Abstract
Performance assessment attracts rising attention in recent English language education, but little has been introduced to English language program in Japan. This is because such assessment imposes severe burdens on instructors: conducting interview, rating students' performance, and analyzing and delivering scores. An automated scoring system for students' performance can be a solution to this problem. To present the solution, we developed an automated scoring system for speech by Japanese learners of English and examined the practicality of the system in an English language program. The system assesses the students' responses to the discourse completion tasks that were created to elicit the expressions learned in the lessons. The responses are represented as vectors applying the bag-of-words model, and the system scores the response using Support Vector Machine and Naive Bayes Classifier. The accuracy of these two prediction methods is estimated using cross-validation. The exact agreement is 74% between our system and the human scores.
