New digital system deciphers 3,500-year-old Hittite script with 90% accuracy

Researchers built a machine learning system to read Hittite cuneiform from tablet images and 3D scans with about 90% accuracy.

The model was trained on labeled tablets using data augmentation to better recognize cuneiform signs.

It works best on clear tablets and outputs standard sign codes for transcription, not full translation.

Human experts still need to review the results, but the tool could help speed work on thousands of untranslated fragments.