mlsanityData to 7 Oct 2026

The Limits of Word Level Differential Privacy

arXiv 2205.02130 ↗First posted 2 May 2022cs.CR cs.CL cs.LGcited by 63

Not on the map. Why?

Abstract (arXiv)

As the issues of privacy and trust are receiving increasing attention within the research community, various attempts have been made to anonymize textual data. A significant subset of these approaches incorporate differentially private mechanisms to perturb word embeddings, thus replacing individual words in a sentence. While these methods represent very important contributions, have various advantages over other techniques and do show anonymization capabilities, they have several shortcomings. In this paper, we investigate these weaknesses and demonstrate significant mathematical constraints diminishing the theoretical privacy guarantee as well as major practical shortcomings with regard to the protection against deanonymization attacks, the preservation of content of the original sentences as well as the quality of the language output. Finally, we propose a new method for text anonymization based on transformer based language models fine-tuned for paraphrasing that circumvents most of the identified weaknesses and also offers a formal privacy guarantee. We evaluate the performance of our method via thorough experimentation and demonstrate superior performance over the discussed mechanisms.

Key facts. Cited by 63 papers in the map (data through 7 Oct 2026). Builds on 30 of its 83 references that are in the map. First posted on arXiv 2 May 2022.

Builds on

30 of 83 references.

References · 30

  1. 2017Attention Is All You Needcited by 59,932
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,049
  4. 2016Deep Learning with Differential Privacycited by 2,682
  5. 2020Extracting Training Data from Large Language Modelscited by 2,144
  6. 2015A large annotated corpus for learning natural language inferencecited by 1,837
  7. 2017Poincar'e Embeddings for Learning Hierarchical Representationscited by 674
  8. 2019The Woman Worked as a Babysitter: On Biases in Language Generationcited by 540
  9. 2019How to Fine-Tune BERT for Text Classification?cited by 409
  10. 2021Large Language Models Can Be Strong Differentially Private Learnerscited by 405
  11. 2021Differentially Private Fine-tuning of Language Modelscited by 392
  12. 2017Fake News Detection on Social Media: A Data Mining Perspectivecited by 334
18 more
  1. 2021Towards Understanding and Mitigating Social Biases in Language Modelscited by 294
  2. 2020Information Leakage in Embedding Modelscited by 199
  3. 2019Privacy- and Utility-Preserving Textual Analysis via Calibrated Multivariate Perturbationscited by 115
  4. 2017CSI: A Hybrid Deep Model for Fake News Detectioncited by 101
  5. 2018Privacy-preserving Neural Representations of Textcited by 54
  6. 2018Generalised Differential Privacy for Text Document Processingcited by 52
  7. 2019Leveraging Hierarchical Representations for Preserving Privacy and Utility in Textcited by 49
  8. 2017A4NT: Author Attribute Anonymity by Adversarial Training of Neural Machine Translationcited by 41
  9. 2020A Differentially Private Text Perturbation Method Using a Regularized Mahalanobis Metriccited by 41
  10. 2019ER-AE: Differentially Private Text Generation for Authorship Anonymizationcited by 35
  11. 2019Paraphrasing with Large Language Modelscited by 35
  12. 2018SynTF: Synthetic and Differentially Private Term Frequency Vectors for Privacy-Preserving Text Miningcited by 20
  13. 2021On a Utilitarian Approach to Privacy Preserving Text Generationcited by 17
  14. 2015Optimality of the Laplace Mechanism in Differential Privacycited by 15
  15. 2019A framework for fake review detection in online consumer electronics retailerscited by 11
  16. 2019Syntactic Recurrent Neural Network for Authorship Attributioncited by 10
  17. 2014Fighting Authorship Linkability with Crowdsourcingcited by 8
  18. 2020Differentially Private Adversarial Robustness Through Randomized Perturbationscited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 3; 2023: 9; 2024: 18; 2025: 21; 2026: 12.

63 papers.

Sort:

By area:

  1. 2023Privacy-Preserving In-Context Learning with Differentially Private Few-Shot Generationcited by 94
  2. 2024Differentially Private Synthetic Data via Foundation Model APIs 2: Textcited by 74
  3. 2022Differentially Private Language Models for Secure Data Sharingcited by 68
  4. 2023Locally Differentially Private Document Generation Using Zero Shot Promptingcited by 63
  5. 2023DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineercited by 56
  6. 2022DP-Rewrite: Towards Reproducibility and Transparency in Differentially Private Text Rewritingcited by 28
  7. 2023DP-BART for Privatized Text Rewriting under Local Differential Privacycited by 27
  8. 2024PrivacyRestore: Privacy-Preserving Inference in Large Language Models via Privacy Removal and Restorationcited by 27
  9. 2024DP-MLM: Differentially Private Text Rewriting Using Masked Language Modelscited by 24
  10. 2023Differentially Private Natural Language Models: Recent Advances and Future Directionscited by 20
  11. 2024Reconstruct Your Previous Conversations! Comprehensively Investigating Privacy Leakage Risks in Conversations with GPT Modelscited by 16
  12. 2025Synthesizing Privacy-Preserving Text Data via Finetuning without Finetuning Billion-Scale LLMscited by 15
38 more
  1. 2023Guiding Text-to-Text Privatization by Syntaxcited by 14
  2. 2024The Fire Thief Is Also the Keeper: Balancing Usability and Privacy in Promptscited by 13
  3. 2024Evaluating Differentially Private Synthetic Data Generation in High-Stakes Domainscited by 12
  4. 2024A Comparative Analysis of Word-Level Metric Differential Privacy: Benchmarking The Privacy-Utility Trade-offcited by 11
  5. 2024NAP^2: A Benchmark for Naturalness and Privacy-Preserving Text Rewriting by Learning from Humancited by 8
  6. 2025On the Impact of Noise in Differentially Private Text Rewritingcited by 8
  7. 2024Estimating Privacy Leakage of Augmented Contextual Knowledge in Language Modelscited by 6
  8. 2024Delving into Differentially Private Transformercited by 5
  9. 2024TAROT: Task-Oriented Authorship Obfuscation Using Policy Optimization Methodscited by 5
  10. 2025DP-Fusion: Token-Level Differentially Private Inference for Large Language Modelscited by 5
  11. 2025Evaluating Differentially Private Generation of Domain-Specific Textcited by 5
  12. 2025How do we measure privacy in text? A survey of text anonymization metricscited by 5
  13. 2024Reconstruction of Differentially Private Text Sanitization via Large Language Modelscited by 4
  14. 2023ConfusionPrompt: Practical Private Inference for Online Large Language Modelscited by 3
  15. 2024DiDOTS: Knowledge Distillation from Large-Language-Models for Dementia Obfuscation in Transcribed Speechcited by 3
  16. 2024Private Language Models via Truncated Laplacian Mechanismcited by 3
  17. 2025Leveraging Semantic Triples for Private Document Generation with Local Differential Privacy Guaranteescited by 3
  18. 2024A Collocation-based Method for Addressing Challenges in Word-level Metric Differential Privacycited by 2
  19. 2025Improving Privacy Benefits of Redactioncited by 2
  20. 2025DP-GTR: Differentially Private Prompt Protection via Group Text Rewritingcited by 2
  21. 2025Inspecting the Representation Manifold of Differentially-Private Textcited by 2
  22. 2025Privacy Preserving In-Context-Learning Framework for Large Language Modelscited by 2
  23. 2025Controlled Generation for Private Synthetic Textcited by 2
  24. 2023Defending Against Authorship Identification Attackscited by 1
  25. 2025Enhancing Privacy in the Early Detection of Sexual Predators Through Federated Learning and Differential Privacycited by 1
  26. 2025Investigating User Perspectives on Differentially Private Text Privatizationcited by 1
  27. 2025Empirical Calibration and Metric Differential Privacy in Language Modelscited by 1
  28. 2025Improved Algorithms for Differentially Private Language Model Alignmentcited by 1
  29. 2025The Double-edged Sword of LLM-based Data Reconstruction: Understanding and Mitigating Contextual Vulnerability in Word-level Differential Privacy Text Sanitizationcited by 1
  30. 2025Differentially-private text generation degrades output language qualitycited by 1
  31. 2026NOIR: Privacy-Preserving Generation of Code with Open-Source LLMscited by 1
  32. 2026OSNIP: Breaking the Privacy-Utility-Efficiency Trilemma in LLM Inference via Obfuscated Semantic Null Spacecited by 1
  33. 2026Adaptive Text Anonymization: Learning Privacy-Utility Trade-offs via Prompt Optimizationcited by 1
  34. 2026A Systematic Exploration of Text Decomposition and Budget Distribution in Differentially Private Text Obfuscationcited by 1
  35. 2022I still know it's you! On Challenges in Anonymizing Source Codecited by 0
  36. 2023Disentangling the Linguistic Competence of Privacy-Preserving BERTcited by 0
  37. 2024Silencing the Risk, Not the Whistle: A Semi-automated Text Sanitization Tool for Mitigating the Risk of Whistleblower Re-Identificationcited by 0
  38. 2024Characterizing Stereotypical Bias from Privacy-preserving Pre-Trainingcited by 0

50 most cited of 63 All 63 (JSON)