Profiling English sentences based on CEFR levels

Abstract

The study aims to demonstrate the procedure for constructing the CEFR-based Sentence Profile (CEFR-SP), a dataset with the CEFR levels assigned for sentences, and to identify the characteristics at each level. Basic statistics such as word length and sentence length are presented for each CEFR level for 7,511 carefully selected sentences, and statistical tests are conducted between adjacent levels to identify criterial features. The findings reveal significant differences in word length between adjacent levels, while word difficulty is not significant in discriminating levels at either end (A1–A2, C1–C2). Sentence length and depth are also not significant discriminators for higher levels (B2–C1, C1–C2). Notably, sentence-level data generally exhibit discriminative values compared to text-level statistics, indicating their direct capture of characteristics at each CEFR level.