Agentic extraction for fine-grained intertextuality in classical texts
July 31, 2026
An agentic framework uses LLMs with tool interfaces to identify character-level spans and five-dimensional reuse typologies in classical Chinese histories. Evaluation against an expert-adjudicated benchmark of 2,533 pairs shows precision between 56% and 93% across twelve models.
HOW THIS AFFECTS YOU
●
researcherThis demonstrates how constrained tool use can move LLM analysis from simple similarity scores to grounded, structural extraction tasks.