Missing Generalizations: A Supervised Machine Learning Approach to L2 Written Production

Recent years have witnessed a growing interest in usage-based models of language, which characterize linguistic knowledge in terms of emerging generalizations derived from experience with language via processes of similarity-based distributional analysis and analogical reasoning. Language learning then involves building the right generalizations, i.e. the recognition and recreation of the statistical regularities underlying the target language. Focusing on the domain of relativization, this study examines to what extent the generalizations of advanced second language learners pertaining to the usage of complex constructions differ from those of experts in written production. We approach this question through supervised machine learning employing as a primary modeling tool random forests with conditional inference trees as base learners.