Intention-based Corrective Feedback Generation using Context-aware Model

In order to facilitate language acquisition, when language learners speak incomprehensible utterances, a Dialog-based Computer Assisted Language Learning (DB-CALL) system should provide matching fluent utterances by inferring the actual learner’s intention both from the utterance itself and from the dialog context as human tutors do. We propose a hybrid inference model that allows a practical and principled way of separating the utterance model and the dialog context model so that only the utterance model needs to be adjusted for each fluency level. Also, we propose a feedback generation method that provides native-like utterances by searching Example Expression Database using the inferred intention. In experiments, our hybrid model outperformed the utterance only model. Also, from the increased dialog completion rate, we can conclude that our method is suitable to produce appropriate feedback even when the learner’s utterances are highly incomprehensible. This is because the dialog context model effectively confines candidate intentions within the given context.