mlsanityData to 7 Oct 2026

Interpreting the Robustness of Neural NLP Models to Textual Perturbations

arXiv 2110.07159 ↗First posted 14 Oct 2021cs.CLcited by 17

Not on the map. Why?

Abstract (arXiv)

Modern Natural Language Processing (NLP) models are known to be sensitive to input perturbations and their performance can decrease when applied to real-world, noisy data. However, it is still unclear why models are less robust to some perturbations than others. In this work, we test the hypothesis that the extent to which a model is affected by an unseen textual perturbation (robustness) can be explained by the learnability of the perturbation (defined as how well the model learns to identify the perturbation with a small amount of evidence). We further give a causal justification for the learnability metric. We conduct extensive experiments with four prominent NLP models -- TextRNN, BERT, RoBERTa and XLNet -- over eight types of textual perturbations on three datasets. We show that a model which is better at identifying a perturbation (higher learnability) becomes worse at ignoring such a perturbation at test time (lower robustness), providing empirical support for our hypothesis.

Key facts. Cited by 17 papers in the map (data through 7 Oct 2026). Builds on 27 of its 46 references that are in the map. First posted on arXiv 14 Oct 2021.

Builds on

27 of 46 references.

References · 27

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  3. 2019XLNet: Generalized Autoregressive Pretraining for Language Understandingcited by 2,801
  4. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
  5. 2021SimCSE: Simple Contrastive Learning of Sentence Embeddingscited by 2,168
  6. 2017Improved Regularization of Convolutional Neural Networks with Cutoutcited by 1,623
  7. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 979
  8. 2019EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Taskscited by 800
  9. 2020An Investigation of Why Overparameterization Exacerbates Spurious Correlationscited by 310
  10. 2021ConSERT: A Contrastive Framework for Self-Supervised Sentence Representation Transfercited by 231
  11. 2017Cut, Paste and Learn: Surprisingly Easy Synthesis for Instance Detectioncited by 209
  12. 2018Contextual Augmentation: Data Augmentation by Words with Paradigmatic Relationscited by 201
15 more
  1. 2016Recurrent Neural Network for Text Classification with Multi-Task Learningcited by 144
  2. 2020An Empirical Study on Robustness to Spurious Correlations using Pre-trained Language Modelscited by 133
  3. 2021Robustness Gym: Unifying the NLP Evaluation Landscapecited by 107
  4. 2019Text Processing Like Humans Do: Visually Attacking and Shielding NLP Systemscited by 99
  5. 2021Evaluating the Robustness of Neural Language Models to Input Perturbationscited by 87
  6. 2020Syntactic Data Augmentation Increases Robustness to Inference Heuristicscited by 85
  7. 2019Perturbation Sensitivity Analysis to Detect Unintended Model Biasescited by 74
  8. 2020A Causal View on Robustness of Neural Networkscited by 53
  9. 2021Robust Open-Vocabulary Translation from Visual Text Representationscited by 47
  10. 2020Evaluating Robustness to Input Perturbations for Neural Machine Translationcited by 36
  11. 2020Can neural networks acquire a structural bias from raw linguistic data?cited by 35
  12. 2015APAC: Augmented PAttern Classification with Neural Networkscited by 23
  13. 2021Code-Mixing on Sesame Street: Dawn of the Adversarial Polyglotscited by 23
  14. 2019Improving Neural Machine Translation Robustness via Data Augmentation: Beyond Back Translationcited by 16
  15. 2021Everything Has a Cause: Leveraging Causal Inference in Legal Text Analysiscited by 14

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 1; 2022: 2; 2023: 5; 2024: 6; 2025: 1; 2026: 2.

17 papers.

Sort:

By area:

  1. 2023Prompt Injection attack against LLM-integrated Applicationscited by 568
  2. 2022A Close Look into the Calibration of Pre-trained Language Modelscited by 57
  3. 2023Assessing Hidden Risks of LLMs: An Empirical Study on Robustness, Consistency, and Credibilitycited by 28
  4. 2023Sociodemographic Bias in Language Models: A Survey and Forward Pathcited by 27
  5. 2021Know Thy Strengths: Comprehensive Dialogue State Tracking Diagnosticscited by 9
  6. 2022Robust Natural Language Processing: Recent Advances, Challenges, and Future Directionscited by 9
  7. 2024An Empirical Study of In-context Learning in LLMs for Machine Translationcited by 9
  8. 2025Evaluating Robustness of Large Language Models Against Multilingual Typographical Errorscited by 9
  9. 2024Robustness of Structured Data Extraction from In-plane Rotated Documents using Multi-Modal Large Language Models (LLM)cited by 7
  10. 2024A Curious Case of Searching for the Correlation between Training Data and Adversarial Robustness of Transformer Textual Modelscited by 5
  11. 2024Stochastic Monkeys at Play: Random Augmentations Cheaply Break LLM Safety Alignmentcited by 5
  12. 2023Greener yet Powerful: Taming Large Code Generation Models with Quantizationcited by 4
5 more
  1. 2026JevOut: Natural Context Can Flip Decision Modelscited by 4
  2. 2024Investigating the Impact of Model Instability on Explanations and Uncertaintycited by 3
  3. 2023How Does Diffusion Influence Pretrained Language Models on Out-of-Distribution Data?cited by 0
  4. 2024Harmonic LLMs are Trustworthycited by 0
  5. 2026CDAE: Enhancing Perturbation Robustness in Pretrained Language Models with Contrastive Denoisingcited by 0