W3C home > Mailing lists > Public > public-swbp-wg@w3.org > December 2004

Re: [OEP] Closed vs Open World $swbpd

From: Pat Hayes <phayes@ihmc.us>
Date: Sat, 18 Dec 2004 21:06:10 -0800
Message-Id: <p06001f11bdeab7b357b7@[192.168.1.7]>
To: "Uschold, Michael F" <michael.f.uschold@boeing.com>
Cc: public-swbp-wg@w3.org

>We just encountered the time-honored 'problem' of what happens when a
>user assumes/expects closed-world inference, and gets open world.
>
>Unless two classes A and B are provably disjoint (possibly via an
>explicit disjoint axiom), a DL reasoner will not conclude that A is in
>the complement of B.

Er..yes, of course. Not just a DL reasoner: any 
correct logical reasoner. What you just said, in 
effect, is that unless P is provable, the 
reasoner will not prove P. That is pretty much a 
definition of 'reasoner', seems to me.

>This can be seen as a feature, because it allows you distinguish between
>"can't prove it" and "can prove that it is false".

No, its a feature because if you could conclude 
that A was disjoint from B for any A and B, then 
all kinds of nonsense would be provable. It would 
follow for example that you were not an employee 
of Boeing (A=human being, B=employee of Boeing, 
start with the assumption that you are human) or 
possibly that you were not human (Same classes, 
but assuming that you are an employee).

I don't think the problem as you have posed it 
has anything to do with provability or 
closed-world reasoning. Closed-world reasoners 
cannot conclude that sets are disjoint, only that 
particular items are not in sets. Im guessing 
that the real source of the confusion is not to 
do with provability, but a mistaken idea that the 
classes in OWL form a disjoint taxonomy or 
classification tree, where the branches never 
overlap. But they don't: they are just sets, and 
something can be in many of them at once. This 
might be worth pointing out, as this 
misunderstanding seems to be quite common in 
people who have grown up with OO programming and 
think that "class" means a Java class.

>It could also be experienced as a problem (read: big pain in the neck)
>by a user who does not need to make that distinction,

What kind of user does not want to make the 
distinction between an arbitrary set and the 
complement of another arbitrary set?

>  because it forces
>them to add bunches of disjoint axioms.

Well, to represent a taxonomy you need one 
disjointness axiom for each branching. So its a 
pain, but only a controllable pain.

>This is a likely source of confusion for some users.

Seems to me that anyone who is this confused is likely to be a hopeless case.

>It would be good if we could say something about this. Questions that
>users may wish to be discussed/answered include:
>
>*	are there any identifiable characteristics of a domain, which
>suggest when you want an open vs. closed world reasoner?
>*	does the Semantic Web infrastructure offer any closed world
>reasoners? 
>*	If not, then what do we say to a user who does not care to
>distinguish not provable from provably not and finds it a nuisance to
>add all those disjoint axioms? They are unlikely view as helpful, a
>comment such as: "You should be glad to be forced to model your domain
>more carefully".

They might not view it as helpful, but it might 
be true nevertheless. Remind them that they are 
publishing information on an open web, and that 
it will be used by agents who have no reason to 
suppose that their domains are closed. Remind 
them that whether or not they care about the 
difference between provably not and not provable, 
other reasoners might very well care a lot about 
that distinction. Suggest to them that in fact, 
they do care about the distinction, when they are 
using knowledge sources without any way to tell 
if they are based on an open- or closed-world 
assumption. Remind them that those assumptions 
are not expressible in OWL/RDF, and there are no 
Web protocols available for negotiating them or 
communicating them

>
>I think that we should try and say something about this somewhere, it is
>bound to come up over and over.
>Is there a note that this topic might be covered in?
>Should we have separate, short not on it?
>Does anyone have enough expertise/experience to say something sensible
>on the topic?

Its a very hot topic in the rules community and 
is being debated intensely right now. Check out 
the email archives of RuleML, for example.

Pat


>
>Alan Rector may know more about this than anyone, with all the users
>encoding bio ontologies in DLs.
>
>Mike
>
>
>=============
>This email is a natural product.  The slight variations in spelling and
>grammar enhance its individual character and beauty and are not to be
>considered flaws or defects.
>=============


-- 
---------------------------------------------------------------------
IHMC		(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.	(850)202 4416   office
Pensacola			(850)202 4440   fax
FL 32502			(850)291 0667    cell
phayes@ihmc.us       http://www.ihmc.us/users/phayes
Received on Sunday, 19 December 2004 05:07:07 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:31:03 UTC