RE: Non-monotonic

Hm... Let's see...

What I was trying to say is not very different from your saying.
You have to conclude that can_fly(Fred) if you're given only
with bird(Fred), but later you're provided with a new fact stating that
Fred is a penguin. Then, with non-monotonic reasoning, you can
retract the previous conclusion, and conclude that can_fly(Fred)
is not true anymore. The first conclusion is valid in its context,
and the second conclusion is also valid in its context.

Non-monotonic reasoning is about the ability to retract previous
conclusions and coming up with new conclusions, as the world
is changing and facts we are given are changing.

Please advise if I missing something here...

Minsu 

-----Original Message-----
From: Adrian Walker [mailto:adrianw@snet.net] 
Sent: Thursday, November 13, 2003 11:56 PM
To: minsu@etri.re.kr
Cc: www-rdf-rules@w3.org; www-rdf-interest@w3.org
Subject: Re: Non-monotonic


Minsu --

You wrote...

Web ontologies *should* independently evolve.
In this regard, I think rule languages and processors for the semantic
web
should address problems of reasoning in the open knowledge space,
which is non-monotonic in its nature.

I often see statements of this kind from the RDF and semantic web folks.

In the real world, if I tell you that  

           can_fly(X) if bird(X) & not penguin(X)
            
           bird(Fred).

you will presumably conclude that Fred can fly.  If I later add
penguin(Fred), you will no longer conclude that Fred can fly.

This may sound naive, but can you argue for me please why the above
should be ruled out on the Semantic Web ?  Seems like you are throwing
out the bird with the bathwater (:-)

Thanks in advance,   -- Adrian

-- 

              INTERNET BUSINESS LOGIC
 Business Rules in English + Your Oracle Databases
              www.reengineeringllc.com

Dr. Adrian Walker
Reengineering LLC
PO Box 1412
Bristol, CT 06011-1412, USA

Phone: 860 583 9677
Fax:   860 314 1029
Cell:  860 830 2085



[Dan Brickley wrote,]
  
* Minsu Jang <minsu@etri.re.kr> [2003-11-08 14:26+0900]
    
- OWLer: an OWL inference system. It uses JTP for inferencing.
- ezOWL: a Protege plugin for visual editing of OWL documents.
- MOA: a merging tool for merging multiple OWL ongologies.
- Buchingae: a web-friendly rule language. not XML-based.
- LogicML: a rule markup language, which is a slight extension of
RuleML's hornlog.
      
...same question as above! Where/when can we find out more? 

    

Most of these are not documented well yet, but ezOWL is
published under plugins directory of Protege.
You can find more about the plugin at http://iweb.etri.re.kr/ezowl/.

  
I notice in
    
http://machine-knows.etri.re.kr/bossam/download/download.html that
  
the engine isn't public yet. Do you have a sense for when it will be
available, or expectations about likely license terms? (opensource
    
would
  
be great...)
    

The engine will be available with the similar licensing terms to that of
Jess.

  
A technical question (for everyone, really): 
The Web Ontology WG spent a lot of time and energy 
on the delicate balance between OWL DL and OWL Full. Do you have any
findings from your implementation work that would help us estimate the
    

  
impact of this aspect of OWL on any future Rules standardisation? Is 
it feasible to keep both traditions/approaches/communities happy
    
within
  
a single "OWL-compatible" rules language? 

    

I don't have good technical background on computational logics or
mechanical theorem proving, so what I'm going to say could be
a disaster for this list. ;-) Even if so, please don't hrurry for
escaping,
but be generous enough to provide your invaluable teachings.

Bossam is just a production rule system with some extended knowledge
representation elements which make the engine suitable for inferencing
over OWL ontology. When I started to write OWL inferencing rules, I did
not
fully understand what makes OWL Full and OWL DL different.

The only thing I knew was that OWL properties, classes and individuals
are
pairwise disjoint in OWL DL, but interchangeable in OWL Full.
This discrimination rule should be reflected in OWL inferencing rules
for OWL DL.
For OWL Full, all the facts and rules of RDFS should be added, and OWL
class
should be declared to be owl:equivalentClass to rdfs:Class.

What I have found until now is that the expressiveness of rule language
is
not very different whether it's for reasoning over OWL Lite, DL or Full.
The difference lies in the rulebases. When you write a rulebase
for reasoning over OWL Lite ontology, you will end up with a smaller set
of rules.

What I have in mind from my implementation experiences is that 
a single rules language is enough for OWL inference.

One thing I'd like to address at this point is that when you start to
write
rules for real applications, the situation becomes very complex.
It's not OWL inferencing, but real knowledge processing. The web is open
and many web users will want to process knowledge in the open web, which
will contain many knowledge elements which contradict each other.
Even if you preset all the ontologies for a reasoning session, the
situation
does not become more optimistic, because web ontologies are not
controlled.
Web ontologies *should* independently evolve.
In this regard, I think rule languages and processors for the semantic
web
should address problems of reasoning in the open knowledge space,
which is non-monotonic in its nature.

  
BTW http://machine-knows.etri.re.kr/bossam/docs/owlinference.html
    
doesn't 
  
explicitly mention OWL DL vs Full or Lite. Which flavour(s) are you
targetting?
    

Um, I did not target any one species (:-\), but the rulebase I'm working
on is
targetting OWL Full. And I have a plan to write two more sets of rules
for
OWL DL and Lite.

Regards,
Minsu



  


-- 

              INTERNET BUSINESS LOGIC
 Business Rules in English + Your Oracle Databases
              www.reengineeringllc.com

Dr. Adrian Walker
Reengineering LLC
PO Box 1412
Bristol, CT 06011-1412, USA

Phone: 860 583 9677
Fax:   860 314 1029
Cell:  860 830 2085

Received on Thursday, 13 November 2003 20:01:17 UTC