An Efficient Multi-agent System Combining POS-Taggers for Arabic Texts

In this paper, we address the problem of Part-Of-Speech tagging of Arabic texts with vowel marks. After the description of the specificities of Arabic language and the induced difficulties on the task of POS-tagging, we propose an approach combining several methods. One of these methods, based on sentences patterns, is original and very attractive. We present, afterward, the multi-agent architecture that we adopted for the conception and the realization of our POS-tagging system. The multi-agent architecture is justified by the need for collaboration, parallelism and competition between the different agents. Finally, we expose the implementation and the evaluation of the system implemented.