Part-of-Speech Tagging for Arabic Gulf Dialect Using Bi-LSTM

Part-of-Speech Tagging for Arabic Gulf Dialect Using Bi-LSTM Randah Alharbi12, Walid Magdy1, Kareem Darwish3, Ahmed AbdelAli3, Hamdy Mubarak3 1School of Informatics, University of Edinburgh 2College of Computer Science & Information Systems, Umm Al-Qura University 3QCRI, Hamad bin Khalifa University raharbi@uqu.edu.sa, wmagdy@inf.ed.ac.uk {kdarwish,hmubarak,aabdelali}@hbku.edu.qa Abstract Part-of-speech (POS) tagging is one of the most important addressed areas in the natural language processing (NLP). There are effective POS taggers for many languages including Arabic. However, POS research for Arabic focused mainly on Modern Standard Arabic (MSA), while less attention was directed towards Dialect Arabic (DA). MSA is the formal variant which is mainly found in news and formal text books, while DA is the informal spoken Arabic that varies among different regions in the Arab world. DA is heavily used online due to the large spread of social media, which increased research directions towards building NLP tools for DA. Most research on DA focuses on Egyptian and Levantine, while much less attention is given to the Gulf dialect. In this paper, we present a more effective POS tagger for the Arabic Gulf dialect than currently available Arabic POS taggers. Our work includes preparing a POS tagging dataset, engineering multiple sets of features, and applying two machine learning methods, namely Support Vector Machine (SVM) classifier and bi-directional Long Short Term Memory (Bi-LSTM) for sequence modeling. We have improved POS tagging for Gulf dialect from 75% accuracy using a state-of-the-art MSA POS tagger to over 91% accuracy using a Bi-LSTM labeler. Keywords: Part-of-Speech (POS), Gulf Arabic (GA), Dialectal Arabic (DA), Bidirectional Long Short Term Memory (Bi-LSTM) 1. Introduction Part-of-speech (POS) tagging is one of the main building blocks in many Natural language processing (NLP) appli- cations (Jurafsky and Martin, 2009). POS tagging of Ara- bic is challenging due to its highly inflectional nature. Ara- bic language has two variants, namely: Modern Standard Arabic (MSA), the formal language that used in news and official business, and Dialectal Arabic (DA), the more in- formal version used in everyday life. Although they share some common characteristics, they differ at many linguistic levels (Katz and Diab, 2011). Most state of the art POS tag- gers for Arabic are designed and trained for MSA. Though the accuracy of MSA taggers is high (around 96%), these taggers fail to achieve high scores for DA (Pasha et al., 2014). For example, as we show in this work, a state-of- the-art Farasa (Abdelali et al., 2016) MSA tagger achieves only 75% accuracy on Gulf Arabic(GA) dialect. With the wide spread of social media websites and chatting applica- tions, DA became widely used (Diab et al., 2010). There is a need to develop NLP tools and applications for them. Hence the need for designing a specific tool for POS tag- ging for DA is of utmost importance. This paper aims at developing a POS tagger for one of the most widely used dialects, namely Gulf Arabic (GA). This work answers the question ”how much gain in accu- racy can be achieved by designing a dedicated …

Content truncated to 500 words