Elamur Research⌕
☰
home / papers / 2025 / Lemmatizing Dialectal Arabic with Sequence-to-Sequence Models

Lemmatizing Dialectal Arabic with Sequence-to-Sequence Models

Paper · 2025 · Mostafa Saeed and Nizar Habash

Summary

Dialectal Arabic is hard to lemmatize. Spelling is not standardized, diacritics are usually missing, and morphological analyzers exist for only a few dialects, often with uneven quality. Saeed and Habash treat lemmatization as character-level generation with Seq2Seq models that output diacritized lemmas. They test the models alone and combined with the CAMeL Tools taggers for Egyptian, Gulf and Levantine. They also examine whether training on related varieties (other dialects, MSA corpora) and combining several taggers helps. Experiments use ARZATB (EGY), Gumar (GLF) and Curras (LEV). Baselines are the CAMeL Tools disambiguators.

The best configurations raise average lemma accuracy by about 12 points over dialect-specific taggers and about 22 points over an MSA tagger applied to dialects. The largest gains appear for Gulf and Levantine, whose analyzers were built automatically. Egyptian, which has an expert-built analyzer, gains little and does best with a single tagger. A manual error audit shows that many remaining "errors" come from flawed gold annotations. Code and models are publicly released.

Key points

  • Baseline gap: the CAMeL Tools MSA disambiguator scores 98.0 on MSA but only 64–69 on dialects.
  • Accuracy gains: dialect-specific taggers reach 90.4 (EGY), 79.1 (GLF) and 58.7 (LEV). The proposed system reaches 90.9, 93.7 and 79.5.
  • Training data sizes: ARZATB has about 134K words, Gumar about 162K, and Curras about 45K. MSA sources include ATB, NEMLAR, the Quranic Corpus, WikiNews, ZAEBUC and BAREC.
  • When pooling helps: cross-dialect training and multi-tagger setups (the DS and AD configs) help the weaker-resourced dialects, but not Egyptian with its high-quality analyzer.
  • Error audit findings:
    • Roughly 23% of errors trace back to bad gold lemmas.
    • Where the gold lemma is valid, about 10% of "wrong" predictions are actually correct and 11% are plausible.
    • The top error types are hallucination (14.0%), nominal pattern or noun–verb confusion (14.7% combined) and verb Form I/II confusion (10.7%).

Interesting ideas

  • I1. Generating a lemma character by character, and adding analyzer candidates where an analyzer exists, gives one model that works both with and without a lexicon, so coverage gaps never fully break the pipeline.
  • I2. Pooling data from related varieties pays off mainly when the target's own resources are weak, and can hurt a well-resourced target, so pooling decisions should be made per target rather than globally.
  • I3. The authors left out corpora whose lemma conventions were inconsistent until those could be normalized to a shared standard (CODA), rather than mixing conflicting annotation styles.
  • I4. Each error in a sample is labeled on three axes: whether the gold label is valid, whether the prediction is correct, plausible or wrong, and the error type. This shows how much of the measured error is really annotation noise.
  • I5. Constraining decoding so outputs stay within valid forms is proposed as the remedy for generator hallucinations, which are the most common failure.

For Mova

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

  • I1 (tokens): an en-crate fallback lemmatizer could generate lemmas for unknown, misspelled or slang words while still preferring lexicon analyses when they are available.
  • I2 (tree): when merging treebanks from different English varieties or genres during UD dialect conversion, measure the gain on each target separately before adopting a combined training set.
  • I3 (learn): silver or external treebanks should pass a convention-consistency check before they are added to training data.
  • I4 (learn): disagreements in evaluation could be split into "gold wrong / acceptable / truly wrong," which would prevent self-induced repair rules from being learned from noisy references.
  • I5 (tokens): any generative lemmatizer or morphology component should restrict its outputs to lexicon-attested lemmas (e.g., WordNet), so every lemma stays explainable and traceable.

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

🎓New York University Abu DhabiAbu Dhabi · United Arab Emirates122 works · 4,602 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 1Research subjects and languages the work is about.Semitic languages · 2,908 works

💡 Concepts 1Key ideas and methods explained on our concept pages.Lemmatization · 20 works

Cite

Cite as Mostafa Saeed and Nizar Habash. 2025. Lemmatizing Dialectal Arabic with Sequence-to-Sequence Models. Proceedings of The Third Arabic Natural Language Processing Conference.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.arabicnlp-main.10

@inproceedings{saeed2025lemmatizing,
  title = {{Lemmatizing Dialectal Arabic with Sequence-to-Sequence Models}},
  author = {Mostafa Saeed and Nizar Habash},
  booktitle = {Proceedings of The Third Arabic Natural Language Processing Conference},
  year = {2025},
  doi = {10.18653/v1/2025.arabicnlp-main.10},
  url = {https://doi.org/10.18653/v1/2025.arabicnlp-main.10}
}

References

46 references, 30 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdelali, Kareem Darwish, Nadir Durrani, and Hamdy Mubarak. 2016. Farasa: A fast and furious segmenter for Arabic. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations, pages 11–16, San Diego, California. Association for Computational Linguistics.
  2. Karim Al-Haff, Mustafa Jarrar, Tymaa Hammouda, and Fadi Zaraket. 2022. Curras + baladi: Towards a Levantine corpus. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 769–778, Marseille, France. European Language Resources Association. ↗
  3. Faisal Alshargi, Shahd Dibas, Sakhar Alkhereyf, Reem Faraj, Basmah Abdulkareem, Sane Yagi, Ouafaa Kacha, Nizar Habash, and Owen Rambow. 2019. Morphologically annotated corpora for seven Arabic dialects: Taizi, Sanaani, Najdi, Jordanian, Syrian, Iraqi and Moroccan. In Proceedings of the Fourth Arabic Natural Language Processing Workshop, pages 137–147, Florence, Italy. Association for Computational Linguistics. ↗
  4. Toms Bergmanis and Sharon Goldwater. 2018. Context sensitive neural lemmatization with lematus. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1391–1400.
  5. Mona T Diab, Mohamed Al-Badrashiny, Maryam Aminian, Mohammed Attia, Heba Elfardy, Nizar Habash, Abdelati Hawwari, Wael Salloum, Pradeep Dasigi, and Ramy Eskander. 2014. Tharwa: A Large Scale Dialectal Arabic-Standard Arabic-English Lexicon. In Proceedings of the Language Resources and Evaluation Conference (LREC), pages 3782–3789, Reykjavik, Iceland.
  6. Shahd Salah Uddin Dibas, Christian Khairallah, Nizar Habash, Omar Fayez Sadi, Tariq Sairafy, Karmel Sarabta, and Abrar Ardah. 2022. Maknuune: A large open Palestinian Arabic lexicon. In Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP), pages 131–141, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics. ↗
  7. Kais Dukes and Nizar Habash. 2010. Morphological Annotation of Quranic Arabic. In Proceedings of the Language Resources and Evaluation Conference (LREC), Valetta, Malta.
  8. Tarek El-Shishtawy and Fatma El-Ghannam. 2014. A lemma based evaluator for semitic language text summarization systems. arXiv preprint arXiv:1403.5596. ↗
  9. Khalid N. Elmadani, Nizar Habash, and Hanada Taha-Thomure. 2025. A large and balanced corpus for fine-grained Arabic readability assessment. In Findings of the Association for Computational Linguistics: ACL 2025, pages 16376–16400, Vienna, Austria. Association for Computational Linguistics.
  10. Ramy Eskander, Nizar Habash, and Owen Rambow. 2013. Automatic extraction of morphological lexicons from morphologically annotated corpora. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1032–1043, Seattle, Washington, USA. Association for Computational Linguistics. find
  11. Andrea Gesmundo and Tanja Samardzic. 2012. Lemmatisation as a tagging task. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 368–372.
  12. Nizar Habash, Mona Diab, and Owen Rambow. 2012a. Conventional orthography for dialectal Arabic. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12), pages 711–718, Istanbul, Turkey. European Language Resources Association (ELRA).
  13. Nizar Habash, Ramy Eskander, and Abdelati Hawwari. 2012b. A morphological analyzer for Egyptian Arabic. In Proceedings of the Twelfth Meeting of the Special Interest Group on Computational Morphology and Phonology, pages 1–9, Montréal, Canada. Association for Computational Linguistics.
  14. Nizar Habash and David Palfreyman. 2022. ZAEBUC: An annotated Arabic-English bilingual writer corpus. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 79–88, Marseille, France. European Language Resources Association. find
  15. Nizar Habash and Owen Rambow. 2007. Arabic diacritization through full morphological tagging. In Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Companion Volume, Short Papers, pages 53–56, Rochester, New York. Association for Computational Linguistics.
  16. 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. ↗
  17. Anton Karl Ingason, Sigrún Helgadóttir, Hrafn Loftsson, and Eiríkur Rögnvaldsson. 2008. A mixed method lemmatization algorithm using a hierarchy of linguistic identities (holi). In Advances in natural language processing: 6th international conference, GoTAL 2008 Gothenburg, Sweden, August 25-27, 2008 Proceedings, pages 205–216. Springer. find
  18. Svanhvít Lilja Ingólfsdóttir, Hrafn Loftsson, Jón Friðrik Daðason, and Kristín Bjarnadóttir. 2019. Nefnir: A high accuracy lemmatizer for Icelandic. In Proceedings of the 22nd Nordic Conference on Computational Linguistics, pages 310–315, Turku, Finland. Linköping University Electronic Press. ↗
  19. Go Inoue, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamor, and Nizar Habash. 2021. The interplay of variant, size, and task type in Arabic pre-trained language models. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 92– 104, Kyiv, Ukraine (Virtual). Association for Computational Linguistics.
  20. 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, pages 1708–1719, Dublin, Ireland. Association for Computational Linguistics.
  21. Mustafa Jarrar, Diyam Akra, and Tymaa Hammouda. 2024. Alma: Fast lemmatizer and pos tagger for arabic. Procedia Computer Science, 244:378–387. find
  22. Mustafa Jarrar, Nizar Habash, Faeq Alrimawi, Diyam Akra, and Nasser Zalmout. 2016. Curras: an annotated corpus for the Palestinian Arabic dialect. Language Resources and Evaluation, pages 1–31. ↗
  23. Mustafa Jarrar and Tymaa Hasanain Hammouda. 2024. Qabas: An open-source Arabic lexicographic database. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 13363–13370, Torino, Italia. ELRA and ICCL.
  24. Mustafa Jarrar, Fadi A Zaraket, Tymaa Hammouda, Daanish Masood Alavi, and Martin Wählisch. 2023. Lisan: Yemeni, iraqi, libyan, and sudanese arabic dialect corpora with morphological annotations. In 2023 20th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA), pages 1–7. IEEE. find
  25. Bart Jongejan and Hercules Dalianis. 2009. Automatic training of lemmatization rules that handle morphological changes in pre-, in-and suffixes alike. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP, pages 145–153.
  26. Salam Khalifa, Nizar Habash, Fadhl Eryani, Ossama Obeid, Dana Abdulrahim, and Meera Al Kaabi. 2018. A morphologically annotated corpus of Emirati Arabic. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA).
  27. Salam Khalifa, Nasser Zalmout, and Nizar Habash. 2020. Morphological analysis and disambiguation for Gulf Arabic: The interplay between resources and methods. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 3895–3904, Marseille, France. European Language Resources Association.
  28. Daniel Kondratyuk, Tomáš Gavenčiak, Milan Straka, and Jan Hajič. 2018. LemmaTag: Jointly tagging and lemmatizing for morphologically rich languages with BRNNs. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4921–4928, Brussels, Belgium. Association for Computational Linguistics. find
  29. Juan Liberato, Bashar Alhafni, Muhamed Khalil, and Nizar Habash. 2024. Strategies for Arabic readability modeling. In Proceedings of the Second Arabic Natural Language Processing Conference, pages 55–66, Bangkok, Thailand. Association for Computational Linguistics.
  30. Mohamed Maamouri, Ann Bies, Tim Buckwalter, and Wigdan Mekki. 2004. The Penn Arabic Treebank: Building a Large-Scale Annotated Arabic Corpus. In Proceedings of the International Conference on Arabic Language Resources and Tools, pages 102– 109, Cairo, Egypt. find
  31. Mohamed Maamouri, Ann Bies, Seth Kulick, Michael Ciul, Nizar Habash, and Ramy Eskander. 2014. Developing an Egyptian Arabic Treebank: Impact of dialectal morphology on annotation and tool development. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC-2014). European Language Resources Association (ELRA). find
  32. Mohamed Maamouri, Ann Bies, Seth Kulick, Dalila Tabessi, and Sondos Krouna. 2012. Egyptian Arabic Treebank DF Parts 1-8 V2.0 - LDC catalog numbers LDC2012E93, LDC2012E98, LDC2012E89, LDC2012E99, LDC2012E107, LDC2012E125, LDC2013E12, LDC2013E21. find
  33. GUIDO MINNEN, JOHN CARROLL, and DARREN PEARCE. 2001. Applied morphological processing of english. Natural Language Engineering, 7(3):207–223. ↗
  34. Hamdy Mubarak. 2018. Build fast and accurate lemmatization for Arabic. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA).
  35. Thomas Müller, Ryan Cotterell, Alexander Fraser, and Hinrich Schütze. 2015. Joint lemmatization and morphological tagging with lemming. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2268–2274, Lisbon, Portugal. Association for Computational Linguistics.
  36. Amal Nayouf, Tymaa Hammouda, Mustafa Jarrar, Fadi Zaraket, and Mohamad-Bassam Kurdy. 2023. Nâbra: Syrian Arabic dialects with morphological annotations. In Proceedings of ArabicNLP 2023, pages 12–23, Singapore (Hybrid). Association for Computational Linguistics. ↗
  37. 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.
  38. Arfath Pasha, Mohamed Al-Badrashiny, Ahmed El Kholy, Ramy Eskander, Mona Diab, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan Roth. 2014. Madamira: A fast, comprehensive tool for morphological analysis and disambiguation of arabic. In In Proceedings of LREC.
  39. Ryan Roth, Owen Rambow, Nizar Habash, Mona Diab, and Cynthia Rudin. 2008. Arabic morphological tagging, diacritization, and lemmatization using lexeme models and feature ranking. In Proceedings of ACL-08: HLT, Short Papers, pages 117–120, Columbus, Ohio. Association for Computational Linguistics.
  40. Mostafa Saeed and Nizar Habash. 2025. Lemmatization as a classification task: Results from Arabic across multiple genres. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, Suzhou, China. ↗
  41. Aleksi Sahala. 2024. Neural lemmatization and pos-tagging models for coptic, demotic and earlier egyptian. In Proceedings of the 1st Workshop on Machine Learning for Ancient Languages (ML4AL 2024), pages 87–97. ↗
  42. Rico Sennrich and Barry Haddow. 2016. Linguistic input features improve neural machine translation. In Proceedings of the First Conference on Machine Translation: Volume 1, Research Papers, pages 83– 91, Berlin, Germany. Association for Computational Linguistics.
  43. Mustafa Yaseen, Mohammed Attia, Bente Maegaard, Khalid Choukri, Niklas Paulsson, Salah Haamid, Steven Krauwer, Chomicha Bendahman, Hanne Fer-søe, Mohsen A Rashwan, et al. 2006. Building annotated written and spoken arabic lrs in nemlar project. In LREC, pages 533–538. Citeseer.
  44. Yin-Lai Yeong, Tien-Ping Tan, and Siti Khaotijah Mohammad. 2016. Using dictionary and lemmatizer to improve low resource english-malay statistical machine translation system. Procedia Computer Science, 81:243–249. find
  45. Nasser Zalmout and Nizar Habash. 2020a. Joint diacritization, lemmatization, normalization, and fine-grained morphological tagging. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8297–8307, Online. Association for Computational Linguistics.
  46. Nasser Zalmout and Nizar Habash. 2020b. Utilizing subword entities in character-level sequence-to-sequence lemmatization models. In Proceedings of the 28th International Conference on Computational Linguistics, pages 4676–4682, Barcelona, Spain (Online). International Committee on Computational Linguistics.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/saeed-2025-lemmatizing-dialectal-arabic-sequence-sequence/