Yuna Bae, Okim Kang
Abstract Although artificial intelligence (AI) technology offers a promising approach to mitigate biases in listener judgments, the impact of different first language (L1) accents on AI understanding has not been widely investigated. Therefore, the current study examined (a) the presence of bias in AI intelligibility of L2‐accented speech and (b) differences in intelligibility as assessed by human experts and AI. Three data sets were used in this study: Data set 1 ( n = 12) with L2 speakers of Chinese, Indian, Spanish, and South African English, Data set 2 ( n = 60) with read‐aloud samples from the same L1s, and Data set 3 ( n = 40) with TOEFL responses from Arabic, Chinese, Korean, and Spanish speakers. Intelligibility was operationalized through transcription, using Apple's Siri and Google Assistant. Word error rates (WER) were calculated by an expert coder, and five human experts transcribed Data set 3 for comparison. Results showed that (a) AI systems' transcription accuracy varied by L1, and (b) human raters did not exhibit this bias. The findings have important implications for L2 learners and teachers by raising awareness of AI‐related fairness issues in L2 classrooms and technology applications.