Artificial Intelligence-Generated Feedback for Second Language Intelligibility: An Exploratory Intervention Study on Effects and Perceptions

Abstract

This study investigated the use of Artificial Intelligence (AI) models and signal detection processes to generate meaningful visual and ChatGPT-like narrative feedback on second language (L2) English intelligibility. To test the effects and perceptions of such techniques, three groups of learners (N = 90) received visual and narrative feedback (n = 30), visual-only feedback (n = 29), and no feedback (n = 31) in an online self-paced intervention with explicit instruction on segmental and suprasegmental features of intelligibility. Pre/postspeaking tasks were evaluated by raters for intelligibility, comprehensibility, and accentedness, as well as segmental and suprasegmental accuracy, in scripted and spontaneous speech. The results indicate that visual feedback improves prominence production, but only those participants who also received the narrative (i.e., ChatGPT) feedback improved in two of the three prosodic features and in intelligibility. However, those who received narrative feedback had the lowest perceptions of the practice activity helpfulness. Implications for the use and improvement of AI-based pronunciation feedback are provided.