UCLA

This study investigates the ability of adult American English listeners to recognize familiar words produced with an allophonic variant. It reports the results of a visual world paradigm experiment. Subjects were presented with /t/- and /d/-final target words produced with a canonical stop, a tap or a one-feature mispronunciation. For both /t/ and /d/-final words, subjects fixated to the target object more in the canonical stop condition than in the mispronunciation condition. Crucially, subjects fixated equally at the target image in both the canonical stop condition and the non-canonical tap condition across both word types. Further analysis, however, showed that listeners were faster at shifting their gaze to the target object when the label was produced with a stop than when it was produced with a tap. Our results suggest that both canonical stop and tap variants are viable labels for a given target word, although the canonical variant seems to be faster at activating a target lexical representation. The implications of these findings for models of spoken word recognition are discussed.