Towards Debiasing Fact Verification Models

By Tal Schuster, Darsh J. Shah, Yunghi Yeo, Daniel Filizzola, Enrico Santus, Regina Barzilay. 2019

236 citations recorded by the source corpus.

Fact verification requires validating a claim in the context of evidence. We show, however, that in the popular FEVER dataset this might not necessarily be the case. Claim-only classifiers perform competitively with top evidence-aware models. In this paper, we investigate the cause of this phenomenon, identifying strong cues for predicting labels solely based on the claim, without considering any evidence. We create an evaluation set that avoids those idiosyncrasies. The performance of FEVER-trained models significantly drops when evaluated on this test set. Therefore, we introduce a regularization method which alleviates the effect of bias in the training data, obtaining improvements on the newly created test set. This work is a step towards a more sound evaluation of reasoning capabilities in fact verification models.

Subfields: Artificial Intelligence, Management Science and Operations Research, 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 Management Science and Operations Research

In Safety, Robustness & Interpretability

Citations in the corpus

Avoiding catastrophic forgetting in mitigating model biases in sentence-pair classification with elastic weight consolidation (2020, methodology): “instances from the symmetric data released by Schuster et al. (2019). The symmetric dataset was constructed from a sample of the FEVER instances with SUPPORTED and REFUTED labels where the claims were retained but the ev-”
Reasoning Over Semantic-Level Graph for Fact Checking (2019, background): “We plan to study fact checking with adversarial attacks (Thorne and Vlachos 2019; Schuster et al. 2019) in the future.”
ProoFVer: Natural Logic Theorem Proving for Fact Verification (2021, methodology): “For robustness related experiments with Symmetric FEVER, five random initialisations of each model are trained.”
Evaluation of Text Generation: A Survey (2020, background): “A recent study (Schuster et al., 2019) reports the results of a fact verification system to identify inherent bias in training datasets that cause fact-checking issues.”
Get Your Vitamin C! Robust Fact Verification with Contrastive Evidence (2021, methodology): “For example, Schuster et al. (2019) All claims in the V ITAMIN C-synthetic are paired with one refuting and one supporting evidence, making it impossible for a claim-only to perform better than random.”
DynaSent: A Dynamic Benchmark for Sentiment Analysis (2020, background): “, 2020), or inherit social biases in the data they were trained on (Kiritchenko and Mohammad, 2018; Rudinger et al., 2017, 2018; Sap et al., 2019; Schuster et al., 2019).”
Elastic weight consolidation for better bias inoculation (2021, methodology): “We show this holds for both the NLI stress-tests, as well debiasing fact-verification systems (Schuster et al., 2019) through fine-tuning.”
Improving Robustness by Augmenting Training Sentences with Predicate-Argument Structures (2020, background): “…examples (He et al., 2019; Clark et al., 2019; Mahabadi and Henderson, 2019), (b) change the importance of the biased training examples in the training objective (Schuster et al., 2019; Mahabadi and Henderson, 2019), or (c) change the conﬁdence of the model on biased examples (Utama et al., 2020a).”
DialFact: A Benchmark for Fact-Checking in Dialogue (2021, methodology): “990 A Lexical Biases 991 Following Schuster et al. (2019), we measure the 992 Local Mutual Information (LMI) to measure the 993 correlation between bigrams in the claims (w) and 994 the categories l, defined as follows: LMI(w, l) = 995 p(w, l)log ( p(l/w)) p(l)) ) .”
Beyond Leaderboards: A survey of methods for revealing weaknesses in Natural Language Inference data and models (2020, background): “, 2019), re-weighting the loss function to incorporate the bias in the data (Schuster et al., 2019; Zhang et al., 2019b), parameter regularisation (Sagawa et al.”
Handling Bias in Toxic Speech Detection: A Survey (2022, methodology): “Sampling Topic proximity [104], Data mixing (augmentation) [4] Annotation Disagreement deconvolution [39] Lexical Length sensitive sampling [24], Knowledge-based term generalization [5], Data iltering [120], Regularization [49], Multi-task learning [103], Ensemble modeling [120] Racial Annotation pr”
Fool Me Twice: Entailment from Wikipedia Gamification (2021, background): “Despite this excitement, Schuster et al. (2019) show that FEVER has many of the same issues as entailment datasets.”
Evidence-based Verification for Real World Information Needs (2021, methodology): “Even when manually constructing claims, making meaning-altering changes may allow allow the veracity to be predicted without evidence (Schuster et al., 2019) due to subcon-scious patterns introduced by annotators.”
Generating Data to Mitigate Spurious Correlations in Natural Language Inference Datasets (2022, methodology): “Consequently, the models risk modelling the datasets, but not the tasks (Gururangan et al., 2018; Poliak et al., 2018; McCoy et al., 2019; Schuster et al., 2019).”