Proficiency, phonological context, and distributional control in Arabic L1 learners' English stop voicing

Arabic and English differ in how they organize stop voicing along the voice onset time (VOT) continuum. Jordanian Arabic typically contrasts prevoiced /b/ with short-lag voiceless stops, whereas General American English uses short-lag /b/ and long-lag aspirated /p/ in stressed singleton onsets, with aspiration suppressed after /s/. Within the revised Speech Learning Model (SLM-r), such learning should appear not only as shifts in mean VOT but also as changes in dispersion, overlap, perceptual boundary location, and context-sensitive control. Sixty-five adults participated: 45 Jordanian Arabic L1 learners of English and 20 General American English controls. Learners completed an institutional English placement test that excluded speaking or pronunciation, and all participants produced real-word English stimuli containing singleton /b/, singleton /p/, and /sp/ cluster /p/. Participants also completed VOT identification and context-naturalness perception probes. VOT was measured from unaltered waveforms in Praat, with strong inter-rater reliability. Learners distinguished /b/ and /p/ in production, but their singleton /p/ VOT remained shorter and more variable than that of controls. Higher proficiency predicted longer singleton /p/ VOT, reduced within-speaker dispersion, reduced /p/-/b/ distributional overlap, lower perceptual /b/-/p/ boundary values, and stronger suppression of aspiration in /sp/ clusters. Learners nevertheless continued to rate overaspirated /sp/ tokens as relatively natural, indicating residual difficulty with context-conditioned aspiration. The findings support a distribution-sensitive account of L2 stop learning: proficiency-related development is reflected in the tightening and separation of VOT distributions and in emerging phonological control, not in mean category placement alone.