Task-oriented spoken dialog system for second-language learning

This paper introduces a Dialog-Based Computer Assisted secondLanguage Learning (DB-CALL) system using task-oriented dialogue processing technology. The system promotes dialogue with a second-language learner for a specific task, such as purchasing tour tickets, ordering food, passing through immigration, etc. The dialog system plays a role of a ticket agent, a waiter, or an immigration officer at each task and the learner completes the task through talking with the system in the second-languages. The system gives a new situation of the task at every trial, so that the learner could have a different experience and learn various expressions in the same task. Our DB-CALL system is based on a task-oriented dialog system consisting of a language understanding module using a structural support vector model and a dynamic dialog graph based dialog management module. Our task-oriented dialog system is trained from the intention-annotated real dialog scripts. The experiments for the performance of our proposed dialog system show the average task success rate of 85.52%, the average turn success rate of 85.32%, and the average turn length of 14.61. In a satisfactory survey targeting the subjects, the subjects agreed that our system makes them learn the second-language more efficiently and with less effort.