Chasing COMET: Leveraging Minimum Bayes Risk Decoding for Self-Improving Machine Translation

TL;DR

This paper introduces a method for self-improving machine translation by iteratively fine-tuning models using Minimum Bayes Risk decoding with COMET as the utility metric, leading to significant quality improvements across languages.

Contribution

It presents a novel approach combining MBR decoding with self-fine-tuning for domain adaptation and low-resource languages in machine translation.

Findings

Significant translation quality improvements across language pairs.

Effective application to domain-adapted and low-resource models.

Potential for iterative self-improvement in MT systems.

Abstract

This paper explores Minimum Bayes Risk (MBR) decoding for self-improvement in machine translation (MT), particularly for domain adaptation and low-resource languages. We implement the self-improvement process by fine-tuning the model on its MBR-decoded forward translations. By employing COMET as the MBR utility metric, we aim to achieve the reranking of translations that better aligns with human preferences. The paper explores the iterative application of this approach and the potential need for language-specific MBR utility metrics. The results demonstrate significant enhancements in translation quality for all examined language pairs, including successful application to domain-adapted models and generalisation to low-resource settings. This highlights the potential of COMET-guided MBR for efficient MT self-improvement in various scenarios.