Why doesn't the conversational agent understand me?: A language analysis of children speech

Abstract

Conversation agents have shifted the way we communicate with ubiquitous services by enabling the use of natural language communication and the analysis of acoustic and linguistic language patterns. Speech skills of children are not yet fully developed; therefore, most conversational agents frequently misunderstand them. In this research, we examined if conversational agents can uncover instances of language discrimination in children. We developed Bolita, a conversational agent using a Google Home and a Sphero Robot to encourage children to practice how to tell a joke. The results of a two week study of the use of Bolita by 37 Mexican children showed a conversational agent is more likely to misunderstand children with speech skills below average. Our results indicate that they speak less, use fewer words, and need more time to answer when interacting with a conversational agent, which may explain the challenges with the conversational agent understanding them. We close discussing the potential of conversational agents to uncover digital markers in children with language differences and suggest ways that conversational agents could be built to be more inclusive.