W3C home > Mailing lists > Public > www-rdf-comments@w3.org > July to September 2003

[closed] re: Pan-01

From: pat hayes <phayes@ihmc.us>
Date: Fri, 11 Jul 2003 18:56:47 -0500
Message-Id: <p06001217bb349c586aa8@[10.0.100.11]>
To: pan@cs.man.ac.uk
Cc: www-rdf-comments@w3.org


Jeff,

with respect to your comment
http://lists.w3.org/Archives/Public/www-rdf-comments/2003JanMar/0266.html
archived as pan-01, the working group has decided not to accept your comment.

This message is an attempt to summarize the primary reasons for this 
decision. A full response to your comment would amount to a research 
paper. We will also note a possible way to reconcile your position 
with the current RDF design.

Your message makes a number of points, but they can be summarized by 
the complaint that the RDF (and RDFS) model theory (MT) is 
"non-standard". While you do not define this exactly, I believe that 
I follow your point, and partly agree with you: there is a body of 
work and opinion, which is based on certain semantic and 
architectural norms, and relative to which the RDF design is somewhat 
unusual in certain respects.  Several other commentators have made 
similar points, speaking also from positions which accept these 
norms.  The norms in question vary in some details but they all 
accept the idea that 'normal' or 'standard' descriptive languages 
must be strictly layered in some sense; this is sometimes expressed 
as a sharp distinction between data and meta-data, or as a sharp 
distinction between first-order individuals and second-order sets or 
predicates. In order to avoid getting into details, let me summarize 
these various points of view as the 'layered' vision of a 
metamodelling architecture.

First, an observation. Proponents of a layered architecture sometimes 
assume that layering is somehow inevitable, or that non-layered 
architectures are inherently faulty. While this may be a defensible 
aesthetic judgement, it is important to note that it is not a 
technical one.  The most extreme non-layered theory yet developed is 
probably Aczel's non-well-founded set theory, in which layering are 
provably impossible; and it has been shown that this set theory is 
relatively consistent with traditional ZF set theory; in other words, 
'layering' is not required for consistency, and a lack of layering 
has nothing to do with avoiding the classical set-theoretic 
paradoxes.  See Aczel, Peter Non-Well-Founded Sets. With a Foreword 
by Jon Barwise. Distributed for the Center for the Study of Language 
and Information. 131 p. 1988 Series: (CSLI-LN) Lecture Notes for an 
excellent summary. Aczel's results have been the source of much 
recent work in the foundations of mathematics and philosophy of 
language and meaning, by the way, and as he is a professor at 
Manchester 
http://www.maths.man.ac.uk/DeptWeb/Groups/Logic/LogicResearch.html 
you should find it easy to find out more: see for example 
http://www.cs.man.ac.uk/~petera/LogicWeb/settheory.html.  As a small 
contribution to this, recent work in the common logic initiative 
(included in the form of the Lbase language in the RDF MT appendix 
http://www.ihmc.us/users/phayes/RDF_Semant_Edit_Weak.html#Lbase) has 
shown that traditional first-order logic does not actually require 
the syntactic layering traditionally associated with FO languages, 
and that simply re-interpreting traditional Tarskian semantics 
relative to Aczel's set theory produces an 'unlayered' version of FO 
logic which maps transparently into the (now classical) 'holds/app' 
encoding of a "higher-order" logical syntax into traditional FO 
logic, a trick which as I expect you know has been widely used 
throughout applied KR work for about two decades or more. I mention 
all this only as background to the more essential point, however.

The key point about the RDF design is that it *allows* the use of 
layering but does not *require* it. This is one aspect of a critical 
design decision taken early in the RDF project: that RDF should 
impose as few restrictions as possible on what can be said in it or 
how it can be used; c.f 
http://www.w3.org/2001/sw/RDFCore/TR/WD-rdf-concepts-20030117/#section-anyone. 
It is not intended that RDF should be an example of any particular 
'style' of metamodelling or knowledge representation: as far as 
possible, it should be permissive rather than instructive.  This aim 
has been our guide in a number of design decisions, and we feel that 
to impose the 'layered' architecture that you espouse would be too 
restrictive, particularly when the current design does not prevent 
RDF users or applications from sticking to a layered discipline if 
they wish to do so (more on this point later) and since the technical 
work mentioned above has shown that there is no *technical* need or 
requirement to adopt the layering assumptions.

To turn to the particular points you raise in your message: all of 
the numbered problems you mention seem to refer to issues which have 
arisen in the layering of OWL onto RDF. As I expect you know, these 
issues have all been resolved by the Webont working group.  I confess 
myself unable to fully follow the critique of the Lbase proposal you 
offer in the message. If you feel there are any errors in the latest 
version of the Lbase translation of the RDF semantics, referenced 
earlier, please point them out.

----

The discussions surrounding the relationship between OWL-DL (which is 
layered) and OWL-Full (which is not) have led to the Webont WG 
formulating a set of conditions on an OWL/RDF graph which suffice for 
it to be a legal OWL-DL document, see for example 
http://www.daml.org/2002/06/webont/owl-ref-proposed#app-DLinRDF . 
These involve providing a labelling of the non-RDFS vocabulary into 
individual, class and property URIrefs and some conditions on the 
occurrence of blank nodes in various contexts. It has occurred to 
several of us that these similar structural conditions could be 
applied to RDFS graphs and might provide a useful way to recognize 
that an RDF graph was 'layered' in the sense that you might accept as 
more "standard". We have not checked what relationship, if any , 
there might be between these criteria and the design of RDFS(FA), but 
mention this work only to suggest that there might be a useful 
embedding of your preferred architectural style into the RDFS design; 
for if so, it would seem that RDF is able to serve its primary design 
aim of being a universal medium of data interchange even for users 
who prefer a layered architecture.

----------

Please reply to this message, copying www-rdf-comments@w3.org, 
indicating whether this response adequately addresses your comment.

Sincerely

Pat
-- 
---------------------------------------------------------------------
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@ihmc.us       http://www.ihmc.us/users/phayes
Received on Friday, 11 July 2003 19:56:49 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 21 September 2012 14:16:32 GMT