Neural Arabic Text Diacritization: State of the Art Results and a Novel Approach for Machine Translation

Ali Fadel, Ibraheem Tuffaha, Bara' Al-Jawarneh, Mahmoud Al-Ayyoub
11/8/2019
cs.CLcs.LG

Official code: AliOsm/shakkelha (99 stars, MIT, last commit 2023-03-24). 1 more implementation below.

Abstract

In this work, we present several deep learning models for the automatic diacritization of Arabic text. Our models are built using two main approaches, viz. Feed-Forward Neural Network (FFNN) and Recurrent Neural Network (RNN), with several enhancements such as 100-hot encoding, embeddings, Conditional Random Field (CRF) and Block-Normalized Gradient (BNG). The models are tested on the only freely available benchmark dataset and the results show that our models are either better or on par with other models, which require language-dependent post-processing steps, unlike ours. Moreover, we show that diacritics in Arabic can be used to enhance the models of NLP tasks such as Machine Translation (MT) by proposing the Translation over Diacritization (ToD) approach.

View on arXivView PDF

Code Implementations(2)

AliOsm/shakkelhaOfficial100%
MIT

Neural Arabic text diacritization

9918Python, Jupyter NotebookFeb 26, 20193 years agoMIT
arabic-languagecomparisondatasetdiacritizationffnn+3 more
AliOsm/translation-over-diacritizationCommunity100%
MIT

Translation-over-Diacritization technique implementation

40Shell, PythonAug 17, 20193 years agoMIT
arabic-languagediacritizationmachine-translationnlprnn+1 more

Similar Papers

Arabic Text Diacritization Using Deep Neural Networks

86% match2 repos3552019

Fine-Tashkeel: Finetuning Byte-Level Models for Accurate Arabic Text Diacritization

85% match2023

Take the Hint: Improving Arabic Diacritization with Partially-Diacritized Text

83% match1 repo42023

Diacritic Recognition Performance in Arabic ASR

81% match2023

QARI-OCR: High-Fidelity Arabic Text Recognition through Multimodal Large Language Model Adaptation

79% match2025

Don't Touch My Diacritics

79% match2024

Cite this paper

@article{fadel2019neural,
  title  = {Neural Arabic Text Diacritization: State of the Art Results and a Novel Approach for Machine Translation},
  author = {Ali Fadel and Ibraheem Tuffaha and Bara' Al-Jawarneh and Mahmoud Al-Ayyoub},
  year   = {2019},
  eprint = {1911.03531},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1911.03531}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1911.03531.svg)](https://codeofpaper.com/paper/1911.03531)

Discussion