Discriminability of non-native tonal contours in low-pass filtered speech

Abstract

Native speakers of a language are able to easily detect non-native “accents”—usually a perceived accent is attributed to phonemic- and phonetic- level information. However, a growing body of work suggests that non-native speakers also exhibit differences from native speakers at the prosodic level. Differences are exhibited in the placement of lexical stress and phrasal accents, as well as the acoustic cues used in prominence production; prosodic differences in production appear to pattern in (native)-language-specific ways (Morrill, ICPhS Proceedings, 2015). In this experiment, we ask (1) whether differences in the realization of tonal contours are perceived by listeners, and (2) whether perceptual discriminability is patterning according to speakers’ native language. Participants listened to low-pass filtered phrases from the Stella passage recordings on the GMU Speech Accent Archive—produced by native speakers of English, Mandarin, Korean, Arabic, and Turkish—and judged whether speakers were saying the same thing. Results indicate that participants were less likely to rate two phrases as the same (even when they were) if they had been produced by speakers of different native languages. Patterns of discriminability across language pairings are attributed to differences in the number and timing of tonal events (e.g., pitch accents).