Contextual Breach: Assessing the Robustness of Transformer-based QA Models

TL;DR

This paper evaluates the robustness of transformer-based question-answering models against various adversarial noises in context, introducing a new dataset and metrics to systematically assess their vulnerabilities in realistic scenarios.

Contribution

The paper introduces a novel adversarial noise dataset with multiple noise types and levels, along with standardized robustness metrics for transformer-based QA models.

Findings

Models show significant performance degradation under adversarial noise.

Certain noise types cause more vulnerability than others.

Robustness varies across different transformer architectures.

Abstract

Contextual question-answering models are susceptible to adversarial perturbations to input context, commonly observed in real-world scenarios. These adversarial noises are designed to degrade the performance of the model by distorting the textual input. We introduce a unique dataset that incorporates seven distinct types of adversarial noise into the context, each applied at five different intensity levels on the SQuAD dataset. To quantify the robustness, we utilize robustness metrics providing a standardized measure for assessing model performance across varying noise types and levels. Experiments on transformer-based question-answering models reveal robustness vulnerabilities and important insights into the model's performance in realistic textual input.