Operations on Ontologies
It is possible that one application uses multiple ontologies, especially when using modular design of ontologies or when we need to integrate with systems that use other ontologies. In this case, some operations on ontologies may be needed in order to work with all of them. We will summarize some of these operations. The terminology in this areas is still not stable and different authors may use these terms in a bit shifted meaning, and so the terms may overlap, however, all of these operations are important for maintenance and integration of ontologies.
- Merge of ontologies means creation of a new ontology by linking up the existing ones. Conventional requirement is that the new ontology contains all the knowledge from the original ontologies, however, this requirement does not have to be fully satisfied, since the original ontologies may not be together totally consistent. In that case the new ontology imports selected knowledge from the original ontologies so that the result is consistent. The merged ontology may introduce new concepts and relations that serve as a bridge between terms from the original ontologies.
- Mapping from one ontology to another one is expressing of the way how to translate statements from ontology to the other one. Often it means translation between concepts and relations. In the simplest case it is mapping from one concept of the first ontology to one concept of the second ontology. It is not always possible to do such one to one mapping. Some information can be lost in the mapping. This is permissible, however mapping may not introduce any inconsistencies.
- Alignment is a process of mapping between ontologies in both directions whereas it is possible to modify original ontologies so that suitable translation exists (i.e., without losing information during mapping). Thus it is possible to add new concepts and relations to ontologies that would form suitable equivalents for mapping. The specification of alignment is called articulation. Alignment, as well as mapping, may be partial only.
- Refinement is mapping from ontology A to another ontology B so that every concept of ontology A has equivalent in ontology B, however primitive concepts from ontology A may correspond to non-primitive (defined) concepts of ontology B. Refinement defines partial ordering of ontologies.
- Unification is aligning all of the concepts and relations in ontologies so that inference in one ontology can be mapped to inference in other ontology and vice versa. Unification is usually made as refinement of ontologies in both directions.
- Integration is a process of looking for the same parts of two different ontologies A and B while developing new ontology C that allows to translate between ontologies A and B and so allows interoperability between two systems where one uses ontology A and the other uses ontology B. The new ontology C can replace ontologies A and B or can be used as an interlingua for translation between these two ontologies. Depending on the differences between A and B, new ontology C may not be needed and only translation between A and B is the result of integration. In other words, depending on the number of changes between ontologies A and B during development of ontology C the level of integration can range from alignment to unification.
- Inheritance means that ontology A inherits everything from ontology B. It inherits all concepts, relations and restrictions or axioms and there is no inconsistency introduced by additional knowledge contained in ontology A. This term is important for modular design of ontologies (see later) where an upper ontology describes general knowledge and a lower application ontology adds knowledge needed only for the particular application. Inheritance defines partial ordering between ontologies.
Not all of these operations can be made for all ontologies. In general, these are very difficult tasks that are in general not solvable automatically -- - for example because of undecidability when using very expressive logical languages or because of insufficient specification of an ontology that is not enough to find similarities with another ontology. Because of these reasons these tasks are usually made manually or semi-automatically, where a machine helps to find possible relations between elements from different ontologies, but the final confirmation of the relation is left on human. Human then decides based on natural language description of the ontology elements or decides only based on the natural language names of the ontology elements and common sense.
An example of such a tool is Chimaera that helps with merging ontologies. It provides suggestions for subsumption, disjointness or instance relationship. These suggestions are generated heuristically and are provided for an operator, so that he may choose which one will be actually used. PROMT or SMART system is a similar system that provides suggestions based on linguistic similarity, ontology structure and user actions. It points the user to possible effects of these changes.