W3C home > Mailing lists > Public > public-webont-comments@w3.org > June 2003

Re: Review section 4.1 of OWL AS&S

From: Graham Klyne <gk@ninebynine.org>
Date: Mon, 23 Jun 2003 11:00:18 +0100
Message-Id: <5.1.0.14.2.20030623100613.02dddb88@127.0.0.1>
To: Ian Horrocks <horrocks@cs.man.ac.uk>
Cc: public-webont-comments@w3.org

Thank you for your consideration.  For the purposes of formal process, I 
find your response is satisfactory.

There are, however, a couple of clarifying points I'd like to make 
concerning your response, which you may choose to ignore.

...

At 11:44 22/06/03 +0100, Ian Horrocks wrote:
> >
> > There was some discussion about the purpose of these transformation
> > rules.  From the description given, I think it is to translate OWL AS 
> to an
> > RDF graph, but others have
> > also suggested that they might be used for defining inverse transformation.
> > I don't think the mapping table is well-suited to that purpose.
>
>Please see [1] for a detailed discussion of this issue.

I accept the point that it is possible to implement a reverse 
transformation (rdf -> owl) that satisfies the rules given, and that any 
such transformation satisfying the relationships defined by the rules will 
assign consistent semantics to the original RDF.  In that respect, I agree 
that specification is sufficient as a formal description of what must be done.

My concern is the possible difficulty of such a transformation.  To deploy 
a technology at Internet scale, not only must it be possible to implement, 
but I believe that the possible means of implementation must be 
sufficiently clear that one does not need to be a researcher at the 
forefront of the corresponding technology in order to construct such an 
implementation.  Standardization is an engineering process, not research.

If the engineering involved to deploy a standardized technology is not 
sufficiently clear, I see two likely consequences:
(1) developers simply do not bother with the standard concerned, and do 
their own thing.
(2) developers try, and get it wrong, resulting in widespread deployment of 
non-interoperable implementations.


Tempered by the above considerations, I do acknowledge and accept your comment:
>This document only provides definitions of various parts of OWL.  Turning
>these definitions into effective procedures is a task for
>implementors (c.f.the OWL Species Validator, available from the WG
>web page or at [5]).

...

>This bit has been changed to
>
>The left column of the <a href="#transformation">table</a> gives a
>piece of <span class="change">abstract</span> syntax (S);
>the center column gives its transformation
><span class="change">into triples</span> (T(S));
>and the right column gives an identifier for the main
>node of the transformation (M(T(S))), for syntactic
>constructs that can occur as pieces of directives.
>
>The new version is available at
>http://www.bell-labs.com/user/pfps/owl/semantics/

(I could not access that document, but the approach you outline seems fine.)

...

> > It has been my assumption that the various flavours of OWL can be used 
> as a
> > "logical layer" to perform some reasoning about things described by
> > common-or-garden RDF.  I've since been told this may not be so.  If it is
> > true that Owl DL and Owl-lite are not suitable for doing any reasoning
> > about things described by ordinary "in the wild" RDF, I think this 
> needs to
> > be clearly stated.
>
>This is addressed by the introduction to the OWL Web Ontology
>LanguageOverview [3]. Test also discusses document conformance in some
>detail [4].

I have taken a look at your [3] and still certain aspects of the role of 
OWL vis-a-vis RDF were not entirely clear.  Maybe it helps if I try to 
explain why.

My uncertainty centres on the role of an "OWL document".

I see a conceptual distinction between two uses of RDF:

(1) "raw RDF data":  used to describe things and concepts in the "real" 
world.  Some examples we have seen are:  tables of airport 
locations,  tables of geopolitical information, descriptions of meeting 
agenda and participants, personal information associated with a web 
homepage, and so on.  This kind of information, considered alone, contains 
no basis for anything more than the most primitive of logical 
reasoning.  But I expect an overwhelming majority of RDF data available 
from the web to be such information, and the utility of Semantic Web 
technology to be driven by such availability.

(2) "RDF logic decription":  specialized RDF vocabularies, with 
corresponding additional semantics, that are used to provide information 
about other RDF vocabularies, and provide the basis for reasoning about 
"raw RDF data".  Examples of such would be RDF schema, OWL, the logical 
rules vocabulary implemented in CWM, etc.

My expectation about an "OWL document" had been that it applies to 
presentation of a particular style of "RDF logic description", which an OWL 
reasoner can combine with arbitrary "raw RDF data" to draw new conclusions 
about the things described there.  So an OWL document that says that a cat 
is a kind of animal combined with raw RDF data that asserts that I own a 
cat can be used by an OWL reasoner to deduce that I own an animal.

My uncertainty from reading the OWL specifications is that it is not clear 
to me whether or not the data that an OWL reasoner can process must *all* 
conform to the OWL abstract syntax, though the implication seems to be that 
it must.  To my mind, this raises a real possibility that a large amount of 
RDF published on the web will be opaque to (say) an OWL-DL reasoner.  There 
is, I believe, a lot of RDF data published on the web that does not have 
explicitly an rdf:type property on every thing described (though these 
could often be inferred given appropriate RDF schema information and an RDF 
schema reasoner), yet the OWL specification seems to say that this type 
information must be explicitly provided for the raw data to be accessible 
to an OWL-DL reasoner.

It may be that I'm not properly understanding the OWL specification, but I 
do think this is a sufficiently important issue that the future relevance 
of OWL will be well-served if the relation between OWL ontology 
descriptions and the form of raw data consumed by an OWL reasoner is more 
clearly stated.

>[1] 
>http://lists.w3.org/Archives/Public/public-webont-comments/2003Jun/0025.html
>[2] http://www.w3.org/TR/2003/WD-owl-semantics-20030331/syntax.html
>[3] http://www.w3.org/TR/owl-features/#s1.3
>[4] http://www.w3.org/2002/03owlt/editors-draft/draft/#conformance
>[5] http://phoebus.cs.man.ac.uk:9999/OWL/Validator


#g


-------------------
Graham Klyne
<GK@NineByNine.org>
PGP: 0FAA 69FF C083 000B A2E9  A131 01B9 1C7A DBCA CB5E
Received on Monday, 23 June 2003 06:59:21 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:43:29 GMT