AI Deciphers Text from 2,000-Year-old Roman Scroll Burned During the Eruption of Mount Vesuvius

Greek writing inside a 2,000-year-old scroll burned during the 79 CE eruption of Mount Vesuvius has been deciphered by a team of student researchers. This latest development may lead not only to the discovery of a previously illegible text, but also pave the way for new applications of artificial intelligence to the archaeological field.

The scroll hails from the luxury Roman villa Herculaneum and is one of more than 1,800 intact papyri turned carbonized ash, known as the Herculaneum scrolls, excavated in the 18th century. The scrolls that have previously been read relate to the Athenian philosophy of Epicurus who lived from 341 to 270 BCE. They are the only known library from antiquity that survived, but most scrolls are too fragile to unfurl.

As part of the Vesuvius Challenge contest, the winners trained machine-learning algorithms on scans of the rolled-up papyrus, wherein they uncovered a previously unknown philosophical work on senses and pleasure. The text discusses music, the taste of capers, and the color purple, along with a description of a possible known flautist Xenophantus who has been mentioned in texts by the ancient authors Seneca and Plutarch.

The students were able to decipher hundreds of words across more than 15 text columns, which accounts for about 5 percent of the scroll. For their discovery, the three students from Egypt, Switzerland, and the United States won a $700,000 grand prize.

“The contest has cleared the air on all the people saying will this even work,” Brent Seales, a computer scientist at the University of Kentucky, Lexington, and cofounder of the prize, told Nature. “Nobody doubts that anymore.”

For the last 20 years, Seales has been working to read these concealed texts by creating software with his team that virtually maps the surfaces of the rolled papyri with 3-D computed tomography (CT) images. In 2019, he mapped the surfaces of two scrolls, but since the carbon-based ink has the same density as papyrus in CT scans, it became impossible to differentiate the two during the imaging process.

Seales was then approached by Silicon Valley entrepreneur Nat Friedman, who donated $125,000 to launch the Vesuvius Challenge and raised thousands more on social media. Seales released his software and high-resolution scans for those participating to use.

The challenge asked competitors to train machine-learning models to unwrap the scrolls to distinguish the text. They set a grand prize for reading four passages of at least 140 characters each before the end of 2023. The winning machine-learning code was released at various stages of the competition so that participants to build on each other’s work, with smaller prizes awarded throughout to incentivize continual progress.

In the middle of last year, US entrepreneur and former physicist Casey Handmer noticed a texture like cracked mud in the scans that formed Greek letters. Luke Farritor, an undergraduate computer science student at the University of Nebraska-Lincoln, then used this texture to train a machine-learning algorithm, which picked up on the word porphyras (purple) for which he won a prize. Berlin-based PhD student Youssef Nader then developed clearer images of the text.

Eighteen submissions were received and, after a jury checked the code, 12 submissions were then presented to a committee of papyrologists who assessed legibility and transcribed the text. Only one team—formed by Farritor, Nader, and Swiss robotics student at the Swiss Federal Institute of Technology Zurich Julian Schilliger—met the prize criteria.

“We were all completely amazed by the images they were showing,” said Federica Nicolardi, judge and papyrologist at the University of Naples Federico II, who is currently analyzing the revealed text with colleagues.

This latest development could spur further investigations at Herculaneum, as entire sections have yet to be excavated and the villa’s primary library has not been identified. These new techniques could also be applied to decipher other texts as well.

A new step of the Vesuvius Challenge prizes have been announced for 2024, with the goal of deciphering at least 85 percent of one scroll by the end of the year.