News Article

Abstract

Deciphering Burnt and Carbonized Scroll Books with Deep Learning-Assisted X-ray Imaging (February 5, 2024).

There is a program called the Vesuvius Challenge in which investors are funding the deciphering of what was written in the extremely fragile scrolls that were carbonized when Mount Vesuvius erupted in 79 A.D., some 2000 years ago (https://scrollprize.org/). Pompeii is famous for the eruption of Mount Vesuvius, which was buried by volcanic ejecta, etc. Not only Pompeii, but also towns near Mount Vesuvius were buried in the same way. The city of Herculaneum is one of them. The scrolls discovered in Herculaneum in the 1750s are one of the most important research subjects. Looking back in history, one would immediately think of using non-destructive methods of analysis such as X-rays, but in the days before the discovery of X-rays, this was obviously not possible. The method of dismantling the scrolls was unavoidable, and it seems that actual dismantling was done. Afterwards, they were probably restored by hand, but it is still difficult to read the text on them. In 2015, other ancient scrolls, though not from the Vesuvius eruption, were successfully read by X-ray imaging without touching them at all. A commercially available micro X-ray CT device was used (for details, see the paper, William Brent Seales, Clifford Seth Parker, Michael Segal, Emanuel Tov, Pnina Shor, and Yosef Porath, “From damage to discovery via virtual unwrapping: reading the scroll from En-Gedi”, Science Advances, 2, e1601247 (2016). https://doi.org/10.1126/sciadv.1601247). Using X-ray imaging, it is possible to read the inside of a scroll-like book without touching it to open it and reveal its contents. The Vesuvius Challenge program appears to have been inspired by this 2015 success story. Furthermore, while X-ray CT simply images three-dimensional electron density contrasts, deep learning techniques can be used to read text from them (for details, see the paper, Yannis Assael, Thea Sommerschield, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, Jonathan Prag and Nando de Freitas, “Restoring and attributing ancient texts using deep neural networks”, Nature 603, 280–283 (2022). https://doi.org/10.1038/s41586-022-04448). Recently, the Vesuvius Challenge experiment was conducted using the imaging beamline of the Diamond Light Source synchrotron radiation facility in the UK to collect a large number of 3D CT images of carbonized scrolls. Deep learning was used to decipher the text.

Eventually, the first success was achieved, although only a small part of the text was deciphered. What was written on the scroll turned out to be a philosophical statement about sensation and pleasure. It was announced that three 21-year-old graduate students from Egypt, Switzerland, and the U.S. were awarded $700,000 for their success in using X-rays to decipher what was written in such extremely fragile ancient burnt scrolls. For more details, see the article “First passages of rolled-up Herculaneum Scroll revealed”, Nature 626, 461–462 (2024). https://doi.org/10.1038/d41586-024-00346-8