Abstract
In this paper we present our approach towardsArabic Dialect identification which was part ofthe The Fourth Nuanced Arabic Dialect Identi-fication Shared Task (NADI 2023). We testedseveral techniques to identify Arabic dialects.We obtained the best result by fine-tuning thepre-trained MARBERTv2 model with a mod-ified training dataset. The training set wasexpanded by sorting tweets based on dialects,concatenating every two adjacent tweets, andadding them to the original dataset as newtweets. We achieved 82.87 on F1 score andwe were at the seventh position among 16 par-ticipants>