Construction of a Dialog System for Talking Using a Topic Maps-Based Online Learning System

抄録

This study attempts to construct a dialog system to retrieve information from an online learning system. It uses a humanoid robot, NAO, as an interface to convey information.

The learning system was built based on a Topic Maps ontology (ISO/IEC 13250:2002). The dialog model was constructed using the development environment of Choregraphe 2.1.4. The utterance-response interaction between the human and the robot was constructed based on a three-step model comprising (1) human utterance, (2) information retrieval and talking, and (3) updating the list of words recognizable by the robot.

To reduce the misrecognition of the human utterance and allow the theme to be changed flexibly, we introduced a focus interest model that restricted the list of words that could be recognized within a focus of interest. In this model, the theme of the dialog was defined by a topic type, and the members of the theme were determined as its instance topics, which include the focus topic. The list of recognizable words, which is the theme of the dialog, is updated by a transition in the focus topic or the theme.

These two models enabled an efficient dialog for retrieving information from the online learning system. However, the fluency of the dialog depends on how well the user knows the topic map ontology.