Re: Fragments - specific proposal

From: Jim Hendler <hendler@cs.rpi.edu>
Subject: Re: Fragments - specific proposal
Date: Mon, 10 Dec 2007 09:25:39 -0500

> I did indeed refer to that page - the language features I outlined  
> clearly can fall into polynomial with certain restrictions (in fact,  
> they are all doable in various datalog and Horn subset)

Actually not, at least not for Datalog or safe Horn rules, because you
can have contradictions.  Note that RDFS (even without datatypes) is
itself not in Datalog or safe Horn for the same reason.

> - those  
> restrictions are what would need design - in the sense that the  
> unrestricted use of the feature set (including the ability to  
> redefine the features in RDF) would clearly be undecidable - 

Hmm.  I wonder about this.  The language you specify is indeed quite
limited, so it may not be possible to mess with the language from within
itself too much.

> the  
> question is can the restrictions be specified simply enough.
> 
> btw, let me be clear to the some of the assembled, you talk about  
> semantics as if there is only one such thing - I point out that the  
> world include programming language semantics, operationally defined  
> semantics, database semantics, axiomatic semantics and many other  
> things beyond model theory.  I am perfectly content for the langauge  
> to only specify an operational semantics normatively, and then let  
> researchers determine the formal model-theoretic semantics.  So when  
> you or Peter says the language is not defined, you are wrong, it is  
> carefully designed against the operational semantics clearly  
> specified in the OWL 1.0 documents.  

What operational semantics?  

On http://www.w3.org/2007/OWL/wiki/Fragments you say 

********************************************************

Requirements
   [...]
   2. [RDFS 3.0] can have a clean operational semantics defined via a
     axiomatic specification.

********************************************************

There is no normative axiomatic specification for any species of OWL,
I don't know how you would generate this operational specification.  In
any case, an axiomatic specification is different from an operational
semantics. 

> Since I defined it as an OWL  
> Full subset, I would point out that it is also formally defined  
> through the OWL Full (RDFS) semantics.

Well, it is possible to produce a subset of OWL Full based on a subset
of the OWL Full vocabulary, V, in several ways:

1/ The language of the fragment could be either a) all RDF graphs, or b)
   RDF graphs that do not use the OWL Full vocabulary not in V.

2/ The semantics of the fragment could be either a) the OWL Full
   semantics in total, or b) include also interpretations that do not
   obey semantic conditions that involve OWL Full vocabulary not in V.

Which of these do you mean (if any)?

(Note that defining the semantics of fragments is simpler in the
direct model-theoretic semantics, particularly because the syntax of the
language does not generate objects in the domain of discourse.)

> To reach consensus, I've expressed a willingness to see us work on a  
> more formal semantics for a restricted subset of this.  FWIW, I'd  
> personally be happier seeing us just define this as a named subset of  
> OWL Full w/o proof of complexity or model-theory, it was my feeling  
> from the WG f2f minutes and parts I heard, that some in the group  
> could not live with it.

But then again I ask what would the meaning of the constructs be?  OWL
only has a model-theoretic semantics.

> But just so we're totally clear - the RDFS 3.0 that I proposed is a  
> fully defined language with all the semantics it needs (from the OWL  
> 1.0 and RDF documents) - so please don't say I haven't defined the  
> language.

I don't think that you have fully defined the language.

>    -JH

Peter F. Patel-Schneider
Bell Labs Research

Received on Monday, 10 December 2007 15:37:54 UTC