Chains Start at S-Selection Positions
Question (1a) does not concern A ¯-chains uniquely. It can be asked of phrasal chains in general, regardless of whether they stop inside IP or proceed to reach the left periphery of the clause. The question breaks up into two subcases (I will initially focus on arguments for simplicity):
(1)
(2)
The first case is addressed by Chomsky (2000). Why can’t (3b) be derived alongside (3a), given the choice of lexical array appropriate for (3a)?
(3)
The critical point in the derivation is when the embedded T is merged, endowed with an EPP feature requiring its Spec to be filled.
(4)
If the pronoun I is part of the lexical subarray of the same phase, as under Chomsky’s assumptions, why can’t it be merged in Spec,T, thus satisfying T’s EPP requirement, and then be raised to the subject position of expect, thus receiving the y-role ‘‘experiencer of expect,’’ with representation (3b)? Why is movement of a proof the only viable option in (4), yielding (3a)? The question is particularly relevant given the idea that Merge, when possible, prevents Move, which Chomsky uses to account for the following pair:
(5)
If the expletive there is part of the relevant lexical subarray, Chomsky suggests, it must be merged at stage (4), to fulfill the EPP requirement on T (on the derivation of the more natural order ...to be a proof dis covered, see the discussion in Chomsky 2000). Then the expletive raises to the main subject position, yielding (5b). (5a) is excluded because the possibility of merging there in (4) precludes the more complex operation of moving a proof.
But why is the pattern reversed in (3)? Chomsky (2000, 103) intro duces the following principle:
(6)
So the argument I cannot be introduced into the structure by pure Merge in a non-y-position in (4), and the only available derivational option is movement of a proof to yield (3a).
This approach must clearly be generalized to first Merge of non-argumental material: an adverb cannot be inserted in a position lower than the position it is interpreted as a modifier of. For instance, in a verb-second (V2) language a temporal adverbial cannot be first-merged to satisfy embedded V2 and then be interpreted as construed with the main tense structure. Consider the following paradigm (provided by Günther Grewendorf, pers. comm.):
(7)
(17b) shows that a temporal adjunct can be extracted from an embedded clause, satisfying ‘‘in passing’’ the V2 constraint in that clause, in which case the adjunct is interpreted as construed with the embedded clause. The deviance of (7c) shows that an adjunct satisfying embedded V2 must be construed with the embedded clause: vor zwei Jahren ‘two years ago’, semantically compatible with the main clause tense but not with the embedded tense, gives rise to ill-formedness in this case, while it is fine with main clause construal in (17d), where embedded V2 is satisfied by another element.
In conclusion, (2a) generally seems to require a negative answer. As for (2b), it is quite generally assumed to require a negative answer as well. For instance, languages do not express reflexive readings by moving an argument from object to subject position, to pick up both y-roles, and control structures cannot be analyzed through movement from the lower subject position to a main clause argument, as the lack of reconstruction effects typical of raising shows (see (8)). (The analysis of control as feature movement, whether right or wrong, is a different matter (for discussion, see Hornstein 1999; Landau 2003); and the unaccusative-like analysis of Romance reflexives proposed by Burzio (1986) and Kayne (2001, chap. 8) does not involve the picking up of two y-roles via movement.)
(8)
Consider, for instance, the familiar raising-control asymmetry with respect to the possibility of anaphor binding under reconstruction (Langen doen and Battistella 1982; Belletti and Rizzi 1988). The anaphor each other can be bound by the men under reconstruction in the raising structure (9a), a result immediately captured by the copy theory of traces, but not in the control structure (9b).
(9)
(See also Burzio’s (1986) discussion of the construal of each under reconstruction, excluded in control structures.)
We can now try to generalize Chomsky’s principle (6) to include these cases, and also to extend it to analogous patterns with nonarguments, adverbial material, and the like.
Let us continue to refer to the traditional notion of s-selection to encompass y-marking of arguments and the licensing of adverbials (ana lyzed along the lines developed in Cinque 1999) and other nonargumental material.
(10)
(We can understand ‘‘s-selection’’ here in a maximally general way, as involving both the selector and the selected element. Note that principle (10) basically expresses and generalizes the GB assumption that the y Criterion applies at D-Structure.) Therefore, the subject I cannot pick up its y-role ‘‘experiencer of expect’’ by second Merge in (3b), John cannot pick up the y-role ‘‘experiencer of see, want’’ in (8), the temporal adverb cannot pick up a selectional relation with the main clause T in (7c), and so on. On the other hand, the expletive there, which is not subject to s-selection, can be first-merged in the subject position of (5) (and must be, according to Chomsky’s analysis).
