Abstract
<p>Recent experience rapidly shapes speech perception, but whether this learning influences speech production remains a central question for sensorimotor theories of speech. Prior work has shown that dimension-based statistical learning can transfer from perception to production for consonants. The present research tested whether this transfer extends to vowels and to vowel duration, a temporal dimension whose role in speech motor control is not fully specified in current models. Across four experiments, listeners were passively exposed to short-term distributional regularities that either preserved or reversed the typical English relationship between spectral quality and duration for /æ/–/ɛ/ vowels. Exposure to the reversed regularity reduced listeners’ reliance on duration in vowel categorization and reduced durational differentiation in their own vowel productions. These perception–production effects replicated across four experiments. Experiments 2–4 further showed that learning generalized beyond the exposure context to novel phonological environments that varied in coda, onset–coda context, and segmental structure, and that this generalized learning also transferred to production. The findings indicate that statistical regularities in others’ speech can dynamically reshape the weighting of temporal information in both vowel perception and production, placing new constraints on sensorimotor accounts of speech learning and control.</p>