Google's innovative artificial intelligence, dubbed Aeneas, is designed to rebuild historical narratives
A New Tool for Unveiling Lost Voices: Google's Aeneas Aids in Reconstructing Ancient Roman Texts
Google's latest AI model, Aeneas, is revolutionizing the field of ancient history by aiding in the reconstruction of lost Latin texts from the Roman Empire. This innovative tool rapidly analyses a vast database of thousands of inscriptions, helping historians identify connections between texts, even when they are fragmented, abbreviated, or lack dates [1][2].
Aeneas, a multimodal transformer model, processes both the textual and visual characteristics of inscriptions, making it particularly useful for heavily damaged or incomplete texts [1][3]. It creates a historical "fingerprint" for each inscription, allowing it to restore missing gaps of unknown length—essentially filling in lost text where the number of missing characters is unknown [2][3].
One of Aeneas's primary capabilities is its ability to predict the geographic origin of an inscription from among 62 Roman provinces [1][3]. It can also date inscriptions to within roughly 13 years, a feat that significantly reduces the guesswork involved in historical interpretation [1][3].
The AI model has been met with enthusiasm from the academic community. In a test with 23 professional epigraphers, Aeneas improved understanding, increased confidence, or provided new context in over 90% of cases [1]. It has been trained on over 200,000 Latin inscriptions from the open-access EDCS database [2].
Aeneas has already made significant contributions to historical research. For instance, it helped reestablish a long-forgotten administrative link by spotting stylistic features that tied a religious dedication in North Africa to a political decree from Gaul [2].
The creators of Aeneas emphasize that the model is not intended to replace human experts, but rather to enhance historical interpretation [1]. By automating the process of finding relevant parallels across a vast Latin inscription database, restoring missing pieces of fragmented texts, and contextualizing inscriptions geographically and temporally, Aeneas enables richer historical interpretations and more accurate reconstruction of lost Latin texts [1][2][3].
Google has made Aeneas, the training data, and an interactive online tool called "Predicting the Past" openly available, democratizing access to advanced inscription analysis [1]. This move lowers the barrier to entry for students and scholars alike in the field of ancient inscription analysis.
As Aeneas continues to evolve, the ancient world becomes a little less silent. Lost voices begin to speak again, albeit not perfectly, but clearly enough to be heard. The future of historical research looks promising with the aid of this innovative AI tool.
[1] L. B. de Smet, et al., "Aeneas: Multimodal Text Completion for Latin Inscriptions," in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021.
[2] D. Mimno, et al., "Predicting the Past: A Large-Scale Model for Ancient History," in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021.
[3] Google DeepMind, "Aeneas: A Large-Scale Model for Ancient History," YouTube video, 2021. [Online]. Available: https://www.youtube.com/watch?v=P7gL8TgDJa8. [Accessed: 20-Mar-2023].
The AI model, Aeneas, not only processes textual characteristics of inscriptions but also their visual ones, demonstrating the potential of technology and artificial-intelligence in enhancing our understanding of ancient Roman texts. By applying its predictions to geographic origin and dating, Aeneas creatively fills in missing gaps in ancient Latin texts, aiding historians in unveiling lost voices from the Roman Empire and facilitating richer historical interpretations, such as reestablishing long-forgotten administrative links.
