TransFool: An Adversarial Attack against Neural Machine Translation Models

Sahar Sadrizadeh, Pascal Frossard, Ljiljana Dolamic

Keywords: Adversarial attack, deep neural network, language model, natural language processing, neural machine translation, robstness.

TL;DR: We propose TransFool to build adversarial attacks against neural machine translation systems, which are fluent sentences and semantically similar to the original sentence, but highly degrade the translation quality.

Abstract: Deep neural networks have been shown to be vulnerable to small perturbations of their inputs known as adversarial attacks. In this paper, we consider the particular task of Neural Machine Translation (NMT), where security is often critical. We investigate the vulnerability of NMT models to adversarial attacks and propose a new attack algorithm called TransFool. It builds on a multi-term optimization problem and a gradient projection step to compute adversarial examples that fool NMT models. By integrating the embedding representation of a language model in the proposed attack, we generate fluent adversarial examples in the source language that maintain a high level of semantic similarity with the clean samples and render the attack largely undetectable. Experimental results demonstrate that, for multiple translation tasks and different NMT architectures, our white-box attack can severely degrade the translation quality for more than 60% of the sentences while the semantic similarity between the original sentence and the adversarial example stays very high. Moreover, we show that the proposed attack is transferable to unknown target models and can fool those quite easily. Finally, our method leads to improvement in terms of success rate, semantic similarity, and fluency compared to the existing attack strategies both in white-box and black-box settings. Hence, TransFool permits to better characterize the vulnerability of NMT systems and outlines the necessity to design strong defense mechanisms and more robust NMT systems for real-life applications.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Social Aspects of Machine Learning (eg, AI safety, fairness, privacy, interpretability, human-AI interaction, ethics)