‎”Pythia” reads better than people half-ruined ancient Greek inscriptions‎

Researchers at Deep Mind, Google’s subsidiary artificial Intelligence company‎‎, led by a Greek, have created “Pythia”, a “smart” system that seems to do better than people expert at reading half-destroyed inscriptions in ancient Greek.

The fully automated system, the first of its kind in the world, fills in the missing words by making alternative suggestions, which can greatly help scientists in the future read ancient inscriptions. There are thousands of such inscriptions from around the world and every year dozens of new ones are discovered. Unfortunately, many have been eroded or destroyed over the centuries, making them illegible or part of their text permanently lost. Filling in the gaps is a difficult job, requiring great knowledge, comparison with other texts, intuition and more.

The researchers, led by Greek Yiannis Asael, trained a neural network (a kind of artificial intelligence algorithm) to guess the words or letters of ancient Greek missing on inscriptions from stone, ceramics or metal aged 1,500 to 2,600 years‎‎ (7th century BC to 5th century A.D.). “Pythia” has thus learned to recognize patterns in 35,000 inscriptions containing more than three million words.

When confronted with a new incomplete inscription, “Pythia” (named after the High Priestess of Apollo at the Oracle of Delphi) proposes up to 20 different words or letters and lets the experts choose the one they consider the best solution.

In a comparative trial, where both “Pythia” and experts were called to fill gaps in 2,949 inscriptions, the latter had an error rate of 57.3%, while artificial intelligence had almost half (30.1%). In addition, while people needed two hours for only 50 inscriptions, “Pythia” made its suggestions for the entire inscription within seconds!

As Thia Somersild of Oxford University, a member of the research team, said, this shows the potential of artificial intelligence in the field of digital epigraphy. Philip Steel of the University of Cambridge has agreed that “Pythia” can make a significant contribution to the restoration of inscriptions, although specialist people will always be needed. “Pythia” could also be used in related fields, such as ancient philology, papyrology, codification, etc.

Yannis Asael graduated from the Department of Applied Informatics of the University of Macedonia in Thessaloniki in 2013 and did postgraduate studies at the Imperial and Oxford British universities, where he is currently pursuing his doctorate in Computer Science. At the same time, since 2015 he is a Deep Mind researcher in the field of artificial intelligence.

Source: ANA-MNA