A Corpus-Based Approach to Investigate the Cohesive Features Across Different Levels of CEFR

Abstract

Despite plenty of previous studies pointing out the importance of validating the CEFR scale, scant attention has been given to the validation of the CEFR cohesion scale based on learners’ corpus. This study aims to examine the cohesive features of written texts at different levels of the CEFR using a corpus-based approach. Employing the TAACO and Coh-Metrix tools, this study identified identifies seven categories of key cohesive features, namely connectives, lexical overlap (sentence), Type-token ratio (TTR) and Density, givenness, semantic overlap, hypernymy and deep cohesion of the CEFR. The results showed that hypernymy and deep cohesion were the strongest predictors to distinguish CEFR levels and these categories generally kept a nonlinear relationship with CEFR levels. This study provides empirical evidence to further validate and refine the CEFR cohesion scale and casts light on the development of cohesive competence across different levels of the CEFR from the perspective of second language acquisition. More importantly, this study can provide pedagogical implications for learning and assessing cohesive competence.