Elamur Research⌕
☰
home / papers / 2026 / AGS-KSU at QIAS 2026: A Comparative Study of Prompting and LLM Approaches for Structured Islamic Inheritance Reasoning

AGS-KSU at QIAS 2026: A Comparative Study of Prompting and LLM Approaches for Structured Islamic Inheritance Reasoning

Paper · 2026 · Hicham Ghazi Sidaoui

Summary

The paper addresses the QIAS 2026 shared task. Systems must solve Arabic Islamic inheritance (Ilm al-Faraid) cases end to end and output structured JSON covering eligible heirs, blocked heirs, fractional shares, awl/radd adjustment and the final distribution. The data come from the MAWARITH benchmark: 12,500 cases, with 200 annotated dev items and 500 test items. Scoring uses MIR-E, which evaluates each stage separately. The author compares four submitted configurations:

  • a Qwen2.5-3B-Instruct model fine-tuned with QLoRA on a laptop GPU;
  • a three-call Fanar-Sadiq pipeline followed by deterministic validation and repair;
  • two GPT-5.4 prompting variants.

The best run gave GPT-5.4 a rule block for eight difficult case types plus 20 solved dev examples per batch. It reached 0.84 MIR-E, against 0.76 for a bare instruction. Only the prompt differed between these two runs, so this pair gives the cleanest evidence that rules and demonstrations help. Differences across model families are confounded by other factors. Error analysis shows that remaining failures concentrate in remainder allocation, especially grandmother blocking and unwarranted radd. The paper is a practical reference for LLM-based structured legal reasoning in Arabic.

Key points

  • Official MIR-E scores: Qwen2.5-3B + QLoRA 0.30, Fanar-Sadiq pipeline 0.46, GPT-5.4 minimal prompt 0.76, GPT-5.4 with rules and dev examples 0.84.
  • The Qwen baseline was trained on 2,000 examples on an RTX 4060 (8GB) in about 5.5 hours. It often ran out of generation budget before emitting valid JSON, which explains its low score.
  • The full prompt's rules covered cases such as grandfather–sibling shares, al-Akdariyyah, al-Mushtarakah and the Umariyyatan rule. Dev demonstrations rotated by a fixed circular offset and were used only at inference, with no weight updates.
  • Exact-match diagnostics for the 0.84 system:
    • awl_or_radd: 93.6%
    • heirs: 70.0%
    • blocked heirs: 69.0%
    • normalized distribution: 71.0%
    • all four jointly: 56.8%
  • The largest error categories were grandmother-line errors (63 cases, 12.6%), false radd (37), awl_or_radd mismatches (32) and collateral continuation errors (20). Code, prompts and predictions are available at github.com/xAGS1/qias2026-ags-ksu.

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

🎓King Saud UniversityRiyadh · Saudi Arabia38 works · 527 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksPolitical Science and International Relations · 206 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,894 worksReasoning · 1,685 worksSemitic languages · 3,066 works

Cite

Cite as Hicham Ghazi Sidaoui. 2026. AGS-KSU at QIAS 2026: A Comparative Study of Prompting and LLM Approaches for Structured 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 10.63317/2irfqjkfpyyn

@inproceedings{sidaoui2026ags,
  title = {{AGS-KSU at QIAS 2026: A Comparative Study of Prompting and LLM Approaches for Structured Islamic Inheritance Reasoning}},
  author = {Hicham Ghazi Sidaoui},
  booktitle = {The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks},
  year = {2026},
  doi = {10.63317/2irfqjkfpyyn},
  url = {https://doi.org/10.63317/2irfqjkfpyyn}
}

References

10 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Hanan Abbas et al. 2025. Fanar: A large language model for arabic. Technical report, Qatar Computing Research Institute. find
  2. Mohammad AL-Smadi. 2025. QU-NLP at QIAS 2025 shared task: A two-phase LLM fine-tuning and retrieval-augmented generation approach for islamic inheritance reasoning. arXiv preprint arXiv:2508.15854. ↗
  3. Nouar AlDahoul and Yasir Zaki. 2025. Benchmarking the legal reasoning of LLMs in arabic islamic inheritance cases. arXiv preprint arXiv:2508.15796. ↗
  4. Salah Eddine Bekhouche et al. 2025. CVPD at QIAS 2025 shared task: An efficient encoder-based approach for islamic inheritance reasoning. arXiv preprint arXiv:2509.00457. ↗
  5. 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. ↗
  6. 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.
  7. 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.
  8. Tom B. Brown et al. 2020. Language models are few-shot learners. In Advances in Neural Information Processing Systems, volume 33, pages 1877–1901. ↗
  9. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. QLoRA: Efficient finetuning of quantized LLMs. In Advances in Neural Information Processing Systems, volume 36.
  10. Qwen Team. 2024. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sidaoui-2026-ags-ksu-qias-2026-comparative/