A Multi-Agent System for POS-Tagging Vocalized Arabic Texts

Chiraz Ben Othmane Zribi, Aroua Torjmen, and Mohamed Ben Ahmed

RIADI laboratory, University of La Manouba, Tunisia

Abstract: In this paper, we address the problem of Part-Of-Speech(POS) 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 that combines several methods (stochastic and rule-based). For the implementation of these methods and the global POS-tagging system, we adopted a multi-agent architecture. In which, five tagger agents work in parallel, each one applies its own method, in order to propose for each word in a sentence the suitable tag among those proposed by the morphological analyzer. The tagger agents cooperate together and with the unknown words solver agent to resolve unknown words. A voting agent decides in the end, which tag to affect to each word. Finally, we present the experimental protocol we used to evaluate the system carried out in this work and the obtained results that we consider very satisfactory.

Keywords: Natural Language Processing (NLP), Arabic language, morphological analyzer, part-of-speech tagging, hybrid methods, multi-agent system.

Received February 3, 2006; accepted April 22, 2006