Future research could profitably explore the hypothesis that syntactic categories should be eliminated from linguistic theory and their work taken over largely by the independently motivated system of semantic types. This would be a notable gain in theoretical economy, provided that their elimination does not necessitate innovations of equivalent complexity elsewhere in the theory.

This article analyzes three phenomena that are troublesome for some theories of ellipsis: the existence of sloppy readings when the relevant pronouns cannot possibly be bound; cases where the antecedent of ellipsis does itself contain an ellipsis site, but in resolving the larger ellipsis the interpretation understood at the ellipsis site in the antecedent is not used; and cases where an ellipsis site draws upon material from two or more separate antecedents. These cases are accounted for by an analysis of silent VPs and NPs that makes them into higher-order definite descriptions that can be bound into.

Chien and Wexler (1990) reported that children obeyed Principle B of binding theory when the antecedent was a quantifier but not when the antecedent was referential. This was argued by Grodzinsky and Reinhart (1993) to support Reinhart's (1983) theory according to which Principle B affects only bound pronouns. Since then, other studies have supported the asymmetry between referential and quantifier antecedents. This article, however, argues that previously unremarked experimental factors lessen the force of all these studies, and it points to other relevant experiments that seem to showthat children do not obey Principle B at all. It reviews previously offered theories on the acquisition of Principle B that are compatible with the latter view of the facts.

Theories of total reconstruction have generally supposed that movement can be followed by an undoing operation like LF lowering (May 1977, 1985) or deletion of higher copies (Chomsky 1993). We argue that reconstruction effects can be derived only if the original movement is purely phonological. There are no undoing operations. We present three distinct arguments, based on an interaction between raising and wh -movement in English, facts from agreement with group terms in British English, and multiple scrambling in Japanese. The arguments imply that the T-model is correct in supposing that movement that affects both LF and PF must precede movement that affects only PF.