After evaluating the generated explanations, we concluded that certain improvements are impossible without prior elaboration of the knowledge base (mostly with respect to detailness and granularity of the encoded facts). One possible improvement of the knowledge `quality' is to refine the ontological relationships. Therefore, we proposed above elaboration of the ISA relation and also explicit representation of roles and perspectives. Naturally, these finer-grained distinctions provide a better basis for semantic and syntactic choices.
General-specific distinction:
We distinguish the characteristics, valid for all instances of a certain type, and the individual attributes. Using inheritance of characteristics, we provide explanations similar to, e.g., ``Oil-separators are characterised with efficiency and capacity. The separator FREYLIT has a purification class III according to the standard B5101 (i.e., it cleans polluted waters with concentration below 5 mg/l)''.
Explanation of exclusive types:
When the mutually-exclusive classification is explicitely encoded (see Figure 4), explanations of the following kind can be generated: The physical state of objects is either liquid or solid.
Roles, natural types and perspective:
Figure 4 shows how the IS-A relation can be
specified
from a particular perspective. Similarly, the IS-A taxonomy
can distinguish explicitly between ``role types''
and ``natural types''. In this way we capture the phenomenon that artifacts
have natural types that cannot be changed, as well as role types
which are subject to temporal changes. For instance, one PERSON can be
characterised by the role types CHILD and ADULT depending
on his age.
However, such distinctions are quite rare for technical domains and
usually the different viewpoints to objects are much mode important.
Thus explanations from a given perspective can be based on the
enhanced ontological information, maybe also combined with
corresponding
``perspective-specific'' operations. For instance, definitions can be
enhanced with more elaborate explanation of the ISA relations
and, more specifically, multiple parent types:
Objects by physical state are either liquid or solid while in
ecology they might be
polluted or non-polluted.
Perspectives also support the representation of intersecting
``domains'' in a common knowledge base where the explanations can be
tailored to the interests of the particular user in each of these
viewpoints. For example, users
interested in ecology can receive only relevant explanations by
application of a special
type of inheritance which accounts for the ISA-KIND relation.
Similarity of objects:
The detailed classification in the type hierarchy provides adequate explanations of the concept position. Thus we can explain objects and characteristics inherited from several supertypes, e.g., ``Oil particles, viewed as oil, are lighter than water; while as physical objects they are particles and have dimension and weight.''
Based on the position in the type hierarchy, DB-MAT also generates explanations of concept similarity: ``Oil separators collect oil from the water surface, while precipitators exploit the principle that oil and admixtures precipitate at the bottom of the chamber.''This explanation is produced for the sister types OIL_SEPARATOR and PRECIPITATOR by verbalising their type definitions.