the hypothesis in (3) does not say that an element can only move once. So long as no feature checking is involved, the number of movement steps is potentially inWnite. This is the same point I made in previous chapters con-cerning Merge and adjunction. QED-considerations impose a single complement and single speciWer requirement, but they do not impose any upper bound on adjunction (Merge situations not involving featural transactions).
Building on my previous work (Boeckx 2003a) as well as on Takahashi (1994) and Bosˇkovic´ (2002a, 2007), I have argued at length in Boeckx (2007a) that intermediate landing sites of move-ment do not involve feature checking.5 Repeating the evidence here would take me too far aWeld. I will simply state that the correctness of the present analysis crucially depends on the claim that the formation of intermediate chain links is not motivated by immedi-ate feature-checking considerations. SpeciWcally, I will follow Boeckx (2007a) and assume that an element is free to move/adjoin so long as it contains an unvalued feature.
5 The claim that movement can proceed solely in virtue of the fact that some feature remains unchecked appears in a variety of recent minimalist frameworks; see Lo´pez (2002, 2007); Preminger (2007); Putnam (2007); van Craenenbroeck (2006), Stroik (1999, forthcoming); and Hornstein (forthcoming). Thanks to Angel Gallego for bringing some of these works to my attention.
Second, the hypothesis in (3) does not say that an element can only participate in one feature-checking relation. It only says that an element cannot be internally merged in more than one checking site.
If all instances of feature checking required movement, as was assumed in early minimalism (cf. Chomsky’s (1993) generalized spec-head relation), (3) would not be empirically viable. But as soon as the existence of Agree is recognized, it can be used in service of (3) to handle potentially problematic instances of multiple feature checking. As a matter of fact, once Agree is assumed, we face a diVerent problem: if Agree exists as a mode of feature checking, doesn’t it render (3) vacuous? If every time an extra checking relation is established Agree can be appealed to, (3) will always be satisWed. We could even say that no checking relation is established under movement. The issue is this: Since the existence of Agree essentially dissociates movement from feature checking, what counts as a checking site for Internal Merge? The key then is to deWne the connection between feature checking and movement, to which I now turn my attention.