Rule-Based Enigmas: Enhancing Complex Task Reasoning in Large Language Models Through Constrained Frameworks

ACL ARR 2025 February Submission8379 Authors

Abstract: This paper investigates the ability of large language models (LLMs) to solve complex tasks under strict rule-based constraints. Focusing on enhancing the reasoning capabilities of LLMs, it proposes an innovative framework that combines cognitive learning and knowledge-guided optimization to improve task completion and traceability. The research introduces a benchmark dataset that integrates multi-domain tasks, explicit rules, and traceable question-answer pairs to evaluate LLMs performance in constrained problem-solving scenarios, requiring creative responses. Empirical experiments demonstrate that the proposed framework significantly enhances LLMs reasoning consistency, knowledge completeness, and adherence to rules. This study provides useful insights for improving the effectiveness of LLMs in tackling real-world challenges, where problem-solving often involves navigating complex constraints and innovative solutions.

Paper Type: Long

Research Area: Question Answering

Research Area Keywords: Large Language Models (LLMs)，Question Answering

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Approaches to low-resource settings, Approaches low compute settings-efficiency, Data resources, Data analysis

Languages Studied: English

Submission Number: 8379