TopQuadrant presented another of our Semantic Web trainings last week. As usual, the discussion with the students taught me more than I felt I was teaching them.
Many of the students in this week's class were well-versed in MDA and MOF, so the were no strangers to modeling. At one point, one of them showed me a diagram of a small model he had drawn; he had two relationships in it, one called "hasAccommodation" (taken from the travel agency example domain we were using), and the other was "hasPart". He had an intuition that these properties were somehow different, and didn't we need to consider different "kinds" of relations.
This discussion reminded me of what I like about modeling in OWL, as opposed to other modeling languages I have used in the past. In OWL, the semantics of the model is given formally by a set of axioms for an inference engine to follow. The meaning of any construct is given by the model-theoretic semantics of the language. This allowed me to answer his question with a precise retort. "Figure out what it is that is different about this 'kind' of property," I said, "then we will try to model that in OWL. If we succeed, then we have expressed your 'kinds' in OWL. If we fail, then we have a concrete spec for how we might want to extend the language. But usually, OWL is strong enough to cover the sorts of things you want from your new kind of property."
In many modeling languages, the way one constructs the modeling framework is to have discussions like this, in which one argues for different "kinds" of properties, or different "sorts" of concepts, etc. But in the absense of a formal semantics for the modeling language, the framework is extended only by discussion and rhetoric; if you argue well enough for a new kind of modeling construct, then it makes it into the modeling system. I have seen whole dissertations written in this style. I liken this to (my goyim understanding of) the Talmud. There is some scripture, but real understanding of it doesn't come just from studying it, but from layers and layers of commentary. The commentary continues over generations, until it is as important or even more so than the original. This is the sort of process that has led to the STEP standard having hundreds of pages in several language commenting, extending and describing the standard.
This process is great for building cultural identity and exploring the multifaceted aspects of human spiritual nature, but it is unwieldy for managing a modeling standards process. For all its warts, OWL has the virtue of a very simple description of its semantics, and one that is based on generations of mathematical rigor.