Self-Learning for Zero Shot Neural Machine Translation

TL;DR

This paper introduces a novel zero-shot neural machine translation method that learns without relying on pivot language data, significantly improving translation quality for low-resource language pairs through a self-learning cycle.

Contribution

The work presents a new zero-shot NMT approach that does not require shared parallel data with a pivot language, using a self-learning cycle to enhance translation performance.

Findings

Up to +5.93 BLEU improvement over supervised baseline

Effective across diverse language pairs with different scripts and relatedness

Outperforms unsupervised NMT, especially in domain-mismatch scenarios

Abstract

Neural Machine Translation (NMT) approaches employing monolingual data are showing steady improvements in resource rich conditions. However, evaluations using real-world low-resource languages still result in unsatisfactory performance. This work proposes a novel zero-shot NMT modeling approach that learns without the now-standard assumption of a pivot language sharing parallel data with the zero-shot source and target languages. Our approach is based on three stages: initialization from any pre-trained NMT model observing at least the target language, augmentation of source sides leveraging target monolingual data, and learning to optimize the initial model to the zero-shot pair, where the latter two constitute a self-learning cycle. Empirical findings involving four diverse (in terms of a language family, script and relatedness) zero-shot pairs show the effectiveness of our approach with up to +5.93 BLEU improvement against a supervised bilingual baseline. Compared to unsupervised NMT, consistent improvements are observed even in a domain-mismatch setting, attesting to the usability of our method.