Parameter Differentiation based Multilingual Neural Machine Translation

TL;DR

This paper introduces a dynamic parameter differentiation method for multilingual neural machine translation, enabling the model to adaptively determine which parameters should be language-specific, leading to improved translation performance.

Contribution

It proposes a novel, training-driven approach inspired by cellular differentiation that automatically identifies language-specific parameters based on gradient similarity.

Findings

Significantly outperforms baseline models with various sharing configurations.

Parameter sharing aligns with linguistic similarities among languages.

Dynamic differentiation improves translation quality across multiple languages.

Abstract

Multilingual neural machine translation (MNMT) aims to translate multiple languages with a single model and has been proved successful thanks to effective knowledge transfer among different languages with shared parameters. However, it is still an open question which parameters should be shared and which ones need to be task-specific. Currently, the common practice is to heuristically design or search language-specific modules, which is difficult to find the optimal configuration. In this paper, we propose a novel parameter differentiation based method that allows the model to determine which parameters should be language-specific during training. Inspired by cellular differentiation, each shared parameter in our method can dynamically differentiate into more specialized types. We further define the differentiation criterion as inter-task gradient similarity. Therefore, parameters with conflicting inter-task gradients are more likely to be language-specific. Extensive experiments on multilingual datasets have demonstrated that our method significantly outperforms various strong baselines with different parameter sharing configurations. Further analyses reveal that the parameter sharing configuration obtained by our method correlates well with the linguistic proximities.