Efficacy of Arabic named-entity recognition

Abstract

Named entry recognition research is a relatively new field for the Arabic language, although it has reached a mature stage for other languages. As Arabic has more speech sounds than many other languages, there is some lack of uniformity in Arabic writing styles. Transcription can become ambiguous, and the same word can be written in several different ways. Spelling mistakes can arise as a result of this same phenomenon. There are also both long and short vowels in Arabic, which can lead to further ambiguity. In the Arabic world, NER research has typically been of limited capacity or coverage. With this in mind, in this paper, we propose a method for analysing the structure of Arabic named-entity recognition and sentence object recognition by combining prior information and conditional random fields. We present a proposed method that leads to a 2.67% performance improvement per sentence, as compared with existing methods.