Introducing fluency measures to the elicited imitation task

Abstract

The elicited imitation (EI) task has been widely used as a measure of automatized L2 knowledge. However, the scoring of the task has relied exclusively on product-based measures (i.e., accuracy of L2 production), without considering any process-based indices of automatization, such as fluency. To fill this gap, our study develops a written version of the EI task and innovatively draws on keystroke logging techniques to introduce new measures of fluency in EI production. To test whether the addition of fluency measures improves task sensitivity, we examined the degree to which fluency and accuracy predicted L2 proficiency among 40 L1 Polish speakers of English, living in the UK (Mage = 31, 20–60). The participants were late learners of English at intermediate-to-advanced level (CEFR B1–C2) with varying lengths of residence (0.5–18 years). Their L2 proficiency was measured through self-evaluation according to CEFR scales and through test-evaluation by DIALANG English grammar and vocabulary tests. Their written production of English article and tense-aspect target structures was coded for grammatical accuracy, speed and pausing fluency, and consistency in fluency. Generalized Additive Modelling revealed a nonlinear interaction between speed and pausing fluency and grammatical accuracy as predictors of selfevaluated (but not test-evaluated) proficiency, which suggests that participants tended to be more accurate and fluent from low to average proficiency, after which their accuracy plateaued while fluency continued to improve. The results support the importance of assessing fluency to maintain the sensitivity of the EI task especially among advanced learners.