Curriculum-Driven Edubot: A Framework for Developing Language Learning Chatbots through Synthesizing Conversational Data

Chatbots are increasingly used in education, but general-purpose systems often fail to align with structured curricula and learners’ proficiency levels. This paper presents Curriculum-Driven EduBot, a framework that extracts topics from English textbooks, synthesizes dialogues around those topics using large language models (LLMs), and fine-tunes an open-source LLM to serve as a conversational partner aligned to curriculum content and CEFR proficiency levels. Dialogues incorporate fixed-format personas and target vocabulary from the textbook to ensure topic relevance and appropriate difficulty. User studies show that EduBot outperforms ChatGPT at leading curriculum-based dialogues and adapting to learners’ English proficiency, offering an interactive, user-tailored tool that enhances conversational practice within a pedagogically coherent framework.

Y U Li, Shang Qu, Zhou Yu, Yu Li, Jili Shen, Shangchao Min