Re: importing and entialment

Jonathan Borden wrote:
> 
> Jeff Heflin wrote:
> >
> > ... However, I think that it
> > is still desirable to say that if the import "fails," any inferences are
> > inherently incomplete. Whether it was because the author mistyped the
> > URL or because the document being referenced was moved by its owner,
> > there is something that the author of the importing document wants
> > referenced, but the deductive system is unable to take it into
> > consideration. I believe saying the inferences are incomplete is much
> > more graceful than having a defined error condition in such cases
> > (although specific systems are free to raise warnings or errors in such
> > cases). However, I believe an error condition is still better than just
> > saying "well let's assume the imports statement refers to an empty
> > document."
> >
> 
> The reason monotonicity is emphasized for Web applications is that we need
> to make such assumptions.
>
> Perhaps when an import is successful, we ought add a triple to the KB that
> says
> 
> <http://example.org/ex.owl> owl:importStatus true .
> 
> or something to that effect?
>
> I _wouldn't_ flag failures, as they might succeed at a later time (and so
> the absense of success is a better flag than explicit failure).
> 
> Jonathan

I'd rather not tell developers exactly how they must treat an import
failure. The beauty of the "A imports B means anything entailed by B is
also entailed by A" semantics of imports is that developers can decide
for themselves how to handle an imports failure, whether it be by error
condition (as Jeremy seems to desire) or by special triples on success
(as you suggest). The only requirement the proposed semantics enforce is
that if the imports fails, a system cannot claim to have deduced the
complete set of inferences from the document.

As a sidebar, for those of you who are not so familiar with the
terminology of logic, here's what is meant by "complete":

ENTAILMENT is a theoretical notion that describes what can be deduced
from a set of statements using the semantics of a specific logic.
Entailment can be used to verify the correctness of inference engines /
deductive systems. If the inference engine can deduce all of the
ENTAILED statements, then it is said to be COMPLETE. If it does not
deduce any statements that are not entailed, then it is said to be
SOUND.

Received on Wednesday, 23 October 2002 09:47:16 UTC