It is well known that in sluicing constructions wh-dependencies can cross certain projections that are otherwise barriers to movement (Ross 1969, Chomsky 1972). This fact would follow under the assumption that the relevant barriers are somehow deactivated when phonologically deleted (“island repair”). The problem, however, is that another form of phonological deletion (VP-ellipsis; VPE) seems to be impossible in certain contexts where sluicing allows for island repair (Chung, Ladusaw, and McCloskey 1995, Merchant 2001).
Nevertheless, we argue against the conclusion that island repair is a special property of sluicing. The argument is based on two observations. First, the difference between sluicing and VPE seems too broad to warrant the conclusion that island repair is the distinguishing factor (Lasnik 2001). Second, the conclusion is directly refuted by other VPE environments where island repair is possible (Kennedy and Merchant 2000; Fox, in preparation). The argument leaves us with a puzzle that we attempt to resolve while still maintaining the null hypothesis that VPE and sluicing involve the same operation of deletion, differing only in the size of the deleted constituent. Our proposed resolution capitalizes on a special property of the relevant sluicing contexts—namely, the presence of an indefinite NP in the antecedent clause in a position parallel to that of a trace in the elided clause. We argue that given the parallelism conditions on ellipsis, this fact prevents the wh-phrase in the elided clause from undergoing successive-cyclic movement. The remaining option (one-fell-swoop movement) requires the deletion of all barriers, including those that would otherwise be circumvented via an intermediate landing site. Such deletion occurs in sluicing but not in VPE, which targets a smaller constituent.