Elamur Research⌕
☰
home / papers / 2026 / AlexandriaX 2026: The First Shared Task on Dialectal Arabic Machine Translation

AlexandriaX 2026: The First Shared Task on Dialectal Arabic Machine Translation

Summary

AlexandriaX 2026 is a shared task on machine translation for Arabic dialects. Most Arabic tools were built for MSA. Dialect output can keep the meaning and still sound wrong, for example through incorrect gender agreement, the wrong level of formality, or MSA slipping in. Aggregate metrics also say little about why an output fails.

The organizers set up three subtasks:

  • Subtask 1: English dialogue into 13 Arabic varieties, conditioned on prior turns and metadata (domain, speaker/addressee gender, persona). It uses the Alexandria corpus.
  • Subtask 2: Direct translation between six dialects for banking queries, built on an expanded ArBanking77. No full parallel training data was given.
  • Subtask 3: Locating erroneous spans in MT output and labelling them with six LQM-based linguistic categories across five varieties.

Twelve teams published system papers. The best systems clearly beat the baselines. The strongest approaches relied on dialect-specific adaptation, use of conversational context, retrieval with reranking, and hybrid error analysis. Pinpointing exact error spans remains difficult. All data, baselines and scoring code are public, which gives the field a shared benchmark for dialect-aware, explainable MT.

Key points

  • Alexandria corpus (Subtask 1):
    • About 107K parallel turns covering 13 countries and 11 domains.
    • English dialogues were generated by Gemini-2.5 Pro, translated by native speakers, then checked by a second reviewer from the same country.
    • The private test set has about 1,000 turns per dialect. Libyan and Sudanese appear only in the test set.
  • Subtask 1 tracks and results:
    • The constrained track allows only Alexandria data and models of at most 5B parameters.
    • Best scores were 30.42 spBLEU (constrained) and 33.49 spBLEU (unconstrained).
  • Subtask 2:
    • Covers Egyptian, Lebanese, Moroccan, Palestinian, Saudi and Tunisian.
    • The blind test set has 3,574 instances over 30 directed pairs.
    • The top system reached 28.40 spBLEU.
  • Subtask 3:
    • The six LQM categories are sociolinguistic, pragmatic, semantic, morphosyntactic, orthographic and graphetic.
    • Spans are scored with Exact Match F1 and Overlap F1; classes are scored with Error Class Macro-F1.
    • The best overall score was 49.82 against a 24.29 baseline.
  • Participation: 38, 33 and 35 registrations for the three subtasks. Subtasks 1 and 2 are scored with spBLEU and chrF++, which the authors admit penalize valid dialectal variation.

Interesting ideas

  • I1. Keep some varieties completely out of training and development and show them only at test time, so that generalization is measured directly rather than assumed. — For Mova (tree): Withhold one UD dialect or genre treebank entirely from the training of the perceptron parser and gparse, and report it as a separate unseen-variety score.
  • I2. Condition translation on explicit speaker and addressee attributes (gender, role, register), because a translation can be faithful in meaning yet still wrong socially. — For Mova (back): When generating text from the meaning graph, read the gender and role of participants from the world model so that agreement and politeness are chosen deliberately rather than by default.
  • I3. Replace one quality score with error spans labelled by linguistic layer, from semantics down to orthography, so failures become diagnosable. — For Mova (learn): Tag each evaluation error by the stage that produced it (morphology, attachment, role, pragmatics), so that self-induced repair rules can target a specific layer.
  • I4. Leading error detectors split the job: a generative model decides the error type, while an encoder anchored to the input fixes the exact offsets. — For Mova (learn): Let the LLM teacher tier propose what is wrong with a silver parse, but let a token-anchored component decide exactly which tokens and arcs are affected.
  • I5. Systems trained on clean dialogue histories but run on their own generated histories suffer from that mismatch, so training should resemble inference conditions. — For Mova (frames): Train coreference and frame linking on the system's own predicted analyses of earlier sentences, not on gold ones, so errors in the context do not catch the model off guard.

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

Fields: Linguistics & languages

Topics: Machine translation

Cited by

2 papers of our library cite it (of the 140884 whose bibliographies we read); each has a page here.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2609.22796 · arXiv 2609.22796

@misc{mekki2026alexandriax,
  title = {{AlexandriaX 2026: The First Shared Task on Dialectal Arabic Machine Translation}},
  author = {Mekki, Abdellah El and Elmadany, AbdelRahim A. and Magdy, Samar M. and Ezzini, Saad and El-Haj, Mo and Jarrar, Mustafa and Alyafeai, Zaid and Ghanem, Bernard and Abdul-Mageed, Muhammad},
  year = {2026},
  doi = {10.48550/arxiv.2609.22796},
  eprint = {2609.22796},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2609.22796}
}

References

38 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mohamed A. Abdallah and Samhaa R. El-Beltagy. 2026. LahjaMT at AlexandriaX-2026: A Context-Aware English-to-Dialectal Arabic Machine Translation System with Lightweight Routing of LoRA Experts. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  2. Muhammad Abdul-Mageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2021. ARBERT & MARBERT: Deep bidirectional transformers for Arabic. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 7088–7105, Online. Association for Computational Linguistics. ↗ our card
  3. Muhammad Abdul-Mageed, AbdelRahim Elmadany, Chiyu Zhang, El Moatez Billah Nagoudi, Houda Bouamor, and Nizar Habash. 2023. NADI 2023: The fourth nuanced Arabic dialect identification shared task. In Proceedings of ArabicNLP 2023, pages 600– 613, Singapore (Hybrid). Association for Computational Linguistics. find
  4. Ali Abusaleh, Bhuvanesh Verma, and Alexander Mehler. 2026. TTLab at AlexandriaX-2026: A Fine-Tuned Surface Tagger for Arabic Machine-Translation Error-Span Detection and Classification. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find our card
  5. Abdullah Alabdullah, Lifeng Han, and Chenghua Lin. 2026. Ara-HOPE: Human-centric post-editing evaluation for dialectal Arabic to Modern Standard Arabic translation. In Proceedings of the 13th Workshop on NLP for Similar Languages, Varieties and Dialects, pages 157–171, Rabat, Morocco. Association for Computational Linguistics. find our card
  6. Meshal Alamr, Maryam Altuwaijri, Hassan Alqaeri, and Abdullah Aldahlawi. 2026. Thakaa at AlexandriaX-2026: Who Gets the Final Say? Assigning Output Authority in Dialectal Arabic Machine Translation and Error Detection. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  7. Hasan Alkhder, Maher Pasa, Maria Hanini, Amro Najjar, and Ahmet Zengin. 2026. alkhder at AlexandriaX-https://alliancecan.ca 2026: Context-Aware English-to-Dialectal Arabic Dialogue Translation with QLoRA-Fine-Tuned NileChat-3B. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. ↗
  8. Ranwa Aly, Seif Eldin Maan, Farida Aly, and Isis Ashraf. 2026. FCDS at AlexandriaX-2026: Improving Context Aware Dialogue English To Dialectal Arabic Machine Translation Via Context Strategies And Quality Aware Decoding. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  9. A. Bergman and Mona Diab. 2022. Towards responsible natural language annotation for the varieties of Arabic. In Findings of the Association for Computational Linguistics: ACL 2022, pages 364–371, Dublin, Ireland. Association for Computational Linguistics. find
  10. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, and Kemal Oflazer. 2018. The MADAR Arabic dialect corpus and lexicon. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find
  11. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, Luke Marris, Sam Petulla, Colin Gaffney, Asaf Aharoni, Nathan Lintz, Tiago Cardal Pais, Henrik Jacobsson, Idan Szpektor, Nan-Jiang Jiang, and 3416 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. Preprint, arXiv:2507.06261. ↗ our card
  12. Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, et al. 2024. Scaling neural machine translation to 200 languages. Nature, 630(8018):841–846. find our card
  13. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. Qlora: efficient finetuning of quantized llms. In Proceedings of the 37th International Conference on Neural Information Processing Systems, NIPS ’23, Red Hook, NY, USA. Curran Associates Inc. find
  14. Abdellah El Mekki, Houdaifa Atou, Omer Nacar, Shady Shehata, and Muhammad Abdul-Mageed. 2025. NileChat: Towards linguistically diverse and culturally aware LLMs for local communities. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 10978–11002, Suzhou, China. Association for Computational Linguistics. find
  15. Abdellah EL Mekki, Samar M. Magdy, Houdaifa Atou, Ruwa AbuHweidi, Baraah Qawasmeh, Omer Nacar, Thikra Al-hibiri, Razan Saadie, Hamzah A. Alsayadi, Nadia Ghezaiel Hammouda, Alshima Mohammed Alkhazimi, Aya Hamod, Al-Yas Yaqoob Al-Ghafri, Wesam El-Sayed, Asila Ismail al Sharji, Mohamad Ballout, Anas Belfathi, Karim Ghaddar, Serry Sibaee, and 28 others. 2026. Alexandria: A multi-domain dialectal Arabic machine translation dataset for culturally inclusive and linguistically diverse LLMs. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 32567–32592, San Diego, California, United States. Association for Computational Linguistics. find
  16. Fatimah Emad Eldin, Khloud Al Jallad, Mona A. Azim, and Omer Nacar. 2026. NAMAA Community at AlexandriaX-2026: Prompting, Fine-Tuning and Agreement Voting for Dialectal Arabic Translation and Evaluation. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  17. AbdelRahim Elmadany, El Moatez Billah Nagoudi, and Muhammad Abdul-Mageed. 2023. Octopus: A multitask model and toolkit for Arabic natural language generation. In Proceedings of ArabicNLP 2023, pages 232–243, Singapore (Hybrid). Association for Computational Linguistics. find
  18. Ashraf Hatim Elneima, AhmedElmogtaba Abdelmoniem Ali Abdelaziz, and Kareem Darwish. 2024. OSACT6 dialect to MSA translation shared task overview. In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024, pages 93–97, Torino, Italia. ELRA and ICCL. find
  19. Nada Esmaeil, Fathima Rena, Sibi Subhash, Osama Elgendy, Mina Naguib, Salma Omar, and Muhammad Arif. 2026. Rosetta at AlexandriaX-2026: LoRA-Adapted NileChat for Context-Aware Dialectal Arabic Dialogue Translation. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find our card
  20. Gemma Team, Sherif El Abd, Vaibhav Aggarwal, Robin Algayres, Alek Andreev, Olivier Bachem, Ian Ballantyne, Cormac Brick, Victor Cărbune, Michelle Casbon, Mayank Chaturvedi, Aditya Chawla, Victor Cotruta, Alice Coucke, Phil Culliton, Robert Dadashi, Lucas Dixon, Mohamed Elhawaty, Utku Evci, and 304 others. 2026. Gemma 4 technical report. Preprint, arXiv:2607.02770. ↗
  21. Nuno M. Guerreiro, Ricardo Rei, Daan van Stigt, Luisa Coheur, Pierre Colombo, and André F. T. Martins. 2024. xCOMET: Transparent machine translation evaluation through fine-grained error detection. Transactions of the Association for Computational Linguistics, 12:979–995. find our card
  22. Mohamed Motasim Hamed, Muhammad Hreden, Khalil Hennara, Zeina Aldallal, Sara Chrouf, and Safwan AlModhayan. 2025. Lahjawi: Arabic cross-dialect translator. In Proceedings of the 4th Workshop on Arabic Corpus Linguistics (WACL-4), pages 12–24, Abu Dhabi, UAE. Association for Computational Linguistics. Md. find
  23. Ajwad Hossain. 2026. Axiom at AlexandriaX-2026: Confidence-Thresholded Dialectal Encoder Fine-Tuning for MT Error Span Detection and Classification. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  24. Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations. find
  25. Momen Ibrahim, Mariam E. Hassib, Ahmed Salem, and Marwan Torki. 2026. ALEXIS at AlexandriaX-2026 Subtask 2: Capacity Discovery and Calibrated Preference Optimization for Cross-Dialect Arabic MT. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  26. Mustafa Jarrar, Ahmet Birim, Mohammed Khalilia, Mustafa Erden, and Sana Ghanem. 2023. ArBanking77: Intent detection neural model and a new dataset in modern and dialectical Arabic. In Proceedings of the ArabicNLP 2023, pages 276–287, Singapore (Hybrid). Association for Computational Linguistics. find
  27. Josef Jon and Ondřej Bojar. 2026. CUNI at AlexandriaX-2026: Finetuning LLMs for dialectal Arabic Translation. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  28. Salam Khalifa, Nizar Habash, Dana Abdulrahim, and Sara Hassan. 2016. A large scale corpus of Gulf Arabic. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 4282–4289, Portorož, Slovenia. European Language Resources Association (ELRA). find
  29. Shankar Kumar and William Byrne. 2004. Minimum Bayes-risk decoding for statistical machine translation. In Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages 169–176, Boston, Massachusetts, USA. Association for Computational Linguistics. find our card
  30. Mariam Labib, Bakhom Ramzy, Nsrin Ashraf, Ahmed M. Fetouh, and Hamada Nayel. 2026. REGLAT at AlexandriaX-2026: A Two-Stage Multi-Encoder Framework for Character-Level Error Detection and LQM Classification in Dialectal Arabic Machine Translation. In Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks, Budapest, Hungary. Association for Computational Linguistics. find
  31. Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2020. Focal loss for dense object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(2):318–327. find our card
  32. Samar M. Magdy, Fakhraddin Alwajih, Abdellah El Mekki, Wesam El Sayed, and Muhammad Abdul-Mageed. 2026. LQM: Linguistically motivated multidimensional quality metrics for machine translation. In Findings of the Association for Computational Linguistics: ACL 2026, pages 40470–40493, San Diego, California, United States. Association for Computational Linguistics. find
  33. Sanad Malaysha, Mo El-Haj, Saad Ezzini, Mohammed Khalilia, Mustafa Jarrar, Sultan Almujaiwel, Ismail Berrada, and Houda Bouamor. 2024. AraFinNLP 2024: The first Arabic financial NLP shared task. In Proceedings of the Second Arabic Natural Language Processing Conference, pages 393–402, Bangkok, Thailand. Association for Computational Linguistics. find
  34. Josef Jon, Rawan Bondok, and Ondřej Bojar. 2026. Current state of LLMs for Arabic dialectal machine translation. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script, pages 329–363, Rabat, Morocco. Association for Computational Linguistics. find
  35. Karima Meftouh, Salima Harrat, Salma Jamoussi, Mourad Abbas, and Kamel Smaili. 2015. Machine translation experiments on PADIC: A parallel Arabic DIalect corpus. In Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation, pages 26–34, Shanghai, China. find
  36. Karima Kadaoui, Samar M. Magdy, Abdul Waheed, Md Tawkat Islam Khondaker, Ahmed Oumar El-Shangiti, El Moatez Billah Nagoudi, and Muhammad Abdul-Mageed. 2023. TARJAMAT: Evaluation of bard and ChatGPT on machine translation of ten Arabic varieties. In Proceedings of ArabicNLP 2023, pages 52–75, Singapore (Hybrid). Association for Computational Linguistics. find our card
  37. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. 2022. TURJUMAN: A public toolkit for neural Arabic machine translation. In Proceedinsg of the 5th Workshop on Open-Source Arabic Corpora and Processing Tools with Shared Tasks on Qur’an QA and Fine-Grained Hate Speech Detection, pages 1–11, Marseille, France. European Language Resources Association. find
  38. OpenAI. 2026. GPT-5.5 System Card. https://openai.com/index/gpt-5-5-system-card/. Accessed: 2026-08-31. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mekki-2026-alexandriax-2026-first-shared-task/