Structure and Acceptability in Verb Phrase Ellipsis

Verb phrase ellipsis is a phenomenon in English where a verb phrase is omitted from an utterance, and its intended meaning is recovered from context. The target, or elided, verb phrase need not be a perfect match for its antecedent, and morpho-syntactic differences of a variety of types (e.g. tense or aspectual marking) may be tolerated (Quirk et al. 1972). The case of voice mismatch, where the antecedent is realized as a passive and the target as an active (or vice versa), has been of particular interest since Sag (1976), where cases of unacceptable mismatch served as crucial evidence in support of a theory where ellipsis is licensed by syntactic identity between antecedent and target. This picture is complicated, however, by cases of apparently acceptable mismatch, often drawn from corpora or other naturally occurring data. (See, e.g., Dalrymple et al. 1991, Hardt 1993, Kehler 2000, 2002.) In this brief study I offer a re-examination of representative voice mismatch data and the role they have played in shaping theories of ellipsis. I present results from magnitude estimation experiments which address the following questions: Under what circumstances are voice-mismatched ellipses judged acceptable, unacceptable, or something in between? Does the penalty associated with voice mismatch stem from a lack of syntactic identity or from some other source? And finally, are the degradation effects of voice mismatch limited to ellipsis contexts? The results from this study support a novel claim that information structure plays a role in determining ellipsis acceptability and that information structural effects are not unique to ellipsis.