Adapting to Non-Centered Languages for Zero-shot Multilingual Translation

TL;DR

This paper introduces a lightweight method to improve zero-shot multilingual translation by adapting to non-centered languages, effectively addressing instability issues and enhancing performance across various datasets.

Contribution

The paper proposes a novel language-specific modeling approach that counteracts zero-shot translation instability by adapting to non-centered languages and combining shared and specific information.

Findings

Outperforms strong baselines in centered data conditions

Effectively handles non-centered data conditions

Disentangles coupled representations in the correct direction

Abstract

Multilingual neural machine translation can translate unseen language pairs during training, i.e. zero-shot translation. However, the zero-shot translation is always unstable. Although prior works attributed the instability to the domination of central language, e.g. English, we supplement this viewpoint with the strict dependence of non-centered languages. In this work, we propose a simple, lightweight yet effective language-specific modeling method by adapting to non-centered languages and combining the shared information and the language-specific information to counteract the instability of zero-shot translation. Experiments with Transformer on IWSLT17, Europarl, TED talks, and OPUS-100 datasets show that our method not only performs better than strong baselines in centered data conditions but also can easily fit non-centered data conditions. By further investigating the layer attribution, we show that our proposed method can disentangle the coupled representation in the correct direction.