Detection of nonnative speaker status from content-masked speech

Listeners are highly sensitive to divergences from native-speaker patterns of speech production, such that they can recognize second-language speakers even from very short stretches of speech. However, the processes by which nonnative speaker detection is accomplished are not fully understood. In this investigation, we used content-masked (backwards) speech to assess listeners’ sensitivity to nonnative speaker status when potential segmental, grammatical, and lexical cues were removed. The listeners performed at above-chance levels across utterances of varying lengths and across three accents (Mandarin, Cantonese, and Czech). Reduced sensitivity was observed when variability in speaking rates and F0 was removed, and when temporal integrity was severely disrupted. The results indicate that temporal properties, pitch, and voice quality probably played a role in the listeners’ judgments.