Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning

2026-07-21

AI Summary

Legal machine translation poses significant challenges due to linguistic complexity and stringent accuracy requirements, which conventional neural approaches struggle to address effectively. This work introduces, for the first time, reinforcement learning with verifiable rewards into legal translation and systematically evaluates supervised fine-tuning, reinforcement learning, and state-of-the-art reasoning-augmented models on multilingual Swiss legal texts. Experiments are conducted using relatively small-scale language models—including Qwen3.5 (4B/9B) and Gemma 3 (12B)—enhanced with structured reasoning to improve translation quality. Results demonstrate that reinforcement learning substantially outperforms supervised fine-tuning, enabling compact models to approach the performance of advanced reasoning-based systems, while also revealing diminishing returns from increasing model scale.

Abstract

Neural machine translation (NMT) in the legal domain is a linguistically and conceptually demanding task, primarily due to the complexity of legal language and the high level of precision it requires. The recent emergence of reasoning-capable language models opens new possibilities for tackling such challenges. They add to a set of other previously proposed techniques to enhance the translation quality, which includes supervised fine-tuning and reinforcement learning. In this work, we perform a comparison between these various approaches. More particularly, we evaluate small language models such as Qwen3.5 4B, Qwen3.5 9B, and Gemma 3 12B enhanced with various re-training paradigms and compare their performances against frontier reasoning models. We focus on the Swiss legal system, which -- with its unique multilingual statutes -- offers a particularly challenging testbed for reasoning-augmented models. Our results show that the quality of small ``base'' models can be greatly enhanced, and that reinforcement learning with verifiable rewards can be applied to NMT in the legal domain and surpasses the translation quality of supervised fine-tuning. The performance of enhanced small models is close to the one of state-of-the-art reasoning models yet remains inferior. We also note that re-training paradigms yield diminishing returns as model size increase. The code and models are publicly available at https://github.com/aixiuxiuxiu/Legal-MT-SFT-RL.

Problem

• legal machine translation
• neural machine translation
• translation quality
• reasoning
• multilingual legal statutes

Innovation

• reasoning-augmented translation
• reinforcement learning
• legal machine translation
• structured reasoning
• verifiable rewards