Perusing the inputs....

[Steven Willmott]

Dear all,

Reading through the contributions ahead of the meeting its interesting
to see quite a number of patterns emerging and also a range of
perspectives. At a high level there clearly seems to be consensus
on a number of issues:

* no single standard will emerge (possible this is not even desirable)
* almost all favour a notion for a "core" or "abstract" model
which is later extended to different specific uses (options for
what the core might look like vary of course - an ontology,
a protocol structure etc.).
* Although a number of people mention XML/RDF/OWL there also seems to
be a consensus that the core needs to be above the representation
level (and a mapping to concrete syntax is done afterwards).
* Gerard and co-authors points on a similar process.
* We need to be to an extent at least content neutral - to allow
users of the format to include domain specific statements within
the framework.

it is also clear that the inputs take different views. For example our
inputs were primarily inspired by capturing conversation (or possible
monologue) in a manner which is machine processable and could
be linked to appropriate semantics. Others (Dave R. et. al. Sanjay and
co-authors, Iyad, David H.) focused more on the meaning of argument, the
relationships between arguments etc. The format no doubt has to account
for both in some form (it is the link which would make it useful), so
one of the tasks tomorrow i think will be to situate all the elements
needed and the "options" for them.

There is also (it seems to me) a significant difference between the
requirements for an Agent-Agent format and a format for interchange
between Tools. I think it will take discussion to understand if we
can easily accomodate them both.

Gerard et. al.'s points about BBN standardisation are definitely welcome
and important. I think this experience is mirrored a great deal in
other communities. Interestingly though the experience was somewhat
different in the FIPA standardisation process of 96-onwards. There
the focus was really on agent-agent communication and this communication
was the core of any applications being built, rather than something
additional (exchange of data files between tools). It's hard to say if
this is salient difference but:

* building system with agents hosted by different parties wasn't
viable without consensus.
* the option of supporting multiple formats was available to but
but with much of the investment in an application going into the
communication itself the motivation to comply to "the" standard
seemed to be greater.

Still, although it didn't happen to FIPA-ACL, with KQML a number of
dialects certainly emerged (generally in each seperate large scale
project that made use of it).

I certainly agree that "one standard" is unlikely to happen and the
convergence to a common core would already be a good achievement. However
we should keep our eye on exactly how "variations" occur.
Since if they work in a controlled way they can be positive, if they
end up fragmenting the effort arbitrarily (there is not enough
common ground) then the effort may be wasted.

In any case - looking forward to interesting discussion!

thanks and best regards,
steve.