AI Deciphers 3,500-Year-Old Cuneiform Tablets Using New System

Researchers in Germany have created an artificial intelligence system capable of decoding and reassembling cuneiform clay tablets, some dating back 3,500 years. These tablets originate from advanced civilizations in the Near East during the third millennium BCE, which used cuneiform writing by inscribing wedge-shaped symbols onto wet clay with a reed stylus before drying the tablets for permanent records.

After approximately 3,500 years since some of these tablets were made in the Anatolia region of Turkey, AI has become the most effective tool to unlock their secrets. The clay tablets, whose fragments numbered up to 30,000 pieces, suffered breakage and dispersion across various museums worldwide. This distribution made the manual reconstruction of these fragments an arduous task for researchers studying the ancient Near East, as they sought to restore complete texts to gain insights into human life thousands of years ago. However, a research team from the University of Würzburg and the Academy of Sciences and Literature in Mainz, Germany, has transformed this process.

Over several years, the team developed a suite of digital tools, culminating in a system named "Palaeographicum." This AI-based system analyzes digital images of over five million documented cuneiform characters contained in 70,000 photographs. The system isolates individual signs, compares them against a vast database, and organizes them into visual tables to identify similarities and differences.

The key lies in the fact that the shape of each cuneiform sign reveals the writer's handwriting style. Although cuneiform was impressed into clay rather than written with ink, personal writing styles were evident. Some scribes dragged the stylus sharply, leaving decorative marks, while others spaced their signs distinctively.

The impact of this system is substantial in terms of time and accuracy. Previously, comparing handwriting across five clay tablet fragments required three full days of meticulous manual work, whereas now it takes only five minutes. Professor Daniel Schwemer, head of Near Eastern Studies, stated, "Our work radically changes the process and saves us thousands of hours."

Additionally, the system aids in more precise dating of fragments. Since Hittite tablets rarely contain explicit dates, researchers rely on changes in handwriting styles over time—a method known as palaeographic lineage—to estimate the text's age. AI accelerates these complex comparisons.

The team also aims to train the AI to automatically recognize the handwriting of individual scribes. This is challenging because a single scribe's handwriting varied depending on circumstances, being more precise in quiet settings and faster and less formal when writing quick field reports. Achieving this goal would, according to Professor Schwemer, "allow us to better understand the output of scribes throughout their careers and assemble a social history of Hittite writing culture."

The system remains under continuous development, with the team regularly retraining the AI based on user feedback to improve its performance.