Eye Gaze Analysis in Human-Human Interactions

This thesis deals with gaze behaviour in human-human interactions. The aim of this work was to study and statistically analyse gaze direction during dialogues in order to suggest possible explanations about the role of gaze in social interactions. Previous research on gaze behaviour showed that visibility improved information transfer during human-human interaction. Recent works concentrated on the role of gaze during dialogues, and especially its link with turn-taking. In this thesis, eight task-oriented conversations were recorded using a camera and an eye-tracking device. The gaze data collected was first analysed with video tracks to evaluate gaze behaviour in general and the factors which have an influence on it. Furthermore, other analyses were carried on to examine the relationship between gaze and the type of utterances, and gaze and turn-taking. Then, the data collected with the eye-tracker were processed with a Matlab program and then compared to the video data to check the reliability of the eye-tracking system. This analysis work is a preliminary step to the enhancement of human-computer communication. These findings could be incorporated into the design of autonomous human-like conversational agents which would exhibit appropriate gaze behaviours during dialogues with human users.