Re: A plea for peace. was: RE: DAML+OIL (March 2001) released: a correction

From: "Jonathan Borden" <jborden@mediaone.net>
Subject: Re: A plea for peace. was: RE: DAML+OIL (March 2001) released: a  correction
Date: Thu, 5 Apr 2001 16:37:23 -0400

> Peter F. Patel-Schneider wrote:
> 
> > From: "Jonathan Borden" <jborden@mediaone.net>
> >
> ...
> >
> > > I suppose I've always seen as one of the benefits of RDF's triple model
> the
> > > very fact that it maps so easily onto a relational table -- and admit
> that I
> > > assumed this abstract syntax would in some sense inherit the formalism
> of
> > > the underlying database (e.g. this very relational model you mention).
> If
> > > you say this _isn't_ the case then I certainly agree things need to be
> > > fixed, it just seems as though it shouldn't be that hard to do.
> >
> > I would be very interested in hearing about the details of this easy
> > mapping.  (Yes, you should consider me to be very skeptical about this.)
> > I see a number of mismatches between the RDB model and the RDF model,
> > including open-world versus closed-world, finite versus infinite domains,
> > notions of identity, how to handle URIs and the things they refer to,
> > reification, transitivity, inference of types, typing (particularly
> > subtyping), and domain and range.
> >
> 
> Part of the issue is that you are mixing up RDF Schema (e.g. domain, range)
> with RDF... 

OK, so you are claiming that transitivity, type inference, subtyping, and
domain and range are not RDF.  That still leaves open-world, infinite
domains, identity, URIs, reification, containers (especially alternative),
distributive referents, and URI patterns.

> For RDF alone:
> 
> DEFINE TABLE triples AS
>     predicate : URI
>     subject : URI
>     object : URI
> 
> -- note that RDF literals can be encoded as "an example": data:text/plain,an
> example so that all objects can be represented as URIs, literals using the
> "data:" scheme.
> 
> -- a URI is a string having the syntax described in RFC 2396 (the EBNF isn't
> quite perfect but close enough)
> 
> Let's start with this alone, and add concepts only as absolutely needed.
> 
> -Jonathan

The mapping above handles none of the above, not even URIs --- not all
strings are valid URIs, which you indicate above, but the RDB schema doesn't
capture that; and string ordering is not a valid operation.

Peter F. Patel-Schneider 
Bell Labs Research

Received on Thursday, 5 April 2001 17:16:09 UTC