Cognitively Informed Cross-Lingual Curriculum Learning for Adaptive Reading Comprehension

Adaptive educational systems rarely incorporate direct psychophysiological evidence about learners while also handling multilingual content with rigor. We propose NeuroCL, a cognitively informed curriculum learning framework that uses moment-to-moment estimates of cognitive load (derived from pupil dilation and eye movement features) to steer a cross-lingual reading comprehension curriculum. The framework combines (i) neurocognitive signals, (ii) linguistically grounded difficulty features (lexical overlap, morphological complexity, dependency length), and (iii) model uncertainty (surprisal/entropy from multilingual encoders) to adaptively select texts and questions in multiple languages. We outline theory, algorithms, and an empirical protocol, and we report results from a privacy-preserving pilot with synthetic-but-plausible data that demonstrate improved comprehension, faster reading times, and reduced cross-language performance gaps relative to static or heuristic curricula. We discuss implications for fair and transparent multilingual learning technologies.