Top-down effects of syntactic sentential context on phonetic processing.

Although much evidence suggests that the identification of phonetically ambiguous target words can be biased by preceding sentential context, interactive and autonomous models of speech perception disagree as to the mechanism by which higher level information affects subjects' responses. Some have suggested that the time course of context effects is incompatible with interactive models (e.g., TRACE). Two experiments examine this issue. In Experiment 1, subjects heard noun- and verb-biasing sentence contexts (e.g., Valerie hated the . . . vs. Brett hated to . . .), followed by stimuli from 2 voice-onset time continua: bay-pay (noun-verb) versus buy-pie (verb-noun). Consistent with prior research, identification of phonetically ambiguous targets was biased by the preceding context, and the size of this bias diminished in slower compared with faster responses. In Experiment 2, tokens from the same continua were embedded among filler target words beginning with /b/ or /p/ to elicit phonemically driven identification decisions and discourage word-level strategies. Results again revealed contextually biased responding, but this bias was as strong in slow as in fast responses. Together, these results suggest that phoneme identification decisions reflect robust, lasting top-down effects of lexical feedback on prelexical representations, as predicted by interactive models of speech perception.