An Analysis of Non-Task-Oriented Dialogs and a Computational Model of Generating Affective Utterances
良子 徳久 · NAIST Digital Library (Nara Institute of Science and Technology) · 2009
Previous research into human-computer interaction has mostly focused on taskoriented dialogs, where the goal is to achieve a given task as precisely and efficiently as possible by exchanging information required for the task through dialog.On the other hand, the necessity of non-task-oriented dialog systems has been raising in recent years.Specially, non-task-oriented dialog systems are needed for home robots.We aim at realization of the verbal communication with robots.This thesis takes up the problem of the non-task-oriented dialog by two steps: the first step is conversation analysis and the second step is its engineering implementation.In the conversation analysis step, the following two issues are investigated: a) What makes a non-task oriented human-to-human conversation to be an enthusiastic one; b) What is the difference between task-oriented and non-task-oriented dialogs.The first issue is explained by investigating what type of utterances contributes to enthusiasm in a non-task-oriented human-to-human dialog.For this end, we first create a non-taskoriented human-to-human dialog corpus.We then analyze the relationship between utterances and enthusiasm by studying the instances by studying the instances in the corpus.As a result, we found that "affective utterance" and "cooperative utterance" were related to enthusiasm.On the other hand, concerning the second issue, we investigate what type of utterances appear saliently in non-task-oriented dialog.We first create two types of human-to-human dialog corpora: a task-oriented dialog corpus and a non-task-oriented dialog corpus.We investigate what are the discriminating characteristics that differentiate them.We found that initiation/response utterance appeared