Modelling Performance: Opening Pandora's Box

The widespread adoption of performance-based approaches to language testing has meant that debate about formal models of language performance in first and second language contexts is assuming a greater importance, despite a preference for relatively atheoretical approaches among some workers in this area This paper argues that it is necessary for researchers and developers in the area of language performance testing to have a clear understanding of the role of underlying performance capacities in second language performance, and critically evaluates the models of language ability proposed by Hymes, Canale and Swain, and Bachman in relation to this issue The analysis identifies a number of problematic features in the main models proposed in the second language assessment context. The paper concludes with suggestions for broadening the scope of the analysis embodied in such models, particularly in the direction of addressing more fully the interactional aspect of performance