Metamodels for RDF Schema and OWL

This paper presents the working draft MOFTM (Meta-Object Facility) metamodels for the Resource Description Framework (RDF Schema) and the Web Ontology Language (OWL), two of the six metamodels currently envisioned for the Ontology Definition Metamodel (ODM) standards effort in the Object Management Group (OMG®), which enable model-driven development of RDF vocabularies and OWL ontologies, respectively. We provide insight into some of the design principles used in developing these metamodels, major challenges addressed to date, and the resolution of some of these issues that has influenced the resultant products. We also briefly review ongoing and future work needed to complete the subset of the ODM specific to these representation formalisms and fully support model driven development of RDF vocabularies and OWL ontologies. INTRODUCTION Over the course of the last five years, and more specifically, since the emergence of Semantic Web Activity from the World Wide Web Consortium (W3C) [1], the development of ontologies—explicit formal specifications of the concepts in a domain and relations among them [2]—has been moving from the research community to early adoption by industry. Increasing evidence of collaborative development of large, standardized controlled vocabularies and ontologies for specific applications and domains, such as in bioinformatics and pharmacogenomics research, is appearing in the literature. Broadly applicable general-purpose ontologies, for example those supporting the Semantic Web Services Initiative [3], are emerging as well. Ontologies are primarily captured in knowledge representation languages developed by the artificial intelligence community. Most of the more commonly-used authoring languages, such as the description logics family of languages [4] or first-order and predicate logic languages, such as Simple Common Logic (SCL) [5] or its predecessor, the Knowledge Interchange Format (KIF) [6], were designed to support machine reasoning. Their text and logic based structure, however, has led to a language syntax that is unfamiliar and awkward for subject matter experts to learn and use effectively, which has been a major hindrance to the development of ontologies. Notably, the recently adopted W3C specifications for the Resource Description Framework (RDF), RDF Schema [7] and the Web Ontology Language (OWL) [8,9] in particular, are somewhat less intimidating to those familiar with XML syntax or constructs used in object-oriented programming, such as classes, properties, and individuals. OWL is an extension of RDF Schema, which is the vocabulary description language for RDF. Both RDF Schema and OWL are key components of the W3C Semantic Web initiative and are likely to gain increasing support in the industry. Having domain expert friendly languages, however, is necessary but not sufficient to promote widespread adoption of these technologies. Current approaches to ontology development are at best, more art than science, and in general, ad hoc. First, the process of ontology development is extremely time consuming and not at all visually intuitive. Any non-trivial ontology represented in OWL, as is, is challenging for domain experts to understand and maintain. Secondly, developing ontologies in isolation of business requirements is of little practical value. Ontology development must become an integral part of the systems analysis and engineering activities of the CIO function. That is, the ontologies that an enterprise develops must form an integral part of that enterprise’s information and application infrastructure This paper promotes the use of Model Driven Architecture (MDA®) and related methodologies for ontology development. This is prompted by current trends towards MDA in software engineering and best practices, as a result of the related standards efforts in the OMG, and the availablity of EMF (Eclipse Modeling Framework) as an open source model-driven software development platform. Key standards in the MDA family include the Meta-Object FacilityTM (MOF®) [10] and the Unified Modeling LanguageTM (UML®) [11]. UML is an industry-standard, graphical language that is used by software engineers for conceptual modeling. The similarity of UML constructs to constructs used in defining ontologies suggests that UML could be leveraged by the large community of existing practitioners to promote broader use and increasing development of ontologies. UML is well established in many commercial and government software engineering organizations with extensive tool support from both commercial and open source vendors. We believe, therefore, that UML is an excellent candidate notation for the graphical development and maintenance of ontologies. To facilitate the development of tools and methods for MDA-based ontology development, as UML is defined using MOF, what is needed is a MOF based metamodel for OWL. Such a metamodel will enable: ! Forward engineering: development of OWL ontologies using MOF based (in particular, UML) modeling tools. ! Reverse engineering: leveraging existing OWL ontologies for ontology modeling and UML modeling. ! Integrated ontology/software development: making OWL ontology development an integral part of software development. This paper presents the draft set of MOF metamodels for RDF Schema and OWL that are currently proposed as a part of the Ontology Definition Metamodel (ODM) activity in the OMG, enabling model-driven development of RDF vocabularies and OWL ontologies, respectively. Because of space limitations, we provide only a brief overview of the metamodels here. We present insight into some of the design principles used in developing these metamodels, major challenges addressed to date, and the resolution of some of these issues that has influenced the resultant products. We also briefly review ongoing and future work needed to complete the subset of the ODM specific to these representation formalisms and fully support model driven development of RDF vocabularies and OWL ontologies. DESIGN PRINCIPLES AND MAJOR CHALLENGES Both RDF Schema and OWL are defined using RDF Schema. That is, RDF Schema serves as the meta language that defines itself and OWL. However, to leverage MDA, EMF, and UML technologies, MOF serves as the meta language, or meta-metamodel used to define the metamodels for RDF Schema and OWL. In developing these metamodels, we encountered a number of distinctions between MOF and RDF Schema that have either challenged our ability to leverage MOF for this purpose or have resulted in metamodels that are less than ideal, as discussed below. RDF Schema uses URI references for naming. Definitions specified in RDF documents (class, property, individual) have globally unique names qualified by namespaces, such as rdfs:Class, rdf:Property, rdfs:subClassOf, owl:Class, and owl:equivalentClass. MOF, on the other hand, uses package-scoped naming for classes and locally scoped naming for associations. In the current set of metamodels, to clearly indicate what a particular notion, such as a class or an association, describes (i.e., its source), we have elected to use prefixes to denote namespaces. Thus, the entities itemized above are called, respectively: RDFSClass, RDFProperty, RDFSsubClassOf, OWLClass, and OWLequivalentClass. RDF Schema and OWL (and actually most knowledge representation formalisms and methodologies) do not make distinctions between meta-levels. For example, rdfs:Class is used to define owl:Class (i.e., owl:Class is an instance of rdfs:Class), but owl:Class is also defined as a subclass of rdfs:Class. From an MDA perspective, however, these distinctions are extremely important: MOF class, at the M3 level, is used to define M2 level classes, such as RDFSClass and OWLClass. UML does not support representation of classes and objects on the same diagram, which is a frequent requirement in knowledge representation. As a result, we have attempted to respect separation of meta-levels whenever possible in the ODM metamodels, for example, OWLClass is defined as a subclass of RDFSClass, but it is not an instance of RDFSClass. RDF Schema and OWL are based on concepts from formal logic, as mentioned above. They provide unary predicates, i.e., classes and individuals, for representing concepts in a domain and binary predicates, in the form of properties, for representing relations between concepts. Both RDF classes and RDF properties are first-class entities and have globally unique identifiers. MOF, on the other hand, is class based. It provides classes for representing concepts in a domain, attributes for representing characteristics that are common to all instances of a given concept, and associations for representing relations between concepts. MOF classes are first-class entities that may have globally unique identifiers (package scoped). Attributes and associations, however, are not first-class citizens. They have only class-scoped, local names, as mentioned previously. This raised a fundamental question as to how best to represent RDF properties in the metamodels: reified as MOF classes so that they have globally unique identifiers or as MOF associations so that they appear more natural to MOF/UML users. In the current metamodels, we have described predefined RDF properties, e.g., rdfs:subClassOf and owl:equivalentClass, as MOF associations in order to make the metamodels easier to understand, simplify the serialized XMI rendering of the metamodels, and for ease of use in MOF and UML tools. Three prefixes (RDF, RDFS and OWL) have been used to denote namespaces, for example RDFSsubClassOf and OWLequivalentClass, to “simulate” globally unique names, however. Again, the primary goal of this work is to enable model driven development of RDF vocabularies and OWL ontologies. As such, we want the metamodels to be as directly representive of RDF Schema and OWL, respectively, as possible. Therefore, we have kept to a minimum any construct that is not explicitly defined in RDF Schema or OWL. Where we felt such a construct was necessary, for example, Vocabulary / Graph for scoping purposes in the RDF Schema metamodel, we have named these artifacts without any global prefix for further clarification. THE RDF SCHEMA METAMODEL The RDF Schema metamodel uses diagrams to control complexity and promote understanding, as exemplified in Figures 1 and 2. The MOF classes and MOF associations are grouped into seven diagrams: ! Classes — Contains classes and associations that can be used to define RDF classes and RDF datatypes. ! Properties — Contains classes and associations that can be used to define RDF properties. ! Containers — Contains classes and associations that can be used to define RDF containers and their members. ! Collections — Contains classes and associations that can be used to define RDF collections (i.e., lists) and their members. ! Reification — Contains classes and associations that can be used to define RDF statements. RDFXMLLiteral RDFSDat atype RDFSCl ass 0..* +RDFSsubClassOf