Passive do so

Abstract

The received view is that the VP pro-form do so cannot be a verbal passive, although it can be unaccusative. I show that this is incorrect: do so can be passive. It can also take a raising to subject verb as its antecedent. This means that do so is compatible with all types of A-movement, although it does not permit A-bar movement of an object. I construct an analysis of do so where it is simply an intransitive verb plus an adverb. Do so combines with a Voice head, which can be unaccusative, passive, or active transitive. The subject of do so is base-generated in Spec-VoiceP, and does not move in unaccusatives or passives. Instead, do so must copy a function from its antecedent in the semantics. The subject of do so can be interpreted semantically as an internal argument if Voice is passive or unaccusative and the antecedent includes a trace, because of the way lambda abstraction works in A-movement. This analysis reconciles the evidence against movement in do so itself with arguments for A-movement in its antecedent. The copy mechanism explains voice and category mismatches, as well as split antecedents and ellipsis-containing antecedents.