Native Language Identification on L2 Portuguese

摘要

This study advances on Native Language Identification (NLI) for L2 Portuguese. We use texts from the NLI-PT dataset corresponding to five native languages: Chinese, English, German, Italian, and Spanish. We include the same L1s as in previous works, and more texts per language. We investigate the impact of different lexical representations, the use of syntactic dependencies and the performance of diverse classification methods. Our best model achieves an accuracy of 0.66 including lexical features, and of 0.61 excluding them. Both results improve previous works on NLI for L2 Portuguese.