Digital Museums: Why Don't Machines Understand Art?

Imagine searching for an ancient vase in a global database. You find "ceramic, 500 BC," but lose the potter's story, the local clay, the experts' doubts about the exact dating. This is the "semantic gap": the divide between what we humans grasp observing an object—nuances, contexts, uncertainties—and the structured data a machine can process.

A study published in *Heritage* analyzed national museum databases in Latvia, Estonia, and Finland. The picture is fragmented: each institution catalogs by its own rules, different terms, languages, incompatible classification systems. When attempting to aggregate this data into a single transnational archive, descriptive richness flattens. Curatorial notes, dense with cultural meaning, struggle to become readable code.

Digitizing heritage isn't just taking photos and filling mandatory fields. It's an interpretive act: choosing vocabularies, standards, detail levels. If every museum "speaks" a different data language, knowledge stays isolated. The key to unlocking the value—the "aura"—of digital objects is metadata standardization: common rules to describe materials, chronologies, images.

Technology promises vast archives and instant searches, but this flexibility costs specificity. The real challenge for future digital museums is designing data infrastructures that connect institutions without betraying human complexity, turning information silos into a living, shared knowledge network.

The study, conducted by researchers from the University of Tartu and the Latvian Academy of Culture, is based on an in-depth comparative analysis of three national data infrastructures: the Latvian National Museum of History (LNVM), the Estonian National Museum (ERM), and the Finnish Heritage Agency (Museovirasto). The goal was to map how physical catalog cards—often narrative, polysemic, written in various national languages (Latvian, Estonian, Finnish, Russian, German, Swedish)—are converted into queryable digital records.

The core of the problem lies in the heterogeneity of metadata standards. While Finland has long adopted the CIDOC-CRM (Conceptual Reference Model) standard for semantic interoperability, Latvia and Estonia show varying degrees of adoption, often overlaying local schemas (such as LIDO or proprietary schemas) onto misaligned controlled vocabularies. This creates friction in aggregation: the same ceramic object may be classified as a "vase" in one database, "container" in another, and "archaeological find" in a third, with dates expressed in incompatible formats (e.g., "Iron Age", "500 BC", "Pre-Roman Period").

Case studies on temporal, material, and image-based metadata highlight an unresolved tension. Human description tolerates uncertainty ("probably", "style of", "attributed to"), while relational databases require discrete values. The strength of the digital—the structured query, the material filter, the timeline—thus becomes its weakness: it forces complexity into rigid boxes.

The proposal is not to return to analog, but to embrace Linked Open Data (LOD) and Knowledge Graphs. Modeling museum knowledge as a graph of related entities (object – has material – clay; object – found at – site; object – has dating – time interval with confidence level) allows preserving ambiguity and context. The digital "aura" lies not in photorealistic fidelity, but in the ability of an artifact to become a node in a global network of meanings, accessible to researchers, students, and the curious everywhere, transcending linguistic and institutional boundaries.