Adaptation to accent is proportionate to the prevalence of accented speech

Abstract

Listeners rapidly reweight the mapping of acoustic cues to speech categories in response to abrupt introductions of accented speech. For instance, when encountering an accent that reverses the typical correlation of acoustic cues to speech category membership, listeners rapidly down-weight reliance on secondary cues [Idemaru and Holt, 2011; 2014; Liu and Holt 2015]. Here, we examined the impact of experiencing mixtures of typical and accented speech in evoking cue down-weighting to test how much information the system requires to adapt. Listeners recognized words as beer versus pier across 41 blocks of 20 trials. In the first block, participants exclusively encountered tokens with the canonical English relationship between fundamental frequency (F0) and voice onset time (VOT); low F0 frequencies were associated with short VOT durations whereas high F0 frequencies were associated with long VOT durations. In subsequent blocks, the ratio of canonical to accented tokens (for which the F0/VOT correlation was reversed) was changed by 5% per block until participants experienced exclusively accented speech. Canonical tokens were then reintroduced incrementally, until participants exclusively heard canonical speech. Reliance on F0 for word recognition linearly decreased as the proportion of accented speech increased. Implications for models of the speech perception system are discussed.