Comparison of child-human and child-computer interactions based on manual annotations.

ABSTRACTTechnological advancements in recent years have been accompanied by a notable increase in research related to conversational child-machine interfaces. The technology has many applications from entertainment to education. In order to integrate this technology successfully we, however, need to understand the key differences (if any exist) in how children interact with machines versus how they interact with humans. Such knowledge could inform the design of more child-appropriate interfaces as well as highlight any distinct characteristics of child-computer interactions that may be crucial for specific applications. In this paper, we analyze a subset of the Little CHIMP corpus, in which preschool aged children have a series of conversations with a human moderator and a Wizard-of-Oz controlled computer character. We first manually transcribed and annotated the data using an objective audio-visual behavior coding scheme. We next extracted features exemplifying language and social communication from these transcriptions and annotations and performed statistical hypothesis tests comparing the child-human and child-computer interactions. Finally, we discuss the differences between these two dyadic conversations.