High-Performance Configurations in Game-Based English Learning: How AI Support and Task Complexity Matter

Abstract Artificial intelligence is not only empowering education but also challenging traditional educational designs and learning models. Against this backdrop, game-based learning needs to pay greater attention to how different learning behaviors adapt to increasingly complex tasks to facilitate high task performance. This study focuses on EFL learning in higher education and develops a configurational framework based on internally driven conditions and external support conditions in GBL-based EFL learning. To compare changes in learning-behavior configurations across different task stages, 33 students completed three tasks with low, medium, and high levels of complexity. Fuzzy-set qualitative comparative analysis (fsQCA) was then employed to identify behavioral configurations associated with high task performance. The results show that, in an AI-scaffolded learning environment, the progressive change in task complexity was accompanied by changes in the number of learning-behavior configurations. As task complexity increased, the configurational roles of internally driven conditions and AI scaffolding conditions became increasingly differentiated within high-performance configurations. These findings suggest that GBL-based EFL performance does not depend on any single learning behavior but rather on the dynamic configuration of learner behaviors and external support conditions in response to task demands.