Neural Machine Translation with Pivot Languages

TL;DR

This paper introduces a joint training algorithm for pivot-based neural machine translation, enabling source-to-pivot and pivot-to-target models to interact during training, which improves translation quality for resource-scarce language pairs.

Contribution

It proposes three methods to connect and jointly train the two models, addressing the data scarcity problem in neural machine translation.

Findings

Joint training improves translation quality over independent models.

Significant performance gains on Europarl and WMT datasets.

Effective for resource-scarce language pairs.

Abstract

While recent neural machine translation approaches have delivered state-of-the-art performance for resource-rich language pairs, they suffer from the data scarcity problem for resource-scarce language pairs. Although this problem can be alleviated by exploiting a pivot language to bridge the source and target languages, the source-to-pivot and pivot-to-target translation models are usually independently trained. In this work, we introduce a joint training algorithm for pivot-based neural machine translation. We propose three methods to connect the two models and enable them to interact with each other during training. Experiments on Europarl and WMT corpora show that joint training of source-to-pivot and pivot-to-target models leads to significant improvements over independent training across various languages.