Asha:
May
I point out the essence of my message in a more specific way: relations
are fundamental to any robust knowledge modeling system, since what it
knows about the class of relations and how it models this elusive kind of
entity are the crucial indicants of its quality and commercial usage. It must
be noted that the extant literature on the ontology of relations is poor
comparing with the formal calculi (algebra and logic) of relations, mostly due
to the pioneering works of De Morgan and Peirce; as far as we know, there is
no systematic classification of the fundamental species of relations.
Despite of this, the reason to have a fundamental theory of relations is
perfectly clear; not having developed a general ontology of relations may
result in a variety of schemes mostly based on practical purposes or
everyday intuition, as systems like RDF/OWL and UML avoiding, or rather
unable, to define and classify relations, considering them as
the class of things having formal existence.
Danny:
I don't disagree that relations are fundamental. But
I don't know why you assume that a highly sophisticated, general model is
essential for useful (including commercial) applications. Surely a more
immediate requirement in those contexts are "schemes mostly based on practical
purposes". I agree that the modelling of relations in RDF/OWL and UML is
limited, where I disagree is that this means that this precludes the use of
those languages in practical applications. In fact, I'd suggest that the
evidence (the growing use of RDF/OWL in software applications, the use of UML
notably in OO development systems) points the opposite way. As well as the
'quality' dimension there are also the dimensions of 'usability' and
'interoperability' to take into account. By 'usability' I mean how
straightforward it is to apply the model/language to a given task; by
'interoperability' I mean the possibility of communication and integration
with existing systems. Though both UML and RDF may be lacking in 'quality',
I'd say they both score pretty highly on these other
two measures.
Asha:
Now, carefully researching the matter, we found out
that the universal class of
relations consists of sixteen classes of ordered classes of the
world variables (objects O, states S,
changes C, and relations R) arrayed in
a matrix structure:
|
O :
O |
O :
S |
O :
C |
O :
R |
|
S :
O |
S :
S |
S :
C |
S :
R |
|
C :
O |
C :
S |
C :
C |
C :
R |
|
R :
O |
R :
S |
R :
C |
R :
R |
Danny:
I'm
not altogether sure what you mean here, but it looks like at a meta level you
are describing four classes of entity in terms of 16 different *simple*
relations.
Asha:
Depending on the nature of correlatives, the general
relational formula allows for a complete extent of possible relations, so
richly listed, but not properly ordered, in the WordNet' lexical
database:
[lots of kind of
relations]
As one
can see, the mathematical (logical) relations between sets (classes)
so actively used by OWL make only small part of an enormous universe of
relations, real, ideal, and formal or external and internal. Here emerges
the conclusion that 'proposing a general conceptual scheme by using the class
/property distinction, the data type languages like OWL are missing the
meat of things', accordinly, using such a language is harmful with
badly constructed Web-based applications.
Danny:
There
certainly are a lot of different kinds of relations. But I think you are
making two mistakes in your conclusion - firstly that RDF/OWL should
provide a truly general conceptual scheme - I can't speak for the developers
of these systems, but I'd say that all that is needed is something that
provides a descriptive capability allowing practical applications to be built.
Something good enough to satisfy a high proportion of current
demands. The web environment brings with it constraints on what is feasible in
technical as well as social terms. The other mistake is in suggesting that
using such languages is in some way harmful. The existing web only has the
REST architecture and countless controlled-vocabulary schemes in
many places where good quality ontology languages could be applied.
Is that harmful too? RDF and OWL offer a small improvement in
sophistication locally, a significant improvement for the network as a whole.
Or should the web be switched off until we get the model
right?
Cheers,
Danny.