textual metanoia

Although XML and XSL isn’t sexy stuff, ontologies are. Chapter 8 of Building Enterprise Taxonomies (Stewart) provides a definition of ontologies, explains the conceptual ideas behind them, and offers a description of two implementations (Resource Description Framework and OWL). Stewart gives the canonical, if somewhat unhelpful, definition of an ontology: “an explicit and formal specification of a shared conceptualization”. Stewart notes the unhelpfulness of the definition, and further breaks out the definition into its core pieces, namely, that ontologies are explicit, formal, and provide a conceptualization of a domain.

An ontology is formal, Stewart argues, because the ontology “defines and details all of the concepts, properties and constraints that comprise a particular domain”. While I think any ontology that attempts to define all the properties of a domain is sure to fail (on theoretical grounds if not practical ones), the essence of the idea strikes me as correct---an ontology attempts to be usefully exhaustive. An ontology is formal because “it must be documented in a form that is both machine readable and interpretable”. Ontologies are, by necessity then, artifacts of computer processing (perhaps interpretable to non-electronic machines, however). An ontology provides a conceptualization because it “organizes… [explicitly and formally declared properties] into an abstract model that shows how they relate to one another”. That an ontology is a conceptualization of a domain is the key to understanding their utility, but also muddies the water between the differences of a thesaurus and an ontology. Thesauri are supposed to formally and explicitly declare the relationships within a domain of knowledge, which sounds a lot like an ontology. Stewart argues that ontologies take this conceptualization a step further, he argues that ontologies codify relationships into formal classes and subclasses which can be translated into logical operations. Of course, a faceted thesaurus is a form of conceptual classes and subclasses, all which can be codified into various classification schemes. As a matter of history, thesauri haven’t tended to be machine readable in the robust way that these-things-called-ontologies have been, but that’s because “ontology” is a philosophical loan word that some computer scientists picked up when they rediscovered what the librarians had been up to since the turn of the 20^th century. Whatever the hair splitting about distinctions and providence, ontologies are those things that explicitly and formally declare a conceptualization of a domain that is easily machine-readable.

Throughout the remainder of the chapter Stewart provides one of the best descriptions of the RDF that I have ever read. Stewart explains the essential and complicated way that RDF triples uniquely identify objects and their relationships and gives helpful diagrams and examples of potentially real-world applications of RDF. Like all of this book, reading this chapter on RDF will by no means provide you with the necessary tools to actually use or develop RDF ontologies, but it will provide you with enough background to potentially make purchasing decisions or lead strategy. Stewart only provides the roughest of overviews of the Web Ontology Language (OWL), but his detailed look at the related RDF technology leaves the reader feeling satisfied about current implementations of ontologies.