UNMASK: Discovering and Causally Verifying Spurious Shortcuts in Text Classifiers

• URL: http://arxiv.org/abs/2608.09209v1
• Date: Mon, 10 Aug 2026 07:31:08 GMT
• Title: UNMASK: Discovering and Causally Verifying Spurious Shortcuts in Text Classifiers
• Abstract summary: Neural language models trained on large crowdsourced corpora frequently exploit spurious surface patterns tied to target labels without true linguistic or causal relevance.<n>Existing approaches either require manual specification of the feature vocabulary or automate discovery only partially.<n>We present U N M ASK, a fully automated pipeline that discovers, causally verifies, and mitigates spurious correlations in text expressions.
• Score: 9.827138852806305
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Neural language models trained on large crowdsourced corpora frequently exploit spurious surface patterns tied to target labels without true linguistic or causal relevance, boosting benchmark performance while failing on adversarial or out-of-distribution inputs. Existing approaches either require manual specification of the feature vocabulary or automate discovery only partially, leaving the gap between dataset-level correlation and model-level exploitation unaddressed. We present U N M ASK, a fully automated pipeline that discovers, causally verifies, and mitigates spurious correlations in text classifiers without additional human annotation. Given unlabeled training examples, U N M ASK generates candidate surface patterns as executable boolean expressions, filters them through a statistical validation protocol with independent replication, and establishes causal model dependence via verified counterfactual interventions. Causally confirmed features then serve as annotation-free group definitions for Deep Feature Reweighting, eliminating the group labels that standard DFR requires. Applied to BERT and RoBERTa trained on MNLI, our pipeline independently rediscovers established lexical-overlap and negation biases, verifying 9 of 10 features on BERT and 6 on RoBERTa, and improving HANS accuracy by up to 12.58 pp. On CivilComments-WILDS, programmatic groups match the 70.1% worst- group accuracy of hand-labeled DFR (Kirichenko et al., 2023) without demographic annotation. We further demonstrate that the discovery and validation stages generalize to reward model preference data, surfacing interpretable spurious correlations in RewardBench2.

Related papers

• Generative Classifiers Avoid Shortcut Solutions [84.23247217037134]
  Discriminative approaches to classification often learn shortcuts that hold in-distribution but fail under minor distribution shift.<n>We show that generative classifiers can avoid this issue by modeling all features, both core and spurious, instead of mainly spurious ones.<n>We find that diffusion-based and autorerimigressive generative classifiers achieve state-of-the-art performance on five standard image and text distribution shift benchmarks.
  arXiv  Detail & Related papers  (2025-12-31T18:31:46Z)
• Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples [11.727747752958436]
  Deep learning models often learn features that spuriously correlate with the class label during training but are irrelevant to the prediction task.<n>Existing methods typically address this issue by annotating potential spurious attributes, or filtering spurious features based on some empirical assumptions.<n>We propose a data-oriented approach to mitigate the spurious correlation in deep learning models.
  arXiv  Detail & Related papers  (2025-12-28T10:54:51Z)
• You only need 4 extra tokens: Synergistic Test-time Adaptation for LLMs [50.54173262572369]
  Large language models (LLMs) are increasingly deployed in specialized domains such as finance, medicine, and agriculture.<n>We study label-free test-time adaptation for language models and present SyTTA, an inference-time framework that adapts models on-the-fly without additional supervision.
  arXiv  Detail & Related papers  (2025-10-11T14:00:39Z)
• Fighting Spurious Correlations in Text Classification via a Causal Learning Perspective [2.7813683000222653]
  We propose the Causally Calibrated Robust ( CCR) to reduce models' reliance on spurious correlations.<n> CCR integrates a causal feature selection method based on counterfactual reasoning, along with an inverse propensity weighting (IPW) loss function.<n>We show that CCR state-of-the-art performance among methods without group labels, and in some cases, it can compete with the models that utilize group labels.
  arXiv  Detail & Related papers  (2024-11-01T21:29:07Z)
• Trained Models Tell Us How to Make Them Robust to Spurious Correlation without Group Annotation [3.894771553698554]
  Empirical Risk Minimization (ERM) models tend to rely on attributes that have high spurious correlation with the target. This can degrade the performance on underrepresented (or'minority') groups that lack these attributes. We propose Environment-based Validation and Loss-based Sampling (EVaLS) to enhance robustness to spurious correlation.
  arXiv  Detail & Related papers  (2024-10-07T08:17:44Z)
• Noisy Correspondence Learning with Self-Reinforcing Errors Mitigation [63.180725016463974]
  Cross-modal retrieval relies on well-matched large-scale datasets that are laborious in practice. We introduce a novel noisy correspondence learning framework, namely textbfSelf-textbfReinforcing textbfErrors textbfMitigation (SREM)
  arXiv  Detail & Related papers  (2023-12-27T09:03:43Z)
• Model-based causal feature selection for general response types [8.228587135343071]
  Invariant causal prediction (ICP) is a method for causal feature selection which requires data from heterogeneous settings. We develop transformation-model (TRAM) based ICP, allowing for continuous, categorical, count-type, and uninformatively censored responses. We provide an open-source R package 'tramicp' and evaluate our approach on simulated data and in a case study investigating causal features of survival in critically ill patients.
  arXiv  Detail & Related papers  (2023-09-22T12:42:48Z)
• Bring Your Own Data! Self-Supervised Evaluation for Large Language Models [52.15056231665816]
  We propose a framework for self-supervised evaluation of Large Language Models (LLMs) We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence. We find strong correlations between self-supervised and human-supervised evaluations.
  arXiv  Detail & Related papers  (2023-06-23T17:59:09Z)
• Class-Adaptive Self-Training for Relation Extraction with Incompletely Annotated Training Data [43.46328487543664]
  Relation extraction (RE) aims to extract relations from sentences and documents. Recent studies showed that many RE datasets are incompletely annotated. This is known as the false negative problem in which valid relations are falsely annotated as 'no_relation'
  arXiv  Detail & Related papers  (2023-06-16T09:01:45Z)
• PromptORE -- A Novel Approach Towards Fully Unsupervised Relation Extraction [0.0]
  Unsupervised Relation Extraction (RE) aims to identify relations between entities in text, without having access to labeled data during training. We propose PromptORE, a ''Prompt-based Open Relation Extraction'' model. We adapt the novel prompt-tuning paradigm to work in an unsupervised setting, and use it to embed sentences expressing a relation. We show that PromptORE consistently outperforms state-of-the-art models with a relative gain of more than 40% in B 3, V-measure and ARI.
  arXiv  Detail & Related papers  (2023-03-24T12:55:35Z)
• Breaking the Spurious Causality of Conditional Generation via Fairness Intervention with Corrective Sampling [77.15766509677348]
  Conditional generative models often inherit spurious correlations from the training dataset. This can result in label-conditional distributions that are imbalanced with respect to another latent attribute. We propose a general two-step strategy to mitigate this issue.
  arXiv  Detail & Related papers  (2022-12-05T08:09:33Z)
• Examining and Combating Spurious Features under Distribution Shift [94.31956965507085]
  We define and analyze robust and spurious representations using the information-theoretic concept of minimal sufficient statistics. We prove that even when there is only bias of the input distribution, models can still pick up spurious features from their training data. Inspired by our analysis, we demonstrate that group DRO can fail when groups do not directly account for various spurious correlations.
  arXiv  Detail & Related papers  (2021-06-14T05:39:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.