Text-based dialogue structure and repair in human-computer and human-human task-oriented interaction

Abstract

There is currently very little research on the comparison of human-human (HHI) and human computer interaction (HCI) in the textual modality. The aim of this study is to give an overview of the behaviour of human interlocutors when they speak to artificial intelligence agents in a text chat, and how the agents react to certain conversational norms and anomalies. I hypothesise that HHI conversations feature shorter exchanges, longer opening and closing sequences, and that human participants in HHI are able to resolve errors much more rapidly than chatbots. To investigate these hypotheses, I compared conversations in which a human spoke with an information-providing conversational agent, and conversations in which a human spoke to another human in the same information-providing role, both chat-mediated. My findings indicated that humans are able to resolve queries within a lower amount of turns, and that they require conversations to be halted much less frequently. This study fills a gap in existing research and suggests promising directions in which future study can be conducted. It stresses the importance of taking a pragmatic perspective into account when creating a conversational agent, and emphasises that using parallel HHI data to inform such development is vital to further progress in natural language processing.