Automatic proficiency classification in L2 Portuguese

Abstract

We present the first experiments on automatic proficiency classification for L2 Portuguese. For the experiments, we take advantage of a new version of the NLI-PT dataset, a compilation of L2 Portuguese texts written by learners. We use supervised learning and we approach the task as a classification problem, using the CEFR scale. Different linguistic features are tested, combined with different algorithms. With the best model, we get an accuracy of 72%, a result in line with previous experiments with other languages.