Detecting Fine-Grained Syntactic Features for Predicting Japanese EFL Learners’ Writing Proficiency

Abstract

In the field of foreign language writing research, linguistic features (e.g., mean length of sentence, type-token ratio) that appear in learners’ performance have been utilized to gauge learners’ development. Among them, syntactic aspect of linguistic structure has prominently been investigated. While there are wide variety of features, however, it is not clear what kind of features should be implemented to improve the quality of writing assessment. This study aims at detecting fine-grained syntactic features for predicting Japanese English-as-a-foreign-language (EFL) learners’ writing proficiency. In the analysis, we used 5,000 argumentative essays written by Japanese learners of English, which are assigned five proficiency levels. A total of 78 fine-grained syntactic features are computed from the essays with TAASSC (Kyle, 2016) and employed as predictors of proficiency levels in a random forest classifier. The results suggest that noun phrase elaboration, use of modal, use of passive voice, and verb based syntactic knowledge contribute to the prediction of the proficiency levels.