5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Emergent Computational Dialogue Management Architecture For Task-Oriented Spoken Dialogue Systems

Takeshi Kawabata

NTT Basic Research Laboratories, Japan

This paper proposes a new dialogue management architecture for human-machine speech communication systems. In our daily speech communication, incremental, non-deterministic and quick-response behaviors are required for effortless information interchange. Emergent computational architectures, proposed in the robot control domain, are promising to enable such features. The dialogue manager (ECL-DIALOG) consists of multiple "phrase pattern" detectors as input sensors. The CFG driven phrase detectors search for phrase patterns in user utterances and generates numerous emergent slot-filling signals. The system integrates them according to their "phrase pattern" priorities and updates the current task-completion context. When a slot value is updated, the system generates an appropriate response. For example, when the system finds a new slot value from user utterances, the system generates a chiming utterance "yeah". When the context slot is replaced by a different value "Tuesday" that has a lower priority, the system asks for confirmation "On Tuesday?".

Full Paper

Bibliographic reference.  Kawabata, Takeshi (1998): "Emergent computational dialogue management architecture for task-oriented spoken dialogue systems", In ICSLP-1998, paper 0143.