Most current models of syntactic parsing assume that the language comprehension system computes detailed, veridical representations of the linguistic input. Contrary to this assumption, numerous studies have found that people systematically misinterpret unambiguous sentences that require a noncanonical order of thematic role assignment as well as sentences that require the construction of elided linguistic content, suggesting that the parser may perform a shallow, semantic-based analysis in addition to a detailed compositional analysis. However, it remains unclear how the syntactic and semantic streams are reconciled to generate a final interpretation that is sometimes incompatible with the veridical interpretation. The goal of this dissertation was to assess the hypothesis that the misinterpretation of noncanonical sentences reflects a competition between the veridical interpretation and an alternative, highly plausible interpretation that are considered in parallel (i.e., the competing interpretations hypothesis).

Chapter 1 provides a brief overview of previous research and theoretical perspectives on the comprehension of noncanonical structures and describes the motivations of the competing interpretations hypothesis. Chapter 2 describes two experiments that assessed the competing interpretations hypothesis in the comprehension of passive sentences. Results showed that misinterpretation rates were higher as the plausibility of the alternative interpretation increased, but this effect was observed only when the alternative interpretation was more plausible than the veridical interpretation, and not when the two interpretations were equally plausible. Chapter 3 outlines the rationale for a follow-up study that extends these findings by investigating the competing interpretations hypothesis in ellipses, another type of movement-derived structure that may also be subject to misinterpretations. Taken together, these experiments provide evidence against the competing interpretations hypothesis that assumed a parallel, competitive architecture in sentence processing, and instead lends support to two-stage models of parsing that assume an initial commitment to one interpretation that may be subsequently revised when a syntactic or semantic anomaly is detected.