Re: [xml-dev] XML Schema as a data modeling tool

[Michael Kay <mike@saxonica.com>]

Data modelling for me is all about achieving a shared understanding of =
the problem. It's something I usually do interactively on a whiteboard. =
The resulting model needs to fit on the whiteboard and it needs to be =
thrown away when you've finished - if it wasn't simple enough to be =
memorised, then you have failed.

For example I once did a workshop with people from a broadcaster. It was =
all about programme scheduling. It was primarily about getting to a =
point where everyone in the room had a common understanding of what they =
meant by words like "programme", "advertisement", "channel", etc -- =
which all led to some very interesting debates; and of course you =
sometimes agree to split the concept, so =
what-engineers-mean-by-a-channel becomes a different concept from =
what-marketeers-mean-by-a-channel.

The whiteboard ends up with lots of boxes and arrows because ultimately =
the way to define a concept is to show how it relates to other concepts: =
an-ad-is-a-kind-of-programme, a-program-has-many-transmissions, =
a-transmission-happens-on-a-channel.

I simply can't imagine doing this kind of exercise and restricting the =
stuff on the whiteboard to be a hierarchy. It would be like restricting =
yourself to only use one colour of felt pen: why would anyone want to do =
that?

Yes, there's a sense in which the term "real world" is unfortunate; the =
model is always synthetic, and abstraction is vital. But I find the term =
useful, because it reminds people that we are trying to improve our =
understanding of the way things "out there" behave, not to design data =
structures that are convenient to implement.

Michael Kay
Saxonica