Re: SEM: Reaching consensus (was Re: SEM: "natural" entailments)

I should perhaps have indicated that my extended flame to Ian about 
Webont being in the Vale of Death and so on was of course a highly 
personal opinion, and was not meant to be an attempt to influence the 
direction of this WG or to disrupt any consensus that has been 
achieved. (If I ever take myself that seriously then I will simply 
leave the WG.) I think that the technical part of this discussion - 
the nature of the relationship between OWL syntax and FOL - is in 
fact directly relevant to the WG's business (it certainly is to what 
I am trying to do) so I hope it will be OK to continue that.

Concerning the technical points, and in a spirit of 
consensus-movement, I hope that Ian and Peter will not mind if I try 
to summarize a long conversation we had today, and make a suggestion. 
To cut to the chase, read the last part first.

1. There are no doubt several details in the accounts of weak and 
fast OWL/RDF which may need to be tweaked, but Peter and I are 
optimistic that they can be fixed and that we are converging on 
something useable; Ian is less sure that we will ever get it right 
(or perhaps better, ever be able to know if we have got it right) but 
he has no deep theoretical objections.

2. Ian and Peter however disagree much more strongly with the idea of 
allowing large OWL into the mix. I confess that I did rather spring 
this on them, but it seemed to me to be a simple and harmless idea, 
and I was (and still am) genuinely surprised by the extent and depth 
of their dislike for what might called the RDF style of semantics (in 
which one can have classes of other classes and so on, and in 
particular non-well-founded classes which contain themselves, such as 
rdfs:Class.) That is, not only do they not want to use it themselves, 
they resist any such leakage of this RDF style into OWL. I, on the 
other hand, feel rather strongly in the other direction, ie that the 
restricted nature of the 'layered' fast-OWL universes is an 
unnecessary and limiting constraint on practical ontology design, and 
that one of the few attractive features of RDF is precisely this 
freedom.

3. The technical semantic issues that Peter raises about large OWL 
are basically an expression of disbelief that the large-OWL semantic 
framework can possibly be made coherent, or at any rate that to show 
that it can be is a major research effort and not something that we 
should be engaged in. Coming to this from several years working on 
the CL/ISO-KIF work, I am sure that there are no fundamental problems 
with this approach and that the basic work has already been done, but 
I will admit that there is much less of it published than Ian would 
be happy with. I am working on a document that will attempt to 
provide a convincing case for the internal coherence of the model 
theory, basically by sketching a version of Herbrand's theorem for 
large-OWL. (For non-logicians: this is a very general technique for 
showing how to construct interpretations from pieces of syntax. This 
does several useful jobs at once. It shows that interpretations 
exist, but also it provides the basic connection between 
interpretations and possible syntactic inference rules. It is in any 
case of independent interest and will I think be of utility in 
thinking about OWL more generally.)

4. One particular concern raised by Peter is the freedom that one has 
in large OWL to use the 'logical' vocabulary such as rdf:type and 
rdfs:subClassOf inside OWL constructs such as restrictions; for 
example, you can define the class of all classes that have more than 
57 subclasses. Peter is worried that allowing such talk might 
inevitably lead to incoherence. I am confident that it will not, but 
concede that it will be difficult to persuade Peter of this. One 
possible option which we discussed, and I promised to write up, is to 
simply disallow such combinations by a syntactic rule. I no longer 
like this idea, as such constructions have obvious uses. For example, 
it makes perfect sense to intersect a class with a cardinality 
restriction on rdf:type, as a way to identify all the categories in a 
database which have less than some number of instances. It is easy to 
think of plenty of other examples of this kind. I therefore do not 
support this idea, which in any case will probably not suffice to 
alleviate the very fundamental concerns that Ian brings to any 
large-OWL-like proposal. I will write it up if people think it is 
worth considering, however.

5. In the interests of achieving consensus and given the limited time 
available and the very strongly held opinions, the simplest option 
would be to simply forget about large OWL, and only consider weak OWL 
(the option which allows unrestricted use of the OWL vocabulary in 
RDF, but only sanctions relatively weak conclusions about OWL 
restrictions) and fast OWL (the option which uses RDF simply as a 
carrier notation for the abstract OWL syntax). These seem to be the 
least contentious options, and we could then restrict the debate to 
whether to allow only one of them or to allow both. I think that this 
is the option which Ian, Peter and I could most easily agree to stop 
arguing about. IMO it misses an opportunity for a useful extension to 
RDF, but if I am right then it would be easy to add it later (users 
just have to take less trouble, basically, and are allowed to draw 
more conclusions, so I doubt it they will complain.)  And in the 
meantime, we can move forward. If the WG thinks this is wise, I could 
produce a re-draft of the document based on this decision in a 
relatively short time, which would then only require tweaking. I 
think that the passionate defenses of intellectual territory could 
them be reduced to a kind of low grumbling noise which could be 
fairly quickly ignored.

Pat
-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes

Received on Tuesday, 24 September 2002 23:09:55 UTC