Re: question about normative status of entailment rules

From: Brian McBride <bwm@hplb.hpl.hp.com>
Subject: Re: question about normative status of entailment rules
Date: Mon, 29 Sep 2003 14:52:17 +0100

> 
> 
> Peter F. Patel-Schneider wrote:
> > I noticed that the entailment rules have changed status from informative to
> > normative.
> 
> Yes.
> 
>    I view this as a serious mistake, and will be objecting to this
> > change if it survives.
> 
> Thank you for the heads up Peter.  As time is short, RDFCore is hoping 
> to decide on a second last call this week, could you provide some 
> substantantive argument in support of your view.

[...]

Here are some initial arguments.  I expect to revise and expand these
arguments when it comes time to create something formal.


1/ There should be a single normative definition of the meaning of RDF.
Making both the model theory and the entailment rules normative means that
divergent RDF implementations are possible, some based on the model theory
and some based on the entailment rules, with both groups claiming to be
correct.

2/ It is much more likely that the entailment rules have internal problems
than for the model theory to have internal problems.

3/ The entailment rules do not form a complete characterization of
entailment in RDF.  They are thus unsuitable for use as a normative
characterization of entailment.  In particular, the entailment rules for
RDFS are completely only for rdfs-consistent graphs.  Implementors are thus
likely to either ignore the incompleteness or to implement consistency
checkers that are incorrect.


Peter F. Patel-Schneider

Received on Monday, 29 September 2003 11:55:07 UTC