--- title: "ANLPers at QIAS: CoT for Islamic Inheritance (2025)" description: "The paper describes a QIAS 2025 SubTask 1 entry: Arabic multiple-choice questions on dividing an estate under Islamic… 7 references." authors: ["Serry Sibaee", "Mahmoud Reda", "Omer Nacar", "Yasser Alhabashi", "Adel Ammar", "Wadii Boulila"] year: 2025 venue: "Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks" source: "https://aclanthology.org/2025.arabicnlp-sharedtasks.120/" doi: "10.18653/v1/2025.arabicnlp-sharedtasks.120" 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/sibaee-2025-anlpers-qias-cot-islamic-inheritance/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sibaee-2025-anlpers-qias-cot-islamic-inheritance/) [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/17/) / ANLPers at QIAS: CoT for Islamic Inheritance # ANLPers at QIAS: CoT for Islamic Inheritance **Authors:** [Serry Sibaee](https://md.research.elamur.ai/free/en/authors/serry-sibaee/), [Mahmoud Reda](https://md.research.elamur.ai/free/en/authors/mahmoud-reda/), [Omer Nacar](https://md.research.elamur.ai/free/en/authors/omer-nacar/), [Yasser Alhabashi](https://md.research.elamur.ai/free/en/authors/yasser-alhabashi/), [Adel Ammar](https://md.research.elamur.ai/free/en/authors/adel-ammar/) and [Wadii Boulila](https://md.research.elamur.ai/free/en/authors/wadii-boulila/) · **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.120/) · [PDF](https://aclanthology.org/2025.arabicnlp-sharedtasks.120.pdf) · [DOI 10.18653/v1/2025.arabicnlp-sharedtasks.120](https://doi.org/10.18653/v1/2025.arabicnlp-sharedtasks.120) · [OpenAlex](https://openalex.org/W4416037599) **Citations:** 0 citations (OpenAlex, 2026-10-14) ## Summary The paper describes a QIAS 2025 SubTask 1 entry: Arabic multiple-choice questions on dividing an estate under Islamic jurisprudence. Solving them requires understanding the text and then carrying out multi-step fraction arithmetic. The authors compare two prompting setups on Claude 3.7 Sonnet and GPT-4o. In the first, the model returns only an answer letter. In the second, the model works through the solution step by step and ends with a fixed "Final Answer" line, which a regex reads. On the dev set, step-by-step prompting raised Claude from 0.67 to 0.81 and GPT-4o from 0.63 to 0.74. The authors also tried a RAG variant that retrieves solved training examples. It reached only 0.737, so they dropped it. Error analysis shows the hardest point is Tasheeh, the rescaling of shares to whole numbers. The paper's main argument is that prompting alone, with no fine-tuning or retrieval, gives large gains on a complex legal-arithmetic task, and does so cheaply. ## Key points - Task: QIAS 2025 SubTask 1, Islamic inheritance MCQs with options A–F, in Arabic. - Dev scores without and with reasoning: Claude 3.7 went from 0.67 to 0.81, and GPT-4o from 0.63 to 0.74. The authors report these as relative gains of about 21% and 17%. - Test scores with reasoning were noticeably lower than on dev: Claude 3.7 scored 0.71 and GPT-4o 0.55. - RAG setup: - The index lived in ChromaDB, with each chunk made of a question plus an enriched answer. - Three embedders were tested: Arabic-Triplet-Matryoshka-V2, all-MiniLM-L6-v2 and BGE-M3; BGE-M3 retrieved best. - The top-3 chunks were added to the reasoning prompt. - RAG+CoT scored 0.737, below plain CoT. - Running the full benchmark with reasoning cost about 10 USD in API fees. ## Interesting ideas - The authors trace where a reasoning chain breaks. In one case the model correctly computed the corrected estate of 192, yet answered with the share from before correction; the arithmetic was right but the question was mismatched. - Free-form reasoning is kept apart from a rigidly formatted final line that a simple pattern can extract, which makes scoring reliable. - Adding retrieved similar solved cases lowered accuracy compared with plain reasoning. The authors blame reliance on partially relevant context and sensitivity to the choice of embedder. - Tasheeh is a separate, explicit normalization step: shares are moved to a common denominator, then scaled until every heir gets an integer. - Scores fell sharply from dev to test (GPT-4o dropped from 0.74 to 0.55), and the authors report both sets of numbers. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (answer): Before giving an answer, check that the chosen value fills the exact slot the question targets (here, corrected vs. original total), not just any value derived along the way. - **I2** (learn): Ask LLM teacher tiers for a fixed, machine-checkable verdict line after their explanation, so silver labels can be harvested without fragile parsing. - **I3** (learn): Treat every added component as a hypothesis and test it against the simpler baseline as a control, keeping it only when it beats that baseline. - **I4** (world): Represent holdings and divided quantities as fractions with an explicit, auditable rescaling step, so questions about distributions can be answered and explained from the stored state. - **I5** (learn): Always report held-out results next to tuning-set results, so gains from prompt or rule tweaks are not overstated. *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 - [Prince Sultan University](https://md.research.elamur.ai/free/en/universities/prince-sultan-university/) — University, Riyadh, Saudi Arabia · 20 works · 120 citations - [Zagazig University](https://md.research.elamur.ai/free/en/universities/zagazig-university/) — University, Zagazig, Egypt · 4 works · 51 citations - [Tuwaiq Academy](https://md.research.elamur.ai/free/en/universities/tuwaiq-academy/) — University, Riyadh, Saudi Arabia · 13 works · 7 citations 📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,954 works · [🤝 Political Science and International Relations](https://md.research.elamur.ai/free/en/shelves/social-sciences/political-science-and-international-relations/) · 206 works 🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works ## Cite Cite as: Serry Sibaee, Mahmoud Reda, Omer Nacar, Yasser Alhabashi, Adel Ammar and Wadii Boulila. 2025. ANLPers at QIAS: CoT for Islamic Inheritance. 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{sibaee2025anlpers, title = {{ANLPers at QIAS: CoT for Islamic Inheritance}}, author = {Serry Sibaee, Mahmoud Reda, Omer Nacar, Yasser Alhabashi, Adel Ammar and Wadii Boulila}, booktitle = {Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks}, year = {2025}, doi = {10.18653/v1/2025.arabicnlp-sharedtasks.120}, url = {https://doi.org/10.18653/v1/2025.arabicnlp-sharedtasks.120} } ``` - [← previous ANLPers at MAHED2025: From Hate to Hope: Boosting Arabic Text Classification](https://md.research.elamur.ai/free/en/papers/alhabashi-2025-anlpers-mahed2025-hate-hope-boosting/) - [next → Annotating English Verb-Argument Structure via Usage-Based Analogy](https://md.research.elamur.ai/free/en/papers/hsiao-2025-annotating-english-verb-argument-structure/) - [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/17/) ## References 7 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Hazem Abdelazim, Tony Begemy, Ahmed Galal, Hala Sedki, and Ali Mohamed. 2024. Multi-hop arabic llm reasoning in complex qa. Procedia Computer Science, 244:66–75. 6th International Conference on AI in Computational Linguistics. [find](https://scholar.google.com/scholar?q=Hazem+Abdelazim%2C+Tony+Begemy%2C+Ahmed+Galal%2C+Hala+Sedki%2C+and+Ali+Mohamed%2E+2024%2E+Multi%2Dhop+arabic+llm+reasoning+in+complex+qa%2E+Procedia+Computer+Science%2C+244%3A66%E2%80%9375%2E+6th+International+Conference+on+AI+in+) 2. Manswab Mahsen Abdulrahman and Abdul Hafiz Musa Walusimbi. 2024. Evaluating the use of artificial intelligence for issuing fatwas in islamic inheritance cases: A juristic study with a comparison to gpt-3.5. Asy-Syari’ah, 26(2):121–146. [find](https://scholar.google.com/scholar?q=Manswab+Mahsen+Abdulrahman+and+Abdul+Hafiz+Musa+Walusimbi%2E+2024%2E+Evaluating+the+use+of+artificial+intelligence+for+issuing+fatwas+in+islamic+inheritance+cases%3A+A+juristic+study+with+a+comparison+to+gp) 3. Ahmet Yusuf Alan, Enis Karaarslan, and Ömer Aydın. 2025. [Improving llm reliability with rag in religious question-answering: Mufassirqas](https://md.research.elamur.ai/free/en/papers/bouchekif-2025-qias-2025-overview-shared-task/). Turkish Journal of Engineering, 9(3):544–559. 4. 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 (Arabic-NLP 2025), Suzhou, China. Association for Computational Linguistics. 5. Serry Sibaee, Omer Nacar, Adel Ammar, Yasser Al-Habashi, Abdulrahman Al-Batati, and Wadii Boulila. 2025. [From guidelines to practice: A new paradigm for arabic language model evaluation](https://md.research.elamur.ai/free/en/papers/sibaee-2025-guidelines-practice-new-paradigm-arabic/). Preprint, arXiv:2506.01920. [↗](https://arxiv.org/abs/2506.01920) 6. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). Preprint, arXiv:2201.11903. [↗](https://arxiv.org/abs/2201.11903) 7. 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+%2DComputer+and+Information+Sciences%2C+33%281%29%3A68%E2%80%9376%2E) *Updated 2 November 2026*