Integrating LLM-based tutor into digital game-based learning for Chinese as a foreign language: impacts on language acquisition, motivation, and learning behaviours

Although Digital Game-Based Learning (DGBL) has been widely recognized as an effective approach for foreign language learning, its feedback mechanisms are often predefined and limited in adaptivity. Recent advances in large language models (LLMs) offer new opportunities to provide adaptive support within DGBL environments. This study investigates the impact of integrating an LLM-based tutor into a DGBL environment for Chinese as a foreign language (CFL) learning, focusing on language acquisition, motivation, and process-oriented learning behaviours. A quasi-experimental study was conducted with 66 CFL learners, who were randomly assigned to either a DGBL + LLM-based tutor group or a DGBL-only group. Participants completed six 20-minute gameplay sessions across a two-week intervention. The LLM-based tutor provided real-time corrective feedback, metalinguistic explanations, and free-form Q&A. Results showed that, after controlling for baseline performance, learners in the DGBL + LLM-based tutor group achieved significantly higher discourse comprehension and motivational intensity than those in the DGBL-only group. Within the LLM-based tutor group, learners also showed significant improvements in extrinsic motivation, intrinsic motivation, and self-confidence over the intervention. Behavioral sequence analysis further showed that learners in the DGBL + LLM-based tutor group were more likely to transition from error correction to subsequent correct responses, exhibited fewer voluntary gameplay interruptions, and alternated between asking questions and reviewing previous tutor explanations. Together, these findings suggest that integrating an LLM-based tutor into DGBL can strengthen discourse learning while supporting learner motivation and more adaptive learning behaviours in CFL learning.