Therapist Role-Reversals in an Autism Spectrum Pilot Study: Robot Malfunctions Prompt Enhanced Social Speech Performance

Background:  Recent studies (e.g., Kim et al. 2015) show increased social communicative behavior by children with Autism Spectrum Disorder (ASD) following human robot interaction (HRI). To test these results with a specific focus on social speech development, we designed and implemented a multi-month pilot study using HRI, hypothesizing that children with ASD would show improved linguistic and pragmatic communication skills through time. Tanaka and Matsuzoe (2013) find that using HRI in standard primary school settings enhances vocabulary, suggesting this may be due to role-reversal opportunities afforded by HRI, allowing children to serve as robot care-givers. This effect holds relevance for our findings. 

Objectives:  To use native NAO robot programming behaviors to elicit and observe social speech performance in three children with  ASD through a pilot study hosted by an equal care (ABA, Therapy directed learning) center, allowing participants to interact with the small humanoid robot  by playing semantic-domain card games, spontaneous dialogues and other routines.

Methods:  The study used a grounded-theory, mixed-methods approach, combining ethnographic observation, with first language acquisition metrics, cognitive linguistic analyses and speech pragmatic analyses. Three investigators were required to facilitate each field data collection session, including an interface technician and two participant observers. Session programming was standardized across participants. Ethnographic field notes were rendered into project reports and compared with WAV audio recordings of HRI sessions to produce detailed transcripts coded to CHILDES CHAT standards (MacWhinney 2000). At least two investigators coded each transcript for multiple layers analysis, including Mean Length of Utterance (MLU) analysis, structural discourse analysis, content analysis of discourse pragmatics (with a focus on implicature), and cognitive semantic analysis of conceptual blends and script-frame dynamics.

Results:  Relative levels of speech performance were operationalized after two visits to define varying degrees of proficiency per child: high [H], medium [M] and low-functioning [L], corresponding with preliminary MLU counts of 1.8[H], 1.0[M] and 0[L], further operationalized via successive layers of analysis. Modest net increases in MLU were apparent by the third visit (MLU=2.1[H] and 1.2[M]), and utterance counts increased substantially (n=82[H], n=69[M]). Speech pragmatic analyses indicated concrete instances of increased linguistic performance for social ends by participants H and M, including (1) increased usage of spontaneous vocatives, (2) increased verbal strategies to initiate conversation and (3) increased response rate to content questions. Linguistic strategies for the expression of concern, humor, empathy and care-giving also emerged following unplanned instances of robot malfunction, including unprompted assumption of overt role reversal by participant H, who offered a behavioral modification reward to the robot following technical correction of a software malfunction by investigators. Using insights from Cognitive Linguistics, such behavior is analyzed as a conceptual blend or interchange between Idealized Cognitive Models.  

Conclusions: Using a range of fresh linguistic evidence, including first language acquisition metrics, cognitive linguistic analyses and speech pragmatic analyses, our study confirms that children with ASD who interact with robots demonstrate increased social communicative behaviors.  We further find that robot malfunctions can give rise to care-giving role-reversals. Children assuming such roles improved in linguistic performance.