DEVELOPMENT AND EVALUATION OF ADAPTIVE ENGLISH PRACTICING SYSTEM FROM ERROR-ANNOTATED LEARNER CORPUS

Learners of English as a second language usually make some errors during their learning process. Some of them do not affect the communication while the others should be corrected. This study proposes a computer-assisted language learning (CALL) system for increasing error awareness, aiming for correct English usage and improvement of English proficiency of learners. The system incorporates two principles: learning from negative evidences, and prioritizing communication errors. Negative examples, specifically erroneous sentences, are extracted from learner corpus, to create practice exercises. These exercises constitute an intelligent tutoring system, by exposing adaptively to abilities and weaknesses of each user, determined by his responses while using the learning system. Among all type of errors in the exercises, ones that impaired understanding, so-called communication errors, gain priority for being studied. Unlike traditional language instructions, actual frequent errors made by learners in general are emphasized, while individual errors are corrected. Moreover, learners are directed to focus on their weaknesses and unintelligible errors, thus it is considered as a straightforward and effective method. To evaluate, experiments are conducted by target users, which are Japanese learners. After using prototypes, evaluation is done in aspects of adaptive mechanism, learning achievement, and satisfaction of learners. The overall result concludes that the system has more potential to improve English ability than traditional non-adaptive approaches.