The derivation of accusative unaccusatives
We now return to the derivation of Slavic accusative unaccusatives. Russian examples are given in (1a—b) for reference:
(1)
The structure for accusative unaccusatives is given in (2) (which combines
The predicate V merges first with the indirect internal argument, and then the resulting constituent merges with the direct object. On the first merger, V checks the “inherent” case that is associated with its indirect internal argument. On the next merger, V assigns its remaining theta role. The maximal projection of V then merges with light-υ, a null affix-like element to which lexical V subsequently attaches. Light-v checks structural ACC on the direct object via Agree, assuming that υ is This operation does not involve movement; no Spec-vP is projected because υ does not have an EPP requirement in these languages (i.e., these are not Object-Shift languages). At this point Tdef merges with υP forming TP. Because T is defective it has no agreement features to check. However one of the internal arguments of υP must concatenate with TP to form a specifier position in order to satisfy the EPP requirement of T. This derivation establishes that displacement need not be motivated by either Case or agreement. As noted above, it further shows that an NP whose Case has been valued previously is not necessarily frozen in place. In this case a goal need not contain an unchecked feature to be active with respect to some probe.
If T were in (2), the derivation would fail to converge. The exact nature of this failure is not obvious. Given that the NP moving to SpecTP does so only to satisfy the EPP, it’s not obvious why the interpretable agreement features on the NP couldn’t also check the agreement features of T. To address this puzzle, let us consider three possible solutions. Suppose that T assigns NOM to the NP that checks its agreement features. This would create a form N+AGG+NOM—surely impossible in Slavic and likely universally. We might further assume that the ACC Case affix blocks the attachment of the NOM affix. Thus, the derivation violates Lasnik’s stranded affix prohibition (Lasnik 1981). Now if a stranded affix constitutes a piece of phonological material that can’t be interpreted as a word, then it might be construed as a violation of Full Interpretation at PF.
Alternatively, we might assume that only a NOM Case-marked NP can check the agreement features of T. Then the derivation fails to converge at LF because of the unchecked uninterpretable agreement features of T that survive. This analysis presupposes NOM Case as a precondition for checking the agreement features of T, rather than as a reflex of agreement with T (as in standard analyses). If we abandon the analysis in which agreement values Case, then we lose the explanation for why T and υ cannot both be defective in the same clause—i.e., the uninterpretable Case feature of the internal argument could not be eliminated from the derivation.
As a third alternative, consider the possibility that there is a universal prohibition against the of an NP entering into more than one agreement relation. Thus, in the case of the accusative unaccusatives, the ACC argument, which gets its Case valued via establishing an agreement relation with υ, could not enter into an agreement relation with T—regardless of NOM Case. This unique agreement constraint generalizes to other instances of improper movement, for example (3).
(3) *John was believed t was content.
By hypothesis, the deviance of (3) concerns agreement rather than Case—i.e., that the agreement features of the NP can only be used once to value the agreement features of T (see Freidin and Vergnaud (2001) for additional discussion).
