The dialogue management module of current practical dialogue systems tend to be implemented by using relative simple rules, such as, state transition or dialogue grammar [1]. These rules were designed for particular task domain and were paid little attention to the modularity. As a result, it is difficult to port the dialogue manager to another task domain. However, based on recent success of speech recognition technology, spoken dialogue interface is demanded as user-friendly interface for several task domain. Therefore, the needs of the way of easy construction of domain model are now increasing.
The question is how we can implement dialogue manager independent to particular task domain. Our approach to this question is to construct dialogue library for general task domain. The first step of constructing general library is to classify several domains according to the direction of information. As a consequence of the analysis, we have got three abstract tasks: slot-filling, database search and explanation. Next, we specified the type of subdialogue and the type of interaction in each abstract task by referring task-oriented dialogue corpus [2]. In the last step, we examine the portability of this library by constructing a dialogue manager for new domain.
This paper was constructed as follows. Section 2 explains each abstract task according to the direction of information. Section 3 describe the representation of our dialogue library and the application interface. Section 4 describes the evaluation of portability of our approach in the most simplest case. Conclusion and future works are given in Section 5.
