Parsing coarticulated speech in perception: effects of coarticulation resistance

A speaker produced schwa-CV disyllables in which consonants were low or high in coarticulation resistance. Articulatory and acoustic measurements verified that the magnitude, but not the extent, of anticipatory coarticulation from the stressed vowel to the schwa was modulated by coarticulation resistance. In a perception experiment, listeners heard schwa-CV disyllables that ended in stressed /i/ or /a/ and made speeded identifications of the stressed vowels. Some disyllables had been cross-spliced so that a schwa vowel that had originally been produced in the context of C/i/ was spliced onto a C/a/ syllable or vice versa. Other disyllables were spliced so that a schwa originally from a C/i/ (or C/a/) context was spliced onto a different C/i/ (C/a/) token. Listeners' response latencies were slower to cross spliced than to spliced disyllables but only in the context of low rather than high resistant consonants. The outcome is generally consistent with a hypothesis that listeners to speech “parse” the acoustic signal along coarticulatory or phonetic gestural lines and that success in parsing varies with the amount of acoustic evidence talkers provide. However, findings did not suggest “perfect” parsing. Correlations between the amount of acoustic information provided by the speaker and the extent to which listeners were disrupted by cross splicing were nonsignificant.