[
Lists Home |
Date Index |
Thread Index
]
> From: Peter J. Gale [mailto:thesemanticweb@yahoo.co.uk]
>
> --- Bill_de_hOra <dehora@eircom.net> wrote: >
> > I'd rather see a classification of web services data
> > based on speech act theory.
>
> That sounds promising to me, would you care to
> elaborate a little on how *you* think this might be
> applied.
I've involved in jsr-87 under Sun's jcp effort to produce an API based
on FIPA's Abstract Architecture (Java Agent Services or Jas). The
similarities between what we're specifying and how the
architecture/stack for web services has fallen out is remarkable. We
have a mandate to keep the agent's message fully devolved from the
message transport, ditto for an agent description, but really that's
just a bit more indirection thrown in. In one sense the web services
architecture is pretty what agent researchers and FIPA have been working
on for years (a parallel: it's a bit like SGML folk being bemused at the
success of XML). That's to say agent research and the web are converging
with regard to base architecture.
The key difference is what goes inside the body of a message. In the Jas
there is an abstract syntax tree for agent messages (a DOM for
communicative acts if you will); this is a container for stuff like RDF,
SL or KIF. The assumption has been that the Jas will be used for highly
structured agent messages (speech acts), but you could use this API
happily for web services and migrate towards communicative acts over
time as you need them. And SOAP doesn't care what you put in its body as
long as its XML.
Where web services lack is in the richness of service description and
discovery, the absence of a message structure, and semantics for both;
SOAP's only an envelope after all and I think the limitations of
XML+namespaces+schema in this regard are well documented. I've seen Clay
Shirky write that it's turtles all the way down with regard to have much
meaning you can automate, but amazing things can happen the more turtles
you pass by.
It's really a matter of finding the 80/20 strike between expressive
power and practical deployment for 1) service descriptions 2) service
messages. Neither WebService nor Agent standards strike that balance
today. In the meantime, there's some collaboration going on between FIPA
(www.fipa.org) and the DAML/web-ont effort. In the near term I'd hope
that future versions of WSDL or WS-Inspection would cherry pick the
useful bits from stuff like DAML-S, and that SOAP stays properly
abstracted from the transport below and the payload above it. But I
think standardizing on message semantics will takes years and years.
Bill de hÓra
|
