Transliteration of Judeo-Arabic Texts into Arabic Script Using Recurrent Neural Networks

What it is about

Judeo-Arabic texts are written in Hebrew letters, and converting them to Arabic script is mostly letter-for-letter, but the letter mapping is ambiguous (alef, yod, hamza, alif maqsura, missing shadda and tanwin). The authors train a four-layer bidirectional GRU with CTC loss, using letter doubling to handle outputs longer than inputs, and a pretraining stage on single letters. Their only parallel source is Judah Halevi's Kuzari (47,083 aligned word pairs after edit-distance alignment). A simple letter-map baseline gives a 9.5% label error rate; the best model gives 2% on the Kuzari test split, and 3.24% on Sa'adiah's Beliefs and Opinions. Synthetic pseudo-Judeo-Arabic data from other medieval Arabic texts hurt accuracy; 15% input dropout helped the unseen text (3.14%). Shuffling test words raised the error by about 0.5%, which shows some context use.

Ideas

• I1. Make the baseline an explicit, inspectable table: each Hebrew letter with its ambiguity list (alef to alif, hamza-alif or nothing; yod to ya or alif maqsura). Write it as a finite-state mapping with context rules, so every chosen Arabic letter carries its reason.
• I2. Use the word-shuffle control as a measure for any context-based rule: shuffle words, re-run, and report how much the output depends on context. Mova's own world decisions could use the same check.
• I3. The paper's warning about synthetic data (pseudo-transliterated text from another era lowered accuracy) is a rule for us: add out-of-genre data to training only after a held-out check on a different text.
• I4. Keep the top-5 beam candidates and let a word-level lookup choose among them, which the authors suggest; a dictionary check is deterministic and can say which candidate it kept.

For Mova

No chapter covers Judeo-Arabic yet. I1 would need a new note (for example languages/ar-x-judeo/, which the operator would have to approve); the rest is a method for any transliteration tool, including tools/ocrfix. The neural model itself is not for Mova, but its error analysis and the ambiguity list are reusable.

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{terner2020transliteration,
  title = {{Transliteration of Judeo-Arabic Texts into Arabic Script Using Recurrent Neural Networks}},
  author = {Terner, Ori and Bar, Kfir and Dershowitz, Nachum},
  booktitle = {Proceedings of the Fifth Arabic Natural Language Processing Workshop},
  year = {2020},
  url = {https://aclanthology.org/2020.wanlp-1.8/}
}

References

19 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Nasreen AbdulJaleel and Leah Larkey. 2002. English to Arabic transliteration for information retrieval: A statistical approach. Technical Report IR-261, Center for Intelligent Information Retrieval Computer Science, University of Massachusetts. find
2. Mohamed Al-Badrashiny, Ramy Eskander, Nizar Habash, and Owen Rambow. 2014. Automatic transliteration of romanized dialectal Arabic. In Proceedings of the Eighteenth Conference on Computational Natural Language Learning (CoNLL-2014), pages 30–38, June. find
3. Mansur Arbabi, Scott M. Fischthal, Vincent C. Cheng, and Elizabeth Bar. 1994. Algorithms for Arabic name transliteration. IBM Journal of Research and Development, 38(2):183–193. find
4. Kfir Bar, Nachum Dershowitz, Lior Wolf, Yackov Lubarsky, and Yaacov Choueka. 2015. Processing Judeo-Arabic texts. In Proceedings of the First International Conference on Arabic Computational Linguistics (ACLing ’15, Cairo), pages 138–144, April. find
5. Ann Bies, Zhiyi Song, Mohamed Maamouri, Stephen Grimes, Haejoong Lee, Jonathan Wright, Stephanie Strassel, Nizar Habash, Ramy Eskander, and Owen Rambow. 2014. Transliteration of Arabizi into Arabic orthography: Developing a parallel annotated Arabizi-Arabic script SMS/chat corpus. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 93–103. find
6. William Chan, Navdeep Jaitly, Quoc V. Le, and Oriol Vinyals. 2016. Listen, attend and spell: A neural network for large vocabulary conversational speech recognition. In 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4960–4964. find
7. Kyunghyun Cho, Bart van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder–decoder for statistical machine translation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1724–1734, Doha, Qatar, October. Association for Computational Linguistics. find
8. Thomas Deselaers, Saša Hasan, Oliver Bender, and Hermann Ney. 2009. A deep learning approach to machine transliteration. In Proceedings of the Fourth Workshop on Statistical Machine Translation, StatMT ’09, pages 233–241. Association for Computational Linguistics. find
9. Alex Graves, Santiago Fernández, Faustino Gomez, and Jürgen Schmidhuber. 2006. Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks. In Proceedings of the 23rd International Conference on Machine Learning, pages 369–376. ACM. ↗
10. Alex Graves, Santiago Fernández, Marcus Liwicki, Horst Bunke, and Jürgen Schmidhuber. 2007. Unconstrained on-line handwriting recognition with recurrent neural networks. In J. C. Platt, D. Koller, Y. Singer, and S. T. Roweis, editors, Advances in Neural Information Processing Systems 20 (NIPS 2007), volume 20, pages 577– 584, January. find
11. Nizar Y. Habash. 2010. Arabic Natural Language Processing. Synthesis Digital Library of Engineering and Computer Science. Morgan & Claypool Publishers. find
12. Mohamed Seghir Hadj Ameur, Farid Meziane, and Ahmed Guessoum. 2017. Arabic machine transliteration using an attention-based encoder-decoder model. In Third International Conference On Arabic Computational Linguistics (ACLING 2017, November 2017, Dubai, United Arab Emirates), volume 117 of Procedia Computer Science, pages 287–297, November. find
13. Yehuda Halevi. 2012. The Kuzari – In Defense of the Despised Faith. Al-Kamel Verlag, Beirut. Transliterated, translated into Arabic and annotated by Nabih Bashir. find
14. Kanishka Rao, Fuchun Peng, Haşim Sak, and Françoise Beaufays. 2015. Grapheme-to-phoneme conversion using long short-term memory recurrent neural networks. In IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4225–4229. find
15. Mihaela Rosca and Thomas Breuel. 2016. Sequence-to-sequence neural network models for transliteration. CoRR, abs/1610.09565. ↗
16. Mehdy Shaddel. 2018. Traces of the hamza in the early Arabic script: The inscriptions of Zuhayr, Qays the Scribe, and “Yazd the King”. Arabian Epigraphic Notes, 4:35–52. find
17. Baoguang Shi, Xiang Bai, and Cong Yao. 2017. An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition. IEEE Transactions on Pattern Analysis & Machine Intelligence, 39(11):2298–2304, November. find
18. Bonnie Glover Stalls and Kevin Knight. 1998. Translating names and technical terms in Arabic text. In Proceedings of the COLING/ACL Workshop on Computational Approaches to Semitic Languages (Semitic ’98), pages 34–41, August. find
19. Ori Terner. 2020. Transliteration of Judeo-Arabic texts to Arabic using RNN. M.Sc. thesis, School of Computer Science, Tel Aviv University. *Figure 6: First page of Maimonides’ The Guide for the Perplexed.* *Table 2: Simple mapping rules for baseline transliteration.* find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2020-terner-transliteration-judeo-arabic-texts-into/