The effect of contextual mismatches on lexical activation of phonetic variants

Listeners' everyday experience of speech is highly varied and much of this variation is richly informative. We routinely make accurate use of phonetic cues in fast speech, for example, that would be ambiguous if decontextualized. In an influential study, Andruski et al. (1994) showed that a word like “king” facilitates the recognition of a semantically-related target like “queen” better with a canonically long VOT on the initial [k] than when portions of this VOT are spliced out. These and similar findings are cited as evidence for idealized or canonical lexical representations–a surprising result given the rarity of these forms in listeners' experience. In a series of semantic priming experiments, we test the hypothesis that such findings are attributable not to a benefit for the canonical variant but to a mismatch between the manipulated variants and the contextual phonetic frame in which they are presented. The short VOT variants facilitate recognition equally robustly as their long VOT counterparts wh...