-
This paper considers the problem of tracking the changes between di erent
versions of an ontology, which can be done based on the notion of logical di erence.
The logical di erence between two ontologies are the axioms in one that are not
entailed by the other, re ecting information gained and information lost between
the ontologies [
5
].
For ontology engineers, tracking the logical di erence in industrial ontologies
is a critical task for the purpose of maintaining them: to track what has changed
in a new version of an ontology, to ensure that the extension is safe in the sense of
being a conservative extension [
9
], and to identify information gained and
information lost [
4
]. This provides a means of discovering issues in the ontologies and
supports quality assurance reviews. Being able to compute the logical di erence
between two ontologies is also important when merging/aligning ontologies and
integrating di erent ontologies [
3,12
].
Presently available ontology di erence and alignment/merging tools compute
the structural di erence between ontologies [
11,1
], which is a syntactic notion,
or approximate the logical di erence between ontologies [
5,3,7,12
], which is a
semantic notion. A particular challenge is the size of ontologies used in
realworld applications. Our targets are the Snomed CT and NCIt ontologies. Being
the most comprehensive, multi-lingual medical ontology in the world, the core
Snomed CT ontology contains over 335K concepts de nitions [
13
]. The NCIt
ontology de nes terminologies in the biomedical domain and contains more than
60K concept de nitions [
2
].
This paper explores how the logical di erence between two ontologies can be
computed using a forgetting- or uniform interpolation-based approach, as
proposed in [
8
]. Forgetting is an ontology re-engineering technique which preserves
the semantics of de nitions of concepts and relationships among them [
6,10,14
].
Current forgetting tools have di culties in tracking the logical di erence in
very large ontologies. To address this shortcoming we introduce a new forgetting
method designed for the task of tracking the logical di erence in large-scale
ALContologies. The method is sound and terminating, and can compute forgetting
solutions for ontologies as large as Snomed CT and NCIt. Our evaluation shows
? This is an extended abstract of a paper published in the proceedings of AAAI 2019.
the method can achieve considerably better success rates (>90%) over existing
tools such as Lethe and Fame 1.0 on the restrictions to ALC of Snomed CT
and NCIt. Based on this forgetting method we have developed a new logical
difference tool to analyze the di erences between the Australia and Canada country
extensions of the base Snomed CT International ontology and recent versions
of the NCIt ontologies. While the Canada extension was found to be a
conservative extension of Snomed CT International, there is signi cant diversion in
the Australia extension, as is re ected in large logical di erence sets.