In this paper, we investigate the relationship between tutors' gaze behavior and particular kinds of linguistic behaviors. In particular, we describe how word classes are distributed over different gazing classes. For this, we collected data from human-robot interactions and used a classification of our participants' gazing behavior to create subsets of their utterances. The participants' speech was transcribed for 3 sessions (2 min each) of interaction with the robot and classified based on the detected gazing classes (looking at the robot, looking at the object or looking somewhere else). The analysis shows that there are, for instance, more object related keywords when people are gazing at an object, and more personal pronouns when people are looking at the robot. Understanding the relationship between human tutors' linguistic and gazing behavior can facilitate bootstrapping the one capability from the other, such that gazing behavior is exploited for narrowing down the search space for object-relevant linguistic material.
|Status||Udgivet - 2013|
|Begivenhed||Human-Robot Interaction Conference - Tokyo, Japan|
Varighed: 3. mar. 2013 → 7. mar. 2013
|Konference||Human-Robot Interaction Conference|
|Periode||03/03/2013 → 07/03/2013|