Previous Page Table of Contents Next Page


3. The ontology: Modelling and representation


In the context of the AOS, an ontology is a system of terms, the definition of these terms and the specification of relationships between the terms. It extends the approach of classical thesauri by providing the opportunity of creating an infinite number of different semantic relationships. For an overview about different types of ontologies, refer to (Guarino 1998). The following gives a detailed description of the modelling approach used for our representation of the prototype ontology:

Semantic robustness towards representational changes, as well as multilingualism, are crucial for the development of this domain ontology (see section 1.1). Therefore, we distinguish between terms, and the concepts these terms represent. Whereas terms might change, and are different in each language, the semantic meaning and interpretation of the terms’ abstract concept stays the same[2]. In the presented modelling approach, a concept’s term representations are called Lexical Entries. These Lexical Entries are limitless and may be characterized as labels, synonyms or word stems. Furthermore, each Lexical Entry has at least two attributes: the concept it refers to and its language. Lastly, relationships between concepts can be established, annotated by the same lexical entries. This approach can be described as a two layered model, in which the semantic layer of the ontology is totally independent from its representation layer and hence, robustness against changes can be achieved.

Ontologies can be represented in different representation languages. (Palmer 2001) gives a brief overview about these languages and provides further information. RDFS[3], the language that was chosen to be used within the AOS framework, is used to define vocabularies of resources and relationships amongst them. Resources can be documents, web pages or parts of them, basically anything, which can be referenced by a URI[4]. RDFS provides a basic set of modelling primitives, which can be easily extended by users to include domain specific semantics in terms of relationships among concepts. Furthermore RDFS models are exchanged via XML and therefore provide interoperability between communities. Although still under development, RDFS evolves to serve as a standard representation in the context of the Semantic Web. For a detailed discussion about modelling ontologies in RDFS, refer to (Staab et al. 2000a).

Figure 2 gives an overview of the above-discussed layered modelling approach in RDFS. The top layer represents an extract of the basic layer provided by the RDFS language. The lexical layer creates the needed abstraction of lexical and language representation from conceptual domain semantics. The lowest layer finally constitutes the domain. The most generic class in RDFS is rdfs:Resource[5], from which every other class is derived An rdfs: Class can be instantiated to define domain specific concepts. Lexical Entries are separate classes which can be instantiated and attached to concepts using the properties kaon:references and kaon:inLanguage. Each property has a domain and a range, which determine the source and the target of the relationship respectively. In that way, an infinite number of lexical entries can be instantiated and related to domain concepts and different languages. If a representation of a concept in terms of its lexical entry changes, the semantics of the ontology are not affected, since it still refers to the same concept. Furthermore, additional domain properties can be derived from rdf:Property in the application layer to relate the domain concepts and build the semantic network.

Figure 2: Layered RDFS model for multilingual ontology representation

This generic, multilingual ontology model establishes the basis for our engineering methodology framework, which will be presented in the following section.


[2] This holds in most cases. There are however cases, where a concept does exist in one culture, even though there is not adequate concept in another one. This is however more evident in humanity domains, since concepts there are richer and less well defined. The project environment here is rather technical and hence chances for this can be neglected.
[3] http://www.w3.org/TR/rdf-schema/#intro.
[4] Uniform Resource Identifier. See also http://www.w3.org/Addressing.
[5] The prefixes rdfs: , rdf: , kaon: , bio: represent XML namespaces and are to uniquely identify each resource. Refer to (RDFSchema 2002) to learn more about RDFS and namespaces.

Previous Page Top of Page Next Page