Phonological variation and inference in lexical access.

Recent experiments have indicated that lexical access in speech is highly intolerant of mismatch. An isolated sequence such as [symbol: see text] strongly disrupts access to the underlying lexical entry (wicked). This observation seems inconsistent with the systematic variability found in the phonetic form of words. Two cross-modal priming experiments tested the hypothesis that phonologically regular variation is perceptually acceptable. Participants heard tokens like [symbol: see text] embedded in contexts that either licensed the change as a result of a regular assimilation process (e.g., [symbol: see text]) or rendered the change phonologically unviable (e.g., [symbol: see text]). The tokens with contextually unviable deviations did not effectively access lexical representations. In contrast, the same tokens in viable phonological context primed as strongly as unchanged controls. These results suggest that mapping speech onto lexical representations involves on-line phonological inference.