Cross-corpus Native Language Identification via Statistical Embedding

In this paper, we approach the task of na- tive language identification in a realistic cross- corpus scenario where a model is trained with available data and has to predict the native lan- guage from data of a different corpus. We have proposed a statistical embedding representa- tion reporting a significant improvement over common single-layer approaches of the state of the art, identifying Chinese, Arabic, and In- donesian in a cross-corpus scenario. The pro- posed approach was shown to be competitive even when the data is scarce and imbalanced.