Long-lasting gradient activation of referents during spoken language processing

Abstract During speech processing, listeners must map a fundamentally continuous acoustic signal onto discrete symbols, such as words. A current debate concerns the time-course over which sub-phonemic (i.e., gradient) acoustic information continues to influence symbolic (i.e., linguistic) interpretation, which can provide evidence regarding the level of representation at which gradient information is maintained. In a visual-world paradigm experiment, participants indicated whether a spoken sentence matched a display while eye-gaze was monitored. Participants heard an acoustically ambiguous stimulus (a pronoun referring to either a male or female referent in the display), which was not disambiguated until later in the discourse. The acoustic properties of the pronouns and length of the ambiguous period were varied while responses and eye-movements to the discourse-relevant items were recorded, providing a measure of whether gradient referential uncertainty is maintained over time. Fixation patterns during the ambiguous period and latencies to fixate the target at the end of the trial varied linearly with the acoustics of the earlier pronoun, indicating that gradient information can be maintained over intervening periods of 35 syllables. These results provide strong evidence that gradient uncertainty is maintained at the level of referent representations.