Database Reference
In-Depth Information
2004) and the Web Ontology Language OWL. There was also significant effort in
the development of reasoners to reason over OWL such as Pellet (Sirin et al., 2007)
and Racer (Haarslev and Möller, 2001), as well as the development of query lan-
guages to interrogate the data encoded in RDF, such as SPARQL (Prud'hommeaux
and Seaborne, 2008) and the less-used RQL (Karvounarakis et al., 2002), RDQL
(Seaborne, 2004), and SeRQL (Broekstra and Kampman, 2003). There were also
many new ontologies appearing, representing the knowledge in domains such
as bioinformatics, for example, the Gene Ontology,
3
law (Hoekstra et al., 2007),
and geography.
4
This led to many interesting results in knowledge representation,
particularly the rise of ontology design patterns (Gangemi, 2005).
An ontology design pattern is a “reusable successful solution to a recurrent mod-
elling problem”
5
that can help authors to communicate their knowledge within the
confines of their chosen ontology language. However, at this time there was also
something of a struggle between the various ontology languages for dominance.
Suffice it to say for now that there was the choice between more complex and expres-
sive languages (such as the Knowledge Interchange Format or KIF [Genesereth and
Fikes, 1992], which used first-order predicate calculus; or CYCL [Cycorp, 2002],
which used higher orders of logic); simpler languages such as the precursor to OWL
called DAML+OIL (Connolly et al., 2001) and RDFS (Brickley and Guha, 2004)
used to describe RDF vocabularies; or ones based on Description Logics, such as
the OWL variant OWL-DL, which balanced expressivity against complexity. That
is, the authors of OWL-DL tried to strike a balance between expressing knowledge
of the domain accurately and guaranteeing, within a finite time, that the reasoner
operating on the language would find all the correct answers to the query posed.
Another research theme at this time was the development of so-called upper ontol-
ogies such as the early UML-based knowledge model CommonKADS (Schreiber
et al., 1999) or more recently SUMO (Pease, Niles, and Li, 2002) and DOLCE
(Gangemi et al., 2002). These ontologies, also known as “top-level” or “foundation”
ontologies, describe concepts at a very general level to lay a “foundation” for knowl-
edge above any individual domain. An ontology that describes knowledge about a
specific domain can then inherit from the upper ontology. That is, all concepts in
the domain ontology can be expressed as subclasses of concepts in the upper ontol-
ogy. This then assists interoperability between domain ontologies based on the same
upper ontology. For example, the DOLCE ontology specifies concepts like
endurant
(a concept that can change over time while remaining the same concept, such as a
person) and
perdurant
(a concept that occurs for a limited time, such as an event or
process). An ontology describing a specific domain would then inherit from DOLCE;
for example, a river would be a kind of endurant and a flood would be a kind of perdu-
rant. A second ontology, say on habitats, that is also based on DOLCE might contain a
concept like freshwater habitat, which because it is also a kind of endurant could then
be related to the river concept in the first ontology as they share a common ancestral
concept. Clearly, this example is very simplified, and many more layers of granularity
would necessarily be included in the ontology to fully describe the domain.
There are, however, some drawbacks to using upper ontologies, not least because
it can be very difficult for an expert in a particular domain such as GI to understand
exactly which of the oddly termed classifications to assign to their concepts. Should a
