Re: Mapping a UML model to a DTD or Schema

[Robin Cover <robin@isogen.com>]

The ability to generate XML Schema documents and DTDs from UML
diagrams is interesting, and no doubt represents one step forward.
I've seen a number of software tools advertised which are said
to support this feature.

In my view, though, the important question to ask of the tool
which manages the UML (class) diagrams is: "What kind of
understanding do you have of the semantic relations being
represented visually?  What kind of integrity constraints can
you express with respect to class definitions and attribute
definitions such as to test the conceptual integrity of the
design?  How complete is the expression language you have for
modeling these constraints?  The last time(s) I looked,
OCL was simply not complete.

The EXPRESS schema language (referenced below) is one kind of
framework for modeling these constraints; there are graphical
tools as well.

The scenario I envision in terms of conceptual modeling would
be a (formal) language-based framework which supports the
ability to model/express semantic relations inside a testable,
executable system.  From this knowledge base, it should be
possible to generate a variety of (UML) diagrams as well as
commmon schema representations (XML Schema, DTD, RELAX, TREX,
Schematron) -- in whatever markup-based syntax you want,
using whatever facilities the schema language has for
modeling the semantic relations and other constraints.

Of course, these schema formalisms already are supported by
tools, and they have validation engines.  The key insight of
conceptual modeling is that it operates NOT at the level
of the implementation (syntax representation, serialization
format) but at the semantic level, using "transparent"
vocabulary drawn from the lexicon of the domain expert and
user, and object models which *directly* reflect the user's
conceptual model of the (abstracted) "real world" entities.

I recently extracted a portion of Alexander Borgida's
1983/1985 article [dated now in some respects] which
clarifies this perspective: how to model the problem
domain in way that matches (one-to-one) the conceptual
model of the problem as understood/experienced by the
user -- not by the database designer or markup language
guru.  And why this is an important consideration in
data modeling.  Borgida has since moved on to
description logics and requirements engineering (where
some of the '1985' principles are still alive), but I
feel that his insights, and related activity in the
area of conceptual modeling, are more relevant to the
(Web) markup-language enterprise than is commonly
acknowledged.

This excerpted paper from Borgida is referenced in
"Conceptual Modeling and Markup Languages"
<http://xml.coverpages.org/conceptualModeling.html>

"Features of Languages for the Development of Information
Systems at the Conceptual Level."

Robin Cover

---------------------------------------------------------------------------

On Fri, 26 Jan 2001, RENOUPREZ Jean-Luc wrote:

> At S.W.I.F.T., we are reviewing our approach at developing Message Standards;
> the new paradigm is based on modelling (using UML),
> and we transform class diagrams modelling the messages to DTDs (today)
> and to Schema's (to morrow).
> 
> Our conversion rules can be found on http://www.swift.com
> (go to the 'Products & Services' page. You'll find our document as a .pdf: swiftML).
> 
> I'll be very interested to get some feedback on our approach.
> 
> 
> "Vegt, Jan" wrote:
> 
> > On Thursday, January 25, 2001 4:32 PM Arnold, Curt wrote :
> > >The Cover pages have a fairly lengthy list of STEP and XML initiatives
> > >http://xml.coverpages.org/related.html#step
> >
> > Arnold, thanks. In my understanding XML and STEP as system neutral data
> > wrappers are
> >  roughly similar mechanisms. I am just wondering if the concept of a
> > modeling level is
> > feasible/practical for a *generalized* markup language.
> > In ebXML and BizTalk you see that modeling levels appear.
> >
> > Also with richer data type support soon available in XML via XSchema, I
> > wonder if
> > mechanisms like EXPRESS-X are actually avoidable (to guarantuee against
> > critical
> > data loss). Any thoughts on this? Len?
> >
> > >The approach I favor is to generate an UML model from the EXPRESS schema
> > (SELECT
> > >maps to Interface for example), refine and iterate on the UML model and
> > then
> > >produce XML Schema from the UML model.
> >
> > Very interesting.
> >
> > >The aecXML (Architectural, Engineering and Construction XML,
> > http://www.aecXML.org)
> > > initiative is currently exploring this approach with mapping the
> > International
> > >Alliance for Interoperability's
> > >Industrial Foundation Classes schema to an XML representation.
> >
> > Thanks, I'll check that out.
> >
> > Jan
>