An upper ontology is a an ontology which describes concepts which remain the same across multiple domains [37]. These concepts are general enough that their meaning, or what they represent, is not dependent on the context or domain in which they are used.
For example, it could be argued that a Person concept is general enough to be applicable to many domains without having to change the definition. A Person within the home domain would be the same Person within the work domain. This Person concept would remain unchanged if moving from a location-based domain to a more natural domain such as a legal or family domain. The Person concept would be relevant to all domains. (This does not mean all attributes of the Person concept are relevant to all domains, but that the definition of the Person concept is true in
all domains).
Upper ontologies are useful for building large scale ontologies [37, 4]. The upper ontologies can be used as foundations for future development. The use of a third- party ontology as a foundation increases the chance of agents, who use one set of ontologies, being able to act in an environment described using another set [4]. When encountering an unknown concept or term, an agent may be able to reduce the concept into terms found within the third party ontology. If successful, the agent can then act on lower-level information, perhaps reconstructing the terms into concepts found within its own ontologies. (Other advantages of reusing ontology terms has been previously discussed in sections 5.5.2 and 5.5.3).
In simple terms, upper ontologies are a standardised set of ontologies designed to provide a common foundation for applications or further ontologies across a range of domains. This section will review the process involved in creating an upper ontology, and discuss existing upper ontologies used within the research domain.
7.3.1 Developing and Using Upper Ontologies
In creating an upper ontology, concepts, properties and rules are created and devel- oped. These concepts may be simply stated entities, such as those in a taxonomy, or may have meaning defined through properties and rules. As already mentioned, upper ontologies may be wide ranging, but are rarely constrained to a particular domain. Upper ontologies may consist of a number of domain specific ontologies in order to be as inclusive as possible.
Developers can create their own ontologies using terms defined within an upper ontology [27]. At this stage, developers may tailor their ontology to a specific domain without invalidating the purpose of an upper ontology. This is because ontologies built upon an upper ontology will share a common definition of terms. A reasoner application, which can understand the ontology data and infer new information, can utilise the new domain specific information. The reasoner application can then draw conclusions based upon both the protocol specific ontology and the upper ontology data. Conclusions may be the result of a user query, or the adding of new
Figure 7.3: Customising an Upper Ontology for Use information.
A number of upper ontologies have emerged within the research domain. Some have been developed for a specific purpose, while others are concerned with providing a consistent and structured approach to describing elements in multiple domains.
7.3.2 Cyc and OpenCyc
The Cyc project began in 1984, with the aim of developing an extensive set of ontologies capable of supporting artificial life [55, 84]. One main project aim is to support reasoning and recognition over natural language data. Using an example given by [55], the following sentences can cause problems for artificial intelligence agents:
• Fred saw the plane flying over Zurich. • Fred saw the mountains flying over Zurich.
The Cyc project expresses information through CycL, a first-order logic lan- guage designed specifically for the project. Using CycL, facts about objects can be expressed, which can then be used to solve syntactic ambiguities like that shown above. Planes, of course, can fly, while mountains cannot. These facts can be known by a Cyc agent, and in turn the meaning of the sentences can be derived. In the first case, Fred is looking up at a plane. In the second case, Fred must be in a plane
looking down over the mountains (while on his way to Zurich!). Understanding the sentence allows new information to be gained.
At present, the Cyc knowledge base contains 300,000 concepts, and nearly 3 million assertions. The Cyc project provides a reduced set of concepts through OpenCyc, which is an open-source version of the Cyc knowledge base. There exists an open-source version of the Cyc ontology, named OpenCyc. OpenCyc has also been translated from CycL into OWL, and contains a set of permanent end-points representing various concepts within the ontology [56]. Despite being an reduced version of the full Cyc knowledge base, OpenCyc still contains a substantial amount of information [57].
7.3.3 The Suggested Upper Merged Ontology (SUMO)
The Suggested Upper Merged Ontology is another movement towards creating an upper ontology to support multiple domains and applications [73]. It is written in a first-order logic language called SUO-KIF and is controlled by the IEEE, although SUMO is in fact open source. SUMO is a collection of ontologies from various general and specific domains, such as Economy, Geography and Government. It can also supply information from external sources, such as Wikipedia [79]. It boasts 20,000 terms (concepts) and 70,000 axonims (relationships). A translation of SUMO from SUO-KIF into OWL is also available [79]. SUMO, like Cyc, is concerned with expressing information in a structured way, so as to allow artificial intelligence to act on the facts, and inferences present within the knowledge base. SUMO is also well grounded within the WordNet lexicon, which is a large database of English terms, relationships and meanings designed to support artificial understanding of words.
7.3.4 DOLCE
DOLCE stands for a Descriptive Ontology for Linguistic and Cognitive Engineer- ing. It is heavily engineered towards assisting machines in understanding the human language [75]. In this manner, as machines understand the language, they are able to identify subjects (such as a Person), actions and able to predict consequences
Figure 7.4: Boundaries Between Protocols
of existing facts and statements. The DOLCE project is also involve in integrat- ing WordNet into the DOLCE ontology (through the OntoWordNet project [75]). Rather than developing ontologies for a specific domain, DOLCE its aimed at defin- ing what words mean within the general human language. These terms can then be carried into specific domains, as developers can be confident that the terms are both correct, and well defined.