More than 2,000 years after Plato died, the towering figure of classical antiquity and founder of the Academy, regarded by many as the first university in the west, can still make front-page news.
Researchers this week claimed to have found the final resting place of the Greek philosopher, a patch in the garden of his Athens Academy, after scanning an ancient papyrus scroll recovered from the library of a Herculaneum villa that was buried when Mount Vesuvius erupted in AD79.
The project belongs to a new wave of efforts that seek to read, restore and translate ancient and even lost languages with cutting-edge technologies. Armed with modern tools, many powered by artificial intelligence, scholars are starting to read what had long been considered unreadable.
“It’s going to have a huge impact,” said Dr Kilian Fleischer, a papyrologist who worked on The History of the Academy, the scroll that revealed details of Plato’s life. “There will be scrolls that will be read with these new techniques that contribute to our knowledge of antiquity, and to our knowledge of literature in general. This might be a second renaissance.”
The History of the Academy, written by the epicurean philosopher Philodemus from the first century, has been studied for many years and modern editions exist. The researchers’ goal was to produce a more comprehensive edition. It’s no easy task when the scroll is in pieces from being unrolled and the papyrus as black as the ink written on it. Substantial portions of text are faded, missing or illegible.
Prof Graziano Ranocchia, project leader at the University of Pisa, used hyperspectral imaging to illuminate the scroll fragments with broadband infrared light. The images reveal letters that are invisible to the naked eye, giving scholars crucial clues as to the missing words. Fleischer likens it to completing a crossword or the game hangman: sometimes it takes only a single letter to be confident of the answer.
“It’s a wonderful feeling, this moment of reading something new and knowing this was information other researchers have wanted for decades or centuries,” Fleischer said. “We are travelling back and seeing text which hasn’t been read for 2,000 years.”
Armed with the scans, the team reconstructed 20% to 30% more of the text, with the additional words slotted into place amounting to 1,000 extra letters. The words for “buried” and “garden” do not appear in the scroll itself – they are conjectured from other characters and context.
Seeing the manuscript emerge was “a marvel”, Ranocchia said. A passage at the end of the scroll was particularly exciting, he added. Philodemus mentions one or possibly two as yet unknown books on the Megarians and the Cynics. “That’s very important for us,” he said. The books may be among the hundreds of charred and unopened scrolls in the National Library in Naples, he said, or perhaps even still buried at the doomed villa.
For many scholars, the prospect of reading the unopened Herculaneum scrolls is profoundly exciting. Carbonised in the blast that overwhelmed Herculaneum, the scrolls are too fragile to physically unroll. But researchers led by Brent Seales, a computer scientist at the University of Kentucky, showed there is no need. The team developed techniques to virtually unwrap CT scans of the scrolls, and trained machine learning algorithms to detect ink on the warped, blackened pages, often by spotting subtle changes in the patterns of the papyrus fibres.
The work led to the Vesuvius Challenge, a competition backed by Silicon Valley entrepreneurs, with lucrative prizes for the team that deciphered the most text from scans of scroll fragments. In February, three computer-savvy students shared the $700,000 (£557,000) grand prize after reading hundreds of Greek words across 15 columns of the scroll. Dr Federica Nicolardi, a papyrologist at the University of Naples Federico II, is studying one of the columns, in which Philodemus discusses how perception begets knowledge when images “bump into our sensory organs”.
The contents of the remaining scrolls are up for debate. Some may be Latin texts. There could be poems by Sappho, Mark Antony’s treatise on drunkenness, perhaps early writings on the Judeo-Christian tradition. Some of the scrolls are stuck together, which could shed light on how ancient libraries were organised.
Nat Friedman, a founding sponsor of the challenge, has announced a further set of prizes to spur researchers on to read 90% of a scroll by the end of this year. A major bottleneck is the tedious process of “segmentation”, which involves manually tracing layers of the digital scroll so the text-reading algorithm doesn’t confuse one warped layer with another. Work is under way to automate the process, which Friedman says will be a massive breakthrough. “That should unlock gobs and gobs of text,” he said. In anticipation, Friedman has booked nearly two weeks of beam time on the Diamond Light Source in Oxfordshire to scan potentially dozens more Herculaneum scrolls.
The new wave of technologies will shed light on more than the Herculaneum scrolls. The same approach could read papyrus wrapped around Egyptian mummies, with sheets that range from letters to laundry lists. There are boxes of the stuff in the back rooms of museums.
Dr Thea Sommerschield, a historian and epigrapher at the University of Nottingham, and her colleagues recently surveyed how machine learning is being applied to ancient languages. The tools, they concluded, are reshaping the field in much the same way as the microscope and telescope transformed science.
Sommerschield and Yannis Assael at Google DeepMind are co-leads on a project called Ithaca, a Transformer-based architecture, the current state of the art in AI. It is available to all and can predict missing characters in ancient Greek inscriptions and propose times and places where they may have been produced. The tool, which is used hundreds of times a week, promises to shed fresh light on the ancient world. Inscriptions in stone, ceramics and metals preserve writing from around the globe and from a cross-section of society, including women and enslaved people, not just emperors and the elite. “They give us information about thought, language, society and politics of the ancient world at large,” said Sommerschield.
So far, the best results have come from multi-disciplinary teams with computer engineers working alongside scholars. Such mixed teams were vital, said Sommerschield, as were efforts to build clean and well-curated datasets that cover not only ancient Greek and other popular languages, but writings from around the world. “If we have that kind of interaction then we will earn the trust and interest and engagement of computer scientists, the general public and the scholarly communities,” she added.