Artificial intelligence software can help historians interpret and date ancient texts by reconstructing works that have been destroyed over time, according to research published in Nature.
A team of computer scientists and classical studies experts led by DeepMind and Ca' Foscari University of Venice trained a neural network to be able to restore inscriptions written in ancient Greek between the 7th century B.C. and 5th century AD
The model, which was named "Ithaca" (Ithaca) can also approxpricewhen the text was written and where it may have come from. Thus researchers can recover texts found on broken ceramic pieces, or obscure texts found on various inscriptions.
First, the text should be transcribed by scanning an image of an old object. It is then fed to Ithaca for analysis. It works by predicting lostys or blurred characters and restores words. The software generates and ranks a list of its top predictions. Archaeologists can then read them and judge whether the model's predictions are accurate or not.
Of course better results are achieved with cooperation human and machines. Back when archaeologists were working alone, they were 25 percent accurate in piecing together ancient texts. Working with Ithaca the accuracy level rose to 72 percent. The performance of the machine learning model alone is about 62 percent accurate. It can also pinpoint with 71 percent accuracy the location where the text was written and can date the works to within 30 years of their creation between 800 B.C. and 800 AD
Ithaca trained on more than 63.000 Greek inscriptions containing more than three million words from the Packard Humanities Institute's Searchable Greek Inscriptions repositories. The team filled in parts of the text and instructed the model to fill in the blanks.
Google's DeepMind is now adapting its model to other types of ancient writing systems, such as Akkadian developed in Mesopotamia, language of ancient Egypt, the language of the Maya from Central America and the ancient Hebrews.
Inscriptions