Fwd: Re: CHAIR-NOTE: Defaults and etc.

>Date: Fri, 25 Jan 2002 08:45:55 -0500
>To: "Peter F. Patel-Schneider" <pfps@research.bell-labs.com>
>From: Jim Hendler <hendler@cs.umd.edu>
>Subject: Re: CHAIR-NOTE: Defaults and etc.
>Cc: w3c-semweb-ad@w3.org
>>Summary:  I feel that I'm in a timewarp back to the '70s.  I sentence Jim
>>	  to go back and read all the ``What's in a ...'' papers and also
>>	  papers on first-order logic.
>>There is absolutely no reason that anyone should think that any properties
>>of objects except for some RDF(S) structural properties, such as rdfs:type,
>>can be deduced on subclasses of those objects.  This axiom certainly says
>>absolutely nothing about the inference of arbitrary properties on
>>subclasses, as can by easily determined by observing that there are no free
>>variables in the first position of PropertyValue predications.
>>There is no reason that any AI person should even be trying to make such
>>inferences go through without a ton of qualifications.  Properties of
>>classes, like cardinality, are very seldom also properties of their
>>subclasses, unless they are really characteristics of the members of the
>I realized that my earlier response may confuse some readers, let me 
>try to provide some perspective.  I realize this is open to 
>interpretation and present it simply as my view of the field as one 
>who lived it, and not as any kind of historical writ.  Others are 
>free to respond and elaborate, this is the last message I'll send on 
>the subject - it's taking us too far from the needs of the WG, which 
>should me my first and sole concern...
>Let me provide some background for those of you who think that this 
>is a one-sided issue:there were some nice papers written by Gene 
>Charniak and Bonnie Webber back in the late 1970s that talked about 
>the logical equivalences of these things, and those predated Ron 
>Brachman's work in KL-ONE (which was also predated by a number of 
>frame systems). In those papers, their were some gaps which 
>reflected on exactly these issues we are discussing.  One group of 
>people, inspired largely by Brachman's seminal work, took to the 
>more logical approach and worked on theoretical fixes to these 
>issues as well and focused on languages with either good theoretical 
>expressivity, or with nice computational properties (in which 
>Peter's work is generally considered seminal).
>  A second group, however, worked more on what we now call the 
>frame-based approach, but whcih really focused more on Charniak's 
>approach to this, which led to langauges like FRAIL, which 
>formalized this stuff in other ways.  Some other important work in 
>this area included the development of formalizations of exactly the 
>inheritance algorithms for class/property inheritance, for example 
>the IDO algorithm Peter and I alluded to recently.
>  My own work in the area is reflected in dozens of papers which not 
>only discuss various aspects of inheritance, but also focus on the 
>algorithms for scaling these, the applications these can be put to 
>etc. (I even have a pending patent on the worlds fastest and most 
>scalable inference algorithms for inheritance, which is what led me 
>into SHOE and then the Sem Web).
>In fact, if one goes back to a famous talk given by Roger Schank 
>somewhere in the late 1970s (sorry. my memory fails me as to what 
>meeting it was  - I know I was there and remember it was in 
>California), he referred to the neats and the scruffies and at that 
>time, this particular debate we're still having today was one of the 
>examples used as a differentiator!  For thos who still haven't 
>figured it out, Peter is the quintessential neat and I'm the 
>archetype for a scruffy.
>Thus, the issue being discussed here gets so much to the heart of 
>things and the differences in how frame and DL folks see the world 
>that it's hard to even know where to begin.  In the frame way of 
>looking at the world, properties on a class refer to properties of 
>the INSTANCES of those classes.  Special properties may be put on a 
>class in some systems (i.e. I can say the class of cows has 56 
>types, but that this is not true of the instances).  I assumed in my 
>mammal example that people would use the standard ideas about 
>inheritance, etc. that they learned in their first AI courses (which 
>I've been teaching since 1980).  Thus, when I state that mammal has 
>the property livebearing, I assume it will be a property assumed 
>true of all instances of mammals.  I believe I can create systems 
>which let me live and assert these sort of facts and classes in the 
>langauges the DL folks love, and have been doing so for a couple of 
>years now with the DAML family of languages (to which I have been a 
>contributor, and was part of the consensus).
>Perhaps the details of my example were wrong, but the use and intent 
>are still there, and no matter what our language looks like, I will 
>use these approaches in my work, as will a sizable (if not dominant) 
>proportion of implementors.  And the good thing is - WHO CARES!  The 
>DAML+OIL ontology library has 170 or so ontologies which have close 
>to 3 million facts linked to them from applications all over the 
>world.  I use those routinely with my students to create instances 
>and to infer the properties those instances have (represented in the 
>real world by values of slots in web forms that are filled in, but 
>which users can overwrite).  The biggest value for me is the fact 
>that the ontologies give me a machine readable way to know the 
>properties "legitimized" by various classes, and thus help me in 
>creating large repositories of instances with appropriate values 
>entered by my users.  Adding the restrictions allowed by DAML+OIL 
>has proved to be useful, because it lets me know about 
>cardinalities, check for consistency between users inputs and 
>ontological definitions, etc.
>My key point I am once again trying to make is that I use exactly 
>the same markup as Peter does for a different class of applications 
>and they coexist just fine on the web -- it is why I like DAML+OIL 
>so much.  If Peter thinks I make "bad inferences" or if I think his 
>stuff is unworkable in the large -- again, WHO CARES??  On the web 
>we both do our thing, and let the market decide.
>Our language MUST service as many users as possible without 
>constraining them to any one view of ontologies.  Is that just my 
>opinion??   NO!
>  AGAIN I QUOTE OUR CHARTER (the only document that has true binding 
>status on our decision making unless we wish to appeal it)
>>     *  The products of the WebONT group should not presuppose any 
>>particular approach to either ontology design or ontology use. In 
>>addition, the language must support the development and linking of 
>>ontologies together, in a web-like manner.
>The exciting thing about DAML+OIL is that both neats and scruffies 
>seem able to live with it, and have started using it -- I've seen 
>major implementations that are in daily use using some approaches, 
>and exciting theoretical papers submitted to conferences based on 
>either extending it or proving things about its coverage.  The 
>excitement of the consensus process is that we can do this, but only 
>if we keep a focus on the mission, and on the fact that we only 
>succeed if we build a langugae that ALL the members of WOWG (and the 
>communities we represent) can live with!!!
>  -Jim H.
>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

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

Received on Friday, 25 January 2002 10:45:47 UTC