The complexities of task-based language performance assessment (TBLPA) are leading language testers to reconsider many of the fundamental issues about what we want to assess, how we go about it and what sorts of evidence we need to provide in order to justify the ways in which we use our assessments. One claim of TBLPA is that such assessments can be used to make predictions about performance on future language use tasks outside the test itself. I argue that there are several problems with supporting such predictions. These problems are related to task selection, generalizability and extrapolation. Because of the complexity and diversity of tasks in most ‘real-life’ domains, the evidence of content relevance and representativeness that is required to support the use of test scores for prediction is extremely difficult to provide. A more general problem is the way in which difficulty is conceptualized, both in the way tasks are described and in current measurement models. The conceptualization of ‘difficulty features’ confounds task characteristics with test-takers’ language ability and introduces a hypothetical ‘difficulty’ factor as a determinant of test performance. In current measurement models, ‘difficulty’ is essentially an artifact of test performance, and not a characteristic of assessment tasks themselves. Because of these problems, current approaches to using task characteristics alone to predict difficulty are unlikely to yield consistent or meaningful results. As a way forward, a number of suggestions are provided for both language testing research and practice.
Stored only when both papers are in the corpus — 318 citations exist upstream.