The increasingly widespread use of semantic technologies and Linked Open Data (LOD) led Digital Humanities to rethink their approach to knowledge management and sharing [1]. These technologies give Digital Humanities a means for representing their knowledge and include it into a network of connected data on the web, thus encouraging its reuse and further enrichment. In this context, ontologies play an essential role, as a technology for organizing knowledge by abstracting data and information of a certain domain.
An increasing number of cultural institutions is choosing ontologies and LOD for modelling and publishing their data, e.g. in Italy the Institute of artistic, cultural and naturalistic heritage of Emilia-Romagna (IBC-ER) [2] and the Fondazione Federico Zeri [3,4], and, in Europe, a lot of institutions within the project Europeana [5].
In this paper we report the results of ArCo (Architecture of Knowledge) [6], a collaborative project that involves the institute of the Italian Ministry of Cultural Heritage ICCD (Institute of Catalogue and Documentation) and the Institute of Cognitive Sciences and Technologies of CNR (Italian National Research Council).
ArCo aims at modelling the wide domain of Italian cultural heritage for two main purposes: (i) building a network of ontologies, compatible and aligned whenever possible with existing ontologies, that can be used as a de facto standard for representing cultural heritage data; (ii) publishing ICCD data as LOD: about 800.000 publishable files stored in a database, i.e. the General Catalogue, each describing a specific cultural property from diverse perspectives.
The cultural heritage domain has an intrinsic complexity, due to the high number of different types of cultural properties that a cataloguer may record, e.g. anthropological material, coin, park, painting, traditional music. They have a lot of shared information types (e.g. location, bibliography, dating), but also many peculiar characteristics (e.g. staircases and floors in a building). Moreover, their description may be very detailed: for a cataloguer is possible to gather information about measurements, exhibitions, documentation, authorship, inventories, relations between cultural properties, and so forth.
There are many projects and models developed in the context of cultural heritage (CH), to model, publish and connect data on the web: CIDOC-CRM [7,8], EDM [9,10], Cultural-ON [11], Fentry [12] and OAEntry [13] ontologies are some relevant examples. A recent paper [14] discusses the main requirements that a model representing cultural heritage should address, based on an analysis of CIDOC and EDM. Although we build on the good practices of such existing effort, our use case required a level of granularity and a diversity of cultural property types that needed new modeling effort.
To build ArCo, we directly reuse classes and properties from the core (roles, agents, locations) modules of OntoPiA [15], an ontology and controlled vocabulary network for Italian Public Administration, and from Cultural-ON, an ontology that models cultural events and sites [16]. We indirectly reuse patterns from existing ontologies, e.g. CIDOC and Cultural-ON and include explicit alignments to them within ArCo.
In the development of the project, we followed the principles of eXtreme Design (XD) [17], an ontology engineering methodology based on ontology design patterns [18]. Fig. 1 depicts as XD applied to ArCo. During the project initiation and scoping, domain experts shared with the ontology engineers' team their knowledge of the domain, providing guidelines and data model regulations for interpreting their data. A generic timeline and a release plan with priorities were defined. As recommended, we worked in tight collaboration with our main "customer", i.e. the ICCD. However, given that the ICCD data will be openly published and have high potential for reuse by several other stakeholders, we decided to interact with some representative of them since the very beginning of the process. In addition to domain experts, other agents, such as companies, were involved in the definition of ontology requirements, initially expressed in the form of user stories. The same requirements are reused in the ontology testing phase. Extending XD, four selected companies were also included in an "Early Adoption Program" (EAP) that worked with the incremental unstable releases of ArCo ontologies and data to test them for e.g. publishing their data according to ArCo ontologies, linking their data to ArCo. The EAP members and all the other interested stakeholders created an active community that interacts by means of a dedicated mailing-list [19], GitHub issues tracker [20] and meetups [21].
Pattern-based ontology design plays a central role [22]: by ontology design patterns we mean reusable successful solution to a recurrent modeling problem [23] [24]. XD encourages the reuse of existing ODPs from online repositories [25] as well as the development of new ODPs, when needed. Reused patterns are annotated with OPLa ontology [26], to support users in identification, reuse and ontology mapping.
Since XD is iterative and incremental, ArCo ontology modules and ICCD data are periodically published as unstable releases: this allows us to involve customers and stakeholders in giving us continuous feedback on modeling and testing activities, and to detect new emerging requirements at early stage.
ArCo release consists of a docker container, available on GitHub [27] and its running instance online [28], which contains:
the user guide accompanying the release, with diagrams and explanations on the content of the release and of each ontology module; the ontologies, including their source code and a human-readable HTML documentation; a SPARQL endpoint storing the General Catalogue data in RDF format, generated according to our ontologies; examples of Competency Questions (CQs), with the corresponding SPARQL queries, for supporting the data query from the community; a RDFizer tool converting XML data represented according to ICCD cataloguing standards to RDF.
ArCo knowledge graph is also available on the MiBAC official portal [29] with its SPARQL endpoint [30].
ArCo ontology network consists of seven ontology modules connected by owl:imports axioms. In Fig. 2, blue circles depict ArCo modules; the green circle indicates directly reused ontologies; the orange circle indicates indirectly reused and aligned ontologies. The network base namespace is https://w3id.org/arco/ontology/, and each module has its own namespace (e.g. https://w3id.org/arco/ontology/core/). The arco module [31] represents the network, importing all the other modules. It models top-level concepts from the CH domain, according to the ICCD cataloguing standards [32]. In particular, the hierarchy of the different types of cultural properties is modeled as follows. The core module [34] represents general concepts orthogonal to the whole network, which are imported by all other ontology modules. This module reuses a number of patterns, such as the Part-of [35], the Classification [36] and the Situation [37] patterns.
The catalogue module [38] models concepts related to the ICCD Catalogue, and in particular catalogue records, that is the XML files recording all data gathered by a cataloguer on a particular Italian cultural property. The Sequence [39] pattern is reused to model the different versions of the same catalogue record, represented by the class a-cat: CatalogueRecordVersion.
The location module [40] is intended to cover spatial and geometry information. A cultural property may have multiple locations, represented by the class a-loc:LocationType. In addition, the fact that a type of cultural property location holds during a time interval is modeled by the a-loc:TimeIndexedTypeLocation, which implements and specialises the TimeIndexedSituation [41] pattern.
The denotative description module [42] encodes the characteristics of a cultural property observed during the cataloguing process, e.g. measurements, materials, techniques, etc. To represent those characteristics we reused and specialised the Description&Situation [43] pattern for modeling both the technical status (a-dd:CulturalEntity TechnicalStatus) and the technical description (a-dd:CulturalEntityTechnicalDescription) of a cultural property.
The context description module [44] represents the context of cultural properties, in a broad sense, including the information related to: authors, collectors, copyright holders, inventories, bibliography, etc. For example, in order to represent the concept of an a-cd:Archival-RecordSet, i.e. fonds, series, subseries, etc., we reuse the Born Digital Archives [45] pattern.
The cultural events module [46] is dedicated to cultural events and exhibitions involving a cultural property. It extends, with some classes and properties (e.g. a-ce:Exhibition), the Cultural-ON ontology [11].
ArCo knowledge graph currently counts: 7 ontology modules, 327 classes, 379 object properties, 154 datatype properties, 395 restrictions. It counts about 170M triples and provides 24,008 owl:sameAs axioms linking to other datasets, such as DBpedia [47], Wikidata [48], the ULAN [49] and TGN [50] Getty Vocabularies, Zeri&LODE [4], YAGO [51], Europeana [52], Geonames [53]. The Entity linking is performed with LIMES [54], and the LIMES configuration files used in the linking process are available on Zenodo [55].
Fig. 3 depicts an example of information of a painting with subject "Madonna con bambino" (tr.en. Madonna with child). On the left side, there is the XML data, expressed as string and stored in the ICCD General Catalogue, and on the right side there is the correspondent data in RDF format generated according to ArCo ontologies.
In order to involve different stakeholders, we have organised a series of meetups associated with the ArCo releases. So far, we had 5 meetups, each attended by about 20 participants, and 1 webinar; we received 35 GitHub issues, and 27 people joined the mailing-list.
ArCo has a potentially very strong impact on both Cultural Heritage and Digital Humanities fields and related domains. At international level, ArCo ontologies allow to represent very detailed information on cultural heritage of many different types and ArCo data can be aligned to other CH data, ensuring a high reliable provenance. These ontology network and dataset will be used by institutions (such as museums, designated for cultural heritage preservation and enhancement), which intend to publish their data as LOD and/or link them to ArCo, as well as by companies and individual consumers (i.e. researchers, students, practitioners, citizens) that own and use CH data for different purposes.
Good examples, among others, of ArCo early adopters are: Synapta team [56], which reuses ArCo ontologies for representing musical instruments belonging to Sound Archives & Musical Instruments Collection (SAMIC) [57], and Ricostruzione Trasparente project [58], which aims at linking its data about areas of Italy damaged by the earthquakes in 2016 to ArCo data.
Currently, an extraordinary amount of data on Italian cultural heritage, in the form of a LOD dataset, is available to anyone interested in querying, consulting and reusing them. ArCo ontologies are released and adopted directly by ICCD, which provides Italian regulations for cataloguing cultural properties. Therefore, ArCo has become, in LOD context, a standard for Italian cultural institutions aiming at creating Linked Data, according to ministerial regulations.
Since the valorization of cultural heritage through LOD enables sharing and reusing of cultural heritage data in an open interconnected and multi-domain knowledge base on the Web, we plan to improve ArCo ontology network and LOD. Future efforts will be directed to: (i) model peculiar information regarding natural heritage and information related to archive and library domains, (ii) improve entity-linking, and (iii) provide tooling support for CH data owners in order to encourage and simplify the adoption of ArCo and other ontologies by domain experts.