Re: defaults

At 7:31 AM -0500 1/23/02, Peter F. Patel-Schneider wrote:
>From: Jim Hendler <hendler@cs.umd.edu>
>Subject: Re: defaults
>Date: Tue, 22 Jan 2002 22:04:07 -0500
>
>>  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.
>
>[...]
>
>>  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).
>
>Hmm.  I seem to remember lots of problems with defaults in frames.
>
>
>Taken from the Parka 3.2 Manual
>(http://www.cs.umd.edu/projects/plus/Parka/parka3-manual.ps):
>
>	A slot for predicate p is inherited only by frames which do not
>	already have an explicit slot for p [even if p can have multiple
>	values].  [p. 19]
>
>	If an inheritance conflict cannot be resolved [because Parka's
>	inferential distance ordering algorithm computes identical numbers
>	for multiple inheritable frames], Parka randomly chooses one of the
>	frames to inherit the slot from. [p. 19]
>
>In Parka, as far as I can tell, all frame-specified slot information for
>slots that inherit using IDO inheritance can be overridden.
>
>
>I would not call this problem-free by any stretch of the imagination.
>
>
>peter


Peter - this is because you're thinking as a logician, not a 
programmer.  Parka actually allows several different inheritance 
models from "All possible properties which I might have" (i.e. naive 
inheritance of all properties, which is also what SHOE used) to the 
Inferential Distance Ordering algorithm (the one above) which was 
specified by Horty and others in an issue of AIJ.  The IDO 
inheritance, which you refer to above, has no tie breaking rule - so 
in the VERY unlikely case of a perfect tie by that heuristic, we 
would flip a coin.  Parka also contained a switch to turn off the 
coin toss and return multiple values.
  In all the uses of Parka, including recent fielding in daily 
operation in a US military seting, the inheritance of defaults has 
never caused a problem, and they have been rigorously tested 
w/respect to these real world problems.  The fact that there exists a 
very unlikely to occur case was considered, and solved by using the 
multiple value flag -- if more than one value is returned, the user 
is asked which one they prefer -- which makes great sense in an 
interactive system
  btw, this use of Parka, although details are classified, is used in 
a scenario where it routinely manages large ontologies with ties into 
operational databases, making both the A box and T box quite large - 
millions of assertions being handled at any given moment.
  So yes, I would call this comparatively problem free, and I would 
bet that a web company would be willing to consider using a system 
that has been proven in mission-critical applications over one that 
some DL person can assure them won't return a bad answer, even though 
it can't be proven to scale to their needs in real world application.
  -JH
p.s. This is the last message I will send on this in public -- I am 
putting my neutrality hat back on.






-- 
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 Wednesday, 23 January 2002 12:20:38 UTC