Elamur Research⌕
☰
home / papers / 2026 / QIAS 2026: Overview of the Shared Task on Islamic Inheritance Reasoning

QIAS 2026: Overview of the Shared Task on Islamic Inheritance Reasoning

Paper · 2026 · Abdessalam Bouchekif, Somaya Eltanbouly, Shahd Gaben et al.

Summary

QIAS 2026, a shared task at OSACT7 (co-located with LREC 2026), tests whether large language models can solve Islamic inheritance (ʿilm al-mawārīth) cases written in Arabic from start to finish. A system reads a natural-language case and outputs a reasoning trace and a structured answer. The answer must cover four things: which heirs inherit and which are blocked (hajb), each heir's legal share (furūḍ), whether a global adjustment (ʿawl or radd) is needed, and the final distribution of the estate.

The data come from MAWARITH, a set of 12,500 cases. These were produced with a rule-based calculator, rewritten as natural questions, and annotated by an expert. Submissions were scored with MIR-E, a weighted metric that grades each reasoning stage separately.

Sixteen teams took part, using prompting, retrieval-augmented generation (RAG), fine-tuning and neuro-symbolic hybrids. A RAG pipeline won with a score of 0.935. A small fine-tuned open model came close to a commercial model. The task shows that legally precise, numerically exact reasoning is still hard for LLMs.

Key points

  • MAWARITH has 12,000 training and 500 test cases. Most are simple (11,535), with 616 ʿawl and 349 radd cases. The test split contains only 39 ʿawl and 5 radd cases. All cases follow the majority (al-jumhūr) opinion.
  • MIR-E has four components:
    • S_h scores heirs and blocking, including heir counts.
    • S_s scores share assignment.
    • S_a scores the adjustment type, and only counts if the first two stages are fully right.
    • S_f scores the final distribution.
  • The organisers' baseline is prompting-only Fanar-Sadiq. Data and baseline code are on GitLab (islamgpt1/qias_shared_task_2026).
  • Top of the leaderboard:
    • CVPD won with RAG on Qwen-9B (0.935).
    • Simplicity came second with a neuro-symbolic pipeline (0.931).
    • KMS was third with prompting (0.916).
    • QU-NLP scored 0.907 with multi-stage QLoRA on a 4B Qwen3, close to Gemini-2.5-Flash's reported 0.901.
  • Fine-tuning results were mixed:
    • Team Silah found fine-tuning alone beat retrieval, with a fine-tuned Fanar model doing best.
    • AGS-KSU's fine-tuned Qwen2.5-3B reached only 0.30, while their GPT-5.4 Thinking prompting setup reached 0.84.

Interesting ideas

  • I1. Build gold data from a deterministic solver first, then have the structured cases rewritten as fluent questions; correctness comes from the solver, while realism comes from the rewriting.
  • I2. Score a dependent pipeline stage by stage, and award credit for a later stage only when earlier stages are fully right, so lucky guesses on downstream steps are not rewarded.
  • I3. Let the language model only extract entities, normalise them to a fixed inventory of categories, and hand them to a symbolic rule engine for all computation; this nearly matched the winning system.
  • I4. Require one schema-constrained output, then parse and validate it before scoring, so that format failures are caught and legal errors can be told apart from arithmetic errors.
  • I5. Adapt a small model in two steps: first on a broad in-domain corpus (fatwas), then on the structured target task; this let a 4B model approach a commercial one.

For Mova

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

  • I1 (answer): Maths word problems could be produced by generating world states with known solutions and then verbalising them, giving exact answer keys for the readers.
  • I2 (learn): Evaluation could grade tokens, tree, frames and world separately, crediting later stages only on correct upstream input, so we see exactly where errors start.
  • I3 (world): Mova can map extracted entities to canonical ontology classes and compute holdings, shares or quantities with explicit rules, keeping every number explainable.
  • I4 (answer): Answers could be emitted as typed structures checked against a schema, so the test harness can label each failure as a reasoning, extraction or calculation fault.
  • I5 (learn): The gparse or tagger training could pretrain on large unlabelled in-domain text before fine-tuning on silver treebank data for a specific genre such as fables.

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

🎓Hamad bin Khalifa UniversityDoha · Qatar141 works · 14,175 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksPolitical Science and International Relations · 182 works

🧭 Topics 3Research subjects and languages the work is about.Reasoning · 1,426 worksSemitic languages · 2,908 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Abdessalam Bouchekif, Somaya Eltanbouly, Shahd Gaben, Mohammed Ghaly, Samer Rashwani, MOHAMED Emad and Heba Sbahi. 2026. QIAS 2026: Overview of the Shared Task on Islamic Inheritance Reasoning. The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/55e9fi9ftwnm

@inproceedings{bouchekif2026qias,
  title = {{QIAS 2026: Overview of the Shared Task on Islamic Inheritance Reasoning}},
  author = {Abdessalam Bouchekif, Somaya Eltanbouly, Shahd Gaben, Mohammed Ghaly, Samer Rashwani, MOHAMED Emad and Heba Sbahi},
  booktitle = {The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks},
  year = {2026},
  doi = {10.63317/55e9fi9ftwnm},
  url = {https://doi.org/10.63317/55e9fi9ftwnm}
}

References

34 references, 26 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ummar Abbas, Mohammad Shahmeer Ahmad, Minhaj Ahmad, Abdulaziz Al-Homaid, Anas Al-Nuaimi, Enes Altinisik, Ehsaneddin Asgari, Sanjay Chawla, Shammur Chowdhury, Fahim Dalvi, et al. 2026. Fanar 2.0: Arabic generative AI stack. arXiv preprint arXiv:2603.16397. ↗
  2. Ali Abdelaal, Mohammed Nader Al Haffar, Mahmoud Fawzi, and Walid Magdy. 2026. Islamicmmlu: A benchmark for evaluating llms on islamic knowledge. arXiv preprint 2603.23750. ↗
  3. Nouar AlDahoul and Yasir Zaki. 2025. NYUAD at QIAS shared task: Benchmarking the legal reasoning of LLMs in arabic islamic inheritance cases. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 861–866.
  4. Omar Alkhamis. 2026. KMS at QIAS 2026: Evaluating llm reasoning for islamic inheritance division. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain. find
  5. Muath Almansour. 2026. Simplicity at QIAS 2026: Decoupling language extraction from mathematical logic in islamic inheritance law. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain. find
  6. Ameera Almasoud, Sharefah Al-Ghamdi, Reem Alqifari, Noof Alfear, and Hend Al-Khalifa. 2026. MirathQA: A dataset for evaluating large language models on hanbali islamic inheritance reasoning tasks. Data in Brief, page 112589. find
  7. Sanaa Alowaidi. 2025. SEA-Team at QIAS 2025: Enhancing LLMs for question answering in islamic texts. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 940–946, Suzhou, China. Association for Computational Linguistics.
  8. Mohammad Alsmadi. 2026. QU-NLP at QIAS 2026: Multi-stage QLoRA fine-tuning for arabic islamic inheritance reasoning. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain.
  9. Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. 2023. Gemini: A family of highly capable multimodal models. arXiv:2312.11805. arXiv preprint Salah Eddine Bekhouche, Abdellah Zakaria Sellam, Hichem Telli, Cosimo Distante, and Abdenour Hadid. 2025. CVPD at QIAS 2025 shared task: An efficient encoder-based approach for islamic inheritance reasoning. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 929–934. ↗
  10. Gagan Bhatia, Hamdy Mubarak, Mustafa Jarrar, George Mikros, Fadi Zaraket, Mahmoud Alhirthani, Mutaz Al-Khatib, Logan Cochrane, Kareem Darwish, Rashid Yahiaoui, et al. 2026. From RAG to agentic RAG for faithful islamic question answering. arXiv preprint arXiv:2601.07528. ↗
  11. Abdessalam Bouchekif, Shahd Gaben, Samer Rashwani, Somaya Eltanbouly, Mutaz Al-Khatib, Heba Sbahi, Mohammed Ghaly, and Emad Mohamed. 2026. MAWARITH: A dataset and benchmark for legal inheritance reasoning with LLMs. arXiv preprint arXiv:2603.07539. ↗
  12. Abdessalam Bouchekif, Samer Rashwani, Emad Soliman Ali Mohamed, Mutaz Alkhatib, Heba Sbahi, Shahd Gaben, Wajdi Zaghouani, Aiman Erbad, and Mohammed Ghaly. 2025a. QIAS 2025: Overview of the shared task on islamic inheritance reasoning and knowledge assessment. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 851–860.
  13. Abdessalam Bouchekif, Samer Rashwani, Heba Sbahi, Shahd Gaben, Mutaz Al-Khatib, and Mohammed Ghaly. 2025b. Assessing large language models on islamic legal reasoning: Evidence from inheritance law evaluation. In Proceedings of The Third Arabic Natural Language Processing Conference, pages 246–257.
  14. Iheb Chaabane, Puneesh Khanna, Suhail Mohmad, Slim Frikha, Shi Hu, Abdalgader Abubaker, Reda Alami, Mikhail Lubinets, Mohamed El Amine Seddik, Hakim Hacid, et al. 2026. Falcon-H1R: Pushing the reasoning frontiers with a hybrid model for efficient test-time scaling. arXiv preprint arXiv:2601.02346. ↗
  15. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. Training verifiers to solve math word problems. CoRR, abs/2110.14168. ↗
  16. DeepSeek AI. 2024. DeepSeek-R1: Incentivizing reasoning capability in large language models. Technical report, DeepSeek AI. Technical Report. find
  17. Eman Elrefai, Mohamed Lotfy Elrefai, and Aml Hassan Esmail. 2025. Gumball at QIAS 2025: Arabic LLM automated reasoning in islamic inheritance. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 953–959, Suzhou, China. Association for Computational Linguistics.
  18. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the MATH dataset. ↗
  19. Shehenaz Hossain and Haithem Afli. 2025. ADAPT–MTU HAI at QIAS2025: Dual-expert LLM fine-tuning and constrained decoding for arabic islamic inheritance reasoning. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 923–928, Suzhou, China. Association for Computational Linguistics.
  20. Ghader Kurdi, Hanan Justanieah, and Hala Justanieah. 2026. Silah at QIAS 2026: Fine-tuning vs. retrieval-augmented generation for islamic inheritance reasoning. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain.
  21. AL-Smadi Mohammad. 2025. QU-NLP at QIAS 2025 shared task: A two-phase llm fine-tuning and retrieval-augmented generation approach for islamic inheritance reasoning. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 892–898.
  22. Mohamed Motasim Hamed, Nada Ghneim, and Riad Sonbol. 2025. HIAST at QIAS 2025: Retrieval-augmented LLMs with top-hit web evidence for arabic islamic reasoning QA. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 883–891, Suzhou, China.
  23. Mohamed Lamine Mouhoub and Chahinez Bouchekif. 2026. PSL at QIAS 2026: Which models perform better in arabic inheritance reasoning? In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain.
  24. Hamdy Mubarak, Rana Malhas, Watheq Mansour, Abubakr Mohamed, Mahmoud Fawzi, Majd Hawasly, Tamer Elsayed, Kareem Mohamed Darwish, and Walid Magdy. 2025. IslamicEval 2025: The first shared task of capturing LLMs hallucination in islamic content. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 480–493.
  25. Yossra Noureldien, Hassan Suliman, Farah Attallah, Abdelrazig Mohamed, and Sara Abdalla. 2025. Athar at QIAS2025: LLM-based question answering systems for islamic inheritance and classical islamic knowledge. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 914–922, Suzhou, China. Association for Computational Linguistics.
  26. Nguyen Xuan Phuc and Thìn Đặng Văn. 2025. PuxAI at QIAS 2025: Multi-agent retrieval-augmented generation for islamic inheritance and knowledge reasoning. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 905–913.
  27. Jihad R’baiti, Chouaib El Hachimi, Youssef Hmamouche, and Amal El Fallah Seghrouchni. 2025. MorAI at QIAS 2025: Collaborative LLM via voting and retrieval-augmented generation for solving complex inheritance problems. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 947–952, Suzhou, China. Association for Computational Linguistics.
  28. Weizhou Shen, Ziyi Yang, Chenliang Li, Zhiyuan Lu, Miao Peng, Huashan Sun, Yingcheng Shi, Shengyi Liao, Shaopeng Lai, Bo Zhang, et al. 2025. QwenLong-L1.5: Post-training recipe for long-context reasoning and memory management. ↗
  29. Hicham Ghazi Sidaoui. 2026. AGS-KSU at QIAS 2026: A comparative study of prompting and LLM approaches for structured islamic inheritance reasoning. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain.
  30. Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, et al. 2025. OpenAI GPT-5 system card. arXiv preprint arXiv:2601.03267. ↗
  31. Wassim Swaileh, Mohammed-En-Nadhir Zighem, Hichem Telli, Salah Eddine Bekhouche, Abdellah Zakaria Sellam, and Fadi Dornaika. 2026. CVPD at QIAS 2026: RAG-guided LLM reasoning for al-mawarith share calculation and heir allocation. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with LREC-COLING 2026, Palma de Mallorca, Spain. find
  32. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, 35:24824–24837.
  33. Jesse Woo, Fateme Hashemi Chaleshtori, Ana Marasović, and Kenneth Marino. 2025. BriefMe: A legal NLP benchmark for assisting with legal briefs. arXiv preprint arXiv:2506.06619. ↗
  34. Osama Farouk Zaki. 2025. CIS-RG at QIAS 2025 shared task: Approaches for enhancing performance of LLM on islamic legal reasoning and its mathematical calculations. In Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 935–939. A. Related Work

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bouchekif-2026-qias-2026-overview-shared-task/