Researchers from Tel Aviv University and Ariel University, Israel, have used artificial intelligence to translate fragments of ancient cuneiform texts on stone tablets into English with what they say is a high degree precision. They call the project “another major step towards the preservation and dissemination of the cultural heritage of ancient Mesopotamia”.
Researchers presented the first neural machine translation from Akkadian to English in the May issue of Nexus PNAS. Their results are “on par with those produced by an average machine translation from one modern language to another,” noted Arkeo news.
Over the past 200 years, archaeologists have found hundreds of thousands of texts that tell the story of ancient Mesopotamia, most written in Sumerian or Akkadian, the authors explained. But most remain untranslated due to their large quantity and the small number of experts who can read them, as well as the fact that most of the texts are fragmentary. Additionally, cuneiform signs are versatile, there are many different types of text, and even the names of people and places can be written as complex sentences.
“First, let me say that we believe AI will not replace philological work,” said Luis Sáenz, of the Digital Pasts Lab at the Land of Israel Department of Studies and Archeology. at Ariel University, one of the authors, in an email to Artnet News. “We want to speed up the process. Our hope is that AI may eventually help both Assyriologists and non-Assyriologists to read cuneiform texts in the future.
This is just the latest example of scientists using the newest tools to work with the oldest materials. University of Kentucky researchers developed an AI system to read scrolls that were incinerated during the eruption of Vesuvius in AD 79, and archaeologists in Italy are work on a robot which uses AI to reconstruct ancient relics from their scattered fragments.
“There are of course limits to the model,” says Sáenz. “The lack of context makes ancient languages difficult to translate, since we only have fragments of texts. Fragments with only one or two lines are extremely difficult for AI to work with. The future will require more tools to digitize data published in articles to continue training the model and improve results. Also, an audience-friendly web platform is important. »
Follow Artnet News on Facebook:
Want to stay one step ahead of the art world? Subscribe to our newsletter to receive breaking news, revealing interviews and incisive reviews that move the conversation forward.
