There are hundreds of thousands of inscriptions at archeological sites and museums around the world that could provide valuable information about ancient civilizations. The problem is that only a small number of people possess the expertise necessary to decipher them, so most remain unread.
Now, an interdisciplinary team of researchers that includes Morris Alper, an assistant professor in the Department of Computer Science at the University of Miami College of Arts and Sciences, is developing artificial intelligence tools to help scholars understand these ancient texts.
"There are very few people who can do this, and this is a really slow process, so there's a real need to have computers help people," Alper said. "AI seems really promising because it can do things like read scans of books really well. But when we start looking at ancient languages and ancient texts, which aren't like a printed book, we see that even the leading AI systems struggle with them. We have to develop better methods that can help scholars to read the texts."
The project, which is supported by a $300,000 grant from Schmidt Sciences through its Humanities and AI Virtual Institute (HAVI)—of which the University of Miami will receive approximately $100,000—will teach AI models to read ancient inscriptions that are inscrutable to current AI systems. The researchers are focused on cuneiform clay tablets from the ancient Near East and Safaitic rock carvings from pre-Islamic Arabia but hope that the tools they develop could eventually be used to analyze inscriptions from other ancient civilizations as well.
Existing AI models struggle to make sense of inscriptions on stone or in clay, text that runs in spirals—like the Safaitic rock carvings from pre-Islamic Arabia—and other ancient scripts. The research team, which includes James Moore from Ohio State University, Letizia Cerqueglini from Tel Aviv University, Ethan Fetaya from Bar-Ilan University and Shai Gordin from Ariel University, plans to build AI models that can learn from few examples and turn photos of unread inscriptions into readable transcripts. They hope the project will help reveal new information about ancient civilizations that used cuneiform, such as the Babylonians and the Assyrians who lived in Mesopotamia, as well as what life was like in ancient North Arabia before the rise of Islam.
The AI models could also help researchers gain a better understanding of how written communication first developed and spread.
"Cuneiform is one of the oldest and longest-attested writing systems in the world," Alper said. "We can also learn things about how writing developed—for example, if the same sign was written differently in different places, or even in different times—because the cuneiform writing system was used continuously for over three millennia."
Once the researchers have created the AI models, they will share them publicly so other scholars can use them to read ancient inscriptions. The researchers are also creating an evaluation system that will gauge the efficacy of new AI models trained to read cuneiform.
Two undergraduate students at the University, Juan Giraldo and Christian Echavarria, are also working with Alper on this project.
Alper and his collaborators are one of 19 interdisciplinary teams that were recently awarded a combined $10 million through HAVI for projects that bring together researchers from the humanities and artificial intelligence.
