To Ship or Not to Ship: An Extensive Evaluation of Automatic Metrics for Machine Translation

TL;DR

This paper evaluates the reliability of automatic metrics for machine translation by comparing them against a large set of human judgments, revealing their strengths and limitations across languages and domains.

Contribution

It provides the largest dataset of human judgments for MT evaluation and systematically assesses how well automatic metrics predict human-perceived quality.

Findings

Metrics vary in accuracy across language pairs.

BLEU's exclusive use hindered model development.

Some metrics closely align with human judgments.

Abstract

Automatic metrics are commonly used as the exclusive tool for declaring the superiority of one machine translation system's quality over another. The community choice of automatic metric guides research directions and industrial developments by deciding which models are deemed better. Evaluating metrics correlations with sets of human judgements has been limited by the size of these sets. In this paper, we corroborate how reliable metrics are in contrast to human judgements on -- to the best of our knowledge -- the largest collection of judgements reported in the literature. Arguably, pairwise rankings of two systems are the most common evaluation tasks in research or deployment scenarios. Taking human judgement as a gold standard, we investigate which metrics have the highest accuracy in predicting translation quality rankings for such system pairs. Furthermore, we evaluate the performance of various metrics across different language pairs and domains. Lastly, we show that the sole use of BLEU impeded the development of improved models leading to bad deployment decisions. We release the collection of 2.3M sentence-level human judgements for 4380 systems for further analysis and replication of our work.