Re: DAMl "Thing" should be Top, Universal class - including concrete types

On February 1, Tim Berners-Lee writes:
> 
> ----- Original Message -----
> From: "Ian Horrocks" <horrocks@cs.man.ac.uk>
> To: "Tim Berners-Lee" <timbl@w3.org>
> Cc: "RDF Logic list" <www-rdf-logic@w3.org>
> Sent: Tuesday, January 30, 2001 1:26 PM
> Subject: Re: DAMl "Thing" should be Top, Universal class - including
> concrete types
> 
> 
> > Tim,
> >
> > I agree with a lot of what you say - it would clearly be "nicer" not
> > to separate abstract and concrete domains. As I understand it, the
> > idea behind daml+oil is that it provides clear semantics and potential
> > implementation paths (e.g., for reasoning services) for some
> > restricted subset of what can be written in RDF(S). If you want the
> > semantics and the services, the cost is the restriction.
> 
> Why?

The main problem is providing any kind of reasoning services. The idea
of the separation comes from work on combinations of modal logics. If
we simply take the union of two logics, this almost always leads to a
complexity explosion and requires the design of completely new
reasoning procedures: see the work of Marx [1] for example. An
alternative is to use a Fusion of the logics in which modalities
(properties) are used to "isolate" the logics, one from the
other. This has the advantage that the fusion generally avoids the
complexity explosion and allows for reasoning to be reduced to
reasoning in the component logics; so if we have reasoners for the two
components we immediately have a reasoner for the fusion (see work by
Wolter [2] and Baader [3]).

> > The proposal
> > simply suggests a way to extend daml+oil with (a restricted form of)
> > concrete domains while still retaining the above properties.
> 
> However, it loses the ability to be a general unconstraining
> langauge for unifying a very wide range of systems present and future.
> This is the requirement of the semantic web
> 
> It is not a requirement to give an "implemention path" for a specific
> reasoning system.  We are not designing a reasoner. We are making
> a universal language which will allow the expression of information
> from many difefrent systems. When a given system has limited descriptive
> power, then its input and output will be limited to a subset of the
> language.

As I understand it, we are working bottom up by defining a logical
language (actually, an ontology language) that has some desirable
properties, which will meet the requirements of many users, and which
can be the basis for future extensions. The final result of such
extensions may be a UWL, if such a thing is possible. In the meantime,
users are free to exceed the bounds of the language if they don't care
about the properties.

You seem to suggest that we are or should be working top down by
defining a UWL and then perhaps specifying certain subsets of this
language that have useful properties. 

We obviously need clarification as to which of these it is. I have to
say that I thought it was pretty clear that we were doing the first:
see my earlier email with the extract from the minutes of the
Washington meeting.

Ian

[1] http://turing.wins.uva.nl/~marx/papers.html

[2] http://www.informatik.uni-leipzig.de/~wolter/

[3] http://www-lti.informatik.rwth-aachen.de/ti/baader-en.html

Received on Thursday, 1 February 2001 17:26:52 UTC