---
title: "ADAPT–MTU HAI at QIAS2025: Dual-Expert LLM Fine-Tuning"
description: "This paper presents a system entered in the QIAS 2025 shared task, specifically its first subtask. 30 references, cited by 2 papers here."
authors: ["Shehenaz Hossain", "Haithem Afli"]
year: 2025
venue: "Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks"
source: "https://aclanthology.org/2025.arabicnlp-sharedtasks.127/"
doi: "10.18653/v1/2025.arabicnlp-sharedtasks.127"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/hossain-2025-adapt-mtu-hai-qias2025-dual/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/hossain-2025-adapt-mtu-hai-qias2025-dual/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/13/) / ADAPT–MTU HAI at QIAS2025: Dual-Expert LLM Fine-Tuning and Constrained Decoding for Arabic Islamic Inheritance Reasoning

# ADAPT–MTU HAI at QIAS2025: Dual-Expert LLM Fine-Tuning and Constrained Decoding for Arabic Islamic Inheritance Reasoning

**Authors:** [Shehenaz Hossain](https://md.research.elamur.ai/free/en/authors/shehenaz-hossain/) and [Haithem Afli](https://md.research.elamur.ai/free/en/authors/haithem-afli/) · **Year:** 2025 · **Venue:** [Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2025](https://md.research.elamur.ai/free/en/venues/arabicnlp/2025/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2025.arabicnlp-sharedtasks.127/) · [PDF](https://aclanthology.org/2025.arabicnlp-sharedtasks.127.pdf) · [DOI 10.18653/v1/2025.arabicnlp-sharedtasks.127](https://doi.org/10.18653/v1/2025.arabicnlp-sharedtasks.127) · [OpenAlex](https://openalex.org/W4416037643)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [2](https://md.research.elamur.ai/free/en/papers/hossain-2025-adapt-mtu-hai-qias2025-dual/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
This paper presents a system entered in the QIAS 2025 shared task, specifically its first subtask. The task is Arabic multiple-choice questions on Islamic inheritance law, with six options (A–F) and one correct answer. Solving the cases means applying fixed fractional shares, exclusion (ḥajb), redistribution (radd) and proportional reduction (awl), and doing the arithmetic correctly. The authors fine-tune ALLaM-7B-Instruct-preview with LoRA on the 20,000 official training questions and run it next to the unmodified base model. Both models get the same four-shot Arabic prompt. Its exemplars were picked to cover eligibility, fixed shares, residuary heirs and LCM normalisation. A custom logits processor and greedy decoding force each model to output exactly one letter from A–F. An arbitration rule merges the two predictions and sides with the fine-tuned model when they disagree. The system reaches 60.0% on dev and 54.7% on the hidden test set, against 42.9% reported elsewhere for zero-shot ALLaM-7B. Error analysis shows that share arithmetic and numerically close distractors cause most failures. The authors conclude that symbolic share calculators are the natural next step.

## Key points
- Data: the official QIAS 2025 Subtask 1 set has 20k training MCQs, 1k dev and 1k hidden test items. Questions are split into Beginner and Advanced levels and cover heir eligibility, share fractions and money allocation.
- Fine-tuning setup:
  - LoRA on q_proj/k_proj/v_proj/o_proj with r = 16, α = 32 and dropout 0.05.
  - CLM objective over the prompt plus the gold letter.
  - 5 epochs, effective batch 16, LR 3×10⁻⁵ with cosine decay, max length 512, bf16.
  - Hardware and software: one A100 80GB, Hugging Face Transformers and PEFT.
- Decoding: max_new_tokens=1, temperature 0 and do_sample=False. Every vocabulary token except the six option letters is masked out.
- Results: 60.0% dev and 54.7% test, versus a 42.9% zero-shot ALLaM-7B baseline reported by Bouchekif et al. (2025b). This is still below the leaderboard top.
- Error taxonomy: eligibility/ḥajb mistakes, fraction errors (especially awl), wrong radd redistribution, and confusion by near-miss numeric options. Fraction and numeric errors are the majority, and they are worst on Advanced items.

## Interesting ideas
- Restricting the output vocabulary to the valid labels makes format errors impossible by construction, so accuracy measures only the reasoning.
- Few-shot exemplars were chosen by rule type rather than at random, giving one example each for exclusion, fixed share, residuary and LCM normalisation.
- Classifying errors by the legal operation that failed showed the bottleneck is arithmetic, not language understanding, which points to handing the computation to a symbolic calculator.
- As described, the arbitration returns the fine-tuned answer whether the two models agree or disagree, so the base model never changes the output; a combination rule should be checked for whether it actually does anything.
- Distractors that are numerically close to the right answer exposed small slips in intermediate steps that rough answers would hide.

## For Mova

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

- **I1** (answer): when a school-test reader must return an option, it can pick from a closed set of option IDs scored against the world of the text, so the answer can never be malformed.
- **I2** (learn): prompts for the LLM teacher tiers could include one exemplar per construction or frame type, so the silver labels cover the space instead of clustering on common cases.
- **I3** (world): fractions, shares and measures in maths word problems should be computed by explicit world-state operations, and errors should be logged by which operation broke.
- **I4** (learn): for witness-parser agreement, report how often each witness changes the final silver tree, as a negative control, and abstain or flag on disagreement instead of quietly defaulting to one parser.
- **I5** (answer): build maths word-problem tests with near-miss numeric options to check that the solver's chain of world updates is exact, not just approximately right.

*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

- [Munster Technological University](https://md.research.elamur.ai/free/en/labs/munster-technological-university/) — Research institute, Cork, Ireland · 4 works · 11 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [🤝 Political Science and International Relations](https://md.research.elamur.ai/free/en/shelves/social-sciences/political-science-and-international-relations/) · 182 works

🧭 Topics (3): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,320 works · [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,426 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Shehenaz Hossain and Haithem Afli. 2025. ADAPT–MTU HAI at QIAS2025: Dual-Expert LLM Fine-Tuning and Constrained Decoding for Arabic Islamic Inheritance Reasoning. Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{hossain2025adapt,
  title = {{ADAPT–MTU HAI at QIAS2025: Dual-Expert LLM Fine-Tuning and Constrained Decoding for Arabic Islamic Inheritance Reasoning}},
  author = {Shehenaz Hossain and Haithem Afli},
  booktitle = {Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks},
  year = {2025},
  doi = {10.18653/v1/2025.arabicnlp-sharedtasks.127},
  url = {https://doi.org/10.18653/v1/2025.arabicnlp-sharedtasks.127}
}
```

- [← previous Adapting Falcon3-7B Language Model for Arabic: Methods, Challenges, and Outcomes](https://md.research.elamur.ai/free/en/papers/boussaha-2025-adapting-falcon3-7b-language-model/)
- [next → Ahasis Shared Task: Hybrid Lexicon-Augmented AraBERT Model for Sentiment Detection in Arabic Dialects](https://md.research.elamur.ai/free/en/papers/ibrahim-2025-ahasis-shared-task-hybrid-lexicon/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/13/)

## References

30 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Zubair Abbasi. 2025. Augmented learning: Generative artificial intelligence and islamic inheritance law. Islamic Law Blog Roundtable Essay. [find](https://scholar.google.com/scholar?q=Zubair+Abbasi%2E+2025%2E+Augmented+learning%3A+Generative+artificial+intelligence+and+islamic+inheritance+law%2E+Islamic+Law+Blog+Roundtable+Essay%2E)
2. Elnaserledinellah Mahmood Abdelwahab, Karim Dagh-bouche, and Nadra Ahmad Shannan. 2016. The algorithm of islamic jurisprudence (fiqh) with validation of an entscheidungsproblem. Preprint, arXiv:1604.00266. [↗](https://arxiv.org/abs/1604.00266)
3. Salako Taofiki Ajani, Bhasah Abu Bakar, and Mikail Ibrahim. 2013. The value of islamic inheritance in consolidation of the family financial stability. IOSR Journal of Humanities and Social Science, 8(3). [↗](https://doi.org/10.9790/0837-0831523)
4. Hayfa Aleid and Aqil Azmi. 2025. Hajj-fqa: A benchmark arabic dataset for developing question-answering systems on hajj fatwas. Journal of King Saud University Computer and Information Sciences, 37:1–28. [↗](https://doi.org/10.1007/s44443-025-00128-w)
5. Sarah Alnefaie, Eric Atwell, and Mohammad Ammar Alsalka. 2023a. [HAQA and QUQA: Constructing two Arabic question-answering corpora for the Quran and Hadith](https://md.research.elamur.ai/free/en/papers/alnefaie-2023-haqa-quqa-constructing-two-arabic/). In Proceedings of the 14th International Conference on Recent Advances in Natural Language Processing, pages 90–97, Varna, Bulgaria. INCOMA Ltd., Shoumen, Bulgaria. on islamic inheritance reasoning and knowledge assessment. In Proceedings of The Second Arabic Natural Language Processing Conference, ArabicNLP 2025, Suzhou, China, November 5–9, 2025. Association for Computational Linguistics.
6. 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](https://md.research.elamur.ai/free/en/papers/bouchekif-2025-assessing-large-language-models-islamic/). In Proceedings of The Second Arabic Natural Language Processing Conference (ArabicNLP 2025), Suzhou, China. Association for Computational Linguistics.
7. Mahmoud Halima Chebet, Joseph Orero, and Anthony Luvanda. 2014. A knowledgebase model for islamic inheritance. [find](https://scholar.google.com/scholar?q=Mahmoud+Halima+Chebet%2C+Joseph+Orero%2C+and+Anthony+Luvanda%2E+2014%2E+A+knowledgebase+model+for+islamic+inheritance%2E)
8. Gael de Francony, Victor Guichard, Praveen Joshi, Haithem Afli, and Abdessalam Bouchekif. 2019. [Hierarchical deep learning for Arabic dialect identification](https://md.research.elamur.ai/free/en/papers/francony-2019-hierarchical-deep-learning-arabic-dialect/). In Proceedings of the Fourth Arabic Natural Language Processing Workshop, pages 249–253, Florence, Italy. Association for Computational Linguistics.
9. Shehenaz Hossain, Fouad Shammary, Bahaulddin Shammary, and Haithem Afli. 2025. [Enhancing dialectal Arabic intent detection through cross-dialect multilingual input augmentation](https://md.research.elamur.ai/free/en/papers/hossain-2025-enhancing-dialectal-arabic-intent-detection/). In Proceedings of the 4th Workshop on Arabic Corpus Linguistics (WACL-4), pages 44–49, Abu Dhabi, UAE. Association for Computational Linguistics.
10. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. [Lora: Low-rank adaptation of large language models](https://md.research.elamur.ai/free/en/papers/hu-2021-lora-low-rank-adaptation-large/). Preprint, arXiv:2106.09685. [↗](https://arxiv.org/abs/2106.09685)
11. Sarah Alnefaie, Eric Atwell, and Mohammad Ammar Alsalka. 2023b. Is GPT-4 a good islamic expert for answering Quran questions? In Proceedings of the 35th Conference on Computational Linguistics and Speech Processing (ROCLING 2023), pages 124–133, Taipei City, Taiwan. The Association for Computational Linguistics and Chinese Language Processing (ACLCLP). [find](https://scholar.google.com/scholar?q=Sarah+Alnefaie%2C+Eric+Atwell%2C+and+Mohammad+Ammar+Alsalka%2E+2023b%2E+Is+GPT%2D4+a+good+islamic+expert+for+answering+Quran+questions%3F+In+Proceedings+of+the+35th+Conference+on+Computational+Linguistics+and+Spe)
12. M Saiful Bari, Yazeed Alnumay, Norah A. Alzahrani, Nouf M. Alotaibi, Hisham A. Alyahya, Sultan Al-Rashed, Faisal A. Mirza, Shaykhah Z. Alsubaie, Hassan A. Alahmed, Ghadah Alabduljabbar, Raghad Alkhathran, Yousef Almushayqih, Raneem Alnajim, Salman Alsubaihi, Maryam Al Mansour, Majed Alrubaian, Ali Alammari, Zaki Alawami, Abdulmohsen Al-Thubaity, and 6 others. 2024. [Allam: Large language models for arabic and english](https://md.research.elamur.ai/free/en/papers/bari-2024-allam-large-language-models-arabic/). Preprint, arXiv:2407.15390. [↗](https://arxiv.org/abs/2407.15390)
13. Abdessalam Bouchekif, Samer Rashwani, Mohammed Ghaly, Mutaz Al-Khatib, Emad Mohamed, Wajdi Zaghouani, Heba Sbahi, Shahd Gaben, and Aiman Erbad. 2025a. Qias 2025: Overview of the shared task Huang Huang, Fei Yu, Jianqing Zhu, Xuening Sun, Hao Cheng, Song Dingjie, Zhihong Chen, Mosen Alharthi, Bang An, Juncai He, Ziche Liu, Junying Chen, Jianquan Li, Benyou Wang, Lian Zhang, Ruoyu Sun, Xiang Wan, Haizhou Li, and Jinchao Xu. 2024. [AceGPT, localizing large language models in Arabic](https://md.research.elamur.ai/free/en/papers/huang-2024-acegpt-localizing-large-language-models/). In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 8139–8163, Mexico City, Mexico. Association for Computational Linguistics.
14. Rana Malhas, Watheq Mansour, and Tamer Elsayed. 2022. [find](https://scholar.google.com/scholar?q=Rana+Malhas%2C+Watheq+Mansour%2C+and+Tamer+Elsayed%2E+2022%2E)
15. Qur’an QA 2022: [Overview of the first shared task on question answering over the holy qur’an](https://md.research.elamur.ai/free/en/papers/malhas-2022-qur-qa-2022-overview-first/). 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 79–87, Marseille, France. European Language Resources Association.
16. Rana Malhas, Watheq Mansour, and Tamer Elsayed. 2023. [find](https://scholar.google.com/scholar?q=Rana+Malhas%2C+Watheq+Mansour%2C+and+Tamer+Elsayed%2E+2023%2E)
17. Qur’an QA 2023 [shared task: Overview of passage retrieval and reading comprehension tasks over the holy qur’an](https://md.research.elamur.ai/free/en/papers/malhas-2023-qur-qa-2023-shared-task/). In Proceedings of ArabicNLP 2023, pages 690–701, Singapore (Hybrid). Association for Computational Linguistics.
18. Marryam Mohammed, Sama Ali, Salma Khaled, Ayad Majeed, and Ensaf Mohamed. 2025. Aftina: enhancing stability and preventing hallucination in ai-based islamic fatwa generation using llms and rag. Neural Computing and Applications, pages 1–26. [↗](https://doi.org/10.1007/s00521-025-11229-y)
19. Omar T Mohammedi. 2012. Sharia-complaint wills; principles, recognition, and enforcement. NYL Sch. L. Rev., 57:259. Basel Mousi, Nadir Durrani, Fatema Ahmad, Md. [find](https://scholar.google.com/scholar?q=Omar+T+Mohammedi%2E+2012%2E+Sharia%2Dcomplaint+wills%3B+principles%2C+recognition%2C+and+enforcement%2E+NYL+Sch%2E+L%2E+Rev%2E%2C+57%3A259%2E+Basel+Mousi%2C+Nadir+Durrani%2C+Fatema+Ahmad%2C+Md%2E)
20. Arid Hasan, Maram Hasanain, Tameem Kabbani, Fahim Dalvi, Shammur Absar Chowdhury, and Firoj Alam. 2024. [Aradice: Benchmarks for dialectal and cultural capabilities in llms](https://md.research.elamur.ai/free/en/papers/mousi-2025-aradice-benchmarks-dialectal-cultural-capabilities/). Preprint, arXiv:2409.11404. [↗](https://arxiv.org/abs/2409.11404)
21. Fanar Team, Ummar Abbas, Mohammad Shahmeer Ahmad, Firoj Alam, Enes Altinisik, Ehsannedin Asgari, Yazan Boshmaf, Sabri Boughorbel, Sanjay Chawla, Shammur Chowdhury, Fahim Dalvi, Kareem Darwish, Nadir Durrani, Mohamed Elfeky, Ahmed Elmagarmid, Mohamed Eltabakh, Masoomali Fatehkia, Anastasios Fragkopoulos, Maram Hasanain, and 23 others. 2025. [Fanar: An arabic-centric multimodal generative ai platform](https://md.research.elamur.ai/free/en/papers/abbas-2025-fanar-arabic-centric-multimodal-generative/). Preprint, arXiv:2501.13944. [↗](https://arxiv.org/abs/2501.13944)
22. Samia Zouaoui and Khaled Rezeg. 2021. Islamic inheritance calculation system based on arabic ontology (arafamonto). Journal of King Saud University - Computer and Information Sciences, 33(1):68–76. [find](https://scholar.google.com/scholar?q=Samia+Zouaoui+and+Khaled+Rezeg%2E+2021%2E+Islamic+inheritance+calculation+system+based+on+arabic+ontology+%28arafamonto%29%2E+Journal+of+King+Saud+University+%2D+Computer+and+Information+Sciences%2C+33%281%29%3A68%E2%80%9376%2E)
23. Busari Muhammad. 2020. The Islamic Law of Inheritance: Introduction and Theories. [find](https://scholar.google.com/scholar?q=Busari+Muhammad%2E+2020%2E+The+Islamic+Law+of+Inheritance%3A+Introduction+and+Theories%2E)
24. Shabaz Patel, Hassan Kane, and Rayhan Patel. 2023. Building domain-specific llms faithful to the islamic worldview: Mirage or technical possibility? Preprint, arXiv:2312.06652. [↗](https://arxiv.org/abs/2312.06652)
25. Faiza Qamar, Seemab Latif, and Rabia Latif. 2024. A benchmark dataset with larger context for non-factoid question answering over islamic text. Preprint, arXiv:2409.09844. [↗](https://arxiv.org/abs/2409.09844)
26. Zhaozhi Qian, Faroq Altam, Muhammad Alqurishi, and Riad Souissi. 2024. [Cameleval: Advancing culturally aligned arabic language models and benchmarks](https://md.research.elamur.ai/free/en/papers/qian-2024-cameleval-advancing-culturally-aligned-arabic/). Preprint, arXiv:2409.12623. [↗](https://arxiv.org/abs/2409.12623)
27. Siti Fatimah Abdul Rahman, Abdul Malek Yaakob, Ahmad Adnan Fadzil, and MS Shaban. 2017. Asset distribution among the qualified heirs based on islamic inheritance law. Contemporary Issues on Zakat Waqf and Islamic Philanthropy, pages 465–474. [find](https://scholar.google.com/scholar?q=Siti+Fatimah+Abdul+Rahman%2C+Abdul+Malek+Yaakob%2C+Ahmad+Adnan+Fadzil%2C+and+MS+Shaban%2E+2017%2E+Asset+distribution+among+the+qualified+heirs+based+on+islamic+inheritance+law%2E+Contemporary+Issues+on+Zakat+Waqf)
28. Zouaoui Samia and Rezeg Khaled. 2018. Islamic inheritance calculation system based on arabic ontology (arafamonto). Journal of King Saud University - Computer and Information Sciences, 33. [↗](https://doi.org/10.1016/j.jksuci.2018.11.015)
29. Neha Sengupta, Sunil Kumar Sahu, Bokang Jia, Satheesh Katipomu, Haonan Li, Fajri Koto, William Marshall, Gurpreet Gosal, Cynthia Liu, Zhiming Chen, Osama Mohammed Afzal, Samta Kamboj, Onkar Pandit, Rahul Pal, Lalit Pradhan, Zain Muhammad Mujahid, Massa Baali, Xudong Han, Sondos Mahmoud Bsharat, and 13 others. 2023. [Jais and jais-chat: Arabic-centric foundation and instruction-tuned open generative large language models](https://md.research.elamur.ai/free/en/papers/sengupta-2023-jais-jais-chat-arabic-centric/). Preprint, arXiv:2308.16149. [↗](https://arxiv.org/abs/2308.16149)
30. Sadia Tabassum, A. Hoque, Sharaban Twahura, and Mohammad Rahman. 2019. Developing an islamic farayez system applying software engineering. Jurnal Kejuruteraan, 31:25–38. [↗](https://doi.org/10.17576/jkukm-2019-31(1)-04)

*Updated 1 November 2026*