Used KENLM on the training data provided by the organizers of the MADAR shared task 2019 (https://sites.google.com/view/madar-shared-task/home?pli=1) to train different language models.
@inproceedings{mishra-mujadia-2019-arabic, title = "{A}rabic Dialect Identification for Travel and {T}witter Text", author = "Mishra, Pruthwik and Mujadia, Vandan", booktitle = "Proceedings of the Fourth Arabic Natural Language Processing Workshop", month = aug, year = "2019", address = "Florence, Italy", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/W19-4628", doi = "10.18653/v1/W19-4628", pages = "234--238" }