Overview

• Aeneas uses specialized generative neural networks to predict missing segments in fragmentary Latin texts without requiring known gap lengths.
• The model was trained on a multimodal dataset of over 176,000 inscriptions and images spanning from the seventh century BC to the eighth century AD.
• In evaluations, Aeneas dated inscriptions with an average error margin of 13 years and assigned them to Roman provinces with 72 percent accuracy.
• In a study with 23 professional epigraphers, historians using Aeneas outperformed both standalone AI and unaided specialists in restoration, dating and localization tasks.
• Released as open-source, Aeneas integrates into scholarly workflows and can be adapted for other ancient languages and inscription corpora.