Abstract

This paper explores how callers formulate information enquiries for an artificial conversational agent in a call centre and compares it with the way enquiries are addressed to human operators of the same call centre. It includes 60 call recordings with human operators and 103 call recordings with the artificial conversational agent, transcribed and analysed using the method of Conversation Analysis. We show that people formulate and reformulate their enquiries differently to an artificial agent, even though the goal in both cases is to get an answer to the same enquiry. When talking to the artificial conversational agent, callers produce short enquiries, similar to web searches. When connected to human operators, callers formulate longer enquiries which include many details. By analysing the differences in the way callers formulate their enquiries to robots and human operators, we show what callers expect artificial conversational agents to process. These expectations affect the way the enquiry is formulated and, as a result, operators and artificial agents encounter different types of problems they have to repair to understand the question correctly and find an answer to it. Our findings have interesting implications for Human Computer Interaction both in terms of “robot-recipient design” and “user-recipient design”.