Re: Semantics of WSDL vs. semantics of service

Hi Bijan, see inline. 8-)
Please note that I'm writing this as my personal opinion, not as the
proposed chair of the SA-WSDL WG.

On Mon, 2006-03-20 at 08:09 -0500, Bijan Parsia wrote:
> I went back and looked and the Scope section is a bit broken (as 
> evidenced by Steve's quote):
> 
> """The Semantic Annotations for WSDL Working Group is chartered to 
> define one or more properties of WSDL 2.0 components to point to 
> additional semantics to concepts represented by those components, e.g. 
> interface, operation, endpoint. Additionally, the Working Group may 
> define annotations to the schema structure to point to external 
> semantics.""""
> 
> "point to additional semantics to concepts" just doesn't parse. 
> Additional semantics *for* concepts represented?

Yes, additional semantics of/for concepts represented.

> I confess to hating the term "external semantics". C'mon.

"externally described"?

> > I for one see some use cases where embedding the annotations would be
> > useful, and I can see at least two ways of embedding them: put a whole
> > semantic description
> 
> I go back to a fight I had in SWSL. What's a *non* semantic description?

Every description expresses some semantics, true, but I believe that
traditionally, when you say "semantic something", it means "semantics
that can, to some degree, be described in a machine-readable form" with
ontologies, axioms etc.

> >  document somewhere in the WSDL document (like we
> > put schemas in the <types> section) and then the annotations will point
> > into the document; or put the full annotations themselves on the spot,
> > instead of referring to them.
> 
> How are the "semantics" to be realized? Via some sort of statement 
> (e.g., axioms in some formalism). So let's say I have a set of concept 
> and property names, but no further axiomization. And I want to say of 
> some operation that is has at least one P relation to a C. Now since 
> there *is* no other axiom, this characterized the terms entirely (thus 
> far). May I inline that? It seems like I should be able to. 
> Alternatively, I could require that I always coin a name for these 
> intermediate expressions (but why?).

I personally think you should be able to inline this, but WSDL-S doesn't
do that so we'll have to start by opening an issue in the WG. 8-)

> (Note that originally I interpreted the discussion as requiring *all 
> parts* of the annotation to be outside the WSDL document, a la OWL-S. 
> There are reasonable reasons for doing this (including supporting third 
> party and alternative annotations seamlessly. Technically, I guess this 
> is not ruled out by the current charter since the concrete syntax of 
> the component properties could be or be required to be in a separate 
> document.

Apart from the RDF mapping, I don't expect the result to have syntax
that will be separable from the WSDL syntax - in WSDL-S you can put
modelReference on an operation, not somewhere outside. Both OWL-S and
WSMO have semantic annotations in their own language, and I don't quite
think that we should try to externalize WSDL-S above "WSDL-based
annotations".

> > While the second option can be seen as out of scope as defined in the
> > charter, at least the first option should be available to us. 8-)
> 
> I find the Out of Scope more disturbing:
> 
> """discuss expression of Web services constraints and capabilities, 
> including precondition and effect."""
> 
> Why? And how can this be at all narrowed? I mean, from the scope, " 
> could have different meanings: calculation of tax on a product, 
> calculation of income tax, etc. " Aren't these expressions of 
> capabilities? (I recognize that constraints and capabilties are a term 
> of art standing for "policy", but still.)

There seems to be little agreement about the modeling of preconditions
and effects, in fact there may be insufficient agreement about the terms
themselves - WSMO distinguishes preconditions, postconditions,
assumptions and effects, so the WSDL-S model is restrictive for WSMO.

However, at least in WSMO, one can give a name to the thing that
contains precondition and effect descriptions, and this name can go into
modelReference on operation, so we don't need explicit precondition and
effect attributes on operations.

In other words, the description of preconditions and effects can be
hidden in the model reference and the SA-WSDL spec will be as useful as
WSDL-S, I believe.

Hope it helps,

Jacek

Received on Monday, 20 March 2006 14:25:39 UTC