A Data-Driven Cognitive Feature–Based Model for English Text Readability Assessment to Support College English Instruction

Text readability assessment is essential for effective college English teaching and instructional material selection. Traditional readability models mainly rely on surface linguistic features and often fail to reflect the cognitive processes involved in reading comprehension. To address this limitation, this study proposes a cognitive feature–based approach for English text readability assessment, aiming to support data-driven English teaching evaluation. The proposed method incorporates cognitive features related to lexical rarity, logical complexity, and comprehension difficulty, and integrates them with conventional linguistic features. Multiple machine learning models are employed and evaluated on four benchmark datasets, including CEFR, CLEC, OneStopEnglish, and RACE. Experimental results show that models combining cognitive and linguistic features consistently outperform those using linguistic features alone across multiple evaluation metrics. The findings indicate that cognitive features provide complementary information for readability assessment and enhance the discriminability of readability levels. This study offers practical implications for college English teaching by enabling more accurate matching between reading materials and learners’ proficiency levels, thereby supporting personalized and data-driven instructional decision-making.