Enhancing Deep Learning with Embedded Features for Arabic Named Entity Recognition

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4904–4912 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4904 Enhancing Deep Learning with Embedded Features for Arabic Named Entity Recognition Ali Lotfy1, Caroline Sabty2, Slim Abdennadher2 1German University in Cairo, 2German International University 1El Tagamoa El Khames, New Cairo, Cairo, Egypt, 2Administrative Capital, Regional Ring Rd, Cairo, Egypt ali.hatab@student.guc.edu.eg, {caroline.sabty, slim.abdennadher}@giu-uni.de Abstract The introduction of word embedding models has remarkably changed many Natural Language Processing tasks. Word embeddings can automatically capture the semantics of words and other hidden features. Nonetheless, the Arabic language is highly complex, which results in the loss of important information. This paper uses Madamira, an external knowledge source, to generate additional word features. We evaluate the utility of adding these features to conventional word and character embeddings to perform the Named Entity Recognition (NER) task on Modern Standard Arabic (MSA). Our NER model is implemented using Bidirectional Long Short Term Memory and Conditional Random Fields (BiLSTM-CRF). We add morphological and syntactical features to different word embeddings to train the model. The added features improve the performance by different values depending on the used embedding model. The best performance is achieved by using Bert embeddings. Moreover, our best model outperforms the previous systems to the best of our knowledge. Keywords: Arabic Natural Language Processing, Named Entity Recognition, Deep learning 1. Introduction The Arabic language is spoken in a large area, includ- ing North Africa and the Arab Peninsula. Although there is no agreed-on statistic estimating the number of speakers of each language, Arabic always comes among the top 10 spoken languages. Arabic is also one of the oldest living languages (Al-Jallad, 2017), which adds to the importance of its processing. The processing of the Arabic language faces many challenges that arise from its unique nature. For in- stance, an Arabic word is often connected to one or many suffixes and/or one, or many prefixes (Awad et al., 2018a). In addition, some words are changed based on their position in a sentence. Due to this complex morphological system, the number of unique words in a document becomes relatively big. Every letter in Ara- bic can have one of four different states (the four basic diacritics). Another diacritic (Shadda) is also used to indicate two letters being written as one letter. Differ- ent diacritics on the same sequence of characters pro- duce other words (Sabty et al., 2018). These diacritics are not written in most Arabic texts, as Arabic speak- ers can recognize them easily from context. Moreover, Arabic is a relatively low-resource language. The named entity recognition (NER) task is one of the classical tasks of Natural Language Processing. There are particular challenges for the NER task, shared be- tween all languages. They include naming organiza- tions and locations after famous person names and the nested named entities. Due to the complex morphol- ogy of the Arabic language, the Arabic NER task faces more challenges than other languages, like …

Content truncated to 500 words