Passing the Pact: Comparing Common Ground Building and Audience Design In Multiparty Human and Human-Agent Collaborative Communication

Abstract

Common ground building and audience design are critical for successful collaborative communication, yet little is known about how these operate during multiparty conversations involving speech agents. Through observing the creation and use of conceptual pacts within a tangram-based referential communication task, we compare how people take varying levels of interlocutor knowledge into account when in multiparty human-human and human-agent collaborative dialogues. Our findings show that, although participants used previously negotiated conceptual pacts similarly with less knowledgeable human and agent partners, people were more sensitive to varying knowledge levels in human-human rather than human-agent multiparty dialogue, elaborating more when using these pacts. The implications of these findings are discussed regarding the use and maintenance of conceptual pacts, common ground building and audience design within human-agent multiparty dialogues.