RE: [RTF] Behavior definition of Services - public discussion

>The bottom line: avoid phrasing the question in terms of equivalence, 
>instead phrase the question in terms of `have I heard of this name 
>before'?

I think this might work or not depending on how well defined is the domain
you are dealing with. This is also true for humans, of course, in the sense
that names are understood within a commonly shared domain (same culture,
same professional affiliation, etc.). But for machines it is particularly
important that the domain be well defined and limited (as various AI efforts
have showed repeatedly in the past).

If I talk about a WebService called "submitPurchaseOrder", and I am in a
very constrained domain (e.g. I am dealing with a business partner I have an
on going business relationship with) then I probably know exactly what that
means down to the smallest semantics details. But if I find a
"submitPurchaseOrder" service offered on a public UDDI repository by
somebody I never dealt with before, that I would not be that sure that the
semantics is the same as the one I am thinking of. In the extreme scenario,
I can think of submitPurchaseOrder being offered in a completely different
cultural context than the one I am familiar with, where the meaning of a
"purchase order" has nothing to do with what I have in mind.

This discussion is somewhat similar to the ones that went on at the time XML
was launched, about the idea that tags can give me information about the
meaning of the corresponding contents. I remember John Bosak responding to
that assumption by showing an XML document where all tags were written in
Japanese, and then asking the audience to make sense of the meaning of the
enclosed information.

Ugo

-----Original Message-----
From: Francis McCabe [mailto:fgm@fla.fujitsu.com]
Sent: Wednesday, July 03, 2002 9:18 AM
To: Champion, Mike
Cc: 'www-ws-arch@w3.org'
Subject: Re: [RTF] Behavior definition of Services - public discussion



It is formally provable that the kind of equivalence that you appear to 
be looking for is undecidable. (This is a corollary of the halting 
problem, and Goedels incompleteness theorem)

However, do not throw the baby out with the bath water. There are many 
things that you can do with semantic descriptions, indeed we MUST do 
with semantic ddescriptions.

If a client wishes to decide whether or not to use a web service, one 
might characterize this as an equivalence problem. However, a simpler, 
more realistic approach is `does it use language that I recognize?' 
(This is the ontology approach)

In effect, there is a graph of concepts that a web service provider 
publishes to describe his or her service. A client applies a matching 
test to that graph -- which might include getting references from other 
graphs -- to see if the graph is congruent with his desired service. If 
it is, then the client can be pretty sure that its OK (assuming that the 
provider's actual service is `faithful' to the description.

Such a graph match is not only decidable, but perfectly tractable too -- 
for most realistic situations. This is because a linguistic approach is 
fundamentally different to the `semantic' approach implied by the 
question: are two programs the same.

In human terms its like this: how do I know that when you use the word 
`foobar' you mean the same as I do when I use it? The answer is, that in 
a fundamental sense we don't know, but in practice the issue doesn't 
seem to come up too often. (Notice, however, that it DOES come up 
occasionally; but that humans resolve questions by linking new words and 
concepts with old concepts)

The bottom line: avoid phrasing the question in terms of equivalence, 
instead phrase the question in terms of `have I heard of this name 
before'?

I hope that this throws light instead of gas on the fire.

Frank McCabe

On Wednesday, July 3, 2002, at 05:31  AM, Champion, Mike wrote:

>
>
>
>> -----Original Message-----
>> From: Ugo Corda [mailto:UCorda@SeeBeyond.com]
>> Sent: Tuesday, July 02, 2002 8:17 PM
>> To: 'Damodaran, Suresh'; 'www-ws-arch@w3.org'
>> Subject: RE: [RTF] Behavior definition of Services - public discussion
>>
>
> +1   or +100 !!!
>
>> For me the bottom line is that semantics is a quicksands area
>> for computing
>> in general (having had direct experience of the AI disillusion of the
>> eighties) and for Web Services in particular. I would be very cautious
>> before introducing concepts like semantic equivalence that
>> could create
>> expectations well beyond what Web Services can actually deliver today.
>
>

Received on Wednesday, 3 July 2002 16:03:50 UTC