On April 24, Xi Wangzhi presented her ongoing project, Automating Fine-grained Historical Event Extraction for Material Infrastructure in Late Imperial Chinese History, in the NLP and Text Mining Seminar at KU Leuven.
This project develops an automated pipeline for document-level event extraction from historical inscriptions. Leveraging Large Language Models (LLMs) for high-recall initial extraction, the system identifies event triggers and arguments based on Information Extraction standards. A subsequent hybrid refinement process, using LLMs and targeted deep learning classifiers, transforms this data to align with research-driven historical schemas. The pipeline, operating in dialogue with manual annotation, aims to produce traceable, modifiable datasets for historical analysis, with outputs designed for review and curation by historians.