RDF and web services

> -----Original Message-----
> From: kifer@cs.sunysb.edu [mailto:kifer@cs.sunysb.edu]
> Sent: 25 May 2002 00:35
> To: Assini, Pasqualino
> Subject: Re: Taking an axe RDF in XML? (no thank you) 


> Hi, Titto,
> thanks for pointing to the URL for your system.
> Unfortunately it crashes (after a very long wait):

Sorry about that.

I think that someone was working on the server :-(

In case you, or anyone else, want to try again (avoiding problems with long
URLs):

- start from http://nesstar2.essex.ac.uk

- click on the Object Browser link at the bottom (the Frames version is
better).


> However, I downloaded your report on NEOOM and browsed through it.
> It looks like a neat system, BUT it corroborates what I 
> conjectured before:
> You are adding objects to the web (and you say as much). The 
> use of RDF is
> not essential here (you could have stayed with XMI).

Yes, I could have (indeed we start by modelling our staff in UML, save it in
XMI and then generate from it RDF, Java code. etc.)

But XMI is much too verbose, was not really made to represent instances
(object state), there are few tools and I would made it much harder to
integrate our system with the semantic web.

All in all I much prefer to use UML/XMI for the modelling and RDF as the
interchange format.


 
> I don't think this is a convincing use of RDF. Without a 
> larger rule-based
> language there will always be this question of why RDF and 
> not something
> else.

This question would remain even if rules where there.

Why RDF and not anyone of the existing ontology/logic language (clips, kif
etc.)?



> Having methods and their invocations as objects is a good 
> idea and a good
> argument in favor of RDF over WSDL. But I don't think this is 
> sufficient to
> convince anybody to switch.

As a matter of fact nobody has for the moment :-)

But I look forward at the RDF model of WSDL ...



> There is another potential barrier to the use RDF. RDF 
> triples that relate
> to the same object can be scattered throughout the same 
> document. This is
> not a problem is you are planning to process RDF using a declarative
> rule-based language. But if you do it in Java then you have 
> to collect all
> triples that relate to the same resource. In contrast, in WSDL things
> related to the same interface are concentrated in one place. 
> This is easier
> to deal with by procedural languages.

Even in RDF you can hold all the authorative information about an object
interface is a single place (that's what we do).

Going from RDF to Java is also easy.

In our sytem we convert back and forth from RDF to Java and use Java objects
for the actual processing.



> However, if RDF were part of a larger rule-based language then 
> one could deploy things much more easily.

Rules are great for the semantic web, so is Google for the Web.

But I would say that the Web was useful even before Google was available.

Any half-decent architecture has to be layered.

1 - The lowest layer of the semantic web is the humble RDF statement.

2 - Most application, like ours, will more likely think in terms of objects
than single properties. They will process and publish object descriptions
rather than arbitrary collections of statements.

3 - On top of this you can define as many advanced services as you like,
including fancy reasoning systems.



> Regarding NLP, I think it is good as an interface, but 
> because it is so
> imprecise I don't believe it can (or should) be part of a 
> specification
> language.

You are probably right (it is a pity though).
 
> 	regards
> 	  --michael  
> 

all the best


     titto

-------------------------
Pasqualino "Titto" Assini - Nesstar Ltd
John Tabor Building - University of Essex
Colchester, Essex  - CO4 3SQ  - United Kingdom
email: titto@nesstar.com personal email: titto@kamus.it

Received on Friday, 31 May 2002 14:23:11 UTC