The ancient scrolls from Herculaneum, buried for nearly 2,000 years following the eruption of Mount Vesuvius, have begun revealing their secrets.
Thanks to AI technology, Luke Farritor, a 21-year-old computer science student at the University of Nebraska-Lincoln, has deciphered a word from the charred papyrus: “porphyras,” the Greek term for “purple.”
Using 3D X-ray scans, Farritor trained an AI model to detect nearly invisible ink-like patterns on the scrolls. The result was a clear identification of porphyras.
“Seeing Luke’s first word was a shock,” said Michael McOske, a research associate at the Department of Greek and Latin at University College London, commenting on the research. McOske continued, “It was immediately convincing, like ‘Good lord, that’s Greek.’”
A second breakthrough came from Youssef Nader, a data science graduate from the Free University of Berlin.
Nader discovered the same word using a different AI technique to detect possible letter shapes.
McOsker lauded Nader’s discovery, stating his snapshot was “even more impressive” than Farritor’s.
The competition to understand the Herculaneum scrolls is heating up – though perhaps not as hot as the 550°C pyroclastic flow that descended from the banks of Vesuvius in 79CE.
The Vesuvius Challenge, an open-source contest designed to decipher the scrolls, provided the platform for these researchers to experiment with innovative machine learning solutions.
A few years ago, studying the enigmatic Herculaneum scrolls meant physically unrolling them, which often damaged the delicate papyri. Even with advances in 3D imaging and computational methods, McOsker noted, “attempts to read the still rolled-up papyri were mirages.”
With these latest developments, the Vesuvius Challenge’s top prize of $700,000 might soon be claimed. It’s awarded to anyone who can read four passages from two intact scrolls by the close of 2023.
“I’m confident that Luke, Youssef, and the other competitors can solve a whole roll,” McOsker added. “Up until now, all the unrolled papyri that we study are missing their beginnings and are in bad condition, so the prospect of reading a complete text, from beginning to end, is really quite something.”
Machine learning has found a unique niche in helping researchers decode ancient texts, including ancient Greek, Sumerian, and other lost languages like Elamite.