Dive into the Chasm: Probing the Gap between In- and Cross-Topic Generalization

TL;DR

This paper investigates the significant differences in how pre-trained language models perform when tested on topics similar to training data versus entirely new topics, revealing factors that influence their robustness and generalization capabilities.

Contribution

It provides the first comprehensive analysis of the variability in generalization gaps and robustness across different language models, highlighting the impact of training objectives and regularization techniques.

Findings

Generalization gaps vary significantly across models

Larger models show different robustness patterns

Regularization and data deduplication improve generalization

Abstract

Pre-trained language models (LMs) perform well in In-Topic setups, where training and testing data come from the same topics. However, they face challenges in Cross-Topic scenarios where testing data is derived from distinct topics -- such as Gun Control. This study analyzes various LMs with three probing-based experiments to shed light on the reasons behind the In- vs. Cross-Topic generalization gap. Thereby, we demonstrate, for the first time, that generalization gaps and the robustness of the embedding space vary significantly across LMs. Additionally, we assess larger LMs and underscore the relevance of our analysis for recent models. Overall, diverse pre-training objectives, architectural regularization, or data deduplication contribute to more robust LMs and diminish generalization gaps. Our research contributes to a deeper understanding and comparison of language models across different generalization scenarios.