Evaluating adversarial attacks against multiple fact verification systems

By James Thorne, Andreas Vlachos, Christos Christodoulopoulos, Arpit Mittal. 2019

52 citations recorded by the source corpus.

Automated fact verification has been progressing owing to advancements in modeling and availability of large datasets. Due to the nature of the task, it is critical to understand the vulnerabilities of these systems against adversarial instances designed to make them predict incorrectly. We introduce two novel scoring metrics, attack potency and system resilience which take into account the correctness of the adversarial instances, an aspect often ignored in adversarial evaluations. We consider six fact verification systems from the recent Fact Extraction and VERification (FEVER) challenge: the four best-scoring ones and two baselines. We evaluate adversarial instances generated by a recently proposed state-of-the-art method, a paraphrasing method, and rule-based attacks devised for fact verification. We find that our rule-based attacks have higher potency, and that while the rankings among the top systems changed, they exhibited higher resilience than the baselines.

Subfields: Artificial Intelligence, Sociology and Political Science, Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Sociology and Political Science

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

Survey of Hallucination in Natural Language Generation (2022, background): “However, it has been shown that veriication models are prone to adversarial attacks and are not robust to negation, numerical or comparative words [133].”
The FEVER2.0 Shared Task (2019, methodology): “The submissions were scored using ‘potency’ and ‘resilience’ (Thorne et al., 2019) that compute a weighted average of FEVER scores: accounting for the correctness of adversarial instances.”
Get Your Vitamin C! Robust Fact Verification with Contrastive Evidence (2021, methodology): “Moreover, we use the revision history as distant supervision for word-level rationales, allowing for ﬁner-grained explanations (Camburu et al., 2018; Lei et al., 2016; Portelli et al., 2020; Thorne et al., 2019b).”
Fool Me Twice: Entailment from Wikipedia Gamification (2021, methodology): “To date, adversarial FEVER example generation uses automatic rules to increase their difficulty (Thorne et al., 2019).”
Beyond Leaderboards: A survey of methods for revealing weaknesses in Natural Language Inference data and models (2020, background): “Adversarial Evaluation refers to generating data with the aim to “fool” a target model.”
Explainable Automated Fact-Checking: A Survey (2020, methodology): “In fact, for automated fact-checking there have even been attempts to generate task specific adversarial claims (Thorne et al., 2019a), e.”
DialFact: A Benchmark for Fact-Checking in Dialogue (2021, methodology): “…by mutating the responses in WoW dataset.217 4.1.1 Methods for claim generation 218 Negation We use the 42 rule-based transforma- 219 tions from Thorne et al. (2019) which apply to 220 verb phrases of the claims to convert them to their 221 negated versions by adding words like “not” or 222…”