DeepScribe: AI can translate ancient texts thousands of years old

Scientists at the University of Chicago have recently developed an artificial intelligence system, based on machine learning technology that can analyze and translate ancient texts, found on clay tablets. Thousands of years old into modern languages.

This machine learning model, called DeepScribe, was trained to be able to read the Cuneiform writing system - the writing used in the Achaemenid Empire in ancient Iranian history (550-330 years ago). AD), as reported by the University of Chicago.

Cuneiform writing is really a challenge for experts who are experienced experts, due to the complexity of the word layout as well as the characters that make up the word. In this case, it is even more difficult because Cuneiform texts are engraved on ancient clay tablets instead of writing on paper as usual, making word recognition extremely time consuming. That's why researchers need help from a machine learning system like DeepScribe.

To build this advanced handwriting analysis AI model, the team resorted to a huge training data warehouse, with 6,000 annotations images from ancient Cuneiform documents collected in the relics. Persepolis, 60 km northeast of the modern city of Shiraz, in the province of Fars, Iran. The team believes that DeepScribe can help discover new secrets about ancient Persian, Iranian history, society and languages.

The training data is then edited based on a lexicon developed by researchers and the database of more than 100,000 individual characters that make up the Cuneiform letters. Preliminary results show that DeepScribe can compile ancient texts with an accuracy of about 80% - inferior to 'manual' experts, but the time is greatly improved. to tell. In addition, this level of accuracy can be improved over time as training data increases.

In the future, DeepScribe can even be used to identify the origins of artifacts found in important relic sites, significantly eliminating the difficulties archaeologists face. .