Multilingual Transfer and Domain Adaptation for Low-Resource Languages of Spain

TL;DR

This paper presents Huawei's participation in the WMT 2024 low-resource language translation tasks, employing various transfer and domain adaptation techniques to improve neural machine translation for Spanish to regional languages of Spain.

Contribution

It introduces a comprehensive set of training strategies for neural machine translation tailored to low-resource languages, demonstrating effective domain adaptation and transfer learning.

Findings

Achieved competitive results in translation quality.

Effective use of multilingual transfer and data augmentation.

Improved translation performance for low-resource languages.

Abstract

This article introduces the submission status of the Translation into Low-Resource Languages of Spain task at (WMT 2024) by Huawei Translation Service Center (HW-TSC). We participated in three translation tasks: spanish to aragonese (es-arg), spanish to aranese (es-arn), and spanish to asturian (es-ast). For these three translation tasks, we use training strategies such as multilingual transfer, regularized dropout, forward translation and back translation, labse denoising, transduction ensemble learning and other strategies to neural machine translation (NMT) model based on training deep transformer-big architecture. By using these enhancement strategies, our submission achieved a competitive result in the final evaluation.