The role of verb-specific lexical information in syntactic ambiguity resolution

Four experiments investigated how verb-specific lexical information is used in resolving the noun phrase complement/tensed sentence complement ambiguity, extending prior research (Ferreira & Henderson, 1990; Holmes, Stowe, & Cupples, 1989; Trueswell, Tannenhaus, & Kello, 1993). Predictions from the Constraint Satisfaction Approach (MacDonald, 1994; MacDonald, Pearlmutter, & Seidenberg, 1994a; 1994b; Tannenhaus & Trueswell, 1994; Trueswell, Tannenhaus, & Kello, 1993) and the Lexical Filtering Proposal (Clifton, Speer, & Abney, 1991; Ferreira & Henderson, 1990; 1991; Frazier, 1987; Frazier & Clifton, 1989) were contrasted. The former approach assumes that lexical information is used to guide the analysis of syntactically ambiguous phrases, predicting that comparable effects of verb bias should be observed for ambiguous versus unambiguous tensed sentence complements as for sentences containing temporarily ambiguous noun phrase complements and for sentences containing unambiguous tensed sentence complements. The latter proposal, an extension of the Garden Path Model (Frazier, 1978; Frazier & Fodor, 1978; Frazier, & Rayner, 1982), assumes that lexical information may be used when it becomes available; however, the analysis of syntactically ambiguous phrases is not delayed until lexical information becomes available, but instead is made in accordance with the syntactic parsing principles Minimal Attachment and Late Closure. Therefore, larger effects of verb bias are predicted for ambiguous versus unambiguous tensed sentence complements than for sentences containing temporarily ambiguous noun phrase complements or for sentences containing unambiguous tensed sentence complements. In Experiments 1-3, two self-paced reading methods (phrase by phrase and word by word presentation) and eye tracking were used to compare reading time on sentences containing ambiguous and unambiguous tensed sentence complements, containing either short or long ambiguous noun phrases, preceded by either NP-biased verbs, i.e., verbs generally occurring most frequently with noun phrase complements, or S-biased verbs, i.e., verbs generally occurring most frequently with tensed sentence complements. In Experiment 4, eye tracking was used to compare reading time on sentences containing temporarily ambiguous tensed sentence complements, temporarily ambiguous noun phrase complements, and unambiguous tensed sentence complements, containing either short or long ambiguous noun phrases, preceded by either NP-biased or S-biased verbs. Results from these four experiments are most compatible with the Lexical Filtering Proposal. Implications for models of human sentence processing are discussed