Arabic Text Diacritization In The Age Of Transfer Learning: Token Classification Is All You Need. (arXiv:2401.04848v1 [cs.CL])

Automatic diacritization of Arabic text involves adding diacritical marks (diacritics) to the text. This task poses a significant challenge with noteworthy implications for computational processing and comprehension. In this paper, we introduce PTCAD (Pre-FineTuned Token Classification for Arabic Diacritization, a novel two-phase approach for the Arabic Text Diacritization task. PTCAD comprises a pre-finetuning phase and a finetuning phase, treating Arabic Text Diacritization as a token classification task for pre-trained models. The effectiveness of PTCAD is demonstrated through evaluations on two benchmark datasets derived from the Tashkeela dataset, where it achieves state- of-the-art results, including a 20\% reduction in Word Error Rate

Read Full Article