From Visualisation to Hypothesis Construction for Second Language Acquisition

One research goal in Second Language Acquisition (SLA) is to formulate and test hypotheses about errors and the environments in which they are made, a process which often involves substantial effort; large amounts of data and computational visualisation techniques promise help here. In this paper we have defined a new task for finding contexts for errors that vary with the native language of the speaker that are potentially useful for SLA research. We propose four models for approaching this task, and find that one based only on error-feature cooccurrence and another based on determining maximum weight cliques in a feature association graph discover strongly distinguishing contexts, with an apparent trade-off between false positives and very specific contexts.