Building on RDF (was: Entailment / Datalog Complete, Sound, Terminating / English Reasoning)

At 16:43 23/05/03 -0400, Drew McDermott wrote:
>    At 22:12 22/05/03 -0400, Adrian Walker wrote:
>    >I was surprised to learn that RDF reasoning appears to be very 
> restricted,
>    >as in:
>    >
>    >  "...a more complex fact is expressed in RDF using a conjunction
>    > (logical-AND) of simple binary relationships. RDF does not provide 
> means
>    > to express negation (NOT) or disjunction (OR). The expressive power of
>    > RDF corresponds to the existential-conjunctive (EC) subset of first 
> order
>    > logic"   [http://www.w3.org/TR/rdf-concepts/]
>    >
>    >Surely, for RDF to function as useful service-finding glue in the 
> Semantic
>    >Web, it should at least support discovery of resources by being able to
>    >walk arbitrary depth first- and second-order hierarchies/ontologies ?
>
>    [Graham Klyne replied:]
>
>    I think the strength of RDF
>    here is that it's agnostic enough about reasoning methods to be used with
>    any or all of these -- it lets us exchange information, but doesn't 
> tell us
>    how to "think".
>
>This is a strange statement.  RDF (and its accessories, such as OWL)
>were specifically designed to impose strict limitations on what can be
>said and what kind of reasoning can be done.  The general philosophy
>was to make it impossible for an undecidable reasoning task to be
>posed.

Hmmm... that's not my understanding of RDF's development.
(Though I don't extend my assessment to OWL.)

Based on an accumulation of interactions over the past few years, my 
impression was that RDF was originally designed to be quite loose in 
constraints it applies.  This was recognized as problematic for some 
purposes, so we now have some formal semantics, but I perceive that in 
defining these there was no intent to define a particular style of 
reasoning to be used with RDF.


>Adrian is quite right that large groups of people who want to
>do things outside the RDF mold are forced to operate in a nonstandard
>or quasistandard world.  Check the recent traffic on www-ws@w3.org
>(the web-services mailing list) for examples of the contortions
>required to make DAML-S approximate legal RDF.

Quite so (until we have standards that build upon RDF).  And I think that's 
quite proper.  My point was that RDF *alone* does not attempt to solve all 
these problems, any more than (say) the IP protocol tells you how to 
implement an HTTP web server, even though (most) use of HTTP is built upon 
the IP substrate.  Similarly, it is my view that other standards (and 
non-standards) will be developed that use RDF as a substrate.


>    [Graham again (I've reordered these two snippets of his message):]
>
>    There is work in progress to address the other capabilities you
>    seek:  informally, there's CWM and a number of other software 
> developments
>    that use RDF.
>
>CWM does not just use RDF.  It _extends_ RDF.

I'd say it does both, in the same fashion that HTTP can be viewed as using 
and extending TCP/IP.

>...  To use that extension
>yourself, you have to use CWM; the official version of RDF is not to
>be extended in that way in the foreseeable future.  It's a lot like
>Microsoft keeping to itself certain features of the operating system
>API so that only Microsoft people can take advantage of them ;)

I think this final point confuses openness with standardization.  If one 
develops an extension to some standard and keeps it secret then that is 
pretty much by definition not a standard.  But if one defines an extension 
to a standard and builds a community consensus for its use, then that too 
can become a standard.

Note that I'm not fundamentally disagreeing with anything you say, just 
viewing it slightly differently.  I think there will probably be several 
standards that build upon RDF (e.g. OWL, query, rules are all in varying 
states of discussion), and will be no less interoperable for not being 
strictly limited to features defined by RDF (again cf. HTTP and 
TCP/IP).  OWL is such a work in progress.  The point of my original 
response to Adrian was to try and suggest not looking to RDF alone for all 
of the answers to the (quite reasonable) questions he raised.

#g


-------------------
Graham Klyne
<GK@NineByNine.org>
PGP: 0FAA 69FF C083 000B A2E9  A131 01B9 1C7A DBCA CB5E

Received on Sunday, 25 May 2003 05:16:20 UTC