"Hidden semantics": what can we learn from the names in an ontology?

Despite their flat, semantics-free structure, ontology identifiers are often given names or labels corresponding to natural language words or phrases which are very dense with information as to their intended referents. We argue that by taking advantage of this information density, NLG systems applied to ontologies can guide the choice and construction of sentences to express useful ontological information, solely through the verbalisations of identifier names, and that by doing so, they can replace the extremely fussy and repetitive texts produced by ontology verbalisers with shorter and simpler texts which are clearer and easier for human readers to understand. We specify which axioms in an ontology are “defining axioms” for linguistically-complex identifiers and analyse a large corpus of OWL ontologies to identify common patterns among all defining axioms. By generating texts from ontologies, and selectively including or omitting these defining axioms, we show by surveys that human readers are typically capable of inferring information implicitly encoded in identifier phrases, and that texts which do not make such “obvious” information explicit are preferred by readers and yet communicate the same information as the longer texts in which such information is spelled out explicitly.