A Novel Deep-learning based Approach for Automatic Diacritization of Arabic Poems using Sequence-to-Sequence Model

Abstract

Over the last 10 years, Arabic language have attracted researchers in the area of Natural Language Processing (NLP). A lot of research papers suddenly emerged in which the main work was the processing of Arabic language and its dialects too. Arabic language processing has been given a special name ANLP (Arabic Natural Language Processing). A lot of ANLP work can be found in literature including almost all NLP applications. Many researchers have been attracted also to Arabic linguistic knowledge. The work expands from Basic Language Analysis to Semantic Level Analysis. But Arabic text semantic analysis cannot be held without considering diacritization, which can greatly affect the meaning. Many Arabic texts are written without diacritization, and Diacritizing them manually is a very tiresome process that may need an expert. Automatic diacritization systems became a demand as an initial step for processing Arabic text for any Arabic Language Processing application as Arabic diacritization is very important to get a readable and understandable Arabic text. For this reason, many researchers recently worked on building systems and tools that automatically diacritize un-diacritized Arabic texts. This work presents a novel deep learning-based sequence-to-sequence model to diacritize un-diacritized Arabic poems. The proposed model was tested and achieved high diacritization accuracy rate.