Rules and ontologies (derives from WRL v. SWRL)

In the discussion of DLP and layering - some people have asked me off 
line why I don't think DLP is sufficient - my response is in terms of 
some use cases which come from real applications I've worked on - 
with small documents there's not a lot of issues, but when you come 
up against large companies actually using this stuff in deployed 
apps, a lot emerges, esp when data interoperability comes along -- 
here are the scenarios I worry about:


1 - Consider an organization like the  Natl Cancer Inst which has a 
big OWL ontology (i.e. has a number of full time people working on 
curation, versioning, etc).  They or some other org decide they'd 
like to use it on databases and datasets for datacleansing or other 
"rule based" operation.
    Recoding the whole into a new rules language would be 
prohibitively expensive unless there is some sort of automagic 
translator of some or all of the RDFS/OWL they use.


2 - Consider an organization using a number of rules to maintain 
their processes and transactions, they merge with another 
organization doing the same.  The boss says "this is crazy, we have 
multiple definitions of key things we need in running the business, 
like the definition of a purchase order" - his golfing buddy 
convinces him he needs to use an ontology to do this and to merge 
data.
    Some mapping of rules to RDFS/OWL would make this much easier


3 - A large company has various activities going on.  One group is 
using  RDFS/OWL to manage corporate Web site assets.  Another one is 
implementing the business  rules that run a number of transactions 
through the corporate data infrastructure.  Problems start to emerge 
where the data and the web assets are not consistent.  Someone says, 
we need to see if the RDFS/OWL and the Rules are computing the same 
thing.  (a real example, US Customs is looking at how to rationalize 
the representations used in their business process rules for 
transactions on databases, and their various product and country 
taxonomies)
     Some way to check consistency between some or all of the rules 
and ontologies is needed.

All of these come from real cases I have worked on, all involved 
either large ontologies, large rule stores or both.   Expecting any 
company to simply say "no problem" and do redundant work seems silly, 
and it is clear from examples like these that a synergy of Rules and 
Ontology makes great sense in the real world.

Problem is DLP is too small a subset of OWL (or of datalog) to be of 
much use in most real world cases.   There are other things being 
explored, like the Kaon2 work at Karlsruhe, that reconcile much 
larger subsets of OWL with datalog, and I think approaches like that 
have great value when looking at examples like the above.

To put it another way, the "twin towers" is not just an imaginary 
problem, it comes up in practice and needs our communities to figure 
out approaches that will help people reach synergy between their Web 
and their Data spaces.

  -JH
-- 
Professor James Hendler			  Director
Joint Institute for Knowledge Discovery	  	  301-405-2696
UMIACS, Univ of Maryland			  301-314-9734 (Fax)
College Park, MD 20742	 		  http://www.cs.umd.edu/users/hendler

Received on Wednesday, 22 June 2005 17:09:47 UTC