A team from Martin Luther University Halle-Wittenberg, Johannes Gutenberg University Mainz, and Mainz University of Applied Sciences has revealed an AI system capable of deciphering ancient cuneiform texts.
Unlocking Ancient Scripts
Researchers, via the Eurographics Association journal, delved into deciphering ancient cuneiform tablets from the Frau Professor Hilprecht Collection.
These artifacts, hailing from Mesopotamia, hold languages like Sumerian, Assyrian, and Akkadian, originating in what is often termed the cradle of civilization – present-day Iraq.
AI-Powered Process
Using a pioneering AI process based on the Region-based Convolutional Neural Network (R-CNN) architecture, they employed a dataset of 1,977 3D tablet models, annotating 21,000 signs and 4,700 wedges.
This AI system, employing an intricate two-step pipeline with a RepPoints model and ResNet18 backbone, effectively identified and interpreted the intricate text inscribed on these ancient relics.
Revolutionizing Analysis
- The AI’s accuracy in pinpointing signs was crucial.
- Then, a wedge detector using Point R-CNN, featuring advancements like Feature Pyramid Network (FPN) and RoI Align, predicted wedge positions, forming the foundation of the cuneiform script.
- This allowed the AI to effectively ‘read’ the text by interpreting these fundamental elements.
Challenges Overcome
By utilizing 3D scans, these tools analyzed measurements like impression depth and symbol distances on the tablets, overcoming issues like inconsistent lighting and color distractions present in traditional 2D photographs.
Unlike Optical Character Recognition (OCR) commonly used for ancient texts, which struggles with cuneiform due to lighting and viewing angles, this method offered a more accurate analysis. Ernst Stötzner, co-author, highlighted the challenge: “OCR works well for ink on paper, but cuneiform tablets are more complex due to lighting and viewing angle variations.”
AI Breakthrough
The 3D nature of cuneiform tablets posed unique interpretation challenges, prompting the research team to extensively train their AI system using 3D scans and data, including contributions from Mainz University.
This approach remarkably improved symbol identification on the tablets, democratizing access to ancient records and paving the way for broader historical text analysis. Future advancements may expand this technology’s use to decipher other weathered three-dimensional scripts in various contexts.
Did you subscribe to our daily newsletter?
It’s Free! Click here to Subscribe!
Source: The Debrief
