interaction of sentence context and local acoustic information during sentence comprehension

This study investigated the interaction of acoustic and semantic information for phoneme categorization during sentence comprehension. Voice onset time (VOT) was manipulated to form a goat-to-coat voicing continuum; target stimuli from this continuum were embedded in sentences biased toward `goat' or `coat'. Sentences were presented in conjunction with three distinct experimental tasks and several temporal positions. Experiment 1A used a cross-modal identification task (Borsky, Tuller, and Shapiro, 1998) to cue stimulus identification at the target; the results showed sentence context biases on identifications and sentence context congruency effects on response times for mid-range stimuli. Experiment 1B used a cross-modal identification task 450 ms after the target; results showed sentence biases on identifications that extended to the endpoints and no sentence context congruency effects on response times. Experiment 2 used a cross-modal interference task (CMI) with the same auditory stimuli. The primary task was listening to the sentences for comprehension; the interference task was a word/non-word decision to an unrelated visual probe that appeared at one of three temporal positions. This was the only task for which no explicit judgments about the identity of the target were required. Response times at the embedded target showed a significant effect of VOT only. The 450 ms later probe position showed a significant VOT x Context interaction; response times were significantly longer for endpoint stimuli when sentence meaning was biased toward the opposite endpoint. These results were interpreted as initially context-independent phonological processing followed by context integration. Experiment 3 used a word-monitoring task; subjects saw the word `goat' or `coat' briefly on a computer screen, then listened to the same sentence stimuli used for the other experiments. The task was to press a button as soon as the monitoring target was heard in the sentence. Results showed that sentence context did not bias identifications. However, response times were significantly longer when sentence context was incongruent with the monitoring target. Taken together, the results of the three distinct tasks support an account of phonological processing in which phoneme categorization is initially independent unless an explicit judgment about the identity of the target is required.