Unbound variables under ellipsis and the syntax of sluicing

In the resolution of ellipsis, a number of non-equivalencies between overt antecedents and their elliptical counterparts have been observed, most prominently by Fiengo and May (1994), who name this phenomenon 'vehicle change'. In this talk, I explore such deviances from identity in the elliptical domain of sluicing. I begin with an investigation of the syntax of sluicing, and show that sluiced remnants behave like wh-phrases in full interrogative CPs in general, and in particular that pied-piping requirements in a particular language hold of sluicing constructions as well. With this background, I turn to a particular subset of sluicing cases: those in which the antecedent IP contains a wh-trace, as in (1):

(1) We need to find out who left, and when.

I argue that the unbound trace in the copied IP used to resolve the ellipsis is pronominal in nature, based on binding theory considerations and insensitivity to islands, and that this 'pronominal' variable is interpreted as an E-type pronoun, as in (2):

(2) We need to find out who [t_1 left], and when [he_1 left].

This assimilation to E-type pronominals also correctly predicts that proportion problem effects typically associated with donkey pronouns will be discernable in these cases as well. This is shown by examining quantificational variability effects in sluicing; I show that the cases at hand differ from multiple embedded questions in permitting asymmetric readings, giving rise to the proportion problem. The present data thus provide an additional argument against unselective binding approaches to donkey anaphora. The analysis presented here, if correct, provides new evidence for a particular conception of deviance from identity under ellipsis: certain types of categories are equivalent to their pronominal correlates under ellipsis.

Reference: Fiengo, R. and R. May. 1994. Indices and Identity. MIT Press.