
lesser importance. Issues like synonymy are not
central, and can be dealt with in terms of the
multifunctional attributes of concepts and instances.
In terms of relations, the ontology model still places
a focus on the hierarchical subsumption relation, but
allows for arbitrarily many other relations. The
ontology model thus takes a more appropriate focus
and enhances expressive power by not limiting the
number of relations, while still being able to deal
with the issues central to the thesaurus model.
Finally, semantic networks subscribe to the more
generic view that concepts and instances have so
much in common that it warrants having only one
type of object, also termed “concept”. The
distinction between the concepts and instances of the
ontology model can still be made using attributes.
The semantic network model also liberalizes the role
of the relations. Relations are all given equal status:
there is no a priori focus on the subsumption
relation. Like the ontology model, the semantic
network model does not limit the number of
relations. In comparison to the ontology model, the
semantic network model therefore is more generic: it
retains the expressive power, and makes it much
easier to handle concepts and instances as the same
type of object where appropriate.
In summary, moving from the thesaurus model via
the ontology model to the semantic network model,
we see that each step corresponds to important
additional insights, and each successive model is
more capable to appropriately serve as a model for
conceptualizations. The expressive power of each
previous model is retained, not by extending the
previous model, but by replacing the modeling
primitives with more generic ones. As a conse-
quence, necessary features must be softcoded in
terms of the modeling primitives instead of being
hardcoded as modeling primitives themselves. This
means that in addition to the model itself it must be
specified how to use modeling primitives such as
attributes to implement the required features.
The structure of conceptualizations helps build the
conceptualization. As we have seen the thesaurus is
simple to build, while the ontology and semantic
network call for more effort. Furthermore the
structure of the conceptualization also limits or
extends its inference capabilities. As for the
thesaurus and the ontology, we saw that they were
limited by the top-down tree structure. The semantic
network is not limited by its structure.
The number of relations in a thesaurus is limited.
This results in a simple conceptualization that can be
built more easily. The relations in an ontology and
semantic network are not limited. This implies a
complex structure, especially for a semantic
network. Being able to use lots of relations results in
powerful inference capabilities.
Thesauri do not distinguish between concepts and
instances. They use preferred terms to represent the
concept. Ontologies do distinguish between concepts
and instances, while semantic networks do not. The
distinction between preferred terms and other terms
is simple and straightforward. Distinguishing
between concepts and instances, as in ontologies, is
a more complex approach. In semantic networks no
distinction is made between concepts and instances.
This results in a complex network, but also in the
advantage of being able to abstract from a concept
more easily.
Higher-order processing for knowledge mapping
involves some reasoning. In the thesaurus
configuration, inference is hardcoded into the
software and is limited by the structure of the
conceptualization. Alternatively, in the ontology and
semantic network configurations, we are not limited
by the structure of the network. Inference typically
makes use of the specific relations in the KMDB.
Thus, inference capabilities of the
conceptualizations only make sense if the KMDB is
integrated with the conceptualization, because this
enables the explicit coding of the inference rules.
Now we will judge the conceptualization based on
the criteria listed in section 3. All conceptualizations
can include all terms that describe the concept, e.g.
synonyms and spelling variants. Thesauri do both
explicitly. In ontologies and semantic networks the
spelling variants are treated as synonyms. Thesauri
define a preferred term, ontologies and semantic
networks do not. The conceptualizations all define
other relations. Thesauri usually consist of three or a
limited number of relations. Ontologies and
semantic networks usually also have a predefined
number of relations, but theoretically can account
for an unlimited number of relations.
In our view there are two viable conceptualizations
for knowledge mapping: the thesaurus and the
semantic network. The thesaurus variant is geared
towards simplicity, ease of implementation, and
reduction of work. The semantic network variant is
geared towards maximum expressiveness. It has a
more complex structure, which makes it harder to
implement. It also allows for integration with the
KMDB.
Table 1: Comparison of conceptualization Tools
Thesaurus Ontology Semantic
Net
Complexity low medium medium
Labour
intensity
medium high high
Integrated – + +
EXPLICIT CONCEPTUALIZATIONS FOR KNOWLEDGE MAPPING
235