[om-list] Problems with Concept Relation Form
Mark Butler
butlerm at middle.net
Tue Oct 21 14:25:32 EDT 2003
Hello everybody,
I have been attempting lately to cast my model completely in concept
relation form, where there are no explicit concept attributes, just
concepts and relations between concepts, maintaining the rule of only
one true relation between any two concepts. This normalization has
some very attractive properties, but I have found it difficult to use.
The main problem is that natural language is full of attributes, which
are always interpreted as noun expressions standing for the target
concept, not the relationship itself - so I commonly deal with three
intermediary concepts relating to two concepts and a relationship
between them - two attributes and one relation. For example, in an
employee / employer relationship we have the attribute
employee.employer, the set attribute employer.employees, and the
relationship between an employee and an employer.
Those three things have very different senses that make it difficult to
eliminate the first two and deal only with the latter. For example, let
us say I want to express the idea that an employer has three employees -
I could make an evaluation of the employer->(employee class)
relationship and and say it has cardinality three, but how do I clealy
specify that I mean the employer has three employees rather than
Employee has three employers. That is where attribute concepts come in
handy - if I evaluate the employer.employee attribute and say it has
cardinality three, the ambiguity goes away. Anytime I talk about a
relationship on the other hand, the polarity of the relationship and the
way each statement refers to it is criticial.
Time and space predicates present a similar issue - If I want to express
something about the state of an object as of a particular time, I am
really creating a new concept, an object state, and expressing a
proposition about that rather than the object itself. The problem is
that like an attribute, an object state does not map 1:1 to the
relationship between an object and a specific time. That particular
(O,t) relationship could have other aspects which make it more than just
an object state. However, in order to refer to the object state, I need
a concept that stands for the object state and nothing else. This can
be done in pure CRF, but the number of objects required to represent a
simple statement quickly gets out of hand.
My preliminary conclusion is that using nominative concepts like
attributes and states will be more efficient than representing
everything in terms of pure class/object concepts and relations between
them, initial appearance to the contrary.
- Mark
More information about the om-list
mailing list