A team from Martin Luther University Halle-Wittenberg, Johannes Gutenberg University Mainz, and Mainz University of Applied Sciences has developed an AI system for deciphering ancient cuneiform texts.
The study, published in The Eurographics Association journal, focused on cuneiform tablets from the Frau Professor Hilprecht Collection, originating from ancient Mesopotamia.
The tablets, over 5,000 years old, contain symbols and wedges forming languages like Sumerian, Assyrian, and Akkadian, covering various topics from everyday life to legal matters.
Traditional deciphering methods are challenging due to heavy weathering, prompting the use of AI.
The AI model is based on the Region-based Convolutional Neural Network (R-CNN) architecture and utilizes a dataset with 3D models of 1,977 cuneiform tablets and detailed annotations of 21,000 cuneiform signs and 4,700 wedges.
The AI process involves a two-part pipeline: a sign detector based on the RepPoints model identifies cuneiform characters, and a wedge detector using Point R-CNN predicts wedge positions, enabling the AI to 'read' the text.
The 3D scans of tablets provide nuanced measurements, overcoming challenges posed by 2D photographs in traditional research methods.
The AI system underwent extensive training with three-dimensional scans and supplemental data, contributing to its success in accurately identifying symbols on the tablets.
The technology democratizes access to ancient records and opens new avenues for research, potentially extending its application to other three-dimensional scripts, such as weathered inscriptions in cemeteries.
The research marks a notable progress in the decoding of ancient texts, showcasing the potential of AI in archaeology and linguistic research.