Re: [UCR] RIF needs different reasoning methods

On 10 Mar 2006, at 13:50, Gerd Wagner wrote:

>
> Bijan Parsia wrote:
>> What I don't see is the reason for specifying particular proof
>> procedures *instead of* expressive subsets.
>> If two reasoners perform acceptibly and give the same answers...
>> what *more* do you need?
>
> Ian Horrocks wrote:
>> Given that a particular prover has an adequate response time and
>> provides correct answers (as defined by the semantics), then why
>> would I care what procedure it uses in order to do its work?
>
> Bijan and Ian, you didn't get the point: rule processing is
> not just reasoning (the world is not that simple), and rule
> interchange is not just an exchange of implication formulas
> between "reasoners" (it's an old misconception of AI that
> intelligent life is all about reasoning). Most rule processing
> in real life is not involving (much) reasoning, given the fact
> that production rules and ECA/reaction rules are more important
> in IT systems than any form of derivation/deduction rules.
> Changing the state of a system and detecting complex events is
> not the domain of FOL.

This isn't about reasoning. The point is that rules should have some 
well specified meaning, and that the observed behaviour of a 
"consuming" system should be determined by this meaning. Francois 
insists that it is important to also specify something about the 
internal workings of the consuming system. I argue that this should not 
be part of a rule interchange language - I give you some rules, they 
have a well specified meaning, it is up to you what you do with them. I 
won't try to speak for Bijan - he is more than capable of speaking for 
himself :-)

>
> Francois just tried to point out that there are these different
> types of rules, and they cannot all be processed in the same
> manner (with "reasoners").

According to my reading, this isn't what he said at all. In fact he 
explicitly stated that the issue was simply efficiency, and that 
different "consuming" systems should have the same behaviour. To quote:

> Bry: Query-answering against this database and against the views can be
> Bry: perforemd using the theorem prover. The answers would be the same 
> as
> Bry: with database query answering methods. The efficiency, however, 
> would
> Bry: not be the same.



> And even for "reasoning rules",
> there is a computational logic distinction between
>
> - rules for making definitions: constructive derivation rules,
>   called "views" in SQL
> - rules for constraining the possible states (and execution
>   histories) of a system: integrity rules/constraints
> - rules for expressing defaults and heuristics: nonmonotonic
>   derivation rules
>
> And none of these computational logic distinctions are
> reflected in standard FOL (and neither are they in OWL/SWRL):
> you can't simply combine them and submit their merge to
> a "reasoner"!

If the RIF supports rules with different meanings (i.e., where 
different behaviour of the consuming system is expected), then clearly 
they would need to be distinguished. I don't see anyone disagreeing 
about that.


>
> Does the RIF (or the W3C) intend to ignore these distinctions?

Sandro seems to be answering this one.

Ian


>
> -Gerd
>
> ---------------------------------------------
> Gerd Wagner
> http://oxygen.informatik.tu-cottbus.de/IT
> Email: G.Wagner@tu-cottbus.de
>
>
>

Received on Friday, 10 March 2006 19:37:40 UTC