Dynamic and Goal-oriented Interaction for Multi-modal Service Agents

Form-based dialogue modeling schemes such as VoiceXML specification have been widely used in designs of dialogue management. However, for such schemes the dialogue goals have to be rigidly modeled as form-filling problems, while the data presentation needs to be statically defined in the forms beforehand. This not only limits the application scope of dialogue systems to form-filling tasks, but also constrains the versatility and variety for human computer interactions. In this paper, a dialogue management approach providing dynamic and goal-oriented interaction is proposed. This approach uses the event hierarchy to represent the problem-solving procedures, as in conventional plan-based schemes. However, instead of using the logic programming languages, it constructs the logical relationships among dialogue goals, conditions and objects with XML-tree structure, while using ECMA-script to perform procedural computation, which make this scheme superior in capabilities of object representation and procedural computation. The next action obtained from inferring on the XML-tree is mode-independent, which can be used to generate the presentation dynamically with multiple modes, including speech, text, GUI or expression of talking head. This proposed scheme has been applied to a multi-modal dialogue agent for personal information services providing goal-oriented interaction.