Abstract
Large pre-trained language models, such as BERT, have recently achieved state-of-the-art performance in different natural language processing tasks. However, BERT based models in Arabic language are less abundant than in other languages. This paper aims to design a grammatical tagging system for texts in Arabic language using BERT. The main goal is to label an input sentence with the most likely sequence of tags at the output. We also build a large corpus by combining the available corpora such as the Arabic WordNet and the Quranic Arabic Corpus. The accuracy of the developed system reached 91.69%. Our source code and corpus are available at GitHub upon request.