A line that stood out for me at yesterday's Tim Berners-Lee talk was "Start off with islands and stitch them together." He said this in answer to a question regarding ontology standardization. Later as I talked to people about it, I was surprised at how controversial an opinion it was.

Many thought (or knew people who thought) that modeling efforts should be less haphazard, that we should we aim at a kind of "imperial ontology" and instill a sense of responsible conformity in those around us. One guy even said, when confronted with some "fringe attributes" that people may consider, that the attributes were awful and should never be used. "I'm a fan of the fourth normal form." (which made me think, possible T-Shirt?)

I agree strongly with Tim. We live in a many-to-many world, with a dizzying degree of semantic commotion. At work, I often find myself with clients at a conference table, trying valiantly to model their domain with enough specificity to do something useful. We often spend so much time debating the boundaries of terms like "project" and "task", that I usually give up and start calling things "bananas" and "doo-dads".

I think the "imperialists" (ie, those that want standard ontologies) are falling into the same classic trap that has plagued the software development community for years. They're advocating a waterfall lifecycle, where the design preceeds implementation. Most in my field now believe waterfall is a disastrous and untenable approach. I personally advocate an interative feature-based approach, where we do something, take a look, then refactor things to a more extensible state. Refactoring is key.

But how do you refactor the semantic web? With URIs winging their way around the world, that's not much of an option. Hence the "stitch them together" part, which is the semweb's equivalent to refactoring.

Will this make a big muddled mess of equivalence relations? Will this impact performance? YES. Get used to it.

After all, we live in a real-life big fractal tangle, just look out your hotel window at the NYC streets below. Our models will have to be just as complex.