Google announces 'Aeneas,' an AI that deciphers ancient documents, capable of deciphering missing parts of ancient Roman inscriptions and estimating dates and regions
Google has developed an AI model called ' Aeneas ' that can help decipher inscriptions written in classical Latin. Aeneas can be used to decipher the contents of inscriptions where the number of characters in the missing parts is unknown, and estimate the age and region of the inscription.
Aeneas transforms how historians connect the past - Google DeepMind
In developing Aeneas, Google integrated the contents of ancient document databases such as the Epigraphic Database Roma (EDR), the Epigraphic Database Heidelberg (EDH) , and the Epigraphic Database Clauss Slaby (EDCS-ELT) to create the Latin Epigraphic Dataset (LED) , which contains 176,000 ancient documents written in classical Latin.
Aeneas is trained using LEDs and can input not only text but also images. Users can input 'transcribed text of inscriptions' or 'photos of inscriptions' to the AI to perform tasks such as estimating missing parts of inscriptions and searching for related inscriptions.
Aeneas can also estimate the date and location of the inscription.
In experiments with multiple historians, Aeneas showed higher accuracy than the average historian in three tasks: estimating missing parts, estimating regions, and estimating dates. In addition, Aeneas showed the highest accuracy in estimating missing parts and estimating regions when historians used Aeneas to make the estimates.
A free web app that can use Aeneas is available at the following link. In addition, it has been announced that the results of Aeneas have been incorporated into the ancient Greek deciphering AI ' Ithaca ' that was released in 2022.
Predicting the Past | Contextualising, restoring, and attributing ancient texts
https://predictingthepast.com/aeneas
The Aeneas code and dataset are available at the following links:
GitHub - google-deepmind/predictingthepast
https://github.com/google-deepmind/predictingthepast?tab=readme-ov-file
Related Posts:
