Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: SUO: questions about SUMO




Michal,
   Replies below:

At 11:04 AM 4/25/2002 +0200, Michal Sevcenko SITE wrote:


>Dear Adam,
>  Thanks for your reply. I'm very pleased that you like my browser.
>Now let me reply to your comments:
>
> > > > Competence: How well does it support problem solving? That is,
> > > > what questions can the representation answer or what tasks can it
> > > > support?
> > >
> > >I think that this is very important point. It would be nice to have
> > >some benchmark queries that are of some practical utility, and that
> > >are (or will be) within competence of SUMO. Do you have something
> > >like this?
> >
> > We don't at the moment, but I agree it would be a good idea.  On
> > domain-specific projects we're doing that use SUMO, we do have
> > question sets, but they aren't limited to the content of SUMO.  We'd
> > welcome suggestions for questions that we could include in some sort
> > of test corpus for SUMO itself.
>
>I don't have much experience with using ontologies for automatic
>query answering, so I can hardly suggest some reasonable test suite.
>But I am going to experiment with this in the near future. In fact, I
>am a bit suspitious about the ability of SUMO to answer queries. It
>seems to me that at this stage SUMO is more valuable for exchanging
>knowledge between _people_, but I may be wrong (as I hope :). You
>also mentioned that there are domain-specific projects built on SUMO
>that have some practical test suites. Could you point me to some
>publicly available information about them, please?

You can find our ontologies of computer network Quality of Service (QoS) 
and ecommerce services at <http://ontology.teknowledge.com>.  The QoS 
project is over and we didn't package our test code unfortunately.

It's easy to come up with some queries that test SUMO and aren't that 
sophisticated:

(instance Foo Object)
(instance Bar Object)
(instance Flub Object)
(part Flub Bar)
(part Bar Foo)

Query: (part Flub Foo)

"If Flub is a part of Bar and Bar is a part of Foo, is Flub a part of Foo?"

------------

(instance Joe Human)
(instance Bob Human)
(instance Something23 Object)
(instance JoeBuysFromBob Buying)
(agent JoeBuysFromBob Joe)
(patient JoeBuysFromBob Something23)

Query: (exists (?EVENT)
          (and
            (instance ?EVENT Selling)
            (agent ?EVENT Bob)))

"If Joe buys something from Bob, did Bob sell something?"


> > Is it ok if we link to your site from
> > http://ontology.teknowledge.com ?  Are you planning to keep the site
> > up to date as we release future versions of SUMO?  Maybe we could even
> > talk about sharing code and produce a combined version of our browsers
> > that has "novice" and "advanced" versions of presentation.
>
>Well, it would be an honour for me to have a link to my browser on
>your page.

Great!  I've linked to your site.

>  I'm also willing to share the source code. The
>virtual.cvut.cz site is quite stable, this is not a problem.

What language is it written in?

>But
>there are some issues that need to be solved:
>1) The program still have some bugs I'm aware of and lots of bugs I'm
>not aware of, of course. It was written quite in a hurry. But I think
>these bugs can be fixed soon.
>2) The program relies on special semantic features that are used in
>SUMO by convention, but are not standardized (such as subclass
>relation is named 'subclass', and always have two arguments, etc.) So
>the program works well with SUMO, but should be updated to be robust
>for ontologies that do not conform to these conventions.

But with luck, these will be standardized :-)

>3) The program uses offline copy of sumo (merge.txt), so some
>synchronization policy have to be established.

We generally notify the IEEE SUO list every time we release a new 
version.  Would that work?

>4) The file merge.txt has been manually split into several files
>(according to comments in merge.txt), so that the browser can show to
>which part of ontology belongs a displayed concept, as well as all
>concepts in one part of the ontology. Later on I noticed that you
>have a special version of merge.txt that have concepts wrapped into
>some namespace predicates, so maybe I should exploit this in my
>browser.

That version was done mainly to prove a point and we're not maintaining 
it.  We could agree on the syntax for a flag that indicates a change of 
ontology section.  Because KIF doesn't have contexts, I'd prefer to keep 
that as a convention in a comment, like

;;  Context: BaseOntology

which would signal that all statements after that, but before the next 
context statement would be in that named context.

>5) Maybe you can suggest what should I write at the title page of my
>browser so that it will be formally correct.

It should say something like "Suggested Upper Merged Ontology - a starter 
document in the IEEE Standard Upper Ontology effort".  It would be good if 
you could also include a link back to http://ontology.teknowledge.com


> > A TimePoint is not just a number, which is an abstract mathematical
> > notion, but rather a specific point on the time line.  TimePoint(s)
> > can be specified with respect to a calendar using functions like
> > SecondFn or YearFn.  For example, SecondFn takes a RealNumber and a
> > Minute and returns the number of second into that particular Minute.
>
>Perhaps I did not express myself correctly. I understand that
>TimePoint is more than a RealNumber. I wanted to say this:
>1) TimePoint is similar to RealNumber. In fact I think that
>everything what holds for instances of RealNumber holds for instances
>of TimePoint too. I do not claim that TimePoint should be a subclass
>of RealNumber (perhaps its not possible to crossover Abstract and
>Physical thing), but maybe TimePoint could inherit RealNumber's
>properties some other way.

I'll grant that there is some relationship, but I think it's covered in 
SUMO by using RealNumber(s) as arguments to the appropriate temporal 
specification functions.  Could you suggest a specific property of 
TimePoint(s) that SUMO doesn't have that should be added?

>2) Some axioms that say something about TimePoints holds for
>RealNumbers too, and are therefore missing in axiomatization of
>RealNumbers. For example, every TimePoint is greater or equal than
>negative infinity,

I believe this is already covered by the fact that since TimePoint(s) are 
specified using numbers (including RealNumber) which have that axiom, the 
same constraint is entailed on TimePoint.

>between two distinct TimePoints there exists other
>distinct TimePoint, etc.

This gets into the issue of "dense" time, which is an issue I think Ian 
Niles has deliberately tried to stay away from.  He's back from vacation 
next week and could explain this better than I can.

>And finally, I would have another specific comment about SUMO
>content. I thing that the documentation of overlapsTemporally
>relation is somehow misleading:
>
> > (overlapsTemporally ?interval1 ?interval2) means that the two
> > TimeIntervals ?interval1 and ?interval2 have a TimeInterval in common.
> > Note that this is consistent with ?interval1 and ?interval2 being the
> > same TimeInterval.
>
>It suggests that if we think the intervals as sets of TimePoints,
>they have nonempty intersection. However, the axiomatization of this
>relation:
>
> > (<=>
> >       (overlapsTemporally ?INTERVAL1 ?INTERVAL2)
> >       (or
> >             (equal ?INTERVAL1 ?INTERVAL2)
> >             (during ?INTERVAL1 ?INTERVAL2)
> >             (starts ?INTERVAL1 ?INTERVAL2)
> >             (finishes ?INTERVAL1 ?INTERVAL2)))
>
>says that ?INTERVAL1 is a subset of ?INTERVAL2 (in the set analogy),
>which is indeed something different. Am I right?

The axiom does look more restrictive than the comment, and may be an 
error.  Many thanks for pointing this out.

Adam


>With best regards,
>Michal Sevcenko
>
>----------------------------------------
>Ing. Michal Sevcenko
>Department of Computer Science
>Faculty of Electrical Engineering
>Czech Technical University in Prague
>Tel +420 2 2435 3661
>http://webis.felk.cvut.cz/en/people/sevcenm.html

Adam Pease
Teknowledge
(650) 424-0500 x571