Text Cohesion and Prediction of General Proficiency in Reading-Writing Integrated Tasks

This paper considers the degree to which performance data can help predict Japanese EFL learners’ general receptive proficiency in an integrated reading-writing task. In the previous studies, there have been a great amount of discussions as to the use of Automatic Essay Scoring for independent tasks such as writing test. However, there is little study which deals with reading-writing integrated tasks in order to show the predictability for general linguistic proficiency. Eighty-four Japanese EFL learners participated in this study. Our finding is that the 26 cohesive indices in the TAACO processing tool account for 38% of their proficiency in random forest and XGBoost. This paper concludes that this approach for the use of performance data and cohesive indices is promising for future research. In addition, in the case of Japanese EFL learners, the result of the occurrence of determiners has implications for second language acquisition theory in terms of linguistic factors.