evaluating cohesion as a predictor of writing quality: an analysis of local, global, and text-level indices in ielts writing task 2

This study examines the predictive validity of local, global, and text-level cohesion indices for human ratings of cohesion and coherence (cc) in ielts writing task 2. a corpus consisting of 105 essays rated by official ielts examiners was analyzed using the tool for the automatic analysis of cohesion (taaco 2.0.4). forty cohesion indices were extracted and correlated with cc scores using spearman’s rank correlation. the results revealed that additive connectives and overall connectives use relate positively to cc scores, while too much redundant local lemma overlap relates negatively. in the global category, overlap of function lemmas and semantic similarity between paragraphs were positively correlated with higher ratings. at the text level, lexical diversity (mattr and ttr) positively correlated most strongly, while lexical repetition had a negative effect on perceived quality. this suggests that selective and varied use of cohesive devices enhances writing quality, whereas repetition detracts from it. further implications for automated essay scoring and l2 writing instruction are discussed.