Data-Efficient Cross-Lingual Transfer with Language-Specific Subnetworks

TL;DR

This paper introduces language-specific subnetworks and dynamic updating techniques to improve cross-lingual transfer in multilingual models, reducing conflicts and enhancing positive transfer during fine-tuning.

Contribution

It presents novel methods for controlling parameter sharing with subnetworks and combines them with meta-learning to improve cross-lingual transfer.

Findings

Dynamic subnetworks reduce parameter conflicts.

Meta-learning enhances transfer performance.

Extensive analysis shows improved transfer with proposed methods.

Abstract

Large multilingual language models typically share their parameters across all languages, which enables cross-lingual task transfer, but learning can also be hindered when training updates from different languages are in conflict. In this paper, we propose novel methods for using language-specific subnetworks, which control cross-lingual parameter sharing, to reduce conflicts and increase positive transfer during fine-tuning. We introduce dynamic subnetworks, which are jointly updated with the model, and we combine our methods with meta-learning, an established, but complementary, technique for improving cross-lingual transfer. Finally, we provide extensive analyses of how each of our methods affects the models.