Abstract for Lyn Frazier's talk

Structuring silence: Processing VP ellipsis
Lyn Frazier (University of Massachusetts)
April 21, 2005 (Thursday)
4:15 PM - ; Room 6417, the CUNY Graduate Center

Processing elided constituents requires listeners to figure out the analysis of silence, i.e., the analysis of unpronounced constituents. This talk will focus on how people process VP Ellipsis (VPE) and the implications of the processing results for the grammatical analysis of VPE. In VPE, the antecedent for the elided constituent may appear intrasententially, or the antecedent may appear in an earlier sentence than the elided constituent. This feature of VPE allows the structure to be used to explore the differences between syntactic representations/processes, on the one hand, and discourse representations/processes, on the other. It will be argued that when an elided VP in one sentence has an antecedent in a preceding sentence, the processor favors antecedents that are part of the main assertion of the preceding sentence. This typically implicates constituents high in the tree-structure. This preference for ?ehigh?f constituents contrasts sharply with the low attachment/recency preference that is observed in syntactic processing. It will be suggested that the salience difference between ?ehigh?f constituents and low/recent constituents is a general difference between discourse representation and syntactic representation.

Linguistic approaches to ellipsis fall into several broad categories depending on whether there is syntactic structure at the ellipsis site and on the precise nature of the licensing requirement. Syntactic approaches to ellipsis generally require syntactic structure at the ellipsis site and require the presence of a (matching) syntactic antecedent. These approaches are known to undergenerate (Hardt, 1993). Semantic approaches require only a semantic antecedent (e.g., for VPE, a property). These approaches overgenerate. A hybrid account also exists. It claims that a syntactic antecedent is required only when certain discourse-coherence relations hold, but not when others hold (Kehler, 2002). Processing results will be used to argue for a syntactic approach, supplemented by a particular processing theory: when an appropriate syntactically-matching antecedent does not occur, the processor creates one at LF using the materials at hand. This ?gVP-Recycling?h hypothesis begins to delimit the circumstances under which listeners/readers accept VPE without a matching antecedent and, it will be argued, it offers a solution for the undergeneration problem.