Exploring large language models for L2 metaphonological awareness tutoring

Abstract This is the first observational study to evaluate the feasibility of implementing large language models (LLMs) for second language (L2) metaphonological awareness training. A custom implementation of GPT-4 acting as a homework tutor was piloted in an English phonetics and phonology course for first-year university students. Two novel homework assignments were designed to leverage the LLM’s strengths and explore its weaknesses. Analysis of learner interaction logs, homework reflections, and survey data revealed that most learners perceived the AI Tutor as helpful for its personalised explanations. However, the overall sentiment was mixed due to the LLM’s propensity for confabulation. Despite these challenges, the pilot demonstrated the potential for LLMs to engage learners in active and self-regulated learning. Recommendations for future directions include designing LLM-based learning environments, promoting AI literacy among educators and learners, and experimentally researching long-term effects of AI tutors on learning outcomes.