The Design of a Multi-domain Chinese Dialogue System

For a multi-domain spoken dialogue system, there are two major difficulties in building its dialogue management. One is to interpret users' interested domain correctly and switch between domains smoothly. The other is the high cost of merging the dialogue management of different application domains. To solve these problems, an implicit domain identification and switching mechanism for dialogue management is proposed to make the process of domain identifying and context switching transparent to users. To merge the dialogue management of different application domains efficiently, a merging process applying the goal-oriented table-driven dialogue management model is also proposed in this paper. By integrating two existing single-domain spoken dialogue systems, a multi-domain spoken dialogue system is set up to prove the usability of proposed solutions. Tested on two corpora collected by both single-domain systems, the multi-domain system contributes 5.47% and 2.28% errors comparing with the responses of two single-domain systems respective.