Exploring The Landscape of Distributional Robustness for Question Answering Models

TL;DR

This paper provides a comprehensive empirical analysis of distributional robustness in question answering models, revealing key insights about model variations, training methods, and robustness across diverse datasets.

Contribution

It offers the first large-scale evaluation of over 350 models across multiple datasets, highlighting factors influencing robustness and providing publicly available evaluation resources.

Findings

Zero-shot and in-context learning are more robust than fine-tuned models.

Few-shot prompt fine-tuning outperforms span prediction in robustness.

Parameter-efficient and robustness-focused training methods do not significantly improve robustness.

Abstract

We conduct a large empirical evaluation to investigate the landscape of distributional robustness in question answering. Our investigation spans over 350 models and 16 question answering datasets, including a diverse set of architectures, model sizes, and adaptation methods (e.g., fine-tuning, adapter tuning, in-context learning, etc.). We find that, in many cases, model variations do not affect robustness and in-distribution performance alone determines out-of-distribution performance. Moreover, our findings indicate that i) zero-shot and in-context learning methods are more robust to distribution shifts than fully fine-tuned models; ii) few-shot prompt fine-tuned models exhibit better robustness than few-shot fine-tuned span prediction models; iii) parameter-efficient and robustness enhancing training methods provide no significant robustness improvements. In addition, we publicly release all evaluations to encourage researchers to further analyze robustness trends for question answering models.