[topicmapmail] RE: back to the lists

[Martin Bryan]

Bernard wrote:

>The example you give is maybe borderline one ...

>Relationship A must have an X in role R and a Y in role Q and that no
two A relationships with X can have the same Y."

> would gladly take the challenge to try and express it in OWL (I don't
see why it could not) but please make it more accurate in its
expression: Does A represent an individual association? Or an
association type? Do X and Y represent an individual topic? Or a class?
Does "an X" means "exactly 1 X"? Or "at least 1 X"? etc ...

A has to be an association type (this is, after all, a constraint, not a
query)
X and Y would be best as a class, but equally I can see a case for them
being an individual topic. I can also see cases where both at least one and
exactly one would apply, though I was thinking of the exactly one case.

>In any case, and even supposing effectively that OWL cannot express that
kind of specific constraint, I think dismissing a language simply
because it cannot handle such and such "ad hoc" tricky situation is not
a correct approach.

If the constraint language cannot properly constrain the members of an
association then it has failed :-(

>I would prefer to explore what OWL *can* express,
starting from basic needs that are now expressed in real use cases by
implementers an their customers, and see if more is needed that can't be
expressed.

This definitely needs to be explored. I simply pointed out that there was a
common user need that I did not think OWL was capable of covering.

>Seems to me that sophisticated constraints will be anyway relevant only
in specific environments and be tackeld with specific implementations.

I did not think I was asking for a sophisticate constraint, only the sort of
thing I would want any template to be able to constrain.

>Another strong argument for using OWL is that it seems a good wrapping
of long reflexion in ontologies, and that it is likely to be adopted by
industries already using ontologies.

This last point is the really strong case for OWL, if you can manage it.

>Showing that somehow a topic map schema is nothing but a kind of
ontology -  and I remain to be convinced it's something else - including
more flexibility than the classical class-instance-property paradigm,
and can be expressed using the same language than any other ontology, is
more likely to hit the market, than an ad-hoc sophisticated language
developed only for topic maps, to address all weird TM constructions you
can imagine - and that is likely to be endless.

I actually see it the other way round. OWL has some known limitations. To
overcome those limitations you can use topic maps to introduce new kinds of
relationships between entries within an ontology and, much more importantly,
between ontologies. This will be particularly significant when trying to
link together a) ontologies at different levels of detail (upper-level v
technical) and b) ontologies in different languages.

>Expressivity vs usability tradeoff, as usual. Don't forget Biezunski's
Laws:

>1. You don't gain anything by showing that you are more intelligent than
your interlocutor.
2. There is no point in making a good standard if no one uses it.

But what about Bryan's corrolaries:

A. You don't gain anything by showing that you can do what can be done as
easily some other way
B. There's no point in developing a system which no other system can use the
data from.

Martin