Incorporating BERT into Neural Machine Translation

TL;DR

This paper introduces BERT-fused models that integrate BERT representations into neural machine translation systems, significantly improving translation quality across various datasets and settings.

Contribution

It proposes a novel BERT-fused algorithm that incorporates BERT as a contextual embedding within NMT models, achieving state-of-the-art results.

Findings

Achieved state-of-the-art results on seven benchmark datasets.

Effective across supervised, semi-supervised, and unsupervised translation tasks.

BERT as a contextual embedding outperforms fine-tuning in NMT applications.

Abstract

The recently proposed BERT has shown great power on a variety of natural language understanding tasks, such as text classification, reading comprehension, etc. However, how to effectively apply BERT to neural machine translation (NMT) lacks enough exploration. While BERT is more commonly used as fine-tuning instead of contextual embedding for downstream language understanding tasks, in NMT, our preliminary exploration of using BERT as contextual embedding is better than using for fine-tuning. This motivates us to think how to better leverage BERT for NMT along this direction. We propose a new algorithm named BERT-fused model, in which we first use BERT to extract representations for an input sequence, and then the representations are fused with each layer of the encoder and decoder of the NMT model through attention mechanisms. We conduct experiments on supervised (including sentence-level and document-level translations), semi-supervised and unsupervised machine translation, and achieve state-of-the-art results on seven benchmark datasets. Our code is available at \url{https://github.com/bert-nmt/bert-nmt}.