Domain and subtask-adaptive conversational agents to provide an enhanced human-agent interaction

Abstract

One of the most demanding tasks when developing conversational agents consists of designing the dialog manager, which decides the next system response considering the user's actions and the dialog history. A previously developed statistical dialog management technique is adapted in this work to reduce the effort and time required to design the dialog manager. This technique allows not only an easy adaptation to new domains, but also to deal with the different subtasks by means of specific dialog models adapted to each dialog objective in the domain of a multiagent system. The practical application of the proposed technique to develop a conversational agent providing railway information shows that the use of these specific dialog models increases the quality and number of successful interactions with the agent in comparison with developing a single dialog model for the complete domain.