Background/Objectives: Accented speech reflects systematic deviation from target-language phonetic norms. This study demonstrates that perceived accent strength covaries with selective, gradient differences in phonological feature realization. We examine whether perceived accents in Hindi- and Tamil-accented English reflect uniform segmental deviation or cue-specific patterns of phonological feature realization. Methods: English speech produced by native speakers of Hindi and Tamil was evaluated using native listener accentedness ratings. Phonetic variation was analyzed using posterior probabilities of phonological features derived from a machine learning model, Phonet. The analyses focused on liquids (laterals and rhotics (e.g., /l/, /ɭ/, and /ɻ/) and labial segments in the fricative–glide space (e.g., /v/, /w/, and /ʋ/), with attention to word position and feature-level generalization. Results: Accentedness ratings differed systematically for Hindi- and Tamil-accented English and covaried with a subset of phonological feature dimensions, yielding contrast- and context-specific patterns of perceptually relevant variation. Not all features that varied in production contributed to perceived accent strength. Conclusions: These findings support a cue-specific, perception-grounded account of accentedness and establish phonological feature posteriors derived from Phonet as interpretable phonological categories through which gradient L2 production differences are evaluated by listeners.
Venkateswaran et al. (Sat,) studied this question.