Generative AI in Foreign Language Learning and Teaching: A Summary and Insights from Our Empirical Investigations

Abstract

This paper reviews our empirical investigations into generative AI (GenAI) in foreign language (L2) learning and teaching. Validation of large language model (LLM) performance showed GPT’s strong correlation with human L2 writing accuracy assessments, outperforming Grammarly and demonstrating effectiveness in automated essay scoring. However, LLMs achieved only moderate agreement in classifying open-ended responses, underscoring the need for human oversight. I also identified distinct linguistic features enabling accurate detection of AI-generated texts. Classroom studies revealed Japanese EFL (English as a Foreign Language) learners’ preference for ChatGPT in editing/proofreading, and that their “Ought-to L2 Self” significantly predicted AI tool use. Building on these findings, I proposed the Metacognitive Resource Use (MRU) framework for strategic use of language resources. This framework underlies LexiTracker, a web application providing vocabulary and writing feedback to promote metacognitive engagement. This review offers practical and theoretical insights for the effective integration of GenAI in language education.