SHIZUKO TOMIOKA (SANNO UNIVERSITY)

WORD ERROR RATES IN AI AUTOMATIC SPEECH RECOGNITION ACROSS ENGLISH VARIETIES: IMPLICATIONS FOR NATIVE SPEAKERISM : WORLD ENGLISHES/ENGLISH AS A LINGUA FRANCA

Artificial intelligence (AI) has significantly transformed language learning, with automatic speech recognition (ASR) being one of the key features that aids particularly the development of speaking skills. Despite its benefits, non-native English learners often encounter challenges with ASR accuracy, and ASR failure could lead non-native learners to blame their pronunciation, thereby prompting an investigation into the potential risks associated with native speakerism among non-native English learners. Considering the diverse sociolinguistic landscape of English, it is crucial to shift emphasis away from pronunciation accuracy and towards the ability to adapt speech for effective communication. This study investigated the word error rates of two prominent AI models, Chat GPT 3.5 and Copilot, across three distinct English varieties: General American, Received Pronunciation, and Japanese English. Through a comparative analysis of ASR performance across these varieties, the study sought to unveil disparities that might perpetuate native speakerism among learners. The findings revealed significant differences in word error rates, particularly with Japanese English exhibiting notably higher error rates compared to native speaker varieties. These disparities underscore the necessity of acknowledging and accommodating linguistic diversity when using AI in language classrooms. Based on the results, the study also has implications for English language teaching that are in line with the contemporary sociolinguistic landscape of English.

Shizuko Tomioka is a lecturer at SANNO University. Her research interests include English as a lingua franca (ELF), native speakerism, and English language teaching. More specifically her research focuses on the influence of ELF teaching on learners’ attitudes toward ELF and the development of materials related to ELF teaching.