---
title: "Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.19181v1"
permalink: "https://feed.astralor.com/items/itm_843e322486a6f3462ac80cc1"
published: "2026-07-21T15:15:56.000Z"
score: 77
channel: "tech"
tags: ["Gemma 3","Qwen3.5","Reinforcement Learning","Legal Machine Translation","Switzerland","Supervised Fine-Tuning"]
---

# Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning

## AI commentary

> This paper stands out for its rigorous empirical comparison in the legal AI niche, showing that small models plus RL can approach frontier performance without heavy compute. Teams building practical machine translation systems for law will find this directly actionable.

## At a glance

This paper compares supervised fine-tuning and reinforcement learning for legal machine translation using small language models (Qwen3.5 4B/9B, Gemma 3 12B) on Swiss multilingual statutes. Results show RL with verifiable rewards outperforms SFT, and enhanced small models approach frontier reasoning models, though gains diminish with larger models. Code and models are open-sourced.

## Key points

- Comparative study of supervised fine-tuning vs. reinforcement learning for legal NMT
- Reinforcement learning with verifiable rewards substantially boosts translation quality of small models
- Swiss multilingual legal system serves as a uniquely challenging benchmark
- Diminishing returns from retraining as model size increases; enhanced small models approach frontier reasoning models

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_843e322486a6f3462ac80cc1