Effective Strategies in Zero-Shot Neural Machine Translation

TL;DR

This paper introduces two strategies to improve zero-shot neural machine translation, enhancing performance and resource efficiency in multilingual systems without parallel data, especially for unbalanced language datasets.

Contribution

The paper proposes novel strategies specifically designed to address zero-shot translation challenges in multilingual NMT systems, focusing on reducing language bias.

Findings

Strategies improve translation quality in zero-shot scenarios

Methods reduce computational resource requirements

Effective on unbalanced multilingual datasets

Abstract

In this paper, we proposed two strategies which can be applied to a multilingual neural machine translation system in order to better tackle zero-shot scenarios despite not having any parallel corpus. The experiments show that they are effective in terms of both performance and computing resources, especially in multilingual translation of unbalanced data in real zero-resourced condition when they alleviate the language bias problem.