Leveraging External Knowledge for Historical Document Restoration via Retrieval-Augmented Large Language Models

Researchers introduced ARI, a restoration framework that combines pretrained LLM knowledge with retrieved external context to repair damaged historical text.

The system uses retrieval-augmented information to fill in missing or illegible content that local context alone cannot reliably recover.

On Korean historical documents, ARI outperformed baseline methods in both general character restoration and named-entity recovery.

Expert evaluations also found it useful for historical analysis, especially for names and other context-dependent terms.