The Future Of Archaeology: AI Decodes Ancient Scripts

AI is increasingly being harnessed to unlock the mysteries of ancient texts, offering a new approach to reading and interpreting historical documents that were previously indecipherable.

From deciphering burnt Roman scrolls to reading faded cuneiform tablets, AI techniques are opening doors to ancient archives, potentially rewriting parts of history. A recent review in Nature highlights the Vesuvius Challenge, a project using AI to read scrolls from the ancient Roman villa in Herculaneum, buried by Mount Vesuvius’ eruption in AD 79.

The scrolls, carbonized by the eruption, have been impossible to open without damaging them. For centuries, scholars have been forced to abandon attempts to read many of these texts.

However, AI is now allowing researchers to see beyond the damage. In October 2023, papyrologist Federica Nicolardi was sent an image of a fragment of a papyrus scroll with visible Greek lettering, a breakthrough that revealed entire lines of text that had been inaccessible for 2,000 years.

“It was incredible,” says Nicolardi. “I thought, ‘So this is really happening.’” She realized then that papyrology had changed forever. “In that moment, you really think ‘now I’m living something that will be a historical moment for my field,’” as reported by Nature.

This breakthrough is part of a broader trend where AI neural networks are being applied to a range of ancient texts, including Greek, Latin, and Chinese Oracle Bone Script.

Nature explains that these networks are trained to recognize patterns in images of ancient writings, helping researchers fill in gaps where characters are missing or damaged. AI can analyze texts more quickly and accurately than humans, allowing scholars to process large volumes of ancient documents and potentially identify patterns that would otherwise go unnoticed.

Nature explains that the technology is based on deep learning models, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs).

CNNs capture grid-like data from images, aiding optical character recognition, while RNNs process sequential data, ideal for reconstructing transcribed texts. CNNs are used for analyzing visual data like eroded lettering, while RNNs suggest missing characters in historical documents.

A key development is the creation of tools like Ithaca, a machine learning model designed to suggest missing parts of ancient inscriptions. Ithaca has been able to restore gaps in texts with greater accuracy than human specialists alone.

Nature reported that when paired with human expertise, the model’s suggestions improve the restoration process even further. Ithaca’s creators believe that such AI models could transform the way scholars approach the study of ancient texts.

Nature explains that beyond individual projects, AI is also being used to tackle vast archives of historical data.

In South Korea, for example, researchers are using AI to translate records from the Joseon dynasty, written in Hanja, an ancient writing system. The AI’s translations have proven to be more accurate than earlier efforts, highlighting the potential for AI to transform historical research.

Nature also reports that AI is also being applied to languages with only fragmentary records. In Crete, for instance, AI is being used to restore missing text from Mycenaean tablets written in the Linear B script. Researchers hope to expand this approach to decipher Linear A, a related but still undeciphered script.

The use of AI to read ancient texts is still in its early stages, but it promises to be a powerful tool in uncovering forgotten histories. As technology continues to advance, AI may soon allow researchers to access vast swathes of historical data that were previously out of reach, changing our understanding of the ancient world.