Adapting MARBERT for Improved Arabic Dialect Identification: Submission to the NADI 2021 Shared Task

In this paper, we tackle the Nuanced Arabic Dialect Identification (NADI)\nshared task (Abdul-Mageed et al., 2021) and demonstrate state-of-the-art\nresults on all of its four subtasks. Tasks are to identify the geographic\norigin of short Dialectal (DA) and Modern Standard Arabic (MSA) utterances at\nthe levels of both country and province. Our final model is an ensemble of\nvariants built on top of MARBERT that achieves an F1-score of 34.03% for DA at\nthe country-level development set -- an improvement of 7.63% from previous\nwork.\n