How L1 – L2 Typological Distance Moderates the Interplay Between Cognitive Resources and LLM Predictors in L2 Reading

This study investigates how typological distance between a learner’s first (L1) and second (L2) languages moderates the predictive validity of large language model (LLM)-based metrics in L2 reading, alongside the compensatory role of cognitive resources. Analysis of eye-tracking data from Chinese learners with L1 Lao, Burmese, Thai, and Vietnamese revealed that surprisal significantly accounted for processing difficulty across all L1 groups, whereas contextual entropy was significant only among Vietnamese speakers. Furthermore, the incremental explanatory power of cognitive abilities was significant only in the Vietnamese group, with only verbal working memory showing a robust independent effect. Additionally, working memory and attention significantly interacted with typological distance: higher cognitive capacity mitigated the processing burden associated with greater L1 – L2 divergence. Collectively, these findings demonstrate that the applicability of LLM-derived predictors is constrained by typological factors and modulated by cognitive resources, underscoring the need for adaptive models of L2 processing that incorporate both individual cognitive differences and cross-linguistic profiles, while providing empirical support for the Cognitive Resource Coordination Model (Robinson, 2001). Keywords: second language reading; L1-L2 typological distance; large language models (LLMs); predictive processing; cognitive abilities