AlignFreeze: Navigating the Impact of Realignment on the Layers of Multilingual Models Across Diverse Languages

TL;DR

AlignFreeze is a novel method that selectively freezes layers during realignment in multilingual models, preventing performance degradation and improving PoS tagging accuracy across diverse languages.

Contribution

The paper introduces AlignFreeze, a new approach that improves cross-lingual transfer by selectively freezing layers during realignment in multilingual models.

Findings

Realignment impacts all layers but harms lower layers most.

Freezing lower layers prevents performance degradation.

AlignFreeze improves PoS tagging accuracy in multiple languages.

Abstract

Realignment techniques are often employed to enhance cross-lingual transfer in multilingual language models, still, they can sometimes degrade performance in languages that differ significantly from the fine-tuned source language. This paper introduces AlignFreeze, a method that freezes either the layers' lower half or upper half during realignment. Through controlled experiments on 4 tasks, 3 models, and in 35 languages, we find that realignment affects all the layers but can be the most detrimental to the lower ones. Freezing the lower layers can prevent performance degradation. Particularly, AlignFreeze improves Part-of-Speech (PoS) tagging performances in languages where full realignment fails: with XLM-R, it provides improvements of more than one standard deviation in accuracy in seven more languages than full realignment.