Modeling learning of the English voicing contrast by Spanish listeners living in the United States

The importance of cue covariation in phonetic learning is explored through four experiments investigating perception of stop consonant voicing. Spanish and English show different uses of voice onset time (VOT; the time between consonant burst release and vocalic voicing onset) in cuing voicing perception. English contrasts short lag (20 ms) stops, whereas Spanish contrasts prevoicing (0 ms) stops. Secondary cues may also differ. In English, VOT and onset f0 (fundamental frequency at voicing onset) are positively correlated, and onset f0 plays a role in voicing perception. In Spanish these properties may not be as strongly correlated, meaning that onset f0 may be less relevant to voicing perception. As predicted, Spanish listeners tested in Spain showed a 0 ms VOT boundary with little use of onset f0, whereas English listeners tested in the US showed a 20 ms boundary with moderate use of onset f0. Significantly, Spanish listeners tested in the US showed an English-like VOT boundary, but made even greater use of onset f0 than did English listeners. Computational Hebbian modeling suggests a role for differences in each groups’ experience with specific patterns of VOT/onset f0 covariation.