Syntactic identity, Parallelism and accommodated antecedents

Analyses of the ellipsis identity condition must account for the fact that some syntactic mismatches between an ellipsis site E and its antecedent A are possible while others are not. Previous accounts have suggested that the relevant distinction is between different kinds of heads, such that some heads in the ellipsis site may mismatch while others may not, and they have dealt with this sensitivity to a set of "special heads" with a built-for-purpose syntactic identity condition which holds over and above semantic identity to constrain ellipsis. In this article I argue against this approach and pursue an alternative which holds that identity is syntactic but "loose" in a precisely defined way. I show that the relevant generalization that accounts for syntactic identity effects in sluicing and VP-ellipsis-like constructions concerns the position of variables in the antecedent, rather than the feature content of syntactic heads. I propose an implementation of syntactic identity which allows for the accommodation of additional antecedents, with these being derived by a grammatical algorithm for generating alternatives, and I show that this implementation derives the right kinds of looseness while restricting mismatches with respect to the position of variables, thus deriving both the tolerable and intolerable mismatches between E and A without recourse to a specific condition regulating the content of special heads.