Elamur Research⌕
☰
home / papers / 2026 / Prompting the Past: Linguistic Transformations and Cultural Accuracy in AI-Generated Image Reconstructions for Multivocal Cultural Heritage

Prompting the Past: Linguistic Transformations and Cultural Accuracy in AI-Generated Image Reconstructions for Multivocal Cultural Heritage

Paper · 2026 · Ravini Wimalasuriya, Lea Krause and Gert-Jan Burgers

Summary

The paper asks whether text-to-image models can produce historical reconstructions that show the same heritage from several cultural viewpoints, in line with the critical-heritage idea of multivocality. The authors chose three contested cases: a Dutch colonial plantation in Suriname, Sri Pada Mountain in Sri Lanka (sacred to four faiths), and the 1640 VOC siege of Portuguese-held Galle Fort. They wrote keyword-based inputs for 13 perspectives. DALL-E 3 expanded these into full prompts, and the same expanded prompts were then given to Stable Diffusion V3.5 and Midjourney V.6.

The study had two parts. A linguistic analysis traced how the automatic rewriting step changed the users' content. Heritage experts then rated each image's accuracy. The rewriting step turned out to matter a great deal: it sometimes added useful detail, but it also dropped, swapped or softened culturally specific elements. All models were stronger on the religious case than on the colonial and military ones, and the outputs tended toward Western defaults.

The work matters for museums and digital-humanities projects. It shows that such tools can support exploratory visitor engagement but cannot replace experts or serve as historical sources.

Key points

  • The setup covered 3 models (DALL-E 3, Stable Diffusion V3.5, Midjourney V.6), 3 heritage cases and 13 perspectives, giving 39 images in total. Each case also included a "Random" perspective to simulate a user with little prior knowledge.
  • User inputs were organised into eight keyword categories: topic, landscape, architecture, people/attire, activities, atmosphere, special objects and artistic style. Experts scored each image on these same categories using a 1–5 scale.
  • The failure patterns identified were:
    • unequal representation of cultural groups;
    • neutralisation or amplification of sensitive content;
    • inconsistent depiction of people;
    • misreadings introduced while the prompt was being rewritten.
  • The authors propose a five-step workflow:
    1. Decide what the image is for and whose viewpoint it is meant to convey.
    2. Write equally specific keywords for all eight dimensions.
    3. Check the LLM-expanded prompt for omissions or softening before generating.
    4. Compare outputs across several models.
    5. Have experts review the results.
  • Main limitations: each case was judged by only one expert, keyword-gathering discussions involved just seven participants, and Stable Diffusion and Midjourney were not tested with their own native prompt pipelines.

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

🎓Vrije Universiteit AmsterdamAmsterdam · Netherlands43 works · 3,372 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArts and Humanities · 17,450 worksComputer Vision and Pattern Recognition · 3,486 worksArcheology · 64 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Ravini Wimalasuriya, Lea Krause and Gert-Jan Burgers. 2026. Prompting the Past: Linguistic Transformations and Cultural Accuracy in AI-Generated Image Reconstructions for Multivocal Cultural Heritage. Proceedings of the 6th International Conference on Natural Language Processing for the Digital Humanities.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.nlp4dh-1.9

@inproceedings{wimalasuriya2026prompting,
  title = {{Prompting the Past: Linguistic Transformations and Cultural Accuracy in AI-Generated Image Reconstructions for Multivocal Cultural Heritage}},
  author = {Ravini Wimalasuriya, Lea Krause and Gert-Jan Burgers},
  booktitle = {Proceedings of the 6th International Conference on Natural Language Processing for the Digital Humanities},
  year = {2026},
  doi = {10.18653/v1/2026.nlp4dh-1.9},
  url = {https://doi.org/10.18653/v1/2026.nlp4dh-1.9}
}

References

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

  1. Mark Altaweel, Adel Khelifi, and Mohammad Hashir Zafar. 2024. Using Generative AI for Reconstructing Cultural Artifacts: Examples Using Roman Coins. Journal of Computer Applications in Archaeology, 7(1):301–315. find
  2. David Kojan. 2008. Paths of Power and Politics: Historical Narratives at the Bolivian Site of Tiwanaku. In Junko Habu, Clare Fawcett, and John M. Matsunaga, editors, Evaluating Multiple Narratives: Beyond Nationalist, Colonialist, Imperialist Archaeolo-gies, pages 69–85. Springer New York, New York,… find
  3. Kawsar Arzomand, Michael Rustell, and Tatiana Kalganova. 2024. From ruins to reconstruction: Harnessing text-to-image AI for restoring historical architectures. Challenge Journal of Structural Mechanics, 10(2):69. find
  4. İzzettin Kutlu and Deryanur Şimşek. 2025. Artificial intelligence (AI) assisted digital reconstruction of historical buildings. In 4th International Civil Engineering & Architecture Conference Volume 2: Architecture. Golden Light Publishing. find
  5. Sonya Atalay. 2008. Multivocality and Indigenous Ar-chaeologies, pages 29–44. Springer New York, New York, NY. find
  6. Shanade Bianca Barnabas. 2016. Heritage-making and the dilemma of multivocality in south africa: a case of wildebeest kuil. International Journal of Heritage Studies, 22(9):690–701. find
  7. Shudong Liu, Yiqiao Jin, Cheng Li, Derek F. Wong, Qingsong Wen, Lichao Sun, Haipeng Chen, Xing Xie, and Jindong Wang. 2025. CultureVLM: Characterizing and Improving Cultural Understanding of Vision-Language Models for over 100 Countries. arXiv preprint. find
  8. James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, and 1 others. 2023. Improving image generation with better captions. Technical report. Available at https://cdn.openai.com/ papers/dall-e-3.pdf. ↗
  9. Midjourney, Inc. 2024. Midjourney. https://www.midjourney.com. Version 6. Accessed: [insert date]. ↗
  10. Jacques Derrida. 1994. Specters of Marx: The State of the Debt, the Work of Mourning, and the New International, 1st edition. Routledge, New York. find
  11. Fernando Moral-Andrés, Elena Merino-Gómez, Pedro Reviriego, and Fabrizio Lombardi. 2024. Can Artificial Intelligence Reconstruct Ancient Mosaics? Studies in Conservation, 69(5):313–326. find
  12. Ana Deumert. 2018. The multivocality of heritage – moments, encounters and mobilities. In Angela Creese and Adrian Blackledge, editors, The Routledge Handbook of Language and Superdiversity, 1 edition, pages 149–164. Routledge. find
  13. Jonas Oppenlaender. 2024. A taxonomy of prompt modifiers for text-to-image generation. Behaviour & Information Technology, 43(15):3763–3776. find
  14. Nitin Rane. 2023. Role and Challenges of ChatGPT and Similar Generative Artificial Intelligence in Arts and Humanities. SSRN Electronic Journal. find
  15. Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, Kyle Lacey, Alex Goodwin, Yannik Marek, and Robin Rombach. 2024. Scaling rectified flow transformers for high-r… find
  16. Claire Smith, editor. 2020. Encyclopedia of Global Archaeology. Springer International Publishing, Cham. find
  17. Laurajane Smith. 2006. Uses of Heritage, 1st edition. Routledge: Taylor & Francis Group, London. find
  18. Anna Foka and Gabriele Griffin. 2024. AI, Cultural Heritage, and Bias: Some Key Queries That Arise from the Use of GenAI. Heritage, 7(11):6125–6136. find
  19. Dirk H. R. Spennemann. 2024. Generative Artificial Intelligence, Human Agency and the Future of Cultural Heritage. Heritage, 7(7):3597–3609. find
  20. Ahmad W. Sukkar, Mohamed W. Fareed, Moohammed Wasim Yahia, Salem Buhashima Abdalla, Iman Ibrahim, and Khaldoun Abdul Karim Senjab. 2024. Analytical evaluation of midjourney architectural virtual lab: Defining major current limits in ai-generated representations of islamic architectural heritage. Bui… find
  21. Marta Vila, M. Antònia Martí, and Horacio Rodríguez. 2014. Is this a paraphrase? what kind? paraphrase boundaries and typology. Open Journal of Modern Linguistics, 4(1):205–218. find
  22. Yixin Wan, Arjun Subramonian, Anaelia Ovalle, Zongyu Lin, Ashima Suvarna, Christina Chance, Hritik Bansal, Rebecca Pattichis, and Kai-Wei Chang. 2024. Survey of Bias in Text-to-Image Generation: Definition, Evaluation, and Mitigation. arXiv preprint. find
  23. Linqing Wang, Ximing Xing, Yiji Cheng, Zhiyuan Zhao, Donghao Li, Tiankai Hang, Zhenxi Li, Jiale Tao, QiXun Wang, Ruihuang Li, Comi Chen, Xin Li, Mingrui Wu, Xinchi Deng, Shuyang Gu, Chunyu Wang, and Qinglin Lu. 2025. Promptenhancer: A simple approach to enhance text-to-image models via chain-of-thou… — [arXiv:2509.04545](https://arxiv.org/abs/2509.04545) find
  24. Alison Wylie. 2008. The Integrity of Narratives: Deliberative Practice, Pluralism, and Multivocality, pages 201–212. Springer New York, New York, NY. find
  25. Arman Zarei, Keivan Rezaei, Samyadeep Basu, Mehrdad Saberi, Mazda Moayeri, Priyatham Kattakinda, and Soheil Feizi. 2025. Improving Compositional Attribute Binding in Text-to-Image Generative Models via Enhanced Text Embeddings. arXiv preprint. find
  26. Lili Zhang, Xi Liao, Zaijia Yang, Baihang Gao, Chunjie Wang, Qiuling Yang, and Deshun Li. 2024. Partiality and Misconception: Investigating Cultural Representativeness in Text-to-Image Models. In Proceedings of the CHI Conference on Human Factors in Computing Systems, pages 1–25, Honolulu, HI, USA.… find
  27. Nana Zheng. 2023. Coming to Grips with Diverse Voices in Participatory Heritage Initiatives. PhD thesis, Vrije Universiteit Amsterdam. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wimalasuriya-2026-prompting-past-linguistic-transformations-cultural/