Advancing Translation Preference Modeling with RLHF: A Step Towards Cost-Effective Solution

TL;DR

This paper introduces a cost-effective reinforcement learning with human feedback approach to improve machine translation quality, especially for low-resource languages, by training reward models to better align with human preferences.

Contribution

It proposes a novel preference learning strategy that enhances translation models using RLHF without requiring extensive high-quality human comparison datasets.

Findings

RLHF effectively improves translation quality across multiple languages.

A reward model with strong language skills better captures human preferences.

The approach benefits translation directions not directly trained with RLHF.

Abstract

Faithfulness, expressiveness, and elegance is the constant pursuit in machine translation. However, traditional metrics like \textit{BLEU} do not strictly align with human preference of translation quality. In this paper, we explore leveraging reinforcement learning with human feedback (\textit{RLHF}) to improve translation quality. It is non-trivial to collect a large high-quality dataset of human comparisons between translations, especially for low-resource languages. To address this issue, we propose a cost-effective preference learning strategy, optimizing reward models by distinguishing between human and machine translations. In this manner, the reward model learns the deficiencies of machine translation compared to human and guides subsequent improvements in machine translation. Experimental results demonstrate that \textit{RLHF} can effectively enhance translation quality and this improvement benefits other translation directions not trained with \textit{RLHF}. Further analysis indicates that the model's language capabilities play a crucial role in preference learning. A reward model with strong language capabilities can more sensitively learn the subtle differences in translation quality and align better with real human translation preferences.