| Understanding the workings of the human language system remains a great challenge. One approach to gain insight into the process has focused on discovering the type of information the language processor uses and how that information is integrated over time. The work presented here examines a piece of the puzzle---what are the processes that underlie syntactic ambiguity resolution. Two current theories of sentence processing offer differing explanations. Constraint-based theories of sentence processing have two important tenets: (1) syntactic analyses are guided by a coalescence of influences that include the frequency of items, categories, and syntactic structures, along with the probability of their occurring together, and (2) our language system uses this information immediately. This view contrasts with the influential Garden Path theory of sentence processing which posits that the sentence processor is initially limited to processing structural information in a first pass at parsing the input, to be followed by other kinds of information in a second stage.; Based on the tenets of constraint-based theory we developed a set of predictions and tested them against those made by the Garden Path model in a series of four reading experiments. Participants read sentences that contained syntactically ambiguous regions at the beginnings of sentences, and following a restricted set of verbs. The results support the immediate and exhaustive use of lexical information in sentence processing, as predicted by constraint-based theories. We found that contingent frequency influenced the interpretation of sentences, and that the lexicon is richer in representation than structural models like the garden path model suggest.; To explain how the integration of multiple sources of evidence occurs, we proposed a dynamical systems mechanism that accounts for a range of data that other theories do not. It showed how representations for syntactic notions like categories, phrase types, and verbs could develop and interact. |
