Vague, Incomplete, Subjective, and Uncertain Information in Digital History

This cumulative dissertation investigates how digital infrastructures can accommodate the epistemic complexity of historical knowledge, focusing on the specific challenges posed by vague, incomplete, subjective, and uncertain (VISU) information. Using art provenance as a focused domain of inquiry, the research addresses how such information can be identified, structured, and preserved in digital formats without flattening its interpretative depth. The central hypothesis underpinning this work is that VISU information can be effectively preserved and meaningfully operationalised in digital history only through workflows that integrate computational methods with interpretive oversight. Specifically, this requires combining automated extraction, adequate data modelling, and expert validation to ensure that the epistemic complexity of historical knowledge is not lost in the process of digitisation. This guiding hypothesis gives rise to three interrelated research questions. First, how can automated extraction processes be implemented to identify and flag VISU information for preservation in historical texts? Second, how can VISU information be formally represented in structured data models without sacrificing interpretive complexity? Third, how can expert validation be operationalised to safeguard VISU information during the transformation of historical data? To answer the first research question, the thesis identifies key natural language processing tasks such as sentence boundary detection and span categorisation, and develops a tailored annotation scheme to capture VISU features. This scheme is used in training and evaluation of models for automatic extraction of structured knowledge from provenance records. Addressing the second question, the research explores modelling strategies that extend CIDOC CRM by aligning CRMinf with the Historical Context Ontology (HiCO) and structuring data using nanopublications. This approach supports the formal representation of historical claims, interpretative assertions, and metadata from the digitisation process. Finally, in response to the third question, the thesis introduces PROV-A, a web based tool that integrates automated extraction with expert validation. It allows historians to refine extracted data, annotate epistemic qualifiers, and publish structured provenance information as linked open data, preserving interpretative depth within scalable workflows. These contributions are presented across seven publications that form the basis of this cumulative dissertation. Together, they establish a methodological approach for preserving the historiographical richness of provenance records as they are transformed into digital formats.