The grammatical aspects of word recognition

This thesis uses experimental and computational techniques to explore the hypothesis that much of the syntactic processing of language is the direct result of word recognition. This hypothesis offers an alternative to a longstanding and influential view that lexical processing and syntactic processing are accomplished by separate mechanisms (e.g., Frazier, 1978, 1989). The thesis expands the theoretical and empirical basis for the Constraint-Based Lexicalist approach to the understanding of language comprehension (e.g., MacDonald, Pearlmutter, & Seidenberg, 1994; Trueswell & Tanenhaus, 1994). I describe results from four experiments, which embed a lexical priming technique in “garden-path” experiments, the standard technique for the study of syntactic processing in psycholinguistic research. This combination of experimental techniques allows the manipulation of lexical processing in readers as they engage in a task that is highly sensitive to syntactic processing. I also describe a new computational model of human sentence processing, which characterizes much of sentence processing as lexical ambiguity resolution. The model is an integration of ideas from the Constraint-Based Lexicalist approach to sentence processing, Lexicalized Tree-Adjoining Grammar (Joshi & Schabes, 1996), and statistical processing techniques from computational linguistics (e.g., Marcus, 1995)