Megadose AI progress, ranked and analyzed.

Leveraging External Knowledge for Historical Document Restoration via Retrieval-Augmented Large Language Models

· HF Daily Papers ·
ARI uses retrieval-augmented LLMs to restore damaged Korean historical documents, especially proper names that local context misses.

The paper says masked-language approaches struggle when the missing text depends on outside historical knowledge. Its framework combines pretrained LLM knowledge with retrieved external context to infer context-dependent named entities. Experiments on Korean historical documents report significant gains over baselines for both general characters and named entities. Expert evaluations frame ARI as a practical aid for domain specialists working through damaged records. HF Daily Papers' note

score 4

Categories: Research