Elamur Research⌕
☰
home / papers / 2026 / A Tale of Two Scripts: Transliteration and Post-Correction for Judeo-Arabic

A Tale of Two Scripts: Transliteration and Post-Correction for Judeo-Arabic

Paper · 2026 · Juan Moreno Gonzalez, Bashar Alhafni and Nizar Habash

Summary

Judeo-Arabic, the medieval Arabic of Jewish communities, was written in Hebrew letters. As a result, Arabic readers cannot use these texts, and Arabic NLP tools, which expect Arabic script, cannot process them. Converting the script is difficult for several reasons: letters map ambiguously, the use of the upper dot was never standardized, hamzas are often dropped, and Hebrew words are mixed in. The authors propose a two-stage pipeline that needs no Judeo-Arabic training data:

  1. A rule-based character mapper (CharMapper) converts the text.
  2. Existing Arabic grammatical error correction systems and LLMs clean up the output.

Gold data comes from aligning the Judeo-Arabic Al-Khazari with Nabih Bashir's Arabic edition. Hebrew tokens and punctuation stay in the text but are not scored. The paper also gives the first LLM benchmark for this task. The pipeline beats the only reproducible earlier system, from Mitelman et al. (2024). The transliterated text can then be processed by Arabic morphosyntactic taggers and Arabic-to-English MT. Code and data are released as CAMeL-Lab/jawhar.

Key points

  • The strongest configuration (dotted input ⇒ CharMapper ⇒ GPT-4o) still makes 3,896 word errors, about 9.6%.
  • About 16.6% of words contain dotted letters. On these words, the paper reports a large gap between settings (93.4% vs. 72.1%).
    • When dots are missing, recovery depends on the letter: about 93% for d/ð, but under half for k/x.
    • Words with several dotted letters are recovered only 31.5% of the time.
  • Manual review of 100 errors from the best system found:
    • 36% valid paraphrases of a non-literal reference
    • 21% unnecessary edits by GPT-4o
    • 18% Alif-Hamza mistakes
    • 7% errors already present in the source
    • 6% other
  • Hebrew words make up about 2% of tokens. In a sample of 100, only 16% matched the reference, mostly because the Arabic edition translated them instead of transliterating them.
  • Earlier work used inconsistent preprocessing and metrics, and Mitelman et al. (HeAr-BERT-based) is the only system with released models. The downstream tagging and MT evaluations use silver labels, including GPT-4o translations.

Interesting ideas

  • I1. A cheap deterministic mapper followed by an off-the-shelf general corrector replaces a task-specific trained model, so no in-domain labels are needed.
  • I2. Removing a marker from the input (dotted vs. dotless) and measuring recovery per letter shows how much information that marker carries. Grouping letters by how closely they resemble ordinary native-speaker typos explains why a corrector handles some cases and fails on others.
  • I3. The error taxonomy separates real mistakes from cases where the reference itself is non-literal ("valid paraphrase") or the source is wrong, and these explain a large share of measured errors.
  • I4. An "unnecessary change" category directly measures how often the corrector damages output that was already correct.
  • I5. Out-of-scope tokens (code-switched Hebrew, punctuation) stay in place for context but are excluded from scoring, which keeps the full text intact while keeping the metric clean.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tokens): noisy English input (OCR, archaic spelling, missing apostrophes) could pass through explicit normalization rules first, then a general corrector, with each edit logged so the step stays explainable.
  • I2 (tree): strip capitalization or punctuation from treebank sentences and measure LAS loss per construction, to find which cues the parsers truly depend on.
  • I3 (learn): when witness parsers disagree with silver or gold trees, label each disagreement as a reference problem, a source problem, or a genuine error before using it for training.
  • I4 (learn): self-induced repair rules should be evaluated as a negative control on already-correct parses, and any rule that breaks correct trees should be rejected.
  • I5 (tree): foreign words or garbled spans could remain in the UD tree with a marked relation and be dropped from LAS, so parsing still sees the full sentence.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of CambridgeCambridge · United Kingdom243 works · 121,396 citations🎓New York University Abu DhabiAbu Dhabi · United Arab Emirates122 works · 4,602 citations🎓Mohamed bin Zayed University of Artificial IntelligenceAbu Dhabi · United Arab Emirates124 works · 5,676 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksHebrew & Aramaic · 496 works

Cite

Cite as Juan Moreno Gonzalez, Bashar Alhafni and Nizar Habash. 2026. A Tale of Two Scripts: Transliteration and Post-Correction for Judeo-Arabic. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-long.93

@inproceedings{gonzalez2026tale,
  title = {{A Tale of Two Scripts: Transliteration and Post-Correction for Judeo-Arabic}},
  author = {Juan Moreno Gonzalez, Bashar Alhafni and Nizar Habash},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-long.93},
  url = {https://doi.org/10.18653/v1/2026.eacl-long.93}
}

References

42 references, 23 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 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, pages 30–38, Ann Arbor, Michigan. Association for Computational Linguistics.
  2. Bashar Alhafni and Nizar Habash. 2025. Enhancing text editing for grammatical error correction: Arabic as a case study. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 17892–17914, Vienna, Austria. Association for Computational Linguistics. ↗
  3. Bashar Alhafni, Go Inoue, Christian Khairallah, and Nizar Habash. 2023. Advancements in Arabic grammatical error detection and correction: An empirical investigation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 6430–6448, Singapore. Association for Computational Linguistics.
  4. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. AraBERT: Transformer-based model for Arabic language understanding. In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection, pages 9–15, Marseille, France. European Language Resource Association. ↗
  5. Mohammed Attia, Pavel Pecina, Younes Samih, Khaled Shaalan, and Josef van Genabith. 2012. Improved spelling error detection and correction for Arabic. In Proceedings of COLING 2012: Posters, pages 103– 112, Mumbai, India.
  6. The COLING 2012 Organizing Committee. find
  7. Kfir Bar, Nachum Dershowitz, Lior Wolf, Yackov Lubarsky, and Yaacov Choueka. 2015. Processing Judeo-Arabic texts. In 2015 First International Conference on Arabic Computational Linguistics (ACLing), pages 138–144. find
  8. Joshua Blau. 1961. A Grammar of Mediaeval Judaeo-Arabic. Magnes Press, Jerusalem. [In Hebrew]. find
  9. Joshua Blau. 1965. The Emergence and Linguistic Background of Judaeo-Arabic: A Study in the Origins of Middle Arabic. Oxford University Press. find
  10. Rawan Bondok, Mayar Nassar, Salam Khalifa, Kurt Micallef, and Nizar Habash. 2025. Proper noun diacritization for Arabic Wikipedia: A benchmark dataset. In Proceedings of the 2nd Workshop on Advancing Natural Language Processing for Wikipedia (WikiNLP 2025), pages 31–44, Vienna, Austria. Association for Computational Linguistics. ↗
  11. Achraf Chalabi and Hany Gerges. 2012. Romanized Arabic transliteration. In Proceedings of the Second Workshop on Advances in Text Input Methods, pages 89–96, Mumbai, India.
  12. The COLING 2012 Organizing Committee. find
  13. Daniel Dahlmeier and Hwee Tou Ng. 2012. Better evaluation for grammatical error correction. In Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 568–572, Montréal, Canada. Association for Computational Linguistics. find
  14. Kareem Darwish. 2014. Arabizi detection and conversion to Arabic. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 217–224, Doha, Qatar. Association for Computational Linguistics. ↗
  15. Kareem Darwish, Hamdy Mubarak, and Ahmed Abdelali. 2017. Arabic diacritization: Stats, rules, and hacks. In Proceedings of the Third Arabic Natural Language Processing Workshop, pages 9–17, Valencia, Spain. Association for Computational Linguistics.
  16. Salman Elgamal, Ossama Obeid, Mhd Kabbani, Go Inoue, and Nizar Habash. 2024. Arabic diacritics in the wild: Exploiting opportunities for improved diacritization. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14815–14829, Bangkok, Thailand. Association for Computational Linguistics. pages 1708–1719, Dublin, Ireland. Association for Computational Linguistics.
  17. Moussa Kamal Eddine, Nadi Tomeh, Nizar Habash, Joseph Le Roux, and Michalis Vazirgiannis. 2022. AraBART: a pretrained Arabic sequence-to-sequence model for abstractive summarization. In Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP), pages 31–42, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics.
  18. Ramy Eskander, Mohamed Al-Badrashiny, Nizar Habash, and Owen Rambow. 2014. Foreign words and the automatic processing of Arabic social media text written in Roman script. In Proceedings of the First Workshop on Computational Approaches to Code Switching, pages 1–12, Doha, Qatar. Association for Computational Linguistics.
  19. Salam Khalifa, Ossama Obeid, and Nizar Habash. 2021. Character edit distance based word alignment. https://github.com/CAMeL-Lab/ced_ word_alignment. ↗
  20. Friedberg Geniza Project. 1999. Friedberg Geniza Project. find
  21. Valentina Bella Lanza. 2020. Judeo-Arabic orthographies: Insights from a fifteenth-century šarḣ. Eurasian Studies, 18(1):134–148. ↗
  22. Imane Guellil, Faiçal Azouaou, Mourad Abbas, and Sadat Fatiha. 2017. Arabizi transliteration of Algerian Arabic dialect into Modern Standard Arabic. In Social MT 2017/ First workshop on Social Media and User Generated Content Machine Translation, Prague, Czech Republic. find
  23. Wiktor G˛ebski. 2024. A Grammar of the Jewish Arabic Dialect of Gabes. Open Book Publishers, Cambridge, UK. find
  24. Judah Ha-Levi. 2012. The Kuzari: The Book of Refutation and Proof on the Despised Faith. Al-Kamel Verlag, Freiberg. Nabih Bashir, Transliterated and edited with the assistance of ‘Abed al-Salam Muosa. find
  25. Nizar Habash, Salam Khalifa, Fadhl Eryani, Owen Rambow, Dana Abdulrahim, Alexander Erdmann, Reem Faraj, Wajdi Zaghouani, Houda Bouamor, Nasser Zalmout, Sara Hassan, Faisal Al shargi, Sakhar Alkhereyf, Basma Abdulkareem, Ramy Eskander, Mohammad Salameh, and Hind Saddiki. 2018. Unified Guidelines and Resources for Arabic Dialect Orthography. In Proceedings of the International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan.
  26. Nizar Habash, Abdelhadi Soudi, and Tim Buckwalter. 2007. On Arabic Transliteration. In A. van den Bosch and A. Soudi, editors, Arabic Computational Morphology: Knowledge-based and Empirical Methods, pages 15–22. Springer, Netherlands. ↗
  27. Nizar Y Habash. 2010. Introduction to Arabic natural language processing, volume 3. Morgan & Claypool Publishers. find
  28. Go Inoue, Bashar Alhafni, Nizar Habash, and Timothy Baldwin. 2026. Do diacritics matter? evaluating the impact of arabic diacritics on tokenization and llm benchmarks. In Findings of the Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco.
  29. Go Inoue, Salam Khalifa, and Nizar Habash. 2022. Morphosyntactic tagging with pre-trained language models for Arabic and its dialects. In Findings of the Association for Computational Linguistics: ACL 2022, Geoffrey Khan. 1997. The Arabic dialect of the Karaite Jews of Hit. Zeitschrift für Arabische Linguistik, 34:53–102.
  30. Jacob Mansour. 1991. The Jewish Baghdadi Dialect: Studies and Texts in the Judeo-Arabic Dialect of Baghdad. The Babylonian Jewry Heritage Center & The Institute for Research on Iraqi Jewry. find
  31. Kurt Micallef, Fadhl Eryani, Nizar Habash, Houda Bouamor, and Claudia Borg. 2023. Exploring the impact of transliteration on NLP performance: Treating Maltese as an Arabic dialect. In Proceedings of the Workshop on Computation and Written Language (CAWL 2023), pages 22–32, Toronto, Canada. Association for Computational Linguistics.
  32. Daniel Weisberg Mitelman, Nachum Dershowitz, and Kfir Bar. 2024. Code-switching and back-transliteration using a bilingual model. In Findings of the Association for Computational Linguistics: EACL 2024, pages 1501–1511, St. Julian’s, Malta. Association for Computational Linguistics.
  33. Hamdy Mubarak, Mohamed Al Sharqawy, and Esraa Al Masry. 2009. Diacritization and transliteration of proper nouns from arabic to english. In Proceedings of the Second International Conference on Arabic Language Resources and Tools, Cairo, Egypt. The MEDAR Consortium. find
  34. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. 2022. AraT5: Text-to-text transformers for Arabic language generation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 628–647, Dublin, Ireland. Association for Computational Linguistics.
  35. Ossama Obeid, Nasser Zalmout, Salam Khalifa, Dima Taji, Mai Oudah, Bashar Alhafni, Go Inoue, Fadhl Eryani, Alexander Erdmann, and Nizar Habash. 2020. CAMeL tools: An open source python toolkit for Arabic natural language processing. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 7022–7032, Marseille, France. European Language Resources Association. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, and 262 others. 2024. Gpt-4 technical report. Preprint, arXiv:2303.08774. ↗
  36. Aviad Rom and Kfir Bar. 2024. Training a bilingual language model by mapping tokens onto a shared character space. arXiv preprint arXiv:2402.16065. ↗
  37. Marina Rustow and Rebecca Sutton Koeser. 2022. Princeton Geniza Project. Center for Digital Humanities at Princeton. Neha Sengupta, Sunil Kumar Sahu, Bokang Jia, Satheesh Katipomu, Haonan Li, Fajri Koto, William Marshall, Gurpreet Gosal, Cynthia Liu, Zhiming Chen, Osama Mohammed Afzal, Samta Kamboj, Onkar Pandit, Rahul Pal, Lalit Pradhan, Zain Muhammad Mujahid, Massa Baali, Xudong Han, Sondos Mahmoud Bsharat, and 13 others. 2023. Jais and jais-chat: Arabic-centric foundation and instruction-tuned open generative large language models. Preprint, arXiv:2308.16149. ↗
  38. Ali Shazal, Aiza Usman, and Nizar Habash. 2020. A unified model for Arabizi detection and transliteration using sequence-to-sequence models. In Proceedings of the Fifth Arabic Natural Language Processing Workshop, pages 167–177, Barcelona, Spain (Online). Association for Computational Linguistics.
  39. Norman A. Stillman. 2010. Judeo-Arabic - history and linguistic description. In Encyclopedia of Jews in the Islamic World Online. Brill. Fanar Team, Ummar Abbas, Mohammad Shahmeer Ahmad, Firoj Alam, Enes Altinisik, Ehsannedin Asgari, Yazan Boshmaf, Sabri Boughorbel, Sanjay Chawla, Shammur Chowdhury, Fahim Dalvi, Kareem Darwish, Nadir Durrani, Mohamed Elfeky, Ahmed Elmagarmid, Mohamed Eltabakh, Masoomali Fatehkia, Anastasios Fragkopoulos, Maram Hasanain, and 23 others. 2025. Fanar: An arabic-centric multimodal generative ai platform. Preprint, arXiv:2501.13944. ↗
  40. Ori Terner, Kfir Bar, and Nachum Dershowitz. 2020. Transliteration of Judeo-Arabic texts into Arabic script using recurrent neural networks. In Proceedings of the Fifth Arabic Natural Language Processing Workshop, pages 85–96, Barcelona, Spain (Online). Association for Computational Linguistics. ↗
  41. Clare Voss, Stephen Tratz, Jamal Laoudi, and Douglas Briesch. 2014. Finding Romanized Arabic dialect in code-mixed tweets. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 2249–2253, Reykjavik, Iceland. European Language Resources Association (ELRA). Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, and 3 others. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45, Online. Association for Computational Linguistics.
  42. Jihene Younes, Emna Souissi, Hadhemi Achour, and Ahmed Ferchichi. 2018. A sequence-to-sequence based approach for the double transliteration of Tunisian dialect. Procedia Computer Science, 142:238–245. Arabic Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/gonzalez-2026-tale-two-scripts-transliteration-post/