Lexical training using accented speech improves non-native contrast discrimination

Abstract

Native American-English (AE) listeners have difficulty distinguishing between dental and retroflex stop consonants. For example, when asked to classify voiceless retroflex and dental stops, AE listeners typically label both items as the alveolar /t/. While AE lacks the dental-retroflex distinction, varieties of Indian English (IE) use this contrast to instantiate the AE /t/-/ð/ distinction (tanks vs. thanks). Specifically, the dental /t̪/ is used in instances where /ð/ would be used in AE while the retroflex /ʈ/ is used in place of /t/. This situation provides us with the opportunity to present the dental-retroflex difference in naturally produced tokens of English. In the current set of experiments, we examine the effect of exposure to IE tokens on AE listeners’ discrimination of /t̪/ and /ʈ/. Participants transcribed English words spoken by an IE speaker. The critical items were words that began with either /t/ (e.g., take), /ð/ (e.g., thorough), or were /t/-/ð/ minimal pairs (tanks vs. thanks). An AXB discrimination task performed before and after the transcription showed that AE speakers improved in their discrimination of /t̪/ and /ʈ/ after transcription. Subsequent studies examine this learning effect to investigate the role of specific lexical contexts and of speaker-specific information.