Department of Applied Linguistics Seminar Series: Pauline Foster
The Department of Applied Linguistics and Communication invites you to our Seminar Series. Our first speaker of the academic year is Pauline Foster, St Mary's University.
Measuring accuracy in second language task performance: the case for a weighted clause ratio.
As increasing numbers of research papers address the issue of accuracy in second language performance, it is timely to explore the measures which are commonly employed. These are of two kinds: local measures which track particular grammatical features, and global measures which assess the overall accuracy of performance.
Both have drawbacks. With local measures there is the issue of which grammatical features will best give a proper reflection of underlying competence. How are they to be chosen, what consideration should be given to the influence of first language transfer or interference, what confidence can we have that a chosen feature will occur often enough in a second language performance to make measurement worthwhile? For global measures to be calculated, language data needs to be segmented into units. These units themselves need to be reliable and valid, and identified accurately. Longer units are likely to disadvantage learners who are attempting a more complex L2 performance because the likelihood of error increases with the length of the unit. Moreover, within a single unit, one minor error carries the same weight as repeated major errors.
While it is concluded that, on balance, global errors provide a more reliable evaluation of accuracy in L2 performance, this paper will discuss the need for a more finely tuned global measure which could classify errors at different levels, i.e. those that seriously impede communication, those that impair communication to some degree, and those that do not impair communication at all. The paper will discuss the problem of reliably identifying these levels and will present analysed samples from written and spoken L2 data. The measure proposed here is one that addresses the problem of gravity of error, and should deliver to researchers a reliable way of comparing accuracy levels for within-subject and between-subject designs, longitudinally and cross-sectionally in oral and written data.