Leveraging External Knowledge for Historical Document Restoration via Retrieval-Augmented Large Language Models

The ARI framework restores historical documents by combining LLM implicit knowledge with retrieval-augmented generation to recover damaged named entities. By integrating explicitly retrieved external context, the model outperforms masked language modeling baselines in reconstructing context-dependent proper nouns within Korean historical texts.