Year: 2025
Type: conference-paper
Abstract: This paper presents a Chain-of-Thought (CoT) prompting approach for Islamic inheritance reasoning in multiple-choice question answering.We address the QIAS 2025 SubTask 1, which requires complex legal reasoning to determine correct inheritance shares according to Islamic jurisprudence.Our system employs two prompting strategies: direct answer extraction and step-by-step reasoning with regexbased answer extraction.We evaluate our approach using Claude 3.7 Sonnet and GPT-4o on Islamic inheritance MCQ tasks.Results demonstrate significant performance improvements when incorporating the thinking step: Claude 3.7 improved from 0.67 to 0.81, and GPT-4o from 0.63 to 0.74.Error analysis reveals that while models perform well in basic reasoning, they struggle with complex correction procedures (Tasheeh 1 ) in inheritance calculations.Our findings confirm that structured reasoning substantially enhances LLM performance on complex Arabic legal reasoning tasks without requiring additional training or retrieval-augmented generation.
Institutions: Prince Sultan University, Zagazig University, Tuwaiq Academy
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037599