Gwilliams, L. 1, 2 , Linzen, T. 1 , Neophytou, K. 2 , Poeppel, D. 1, 3 & Marantz, A. 1, 2
1 New York University
2 NYUAD Institute
3 Max-Planck Institute
Behavioural studies posit that commitment to a phonological category can be delayed over 1s after onset, using insensitivity to subphonemic variation (e.g., voicing) as a metric for commitment. Our results support a contrary claim: while sensitivity to phoneme ambiguity indeed extends over 1s, phoneme-category commitment is an independent computation that resolves earlier. Participants (n=25) listened to word-pairs that, apart from the initial consonant, have an identical speech stream until point-of-disambiguation (POD) (e.g.,'parak-eet','barric-ade'). Pre-POD onsets were morphed to create voicing or place-of-articulation continua (e.g. parak-eet<->barak-eet). Magneto-encephalography was recorded concurrently. Phoneme-category ambiguity significantly modulated early responses in left Heschl's gyrus from 40-60ms after word onset and 40-60ms time-locked to POD. Crucially, sensitivity to onset ambiguity did not interact with POD-latency. However, activation of competing lexical items (modelled by surprisal) did significantly interact: before ~500ms, expectations were weighted by subphonemic acoustic detail; after that, expectations were commitment-dependent. Our results implicate maintenance of subphonemic detail until at least 700ms, reflected by reactivation of Heschl's gyrus. By contrast, phonological commitments are restricted to a short integration window. These findings suggest that modelling predictions of upcoming sounds is a reliable metric for phonological commitment, which is separable from retroactive sensitivity to subphonemic detail.