Our paper addresses how artificial intelligence technologies can transform museum records of provenance into structured and machine-readable data, which is the first critical step in undertaking a large-scale cross-institutional analysis of object history. Drawing on research on natural language processing (NLP), we have identified sentence boundary disambiguation and span categorization as highly effective techniques for extracting and structuring information from provenance texts. Our paper focuses on a provenance-specific annotation scheme that enables us to retain historical nuances when constructing provenance linked open data (PLOD)