A Robotic Therapist For Positive, Affective Prosody in High-Functioning Autistic Children

Background:   Deficits in prosody have been considered an identifying characteristic of ASD since Kanner (1943).  Prosodic abilities have been shown to be related to attributions of social competence (Paul et al., 2005) and to success in vocational placements (Shriberg and Widder, 1990). Prosody has been found to remain impaired in speakers with ASD, even when other aspects of language improve (Kanner, 1971; DeMyer et al., 1973). The known preference of children with ASD for mechanical objects (Baron-Cohen and Belmonte, 2005) suggests that intervention with a robotic agent may increase efficacy in prosodic therapy.

Objectives:

1. To assess the accuracy of a robotic agent in classifying child prosody.

2. To compare the effectiveness of prosodic therapy delivered by a human therapist and by a robotic agent, measured by change in blind listeners’ pre- and post-therapy judgments of participants’ positive prosodic production.

Methods:   Six speakers with ASD (ages 7-12), demonstrating deficits in positive prosody production, are randomly assigned to one of two treatment conditions: script-based (McClannahan and Krantz, 2005) training provided by a speech-language pathologist or provided by a robotic agent. The robotic agent has an expressive face, “speaks” prerecorded scripted conversations, and utilizes a signal processor that classifies child prosodic responses as inappropriate or appropriately positive. Both human and robotic therapists provide reinforcement only to appropriate prosody in these scripted conversational contexts.

Results:   Preliminary data demonstrate that the robotic therapist is capable of correctly classifying over 70% of utterances produced by children in this age range as positive or not. Pilot testing of the robotic therapist revealed that participants readily engage with the robot, and that participants significantly modified their prosodic production during therapy.

Conclusions:   Preliminary data support the hypothesis of the efficacy of a robotic agent for prosody intervention. Results of the randomized controlled trial will be presented.