Re: From our comments list

>This comment, send to our public comments list talkin about Guide:
>
>At 15:55 +0100 2/12/03, Ivan Herman wrote:
>>
>>Reviewing the OWL Guide version 1.0,
>>http://www.w3.org/TR/2003/WD-owl-guide-20030210/:
>>
>>As a reader NOT participating in the working group (ie, not being part
>>of all the intricate discussions), the current document is very
>>confusing on the issue on how OWL and RDFS relate. The impression one
>>gets when reading the document is that OWL picks bits and pieces of the
>>RDFS document for its own purposes (rdfs:subClassOf, rdfs:label, etc),
>>but deliberately ignores or deprecates, for example, rdfs:Property or
>>rdfs:Class by introducing owl:Class, owl:ObjectProperty, and owl:Thing,
>>etc. I presume there is a clear model on how these notions relate to one
>>another, but I am afraid the current document creates real confusion to
>>the community. Adding a separate chapter describing the 'onion model' of
>>concentric possibilities offered by RDFS and the various OWL dialect
>>would be beneficial.
>>
>>Ivan
>
>reflects the confusion in our current layering and raises a point we 
>will have to handle -- I have sent him a reply pointing him at 
>overview and ref, and mentioning we are working on this -- I believe 
>we need to create a section for one of our documents, or that could 
>appear separately on our website pointed to by our documents, that 
>discusses this issue -- if we don't do it before LC, we can be sure 
>to get comments like this and have to address it before we move 
>further through the process

The basic problem here is that the WG itself is deeply split on this 
issue. Maybe we ought to be honest about this split and present a 
document which discusses the issue and gives an outline of both 
perspectives.

As I see it, the two basic views are roughly as follows, listed in no 
particular order.

1. OWL is a coherent language with its own abstract syntax and an 
extensional model theory based on the classical mapping from DLs into 
classical FOL syntax. For essentially political reasons, it has been 
integrated with the RDF family of languages and provided with an 
RDF-consistent syntax, but OWL-RDF is less definitive, uglier and 
less important than the purer language which is best characterized by 
the abstract syntax and its model theory. On this view, OWL-DL is the 
'real' language and OWL-Full is a kind of weird bastardization of the 
OWL concept space into RDF, which is not really significant since 
nobody is ever going to implement a complete reasoner for it. On this 
view, the lack of a complete, decideable and hopefully reasonably 
efficient inference engine is seen as crippling, and the cost of 
conforming to a slightly more complex and/or restrictive syntax is 
seen as trivial in comparison.

2. OWL is a semantic extension of RDF(S), representing a kind of next 
step in series of more expressive RDF languages based on a few common 
themes and principles, which include the use of triples-stores as a 
basic medium and the uniform application of basic RDF syntax like 
bnodes and literals, and with a semantic foundation which follows the 
RDF lead, ie is intensional and non-well-founded. On this view, 
OWL-Full is the 'natural' version of the language, the abstract 
syntax and its extensional model theory is is a mere formal 
curiosity, and OWL-DL and OWL-Lite are sub-cases of OWL-Full which 
have been identified because they are easier to implement; but 
fidelity to the syntactic requirements of these subcases is motivated 
purely by pragmatic concerns, to fit inside the competence of some 
inference engine, not by any particular sense that they have a better 
semantics. On this view, it would make perfect sense to think of OWL 
as simply another RDF special vocabulary, with all of the earlier 
special vocabularies retaining their previous meanings. On this view, 
the freedom from a need to consider details of syntactic 
compatibility, the concomitant ease of writing parsers, and the 
overall simplicity are seen as centrally important, and the lack of a 
complete, efficient reasoner guaranteed to work in all cases is seen 
as relatively unimportant, since it is felt that most applications 
will not need such completeness.

----

Now, we - largely Peter - have managed to bring these two pictures 
into a much tighter *formal* alignment than one might have thought 
possible, so that in principle it ought to be the case that people 
taking either position can as it were agree to disagree, without this 
affecting the formal meaning of anything that they can say to one 
another in OWL. This is quite a significant achievement, but it still 
represents a kind of armed truce rather than a reconciliation; and it 
is a fragile truce, even in the formal aspects (viz. the recent 
debate about annotation syntax, where the two views clearly were 
pushing in different directions.) When trying to give an intuitive 
overview of the whole system of languages, sublanguages etc., 
therefore, it might be best to not try to give the false impression 
that there is a single overall picture here which everyone is 
completely happy with.  If we present OWL honestly as a rather 
complicated compromise between two different methodologies, maybe the 
reasons for the complexity will be more apparent, and readers will be 
able to make a more informed choice about which way of thinking is 
most congenial to them, and of the various pros and cons of adopting 
one of a variety of points of view.  (Is it worth the effort to stay 
within the DL subset, for example? Without an honest assessment of 
the real differences between Full and DL, I think it is hard to 
answer this question rationally.)

Pat

PS. I could try to draft a document along these lines, which would 
try to give the pros and cons of each perspective and explain the 
issues involved in a lay-readable form, if the WG feels that it would 
be useful. Obviously it would have to be ad-hominem-purged.


>  -JH
>
>
>
>--
>Professor James Hendler				  hendler@cs.umd.edu
>Director, Semantic Web and Agent Technologies	  301-405-2696
>Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
>Univ of Maryland, College Park, MD 20742	  240-731-3822 (Cell)
>http://www.cs.umd.edu/users/hendler


-- 
---------------------------------------------------------------------
IHMC					(850)434 8903 or (650)494 3973   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola              			(850)202 4440   fax
FL 32501           				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes
s.pam@ai.uwf.edu   for spam

Received on Thursday, 13 February 2003 16:25:02 UTC