Re: [xtm-wg] TM Conceptual Model: Semantics of the UML "modern dinosaur"?

[Robert Barta <rho@bigpond.net.au>]

Sam Hunting wrote:
> 
> I noticed that there were no responses to Alexander's posting, so I
> decided to repost it, on the theory that some of the ideas expressed
> there-in might be useful to the group's deliberations in Paris.

> --- Alexander Sigel <sigel@bonn.iz-soz.de> wrote:
.....
> > what makes me worry is the lack of semantics in UML which might
> > hold us from fully understanding and clearly expressing the semantics
> > of the TM conceptual model.

Sam,

Well, UML certainly helps to understand relationships between items
under specification but - I agree - UML itself is notoriously
"meaningless".

> > to construct a higher-level internal representation). it appears
> > to me that several basic TM processing functions can be built e.g.
> > with XPathScript (part of AxKit). (but code will become complex).

Seeing our own experiences, I'm quite pessimistic here that
XPathScript/XSLT/..
can produce useful output from a typical TM. And even if it would,
the processing costs are prohibitively high. Holding several maps
as DOM objects could become a major nightmare.

It becomes even worse if we allow maps to be combined/merged/filter
with a TM-query language. In this case the representation might differ
heavily: XML documents, legacy database, and "virtual maps" defined
via expressions like

  ( map_a [ joined on map_j with ] map_b | scoped by map_s )

...
> > tied. in OO vs. RDMBS speak: with the right schema you can roll your
> > own
> > object wrapper from relational stuff). however, performance may
> > differ.

And it makes the hell of a difference whenever you're hit by 100
reqs/sec.

> > for me, the essence of this discussion is:
> > - what is the appropriate level of conceptual model?

Excellent question. To rephrase it: "Can we build a topic map algebra?",
i.e. "what are acceptable models (sets, monad, category theory,
whatever) and 
what operators can we define on it?"

Still, I do not think (and expect) that the upcoming meeting will be
able
to address it. Maybe it is better to get things on the road (involved
companies
are already announcing their products) and to gain some employment
experiences.
(And i'm not overly keen to rewrite my parser every two weeks ;-)

I'm rather convinced that most of the involved participants are aware
that for 
a large scale rollout a centralized/proprietary solution won't do. And
'going
distributed' certainly means that we have to check interoperability and
this is
only possible with (a) a reference implementation and/or (b) a clear
(well, more
or less) formal model.

So, whatever the outcome will, still some way to go.

\rho

To Post a message, send it to:   xtm-wg@eGroups.com

To Unsubscribe, send a blank message to: xtm-wg-unsubscribe@eGroups.com