Re: n-ary relations, triples, query languages

Below are some comments on
http://smi-web.stanford.edu/people/noy/nAryRelations/n-aryRelations-2nd-WD.html

First, here's a more general question:
Why are triples preferred to n-ary relations as a basis for RDF
(or anything else)?
Note that in mathematical logic, for instance, n-ary relations seem to
be preferred.  Also in relational databases.  The objective ought to
be to make it straight forward to represent information.  Triples seem
to just require an extra translation step, which introduces
unnecessary complications (as discussed below).

- The idea that one role seems clearly the subject seems to me to
be a hallucination.  I think the distinction in RDF between subject
and object is already problematic.  To me even the most "obvious"
cases like a person related to a birthday seem equally reasonable
as a birthday related to a person.
The idea that Christine should be a subject related to an invented
object (rather than the invented object being the subject related to
ALL of the n arguments) raises the problem that the same semantic
relation could have been represented with ANY of the arguments
distinguished as the subject.  Or even worse, as any TREE involving
the n arguments and any number of other invented objects.
At very least you should minimize the number of different
representation choices with an unambiguous standard.

- temperature high and falling - feels to me like two independent
statements in the sense of database normalization.
The fact that you need to talk about cases in which both are true
doesn't seem like a reason to confuse this issue.  It seems like
a reason for another language in which you can talk about two
conditions both being true.  This language can be used for queries
and definitions of other concepts.  In logic that would be something
like the language of First Order Logic, in databases it seems to be
SQL.

- issue 1 (probability of a statement) feels to me exactly like the
case you describe as the original RDF usage for "reification",
the probability is related to the statement about a diagnosis

- even for issue 2 the difference between creating an object to
represent a statement seems not very different from creating an
object to represent a "relationship"

Received on Monday, 5 December 2005 20:37:28 UTC