Abstract
This study investigates the integration of native language identification into authorship attribution, a previously unexplored aspect that is particularly important in multilingual contexts. We introduce AA-NLI50, a new dataset containing both native language and authorship information. We propose a novel chain-of-thought approach for native language identification. Our findings demonstrate that our system significantly enhances authorship attribution performance, with results showing a mean accuracy improvement of 9% over baseline methods.
Uluslu, Ahmet Yavuz