Linear coherence and relevance: Logic in computer-human ‘conversations’

Abstract This paper describes an examination of computer-human ‘conversations’ - transcripts of searches for information in a hypertext on violent crime and accompanying articulations - for actors' assessments of troubles. Two classes of troubles exist - those that result from violations of expectations of coherent linear flow and those that result from violations of relevance assumptions. These violations are inherent in these computer programs because of their essential linked-file organization, which means that actors commonly traverse links from one topic to another, many times in unanticipated ways, and because information category naming contains etceteras, which may not be mutually acknowledged. In the process of searching and thereby constructing text in this way, actors may start off on seemingly irrelevant paths, but still reach text segments that they are looking for. The error that occurs is thus a collaborative event, a response to disruption of actors' routine understandings of logic.