Automatic Generation of BFO-Compliant Aristotelian Definitions in OWL Ontologies with GPT

Ontologies are representational artifacts that purport to accurately describe some aspect of reality, including the entities and the relations that hold between them. In computer science, ontologies are software artifacts containing the schematic structure for machine-readable knowledge, typically formed as a graph of subject-predicate-object triples, constrained through Description Logics. These resources and their relations are self-defining, i.e., some resource may be defined by considering all its stated relations. Resources are often attended with natural language annotations, that humans may read and interpret, such as labels and definitions. Many long-standing ontologies have useless lexical definitions that define resources cyclically, e.g., a FOAF: Person is simply defined as “A person”. In Aristotelian terms, the definition of a thing should be reducible, by using terms simpler than itself, such that every definition can be unpacked up to the most general thing, which can only be defined by stating examples and use cases. This paper presents an innovative technique that leverages the Generative Pre-trained Transformer (GPT) large language model, GPT -4, for automatically generating Aristotelian definition annotations for OWL classes that engenders compliance with the Basic Formal Ontology standard.