Cultural Social Signal Interplay with an Expressive Robot

Abstract

Social robots are being developed as a form of social skills training for individual's with an autism spectrum condition (ASC). Effective training will therefore require the social signals produced by a robot to be contingent with people's knowledge and expectations of social cognition and behaviour. Designing recognisable facial expressions is an important part of this challenge; ensuring interactions are more believable and motivating. This design process requires - amongst other factors - consideration of how culture and native language affects social signal processing. In this experiment participants offered a full-bodied robot (named 'Alyx') food items to which Alyx reacted autonomously, producing either an approving or disapproving expression. Participant's responded to these expressions (i.e. the robots social signals) by indicating whether Alyx liked or disliked the food. Task performance was examined both quantitatively (response time and accuracy) and qualitatively (participant's reactionary expressions). The results revealed significant cultural differences, as non-native English speakers were less accurate at interpreting expressions, but also a similar response trend between these groups. Qualitative analysis supported the notion that Alyx's expressions were not universally understood. These findings are discussed in the context of social skills training.