Re: basic decisions underlying DAML-ONT

Actually, we did want to include defined classes in this release, but 
there were some problems we couldn't resolve right away, so they are 
on a TO-DO list (which is a good idea, thanks Peter).  The key 
problem was that a goal of DAML as a language is to allow things to 
be stated, but not necessarily to force one particular inference 
model on things.  Consider the case which Ian Horrocks and I 
discussed of "expensive-printer" -- we want to have a defined class 
for this that says a printer is an expensive printer if it costs more 
than $500.
  Easy to express, we would say there is some
   DefinedClass (expensive-printer)
     type printer
     cost >$500
notice this is different than the primitive class "printer" because 
we want it to be inferred that some printer is in this new class when 
we learn that its cost is greater than $500.
  But here's the problem - remember the new game! We're on the web, so 
someone somewhere defines something as a printer, some catalog 
somewhere defines its cost as $1000.  Is the system inconsistent if 
we don't return that printer as an expensive printer???  That is, can 
we insist that there must exist a distributed mechanism that will 
somehow find these two facts (which are likely on different pages, 
maybe even pointing to different name spaces that in turn point to 
other things that eventually both share the same DefinedClass)??
  In DAML-ONT we'd have to say that such a rule "permits" an inference 
engine to decide something is an expensive printer, but does not 
"require" that some inference engine do so.   Deciding how to 
describe, annotate, and "enforce" this kind of thing was too much for 
us to get straight on the first release, so we punted for now.
  Coming soon will be an OIL->DAML document that will show how a lot 
of the Description Logic things (like definedClass) map down (and 
maybe even an OIL namespace built on DAML-ONT) in which case they 
will propose a mechanism and description of defined-classes that I 
hope will find its way into the draft.

  The idea of doing logic and inference in a widely distributed, 
uncontrolled, dynamic, web-grounded way is new, and challenges a lot 
of basic assumptions.   A real challenge is this notion of a 
declarative language that is separate from the inference engine(s) 
that use it.And, if you'll pardon a personal observation for a 
moment, that is what makes it so exciting and so much fun -- many 
things we've taken for granted in KR for many years are suddenly 
called into question.
Dr. James Hendler		jhendler@darpa.mil
Chief Scientist, DARPA/ISO	703-696-2238 (phone)
3701 N. Fairfax Dr.		703-696-2201 (Fax)
Arlington, VA 22203		http://www.cs.umd.edu/~hendler

Received on Thursday, 12 October 2000 11:24:41 UTC