Backtranslation Augmented Direct Preference Optimization for Neural Machine Translation

TL;DR

This paper introduces a reinforcement learning framework using Direct Preference Optimization to improve neural machine translation quality post-training, demonstrated on English-German translation.

Contribution

It presents a novel RL-based post-training method leveraging preference feedback to enhance pre-trained NMT models.

Findings

COMET score improved from 0.703 to 0.747

DPO provides an efficient, stable post-training enhancement

Framework requires only a text corpus and expert feedback

Abstract

Contemporary neural machine translation (NMT) systems are almost exclusively built by training on supervised parallel data. Despite the tremendous progress achieved, these systems still exhibit persistent translation errors. This paper proposes that a post-training paradigm based on reinforcement learning (RL) can effectively rectify such mistakes. We introduce a novel framework that requires only a general text corpus and an expert translator which can be either human or an AI system to provide iterative feedback. In our experiments, we focus specifically on English-to-German translation as a representative high-resource language pair. Crucially, we implement this RL-based post-training using Direct Preference Optimization (DPO). Applying our DPO-driven framework to the gemma3-1b model yields a significant improvement in translation quality, elevating it's COMET score from 0.703 to 0.747 on the English to German task. The results demonstrate that DPO offers an efficient and stable pathway for enhancing pre-trained NMT models through preference-based post-training.