Designing large language model-based agents with 5E framework for ESL learners’ grammar acquisition

This study examines how large language models (LLMs) aid English as a Second Language (ESL) learners in acquiring grammar. Two artificial intelligence (AI) agents were designed: one as a conventional English teacher and another utilizing the 5E framework (engage, explore, explain, elaborate, evaluate) for inquiry-based learning (IBL). Thirty-seven ESL students were randomly divided into two groups: 17 in the AI-facilitated conventional English teacher group and 20 in the AI-facilitated 5E English teacher group. Pre- and post-tests, along with interviews, were used to investigate students’ intrinsic motivation, cognitive changes, and performance transformation with the designed AI agents. The study revealed that high-performing students responded positively to the AI teacher, while low-performing students exhibited mixed attitudes. Additionally, it reported the challenges encountered in applying the 5E framework. This study addresses the gap where LLMs are rarely tested as “instructional facilitators” of structured pedagogical frameworks for grammar learning—moving beyond their traditional role as “resource providers.” It demonstrates that LLMs can operationalize the 5E framework to support systematic grammar acquisition, offering theoretical insights into integrating IBL with LLM technology. Practically, it provides educators with guidance on matching LLM agents to learner proficiency (e.g., 5E-based agents for high-performing students, conventional agents for low-performing ones) and highlights directions for optimizing 5E-based LLMs to better support struggling learners. Furthermore, it addresses the prior neglect of learner diversity in LLM studies by analyzing high- and low-performing subgroups, laying a foundation for evidence-based AI-driven ESL grammar teaching.