RE: Properties of Properties Question

From: Peter F. Patel-Schneider [mailto:pfps@research.bell-labs.com]
> I'm still a bit confused.
> 
> There are a number of things that you could be asking for:
> 
> 1/ The ability to associate information with assertions.  (In 
> RDF terms
>    this would probably be using a statement as the subject of another
>    statement, which can't be done in RDF.)  What you would be 
> trying to do
>    below would be to restrict the kinds of information that can be
>    associated with a particular assertion.  This can't be 
> done in DAML+OIL
>    because assertions cannot have associated information.

I can see a role for things in category 1/.  A simple example, common in
many knowledge bases, would be to say "this assertion is 50% likely to be
true", or "this assertion holds under the following preconditions".
Separate from the logical properties of the assertion**, it might be nice
just to be able to record the provenance of an assertion for auditing or
explanation purposes.  I take it from the analysis above that the only way
to record such information in a DAML+OIL model would be to add an extra
intervening node, from which hangs the properties of the association, so:

<x> abc:someProp <y> .

becomes:

<x> abc:someProp [<> abc:prob "0.5" ;
                     abc:provenence <...whatever...> ;
                     abc:hasValue <y>] .

Actually I'm not sure that the [..] context brackets are necessary, but I
can't remember whether N3 has parentheses to sort out operator precedence.

Is there another way of handling such a requirement?

Cheers,
Ian


** i.e. to side-step the comment "RDF/DAML+OIL doesn't support approximate
reasoning, so tough".

Received on Tuesday, 23 October 2001 12:51:08 UTC