Exaggeration of featural contrasts in clarifications of misheard speech in English

This study investigates the extent to which speakers manipulate featural distinctions when trying to clarify misunderstood speech, focusing on voicing contrasts in stops and height and backness (represented by F1 and F2) and durational contrasts in vowels. Participants interacted with a simulated speech recognizer, repeating words when they were “guessed” incorrectly. Both phonemically voiced and voiceless stops showed more extreme VOT values when elicited by an incorrect guess in which the consonant was a minimal pair in voicing with the target consonant (e.g. subject reads “bit”, computer guesses “pit”), but not when elicited by an open-ended request for repetition (e.g. subject reads “bit”, computer guesses “What did you say?”). A follow-up study showed that the change in VOT between the two repetitions was only present when the incorrect guess contrasted in voicing, but not when it contrasted in place or manner. In contrast, for vowels, the amount and direction of formant change in the F1–F  2 space was not significantly different from zero for either type of incorrect guess. However, when there was a durational component to the vowel contrast (/i/ vs. /ɪ/), speakers exaggerated the durational differences between the segments, as opposed to when there was not a durational contrast (e.g. /i/ vs. /u/). The results show that speakers perform local, systematic, and phonologically informed manipulations of temporal contrasts online when clarifying phonetic segments.

• We investigate whether speakers manipulate VOT or formants in clarifying speech.
• Participants repeat words that have been guessed incorrectly.
• Speakers exaggerate VOT when a segment has been guessed with the wrong voicing value.
• Vowel formant contrasts are not exaggerated, but contrastive durational differences are.
• Speakers use abstract phonetic categories during online production tasks.