Cohesion in the discussion section of research articles: A cross-disciplinary investigation.

Despite its prominence and functionality in academic writing, cohesion is under-researched in academic genres, including research articles (RAs). Moreover, there is little cross-disciplinary research on cohesion in academic discourse. Thus, this study aimed to investigate cohesion in the discussion section of RAs at sentence, paragraph and text levels, across three disciplines (i.e., applied linguistics, chemistry, and economics). To this end, 24 indices of local, global, and text cohesion were analyzed in a corpus of 300 discussion sections (100 from each discipline). MANOVAs identified significant cross-disciplinary variations in local, global, and text cohesion. Specifically, indices of local cohesion were generally higher in applied linguistics discussions, but measures of global, and text cohesion were mostly higher in chemistry and economics texts, respectively. Random forest modeling revealed that negative connectives were the most powerful classifiers of applied linguistics discussions, whereas adjacent sentence overlap noun synonyms and positive connectives were the best predictors of chemistry and economics discussions, respectively. These results are discussed with a view to offering theoretical and pedagogical implications for English-for-specific-purposes researchers and practitioners. • Indices of local cohesion were generally higher in applied linguistics discussions. • Chemistry and economics feature global and text cohesion respectively. • Local cohesion indices were the best classifiers in the three disciplines. • Random forest modeling complemented MANOVA results. [ABSTRACT FROM AUTHOR]