Adversarial Text Generation with Dynamic Contextual Perturbation

TL;DR

This paper introduces Dynamic Contextual Perturbation (DCP), a novel adversarial text attack method that dynamically generates context-aware perturbations to challenge NLP models while maintaining natural language semantics.

Contribution

The paper presents a new context-aware adversarial attack scheme that leverages pre-trained language models to produce more natural and effective adversarial examples across entire texts.

Findings

DCP outperforms existing methods in fooling NLP models.

DCP maintains high semantic fidelity and fluency.

Adversarial examples generated by DCP are more subtle and impactful.

Abstract

Adversarial attacks on Natural Language Processing (NLP) models expose vulnerabilities by introducing subtle perturbations to input text, often leading to misclassification while maintaining human readability. Existing methods typically focus on word-level or local text segment alterations, overlooking the broader context, which results in detectable or semantically inconsistent perturbations. We propose a novel adversarial text attack scheme named Dynamic Contextual Perturbation (DCP). DCP dynamically generates context-aware perturbations across sentences, paragraphs, and documents, ensuring semantic fidelity and fluency. Leveraging the capabilities of pre-trained language models, DCP iteratively refines perturbations through an adversarial objective function that balances the dual objectives of inducing model misclassification and preserving the naturalness of the text. This comprehensive approach allows DCP to produce more sophisticated and effective adversarial examples that better mimic natural language patterns. Our experimental results, conducted on various NLP models and datasets, demonstrate the efficacy of DCP in challenging the robustness of state-of-the-art NLP systems. By integrating dynamic contextual analysis, DCP significantly enhances the subtlety and impact of adversarial attacks. This study highlights the critical role of context in adversarial attacks and lays the groundwork for creating more robust NLP systems capable of withstanding sophisticated adversarial strategies.