Elamur Research⌕
☰
home / papers / 2025 / Developing A German Document-Level Parallel Dataset For Automatic Text Simplification To Generate Easy Language

Developing A German Document-Level Parallel Dataset For Automatic Text Simplification To Generate Easy Language

Paper · 2025 · Vivien Jiranek and Stefan Hillmann

Summary

Automatic conversion of standard German into Leichte Sprache (GE2R) has been held back by a shortage of parallel training data. The authors address this by building EasyGerman, a document-level corpus of news articles from the public broadcaster MDR. Each regular-German article is paired with a professional Easy Language version, using the links the publisher itself provides. They profile the corpus linguistically and benchmark four fine-tuned setups: mBART and mt5, each initialised from either a summarization or a translation checkpoint. Evaluation combines readability formulas, overlap metrics (BLEU, ROUGE) and BLANC. A small reader study, which included people from the target group, judged the best system's output. All four models produced text whose readability matched the human GE2R references and was clearly simpler than the inputs. The mBART translation variant scored best, and readers found its output easy and faithful to Easy Language rules. Its weaknesses were uneven content coverage and logical slips. The resource matters because it gives a reproducible benchmark for neural Easy Language generation in German.

Key points

  • EasyGerman has 3,832 MDR articles from July 2021 to April 2024, forming 1,916 aligned document pairs. Sources are the MDR weekly Easy Language news page, the MDR News Archive and the Internet Archive.
  • To avoid copyright problems, the authors release URL lists and scraping/building scripts instead of the texts. This matters because MDR removes old links on a weekly or monthly schedule.
  • GE2R documents contain about half as many words as their regular-German counterparts and about a quarter of the vocabulary. The sentence count is nearly the same, but sentences average 9 words instead of 17.
  • mBART-tr had the best overlap scores, reaching ROUGE-1 0.48 and BLEU around 0.10–0.11. mBART-sum was the weakest. BLANC stayed low for every model.
  • Human raters were split into a regular-German group and a GE2R group according to self-reported reading difficulties. Both groups felt the source intent was mostly preserved.

Interesting ideas

  • I1. The authors align whole documents rather than sentences, because simplification omits, splits and reorders material, which makes sentence mappings unreliable.
  • I2. Readability scores of system output are placed between two anchors: the complex input and the human simplified reference. This shows both that the text got simpler and that it landed where a human would.
  • I3. A summarization-quality metric (BLANC) is used as a sanity check: low scores showed the models were simplifying rather than merely condensing.
  • I4. Corpus-level statistics give a quantitative target profile for the simpler variety. These include words per sentence, vocabulary ratio and a roughly unchanged sentence count.
  • I5. Human evaluators are grouped by self-reported reading ability, so readers who need simplified text are analysed separately from fluent readers.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (back): Generation from the meaning graph into simplified English could be judged at the document level, so that legitimate reordering of events is not penalised as an error.
  • I2 (back): The text → graph → text compressor could report its output's readability against the source and a gold rewrite, which makes "how much did we simplify" an explainable number.
  • I3 (learn): This works like a negative control. A metric that should stay low if the system does the intended task can catch training that drifts into a shortcut.
  • I4 (back): Such ratios could serve as explicit constraints for an easy-English generation mode, for example splitting clauses while keeping the number of propositions.
  • I5 (answer): When testing Mova's readers and answers on tales or school texts, results could be stratified by intended audience rather than averaged.

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

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Vivien Jiranek and Stefan Hillmann. 2025. Developing A German Document-Level Parallel Dataset For Automatic Text Simplification To Generate Easy Language. Proceedings of the 21st Conference on Natural Language Processing (KONVENS 2025): Workshops.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{jiranek2025developing,
  title = {{Developing A German Document-Level Parallel Dataset For Automatic Text Simplification To Generate Easy Language}},
  author = {Vivien Jiranek and Stefan Hillmann},
  booktitle = {Proceedings of the 21st Conference on Natural Language Processing (KONVENS 2025): Workshops},
  year = {2025},
  url = {https://aclanthology.org/2025.konvens-2.11/}
}

References

30 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Toni Amstad. 1978. Wie verständlich sind unsere Zeitungen? Ph.D. thesis, Universität Zürich. find
  2. Dennis Aumiller and Michael Gertz. 2022. Klexikon: A German dataset for joint summarization and simplification. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 2693–2701, Marseille, France. European Language Resources Association. find
  3. Richard Bamberger and Erich Vanecek. 1984. Lesen-verstehen-lernen-schreiben: Die schwierigkeitsstufen von texten in deutscher sprache. Jugend und Volk Verlagsgesellschaft. find
  4. Alessia Battisti and Sarah Ebling. 2019. A corpus for automatic readability assessment and text simplification of german. Preprint, arXiv:1909.09067. ↗
  5. Alessia Battisti, Dominik Pfütze, Andreas Säuberli, Marek Kostrzewa, and Sarah Ebling. 2020. A corpus for automatic readability assessment and text simplification of German. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 3302–3311, Marseille, France. European Language Resources Association. ↗
  6. Sofia Blinova, Xinyu Zhou, Martin Jaggi, Carsten Eickhoff, and Seyed Ali Bahrainian. 2023. SIMSUM: Document-level text simplification via simultaneous summarization. find
  7. Bundesministerium der Justiz. 2002. Behindertengle-ichstellungsgesetz abschnitt 2 verpflichtung zur gleichstellung und barrierefreiheit. § 11 verständlichkeit und leichte sprache. find
  8. Meri Coleman and T. L. Liau. 1975. A computer readability formula designed for machine scoring. Journal of Applied Psychology, 60:283–284. ↗
  9. Sarah Ebling, Alessia Battisti, Marek Kostrzewa, Dominik Pfütze, Annette Rios, Andreas Säuberli, and Nicolas Spring. 2022. Automatic text simplification for german. Frontiers in Communication, 7. ↗
  10. R. F. Flesch. 1948. A new readability yardstick. Journal of Applied Psychology, 32(3):221–223. ↗
  11. Silvia Hansen-Schirra, Jean Nitzke, and Silke Gutermuth. 2021. An intralingual parallel corpus of translations into german easy language (geasy corpus): What sentence alignments can tell us about translation strategies in intralingual translation. In Vincent X. Wang, Lily Lim, and Defeng Li, editors, New Perspectives on Corpus Translation Studies, New Frontiers in Translation Studies, pages 281–298. Springer. ↗
  12. Renate Hauser, Jannis Vamvas, Sarah Ebling, and Martin Volk. 2022. A multilingual simplified language news corpus. In Proceedings of the 2nd Workshop on Tools and Resources to Empower People with REAding DIfficulties (READI) within the 13th Language Resources and Evaluation Conference, pages 25–30, Marseille, France. European Language Resources Association. find
  13. Neslihan Iskender, Oleg V. Vasilyev, Tim Polzehl, John Bohannon, and Sebastian Möller. 2021. Towards human-free automatic quality evaluation of german summarization. CoRR, abs/2105.06027. ↗
  14. Sarah Jablotschkin, Elke Teich, and Heike Zinsmeister. 2024. DE-lite - a new corpus of easy German: Compilation, exploration, analysis. In Proceedings of the Fourth Workshop on Language Technology for Equality, Diversity, Inclusion, pages 106–117, St. Julian’s, Malta. Association for Computational Linguistics. find
  15. Sarah Jablotschkin and Heike Zinsmeister. 2020. Leiko: A corpus of easy-to-read german. In Poster presented at the Computational Linguistics Poster Session in the course of the 42nd annual conference of the Deutsche Gesellschaft für Sprachwissenschaft (DGfS) in Hamburg. Download of Poster and corpus from: https://doi.org/10.5281/zenodo, volume 3923917. ↗
  16. David Klaper, Sarah Ebling, and Martin Volk. 2013. Building a German/simple German parallel corpus for automatic text simplification. In Proc. 2nd Workshop on Predicting and Improving Text Readability for Target Reader Populations, pages 11–19, Sofia, Bulgaria. Association for Computational Linguistics. find
  17. Daisy Lange. 2018. Comparing ‘leichte sprache’,‘einfache sprache’and ‘leicht lesen’: A corpus-based descriptive approach. In Eds. SU-SANNE J. JEKAT, and GARY MASSEY. Barrier-free communication: methods and products: proceedings of the 1st Swiss conference on barrier-free communication. Winterthur: ZHAW Digital Collection, pages 75–91. find
  18. Chin-Yew Lin. 2004. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics.
  19. Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, and Luke Zettlemoyer. 2020. Multilingual denoising pre-training for neural machine translation. Transactions of the Association for Computational Linguistics, 8:726–742. ↗
  20. Margot Madina, Itziar Gonzalez-Dios, and Melanie Siegel. 2023. Easy-to-read in germany: A survey on its current state and available resources. ↗
  21. Babak Naderi, Salar Mohtaj, Kaspar Ensikat, and Sebastian Möller. 2019. Subjective assessment of text complexity: A dataset for german language. arXiv preprint arXiv:1904.07733. ↗
  22. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA. Association for Computational Linguistics. ↗
  23. Annette Rios, Nicolas Spring, Tannon Kew, Marek Kostrzewa, Andreas Säuberli, Mathias Müller, and Sarah Ebling. 2021. A new dataset and efficient baselines for document-level text simplification in German. In Proceedings of the Third Workshop on New Frontiers in Summarization, pages 152–161, Online and in Dominican Republic. Association for Computational Linguistics. find
  24. Andreas Säuberli, Sarah Ebling, and Martin Volk. 2020. Benchmarking data-driven automatic text simplification for German. In Proceedings of the 1st Workshop on Tools and Resources to Empower People with REAding DIfficulties (READI), pages 41–48, Marseille, France. European Language Resources Association. find
  25. Regina Stodden, Omar Momen, and Laura Kallmeyer. 2023. DEplain: A German parallel corpus with intralingual translations into plain language for sentence and document simplification. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16441–16463, Toronto, Canada. Association for Computational Linguistics. find
  26. Renliang Sun, Hanqi Jin, and Xiaojun Wan. 2021. Document-level text simplification: Dataset, criteria and baseline. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 7997–8013, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics. find
  27. Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, and Angela Fan. 2020. Multilingual translation with extensible multilingual pretraining and finetuning. CoRR, abs/2008.00401. ↗
  28. Vanessa Toborek, Moritz Busch, Malte Boßert, Christian Bauckhage, and Pascal Welke. 2022. A new aligned simple German corpus. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 11393–11412, Toronto, Canada. Association for Computational Linguistics. ↗
  29. Oleg Vasilyev, Vedant Dharnidharka, and John Bohannon. 2020. Fill in the blanc: Human-free quality estimation of document summaries. arXiv preprint arXiv:2002.09836. ↗
  30. Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mT5: A massively multilingual pre-trained text-to-text transformer. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 483–498, Online. Association for Computational Linguistics. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jiranek-2025-developing-german-document-level-parallel/