Researchers from Hugging Face have introduced a novel framework for restoring damaged historical documents using retrieval-augmented generation (RAG). The framework, named ARI, leverages large language models (LLMs) combined with external knowledge retrieval to address the challenge of restoring named entities that require historical context.
Traditional restoration methods based on masked language modeling can effectively use local context but struggle with proper nouns that depend on external historical knowledge. ARI overcomes this by integrating the implicit knowledge of pre-trained LLMs with explicitly retrieved external information, enabling it to infer context-dependent proper nouns more accurately.
The team conducted extensive experiments on Korean historical documents, demonstrating that ARI significantly outperforms existing baselines in restoring both general characters and named entities. The improvements were validated through comprehensive evaluations, including expert assessments, confirming the framework's practical utility for domain experts.
This work promises to accelerate the analysis of historical records by providing a practical tool for historians and archivists. The paper is available on Hugging Face and has been recommended by the Semantic Scholar API alongside related research on RAG-based systems and knowledge-grounded language modeling.