Score Prediction by SVM and its Implication for Japanese EFL Learners’ Essay Evaluation

Abstract

This paper discusses the future possibilities of employing SVM and Significant Word Detection for automatic essay writing for Japanese English-as-a-Foreign-Language (EFL) Learners. After reviewing the limitations of traditional frequency-based scoring using indices related to commonly assumed constructs for learners’ productive performances; that is, Complexity, Accuracy, and Fluency (CAF), this paper suggests the possibility to utilize the SVM model on Significant Word Detection instead of “frequency-based” scoring of the proposed indices on the basis of the essay data of 212 Japanese EFL learners on the Criterion Test. Specifically, the data (F-measure value) shows that the proposed model distinguish more clearly the difference in proficiency between Scores 1 and 2 on the Criterion Test in terms of indices and selected words rankings.