Developing an AI network can help scholars restore ancient Greek texts. Researchers at University of Oxford have recruited new deep learning techniques to decipher symbols of ancient languages.
The research team says there are many artifacts featuring important historical symbols and inscriptions that erode with time, noting that the clear inscriptions that archaeologists can read and understand are few.
In a test, where the AI attempted to fill the gaps in 2,949 damaged inscriptions, human experts managed to explain 30 percent of unclear symbols, whereas the network took two hours to decipher 50 inscriptions, reported the New Scientist website.
The team said the new network named "Pythia" is able to provide 20 different predictions to explain each eroded inscription, leaving the selection of the suitable hypothesis that correspond to the context of the ancient inscriptions to archaeologists.
In addition to providing various options, Pythia evaluates the reliability of each option or hypothesis it offers to archaeologists to help them reach the correct one.
"Pythia is the first electronic model used to restore ancient archaeological inscriptions by deciphering the eroded letters through deep learning networks," the study team wrote.