Modelling Second Language Performance: Integrating Complexity, Accuracy, Fluency, and Lexis

Complexity, accuracy, and fluency have proved useful measures of second language performance. The present article will re-examine these measures themselves, arguing that fluency needs to be rethought if it is to be measured effectively, and that the three general measures need to be supplemented by measures of lexical use. Building upon this discussion, generalizations are reviewed which focus on inter-relationships between the measures, especially between accuracy and complexity, since positive correlations between these two areas have been less common in the literature. Some examples of accuracy–complexity correlations are reviewed. The central issue here is how to account for these correlations, and so the discussion explores rival claims from the Cognition and Trade-off Hypotheses. It is argued that such joint raised performance between accuracy and complexity is not a function of task difficulty, as the Cognition Hypothesis would predict, but that instead it reflects the joint operation of separate task and task condition factors. Extending the theoretical discussion, connection is made with the Levelt model of first language speaking, and it is proposed that the results obtained in the task-based performance literature can be linked to this model, modified to take account of differences between first and second language processing, particularly as these stem from differences in the underlying mental lexicons.