Parallel HMM-Based Approach for Arabic Part of Speech Tagging

In this paper we try to go beyond the classical use of the Hidden Markov Model for Part Of Speech Tagging, particularly for the Arabic language. In fact, most available Arabic tagging systems and tagsets are derived from English and do not make use of the linguistic richness of Arabic. Our new proposed tagging system will consist of two Hidden Markov Models working in parallel: In addition to the main model, a second model is added to serve as a reference for low probabilities tags. Of course, a dual corpus is required to train both models. To do so, we restructure the Nemlar Arabic corpus and extract a new tagset from diacritics and grammatical rules. The approach is implemented by using Java programming environment and several experimentations are conducted to evaluate it. The results of this approach, which are promising, as well as its limitations, are deeply discussed and future possible enhancements are also highlighted. This work will open the door for new promising research perspectives, particularly for the Arabic language processing, and more generally for the applications of Hidden Markov Models.