Image Credit: ACM SIGGRAPH 2025 poster Automatic Interpretation of Ancient Egyptian Texts for Education and Research by Maksim Golyadkin, Innokentiy Humonen, Ianis Plevokas, Ekaterina Bureeva, Ekaterina Aleksandrova, and Ilya Makarov. Example outputs at three levels (glyphs, words, and sentences) with Gardiner codes, transliterations, and translations.
Artificial intelligence is opening new pathways to understanding the past — and at SIGGRAPH 2025, one research team discovered just how far that potential can reach. Developed by scientists at AIRI Institute and ISP RAS in collaboration with Egyptologists and educators, SIGGRAPH 2025 Posters project “Automatic Interpretation of Ancient Egyptian Texts for Education and Research” brings modern computer vision and natural language processing to one of humanity’s oldest written languages. We sat down with the team behind the project, Maksim Golyadkin, Innokentiy Humonen, Ianis Plevokas, Ekaterina Bureeva, Ekaterina Aleksandrova, and Ilya Makarov, to understand their process and hear about their hope for the future.
SIGGRAPH: What inspired you to create a system focused on Ancient Egyptian texts?
We were inspired by the rapid growth of large language models (LLMs) and other large systems that can do many tasks without specific training. This growth raises a simple question: How can they actually help humanity? We see many areas where these models show potential, but performance is still not sufficient to make specialists’ work easier, and awareness of what the new technologies can do remains low. Consequently, we believe dedicated efforts are needed to integrate modern AI systems into such a domain as archaeology.
Specifically, studying ancient languages of past civilizations without living native speakers remains very challenging. This field is not very populated, there are not many specialists across a large variety of languages, and research could benefit a lot from more working hands, whether humans or machines.
We chose ancient Egyptian for several reasons. On one hand, the civilization itself is fascinating: Many people, as children, read about the “curse of the pharaohs,” watched “The Mummy,” or felt intrigued by mysterious hieroglyphs — and some of us still are. On the other hand, despite existing research in this area (such as Thoth AI[1] and work by Google DeepMind[2]), we identified a substantial gap between computer science and Egyptology in which our work can make a meaningful contribution.
In addition, our colleagues have achieved significant progress in optical character recognition (OCR) for old Russian language manuscripts in the project “Digital Peter.” However, their approaches were limited to the domains with richer labeled data.
Our team from AIRI Institute and ISP RAS decided to tackle this problem by combining our experience in modern computer vision, natural language processing, and synthetic data generation with domain experts from the Egyptology community. It turned out that our research group already had the necessary scientific background to reach state-of-the-art solutions, even for a task with such limited data.
Started as a project to create “Google Translate in camera mode” for museum visitors, the project transformed into creating labeling tools for archaeologists and ancient language specialists so that every new digitized data set leads to the improvement of the technology and, as a result, provides an AI-agentic framework for new knowledge discovery revealing the mysteries of the past.
SIGGRAPH: Your system unifies OCR, transliteration, and translation into one, but Ancient Egyptian isn’t just one script — it spans hieroglyphs, hieratic, and demotic. How does your approach manage these diverse writing styles, and what sets it apart from earlier attempts at digitally interpreting ancient languages?
Egyptologists solved the problem of different scripts with transliteration: They convert any script form into a common Latin-based representation. In transliteration, signs are written with letters and special markers instead of drawing the original signs. Because of this, all three scripts can be mapped into one shared textual form, which can then be used for search, linguistic analysis, and translation.
Our system follows this idea by using transliteration as the core representation. At the moment, because of training data limits, our recognition module works only on hieroglyphic texts, and our downstream analysis is also centered on hieroglyphic representation. However, the architecture is modular and flexible. If we extend the training data with hieratic and demotic examples (which we are currently doing), the same pipeline can support these scripts without any conceptual change.
We chose classical hieroglyphic writing as our starting point because it is simpler for OCR, more familiar to a broad audience, and historically earlier than the other script types. Building on this script, we plan to use our models to generate synthetic data for more complex scripts and thus advance OCR for large collections of untranslated and undigitized texts (for example, in the Museo Egizio in Turin[3]) that are still waiting for their moment to be integrated into the catalog of accessible knowledge and hold even greater scientific importance.
Finally, compared with earlier digital approaches, our focus is on practical use across varied real-world inputs, not only on narrow data sets. Many earlier works report good results on small, homogeneous collections. In contrast, we put extra effort into generalization so that the system works on different surfaces and capture conditions — stone reliefs, papyri, ostraca, museum photos, scans from books, and more.
SIGGRAPH: Open source invites global collaboration. How do you imagine students, educators, or hobbyists building on your system?
The success of AI is due to qualitative data. Only by bringing AI specialists and domain experts in the humanities with knowledge of ancient languages together can one truly build a system with sufficient quality.
AIRI’s mission is to create universal AI systems that solve real-world problems. Coupled with the expertise in building robust and trustworthy AI solutions at ISP RAS, we are able to develop novel machine-learning systems and benchmarks to track the progress of AI applications in multidisciplinary projects. While our core competence lies in computer science, our collaboration with experts from HSE University, the University of Cambridge, and other institutions allows us to achieve breakthrough results and integrate them into relevant research communities.
Thus, we invite every expert and hobbyist in ancient languages to join us and collaborate with our tool in analyzing, digitizing, and exploring Egyptian heritage.
We specifically note that our approach can be scaled to other languages by bringing leading experts in ancient cultures and AI together to find new discoveries in humanity’s past.
Focusing on education, we developed a web application that integrates the proposed labeling and OCR system, offering an interactive interface to increase accessibility, and introduced it into the educational process. First, the initial data set was annotated by Egyptology students, giving them intensive hands-on experience with reference literature and monument publications and allowing them to compare different publication methods in practice. From an educational perspective, the system’s key advantage is interactivity: It proposes potential interpretations, and students verify them. It is not a cheat sheet but a working partner with a human in the loop. Interactive engagement is more powerful, and one can envision curricula that help students at early stages train sign recognition on actual monuments with natural variation rather than on standardized typographic fonts. Even during development, we observed an educational effect: The system encouraged students to examine details more closely and to develop visual literacy, which is a crucial skill for any specialist. Moving forward, we plan to build more refined educational products on this foundational principle.
SIGGRAPH: Why is it important to make Ancient Egyptian writing more accessible to learners and researchers today?
First of all, Ancient Egyptian writing is simply cool: With modern AI, we suddenly have the chance to interact with a language that has been silent for thousands of years. What once seemed locked away in stone and papyrus can now be read, searched, and studied in ways that were unimaginable before. This makes the study of hieroglyphs not only exciting but also meaningful, because every improvement in accessibility opens the door to new historical findings, new cultural insights, and a deeper understanding of our ancestors.
For students, it makes hieroglyphic learning more interactive, engaging, and varied. For Egyptology as a discipline, this has significant implications — accelerating corpus digitization, enabling large-scale paleographic analysis previously impossible by hand, and creating new research opportunities in diachronic sign variation, regional scribal practices, and computational philology.
By assisting — not replacing — specialists in initial processing, it allows Egyptology researchers to focus on interpretation and analysis rather than mechanical transcription. This is particularly crucial for underpublished material in museum collections and archaeological sites, where documentation often lags decades behind excavation. Furthermore, these collections are disjointed and communication between institutions is limited, so creating a global platform would make it possible to conduct research on the complete body of data.
In the era of information overload, it’s critically important to distinguish authentic cultural artifacts from later imitations and modern interpretations. Currently, hieroglyphic inscriptions appear to the general public as indecipherable fields on museum artifacts and online photographs. Developing this technology will enable closer engagement with this dimension of Egyptian culture.
To sum up, by studying the thoughts and legacy of ancient civilizations with the help of our AI system, we can support Egyptologists in making new scholarly discoveries while also giving the wider public a more conscious way to engage with artifacts in museums. In this way, advanced technology becomes not only a research instrument but also a bridge between academic knowledge and cultural appreciation, helping to preserve and understand ancient Egyptian heritage in the modern world.
The figure represents the output of the model and contains inaccuracies reflecting different stages of the processing pipeline.
[1] https://somiyagawa.com/thoth
[2] De Cao, M., De Cao, N., Colonna, A., & Lenci, A. (2024, August). Deep learning meets egyptology: a hieroglyphic transformer for translating Ancient Egyptian. In Proceedings of the 1st Workshop on Machine Learning for Ancient Languages (ML4AL 2024) (pp. 71-86).
[3] https://museoegizio.it/en
As SIGGRAPH 2025 celebrates the intersection of art, science, and innovation, this project reminds us that AI’s greatest promise may not only lie in shaping the future, but also in revealing the depths of our shared history. Has this story sparked an idea in you? If you have a project you want to showcase at SIGGRAPH 2026, submissions are opening soon! Stay tuned for more information.
Maksim Golyadkin is receiving a PhD degree in computer science at the HSE University and works as a researcher in the field of Applied GenAI at AIRI Institute and ISP RAS.
Innokentiy Humonen holds a Master’s degree in Computational Linguistics from HSE University. He works as a researcher at AIRI Institute and ISP RAS, specializing in AI for ancient and low-resource languages. His work focuses on adapting and developing natural language processing methods for historical scripts and under-resourced linguistic data.
Ianis Plevokas holds a Master’s degree in Data Science from the Higher School of Economics, where his research focused on OCR and the translation of ancient languages. He is a Data Engineer at Selectel, leading ETL projects and building scalable data pipelines. His research and engineering interests lie at the intersection of computer vision, data systems, and computational linguistics.
Ekaterina Bureeva is an Egyptology student at HSE University. Her research focuses on bridging traditional historical and philological analysis with advanced digital technologies.
Ekaterina Aleksandrova holds a PhD in Cultural Studies and is an Associate Professor and Senior Research Fellow at the Faculty of Humanities, Institute for Oriental and Classical Studies, HSE University. She is the Academic Supervisor of the “Egyptology” Program and an Egyptologist and scholar of religion specializing in Old Kingdom Egypt and the Pyramid Texts. Her research integrates philology, semiotics, digital humanities (DH), and comparative religion to explore myth, ritual, and sacred communication, with a particular emphasis on applying quantitative methods to comparative mythology, such as corpus analysis (topic modelling; Thesaurus Linguae Aegyptiae), myth–ritual dynamics, and the interplay of verbal and visual imagery in ancient Egyptian funerary literature.
Ilya Makarov received the Specialist degree in mathematics from the Lomonosov Moscow State University, Moscow, Russia, and holds a PhD degree in computer science from the University of Ljubljana, Ljubljana, Slovenia.
Since 2011, he has been a lecturer at the School of Data Analysis and Artificial Intelligence, HSE University, where he was the School Deputy Head from 2012 to 2016. He holds Associate Professor and Senior Research Fellow positions at HSE, MIPT, ITMO, Innopolis, MePHI, ISP RAS and SHAD. Now, he is leading research fellow and leads of AI in the Archeology project at AIRI Institute.
