Tim,
Man-made domains have the benefit that the variables that affect these domains are usually reasonable transparent and discrete, whereas with evolutionary systems, there are typically variables that can be maddeningly difficult to qualify, let alone quantify, in part because phase transitions in such systems are seldom discrete. Health classification systems provide a good case in point - cancer is a systemic condition, there are many forms of cancer that present in different ways, and cancers of one type may induce cancers of another type, while there may be some cancers that would never occur if other types of cancer are present. Modeling this requires a significant number of time dependent variables, and the solutions are almost always non-linear as well.
Building an ontology then requires effectively identifying areas of consistent behavior within what is ultimately a piecewise discontinuous membrane. This is true for any system which incorporates non-linear phase transitions. The various phases of water, for instance, can be categorized only in those areas that are sufficiently far from a phase transition boundary (or that encompass just the phase transition boundary with recognition that this is a transitional state), as the physics at these boundaries gets non-linear. The characteristics of water at exactly 100 C are very different from those at 98 C or 102 C. I can be reasonably certain that the properties of liquid water will be consistent (i.e., continuous) over most of its range, I can be reasonably certain that the properties of steam will likewise be consistent. Boiling water at the transition threshold gets .. complicated, and modeling it becomes much harder because there are more external behaviors at work.
Roger,
Back to your original question - in a perfectly idealized world, the SME should be able to provide you with a model for replication within his domain of expertise, which may encompass two or more states that the model can be in, but because a model itself is simply a mechanism for imposing a working order upon reality in order to manipulate that reality in some way, different SMEs will perforce have different models. The welder may have particular insights into the characteristics of a metal at various phases - its plasticity, its brittleness, its heat dissipation envelope, but probably could not tell you the electrical characteristics of the materials he works with, because they are simply not relevant to his models. On the other hand, I'd be leery of giving most theoretical physicists a blow torch and ask them to weld together two pieces of metal.
This brings up a second point. The welder is not likely to retain the knowledge about the electrical characteristics of metals (this is a simplification, welders deal with highly volatile materials, so knowing which volatiles can be set off by a spark could be VERY important to them). The physicist, on the other hand, is probably not trying to learn about welding. He is only interested in the welder's model to the degree that it gives him insight into the model that he is building about the nature of metals at phase transition points. In effect he is receiving not a perfect rendition of the welder's model, but rather is transforming, via metaphor, the model that the welder has into his own model, conceptually connected in different ways.
I'm running into this issue now with Disney. The organization has multiple systems that evolved over time to facilitate very specific tasks, and as a consequence, the implicit models that these systems use are somewhat different. However, with digital convergence, increasingly each of these models is increasingly describing the same thing in different terms and dimensions. My primary goal is to transform these extant models into a more common one that nonetheless can support legacy systems (and models). Model transformation (call it a model tensor operation, as I think there's a lot of analogies between the two) plays a huge part in reconciling these systems. Capturing the time dependent nature of those models is another part of that puzzle - as Tim and Stephen both indicated, models evolve in response to business conditions (uncontrollable external stimulae).
It's one of the reasons I'm increasingly moving to semantic modeling, though I think even that's only a way point. Some assertions are only valid in the right context, which is principally but not exclusively a temporal function.
Kurt
