Presenting texts at an appropriate difficulty level is essential for maintaining reader motivation and promoting effective learning. While recent advances in natural language processing have made it possible to automatically simplify texts, manually identifying and rewriting every passage that readers may find difficult remains time-consuming. To address this, it is important to automatically predict which parts of a text are perceived as difficult and selectively simplify those segments. Previous studies have explored the prediction of subjective reading difficulty using syntactic and eye-tracking features. More recently, advances in computational semantics have enabled the quantification of semantic features. This study proposes a novel approach to predicting subjective text difficulty by incorporating semantic features. The results demonstrate the effectiveness of semantic features in difficulty prediction and reveal a significant relationship between subjective difficulty and semantic similarity.
KIUCHI et al. (Wed,) studied this question.