arXiv cs.CL· Kunquan Li, Yingxue Zhang, Zhibin Lan, Fandong Meng, Jinsong Su·· 12 小时前AI 评分35
ReflectMT：将反思内化，实现高效高质量机器翻译
ReflectMT: Internalizing Reflection for Efficient and High-Quality Machine Translation
AI 导读
ReflectMT 提出两阶段反思内化框架，采用“后思考”范式，通过强化学习培养模型的“翻译-反思-改进”能力，并将反思所得知识内化。在 WMT24 上，其基于 GPT-4o 的 GRF 得分为 86.16，超过 DeepSeek-R1 的 84.03 达 2.13 分，同时输出 token 减少约 92%。独立自动评测、盲测人工评估及另外三个翻译方向的初步实验进一步验证了该框架的有效性。
正文 · 原文
Abstract:Recent years have witnessed growing interest in applying Large Reasoning Models (LRMs) to Machine Translation (MT). While most approaches adopt a 'pre-thinking' paradigm and benefit from explicit reasoning trajectories, they suffer from substantial inference cost and latency. To address these limitations, we propose ReflectMT, a two-stage reflection internalization framework for machine translation that employs a 'post-thinking' paradigm. Our approach develops the model's 'translate-reflect-refine' capability through reinforcement learning. In the first stage, we cultivate the model's capacity for high-quality reflection and refinement, thereby enhancing its semantic comprehension and task-specific knowledge. In the second stage, we train the model to internalize the knowledge acquired during reflection. On WMT24, ReflectMT achieves a GPT-4o-based GRF score of 86.16, outperforming DeepSeek-R1 (84.03) by 2.13 points, while using approximately 92% fewer output tokens. Independent automatic evaluation, a blinded human evaluation, and preliminary experiments on three additional translation directions provide further evidence for the framework's effectiveness.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2604.19144 [cs.CL] |
| (or arXiv:2604.19144v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2604.19144 arXiv-issued DOI via DataCite |
Submission history
From: Kunquan Li [view email]
[v1]
Tue, 21 Apr 2026 06:48:41 UTC (342 KB)
[v2]
Mon, 2 Nov 2026 07:16:55 UTC (697 KB)
来源：arXiv cs.CL · arxiv.org