A cognitive resource trade-off model of AI-assisted language learning in Chinese undergraduates

Abstract

Artificial intelligence is becoming increasingly embedded in language learning, yet its psychological consequences remain insufficiently understood. In a cross-sectional survey of 4,010 Chinese undergraduates who used AI tools for English learning, this study examined the associations among AI-assisted learning engagement, cognitive load, metacognitive monitoring, reinforcement experience, English learner identity, and sustained usage and investment intention. Confirmatory factor analysis supported the six-factor measurement model, and structural equation modeling showed excellent fit. AI engagement was positively associated with both metacognitive monitoring and cognitive load, indicating simultaneous facilitative and constraining pathways. Cognitive load was negatively associated with metacognitive monitoring and learner identity, whereas metacognitive monitoring and reinforcement experience were positively associated with learner identity. The decomposition of indirect effects further showed that AI engagement was positively associated with learner identity through metacognitive monitoring but negatively associated with learner identity through cognitive load. An alternative model adding a direct path from AI engagement to learner identity did not improve model fit, supporting the more parsimonious theoretical model. Reinforcement experience and learner identity were positively associated with sustained usage and investment intention. These findings suggest that the educational value of AI-assisted language learning depends not only on the extent of AI use, but also on whether AI feedback can support reflective regulation without imposing excessive cognitive burden.