Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain

TL;DR

This paper evaluates the effectiveness of domain-adapted multilingual neural machine translation models in the medical field, demonstrating improved translation quality and fewer terminology errors for English-Romanian compared to out-of-domain models.

Contribution

It provides a comparative analysis of in-domain versus out-of-domain MNMT models in the medical domain, highlighting the benefits of domain adaptation.

Findings

In-domain MNMT outperforms out-of-domain in all automatic metrics.

In-domain MNMT produces fewer terminology-specific errors.

Domain adaptation improves translation quality in specialized fields.

Abstract

Multilingual Neural Machine Translation (MNMT) models leverage many language pairs during training to improve translation quality for low-resource languages by transferring knowledge from high-resource languages. We study the quality of a domain-adapted MNMT model in the medical domain for English-Romanian with automatic metrics and a human error typology annotation which includes terminology-specific error categories. We compare the out-of-domain MNMT with the in-domain adapted MNMT. The in-domain MNMT model outperforms the out-of-domain MNMT in all measured automatic metrics and produces fewer terminology errors.