BLEURT Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Training

TL;DR

This paper analyzes neural automatic metrics for machine translation, revealing robustness issues and proposing token-level constraints to improve their reliability and the translation systems they guide.

Contribution

It systematically compares metrics like BLEURT and BARTScore, identifies robustness defects, and introduces token-level constraints to enhance metric robustness and translation quality.

Findings

BLEURT and BARTScore exhibit universal adversarial translations.

Distribution biases and metric paradigms cause robustness issues.

Token-level constraints improve metric robustness and translation performance.

Abstract

Automatic metrics play a crucial role in machine translation. Despite the widespread use of n-gram-based metrics, there has been a recent surge in the development of pre-trained model-based metrics that focus on measuring sentence semantics. However, these neural metrics, while achieving higher correlations with human evaluations, are often considered to be black boxes with potential biases that are difficult to detect. In this study, we systematically analyze and compare various mainstream and cutting-edge automatic metrics from the perspective of their guidance for training machine translation systems. Through Minimum Risk Training (MRT), we find that certain metrics exhibit robustness defects, such as the presence of universal adversarial translations in BLEURT and BARTScore. In-depth analysis suggests two main causes of these robustness deficits: distribution biases in the training datasets, and the tendency of the metric paradigm. By incorporating token-level constraints, we enhance the robustness of evaluation metrics, which in turn leads to an improvement in the performance of machine translation systems. Codes are available at \url{https://github.com/powerpuffpomelo/fairseq_mrt}.