On the Interplay Between Fine-tuning and Sentence-level Probing for Linguistic Knowledge in Pre-trained Transformers

TL;DR

This paper investigates how fine-tuning affects the linguistic knowledge in pre-trained models like BERT, RoBERTa, and ALBERT using sentence-level probing, revealing that fine-tuning can both enhance and diminish linguistic representations depending on the task and model.

Contribution

It provides a detailed analysis of the impact of fine-tuning on linguistic knowledge in pre-trained transformers, highlighting variability across models and tasks.

Findings

Fine-tuning causes substantial changes in probing accuracy for some tasks.

Changes in representations are larger in higher layers of models.

Fine-tuning sometimes improves probing accuracy beyond strong pooling methods.

Abstract

Fine-tuning pre-trained contextualized embedding models has become an integral part of the NLP pipeline. At the same time, probing has emerged as a way to investigate the linguistic knowledge captured by pre-trained models. Very little is, however, understood about how fine-tuning affects the representations of pre-trained models and thereby the linguistic knowledge they encode. This paper contributes towards closing this gap. We study three different pre-trained models: BERT, RoBERTa, and ALBERT, and investigate through sentence-level probing how fine-tuning affects their representations. We find that for some probing tasks fine-tuning leads to substantial changes in accuracy, possibly suggesting that fine-tuning introduces or even removes linguistic knowledge from a pre-trained model. These changes, however, vary greatly across different models, fine-tuning and probing tasks. Our analysis reveals that while fine-tuning indeed changes the representations of a pre-trained model and these changes are typically larger for higher layers, only in very few cases, fine-tuning has a positive effect on probing accuracy that is larger than just using the pre-trained model with a strong pooling method. Based on our findings, we argue that both positive and negative effects of fine-tuning on probing require a careful interpretation.