Form, frequency, markedness and strategies in second language performance modelling

Abstract

Recent research has brought to light a range of factors which underly native language performance phenomena, including formal similarity or difference, linearity and proximity factors, frequency judgements, the availability of unmarked forms, and general cognitive strategies. On the basis of a substantial (70,000 word) machine-readable corpus of second-language written productions by anglophone learners of French, it is shown that such factors can make a substantial contribution to the modelling of second-language learner errors. Detailed examples are discussed in the areas of spelling errors, gender assignment, and gender and number agreement. The tendencies isolated on the basis of the corpus data are then modelled using a natural-language generation environment.