Latin scholars use AI to crack ancient Roman text mysteries

With the proliferation of artificial intelligence (AI) into every pore of our society, it’s no surprise that it has demonstrated exceptional usability in almost any scientific branch, including the analysis of ancient Roman text mysteries.

As it happens, Latin scholars have deployed AI to help them understand a massive body of inscriptions discovered every year, offering an invaluable insight into the daily life of ancient Romans, but which have so far presented a significant challenge, per a Phys report on July 26.

AI tool decodes ancient Roman texts

Specifically, Google researchers have helped develop a new AI tool that can help these scholars piece together puzzles from the ancient past, mainly the common inscriptions across the Roman world, like emperors’ decrees, graffiti on the city streets, and even ‘Beware of the dog’ warnings.

According to the study co-author Yannis Assael, a researcher at Google’s AI lab DeepMind, such inscriptions are “so precious to historians because they offer first-hand evidence of ancient thought, language, society, and history.” That said, the damage over the millennia has made it difficult to “know where and when they were written.”

In the words of Thea Sommerschield, an epigrapher at the University of Nottingham who co-designed the AI model:

“Studying history through inscriptions is like solving a gigantic jigsaw puzzle. You can’t solve the puzzle with a single isolated piece, even though you know information like its color or its shape. To solve the puzzle, you need to use that information to find the pieces that connect to it.”

To address these problems, the researchers have created a generative neural network called Aeneas, an AI tool named after the Trojan hero and mythical son of the Greek goddess Aphrodite that scholars can train to identify complex relationships between types of data.

Furthermore, they trained Aeneas on 176,861 inscriptions – worth up to 16 million characters – 5% of which contained images. The data included dates, locations, and meanings of Latin transcriptions from the area spanning 5 million square kilometers of the 62 Roman provinces over 2,000 years.

According to DeepMind, over 20 historians who tried out the AI model discovered it provided a useful starting point in 90% of cases, producing the best results when combined with their skills as researchers rather than relying exclusively on one or the other.

Meanwhile, AI advances continue, one of the recent ones being AI models that could simplify the generation of realistic motions of human characters or avatars, and solve the problem of models that move in very unnatural ways, far removed from human movement.