SUO: RE: W3C approves RDF and OWL as recommendations
I'll try and up my 25% a little...
> DA> The bottom line is that the W3C's initiatives and
> > the loftier aims you describe aren't mutually exclusive.
> > I'd go further and suggest that the former may indeed
> > help bootstrap development of the latter.
>
> I agree with the first statement, but disagree with the
> second. The biggest weakness of the semantic web is
> illustrated by the popular "layer-cake" diagram (see
> attachment). As that diagram shows, the W3C started
> with a purely syntactic foundation consisting of Unicode,
> the URI naming scheme, and the XML language. There is
> nothing wrong with including those notations as important
> parts of the W3C strategy. But what is wrong is the absence
> of an architecture that addresses the semantic issues.
I'm not sure what kind of thing John has in mind for architecture that
addresses the semantic issues, maybe there is something lacking here. But
the "layer-cake" is only really addressing the relationship between the
W3C's Semantic Web technologies (OWL etc) with the technologies of the Web.
There is a key point here - it isn't that the lower-level blocks such as the
URI naming scheme and the XML language are an optional extra, they are a
necessary prerequisite for a WWW ontology language.
Approaching the 'bootstrapping' point from a very different angle, a web
augmentated by an ontology language (however limited), is likely to be more
conducive to the development of more sophisticated systems than a
poorly-indexed library.
> I agree with Azamat Abdoullaev's statement:
>
> AA> the administrators running the World Wide Web Consortium
> > rushed the matter by recommending OWL as an ontology
> > language standard fit for structuring the Web data,
> > documents, and applications. Since, beside the well-known
> > merits, the language has bad conceptual faults which make
> > it fall short of wide commercial use.
>
> As an example of the weakness of OWL, I would cite the following
> claim, which is dangerously and inexcusably misleading:
>
> DA> As I see it the OWL language attempts to find the sweet-spot
> > between expressive capability and decidability in the context
> > of the web. From my own little experiments, I'd suggest it
> > does quite a good job.
> The most serious misconception is the idea that decidability is
> a property of a language. That is totally misguided. The correct
> statement is that decidability, like solvability or unsolvability,
> is a property of a problem. There is an enormous number of
> problems associated with the WWW, with varying degress of
> solvability and with varying requirements for expressive power.
That's a weakness in my sloppy description, not in OWL. John's absolutely
right that decidability applies to the problem. It is the decidability of
problems when described using OWL that is the issue. But here the layering
is useful, catering for some degree of variation in solvability - there is a
subset of problems that may be expressed reasonably well using OWL DL, and
they will be decidable in finite time (I forget the details of the sums,
sorry). There is a superset of these problems that can be expressed using
OWL Full/RDF many of which are less tractable. Then there are problems that
simply can't be expressed in these languages at all, but could be expressed
using an arbitrary XML language. It is in the coverage of these sets in the
context of the web where the 'sweet-spot' point comes in. At present most
things we might want to describe are web resources, or things that can
relatively easily be described by reference to web resources (e.g. people
can be identified indirectly through their web pages and email address). The
web is an open world, but one which can be temporarily, locally closed for
many problems. Most bits of information take the form of simple descriptive
predicates (in the form of metadata), the main demand being able to join
these together with very simple inference. OWL covers this quite well.
> One example of the need for more expressive power than OWL is the
> query language, of which SQL is the most successful commercial
> example. The WHERE clause in SQL queries and constraints has
> the full expressive power of first-order logic, which goes far
> beyond the capabilities of OWL. Yet all the problems to which SQL
> queries and constraints are applied can be solved in polynomial
> time. By any measure of success, SQL certainly hit a "sweet spot",
> and the failure of OWL to match that sweetness is a serious mistake.
I'm not an expert, but I smell a rat here somewhere - Date and Fabian [1]
have covered ad nauseum the limitations of (implementations) of the SQL
language compared to the capabilities of Codd's relational model. I can
however respond from a pragmatic point of view, that the RDBMS
table-oriented approach is a poor match to the arbitrarily structured data
found in the web environment. The graph model provided by RDF is a
considerably better fit. (Incidentally, a simple query language, RDQL, was
recently submitted as a W3C note [2]).
> Another serious limitation of OWL is the kludgy syntax which was
> inherited from RDF, which is a very watered down version of Guha's
> original proposal, which he only intended as a preliminary step
> toward a more suitable language. Unfortunately, the absence of
> published guidelines for the RDF semantics let developers treat
> the RDF building blocks as a Lego kit that lead to an incoherent
> florescence of exotic forms that had no common semantics. Guha
> and Hayes eventually published a sound semantics, but only after
> many applications of RDF had been built without any semantics.
There are several syntaxes, including OWL's own abstract syntax which aims
to address some of the kludginess concerns. As the language is aimed
primarily for machine consumption, the ugliness of the RDF/XML concrete
syntax is neither here nor there.
> AA> The language is lacking many significant features of
> > relations, both formal and real. Among real, first of all,
> > the relation of cause and effect is fatally missing.
> > Such defects come from the approach used, purely
> > set-theoretical and formal logical, while any content-based
> > (world) ontology distinguishes internal and external relations,
> > avoiding their reduction to relational properties.'Being a
> > parent' is merely a relational property, while 'parentage'
> > (parenthood, but not the act or process of parenting) is
> > a relation of parent to child. Or, more generally, the
> > relational property of being a cause is just a monadic
> > reduction of causality, the relation of cause to effect.
>
> DA> Cause and effect can be modelled using OWL - this is
> > something found in OWL-S:
> >
> > http://www.daml.org/services/owl-s/1.0/owl-s.html
>
> Indeed, a relation named "cause" can be represented in OWL.
> But there is much more to the concept of causality than
> just a simple relation. For a brief outline of some
> of the issues, I suggest a paper I started to write,
> and which I hope to complete sometime:
>
> http://www.jfsowa.com/ontology/causal.htm
> Processes and Causality
Nice paper. I think the diagram at the end relating World-Model-Theory is
particularly pertinent, with the 'Approximation' part between World and
Model. The Web Ontology Language is aimed at solving practical problems, and
the World in question in the context of processes and causality is that
which encompasses problems like (real-world) planning and choreography of
web services. There certainly is more to "cause" than a simple relation, but
for many immediate problems that's a good enough approximation.
> DA> I can't actually see any part of that which couldn't be
> > represented fairly directly using RDF/OWL. As to the processing
> > of the material thus represented, maybe what you describe falls
> > outside of the capability of OWL's Description Logic - but
> > that doesn't mean such logic couldn't be layered on top.
>
> I have no quarrels with using a description logic as a part
> of a more complete system of logic and ontology. But I
> strongly disagree with the idea of starting with the bottom
> layers before any work has been done on the overall
> architecture. For a view of the kind of design that emerges
> from that approach, I recommend the following web site:
>
> http://www.winchestermysteryhouse.com/story.html
The difficulty here is that this is the *Web* Ontology Language, and as such
must begin from existing architecture - in particuar the Representational
State Transfer architecture as descibed by Fielding [3]. Some amount of
"Mystery House" is unavoidable, although there is a fair amount of
cooperation between the logicians of OWL and the technicians of the
Technical Architecture Group which has helped reduce the number of rooms
without doors.
> I agree with Abdoullaev that Aristotle's framework is an
> outstanding example of the kind of work that is needed
> for an upper ontology, and I believe that the Aristotelian
> foundation he recommends should be included in the MSO
> (Multi-Source Ontology) that is being proposed for the SUO:
>
> http://www.eis.com.cy/
I don't disagree, but the ontology and the language in which it is expressed
aren't necessarily one and the same.
> However, I also agree with Ayers' cautionary note:
>
> DA> Whether there is a single universal ontology is a moot point.
> > There's certainly a lot more that can be done in the field,
> > but I would suggest that we can learn a lot from applying
> > what we have already, approaching the problem from the
> > direction of what we do know, rather than that of what we don't.
>
> Where we disagree is whether OWL is a satisfactory example
> of what we know. The Aristotelian basis of the EIS ontology
> has been known for over two millennia. FOL has been known
> for over a century, and the SQL application of FOL has been
> in commercial use for the past 30 years. Another example is
> the Horn-clause subset of logic, which has been implemented
> in many commercial applications for almost as long as SQL.
>
> Although I agree that the possibility of a single universal
> ontology is "a moot point", there is a lot more that can be
> done to accommodate multiple ontologies. I have recommended
> the following paper as an outline of a richer, broader approach,
> which could be supported by something like the MSO and WebKB:
>
> http://www.jfsowa.com/pubs/signproc.htm
> Signs, Processes, and Language Games
Again, I don't disagree, in fact I think support for multiple ontologies is
essential. The extent to which OWL can facilitate this (given its expressive
limitations) remains to be seen. But I think it's a very good compromise for
the requirements of the web at this point in time. From my shallow
understanding I get the impression that the model of OWL (including URIs)
isn't that far from Pierce's semiotics that many of your developments of
those ideas couldn't be used alongside the language.
As a final note, I'd just like to reiterate that OWL isn't intended to
answer all demands for an ontological system, just to add some useful
support to the web. Take the web out of the equation, that's a different
scenario.
Cheers,
Danny.
[1] http://www.firstsql.com/dbdebunk/
[2] http://www.w3.org/Submission/RDQL/
[3] http://www.ics.uci.edu/~fielding/pubs/dissertation/top.htm