Principles of formal ontology are rarely applied to the construction of large ontological or lexical resources such as WordNet or Yago, while it is well-known that their coherence from this viewpoint is far from satisfactory, and any improvement would have a large impact on their many applications.
The OntoClean methodology uncovers some mistakes, but requires costly and expertise-intensive manual work which is unrealistic to implement beyond the top-level. Automatic provers are able to find inconsistencies in logically implemented resources, but extra effort is needed to build an explanation of the error to help fixing it. Crucially, automatic provers can hardly focus on classical ontological mistakes, a fortiori when axiomatization is shallow.
I will explore the feasibility of a semi-automatic method to improve the ontological coherence of ontologies and lexical resources, even not logically-implemented ones. Ontological errors are automatically extracted and classified so that the user engaged in improving the resource's coherence focusses on possible solutions. Such a method could also be implemented within authoring systems.
Laure Vieu is a French National Research Council (CNRS) researcher at Institut de Recherche en Informatique de Toulouse (IRIT). She has been working in formal ontology, formal semantics, lexical semantics, discourse semantics, and their interplay for over 20 years. She has coordinated a long-term joint lab on interacting knowledge systems between (among others) IRIT and the Laboratory for Applied Ontology from the ISTC-CNR in Trento, Italy, where she was on secondment for 7 years.