Ceusters Abstract
Werner Ceusters: The Axiomatization of BFO 2020: The Good, The Bad, and The Ugly
Abstract: The Basic Formal Ontology (BFO) is undoubtedly the most frequently cited upper ontology in the biomedical literature. It is also claimed to be used as the foundation for many biomedical application ontologies. However, a closer look at these ontologies makes it clear that many of them deviate considerably from even the most basic principles upon which the BFO is built. Reasons can be derived from published papers that describe these ontologies. One reason is the popularity of OWL and tools such as Protégé of which the limits are poorly understood. A second one is the loss of semantics that arises from representing the BFO exclusively by means of binary relations, thus ignoring certain principles underlying the BFO, notably the requirement for temporal indexing in relations involving continuants. Although the importance of these and other principles has been stressed in many papers and presentations by BFO authors, education therein seems to be lacking or is at least inadequate.
The recent axiomatization of the BFO in First Order Logic (FOL) might offer an opportunity to remediate the situation. Soon after its public availability in September 2021, I started to use the BFO’s CLIF files in my ontology and referent tracking courses to explain BFO’s foundational principles. For some assignments, students were expected to write axioms for terms and relations they intended to include in the ontologies they were developing as part of their MSc or PhD thesis. I also embarked on a still ongoing project to develop a set of tools to make axiom-based ontology design easier for students lacking a solid background in formal logic and/or computer science. It is therefore encouraging to see that others are following similar paths and, for example, have shown that automatic reasoning with FOL annotations can be used to detect previously unnoticed errors in simple OWL-based classifications.
In this talk, I will focus on some of my and my students’ experiences with the use and development of a parser/generator for BFO2020-style CLIF axioms, and a reasoner for satisfiability testing of both axiomatized ontologies and instance data expressed in their terms. I will discuss how these experiences – good ones (yeah, temporal reasoning becomes possible!), bad ones (is there a bug in the reasoner or are temporal regions not what I thought them to be?), and ugly ones (is the resulting model one that we intended?) – might be used to improve the documentation of the BFO, to make axioms more actionable and informative while still logically equivalent and to provide a smoother path to axiomatization of related ontologies such as the Ontology of General Medical Science, the Information Artifact Ontology and the Ontology of Biomedical Investigations.
