Improving Zero-Shot Translation of Low-Resource Languages

TL;DR

This paper introduces an iterative training method that enhances zero-shot translation in low-resource multilingual neural machine translation by leveraging the model's own generated translations to improve performance.

Contribution

It proposes a simple iterative training procedure that uses the model's own translations to improve zero-shot translation quality in low-resource settings.

Findings

Achieved about 9 BLEU points improvement over baseline

Up to 2.08 BLEU points gain over pivoting with bilingual models

Observed slight improvements in non-zero-shot directions

Abstract

Recent work on multilingual neural machine translation reported competitive performance with respect to bilingual models and surprisingly good performance even on (zeroshot) translation directions not observed at training time. We investigate here a zero-shot translation in a particularly lowresource multilingual setting. We propose a simple iterative training procedure that leverages a duality of translations directly generated by the system for the zero-shot directions. The translations produced by the system (sub-optimal since they contain mixed language from the shared vocabulary), are then used together with the original parallel data to feed and iteratively re-train the multilingual network. Over time, this allows the system to learn from its own generated and increasingly better output. Our approach shows to be effective in improving the two zero-shot directions of our multilingual model. In particular, we observed gains of about 9 BLEU points over a baseline multilingual model and up to 2.08 BLEU over a pivoting mechanism using two bilingual models. Further analysis shows that there is also a slight improvement in the non-zero-shot language directions.