Task graph based task-oriented dialogue system using dialogue map for second language learning

This paper presents a rule-based task-oriented dialogue system for second language learning and a knowledge extraction method which automatically extracts the training data for Natural Language Understanding (NLU) and dialogue rules for dialogue management from a Dialogue Map (DM). The DM consists of turn-byturn utterances between the system and the learner. Therefore, the proposed method can automatically extend a new dialogue domain by constructing a dialogue map with a simple format. We constructed two dialogue maps for English and Korean, respectively, and implemented English and Korean task-oriented dialogue systems using the DMs. In the experiments, although the turn success rates are relatively low (78.1% in English and 78.76% in Korean), the task success rates are 90.83% in English and 99.17% in Korean. The systems constructed by the proposed method should enable learners to communicate successfully in the topic despite some mistakes in the system responses.