Social competence in conversational retrieval

Conversational IR systems — here, meaning systems that work over multiple turns of natural language — are rapidly getting better at the basic functions of understanding searcher utterances, tracking context, and finding and presenting relevant information. However, conversation is much more than an exchange of facts: humans start learning conversational norms in utero, and unlike web UIs, conversational exchanges are laden with social signals. As humans, we cannot help but treat computers as social agents (Nass et al., 1994), so as developers and researchers we must be aware of social conventions.