I present an analysis of antecedent mismatch effects under ellipsis based on information structure, in which apparent syntactic parallelism effects are explained as a consequence of an information structural constraint requiring topic/comment parallelism for contrastive topics. Experimental findings in support of this hypothesis demonstrate first that the penalty associated with antecedent mismatch is greater in contrastive topic structures as compared to simple focus structures and next that mismatched contrastive topics show reduced acceptability in both ellipsis and non-ellipsis contexts. Online reading time results further support this hypothesis showing that mismatched antecedents induce disproportionately larger processing costs in contrastive topic structures as compared to simple focus structures, as evidenced by increased reading times in regions both preceding and following the ellipsis site. The findings that mismatched contrastive topics lead to reduced acceptability offline in the absence of ellipsis and that mismatched contrastive topics disrupt online processing prior to the ellipsis site are consistent with the hypothesis advanced in this thesis but pose a challenge to models which have instead attributed mismatch effects to special-purpose processes supporting antecedent reconstruction or repair