Even though the most important questions about the acquisition of (L2) sound systems concern change over time (like the effect of age of onset; acquisition orders, perception vs. production; implicational relationships), most research has investigated the outcomes of phonological development rather than the process itself. I will argue that only a longitudinal, process-based approach can truly inform us about the developmental process. In this paper I will report on two longitudinal case studies of English learners of Dutch.

Using variability analyses and nonlinear time series analyses, several phonetic correlates (Voice Onset Time, vowel formants) were analyzed over time in two tasks: a word naming task and in a shadowing task. Based on previous studies it could be expected that pronunciation accuracy is first approached in shadowing tasks and gradually spreads to more spontaneous production.

The data show variable developmental patterns in the two tasks, and support the idea that variability is not an irrelevant by-product of development, but can be seen as the motor of change. The data also show how process-based analyses will be able to reveal dynamic interactions that would remain concealed in product based approaches.