Elamur Research⌕
☰
home / papers / 2026 / Linguistic Identity Leakage: When Language Reveals Identity in Anonymized Text

Linguistic Identity Leakage: When Language Reveals Identity in Anonymized Text

Paper · 2026 · Wajdi Zaghouani

Summary

Conventional text anonymization removes names, addresses and other explicit identifiers. This position paper argues that the language of a document still exposes its author. Dialect vocabulary, morphology, code-switching habits and stylometric regularities can point to a person's region, social background or community after redaction. The author calls this residual risk Linguistic Identity Leakage (LIL) and the features that carry it Linguistic Personally Identifiable Information (L-PII).

The paper draws on sociolinguistics, stylometry and NLP privacy research. It sketches a preliminary five-category taxonomy of L-PII, illustrated mainly with Arabic dialect data. It also proposes the Identity Inference Risk (IIR) framework for auditing datasets and models for leftover identity risk. Current LLMs, it argues, make such inference cheap and scalable.

The work reports no experiments. Its contribution is conceptual, and IIR has not been implemented. Its value lies in exposing a blind spot of span-level de-identification, especially for small, linguistically distinctive communities. It recommends that dataset documentation and LLM privacy testing account for this risk.

Key points

  • Running example: a Tunisian sentence that mixes dialect and French passes NER-based anonymization untouched. It still contains three identity cues that an LLM or dialect classifier could exploit:
    • the Tunisian word "barsha";
    • a Maghrebi n- verb prefix;
    • a switch into French.
  • Cited evidence on inference and obfuscation:
    • Staab et al. (2024) found that LLMs infer location, income and sex from Reddit profiles with up to 85% top-1 accuracy, and that anonymization and alignment did not prevent this.
    • Zhai et al. (2022) showed adversarial attributors cut obfuscation success from 20–30% to 5–10%.
  • Cited evidence on dialect granularity: the MADAR Arabic Dialect Corpus covers 25 cities, implying written dialect may reveal a writer's city, not only their country.
  • Recommendations:
    • treat dialect, sociolect and style as quasi-identifiers in Datasheets for Datasets;
    • accompany releases with an IIR audit giving per-category risk scores;
    • restrict dialect-rich subsets to vetted researchers;
    • test LLMs for demographic inference, beyond memorization and membership inference.
  • Limitations stated by the author:
    • there is no empirical measurement of LIL;
    • the taxonomy comes from conceptual synthesis rather than the Nickerson et al. (2013) method;
    • the idea that generative AI could flatten stylistic variation and thereby lower LIL has not been tested.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Northwestern University in QatarDoha · Qatar39 works · 148 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksSociology and Political Science · 626 works

Cite

Cite as Wajdi Zaghouani. 2026. Linguistic Identity Leakage: When Language Reveals Identity in Anonymized Text. Proceedings of the Seventh Workshop on Privacy in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.privatenlp-main.8

@inproceedings{zaghouani2026linguistic,
  title = {{Linguistic Identity Leakage: When Language Reveals Identity in Anonymized Text}},
  author = {Wajdi Zaghouani},
  booktitle = {Proceedings of the Seventh Workshop on Privacy in Natural Language Processing},
  year = {2026},
  doi = {10.18653/v1/2026.privatenlp-main.8},
  url = {https://doi.org/10.18653/v1/2026.privatenlp-main.8}
}

References

23 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Muhammad Abdul-Mageed, Amr Keleg, AbdelRahim Elmadany, Chiyu Zhang, Injy Hamed, Walid Magdy, Houda Bouamor, and Nizar Habash. NADI 2024: The fifth nuanced Arabic dialect identification shared task. In Proceedings of The Second Arabic Natural Language Processing Conference, pages 709–728, Bangkok, Th…
  2. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, and Kemal Oflazer. The MADAR Arabic dialect corpus and lexicon. In Proceedings of the Eleventh International Conference on Language Resources an… find
  3. Nicholas Carlini, Florian Tramèr, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, Úlfar Erlingsson, Alina Oprea, and Colin Raffel. Extracting training data from large language models. In Proceedings of the 30th USENIX Security Symposium (USENIX… find
  4. Springer, Berlin, Heidelberg, 2006. find
  5. Oluwaseyi Feyisetan, Borja Balle, Thomas Drake, and Tom Diethe. Privacy- and utility-preserving textual analysis via calibrated multivariate perturbations. In Proceedings of the 13th International Conference on Web Search and Data Mining (WSDM 2020), pages 178–186. find
  6. ACM, 2020. https://dl.acm.org/doi/ 10.1145/3336191.3371856 — [doi:10.1145/3336191.3371856](https://doi.org/10.1145/3336191.3371856) ↗
  7. Timnit Gebru, Jamie Morgenstern, Briana Vecchione, Jennifer Wortman Vaughan, Hanna Wallach, Hal Daumé III, and Kate Crawford. Datasheets for datasets. Communications of the ACM, 64(12):86– 92, 2021. https://dl.acm.org/doi/10.1145/ 3458723 ↗
  8. Dhiman Goswami, Jai Kruthunz Naveen Kumar, and Sanchari Das. NLP privacy risk identification in social media (NLP-PRISM): A survey. arXiv preprint arXiv:2602.15866, 2026. https://arxiv.org/ abs/2602.15866 ↗
  9. Clive Holes. Modern Arabic: Structures, Functions, and Varieties. Georgetown University Press, Washington, D.C., revised edition, 2004. find
  10. Timour Igamberdiev and Ivan Habernal. DP-BART for privatized text rewriting under local differential privacy. In Findings of the Association for Computational Linguistics: ACL 2023, pages 13914–13934, Toronto, Canada, 2023. Association for Computational Linguistics. https://aclanthology.org/ 2023.fi… ↗
  11. Oleksandra Klymenko, Stephen Meisenbacher, and Florian Matthes. Differential privacy in natural language processing: The story so far. In Proceedings of the Fourth Workshop on Privacy in Natural Language Processing, pages 1–11, Seattle, United States, 2022. Association for Computational Linguistics.… find
  12. William Labov. Sociolinguistic Patterns. University of Pennsylvania Press, Philadelphia, PA, 1972. find
  13. Pierre Lison, Ildikó Pilán, David Sanchez, Montserrat Batet, and Lilja Øvrelid. Anonymisation models for text data: State of the art, challenges and future directions. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conferen… find
  14. Justus Mattern, Benjamin Weggenmann, and Florian Kerschbaum. The limits of word level differential privacy. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 867–881, Seattle, United States, 2022. Association for Computational Linguistics. https://aclanthology.org/ 2022… ↗
  15. Jonathan Schler, Moshe Koppel, Shlomo Argamon, and James W. Pennebaker. Effects of age and gender on blogging. In Computational Approaches to Analyzing Weblogs, Papers from the 2006 AAAI Spring Symposium, Technical Report SS-06-03, pages 199– 205, Stanford, CA, 2006. AAAI. find
  16. Robin Staab, Mark Vero, Mislav Balunović, and Martin Vechev. Beyond memorization: Violating privacy via inference with large language models. In The Twelfth International Conference on Learning Representations (ICLR 2024), 2024. https://openreview.net/forum?id=kmn0BhQk7p Robin Staab, Mark Vero, Mi… ↗
  17. Efstathios Stamatatos. A survey of modern authorship attribution methods. Journal of the American Society for Information Science and Technology, 60(3):538– 556, 2009. https://onlinelibrary.wiley.com/ doi/10.1002/asi.21001 — [doi:10.1002/asi.21001](https://doi.org/10.1002/asi.21001) ↗
  18. Latanya Sweeney. k-anonymity: A model for protecting privacy. International Journal on Uncertainty, Fuzziness and Knowledge-Based Systems, 10(5):557– 570, 2002. https://dl.acm.org/doi/10.1142/ S0218488502001648 ↗
  19. Wanyue Zhai, Jonathan Rusert, Zubair Shafiq, and Padmini Srinivasan. A girl has a name, and it’s . . . adversarial authorship attribution for deobfuscation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7372–7384, Dublin, Ir… find
  20. Carol Myers-Scotton. Social Motivations for Codeswitching: Evidence from Africa. Clarendon Press, Oxford, 1993. find
  21. Tempestt Neal, Kalaivani Sundararajan, Aneez Fatima, Yiming Yan, Yingfei Xiang, and Damon Woodard. Surveying stylometry techniques and applications. ACM Computing Surveys, 50(6):86:1–86:36, 2017. https://dl.acm.org/doi/10.1145/3132039 — [doi:10.1145/3132039](https://doi.org/10.1145/3132039) ↗
  22. Robert C. Nickerson, Upkar Varshney, and Jan Muntermann. A method for taxonomy development and its application in information systems. European Journal of Information Systems, 22(3):336–359, 2013. https://doi.org/10.1057/ejis.2012.26 — [doi:10.1057/ejis.2012.26](https://doi.org/10.1057/ejis.2012.26) ↗
  23. Ildikó Pilán, Pierre Lison, Lilja Øvrelid, Anthi Papadopoulou, David Sánchez, and Montserrat Batet. The text anonymization benchmark (TAB): A dedicated corpus and evaluation framework for text anonymization. Computational Linguistics, 48(4): 1053–1101, 2022. https://aclanthology.org/ 2022.cl-4.19 ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zaghouani-2026-linguistic-identity-leakage-when-language/