Argument Contained Ellipsis Revisited

This paper investigates an unusual identity constraint on English verb phrase ellipsis which imposes the following requirement: when an elliptical relation holds between two verb phrases A and B such that A is contained in an argument b of B, then the corresponding argument a of A must be identical to b. I argue that these facts provide evidence for a semantic theory of the interpretation of variables in which assignment functions are partial, and binding involves adding new variable/value pairs to an assignment, rather than reassigning new values to (possibly previously used) variables. Based on additional data, I argue that the identity conditions on ellipsis must be stated in terms of semantic parallelism between constituents (potentially) larger than the elided phrases, as advocated in Rooth 1992, Fox 1999, and Merchant 2001, and that the syntactic representation of binding structures must crucially encode binding indices on binders, rather than as adjuncts on their scope as in Heim and Kratzer 1998. I conclude by demonstrating that the analysis proposed here has broader empirical coverage than current variable-free analyses, and so potentially constitutes an argument for adopting a semantic framework that makes use of variables.