Restoring and attributing ancient texts using deep neural networks
Problem
Many ancient Greek inscriptions are damaged to the point of illegibility, displaced from their original location, and their dates of writing are uncertain.
Why it matters
Inscriptions are primary historical evidence that historians currently cannot fully read, place or date.
What the AI did
Ithaca, a deep neural network, performs textual restoration, geographical attribution and chronological attribution of Greek inscriptions, designed as a collaborative, interpretable decision-support tool for historians.
Results
62% restoration accuracy alone; historians improved from 25% to 72% with Ithaca; 71% geographical attribution accuracy; dating within less than 30 years of ground-truth ranges.
Impact
Improved historian accuracy substantially and contributed to redating key texts of Classical Athens and topical debates.
Record
| Venue | Nature |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Yannis Assael, Thea Sommerschield, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, JONATHAN R. W. PRAG et al. |
| Citations | 206 (OpenAlex, at import time) |
| References | 73 |
| Model or method | Ithaca (deep neural network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-022-04448-z |
| Open access | hybrid |
| Keywords | epigraphy, deep learning, text restoration, ancient Greek, attribution, dating |
Importance rating
real but narrower advance
Abstract claims transformational impact on studying ancient history; improves historian restoration accuracy from 25% to 72%, but field is specialized.