Targeted Multilingual Adaptation for Low-resource Language Families

TL;DR

This paper demonstrates that targeted multilingual adaptation, focusing on closely related languages, significantly improves performance on low-resource languages, establishing new best practices for language adaptation.

Contribution

It introduces a systematic approach for adapting pre-trained models to language families, showing the effectiveness of targeted multilingual training for low-resource languages.

Findings

Adapted models outperform baselines on downstream tasks.

Vocabulary size has limited impact on low-resource language performance.

Aggressive up-sampling benefits low-resource languages with minimal impact on high-resource languages.

Abstract

The "massively-multilingual" training of multilingual models is known to limit their utility in any one language, and they perform particularly poorly on low-resource languages. However, there is evidence that low-resource languages can benefit from targeted multilinguality, where the model is trained on closely related languages. To test this approach more rigorously, we systematically study best practices for adapting a pre-trained model to a language family. Focusing on the Uralic family as a test case, we adapt XLM-R under various configurations to model 15 languages; we then evaluate the performance of each experimental setting on two downstream tasks and 11 evaluation languages. Our adapted models significantly outperform mono- and multilingual baselines. Furthermore, a regression analysis of hyperparameter effects reveals that adapted vocabulary size is relatively unimportant for low-resource languages, and that low-resource languages can be aggressively up-sampled during training at little detriment to performance in high-resource languages. These results introduce new best practices for performing language adaptation in a targeted setting.