What is "an Ontology"

If you want to really irritate me, refer to a plain taxonomic categorization as "an ontology". It's like calling a case of canned alphabet soup "a literature".

This is part of a general social trend that appropriates the most obvious outward manifestation of something they don't understand and devalues the original concept.

Firstly, ontology is first and foremost a field of philosophy that studies the types of objects and the relations among them. It is called "ontology", not "an ontology". Secondly, Computer Science, and particularly Artificial Intelligence, uses the term to refer to a formal specification of a domain that can be used for a task, typically involving inference. A hierarchical data structure containing classes arranged with subsumption relations (is-a) can be part of it, but the hierarchy alone is useless. In fact, a single hierarchy is a pretty lame representation of most domains.

All sorts of knowledge domains are not hierarchical at all. They can sometimes be forced into a hierarchy but the hierarchy can not be presumed to be useful, and if it is useful for one purpose it may not be useful for another. For instance, say you have 4 groups of people: women lawyers, men lawyers, homeless women, and homeless men. What is the best hierarchy - men and women, then split up by jobs, or lawyers and homeless people then split up by sex? That depends whether the purpose of the categorization is to design washrooms or social services. And who is to say that a lawyer can't be homeless? We are simply projecting our own stereotypes - it is a very powerful cognitive process that attempts to simplify the world according to simple rules. The desire to categorize hierarchically says more about us than it does about the objects of categorization.

In the example above, the "correct" hierarchy is task-specific. When we try to solve problems, it is useful to group things that behave similarly. When it comes to public washrooms, a number of rules apply to all women and not to men, and vice-versa. An ontology will contain both the rule set and the hierarchical categorization. Later, the process of subsumption will be used to apply these rules to subclasses.

But without the rules, a taxonomy is perfectly arbitrary and not particularly useful, even misleading in what it implies. With the rules and other components of an ontology, the hierarchy is probably an arrangement of "is-a" relationships. There can be other types of relationships as well, and they are labelled as such. Without rules, it is a classification made of containers within containers. These are more "part-of" relationships. As such, this type of taxonomy is more accurately called "a mereology" (to misuse a different word).

Where are these real ontologies? Most of them have been developed for relatively small domains. But for the ambitious universal ontologies, there is the Big One - Cyc. It has been decades in the making, and attempts to formalize common sense so that computers can use it. The other one is SUMO, to which I have made minor contributions, which is a top-level ontology that is more based on language than on abstractions.

These ontologies are formal knowledge bases containing both information and knowledge. And don't get me started about people who call a database or document collection a "knowledge base"...

References
http://shirky.com/writings/ontology_overrated.html
http://www.shirky.com/writings/semantic_syllogism.html
http://www.cyc.com/cyc/technology/whatiscyc
http://www.opencyc.org/
http://suo.ieee.org/SUO/SUMO/

[] permanent link Comments: 2