Even the most tightly scripted solo performances involve improvisation; the detailed execution of each note or word cannot be completely determined in advance. In joint performances the challenge of co­ordinating the actions of multiple people in real­time becomes even more complex. One response to this challenge has involved appeal to prediction using ‘forward models’ from computational models of action planning. These models involve automatic activation of motor representations of the future perceptual consequences of an unfolding action. Although normally associated with action production, if a person perceiving the action can also produce a forward model they can predict what word or note will come next. An important problem with this approach is that it is by definition conservative. It only works for familiar or rehearsed actions and cannot account for the production of novel or improvised responses. Using case studies from free jazz improvisation and conversation I will illustrate this problem for natural co­ordinated action. Rather than relying on access to pre­established shared representations, constructive engagement in these situations requires mechanisms that enable people to adapt and create new conventions on the fly i.e. improvise. I will argue that the key processes through which this is achieved are the interactional processes of ‘repair’ that we use to detect and deal with things that do not go as expected. These mechanisms are not auxiliary but rather provide the fundamental foundations on which all successful human interaction depends.
Mot(s) clés libre(s) : interaction linguistique, improvisation