Elamur Research⌕
☰
home / papers / 2026 / AbjadMed: Arabic Medical Text Classification at AbjadNLP 2026

AbjadMed: Arabic Medical Text Classification at AbjadNLP 2026

Paper · 2026 · Pranav Gupta, Niranjan Kumar M, Balaji Nagarajan et al.

Summary

AbjadMed is a shared task at the 2nd AbjadNLP workshop (EACL 2026). It sets up a benchmark for single-label classification of Arabic medical question–answer text into 82 categories. The data reflect real healthcare conditions: class frequencies are strongly skewed, some categories overlap in meaning, and labels carry the noise of ordinary annotation. The organisers built a curated subset of the Arabic Healthcare Dataset (AHD), which was collected from the Altibbi platform. They capped very frequent classes, dropped a few extremely rare ones, and otherwise left the labels as found. Each question and its professional answer are joined into one text. Systems are ranked by macro-averaged F1, so every category counts equally regardless of size. Participating teams mostly fine-tuned Arabic BERT variants, and one used Qwen3-8B, combined with many imbalance-handling tricks. The results show the task is still hard for modern pretrained models, especially on rare and semantically close categories. The benchmark gives a reproducible basis for studying robustness in Arabic healthcare NLP.

Key points

  • The source corpus, AHD, holds more than 808,000 Q–A pairs across 90 categories. The shared-task subset has 46,585 instances: 27,951 for training and 18,634 for testing, covering 82 classes.
  • Frequent classes were downsampled to at most 600 examples each, and 8 rare categories were removed. No relabelling, merging or linguistic normalisation was applied.
  • Texts average about 59 words (58.28 in test). Lengths range from 8 to 2,223 words in training.
  • Category names were originally Arabic and were translated into English by an LLM for inspection only. Systems predict integer labels, and submissions are CSV files with the columns Id and Predicted.
  • Teams used several backbones:
    • CamelBERT, MARBERT/MARBERTv2, ARBERT, AraBERT and Qwen3-8B (with QLoRA).
    • Techniques included focal loss, LDAM loss, class weights, back-translation, kNN retrieval, attention or mean pooling, adversarial training and ensembling.
    • GATech found that a plain AraBERT encoder beat LLM-based reranking.

Interesting ideas

  • I1. Macro-F1 gives every class the same weight, so a system cannot hide poor rare-class performance behind high accuracy on frequent classes.
  • I2. The organisers kept the original overlapping and noisy labels and scored strictly against them, with no merging or smoothing afterwards; this makes the benchmark reflect real-world difficulty rather than a tidied version of it.
  • I3. To build the dataset, the organisers capped frequent classes at a fixed count and dropped only categories too rare to learn, which tames skew while keeping a realistic long tail.
  • I4. Tashkees-AI found that a flat classifier outperformed hierarchical and RAG-based pipelines, because mistakes in early stages carried through to later ones.
  • I5. REGLAT placed a kNN classifier over fine-tuned embeddings and added a specialist MLP that corrects predictions for rare classes. This is a cheap way to give minority labels dedicated attention.

For Mova

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

  • I1 (learn): Report macro-averaged scores over rare frame types, VerbNet classes or dependency relations, so improvements on common constructions do not mask regressions in the long tail.
  • I2 (frames): Evaluate semantic-role and speech-act labelling against labels as annotated, and analyse confusions between near-synonymous frames separately instead of quietly merging them.
  • I3 (learn): Apply a per-construction cap when assembling silver data from witness parsers, so common patterns do not swamp the training mix.
  • I4 (answer): Measure how errors propagate across stages from tree to frames, world and answer, and compare against shortcut baselines to see where pipeline cascades cost accuracy.
  • I5 (frames): A nearest-neighbour lookup over already-resolved examples, followed by a small corrector for rare VerbNet classes, could improve frame assignment while staying explainable through the retrieved neighbours.

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, 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

Cite

Cite as Pranav Gupta, Niranjan Kumar M, Balaji Nagarajan, Imed Zitouni and Mo El-Haj. 2026. AbjadMed: Arabic Medical Text Classification at AbjadNLP 2026. Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.abjadnlp-1.64

@inproceedings{gupta2026abjadmed,
  title = {{AbjadMed: Arabic Medical Text Classification at AbjadNLP 2026}},
  author = {Pranav Gupta, Niranjan Kumar M, Balaji Nagarajan, Imed Zitouni and Mo El-Haj},
  booktitle = {Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script},
  year = {2026},
  doi = {10.18653/v1/2026.abjadnlp-1.64},
  url = {https://doi.org/10.18653/v1/2026.abjadnlp-1.64}
}

References

28 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 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. Association for Computational Linguistics. ↗
  2. Shrouq Al-Fuqaha’a, Nailah Al-Madi, and Bassam Hammo. 2024. A robust classification approach to enhance clinic identification from arabic health text. Neural Computing and Applications, 36(13):7161– 7185. ↗
  3. S Al-Harbi, A Almuhareb, A Al-Thubaity, M. S. Khorsheed, and A Al-Rajeh. 2008. Automatic arabic text classification. In Proceedings of The 9th International Conference on the Statistical Analysis of Textual Data (01/03/08). find
  4. Nashwan Ahmed Al-Majmar, Hezam Gawbah, and Akram Alsubari. 2024. Ahd: Arabic healthcare dataset. Data in Brief, 56:110855. ↗
  5. Ashwag Alasmari, Sarah Alhumoud, and Waad Alshammari. 2024. Aramed: Arabic medical question answering using pretrained transformer language models. In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OS-ACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation@ LREC-COLING 2024, pages 50–56.
  6. Hassan Alhuzali, Farah E Shamout, Muhammad Abdul-Mageed, Chaimae Abouzahir, Mouath Abu Daoud, Ashwag Alasmari, Walid Al-Eisawi, Renad Al-Monef, Ali Alqahtani, Lama Ayash, and 1 others. 2025. Arahealthqa 2025: The first shared task on arabic health question answering. In Proceedings of the Third Arabic Natural Language Processing Conference: Shared Tasks, pages 107–118.
  7. 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 Process-Mahmoud El-Haj, Paul Rayson, Scott SL Piao, and Jo Knight. 2018. Profiling medical journal articles using a gene ontology semantic tagger. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018). ↗
  8. Mo El-Haj. 2025. Proceedings of the 1st workshop on nlp for languages using arabic script (abjadnlp 2025). In Proceedings of the 1st Workshop on NLP for Languages Using Arabic Script, held at COLING 2025, Abu Dhabi, United Arab Emirates. find
  9. Mo El-Haj. 2026. Proceedings of the 2nd workshop on nlp for languages using arabic script (abjadnlp 2026). In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script, held at EACL 2026, Rabat, Morocco. find
  10. Fatimah Mohamed Emad Eldin. 2026. Flat vs. Hierarchical Classification for Fine-Grained Arabic Medical QA. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  11. Ahmed Megahed Fetouh, Mohammed Rahmath, Omer Dawood, Mariam Labib, Nsrin Ashraf, and Hamada Nayel. 2026. Handling Imbalanced Arabic Medical Text Classification via Hierarchical KNN-MLP Architecture. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  12. Jaafar Hammoud, Aleksandra Vatian, Natalia Dobrenko, Nikolai Vedernikov, Anatoly Shalyto, and Natalia Gusarova. 2021. New arabic medical dataset for diseases classification. In International Conference on Intelligent Data Engineering and Automated Learning, pages 196–203. Springer.
  13. 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. Association for Computational Linguistics.
  14. Rahul Jaisy. 2026. Forensic Optimization and Class-Balanced Discovery for Medical Diglossia in Abjad Scripts. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  15. Sujith Kanakkassery. 2026. Imbalance-Aware Transformer Fine-Tuning for Arabic Medical Text Classification. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  16. Abdallah Khallouf. 2026. Efficient Fine-Tuning with All-Linear LoRA for Arabic Medical QA Classification. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  17. Ahmed Khamis. 2026. Bidirectional Encoders vs. Causal Decoders: Insights from 82-Class Arabic Medical Classification. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  18. Daisy Monika Lal, Paul Rayson, Christopher Peter, Ignatius Ezeani, Mo El-Haj, Yafei Zhu, and Yufeng Liu. 2025. Lens: Learning entities from narratives of skin cancer. In Proceedings of the 31st International Conference on Computational Linguistics: System Demonstrations, pages 20–27. find
  19. Minh-Hoang Le. 2026. Combining LDAM Loss and Adversarial Training for Arabic Medical Question-Answer Classification. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  20. Abdulaziz Malik, Mahmoud El-Haj, and Michael K Paasche-Orlow. 2019. Readability of patient educational materials in english versus arabic. HLRP: Health Literacy Research and Practice, 3(3):e170– e173. study. International Journal of Corpus Linguistics, 26(3):305–335. ↗
  21. Nahid Montasir Rifat and Foyez Ahmed Dewan. 2026. Imbalance-Aware Fine-Tuning of Pretrained Arabic Transformers for Arabic Medical Text Classification Task. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  22. Gleb Shanshin. 2026. Fighting Class Imbalance in Arabic Medical Text Classification. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  23. Amrita Singh. 2026. Arabic Medical Text Classification via Data- and Algorithm-Level Fusion. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  24. Olga Snissarenko. 2026. Arabic Clinical Text Classification with AraBERT: Results from the AbjadMed Shared Task. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  25. Ahlam Wahdan, Mostafa Al-Emran, and Khaled Shaalan. 2024. A systematic review of arabic text classification: areas, applications, and future directions. Soft Computing-A Fusion of Foundations, Methodologies & Applications, 28(2). ↗
  26. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, and 8 others. 2025. Qwen3 technical report. arXiv, abs/2505.09388. ↗
  27. Thanh Phu Nguyen, Tuan Thai Huy Nguyen Cu, Son Thai Pham, and Tri Duy Ho Nguyen. 2026. Enhancing Arabic Medical Text Classification Using Fine-Tuned AraBERT. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco. find
  28. Sheryl Prentice, Jo Knight, Paul Rayson, Mahmoud El Haj, and Nathan Rutherford. 2021. Problematising characteristicness: a biomedical association case find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/gupta-2026-abjadmed-arabic-medical-text-classification/