Museums increasingly deploy environmental monitoring systems (e.g., data loggers and sensor networks) to support preservation of cultural heritage objects. However, monitoring data is often captured and accessed through proprietary vendor software, which makes reuse, integration, and controlled cross-institution sharing difficult—particularly in object loan scenarios. This paper reports on MuMo (Museum Monitoring), a three-year applied research project that explored how dataspace principles can be applied to environmental monitoring in real museum settings. Rather than replacing existing systems, MuMo extends a legacy monitoring dashboard with semantic data modeling, Linked Data Event Streams, and Solid-compliant data access management. We present the system design, where environmental monitoring data is semantically described and published with access controlled fragments as a stream. The semantic links between these fragments allow clients to prune irrelevant branches while providers can restrict access at natural boundaries (e.g., per location or sensor). Further, we describe how MuMo is used in practice through a set of in-use scenarios, where the system enables (1) decentralized data publication for longitudinal analysis of environmental conditions, (2) selective cross-institutional data sharing during object loans, and (3) client-side data integration and aggregation across independent deployments. In practice, this dataspace-oriented deployment reveals design trade-offs while preserving institutional autonomy: museums control their data and authorization policies end-to-end, supporting trusted data sharing across organizational boundaries. The deployed system and its insights are thus relevant to a broad range of (Digital Humanities) projects that involve long-running data integration under distributed governance.

This paper introduces the Cultural Heritage Survey Ontology Design Pattern (CHS-ODP), a content pattern that abstracts from specific implementations and provides a reference core for modelling survey processes within cultural heritage knowledge graphs. The pattern is grounded in a set of generalised competency questions derived from two independent real-world use cases: a Reflectance Transformation Imaging (RTI)-based documentation campaign on rock art and a diagnostic investigation on illuminated manuscripts. Designed in alignment with CIDOC-CRM, ArCo, DOLCE UltraLight, the proposed pattern supports interoperable and reusable survey-centric modelling across heterogeneous cultural-heritage domains.

The State Papers Ireland (TNA SP 60- SP 63) constitute a rich and complex historical corpus, including official letters, private papers, petitions, and correspondence preserved at The National Archives UK, with a current focus on 1660–1715. Linking person mentions in these records to entities in the Virtual Record Treasury of Ireland (VRTI) Knowledge Graph (KG) is challenging due to historical spelling variants, temporal uncertainty, and incomplete metadata. This paper investigates methods for linking archival documents to their corresponding historical individuals, which are recorded in the VRTI-KG. We propose a metadata-driven heuristic approach—utilising fuzzy string matching and temporal constraints, and compare it against a neural entity linking method that employs BERT-based embeddings. Our results indicate that while neural models offer semantic flexibility, a metadata-driven heuristic method remains more robust for historical corpora characterised by sparse and inconsistent metadata, spelling variation, and heterogeneous location information. To further enhance linking accuracy, we discuss the potential of Named Entity Recognition (NER) to enrich both entity properties and document features, providing additional cues for mapping archival documents to the corresponding KG person entities.

This work proposes an automated system for generating contextualized narratives from structured museum data. In a first step, a tabular database containing object metadata (Identifier, Name, creator, culture, dating, material, etc.) is transformed into a knowledge graph based on the CIDOC CRM (Conceptual Reference Model) standard. This graph not only represents the objects themselves, but also models the events associated with them (creation, collection, acquisition, etc.), ensuring a rich and temporally anchored semantic understanding. In a second phase, targeted Cypher queries query this graph to extract contextual information related to each object. This information is then integrated into a structured prompt, sent to a pre-trained language model (LLM) via an API, to automatically generate a fluid, coherent, and historically informed narrative. Finally, the produced text is converted into an audio file, providing multi-modal playback accessible to a wide audience. The system has been empirically validated by domain experts who confirmed the accuracy and relevance of the generated narratives, as well as by members of the general museum public, who confirmed the texts’ accessibility.

The Deutsche Digitale Bibliothek (DDB) aggregates objects from libraries, archives, museums, audiovisual archives, monument conservation, and research institutions into a structurally uniform container, the Europeana Data Model (EDM), yet cross-domain discovery is structurally unsupported. Like Borges’ Library of Babel, the DDB has order without orientation. We identify three structural parallels and propose mocho (Mid-level Ontology for Cross-domain Cultural Heritage Objects), a reusable alignment layer between EDM and GLAM-spanning domain ontologies. mocho defines WEMI 1-typed entities and four alignment patterns.An empirical snapshot of 115k records quantifies the heterogeneity mocho addresses.

Uncertainty annotations such as "confidence = 0.7" are increasingly prevalent in semantic infrastructures for Cultural Heritage (CH) and Digital Humanities (DH), where they render uncertainty machine-readable and support ranking, filtering, and aggregation processes. In collection and provenance research, however, attributions frequently rest on fragmentary evidence, competing interpretations, and ongoing scholarly debates, such that numerical precision without an explicit reference frame remains semantically underdetermined. This paper investigates how uncertainty in RDF-based knowledge graph infrastructures can be modelled and published in ways that make it not only machine-readable but also interpretable, revisable, and responsibly reusable. The guiding thesis is that uncertainty is not a property of an object but a feature of a research judgment. We formulate a typology of uncertainty forms relevant to DH and CH research and show why each resists scalar representation. To ground requirements in practice, we report insights from an exploratory workshop with 15 professionals from Galleries, Libraries, Archives, and Museums (GLAM), analysed using Reflexive Thematic Analysis in light of an Human-Computer Interaction (HCI)-informed perspective on interpretability, from which recurring requirements emerge concerning transparent uncertainty marking, authorship, temporality, revisability, and the visibility of alternatives. On this basis, we propose a lightweight modelling pattern that represents uncertainty as an attributed, contextualised, and referenceable evaluation of a claim. We illustrate it with a case of conflicting narratives from Sammlung Kulturen der Welt (SKW) in Lübeck, where at least five epistemically incompatible claims coexist, a complexity no scalar value can capture.

The Repertorium Germanicum (RG), a critical edition of papal registers for the Holy Roman Empire (1378–1484), presents a paradigm case for automated knowledge extraction from historical sources: entries written in dense abbreviated medieval Latin, structured by a domain ontology under active development within the HisQu project (HisQu — Forschungsdateninfrastruktur Historische Quellen - Research Data Infrastructure „historical sources“). This paper evaluates GLiNER2—a schema-driven zero-shot information extraction framework built on a DeBERTa-v3 backbone—across four experimental configurations, varying model variant (large vs. multilingual) and preprocessing strategy (abbreviation expansion vs. raw text), for the task of populating the Ablass (papal indulgence) branch of this ontology. Our results reveal a fundamental tension between act-level detection, which both variants perform with reasonable coverage, and relation-level instantiation, which collapses across all configurations. We document a systematic subtype conflation bias, near-zero extraction of recipient church institutions despite explicit schema instructions, and entity boundary errors specific to abbreviated Latin tokenisation. Counterintuitively, abbreviation expansion degrades archival source reference extraction. We argue that the structural formula of the RG—not the abbreviated Latin per se—is the primary challenge, and conclude that generic zero-shot NER, while useful as a feasibility probe, is insufficient for the semantic depth required by ontology-based indexing of this source type.

We present the Theatre Migrants Knowledge Graph, a structured Linked Data resource documenting the professional migrations of European theatre practitioners in the 19th century. The knowledge graph is developed within the ERC-funded T-MIGRANTS project at LMU Munich and currently comprises 3,163 persons, 1,173 migration events, 1,128 locations, 2,901 organizations, and 3,265 interpersonal relationships. We build on well-known vocabularies to describe migrations, religious affiliations, and temporal uncertainty in historical data. Entities are systematically linked to external authority files, including Wikidata, GeoNames, GND, VIAF, and ISNI. The knowledge graph is generated from a relational database through a reproducible RDF mapping pipeline that uses agentic AI (Claude Opus 4.6) as a design assistant. We investigate three questions: whether an implementer without domain knowledge can use agentic AI to produce a functional knowledge graph, what role the AI tool plays in the process, and what risks this approach introduces. We report our findings and discuss how symbolic validation layers can mitigate the identified risks.

Semantic data are gaining increasing relevance within the fields of Cultural Heritage (CH) and Digital Humanities (DH). Despite their considerable promise and potential, however, these technologies currently tend to introduce additional challenges rather than effectively resolving existing ones. A central issue lies in the limited availability of graphical user interfaces that enable individuals with little prior knowledge of semantic data to explore them through clear, accessible, and easily customizable visualizations. To address this challenge and facilitate broader access to the insights embedded in semantic data, this paper presents methods from the field of Human-Computer Interaction (HCI) and applies them to concrete examples from CH and DH. In particular, the study adopts a Human-Centered Design (HCD) approach, systematically examining the target user group and its needs, identifying recurring issues in existing visualizations of the CIDOC Conceptual Reference Model (CIDOC CRM), and deriving the tasks and requirements for a novel user interface (UI). Based on these analyses, the paper introduces a newly designed UI sketch and justifies its design decisions with reference to established principles from Information Visualization (InfoVis).