Re: Comments on * DRAFT * Rules Working Group Charter $Revision: 1.60 $

Dieter,

I haven't seen counter evidence that one can't put reasoners such as
Prover9 and Eprover at industrial strength applications, at least I
don't see any reason at all to exclude those excellent technologies
from for instance our own advanced clinical applications. They listen
to FOL with equality formulae. Also constraint handling rule systems
do so. In my own implementation experience, I have no problem to map
such formulae to N3 based reasoners that we use since last 6 years.

-- 
Jos De Roo, AGFA http://www.agfa.com/w3c/jdroo/




Dieter Fensel <dieter.fensel@deri.org>
Sent by: public-rule-workshop-discuss-request@w3.org
19/08/2005 18:10

 
        To:     public-rule-workshop-discuss@w3.org
        cc:     (bcc: Jos De_Roo/AMDUS/MOR/Agfa-NV/BE/BAYER)
        Subject:        Comments on * DRAFT * Rules Working Group Charter $Revision:   1.60 $



Comments on * DRAFT * Rules Working Group Charter $Revision: 1.60 $

First, the draft does not at all reflect the majority of the discussions 
during the W3C workshop on rule languages that had the aim to lay the 
ground. In the future, we can save traveling overhead if these workshops 
and their discussions are largely ignored anyway based on predefined 
opinions. What is sketched as a charter draft reflects the opinion of a 
small (but noisy) minority at this workshop. In general, I would like to 
raise the following four issues.

1) Can OWL really provide a lot of support for use cases asking for rules?
The agenda mentions "that some use cases for rules can be addressed with 
OWL". Indeed, it was funny (or frankly spoken, painful) to see during the 
workshop that people used range and value restrictions in OWL as integrity 

constrains excluding certain values for attributes. Fortunately, nobody 
had 
told them that OWL would simply ignore these "constraints" and infer 
artificial equalities of instances instead.

2) A rule language should be based on full first order logic with 
equality?
We will first explain why and then why not.

2.1 Why
We simply do not know.

2.1.a Why full first order logic?
No justification is given in the text! The reader can only guess that the 
reason is to define the rule language as an extension of OWL.

2.1.b Why equality?
No justification is given in the text! The reader can only guess that the 
reason is to define the rule language as an extension of OWL.

2.2 Why NOT
We have good reasons to believe that it is not a good choice to define a 
rule language as a first-order language with equality.

2.2.a Why NOT full first order logic
Without any justification the charter requires that the rule language is 
based on full first order logic. Notice that none of the existing rule 
languages and their implementations are based on such a paradigm. 30 years 

of research and industrial practice are ignored WITHOUT ANY JUSTIFICATION.
The draft charter becomes a parody when it states: "It is understood that 
not all rule engines will offer complete FOL reasoning." This is a funny 
understatement! There will never be any inference engine in our universe 
that will offer complete FOL reasoning. This is precisely one of the 
reasons for 30 years of research on syntactical sub languages of FOL and 
the usage of minimal model semantics. The proposal would destroy all 
useful 
computational properties of rule languages for the sake of defining it as 
an extension of OWL. (a) All problematic design decisions of OWL would be 
inherited by the rule languages (For example, rules could have been 
layered 
on top of OWL Lite, if this language would have been defined more 
appropriate and minimalistc). (b) Instead of nice computational rule 
languages, a first order zombie language is generated that has neither 
complete and correct, nor efficient reasoning support. This applies to 
practical as well as theoretical aspects.
Not only is any justification missing why the rule language should be 
based 
on full first order logic with equality. Moreover, this decision is used 
to 
ignore significant requirements of industry working with 
rules.  "...Negation as failure ... (is) out of scope."
Why? Because "combining it with FOL (is) an unsolved research problem." 
And 
this in  the case where "NAF is essentially the type of negation seen in 
many commercial and research rule systems". Why can industry and many 
research prototypes deal with NAF when the working group can not? The 
reason is that the working group uses without any justification first 
order 
logic with equality and "combining it with FOL is an unsolved research 
problem." Because of this, most research on rule based systems and 
important requirements of industry are ignored. Similarly, update 
operations and evaluation strategies of rule based systems are ignored 
when 
they do not fit into the unjustified theoretical framework. If you only 
have a hammer everything has to be a nail! Why not simply getting rid of a 

non-justified and wrong choice for the underlying logical model instead of 

ignoring most research on rules and all important requirements of 
industry?

2.2.b Why NOT equality
Firstly, many equalities in the context of the web (like addresses in a 
Unix file system used as URIs) are very cumbersome to model in logic. An 
oracle, external to a logical language (that provides a unique identifier 
for equivalence classes of terms), seems to be a much more useful tool. 
Secondly, equality significantly blows up the reasoning costs of a logical 

language when simple syntactical term matching to decide whether two 
identifiers are equal is replaced by costly logical inference over axioms 
and all their many ***many*** consequences on whether the two terms have 
to 
be viewed to be equal. In a nutshell: it costs a lot and it is of very 
limited use. Why would you want to buy into it?

3 OWL and a rule language
It is a well-know result from theoretical and practical work around first 
order logic that the language is undecidable in general and difficult to 
evaluate in computational terms. Therefore, a lot of research has been 
done 
during the last fifty years to come up with more narrowly defined logical 
languages. One example is the Description Logic paradigm that underlies 
OWL 
and a second example is Horn logic with minimal model semantics that 
underlies most implementations of rule based systems. Unfortunately, it is 

also a well-known fact that it is not possible to straight-forwardly 
combine both language paradigms without destroying the interesting and 
useful computational properties of both. Therefore, integration can only 
be 
done at a minimal level ensuring that none of the both language types are 
destroyed in their justification.
SWRL and SWRL-FOL, which are mentioned as positive examples in the draft 
agenda, do precisely the opposite. They naively extend Description Logic 
with a rule syntax. Therefore, they end up in an undecidable language, the 

same way first order logic is. That is, these languages restrict the 
expressive power and enforce cumbersome syntax of FOL without providing 
anything in return in terms of reduced computational complexity. This 
happens when you ignore 30 years of research. These languages are neither 
justified by a proven body of research nor by a body of implemented 
reasoners nor industrial experience. It is quite hard to understand why 
W3C 
wants to commit to such enterprises?
The precise definition of what could be a maximal intersection of rules 
and 
description logic being minimalistic enough for not destroying the 
computational properties of the resulting language is currently the 
battleground for many PhD students and at the same time an interesting 
topic of research for the next years. But it is not at all the task of W3C 

to standardize very early drafts of PhD theses. And for sure, this 
intersection will be less expressive than the ones mentioned in the draft 
agenda.

4) Summary
In general, there is a high risk that a working group with such an agenda 
generates serious damage to
- the semantic web by providing a useless language with unmanageable 
computational properties;
- the rule industry since important features are simply ignored by this 
proposal and a language that has nothing to do with the rule paradigm is 
offered as exchange format;
- the general reputation of W3C as a place where consensus is achieved 
without manipulation; and
- the general reputation of W3C as a producer of recommendations of high 
quality.
Given the current draft, we would rather prefer not seeing a working group 

or renaming the working group to "A working group on extensions of OWL 
towards a not well-justified sub fragment of first order logic with 
equality." However, with OWL-Full we already have a zombie in our 
basement, 
that is, there is no need to create a second one?

Yours sincerely,

Dieter Fensel
Scientific Director of DERI
----------------------------------------------------------------
Dieter Fensel, http://www.deri.org/
Tel.: +43-512-5076485/8

Received on Saturday, 20 August 2005 11:32:32 UTC