Reading the past with AI: Comparative Linking and Extraction of Information in Open historical data
Reading the past with AI: Comparative Linking and Extraction of Information in Open historical data
Name of applicant
Antonia Karamolegkou
Title
Postdoctoral Fellow
Institution
INRIA, Paris
Amount
DKK 1,835,995
Year
2026
Type of grant
Internationalisation Fellowships
What?
This project develops AI tools to extract and link information from historical documents in Ancient Greek, Latin, and Medieval Danish. It creates open benchmarks and models to improve how we extract and connect information from the past, supporting research across various disciplines, as well as applications such as educational tools and fact-checking of historical claims.
Why?
Millions of digitised historical documents remain difficult to use because they are noisy and inconsistently formatted, and existing AI methods involve trade-offs in accuracy, reliability, and cost. They may generate incorrect information, ignore document structure and connections between elements (e.g., links between notes and text), and are rarely developed with input from domain expert users.
How?
The project compares existing and emerging AI methods (traditional multi-stage pipelines, state-of-the-art multimodal models, and agents) across diverse historical corpora and develops new datasets and evaluation tools. Working with domain expert users at Inria Paris and the University of Copenhagen ensures the methods are reliable and useful for real-world applications and research.