Elamur Research⌕
☰
home / papers / 2026 / Revisiting Masking After Fifteen Years: Early Approaches to Non-Reconstructable Linguistic Data in the current context

Revisiting Masking After Fifteen Years: Early Approaches to Non-Reconstructable Linguistic Data in the current context

Paper · 2026 · Georg Rehm, Thorsten Trippel and Andreas Witt

Summary

This position paper revisits two 2007 proposals by the same authors for sharing annotated corpora whose source texts cannot be legally redistributed due to copyright, contracts or privacy. Both proposals mask the text but keep the annotations. The first, presented at TLT 2007, hid the word forms in treebanks while keeping syntactic and morphological structure. The second, presented at DH 2007, extended this to any XML-annotated corpus using configurable random string substitution.

The authors place these ideas within today's Derived Text Formats (DTFs). They argue that LLMs, FAIR principles and new standards now call for more formal methods. These methods should define explicit abstraction levels and reversibility classes, and record provenance in a machine-actionable way. The paper sketches a modern DTF generation workflow, illustrated with the MONA-pipe framework, and covers use cases in linguistics, the digital humanities and NLP. It concludes that DTFs offer a sustainable, infrastructure-ready route to research that is reproducible and legally safe. This matters for anyone who needs to publish benchmarks or derived datasets from protected texts.

Key points

  • Original 2007 masking methods:
    • Masked treebanks swapped tokens for neutral placeholders but kept constituent trees, dependencies and grammatical features.
    • Corpus masking could optionally keep token length, capitalisation and affixes.
    • Both were implemented in the CorpusMasker tool (Dellert, 2007, v0.1).
  • Standards that support DTFs: TEI, ISO annotation models (MAF, LAF, SynAF, SemAF, LMF), CMDI metadata, PISA, and the draft DIN 19,461 DTF specification.
  • New risk from LLMs: LLMs could partially reconstruct masked text from structural or distributional cues. Du et al. (2025) find that reconstruction is still poor, but caution is warranted as models improve.
  • Legal context:
    • GDPR limits sharing of data that contains personal information.
    • The German TDM exception (§ 44b UrhG) allows model training but not redistribution of the corpus, so non-reversible derived formats are needed for reproducibility.
  • Infrastructure context: The relevant infrastructures named are Text+, NFDI, EOSC, CLARIN, META-SHARE, European Language Grid and the European Language Data Space.

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

🔬German Research Centre for Artificial IntelligenceKaiserslautern · Germany77 works · 29,479 citations🔬Leibniz Institute for the German LanguageMannheim · Germany5 works · 22 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksInformation Systems · 1,437 worksLiterature and Literary Theory · 519 works

Cite

Cite as Georg Rehm, Thorsten Trippel and Andreas Witt. 2026. Revisiting Masking After Fifteen Years: Early Approaches to Non-Reconstructable Linguistic Data in the current context. Proceedings of Leveraging Derived Text Formats to Unlock Copyrighted Collections for Open Science (DTF) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/4ywpnvfa3znb

@inproceedings{rehm2026revisiting,
  title = {{Revisiting Masking After Fifteen Years: Early Approaches to Non-Reconstructable Linguistic Data in the current context}},
  author = {Georg Rehm, Thorsten Trippel and Andreas Witt},
  booktitle = {Proceedings of Leveraging Derived Text Formats to Unlock Copyrighted Collections for Open Science (DTF) @ LREC 2026},
  year = {2026},
  doi = {10.63317/4ywpnvfa3znb},
  url = {https://doi.org/10.63317/4ywpnvfa3znb}
}

References

20 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Johannes Dellert. 2007. Corpusmasker: A tool for parameterised masking of linguistic resources. https://www.lingexp.uni-tuebingen.de/sfb441/c2/corpus-masker-0.1.tar. gz. Version 0.1, SFB 441 “Linguistic Data Structures”, University of Tübingen. ↗
  2. ISO 24615-1:2014. 2014. Language resource management — syntactic annotation framework (SynAF) — part 1: Syntactic model. International Standard, International Organisation for Standardizaton (ISO), Geneva. find
  3. DIN 19461:2026-04 (E). 2026. Sprachressourcen und Sprachtechnologie - Abgeleitete Textformate (ATF). Technical report, Deutsches Institut für Normung, Berlin. find
  4. ISO 24617-1:2012. 2012. Language resource management — semantic annotation framework (SemAF) — part 1: Time and events (semaf-time, iso-timeml). International Standard, International Organisation for Standardization (ISO), Geneva. find
  5. Tillmann Dönicke, Florian Barth, Hanna Varachkina, and Caroline Sporleder. 2022. Monapipe: Modes of narration and attribution pipeline for german computational literary studies and language analysis in spacy. In Proceedings of the 18th Conference on Natural Language Processing (KONVENS 2022), Potsda… find
  6. ISO 24619:2011. 2011. Language resource management – Persistent identification and sustainable access (PISA). International Standard, International Organization for Standardization (ISO), Geneva. find
  7. ISO 24622-1:2015. 2015. Language resource management – Component Metadata Infrastructure (CMDI) – Part 1: The Component Metadata Model. International Standard, International Organization for Standardization (ISO), Geneva. find
  8. Keli Du, Sarah Ackerschewski, Uygar Navruz, Nazan Sınır, Julian Valline, and Christof Schöch. 2025. Reconstructing shuffled text. bad results for nlp, but good news for using in-copyright texts. Journal of Computational Literary Studies, 4(1). find
  9. ISO 24622-2:2019. 2019. Language resource management – Component Metadata Infrastructure (CMDI) – Part 2: The Component Metadata Specification Language. International Standard, International Organization for Standardization (ISO), Geneva. find
  10. European Parliament and Council of the European Union. 2026. Regulation (eu) 2016/679 of the european parliament and of the council of 27 april 2016 on the protection of natural persons with regard to the processing of personal data and Timm Lehmberg, Christian Chiarcos, Georg Rehm, and Andreas Witt… find
  11. Timm Lehmberg, Georg Rehm, Andreas Witt, and Felix Zimmermann. 2008. Digital text collections, linguistic research data, and mashups: Notes on the legal situation. Library Trends, 57(1):52 – 71. find
  12. MONAPipe 2023. 2023. Monapipe: Modes of narration and attribution pipeline. GitLab repository. Accessed: 2026-03-20. find
  13. Stelios Piperidis, Harris Papageorgiou, Christian Spurk, Georg Rehm, Khalid Choukri, Olivier Hamon, Nicoletta Calzolari, Riccardo del Gratta, Bernardo Magnini, and Christian Girardi. 2014. META-SHARE: One year after. In Proceedings of the 9th Language Resources and Evaluation Conference (LREC 2014),… find
  14. Christof Schöch, Frédéric Döhl, Achim Rettinger, Evelyn Gius, Peer Trilcke, Peter Leinen, Fotis Jannidis, Maria Hinzmann, and Jörg Röpke. 2020. Abgeleitete textformate: Text und data mining mit urheberrechtlich geschützten textbeständen. Zeitschrift für digitale Geisteswissenschaften (ZfdG), 5. find
  15. Text+ on TDM. 2025. Why the training of large language models falls within the scope of the text and data mining exceptions. Last modified: 23 January 2025. find
  16. UrhG. 2021. § 44b urhg – text und data mining. https://www.gesetze-im-internet.de/urhg/__44b.html. Zugriff am 16. ↗
  17. Georg Rehm, editor. 2023. European Language Grid: A Language Technology Platform for Multilingual Europe. Cognitive Technologies. Springer, Cham, Switzerland. find
  18. Georg Rehm, Stelios Piperidis, Khalid Choukri, Andrejs Vasil , jevs, Katrin Marheinecke, Victoria Arranz, Aivars Bērzin , š, Miltos Deligiannis, Dimitrios Galanis, Maria Gavriilidou, Maria Giagkou, Katerina Gkirtzou, Dimitris Gkoumas, Annika Grützner-Zahn, Athanasia Kolovou, Penny Labropoulou, Andi… find
  19. Georg Rehm, Andreas Witt, Heike Zinsmeister, and Johannes Dellert. 2007a. Corpus masking: Legally bypassing licensing restrictions for the free distribution of text collections. In Digital Humanities 2007. Conference Abstracts, pages 166–170, Urbana-Champaign. University of Illinois. find
  20. Georg Rehm, Andreas Witt, Heike Zinsmeister, and Johannes Dellert. 2007b. Masking treebanks find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rehm-2026-revisiting-masking-after-fifteen-years/