Proceedings of the Annual Meeting of the Cognitive Science Society

Machine learning can identify, with reasonable accuracy, the native language of someone writing in a foreign language(Joel Tetreault et al., 2013). Intriguingly, native language identification (NLI) can be accomplished looking only at thesyntactic structure, ignoring word choice (Swanson, 2013). This finding has potentially broad relevance to cognitivescience since it suggests a broad-based method to empirically study the effects of first language syntax on second language(L1-¿L2 transfer). However, that requires interpretation of the resulting models, which is notoriously difficult (Williamset al., 2017). As a first step, we compare the results of a variety of state-of-the-art machine learning techniques on NLI intwo languages: English and Spanish.