Russian Scientists Taught AI to Read Ancient Hieroglyphs
An interdisciplinary team from AIRI and the Higher School of Economics (HSE) has created an advanced technology designed to work with one of the most complex ancient languages. Recognizing hieroglyphs is an extremely challenging task, as the system must interpret not just 30–40 standard letters but hundreds of unique symbols.
To train the AI model, scientists used the extensive Thesaurus Linguae Aegyptiae database — the most comprehensive collection of ancient Egyptian texts with translations.
To achieve high accuracy, the researchers applied:
• Context-dependent OCR models (optical character recognition).
• Diffusion models that generated realistic images to expand the training dataset. The resulting translations, initially evaluated by automated metrics, were later verified and confirmed by professional Egyptologists.
According to the developers, the project aims to create new handwriting recognition technologies — a crucial step in addressing data scarcity and improving large language models.
In the future, such technologies could be integrated into augmented reality (AR) mobile applications, allowing museum visitors to simply point their smartphone cameras at hieroglyph-covered artifacts and instantly see translations. The system will also greatly simplify the learning process for Egyptology students.
Access to the system is already available upon request for interested specialists, and the first scientific paper on this methodology was presented at the SIGGRAPH 2025 international conference.
ORIENT
