Paper published in a book (Scientific congresses, symposiums and conference proceedings)
How do humans perceive adversarial text? A reality check on the validity and naturalness of word-based adversarial attacks
DYRMISHI, Salijona; GHAMIZI, Salah; CORDY, Maxime
2023In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics
Peer reviewed
 

Files


Full Text
ACL paper.pdf
Publisher postprint (1.19 MB)
Download

All documents in ORBilu are protected by a user license.

Send to



Details



Abstract :
[en] Natural Language Processing (NLP) models based on Machine Learning (ML) are susceptible to adversarial attacks -- malicious algorithms that imperceptibly modify input text to force models into making incorrect predictions. However, evaluations of these attacks ignore the property of imperceptibility or study it under limited settings. This entails that adversarial perturbations would not pass any human quality gate and do not represent real threats to human-checked NLP systems. To bypass this limitation and enable proper assessment (and later, improvement) of NLP model robustness, we have surveyed 378 human participants about the perceptibility of text adversarial examples produced by state-of-the-art methods. Our results underline that existing text attacks are impractical in real-world scenarios where humans are involved. This contrasts with previous smaller-scale human studies, which reported overly optimistic conclusions regarding attack success. Through our work, we hope to position human perceptibility as a first-class success criterion for text attacks, and provide guidance for research to build effective attack algorithms and, in turn, design appropriate defence mechanisms.
Research center :
Interdisciplinary Centre for Security, Reliability and Trust (SnT) > Other
Disciplines :
Computer science
Author, co-author :
DYRMISHI, Salijona ;  University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT) > SerVal
GHAMIZI, Salah ;  University of Luxembourg > Faculty of Science, Technology and Medicine (FSTM) > Department of Computer Science (DCS)
CORDY, Maxime  ;  University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT) > SerVal
External co-authors :
no
Language :
English
Title :
How do humans perceive adversarial text? A reality check on the validity and naturalness of word-based adversarial attacks
Publication date :
2023
Event name :
ACL 2023: The 61st Annual Meeting of the Association for Computational Linguistics
Event organizer :
ACL
Event place :
Toronto, Canada
Event date :
from 09-07-2023 t0 14-07-2023
Main work title :
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics
Publisher :
Association for Computational Linguistics
ISBN/EAN :
978-1-959429-72-2
Collection name :
Volume 1: Long Papers
Peer reviewed :
Peer reviewed
Focus Area :
Security, Reliability and Trust
FnR Project :
FNR14585105 - Search-based Adversarial Testing Under Domain-specific Constraints, 2020 (01/10/2020-30/09/2024) - Salijona Dyrmishi
Available on ORBilu :
since 13 August 2023

Statistics


Number of views
56 (1 by Unilu)
Number of downloads
9 (0 by Unilu)

Scopus citations®
 
1
Scopus citations®
without self-citations
1

Bibliography


Similar publications



Contact ORBilu