CasANER: Arabic Named Entity Recognition Tool
Résumé
Actually, the Named Entity Recognition (NER) task is a very innovative research line involving the process of unstructured or semi-structured textual resources to identify the relevant NEs and classify them into predefined categories. Generally, NER task is based on the classification process, which always refers to the previous categorizations. In this context, we propose CasANER, which is a system recognizing and annotating the ANEs. The CasANER elaboration is based on a deep categorization made using a representative Arabic Wikipedia corpus. Moreover, our proposed system is composed of two kinds of transducer cascades, which are the analysis and synthesis transducers. The analysis cascade, which is dedicated to the ANE recognition process, includes the analysis, filtering and generic transduces. However, the synthesis cascade enables to transform the annotation of the recognized ANEs into an annotation respecting the TEI recommendation in order to provide a structured output. The implementation of CasANER is ensured by the linguistic platform Unitex. Then, its evaluation is made using measure values, which show that our proposed system outcomes are satisfactory. Besides, we compare CasANER system with a statistical system recognizing ANEs. The comparison phase proved that the results obtained by our system are as efficient as those of the statistical system in the recognition and annotation of the person’s names and organization names.