Understanding the effects of language-specific class imbalance in multilingual fine-tuning
Additional information
Authors
Jung V.,
van der Plas L.
Type
Article in conference proceedings
Year
2024
Language
English
Abstract
"We study the effect of one type of imbalance often present in real-life multilingual classification datasets: an uneven distribution of labels across languages. We show evidence that fine-tuning a transformer-based Large Language Model (LLM) on a dataset with this imbalance leads to worse performance, a more pronounced separation of languages in the latent space, and the promotion of uninformative features. We modify the traditional class weighing approach to imbalance by calculating class weights separately for each language and show that this helps mitigate those detrimental effects. These results create awareness of the negative effects of language-specific class imbalance in multilingual fine-tuning and the way in which the model learns to rely on the separation of languages to perform the task.
Conference proceedings
Proceedings of Findings of the Association for Computational Linguistics
Numero ( Mese )
March
Publisher
"Association for Computational Linguistics"
Meeting name
EACL 2024
Meeting place
St. Julian's, Malta
Meeting date
March 2024
Pages (or article number)
"2368-2376"
