Available under a CC-BY-NC license
The linked data paradigm has significantly altered the notion of knowledge representation as well as the role of ontologies as applied in Artificial Intelligence (AI) and in the early stages of the Semantic Web. Recent initiatives such as Wikidata and its underlying software Wikibase have furthered this shift, forcing a reexamination of our approach to how semantic technologies are developed and used. This work provides a brief overview of this transition in the context of cultural heritage and based on the experience of the Pratt Semantic Lab team.
The linked data paradigm has significantly altered the notion of
knowledge representation as well as the role of ontologies as applied in
Artificial Intelligence (AI) and in the early stages of the Semantic
Web. Recent initiatives such as
Ontologies are the core of any knowledge representation system and a research focus in the field of AI. They are also a key component of the Semantic Web vision and the cornerstone of the Linked Open Data initiative—the most recent manifestation of the Semantic Web. In this new data environment, ontologies have morphed into lightweight knowledge tools with minimal or no formalization. Even changes in the lexicon, where ontologies and vocabularies have become interchangeable terms, reflect the more relaxed notion of the semantics they now carry.
The need to represent and interconnect large amounts of heterogeneous data in an open and distributed information space like the web makes formal ontological constraints a potential barrier to interoperability and broad adoption. This is especially true when it comes to the cultural heritage sector, a key player in linked data development. Cultural heritage is a complex and inherently interconnected field, populated by infinite entities and properties with various degrees of granularity. Its domain is typically less structured and more ambiguous when compared to domains such as medicine or the hard sciences. With their multifaceted and often abstract nature, arts and humanities tend to defy high formalization, making it difficult to ground their domain specifications into formal ontological commitments.
All of this makes the cultural heritage domain well suited for linked data ontologies whose semantics is relatively light and relies on rather basic constructs. For example, properties are not constrained logically by axioms and rules and their restrictions are, for the most part, semi-formal, usually consisting of a range, domain, and data type. Most properties are defined even more informally through natural language descriptions.
Agreement is typically achieved through the use of shared ontologies. While
instances via properties expressing equivalence or similarity. This provides a dynamic approach to information sharing as no pre-coordination is needed to stipulate meaning. This is a way for multiple and heterogeneous data sources to interoperate and for multiple points of view to coexist in a shared representation space.
A pragmatic approach drives the development and use of ontologies in the
linked data environment, a slight departure from the original spirit of the
Semantic Web (Pattuelli,
Yet, this lighter approach has its limitations. For example, one drawback of using representation languages that are not highly formalized is the limited capability to detect contradictions that can raise issues of consistency. There is a tradeoff between expressivity—including the ability to support inferencing and thus derive new knowledge—and computational tractability. Simplification is favored for the sake of interoperability and ease of use.
As a common linked data practice, a good deal of data modeling is also
performed through an application profile approach that consists of reusing
terms from existing RDF ontologies. A key tenet of linked data is semantic
shareability, and linked open data principles enable great flexibility in
mixing and reusing classes and properties from suitable sources without the
need of any centralized agreement, as mentioned earlier. One of the criteria
for selecting candidate terms, especially predicates, is the frequency of
use as a way to foster consistency in data modeling. RDF ontologies are easy
to extend. As a result, these models can expand as representation needs
evolve with little impact on systems. Flexibility has been recognized as a
key factor in the exponential growth of the linked data cloud (
A number of RDF-based ontologies, both general and domain-specific, have been
developed for the cultural heritage sector.
It is interesting to note that these reference and top-level ontologies have
shifted the barycenter from the traditional object-centric approach to an
event-centric approach. While the bibliographic (or documentalist) paradigm
has focused on objects and on the cluster of static properties describing
them (e.g., creator, object type, title, date of creation, format, etc.),
centering on events allows a more dynamic representation of culture, making
it possible to better express dimensions such as temporality, provenance and
various contexts of cultural production. Advantages and disadvantages of
current ontologies in cultural heritage domains such as museums and
libraries have been addressed in the literature (
New community-driven and rather intuitive knowledge representation tools are
also making strides in the linked open data community. First and foremost is
Wikidata
The data model Wikibase employs mirrors the Wikidata data structure. Wikidata
is essentially a collection of entity pages that include statements. Each
statement is made up of a claim in the form of a property-value pair and
optional qualifiers and references. The building blocks of a statement are
items and properties, comparable to what RDF defines
as entities (subject or object) and predicates. A
richer model than RDF, Wikidata provides a way to further refine claims
through qualifiers. These qualifiers specify the context in which a claim is
valid (e.g.
From a technical perspective, Wikibase provides a unified infrastructure where data, along with the tools and applications that generate, curate, store and consume it, sit together in one system. Data from different areas—from content to administrative data, including revision tracking and history—are integrated to give context and to convey multiple perspectives to graphs and datasets. We use an instance of Wikibase which makes it possible to work in a live system while retaining control of the data. This provides unprecedented flexibility during the modeling process. Through an intuitive editor, statements are created collaboratively by assigning identifiers to items and properties. Legacy data can be batch loaded. Mapping to corresponding Wikidata entities and properties is performed whenever possible to facilitate the process of later transferring the data to Wikidata. In this phase, however, there is no urgency to make binding decisions on the types of predicates to employ. These decisions can be deferred to the time when the data will be shared publicly, exported to Wikidata or exposed as RDF graphs, providing a way to tailor the semantics to different information contexts. Because Wikidata properties are generally mapped to equivalent RDF predicates via equivalent property (P1628), we intend to capitalize on this ‘parallel’ modeling feature to add multiple views, whenever suitable and consistent, to the representation of our domain of interest. This will enable us to exploit this semantic enrichment at the query time. Our work with Wikibase is still in its initial phase, but we can already see the potential. As we continue to experiment with this new linked data infrastructure, we hope to be able to share use cases and lessons learned and contribute to a new wave of linked open data development.