Google DeepMind open sourced Aeneas, a generative AI model for understanding ancient inscriptions. Aeneas can process both text and image input and outperforms other state-of-the-art models at restoring missing characters in damaged inscriptions.
Aeneas is intended to assist historians with epigraphy, the study of ancient inscriptions. It helps automate several key tasks: dating an inscription; identifying the region of an inscription’s origin; reconstructing partial inscriptions; and identifying parallels, or inscriptions with similar words or phrasing. Aeneas uses a multimodal transformer architecture as its backbone, with dedicated heads for each task. When evaluated on several epigraphic tasks, Aeneas outperformed the state-of-the-art AI model and human historians. When human historians used Aeneas as a tool on these same tasks, the combined performance was even better. According to DeepMind,
Our model can also be adapted to other ancient languages, scripts and media, from papyri to coinage, expanding its capabilities to help draw connections across a wider range of historical evidence…This work was part of a wider effort to explore how generative AI can help historians better identify and interpret parallels at scale. We want this research to benefit as many people as possible, so we’re making an interactive version of Aeneas freely-available to researchers, students, educators, museum professionals and more….
Aeneas builds on DeepMind’s Ithaca project, a text-only model trained to perform epigraphy on ancient Greek texts. Aeneas adds support for image input. Aeneas can also restore inscriptions with an unknown number of missing characters as well as output parallels, capabilities that Ithaca lacked.
To train the model, DeepMind assembled Latin Epigraphic Dataset (LED), a corpus containing 176,861 inscriptions. They started with existing source dataset, then created a “complex pipeline” to clean the records and combine them into a single format. The data contains inscriptions from the 7th century BCE to the 8th century CE, from regions spanning the Roman world from Britain to Mesopotamia.
To evaluate its effectiveness as a research tool, DeepMind performed a study with 23 epigraphic experts who used Aeneas in a “simulation of real-world research workflows” with a time constraint. The human experts manually selected parallels for inscriptions, but usually also incorporated at least one additional one suggested by Aeneas. One researcher claimed:
The parallels retrieved by Aeneas completely changed my historical focus…it would have taken me a couple of days rather than 15 min [to find these texts]. Were I to base historical interpretations on these inscriptions’ readings, now I would have days to write and frame the research questions rather than finding parallels.
In a discussion about Aeneas on Hacker News, one user wrote:
[To] me, these are just educated guesses and thus whenever used the tool needs that disclaimer. That said – a lot of (ancient) history is educated guesses based on partial information. Even when we have a lot of writing available, like Cicero, we have to admit that we’re seeing the events from a particular point of view, someone with his own biases and motives. So we try to infer what happened in history based on data that has a certain amount of ‘data quality issues’ regardless.
The Aeneas code is available on GitHub. There is also an Aeneas interactive demo website.
