Discourse structure and parallelism in VP ellipsis

Sentences containing VP ellipsis (VPE) must find a matching antecedent VP for the elided VP. This is a condition associated with recoverability: when a VP-gap is encountered, a VP representation from the preceding context must be copied into the empty VP position for the sentence to be interpreted. What kind of representation this is has long been debated— syntactic identity [1-3] has been argued for based on sentences where syntactic non-identity results in ill-formedness (1). However other cases seem to need a semantic identity condition to rule in acceptable sentences containing some form of syntactic non-identity (2) [4-5]. One approach to such observations has been to assume syntactic and semantic recovery mechanisms are available to the processor, and which is used depends on how the meanings of the antecedent and ellipsis clauses are related; for Kehler [6], establishing a Resemblance relation between two clauses requires aligning their arguments, and as such enforces syntactic identity (1), while clauses related by Cause-Effect (CE) are parallel at a level of representation that encodes meaning at a propositional or property level, so it's possible for parallelism to be satisfied despite syntactic non-identity (2). A different approach is to associate different (syntactic, semantic) mechanisms for resolving VPE based on the domain where the parallelism condition applies; for example, Frazier and Clifton (F&C) [7] propose different well-formedness condition apply to VPE within connected syntactic structures (1-2) versus across sentences (3). Finally, parallelism might hold at a single level of representation that has the appropriate combination of properties—the types of non-identity possible would result from whether or not particular kinds of information are encoded at the relevant level of representation (see e.g. suggestions in [8]). We present three experiments (two magnitude estimation, one 2-alternative forced choice) that argue for a parallelism condition on VPE that is violable, and applies at the level of discourse structure. Differences in the antecedent-ellipsis relations permitted within and across sentences follow from the unavailability of reflexive-binding across sentences. Experiment 1 tests Kehler's predictions that the acceptability of syntactic mismatch will depend on the coherence relation between the antecedent and ellipsis clauses. Sentences with VPE were compared, with antecedent and elided VP either matched (1a) or mismatched (1b) in voice, and where clauses were related by Resemblance (1) or CE (2); all conditions had corresponding no-ellipsis controls. Participants gave sentences acceptability scores relative to a self-chosen standard. There was an Ellipsis-Match interaction such …