| Popis: |
This paper describes an examination of the HuComTech Corpus of multimodal interactions from the viewpoint of an automated dialogue system. Specifically, it looks at the difference between formal and informal interactions from the point of view of image processing. We show that an autonomous dialogue system is able to make inferences about the engagement states of its interlocutor in order to function efficiently without complete dialogue understanding. The paper shows that a very simple image processing algorithm is capable of distinguishing engagement states with a high degree of reliability. We can infer from the results that future autonomous agents might use a similar method to estimate the degree of engagement a human interlocutor has in their conversations. |
