Can Group Relative Policy Optimization Improve Thai Legal Reasoning and Question Answering?

TL;DR

This paper introduces Group-Relative Policy Optimization (GRPO), a novel training approach that significantly improves Thai legal question answering by enhancing citation accuracy and reasoning robustness in language models, while reducing computational costs.

Contribution

The paper presents GRPO, a new method that aligns LLMs for better legal reasoning and citation accuracy using semantic rewards, outperforming traditional instruction tuning.

Findings

Achieves up to 90% citation-F1 improvement

Increases joint quality metrics by 31%

Reduces computational costs by 2.5x

Abstract

The Retrieval-Augmented Generation (RAG) systems' performance on Thai legal question answering is still limited, especially for questions requiring extensive, complex legal reasoning. To address these limitations, we introduce an approach aligning LLMs toward improved law citation accuracy and better response quality using Group-Relative Policy Optimization (GRPO). Our approach leverages BGE-M3 embeddings as a cost-efficient semantic-similarity reward, significantly reducing computational expenses up to 2.5x compared to large language model judges. Experiments on the NitiBench benchmark demonstrate substantial improvements: GRPO achieves up to 90% citation-F1 gains from the base model and a 31% increase in joint quality metrics over instruction tuning. Crucially, our method shows enhanced robustness on complex legal reasoning tasks compared to instruction tuning, providing an effective and resource-efficient solution for enhancing Thai legal LLMs.