Automatic Adaptation of Online Language Lessons for Robot Tutoring

Teaching with robots is a developing field, wherein one major challenge is creating lesson plans to be taught by a robot. We introduce a novel strategy for generating lesson material, in which we draw upon an existing corpus of electronic lesson material and develop a mapping from the original material to the robot lesson, thereby greatly reducing the time and effort required to create robot lessons. We present a system, KubiLingo, in which we implement content mapping for language lessons. With permission, we use Duolingo as the source of our content. In a study with 24 users, we demonstrate that user performance improves by a statistically similar amount with a robot lesson as with Duolingo lesson. We find that KubiLingo is more distracting and less likeable than Duolingo, indicating the need for improvements to the robot’s design.