Keyword Detection in Human-Robot Tutoring Scenarios

We describe a way of narrowing the search space for descriptive keywords during a human-robot tutoring scenario, where the tutor is explaining names and characteristics of objects to the robot, by employing interaction detection techniques. This system detects attention getting behaviour which is derived from mother-infant interactions and extracts the verbal information during these specific time periods, segmenting it and building up histograms to estimate word frequencies and thus word importance. This method should allow us to create a system that does not rely on a dictionary or normal speech recognition to acquire novel word-object relations but only relies on the pure interaction between the robot and a human tutor.