Arabic Named Entity Recognition Process using Transducer Cascade and Arabic Wikipedia

Proceedings of Recent Advances in Natural Language Processing, pages 48–54, Hissar, Bulgaria, Sep 7–9 2015. Arabic Named Entity Recognition Process using Transducer Cascade and Arabic Wikipedia Fatma Ben Mesmia University of Sfax, Laboratory MIRACL, Multimedia, In- foRmation Systems and Advanced Computing Laboratory, fatmabm@ymail.com Nathalie Friburger University François Rabelais of Tours, LI, Computer laboratory, nathalie.friburger@univ- tours.fr Kais Haddar University of Sfax, Laboratory MIRACL, Multimedia, In- foRmation Systems and Advanced Computing Laboratory, Kais.Haddar@fss.rnu.tn Denis Maurel University François Rabelais of Tours, LI, Computer laboratory, denis.maurel@univ-tours.fr Abstract Transducers namely transducer cascades are used in several NLP-applications such as Ara- bic named entity recognition (ANER). To ex- periment and evaluate an ANER process, a weight coverage corpus is necessary. In this pa- per, we propose an ANER method based on transducer cascade. The proposed transducer cascade is generated with the CasSys tool inte- grated in Unitex linguistic platform. The exper- imentation of our method is done on a Wikipe- dia corpus. The Wikipedia text format is ob- tained with Kiwix tool. The experiment results are satisfactory based on calculated measures. Keywords: Cascade of transducers, Wikipe- dia, Arabic named entities, Unitex, CasSys 1. Introduction Transducers can play an important role in the In- formation Extraction (IE) namely in the Named Entity Recognition (NER). At the same time, transducers can extract and classify the Arabic Named Entity (ANE). Generally, the use of trans- ducers is realized in well defined succession that is called cascade (Friburger and Maurel, 2004). In fact, the identification of necessary transduc- ers is not an easy task because several linguistic phenomenacan interact (Shaalan, 2014; Ben Mes- mia and al, 2015). The free resource Wikipedia is an important in- formation source. Indeed, several text processing applications based on transducer cascade can ben- efit from Wikipedia articles. Therefore, names of people, which are part of proper nouns, appear fre- quently in the Arabic Wikipedia. More efforts by NLP-researchers are concentrated on this type. Person names are considered as the most chal- lenging task for Arabic. In this context, our objective is to propose, us- ing the rule-based approach, a transducer cascade for the recognition of personality’s names. In this approach, we benefit from the robustness of trans- ducers and exploit the free resource, Wikipedia. The recognition requires the identification of dic- tionaries, a list of trigger words and extraction rules allowing the development of a set of trans- ducers acting on the corpus with a certain logic. The present paper is composed of six sections. The second section presents previous work de- scribing the developed systems for the recogniz- ing of the personality names. The third section is dedicated to describing the categorization of per- son’s names. The fourth section devoted to detail the proposed method that is implemented by using CasSys system. The experiment is presented and evaluated in section five. Finally, we give a con- clusion and some perspectives. 2. Previous Work There are several work treating the ANER based on several approaches among which we cite the work of (Shaalan and Raza, …

Content truncated to 500 words