Context and Well-formedness: the dynamics of ellipsis

Using data from ellipsis as evidence, this talk argues that a grammar formalism for natural languages should both articulate the process of building up interpretation in real time, and be sensitive to contextually provided information. Almost every natural language expression displays some form of context dependence. This is most obvious with anaphora and tense construal, but the effects go much further than this. An utterance of e.g. I bumped into Mary yesterday in the park will convey different information according to who the speaker is, who Mary might be, where the park is (and what park) and when the sentence was uttered. Such matters are generally treated as the province of semantics, the syntax merely providing some decontextualised compositional analysis of the string of words that inputs into the semantic interpretation. A string like He cried is thus treated by the syntax to be well-formed irrespective of whether the context provides appropriate antecedents for the construal of the two pronouns, even although a sentence like (1) is peculiar, if not ill-formed, with no prior context to provide a male referent. (1) # Mary hit her head on the doorframe and he cried. There are, however, syntactic phenomena that more obviously require some reference to context to determine well-formedness, in particular elliptical constructions, where the preceding linguistic context is essential in determining the well-formedness of the string: (2) Mary washed her hair and so did Bill. (*Mary was tall and so did Bill.)