Using GPT-3 as a More Knowledgeable Other to Provide Feedback on English Language Learners’ Spoken Transcripts
Paul Raine · Teaching English as a Second or Foreign Language--TESL-EJ · 2025
Generative Pre-trained Transformers (GPT) have shown promise as More Knowledgeable Others (MKO) for English Language Learners (ELLs) by providing on-demand, contextualized feedback within the Zone of Proximal Development (ZPD). In this study, a web app collected spoken responses from 680 adult ELLs, automatically transcribed them and sent them to GPT-3 for corrective and elaborative improvements. Of the 428 instances that participants rated, approximately 74% were judged “very helpful,” and perceived helpfulness rose with learner proficiency. Semi-structured interviews with six users offered illustrative insights into how feedback was interpreted and applied, particularly by intermediate-level learners. Although this research shows promise for utilizing GPT-3 as an MKO for ELLs, the implementation needs to be further refined and developed, and we need to continue to bear in mind the limitations inherent in GPT-based systems.