How children speak with their voice assistant Sila depends on what they think about her*

Children encounter artificial intelligence, e.g., voice assistants, at an increasingly younger age. However, we know far too little about children's communicative behavior with a voice-only interaction partner and whether and how their assumptions about the interaction partner predict how they communicate. To address these aspects, we ran a treasure hunt game with 50 5-6-year-olds who were guided through the hunt by a voice-only interaction partner Sila - acted by a human and connected via loudspeakers. Crucially, Sila was introduced either as a human or a voice assistant. We employed easy communication trials in which communication went smoothly and challenging communication trials that contained staged misunderstandings. Results showed that at the group level, children differed in the way they adjusted their speech across experimental phases - both in quantity (fewer grounding turns and fewer syllables in speech directed to voice assistants) and quality (reduced increase in f0 and intensity in speech directed to voice assistants). Notably, the communicative behavior was predicted by children's individual parasocial relationship with their voice-only interaction partner. We discuss the relevance of our findings for developmental science at the interface with communication science and phonetics, and the direct practical implications we can infer for speech technology.

Automated summary

Children's communicative behavior with a voice-only interaction partner differs based on their assumptions about the partner. The way children adjusted their speech varied in terms of quantity and quality when interacting with voice assistants. Individual parasocial relationships played a role in predicting children's communicative behavior.