Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
157 tokens/sec
GPT-4o
43 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Disentangling the Linguistic Competence of Privacy-Preserving BERT (2310.11363v1)

Published 17 Oct 2023 in cs.CL

Abstract: Differential Privacy (DP) has been tailored to address the unique challenges of text-to-text privatization. However, text-to-text privatization is known for degrading the performance of LLMs when trained on perturbed text. Employing a series of interpretation techniques on the internal representations extracted from BERT trained on perturbed pre-text, we intend to disentangle at the linguistic level the distortion induced by differential privacy. Experimental results from a representational similarity analysis indicate that the overall similarity of internal representations is substantially reduced. Using probing tasks to unpack this dissimilarity, we find evidence that text-to-text privatization affects the linguistic competence across several formalisms, encoding localized properties of words while falling short at encoding the contextual relationships between spans of words.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (77)
  1. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC conference on computer and communications security, pages 308–318.
  2. Higher-order comparisons of sentence encoder representations. arXiv preprint arXiv:1909.00303.
  3. Fine-grained analysis of sentence embeddings using auxiliary prediction tasks. arXiv preprint arXiv:1608.04207.
  4. Guiding text-to-text privatization by syntax. In Proceedings of the 3rd Workshop on Trustworthy Natural Language Processing (TrustNLP 2023), pages 151–162, Toronto, Canada. Association for Computational Linguistics.
  5. Private empirical risk minimization: Efficient algorithms and tight error bounds. In 2014 IEEE 55th annual symposium on foundations of computer science, pages 464–473. IEEE.
  6. What do neural machine translation models learn about morphology? arXiv preprint arXiv:1704.03471.
  7. Plausible deniability for privacy-preserving data synthesis. arXiv preprint arXiv:1708.07975.
  8. Deep rnns encode soft hierarchical syntax. arXiv preprint arXiv:1805.04218.
  9. On identifiability in transformers. arXiv preprint arXiv:1908.04211.
  10. The secret sharer: Evaluating and testing unintended memorization in neural networks. In 28th USENIX Security Symposium (USENIX Security 19), pages 267–284.
  11. Broadening the scope of differential privacy using metrics. In International Symposium on Privacy Enhancing Technologies Symposium, pages 82–102. Springer.
  12. Probing bert in hyperbolic spaces. arXiv preprint arXiv:2104.03869.
  13. A customized text sanitization mechanism with differential privacy. In Findings of the Association for Computational Linguistics: ACL 2023, pages 5747–5758, Toronto, Canada. Association for Computational Linguistics.
  14. What does bert look at? an analysis of bert’s attention. arXiv preprint arXiv:1906.04341.
  15. What you can cram into a single vector: Probing sentence embeddings for linguistic properties. arXiv preprint arXiv:1805.01070.
  16. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.
  17. Calibrating noise to sensitivity in private data analysis. In Theory of cryptography conference, pages 265–284. Springer.
  18. Amnesic probing: Behavioral explanation with amnesic counterfactuals. Transactions of the Association for Computational Linguistics, 9:160–175.
  19. Allyson Ettinger. 2020. What bert is not: Lessons from a new suite of psycholinguistic diagnostics for language models. Transactions of the Association for Computational Linguistics, 8:34–48.
  20. Generalised differential privacy for text document processing. In International Conference on Principles of Security and Trust, pages 123–148. Springer, Cham.
  21. Privacy-and utility-preserving textual analysis via calibrated multivariate perturbations. In Proceedings of the 13th International Conference on Web Search and Data Mining, pages 178–186.
  22. Yoav Goldberg. 2019. Assessing bert’s syntactic abilities. arXiv preprint arXiv:1901.05287.
  23. Colorless green recurrent networks dream hierarchically. arXiv preprint arXiv:1803.11138.
  24. Conditional probing: measuring usable information beyond a baseline. arXiv preprint arXiv:2109.09234.
  25. John Hewitt and Percy Liang. 2019. Designing and interpreting probes with control tasks. arXiv preprint arXiv:1909.03368.
  26. John Hewitt and Christopher D Manning. 2019. A structural probe for finding syntax in word representations. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4129–4138.
  27. Do attention heads in bert track syntactic dependencies? arXiv preprint arXiv:1911.12246.
  28. Visualisation and’diagnostic classifiers’ reveal how recurrent and recursive neural networks process hierarchical structure. Journal of Artificial Intelligence Research, 61:907–926.
  29. Contrastive explanations for model interpretability. arXiv preprint arXiv:2103.01378.
  30. Sarthak Jain and Byron C Wallace. 2019. Attention is not explanation. arXiv preprint arXiv:1902.10186.
  31. What does bert learn about the structure of language? In ACL 2019-57th Annual Meeting of the Association for Computational Linguistics.
  32. Jae-young Jo and Sung-Hyon Myaeng. 2020. Roles and utilization of attention heads in transformer-based neural language models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3404–3417.
  33. Revealing the dark secrets of bert. arXiv preprint arXiv:1908.08593.
  34. Representational similarity analysis-connecting the branches of systems neuroscience. Frontiers in systems neuroscience, 2:4.
  35. The emergence of number and syntax units in lstm language models. arXiv preprint arXiv:1903.07435.
  36. Albert: A lite bert for self-supervised learning of language representations. arXiv preprint arXiv:1909.11942.
  37. Multi-head attention with disagreement regularization. arXiv preprint arXiv:1810.10183.
  38. Tomasz Limisiewicz and David Mareček. 2020. Introducing orthogonal constraint in structural probes. arXiv preprint arXiv:2012.15228.
  39. Open sesame: getting inside bert’s linguistic knowledge. arXiv preprint arXiv:1906.01698.
  40. Assessing the ability of lstms to learn syntax-sensitive dependencies. Transactions of the Association for Computational Linguistics, 4:521–535.
  41. Linguistic knowledge and transferability of contextual representations. arXiv preprint arXiv:1903.08855.
  42. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
  43. Rebecca Marvin and Tal Linzen. 2018. Targeted syntactic evaluation of language models. arXiv preprint arXiv:1808.09031.
  44. The limits of word level differential privacy. arXiv preprint arXiv:2205.02130.
  45. Learned in translation: Contextualized word vectors. Advances in neural information processing systems, 30.
  46. Learning differentially private recurrent language models. arXiv preprint arXiv:1710.06963.
  47. What happens to bert embeddings during fine-tuning? arXiv preprint arXiv:2004.14448.
  48. On the nature of bert: Correlating fine-tuning and linguistic competence. In Proceedings of the 29th International Conference on Computational Linguistics, pages 3109–3119.
  49. Pointer sentinel mixture models.
  50. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781.
  51. Privacy risks of general-purpose language models. In 2020 IEEE Symposium on Security and Privacy (SP), pages 1314–1331. IEEE.
  52. Deep contextualized word representations.
  53. Dissecting contextual word embeddings: Architecture and representation. arXiv preprint arXiv:1808.08949.
  54. Natural language understanding with privacy-preserving bert. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 1488–1497.
  55. Improving language understanding by generative pre-training.
  56. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9.
  57. Probing the probing paradigm: Does probing accuracy entail task relevance? arXiv preprint arXiv:2005.00719.
  58. Attention can reflect syntactic structure (if you let it). arXiv preprint arXiv:2101.10927.
  59. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108.
  60. Naomi Saphra and Adam Lopez. 2018. Understanding learning dynamics of language models with svcca. arXiv preprint arXiv:1811.00225.
  61. Sofia Serrano and Noah A Smith. 2019. Is attention interpretable? arXiv preprint arXiv:1906.03731.
  62. Selective differential privacy for language modeling. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2848–2859, Seattle, United States. Association for Computational Linguistics.
  63. Congzheng Song and Ananth Raghunathan. 2020. Information leakage in embedding models. In Proceedings of the 2020 ACM SIGSAC Conference on Computer and Communications Security, pages 377–390.
  64. Stochastic gradient descent with differentially private updates. In 2013 IEEE Global Conference on Signal and Information Processing, pages 245–248. IEEE.
  65. Investigating transferability in pretrained language models. arXiv preprint arXiv:2004.14975.
  66. Bert rediscovers the classical nlp pipeline. arXiv preprint arXiv:1905.05950.
  67. What do you learn from context? probing for sentence structure in contextualized word representations. arXiv preprint arXiv:1905.06316.
  68. Investigating the impact of pre-trained word embeddings on memorization in neural networks. In International Conference on Text, Speech, and Dialogue, pages 273–281. Springer.
  69. Attention is all you need. Advances in neural information processing systems, 30.
  70. Jesse Vig and Yonatan Belinkov. 2019. Analyzing the structure of attention in a transformer language model. arXiv preprint arXiv:1906.04284.
  71. The bottom-up evolution of representations in the transformer: A study with machine translation and language modeling objectives. arXiv preprint arXiv:1909.01380.
  72. Glue: A multi-task benchmark and analysis platform for natural language understanding. arXiv preprint arXiv:1804.07461.
  73. Stanley L Warner. 1965. Randomized response: A survey technique for eliminating evasive answer bias. Journal of the American Statistical Association, 60(309):63–69.
  74. A non-linear structural probe. arXiv preprint arXiv:2105.10185.
  75. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144.
  76. A differentially private text perturbation method using a regularized mahalanobis metric. arXiv preprint arXiv:2010.11947.
  77. Differential privacy for text analytics via natural text sanitization. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3853–3866, Online. Association for Computational Linguistics.

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now