RCS-CSLT: a language feature-driven readability classification system for international Chinese education

Abstract

The provision of reading materials that are tailored to the proficiency levels of second language (L2) learners is crucial for fostering the development of their language skills. However, there is a limited availability of reliable, and contextually relevant Chinese text readability classifiers that meet the latest demands of Chinese as a Second Language (CSL) learners. Consequently, accurately assessing the readability of uncategorized texts presents significant challenges for both students and educators. This study introduces the development of a Readability Classification System for CSL text (RCS-CSLT), which aims to grade the readability of CSL texts. The RCS-CSLT utilizes a CSL-specific BERT architecture, incorporates Chinese Proficiency Grading Standards (CPGS) for International Chinese Language Education issued by the Ministry of Education of China, and integrates Chinese language features, including lexical richness, syntactic complexity, and syntactic patterns. To evaluate its performance, a CSL text dataset was used and compared with a baseline BERT model. The results demonstrate that the RCS-CSLT model achieves excellent performance in predicting the readability levels of CSL texts, achieving an average accuracy of 89.8%, and consistently outperforms the baseline model in terms of classification performance at each level. An analysis of a wider range of Chinese text reveals that RCS-CSLT attains an accuracy exceeding 90.3% when compared to human experts’ evaluations in readability assessments. These results affirm the proficiency of RCS-CSLT in appraising the readability of Chinese texts, showcasing its efficacy within the realm of international Chinese language instruction and study. This technological advancement offers benefits to both learners and educators. For learners, RCS-CSLT functions as a grading indicator and auxiliary tool for selecting appropriate reading materials, while for educators, it provides insights into text readability levels and aids in designing instructional plans, particularly in computer-assisted language learning.