Re: defaults

At 4:23 PM -0500 1/22/02, Lynn Andrea Stein wrote:
>I believe that we will have a difficult if not impossible time in
>producing a "reasonable" default mechanism.
>
>



************ WARNING WARNING - CHAIR NEUTRALITY VIOLATION ******************

I am violating Chair neutrality for the sake of this message!!

************ WARNING WARNING - CHAIR NEUTRALITY VIOLATION ******************

The overwhelming "anti-default" span of this discussion forces me to 
mention that the "other half" of the KR world (the frames folks) have 
never had any real problem with defaults (or non-monotonicity). In 
fact, the semantic net systems we all derive our heritage from (DL or 
frame!) started with an argument that the nets allowed better 
encoding because one COULD have defaults, for example, it is a useful 
encoding to say

MAMMAL --- birth-method ----> live

AARDVARK --- ISA ---> mammal
AARDWOLF --- ISA ---> mammal
(500  mammal examples deleted)
ZEBRA -- ISA ---> mammal

and thus save ourselves 500+ assertions that these things are each 
live bearing.

The exception
PLATYPUS --- ISA ---> mammal
PLATYPUS --- Birth-method ---> live

causes no one any problems unless they are trying to use a 
classifier, which is why many of us frame folks don't like 
classifiers!!

The key point I want to make is that it seems to me we are assuming 
that this sort of solution is bad a priori, because it is somehow 
"non-monotonic" -- yet there are effective inference procedures for 
this (polynomial if one excludes explicit "is not a" links) -- and, 
in fact, those system have scaled way beyond anything done in DLs to 
date (speed, size, assertion type, etc.) [1][2][3]

I believe it is important that this group make considered decisions 
about how the extra utility of these constructs trades off with the 
logical cleanliness of the other.

I would love to find a mechanism (for example the one Frank suggests 
seems to me to lead to something workable), or a "layering" (the way 
Dieter was using the term) which allows some coexistance of these 
things.  Frame inference systems have been used in a lot more 
deployed systems than DL systems have and over a lot longer span of 
years, and it seems strange to me for us not to at least seriously 
explore the issues -- especially as Guus made very strong arguments 
for the need for this in all his use cases particularly cataloging 
and web site management, which are two of the areas in which actually 
deployed ontology systems are used regularly!

   -Jim H.


[1] http://www.cs.umd.edu/projects/plus/Parka/aaai97.ps
[2] http://www.cs.umd.edu/projects/plus/Parka/ppai.ps
[3] A nice collection of papers on sem networks can be found in: 
Semantic Networks in Artificial Intelligence, edited by Fritz 
Lehmann, Pergamon Press, 1992.

************ WARNING WARNING - CHAIR NEUTRALITY VIOLATION ******************

I am violating Chair neutrality for the sake of this message!!

************ WARNING WARNING - CHAIR NEUTRALITY VIOLATION ******************


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

Received on Tuesday, 22 January 2002 22:07:56 UTC