MARCO-Law: Marginal-Aware Reinforcement Learning for Legal Tool Orchestration

ACL ARR 2025 May Submission7663 Authors

Abstract: Large-scale pretrained language models (LLMs) have achieved significant advances in natural language tasks, yet challenges persist in legal applications that demand high precision. Current approaches enhance model performance through long-chain reasoning and tool invocation, but often struggle with excessive resource consumption and suboptimal tool integration. To address these issues, this paper proposes a reinforcement learning-based framework for multi-tool collaborative invocation. The framework dynamically optimizes tool usage across multiple iterations, selecting tools and refining search terms based on marginal benefits, ensuring the execution of the most effective analysis strategy. Experimental results show that the proposed method improves both the accuracy of legal question answering and resource utilization, demonstrating the potential of multi-tool collaboration and adaptive strategy adjustment in the legal domain.

Paper Type: Short

Research Area: Efficient/Low-Resource Methods for NLP

Research Area Keywords: data-efficient training, NLP in resource-constrained settings

Contribution Types: NLP engineering experiment

Languages Studied: English

Submission Number: 7663