Abstract

Interoperability of digital libraries has been a research challenge for some time, especially its semantic aspect. In a digital library environment like Europeana, where a vast amount of information objects from heterogeneous sources will be made available, it is important to find standards that enable semantic interoperability. This dissertation aims at evaluating the applicability of CIDOC CRM for providing semantic interoperability in a digital library environment. To achieve this, the current use of CIDOC CRM in digital cultural heritage projects will be examined and its role for semantic interoperability will be determined. Limitations, benefits and prerequisites will influence the answer to the question on how CIDOC CRM can be integrated in Europeana in order to enhance semantic interoperability.

Categories and Subject Descriptors

D.2.12 [Interoperability]; H.3.7 [Digital Libraries]: Standards; I.2.4 [Knowledge Representation Formalisms and Methods]: Semantic Networks

General Terms

Documentation, Theory, Verification.

Keywords

Semantic interoperability, digital library, examination, CIDOC CRM, Europeana.

1 INTRODUCTION

One of the goals of the European Commission's i2010 strategy is to "create a single European information space, which promotes an open and competitive internal market for information society and media services [7]. As part of this strategy the project EDLnet / EuropeanaNet was set up aiming at the establishment of the European Digital Library, which will provide "a multilingual common access point to Europe's distributed digital cultural heritage" [6, p. 10]. Based on EDLnet's work a follow-up project cluster has been set up. Part of this is EuropeanaConnect which started in May 2009 and deals with the technical enhancements for the European Digital Library, which was named Europeana . Making Europeana's digital-born and digitised information objects semantically retrievable is one of the main tasks in EuropeanaConnect, which will be achieved by creating a semantic data layer.

Consolidating Europe's digital cultural heritage means including objects from all memory organizations [cf. 8] (museums, archives, audio-visual archives and libraries), i.e. Europeana is intended to be a digital museum, a digital archive, a digital audio-visual archive and a digital library all at once (For simplification reasons the author refrains from defining the term digital library and will use it in this paper as a super ordinate concept for digital library, digital museum, digital archive and digital audio-visual archive). As described in 6the infrastructure of Europeana should be able to handle a vast amount of information objects from various heterogeneous sources, resulting in representations of cultural heritage objects of different metadata standards and quality. This emphasises the importance of establishing an interoperable semantic data layer considering the different technical and library scientific standards. This dissertation would like to contribute to the research in this area by exploring a way to provide semantic interoperability.

In November 2008 a first prototype of Europeana was released providing basic search functionalities over about two million digital objects. After having launched Europeana v1.0 in June 2010, another updated version will be released in 2011, for which basic semantic search functions will be implemented.

2. SEMANTIC INTEROPERABILITY

2.1 Why Semantics?

Traditionally the retrieval possibilities in a digital library are limited to those of a digital object catalogue through the (controlled) metadata fields assigned by information professionals to the objects. Yet, accessing information objects in their context would permit the user to gain new perspectives on the object.

The Web as a construct of HTML-pages hyperlinked to each other --yet, not necessarily reciprocally-- is slowly moving towards a Web that allows the machine to understand the information it is dealing with. The ultimate scenario of the Semantic Web would be intelligent agents that are able to infer new knowledge from digital information and its semantic relations. Tim Berners-Lee in [1] proclaimed the term "Semantic Web" already in 2001, however his vision has only been partially realised in the Web environment. The need for semantic retrieval and interoperability steadily emerges with the commercialisation of the Web, which is also supported by the W3C's efforts on expediting research on the Semantic Web [9]. XML, RDF and OWL are the most important standards in this context.

Consequently a combination of the digital library and the Semantic Web environment seems obvious. The digital object representation layer would gain added value if its objects were semantically contextualised; this can be realised by Semantic Web technology that makes information understandable for machines. Digital libraries offer the possibility to partly carry Berners-Lee's vision of the Web into effect and become a part of it, namely the Semantic Web.

In this context the Linked Data community seeks to link related data on the Web using URIs as well as RDF and hence enhance semantic interoperability. On the one hand new resources can be published and added to the Linked Data cloud, on the other hand existing resources can be reused. Considering the growth of the Linked Data cloud, it seems necessary to focus on the second approach and start reusing existing resources and not reinvent the wheel.

2.2 Semantic Interoperability

Semantic interoperability can be understood from the linguistic, the human communication's point of view, or from the technical, the computer scientific side. In communication semantic interoperability means that sender and recipient share, unambiguous meaning about the information they exchange. In computer science machines are semantically interoperable if they are able to exchange and interpret information unambiguously [12]. This implies both machines would derive the same inferences from a piece of information independently. Europeana and its semantic data layer follow the computer scientific definition of semantic interoperability.

The DELOS Digital Library Reference Model provides only a limited explanation on semantic interoperability [3, p. 55]. It states a digital library can follow three approaches of interoperability: the federated, the harvesting, and the gathering approach. In the federated approach the interoperability consists of a common use of protocols; thus "semantic interoperability is at the level of query and result set". The harvesting approach relies on providing content to third parties in a certain standard, where the semantic interoperability is "usually based on the use of a common ontology, e.g. Dublin Core". The gathering approach does not require interoperability at all, as "no source takes care of its potential consumers".

However, the DELOS project deliverable on "Semantic interoperability in Digital Library Systems" [18] mentions three levels on which digital libraries can interoperate semantically: data structures (e.g. metadata), categorical data (universals e.g. classification) and factual data (particulars e.g. people). It sheds light on the theoretical prerequisites of semantic interoperability that include Standards and Consensus Building, Foundational and Core Ontologies, Knowledge Organization Systems (KOS), Semantic Services, Architecture and Infrastructure as well as Access and Rights issues. Yet, the question of how they should be implemented practically remains unanswered.

2.3 Approaches towards Semantic Interoperability

My research proved that the most important prerequisite for semantic interoperability is the interoperable organisation of subjects, concepts, ideas, but also other controlled metadata: the KOS Ñ see Table 1. [20]

Table 1. The many forms of KOS

Yet, all of them provide different levels of semantic interoperability and some aim only at syntactic interoperability. Syntactic interoperability implies that two systems recognise character strings as tags, relationships or terms, whereas semantic interoperability implies that these systems would also be able to interpret the meaning of the tags, relationships or terms correctly. Soergel in [19] identifies two ways to approach interoperability: the use of a common standard or the mapping of standards to each other. He further claims that KOS are "predominantly syntactic" (p. 38), although in an earlier table (p. 29) he lists being a "semantic road map to individual fields and the relationships among fields" as one of the possible functions of a KOS. This shows the potential of KOS as a solution to semantic interoperability is not fully tapped in practice. The following subchapters will describe some of the above-listed KOS with respect to their role for semantic interoperability in a digital library environment.

2.3.1 DCMI Abstract Model

The Dublin Core Metadata Initiative (DCMI) is the commonly known organisation for providing simple standards and guidelines on metadata. The classic Dublin Core Metadata Element Set consists of 15 elements and has become part of a more comprehensive set of metadata vocabularies and technical specifications, the DCMI Abstract Model (DCAM), as a result of DCMI's efforts to meet the requirements of being compatible with the W3C's information architecture; more precisely RDF.

The DCAM comprises of three interconnected models: the resource model, the description set model, and the vocabulary model (DCMI Metadata Terms). Being able to describe resources and metadata terms with URIs and using standardised fields for those terms takes the digital library one big step into the direction of semantic interoperability. Although the DCAM enhances the syntactic interoperability more, it sets the important basis in a computer scientific way to have systems using and interpreting the DCMI Metadata Terms (e.g. dc:creator, dcterms:provenance, dc.format, dcterms:hasFormat) elements the same way.

Thus, one can say the DCMI Abstract Model fits into the approach of creating a common standard to provide semantic interoperability for a digital library. Yet, semantic relationships remain at a basic level giving references to the corresponding RDF notions. Detailed semantic specifications are not in the scope of the DCMI recommendation. [19]

2.3.2 SKOS

Taking semantics further, an approach has been developed by the W3C community in the context of the Semantic Web: SKOS. It is the most important standard for a digital library as it presents "a common data model for sharing and linking knowledge organization systems, such as thesauri, taxonomies, classification schemes and subject heading systems via the Web" [17] being compatible with the XML, RDF and OWL standards.

SKOS covers both approaches identified by Soergel in [20], being a standard in itself plus providing an appropriate tool set for mapping. The elements of SKOS most important for semantic interoperability are presented in Table 2. For the full list cf. [17]:

Table 2. Important elements of SKOS

SKOS is intended to describe thesauri; thus a representation of a thesaurus can only be as semantic as its original version. SKOS is not the magical tool that adds semantics but only provides a standard to translate existing classifications and thesauri into a machine-readable format. Yet, it provides an efficient tool set for making collections indexed with keywords from one or more thesaurus and from different sources interoperable. Consequently SKOS provides some level of semantic interoperability on the aforementioned computer scientific level.

Archives raise another problematic issue, as they do not tend to index their objects with a thesaurus, but rather use finding aids to describe their collections. Those finding aids might use a controlled vocabulary itself, yet it is not obligatory. [cf. {16}] As a result SKOS' application potential is limited in the archival area; it can only provide very limited semantics as the original indexing sources do not contain structured semantics.

2.3.3 Ontologies

The term ontology is borrowed from philosophy and emerged in computer and information science in the context of Artificial Intelligence Systems. In the Semantic Web world an ontology describes a part of the real world. The often cited T. Gruber defines it in [13] as "an explicit specification of a conceptualization." Concepts and relationships are at the core of it and provide successful communication and common understanding about a domain (or the specific part of the world it represents) [21]. Fensel in [11] additionally states ontologies bring the Web to its full potential by uniting two important aspects: On the one hand ontologies describe real-world semantics; on the other hand they allow machines to process these translated formal semantics for information. On the whole every ontology is an entity-relationship model Ñ derived from relational database modelling Ñ that can be illustrated by an RDF graph. [12, 20]

Blumauer and Pellegrini in [2] define a semantic staircase ("Semantische Treppe", p.16) where the ontology represents the highest semantic richness of all KOS (translated from [2]):

Figure 1. Semantic staircase

Consequently an ontology seems to be the obvious solution to semantic interoperability. In [20] Soergel suggests an ontology should be developed at the start of a new digital library. However, the same arguments that speak against the development of a concept hub [cf. {20}] apply also to the development of a new ontology for Europeana. Furthermore Europeana wants to become part of the (Semantic) Web architecture. Hence, the reuse of existing standards is a priority. One ontology predestined to be reused for Europeana is the CIDOC CRM which the following subchapter will explain in more detail.

2.2.3.1 CIDOC CRM

According to Doerr [8] museum information is more diverse than any other cultural heritage information. As a result, in a long-lasting process the Committee on Documentation of the International Council of Museums (ICOM-CIDOC) has developed a Conceptual Reference Model (CRM, known as the CIDOC CRM) which is "a high-level ontology to enable information integration for cultural heritage data and their correlation with library and archive information" [8]. The model has become an ISO standard in 2006 (ISO 21127:2006; current version 5.0.1 [4]).

CIDOC CRM is a relationship-centric model that mainly describes objects in terms of the entities actors, temporal entities like time-spans and places, conceptual objects and physical stuff. (cf. Figure 2 [15]) As it provides a picture of the cultural heritage domain, it seems to be the perfect solution for enhancing semantic interoperability in Europeana beyond simpler semantic relationships defined by thesauri.

Figure 2. A qualitative metaschema of CIDOC-CRM

Yet, as simple as the qualitative metadata schema appears to be, the whole ontology is complex and comprehensive, which raises again the question of feasibility for Europeana considering the vast amount of sources. Doerr, one of the main creators of CIDOC CRM, recommends not implementing the entire ontology, but restricting entities and relationships to the ones absolutely necessary. He suggests to cherry-pick; thus apply the reference model according to one's needs.

Technically, CIDOC CRM is available as an OWL DL as well as an OWL2 implementation. In addition the CRM can also be encoded in RDFS. The specification proposals of these implementations can be viewed and downloaded. Besides it is compatible with the DC metadata standards. Also, CIDOC CRM has been harmonised with the FRBR model and resulting in FRBRoo , an ontology guaranteeing semantic interoperability between the documentation structures used for library and museum information.

3. CURRENT STATE-OF-THE-ART IN EUROPEANA

Europeana follows two approaches towards semantic interoperability. At the moment, some level of interoperability is assured at data ingestion by requesting the data provider to declare the "Europeana Semantic Elements (ESE)" for their metadata. These elements are currently being substantially redeveloped into a data model that should provide semantics properly speaking (cf. 3.1 Europeana Semantic Elements). Additionally the project works on the creation of a semantic data layer by mapping controlled vocabularies in use by the data providers to each other by using the SKOS standard. These approaches will be explained in the following subsections in more detail.

3.1 Europeana Semantic Elements

It has to be declared the choice to call this metadata specification "semantic" was not appropriate in the first place. In the meantime this mistake has been understood and will be considered when finalising the appellation for the currently named Europeana Data Model (which should rather be declared as an updated version of the Europeana Semantic Elements).

The ESE are used in the current prototype portal. They consist of the 15 original Dublin Core (DC) metadata elements, a subset of the DC terms and twelve elements tailored to meet Europeana's additional needs. They enable the search functions of the prototype portal where the user can search for title, creator, date and subject as well as sort by facets like date, type, provider, country, and language. Yet, the ESE do not support semantic search functions (e.g. related concepts). Hence, they need to be understood as a common metadata schema that is being developed into a more comprehensive data model providing semantics. [10]

3.2 Mapping of Controlled Vocabularies with SKOS

In addition to the ESE, Europeana applies the SKOS vocabulary to build the semantic data layer, i.e. creating a layer of interconnected controlled vocabularies that can be used to enrich existing object metadata. Most important is that metadata from the data providers include URIs to the concepts of controlled vocabularies (e.g. having <dcterms:spatial>Monaco</dcterms:spatial> in the metadata supplemented by the URI to the record ID: 7005758 in the TGN which represents Monaco, the nation.).

As explained in subsection 2.2.2 SKOS is used to describe thesauri and other controlled vocabularies, whereas the mapping vocabulary is used to establish basic semantic interoperability among them. The current activities in this field, particularly at the VUA (Vrije Universiteit Amsterdam) and the NTUA (National Technical University of Athens), include research on methods for aligning as well as mapping vocabularies, (semi-)automating these workflows, thus reducing the human intelligence factor and methods for fuzzy matching.

3.3 Europeana Data Model v5.02

The Europeana Data Model (EDM) is developed within the main Europeana v1.0 project by work package 3. They emphasise an open discussion with the Semantic Web and the Linked Data community. The approach of reuse is pursued, which can be seen in Figure 3 [5] depicting EDM's class hierarchy. Elements of the established standards OAI-ORE , RDFS, DCMI and SKOS are used. Other elements have been defined in the Europeana namespace. In a previous version they attempted to use the IRW ontology, which is a fairly new ontology for resources on the web modelling linked data resources [cf. 14]. However, this step seemed rather puzzling, as IRW is not an established standard yet and other standards have not been used (e.g. classes and properties that clearly derive from other ontologies: FRBR, ABC Harmony , CIDOC CRM etc.). Precise references from the "newly defined" Europeana classes to these equivalent classes need to be defined and added to the specifications.

Figure 3. The EDM Class hierarchy

In Figure 4 EDM's property hierarchy is displayed [5]. The majority of the properties are defined in the Europeana namespace, meaning they are not reused from other ontologies or schema. The properties prove to be an area where further research on the reuse of properties from existing metaschema or ontologies has to be carried out (e.g. the properties wasPresentAt, happenedAt and occurredAt are directly taken from the CIDOC CRM, yet not identified as such).

The EDM will be the data model to which the metadata of the information objects will be mapped. The search engine will search the data according to the EDM and retrieve the results accordingly. Thus semantic interoperability takes place at the search and retrieval level.

In Figure 4 EDM's property hierarchy is displayed [5]. The majority of the properties are defined in the Europeana namespace meaning they are not reused from other ontologies or schema. The properties prove to be an area where further research on the reuse of properties from existing metaschema or ontologies has to be carried out (e.g. the properties wasPresentAt, happenedAt and occurredAt are directly taken from the CIDOC CRM, yet not identified as such).

The EDM will be the data model to which the metadata of the information objects will be mapped. The search engine will search the data according to the EDM and retrieve the results accordingly. Thus semantic interoperability takes place at the search and retrieval level.

Figure 4. The EDM property hierarchy

4. RESEARCH QUESTION AND METHODOLOGY

In summary various factors need to be considered when creating a semantic data layer for a digital library. A crucial success factor is the interaction of the technical and the library scientific standards. This dissertation will investigate the following research questions identified in the context of semantic interoperability and digital libraries:

4.1 Research Question

How has the standard ISO 21127:2006 CIDOC CRM been applied in digital cultural heritage projects so far? What are the prerequisites, limitations, benefits? How does CIDOC CRM influence semantic interoperability?

As it will be analysed how CIDOC CRM is implemented to provide semantic interoperability in digital cultural heritage projects, the general conclusions drawn will answer the following questions: On the one hand, it will be identified whether limitations of the ontology CIDOC CRM exist and if so, what these limitations are. On the other hand, the benefits of a domain ontology like CIDOC CRM will be defined. Subsequently, the impact of these limitations and benefits on semantic interoperability will be explained. It shall also be investigated whether correlations exist between the amount of data as well as the quality of metadata and the level of semantic interoperability to be reached. This dissertation should find general conclusions or even rules on how CIDOC CRM influences semantic interoperability in a digital library environment.

What are the conclusions to be drawn from answering these questions for semantic interoperability in Europeana? Can the standard CIDOC CRM be integrated in Europeana to provide semantic interoperability? If so, what are the prerequisites, limitations, benefits?

It will be examined what parts of CIDOC CRM can be applied for the semantic data layer and how the currently used standards (OAI-ORE, SKOS, RDFS, DC) can interact with it. It shall be verified how and where the highest benefits for semantic interoperability in Europeana can be achieved. Reciprocally it shall be determined whether Europeana and its object representations need to meet certain requirements to take full advantage of CIDOC CRM. The question to be answered in particular is to what degree semantic interoperability can be augmented introducing CIDOC CRM.

4.2 Methodology

The research includes a literature review on semantic interoperability in digital libraries. It identifies and provides a definition of the most important concepts in this research area. It is the basis for the investigation of digital cultural heritage projects having implemented CIDOC CRM. This analysis will build upon the levels of semantic interoperability and their prerequisites identified by the DELOS project deliverable [18]. The projects are analysed, compared and evaluated in terms of background, prerequisites, approach towards the technical standards as well as semantic interoperability, and lessons learned. Furthermore it will be investigated what other standards have been used in these projects in combination with CIDOC CRM. Currently the literature review and the analysis of the digital cultural heritage projects are work in progress.

By observing other digital cultural heritage projects having implemented CIDOC CRM I will find out how they have approached semantic interoperability and determine the role of the CRM in it. The common characteristics will be identified as well as how they have contributed to semantic interoperability. Another important aspect will be if the projects are closed systems or if they integrate data from multiple systems. I will investigate how they have applied CIDOC CRM and if/how it was combined with other standards. Also, the integration into the Linked Data environment will be examined. Beyond the theoretical prerequisites and levels of semantic interoperability described in the DELOS project deliverable [18] the model for search and retrieval will also be matter of investigation, as it needs to be determined how CIDOC CRM supports the retrieval process. Additionally I will examine the richness and the granularity of the metadata and if the full or only parts of the model were applied. Europeana needs to combine information objects from heterogeneous sources originating from the museum, the library and the archival area. Thus the ways of application of CIDOC CRM for those communities will be compared. Differences and similarities will be identified.

The analysis of the projects will be based on project reports and the actual implementation. If open questions remain, experts from these projects will be contacted. The method of interrogation will be determined according to the issues that need to be solved.

This evaluation of CIDOC CRM's current use and technical implementation will shed light on how it provides or enhances semantic interoperability. It will show what criteria need to be met to achieve semantic interoperability and what factors influence it. Furthermore the different levels of semantic interoperability will be considered. For Europeana I will define how to implement CIDOC CRM in order to provide or enhance semantic interoperability. Possible solutions may include but are not limited to an application profile as a middle layer between the metadata and the EDM.

The results shall allow conclusions for CIDOC CRM in terms of semantic interoperability and applicability for a digital library. Interoperability of digital libraries has been a research challenge for some time. Tackling the semantic aspect of interoperability has been neglected. Mostly syntactic or other aspects of interoperability have been investigated so far. Therefore research on semantic interoperability is not very comprehensive, especially in the digital library context. Europeana unites representations of cultural heritage objects from many heterogeneous sources; hence semantic interoperability has proven to be an important issue that needs to be examined in depth for the future. Additionally, summarising the research that has been carried out on CIDOC CRM will allow a better understanding of its implementation potentials. Thus, this dissertation will contribute to the research on semantic interoperability in digital libraries by evaluating the applicability of CIDOC CRM.

5. REFERENCES