Identifying effective cohesive features for task classification in integrated reading-writing tasks

Abstract

Many language proficiency assessments have begun to include integrated writing tasks, which require test takers to read and/or listen to the passage and write a response based on the information in the passage. Especially in the research of integrated reading-writing tasks, linguistic features related to cohesion have been seen as indicators of the writing response because it has a relationship with both reading and writing skills. In this study, an integrated reading-writing task is conducted on Japanese English as a Foreign Language (EFL) students. By using machine learning methods, we investigate the relationship between the cohesiveness of the source text and the use of cohesive devices in writing responses in the task. The results of Decision Tree and Random Forest show some significant cohesive devices for the classification of the task. In future research, we will closely look into the linguistic features of each writing response and participants’ writing strategy use in the task. It will provide important implications for increasing the construct validity of integrated reading-writing tasks and writing pedagogy.