Groundbreaking AI Techniques Transform Ancient Texts

In October 2023, an email landed on Federica Nicolardi’s phone that would radically alter the course of her research as a papyrologist at the University of Naples. The email contained an image revealing a fragment of a burnt papyrus scroll, one of many charred remnants from a luxury Roman villa in Herculaneum. This villa was preserved under volcanic ash and rubble following the catastrophic eruption of Mount Vesuvius in 79 AD. The image not only showcased Greek lettering brightly visible against a dark background but also heralded a new era in the field of ancient text analysis.

The Vesuvius Challenge: A New Approach

Nicolardi was part of the Vesuvius Challenge, a groundbreaking project leveraging artificial intelligence (AI) to read texts that had been largely deemed unreadable for 2,000 years. She described the moment of seeing the text as incredible, realizing that she was witnessing something historic for her field. “You really think, ‘now I’m living through a historical moment,’” she said. This project represents just one facet of a broader trend where AI is changing perspectives on ancient civilizations.

Insights from AI

Artificial intelligence, particularly deep learning techniques, is enhancing how researchers approach ancient texts. Neural networks can analyze vast amounts of data far beyond human capability. These advancements allow researchers to reconstruct missing characters, decipher lost languages, and classify archives that human scholars could struggle to make sense of. “It’s not only the questions we want to answer,” Nicolardi explains, “but the questions we can ask.”

A Historical Collaboration

The integration of computer science with classical studies began gaining momentum in the 2010s. The collaboration largely stems from the work done at universities such as Oxford, although researchers across the globe are also contributing. The initial team led by PhD candidates Thea Sommerschield and Yannis Assael devised a model called Pythia that could reconstruct missing words in Greek inscriptions.

Using tens of thousands of ancient Greek inscriptions, Pythia demonstrated significant promise by revealing previously hidden text with unprecedented accuracy. Following this groundwork, the team developed a more advanced model named Ithaca, which not only fills in gaps but also provides context about the potential dates and origins of unknown texts.

Tools to Aid Scholars

Ithaca leverages a transformer model, allowing it to process complex linguistic patterns effectively. In tests, this tool achieved a success rate of 62% in filling gaps in ancient texts, significantly outperforming human experts. When combined with the skills of these specialists, the accuracy jumped to an impressive 72%. The tool has seen substantial interest since its launch, attracting hundreds of inquiries weekly and contributing to fields as varied as ancient politics and philosophy.

Broader Applications in Ancient Studies

AI’s role is not limited to just Greek and Roman texts. Other projects are actively translating massive archives from different cultural contexts. South Korean researchers are tackling the limitations of accessing historical records that span centuries, documenting the reigns of 27 Korean kings. These archives are written in Hanja, an ancient character-based system. With the help of machine translation networks, researchers have made significant strides in translating these texts into modern Korean, vastly improving the understanding of their contents.

Moreover, researchers like Katerina Papavassileiou are using neural networks to reconstruct missing texts from Mycenaean tablets, which gives insight into early civilization activities and organization. These projects showcase how AI can bridge gaps across cultures, offering a more comprehensive understanding of human history.

Advances in Reading the Herculaneum Scrolls

At the forefront is the Vesuvius Challenge, primarily led by computer scientist Brent Seales and his team at the University of Kentucky. Their technique, known as “virtual unwrapping,” allows scientists to read scrolls without physically opening them — a crucial advancement since opening the Herculaneum scrolls would cause irreparable damage.

By using high-resolution scans, Seales’ team can create detailed images of the scrolls. Despite the scrolls’ carbon-based ink being invisible to CT scans, they trained a neural network to detect subtle differences in the papyrus surface, which gave clues to where the ink might be.

Competition and Innovation

In March 2023, the Vesuvius Challenge invited participants from around the world to submit their algorithms to find the ink on the scrolls. This led to remarkable contributions, including a grand prize awarded in February 2024 for showcasing high-quality text extraction. The importance of such competitions demonstrates the collaborative effort of engineers, historians, and scholars.

By identifying text attributed to a previously unknown work of Greek philosophy, the researchers believe this could offer rich insights into ancient beliefs and practices.

The Future of Ancient Text Reconstruction

Looking ahead, the implications of these technological advancements are immense. Seales expresses optimism about increasing the speed of the virtual unwrapping process, potentially enabling teams to read entire libraries of unopened scrolls. There are potentially hundreds of these scrolls still waiting to be analyzed, which could yield an extraordinary amount of historically significant text.

This progress also speaks to what Seales calls “the invisible library,” referring to texts hidden from view, such as those within medieval bindings or ancient mummy wrappings. Researchers are eager to explore every possibility.

Key Takeaways

The use of AI in reading ancient texts signals a transformative shift in the way scholars will engage with historical documents. With advanced techniques like virtual unwrapping and machine learning algorithms, researchers are finally unraveling the long-held secrets of lost civilizations.

As projects like the Vesuvius Challenge continue, historians and archaeologists will be able to tap into new troves of knowledge that could reshape our understanding of the past. This intersection of technology and humanities hints at a future rich with information, offering a deeper insight into our shared human narrative.