arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2311.08380cs.CL

结合最小贝叶斯风险解码的神经机器翻译直接偏好优化

Direct Preference Optimization for Neural Machine Translation with Minimum Bayes Risk Decoding

University of Cambridge(剑桥大学)

Guangyu Yang, Jinghong Chen, Weizhe Lin, Bill Byrne

AI总结:

针对多语言大语言模型的MBR解码计算成本高的问题,提出用DPO强化学习技术微调模型,仅用小型单语微调集即可在推理无额外计算的情况下提升多个NMT测试集的翻译性能。

摘要:

最小贝叶斯风险(MBR)解码可显著提升多语言大语言模型(MLLMs)的翻译性能,但计算成本高昂。本文展示了如何利用最新的强化学习技术直接偏好优化(DPO)微调MLLMs,在推理阶段无需额外计算即可获得MBR的性能增益。该方法仅使用小型单语微调集,在多个神经机器翻译(NMT)测试集上的性能较未使用DPO的MLLMs有显著提升。

英文摘要:

Minimum Bayes Risk (MBR) decoding can significantly improve translation performance of Multilingual Large Language Models (MLLMs). However, MBR decoding is computationally expensive. We show how the recently developed Reinforcement Learning technique, Direct Preference Optimization (DPO), can fine-tune MLLMs to get the gains of MBR without any additional computation in inference. Our method uses only a small monolingual fine-tuning set and yields significantly improved performance on multiple NMT test sets compared to MLLMs without DPO.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑