mlsanityData to 7 Oct 2026

Evaluating the Robustness of Neural Language Models to Input Perturbations

arXiv 2108.12237 ↗First posted 27 Aug 2021cs.CL cs.AIcited by 87

Not on the map. Why?

Abstract (arXiv)

High-performance neural language models have obtained state-of-the-art results on a wide range of Natural Language Processing (NLP) tasks. However, results for common benchmark datasets often do not reflect model reliability and robustness when applied to noisy, real-world data. In this study, we design and implement various types of character-level and word-level perturbation methods to simulate realistic scenarios in which input texts may be slightly noisy or different from the data distribution on which NLP systems were trained. Conducting comprehensive experiments on different NLP tasks, we investigate the ability of high-performance language models such as BERT, XLNet, RoBERTa, and ELMo in handling different types of input perturbations. The results suggest that language models are sensitive to input perturbations and their performance can decrease even when small changes are introduced. We highlight that models need to be further improved and that current benchmarks are not reflecting model robustness well. We argue that evaluations on perturbed inputs should routinely complement widely-used benchmarks in order to yield a more realistic understanding of NLP systems robustness.

Key facts. Cited by 87 papers in the map (data through 7 Oct 2026). Builds on 25 of its 37 references that are in the map. First posted on arXiv 27 Aug 2021.

Builds on

25 of 37 references.

References · 25

  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. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,049
  4. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
  5. 2019XLNet: Generalized Autoregressive Pretraining for Language Understandingcited by 2,801
  6. 2018Deep contextualized word representationscited by 2,285
  7. 2017SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluationcited by 998
  8. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 979
  9. 2018Universal Sentence Encodercited by 485
  10. 2017Supervised Learning of Universal Sentence Representations from Natural Language Inference Datacited by 428
  11. 2016Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependenciescited by 350
  12. 2017Synthetic and Natural Noise Both Break Neural Machine Translationcited by 304
13 more
  1. 2019Combating Adversarial Misspellings with Robust Word Recognitioncited by 162
  2. 2017Deep Text Classification Can be Fooledcited by 105
  3. 2016Robustness of classifiers: from adversarial to random noisecited by 87
  4. 2020OpenAttack: An Open-source Textual Adversarial Attack Toolkitcited by 84
  5. 2019Perturbation Sensitivity Analysis to Detect Unintended Model Biasescited by 74
  6. 2020Adv-BERT: BERT is not robust on misspellings! Generating nature adversarial samples on BERTcited by 69
  7. 2021Towards Robustness Against Natural Language Word Substitutionscited by 68
  8. 2020Robust Encodings: A Framework for Combating Adversarial Typoscited by 58
  9. 2020Evaluating Robustness to Input Perturbations for Neural Machine Translationcited by 36
  10. 2017How Robust Are Character-Based Word Embeddings in Tagging and MT Against Wrod Scramlbing or Randdm Nouse?cited by 23
  11. 2020Post-hoc explanation of black-box classifiers using confident itemsetscited by 21
  12. 2018Identify Susceptible Locations in Medical Records via Adversarial Attacks on Deep Predictive Modelscited by 15
  13. 2020Explaining Black-box Models for Biomedical Text Classificationcited by 7

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 3; 2022: 12; 2023: 27; 2024: 15; 2025: 22; 2026: 8.

87 papers.

Sort:

By area:

  1. 2023Prompt Injection attack against LLM-integrated Applicationscited by 568
  2. 2023Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolutioncited by 521
  3. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 520
  4. 2023On Evaluating Adversarial Robustness of Large Vision-Language Modelscited by 281
  5. 2022Repairing the Cracked Foundation: A Survey of Obstacles in Evaluation Practices for Generated Textcited by 154
  6. 2022The Fallacy of AI Functionalitycited by 88
  7. 2023Physics of Language Models: Part 1, Learning Hierarchical Language Structurescited by 83
  8. 2022Evaluation Gaps in Machine Learning Practicecited by 48
  9. 2024Changing Answer Order Can Decrease MMLU Accuracycited by 41
  10. 2023In ChatGPT We Trust? Measuring and Characterizing the Reliability of ChatGPTcited by 35
  11. 2024Large Language Models for UAVs: Current State and Pathways to the Futurecited by 33
  12. 2023Text-CRS: A Generalized Certified Robustness Framework against Textual Adversarial Attackscited by 32
38 more
  1. 2024PromptWizard: Task-Aware Prompt Optimization Frameworkcited by 32
  2. 2023Assessing Hidden Risks of LLMs: An Empirical Study on Robustness, Consistency, and Credibilitycited by 28
  3. 2023Revisit Input Perturbation Problems for LLMs: A Unified Robustness Evaluation Framework for Noisy Slot Filling Taskcited by 23
  4. 2024Can LLMs Recognize Toxicity? A Structured Investigation Framework and Toxicity Metriccited by 23
  5. 2022Rank List Sensitivity of Recommender Systems to Interaction Perturbationscited by 20
  6. 2022Open Domain Multi-document Summarization: A Comprehensive Study of Model Brittleness under Retrievalcited by 20
  7. 2024Assessing Dialect Fairness and Robustness of Large Language Models in Reasoning Taskscited by 20
  8. 2025RobustFlow: Towards Robust Agentic Workflow Generationcited by 19
  9. 2026Brittlebench: Quantifying LLM robustness via prompt sensitivitycited by 19
  10. 2021Interpreting the Robustness of Neural NLP Models to Textual Perturbationscited by 17
  11. 2022Not Just Pretty Pictures: Toward Interventional Data Augmentation Using Text-to-Image Generatorscited by 17
  12. 2025Enhancing LLM Robustness to Perturbed Instructions: An Empirical Studycited by 17
  13. 2023Influence of External Information on Large Language Models Mirrors Social Cognitive Patternscited by 16
  14. 2021Invariant Language Modelingcited by 15
  15. 2024Towards Truly Zero-shot Compositional Visual Reasoning with LLMs as Programmerscited by 14
  16. 2024NLP Verification: Towards a General Methodology for Certifying Robustnesscited by 14
  17. 2025Prompt Perturbations Reveal Human-Like Biases in Large Language Model Survey Responsescited by 13
  18. 2023Surfacing Biases in Large Language Models using Contrastive Input Decodingcited by 11
  19. 2023Observations on LLMs for Telecom Domain: Capabilities and Limitationscited by 11
  20. 2024MM-R3: On (In-)Consistency of Vision-Language Models (VLMs)cited by 10
  21. 2025Out of Style: RAG's Fragility to Linguistic Variationcited by 10
  22. 2022Robust Natural Language Processing: Recent Advances, Challenges, and Future Directionscited by 9
  23. 2022PSSAT: A Perturbed Semantic Structure Awareness Transferring Method for Perturbation-Robust Slot Fillingcited by 9
  24. 2024An Empirical Study of In-context Learning in LLMs for Machine Translationcited by 9
  25. 2025Evaluating Robustness of Large Language Models Against Multilingual Typographical Errorscited by 9
  26. 2022A.I. Robustness: a Human-Centered Perspective on Technological Challenges and Opportunitiescited by 8
  27. 2021Improving the robustness and accuracy of biomedical language models through adversarial trainingcited by 7
  28. 2024Evaluating Large Language Models along Dimensions of Language Variation: A Systematik Invesdigatiom uv Cross-lingual Generalizationcited by 7
  29. 2023Model-agnostic explainable artificial intelligence for object detection in image datacited by 6
  30. 2023ANTONIO: Towards a Systematic Method of Generating NLP Benchmarks for Verificationcited by 6
  31. 2023Dual Use Concerns of Generative AI and Large Language Modelscited by 6
  32. 2023Evaluating Concurrent Robustness of Language Models Across Diverse Challenge Setscited by 6
  33. 2025Exploring Robustness of Multilingual LLMs on Real-World Noisy Datacited by 6
  34. 2025Subword models struggle with word learning, but surprisal hides itcited by 6
  35. 2025SAGE: A Realistic Benchmark for Semantic Understandingcited by 6
  36. 2023Towards Robust and Generalizable Training: An Empirical Study of Noisy Slot Filling for Input Perturbationscited by 5
  37. 2023DemoNSF: A Multi-task Demonstration-based Generative Framework for Noisy Slot Filling Taskcited by 5
  38. 2025Position: Stop Acting Like Language Model Agents Are Normal Agentscited by 5

50 most cited of 87 All 87 (JSON)