A context for error: using conversation analysis to represent and analyse recorded voice data

Abstract

Recorded voice data, such as from cockpit voice recorders (CVRs) or air traffic control tapes, can be an important source of evidence for accident investigation, as well as for human factors research. During accident investigations, the extent of analysis of these recordings depends on the nature and severity of the accident. However, most of the analysis has been based on subjective interpretation rather than the use of systematic methods, particularly when dealing with the analysis of crew interactions. This paper presents a methodology, called conversation analysis, which involves the detailed examination of interaction as it develops moment-to-moment between the participants, in context. Conversation analysis uses highly detailed and revealing transcriptions of recorded voice (or video) data that can allow deeper analyses of how people interact. The paper uses conversation analysis as a technique to examine CVR data from an accident flight. The focus accident was a controlled flight into terrain event involving an Israel Aircraft Industries Westwind 1124 jet aircraft, which impacted terrain near Alice Springs on 27 April 1995. The conversation analysis methodology provided a structured means for analysing the crew’s interaction. The error that contributed directly to the accident, an incorrectly set minimum descent altitude, can be seen as not the responsibility of one pilot, but at least in part as the outcome of the way the two pilots communicated with one another. The analysis considered the following aspects in particular: the significance of overlapping talk (when both pilots spoke at the same time); the copilot’s silence after talk from the pilot in command; instances when the pilot in command corrected (repaired) the copilot’s talk or conduct; and lastly, a range of aspects for how the two pilots communicated to perform routine tasks. In summary, the conversation analysis methodology showed how specific processes of interaction between crew members helped to create a working environment conducive to making, and not detecting, an error. By not interacting to work together as a team, pilots can create a context for error. When analysing recorded voice data, and especially for understanding instances of human error, often a great deal rests on investigators’ or analysts’ interpretations of what a pilot said, or what was meant by what was said, or how talk was understood, or how the mood in the cockpit or the pilots’ working relationship could best be described. Conversation analysis can be a tool for making such interpretations.