W3C home > Mailing lists > Public > www-rdf-interest@w3.org > December 1999

[www-rdf-interest] <none>

From: Sean Luke <seanl@cs.umd.edu>
Date: Thu, 23 Dec 1999 18:25:01 -0500 (EST)
To: www-rdf-interest@w3.org
cc: jhendler@darpa.mil, heflin@cs.umd.edu
Message-ID: <Pine.SO4.4.05.9912231728450.11252-100000@jifsan.cs.umd.edu>
Sankar Virdhagriswaran <sv@crystaliz.com> wrote:

> Why do you think 'inferencing' is important. Don't we go a long way by
> providing support for simple querying over well marked up content
> where the meta-data is described in RDF

Hi, Sankar.  Consider the current special-case oddities in RDF right now:
subProperty, subClass, containers with infinite relation sets, aboutEach,
etc.  These are all manifestations of custom semantics to fill certain
inferential deficiencies in the language.  As the language progresses, my
money is on this continuing.  I'm guessing it won't be 6 months after the
spec is finalized that a strong push is made to provide for a number of
specific transitive closure operators, transfers-through features, inverse
relations, etc.  If it's like other similar languages in the past, I'd
guess that at some point there will be enough custom-purpose semantics in
the language that most everyone building an implementation of a query
system for RDF will have just done it using a general inference engine of
some sort, mapping all the custom warts into specific general-purpose
rules.

So why not just dump the special-case stuff and go with simple
(simplistic) general-purpose inferential semantics to begin with?  No
doubt some people will clamor for even stronger semantics, for which
Sergey's first comment below might be then applicable.  But the simple
inferences will cover the _very_ large majority of needs.

I don't have any problem with defining a very basic language consisting of
n-ary relations, resources, and types, with no semantics at all, then
creating a "Level 2" version of the language with a basic general
semantics engine which various schema may opt for if they so choose.  But
lots of little special-purpose, finite inferential semantics like we have
right now, is _not_ a good way to do this IMHO.  It's a recipe for
constant change of the specification as users' needs grow rapidly.


Sergey Melnik <melnik@DB.Stanford.EDU> wrote:

> To remind: you can built *any* languages on top of RDF, including an
> inference language.

Hi, Sergey.

Syntactically.  But then those languages are no longer RDF.  Inferences
require a-priori semantics that are not part of RDF.  And anyway, taking
this worldview would only encourage the splintering of efforts to create
at least some _basic_ semantics, which are not only useful, but, I'd
argue, very soon will be necessary.

> But definitely, it'd be nice to have an inf. lang. everybody agrees
> on.

We would need something that provides low computational complexity, fair
generalization, and possibly the ability to be "optional" or "limited" in
such a way that it cannot be easily abused to create too dense a web of
inappropriate inferential interconnectedness (this third item, actually,
SHOE does not currently do and is a general research area for us).

Sean
Received on Thursday, 23 December 1999 18:25:03 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:51:42 GMT