Arabic Diacritic Recovery Using a Feature-rich biLSTM Model

Diacritics (short vowels) are typically omitted when writing Arabic text, and readers have to reintroduce them correctly pronounce words. There two types of diacritics: The first core-word diacritics (CW), which specify the lexical selection, second case endings (CE), appear at end word stems generally their syntactic roles. Recovering CEs is relatively harder than recovering due inter-word dependencies, often distant. In this article, we use feature-rich recurrent neural network model that a variety linguistic surface-level features recover both core endings. Our surpasses all previous state-of-the-art systems with CW error rate (CWER) 2.9% CE (CEER) 3.7% for Modern Standard (MSA) CWER 2.2% CEER 2.5% Classical (CA). When combining diacritized cores endings, resultant rates 6.0% 4.3% MSA CA, respectively. This highlights effectiveness feature engineering such deep models.