Pages

Thursday, October 8, 2015

Conversation Starter: How many Kinds of Ontology are There?

So here's something that's been bugging me for a while:

Is there more than one kind of a thing of which I can say of it "This is an ontology"?

Simplistically one would have to say Yes. Some people would define an ontology as something which represents things in the world. Someone else would say an ontology has to also be framed in formal logic, or even formal first order logic. Someone else would way it has to be all these things and you also have to be able to reason over it otherwise it is "not an ontology" or it "has no semantics".

What if they are all right, but talking about different things? Is it necessarily the case that ontologies need to be different for different kinds of requirement - for example to create a common language across an industry, or to create a reasoning-based application for car rentals or something?

Personally I think these are all useful things to have. But some of the design techniques that would be appropriate for a reasoning based application, would not be appropriate for a common industry language. For example, in one kind of ontology you might have properties defined with no domain and range and only make them "mean" something when they are referred to in a restriction on a class. This seems to me perfectly reasonable, but from a concept model point of view it is those properties which have "no semantics" - the overall graph becomes meaningful only when you close the world and throw the thing at a reasoner.

If this is the case, what are the implications for ontology re-use? Before I can re-use an ontology from somewhere else, I would need to know whether it was intended to support a stand-alone application or be part of an industry language. I would want to know what theory of meaning was brought to bear in its creation - was meaning considered to arise within the graph as a whole based on the internal structure of that graph, or were some concepts regarded as axiomatic to the others? Is there some grounding of basic, primitive concepts which others are then defined in relation to, or was that not considered necessary for that particular ontology?

No doubt it is possible to create ontologies which fit into multiple views of what it is for and how meaning arises - but can I rely on that?

Assuming that there is no "wrong" definition for ontology, what do others think?

Mike