Everyone Doesn t Have to Agree to the Same ber-Ontology

Everyone Doesn't Have to Agree to the Same " ber-Ontology"

One of the things that dissuades some people is the belief that for this to work, we all have to subscribe to the same ontology. Anyone who has tried to describe the operation of their entire company on a single database (schema) knows by extension how impossible this is. And yet people want to try.

The "Upper Ontology"

Some groups are working on what is called the "upper ontology." Sometimes this looks like a small veneer to describe the rest of the ontologies, and if that is the case, full speed ahead. But just as often the ontologies become bigger when they should be getting smaller and broader when they should be getting narrower.

Some Very Big Ontologies

Cyc, from Cycorp, is the granddaddy of the big ontologies.^[118] To be fair, they haven't set out to create an upper ontology; they set out to build a commonsense database that would allow a computer to reason in a way that is analogous to the way humans reason, in the presence of unstructured data (natural language). This is an ambitious goal. Maybe they've achieved it. Maybe they're close. The problem is that they have seduced many ontologists into using this as a starting point, and it is just too unwieldy to be used in practice.

WordNet is similarly a large ontology.^[119] It is a dictionary of the English language with which systems can do some limited reasoning. It is useful for dealing with English generically, but it isn't set up to be the source authority on anything. For example, WordNet has information on eight types of mushrooms. Other sites have information on thousands, with far more data.

The issue is not that there is anything wrong with large ontologies. They are useful for some things. But we shouldn't let their success in breadth allow us to believe that with a bit more depth they could handle our semantic integration needs. Just the opposite: As they become larger, and especially as they have more constituents, their size will prevent them from being able to specialize. What we need are a manageable number of moderately sized ontologies.

Six Degrees of Separation

If every business and every person had their own ontology (the idiolect issue again), communication would be chaos. But we also don't think it wise to try to force everyone to subscribe to a common ontology. How are we going to communicate? How will we do commerce?

I think we'll do it in a manner similar to how we do it now. Each specialty and subspecialty will rally around some sort of standard bearer, some institution they believe has best captured their collective knowledge. Urologists will reference a few urology ontologies. Brake pad manufacturers will reference industrial materials ontologies and manufacturing methods ontologies. Each of these in turn will reference more general ontologies in which they express their schema. The urology ontology will be expressed in medical terms (not legal terms, despite current trends) from a more general (not a more comprehensive) medical ontology. The medical ontology in turn will be expressed in a few even more general ontologies, perhaps general anatomy, general etiology, and general activity. All the ontologies will be interlinked, but not all to the same set of ontologies. My guess is that as this system matures, the distance between any two ontologies will be on the order of six degrees.

How Many Types of Entities Are There in the World?

If you observe a typical corporation and look at its production databases, you will find, for a company of any size, at least a thousand database tables and tens of thousands of elements (attributes, properties, and relationships, as well as the table entities). From personal observation there seems to be about 10 for every employee in a company. (A company of 1,000 employees could be expected to have 10,000 elements; a company of 100,000 would have a million. These aren't scientific numbers, but I don't think they are too far off.) With a hundred million workers in the United States, we should expect to have about a billion elements floating around our corporate infospace. (Of course many of them are the same, but we have no way of knowing this.)

How may elements are there likely to be in all of commerce? From our discussion on vocabularies, I could easily imagine hundreds of thousands. Maybe even millions. But keep in mind that every distinction we make is not necessarily a new element. Blue pens and red pens are different SKUs, but I don't set up new columns in my database. Where it becomes interesting (and where the question of cross-domain interaction becomes interesting) is, What is the smallest ontology you need to be able to communicate with your trading partners?

A Core Ontology Would Be Helpful

Rather than trying to build an ontology that tries to include everything, we should be looking at ontologies that allow us to describe anything. I don't think this would have to be very large. Elegant, yes. Large, no. I think a few hundred concepts would do it.

^[118]See http://www.cyc.com for further information.

^[119]The WordNet database is available from many sites, including http://www.cogsci.princeton.edu/~wn/.