Re: Cross-ontologies reasoning

At 9:59 -0500 12/29/03, Bill Andersen wrote:
>On Dec 26, 2003, at 1205, Jim Hendler wrote:
>>
>>  Drew,. I agree completely if we use your definition of 
>>ontology-merging.  Partial mappings have a greater
>>  success (particularly allowing heuristic mechanisms), and of 
>>course there's no reason we can't have some human in the loop. 
>>Also, none of the literature I know allows instances to be mapped 
>>against multiple ontologies, which is a new idea that occurs easily 
>>on the Semantic Web, and which opens many opportunities for new 
>>research.
>>   So I guess I'm kidding myself
>>   -JH
>
>Hey, Jim...
>
>I don't often post to this group but this discussion is near to my 
>heart - well,
>at least to my research interests.  I think the answer depends not 
>so much on the
>ontologies, but on the systems that advertise those ontologies as a 
>description
>of what they "know about".  Your prescription would be fine if all you want to
>do is content management.  There, close is good enough.  Close is 
>not good enough
>for many database integration applications, where much could hinge on the
>correctness of the mapping (e.g., a factory floor control application).  Drew
>is likely to be correct that such mappings will be hard to come by 
>automatically,
>but, just like what kind of "ontology" you need depends on what you want to do
>with it, so it is with what kind of mappings you need.  Bottom line is I don't
>think you're kidding yourself so long as you stick to (vary) fault-tolerant
>applications.

Bill, I disagree, I'd actually claim that in the real world virtually 
all of this stuff we're doing is heuristic at best, so I'm less 
convinced that these techniques wouldn't work.  That said, I don't 
see why partial matchings have any less value than total matching. 
If you and I are transacting business (or you are running a factory 
using a machine I manufactured) I don't see why we would need, or 
even want, to merge everything in our ontologies (for example, maybe 
you believe that Rush Limbaugh is a rational being and I cannot live 
with that belief)  -- rather, we would make sure we could translate a 
set of terms critical to our application, and not worry much if there 
is some incocnsitency in the larger set -- only the smaller set of 
terms we transact needs to be conistent between us -- there's a lot 
of reason to believe this latter kind of thing (multiplied across 
many ontologies and applications) would scale better with no (or 
little) loss of capability for the critical applications

Besides, I am arguing for human in the loop in some way, which seems 
to me much more likely to be more powerful and usable in the 
short-term than any fully automated thing -- so I think you are way 
over pessimistic

>
>Question: What does it mean "[to map] instances against multiple ontologies"?

(I sent this to a couple people one-on-one, but since I was asked on 
the list) --

what I had in mind is something more like the fact that on the SW I 
can say something like

:Photo1 foaf:depicts :MyDog.
:MyDog a cyc:Dog,  Petshops:Dog;
  cyc:skinColor  colors:brown;
  cyc:Breed Petshops:Labrador.

 From the above (which cam be easily created in RDF from various 
markup tools floating around - cf http://owl.mindswap.org/images/) I 
postulate a tool should be able to abduce some useful information not 
just about MyDog, but about how cyc and petshops definitions of "dog" 
are related -- note also that this would work if Petshops used some 
other URIref than "Dog" -- canine, celeb, perro, etc...
  There are several people (perhaps you folks included) working on 
mapping database schemas/instances to each other and using that for 
integration, but most of the ones I know map multiple databases to a 
single ontology (or a set of disaparate, but unlinked ones), what I 
haven't seen is a lot that is looking at instances like the above 
which bring together info from multiple ontologies in interesting ways
  -JH



-- 
Professor James Hendler			  http://www.cs.umd.edu/users/hendler
Director, Semantic Web and Agent Technologies	  301-405-2696
Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
Univ of Maryland, College Park, MD 20742	  240-277-3388 (Cell)

Received on Monday, 29 December 2003 13:55:03 UTC