Who fell of the fruit truck earlier this
evening?
From: Francis McCabe
[mailto:]
Sent: Tuesday, June 10, 2008 11:17
PM
To: Ken Laskey
Cc: soa-rm-ra
Subject: Re:
[soa-rm-ra] reaching closure on Action
In Cayenne ...
On Jun 10, 2008, at 7:31 PM, Ken Laskey wrote:
In orange...
On Jun 10, 2008, at 8:41 PM, Francis McCabe wrote:
In Magenta...
On Jun 10, 2008, at 2:52 PM, Ken Laskey wrote:
Then my responses must be in Lime.
Ken
On Jun 10, 2008, at 4:37 PM, Francis McCabe wrote:
My responses in Strawberry...
Frank
On Jun 10, 2008, at 11:21 AM, Ken Laskey wrote:
OK, let's see if we're getting somewhere?
Ken
P.S. This has to be resolved before we run out of distinct
colors.
On Jun 10, 2008, at 1:54 PM, Francis McCabe wrote:
On Jun 10, 2008, at 10:02 AM, Ken Laskey wrote:
I changed the color of a few lines below to red because these
start getting to my questions.
Now while I'd still like you (or someone else) to respond to
the questions below (timestamp 6:59 AM) as stated, I can infer
from your response that anything done by a speaker with the
purpose of applying intent is Action on the part of the speaker,
whether or not it is effective. At a minimum, it requires a
listener to have any chance to be effective. However, from
the context of the speaker, it is an action either way.
Yes.
If a listener hears what the speaker says, we have joint
action, in the form of CA.
Yes.
This still doesn't tell me how CA does something (i.e.
the initiating activity) to get something done, i.e the
RWE.
Also, what is the intent applied by the listener? Is it
simply to hear what a speaker says? At a minimum, it should
probably also be interpreting the message and deciding if it is
part of the Action Model of the service it is listening for.
Yes.
However, I think that this is where the counts-as relation
comes in to play. For the listener to understand that a CA is
supposed to result in a service action, the listener must apply that
relation. In effect, there is a shift going on from communication
actions to service actions.
You just can't say counts-as and sweep it under the
rug.
How specific do we need to get? I do not feel a need to
sweep stuff under the carpet.
As I think Rex noted, we are changing modes. The speaker
successfully sent the message because the listener received and was able
to accurately interpret it. I believe this requires more than would
be required of a traditional listener. Is this
correct?
Not really. But there is an
unknowable gap here: there is no way that an external observer can determine
if a listener has understood something accurately. We should not require
this.
In the nominally successful
(even less than ideal) situation, this is what happens. For the current
discussion, the unknowable gap should also be
unmentionable.
I agree. That implies
that the only semantics that we can be concerned about is so-called public
semantics. That in turn implies that anything that goes on as a result of a
service being acted on is either in the public description of the RWE or does
not happen -- from the perspective of an external observer.
So, any discussion of follow on
activities within a service is out of bounds. That pushes the focus back to the
Service Action.
The CA phase is done and the service action phase begins. The
message is one recognized by the service Action Model and the listener or
the someone/something for which listening is being done will then initiate
activity in response to the interpretation of receiving the message and
processing its payload.
I have heartburn with Rex's
view of events. This phase model may be helpful for many people; but this is
a realizing SOA concept not a EcoSystem concept. For the EcoSystem view, the
counts-as relation seems more
appropriate.
At this point, I don't care
which view we will have which pieces of the discussion, but I am concerned
that we know what we mean. If I'm playing cards and claim the remaining
tricks, this counts-as playing out the hand and winning the tricks
individually. If I leave a message with my daughter to give to my wife,
this does not count-as giving the message to my wife. I feel the same
kind of disconnect with the CA/SA gap. We need to say something about
bridging the gap.
One of the key
assumptions of the public semantics is that the medium of communication 'works'.
If the medium is not reliable, then that must be accounted for.
As far as the CA/SA gap, what
counts-as is doing is connecting two separate systems of 'logic'. I can see
rules that look like:
<msg><service
name="yourService"><action name="doSomething"><parameter>With
This</parameter></action>... counts-as yourService.doSomething(With
This)
(Fictitious example
obviously).
The activity results in RWE which are specific things that constitute
providing the described business functionality. There may be a
Process Model that defines intermediate public steps. Each of these
will require a successful CA that leads to initiating required action and
further RWE, and this continue until the processing is
complete.
absolutely.
You may cover it below, but I still ask whether the details of CA are
needed in the general discussion or only when the specifics of consumer
and service agents are required.
Let me try to answer your questions directly:
- If a speaker does something that s/he believe is an
application of intent but it fails, what has
happened?
Not much has happened; depending on where in the chain of
reasoning the failure occurred.
Does it matter if the application was insufficient
and would never work or if this instance of the application
failed but in general it would
work?
I suspect that you cannot distinguish these cases in
general.
I think something auditing (attempted) interactions could
identify where and how consistently it
failed.
If you succeed, do not forget to patent it
:)
I was thinking more of a Nobel Prize in Services.
Does it matter if there was no listener or if the
listener, say as a defensive maneuver, didn't
respond?
Without an active listener, there is no CA.
Does it matter if there is misunderstanding of the
message or the RWE and what results is not in line with the
original intent?
It seems to me that there are several questions here:
a. If the listener knows that he has not understood, then he
can detect a failure at the level of the CA.
b. If the listener believes that he has understood, but in fact
his understanding is incorrect, then all bets are off. A
higher-level observer might be able to determine that a
misunderstanding has occurred -- similar in a way to monitoring
obligation policies.
c. If the listener has understood appropriately, but the RWE is
not what the speaker intended, then the law of unintended
consequences seems to apply :) My guess is that this is not even
computable in the general case.
For the case where we have an instance failure, we'll
have a challenge explaining how someone doing the exact same
thing is sometimes applying intent and sometimes
not.
Applying intent is independent of the
result.
Glad this is resolved!
- If the application of intent
(1) the creation and sending of the message,
or
For the speaker, there are three understandings of action that
need to be understood:
a. He intends to send the message
b. He intends to play his role in the joint action of
communication
c. He intends to play his role in the action conveyed in the
message.
what action is conveyed by the
message?
I think that to properly resolve this would imply going into speech
act theory more deeply. One approach is to standardize messages a la
FIPA/KQML into performatives and propositional content. I for one would
not be averse to this, but I think that that may not fly very well with
the larger industry :)
No, we need a one line answer, not a formal representation.
From any of our variants of action, what action is conveyed by the
message? From my response above, I would say, "initiate
activity."
No. The action conveyed is
presumably one of the actions in the action model. A listener's response may
be to initiate some activity, but the CA conveys an action from the Action
model.
Let me be stronger: The
action conveyed *is* one of the actions in the action model. That is the
CA. We still need to bridge to the
SA.
As soon as you say
"the action conveyed in the message is one of the actions in the action model"
you *have* bridged the gap. The action model is a service model, not a
communications model.
(2) the sending (speaker) and receiving
(listener) of the message,
or
When both speaker and listener play their roles, they
both intend to take part in the join action of communication
(3) the sending (speaker), receiving
(listener), and activity initiation in response (who wanted
listening to occur)?
As above, except that now we engage the counts-as relation to
shift the CA to the service action.
We have to say something more concrete than counts-as. This
sounds like my daughter trying to get homework credit because at some
point she put the assignment in her
backpack.
Well, we do state somewhere that messages denote actions and
events. Is that sufficient? Almost certainly not in practice. Do we need
to do more? I do not know.
Consider the more I wrote above.
Again, a service action is also a joint action (in my opinion).
How does this relate to Action in the service's Action
Model?
I think that the action model characterizes the set of available
actions against the service. They are all joint actions in that other
participants are (normally) performing those actions. Something like the
actual performance of a Service Action is a Joint Action?
So the actual performance of a service action (i.e., the receiving
and interpreting of the incoming message and initiating activity in
response) is a Joint Action but the Action as described in the Action
Model is a potential.
I think that that
is accurate.
Whether I would say singular depends on our definition of singular.
But if not singular, we still need a compelling example (which seems
to have slipped away) of when is a SOA-relevant action not a joint
action.
I think that the
logic of our discussion suggests that all actions as conveyed by message
interaction are joint.
So I'm back to having lost the
bubble on what is a SOA-relevant single action. Or are all actions
joint?
This would not cause
me heartburn.
The proper role for the listener (service) is to respond as
though the service action had been
intended.
Then I get back to my question of whether we really need all the
redirection through the listener. The consumer invokes/directs
an action against a service by sending it a message and the service
responds with, in the ideal case, the stated RWE or, in a fault case,
some explanation of what went wrong. The joint action is between
the consumer and the service, possibly the service
provider.
The code that is executed by a listener is going to be pretty
similar however you architect it. one benefit of the CA/SA distinction
is that it helps to clarify why the different parts of that code belong
in different places: service action code vs MEP code.
Does this get too concrete? If there is an overall
contextual point to be made, then we should make it and make it clearly.
If not, we have enough other things to write
about.
I was giving an
example...
There is still the corner case where the listener is listening,
understands that a service action is intended, but is not willing to
participate in the service action. This is what we meant in the RM
by willingness.
This is just one of the possible responses that I allude to
above, and I don't think it is really a special
case.
I agree. It is a corner case but not an impossible case. What is
going on there is that the CA is *successful* but the SA is not.
In life, a response is sometimes what you expect and sometimes not.
Except for extreme cases, you just process what happens and carry
on. I tell my daughter to find another copy of her homework
assignment because it's really irrelevant if she didn't put it in her
backpack, she put it in but it fell out, or there really is a miniature
black hole that sucks it to another universe. The RWE of the ice
cream will not happen until the homework is
done.
She is lucky that ice cream
is an option ... :)
Getting late and I'm wandering.
Whichever one (or combination) you pick, please relate it
back to the Action Model.
Or is the service the listener?
Again, I was careful with the questions below because
answering those will get immediately to my pain points, and I hope
those of others too.
Ken
On Jun 10, 2008, at 12:04 PM, Francis McCabe wrote:
I think that we need to be clear about the distinction
between action and effect.
So, whether or not it is effective, action is the
application of intent .... etc.
The effect of performing an action is captured in the
RWE.
So, if a speaker says something, and noone was listening
then:
1. There was no communicative action
2. The RWE was minimal (it is never completely null, but
presumably no answer counts as much less of an effect than
intended).
For there to be a CA, there must have been some level of
engagement, and hence some level of RWE. But it is still
possible for that to be less than expected.
It is possible for a speaker to intend to engage in a CA;
without it happening. Both parties must intend for it to happen
-- hence the joint part of the joint action.
Frank
P.S. To Rex: this discussion is clearly important for our
own understanding. It is a separate question as to how much this
affects the spec.
On Jun 10, 2008, at 6:59 AM, Ken Laskey wrote:
I don't dispute the idea that for there to be
communications and *effective* application of intent, there
needs to be a speaker and a listener and they have to actively
participate and understand each other. The question is
how this relates to the thing(s) we call Action.
Let me try to capture this in a couple questions:
- If a speaker does something that s/he believe is an
application of intent but it fails, what has happened?
Does it matter if the application was insufficient and would
never work or if this instance of the application failed but
in general it would work? Does it matter if there was no
listener or if the listener, say as a defensive maneuver,
didn't respond? Does it matter if there is
misunderstanding of the message or the RWE and what results is
not in line with the original intent?
For the case where we have an instance failure, we'll
have a challenge explaining how someone doing the exact same
thing is sometimes applying intent and sometimes not.
- If the application of intent
(1) the creation and sending of the message,
or
(2) the sending (speaker) and receiving (listener)
of the message, or
(3) the sending (speaker), receiving (listener),
and activity initiation in response (who wanted listening to
occur)?
Whichever one (or combination) you pick, please relate it
back to the Action Model.
Ken
On Jun 10, 2008, at 9:03 AM, Francis McCabe wrote:
I mis-stated something last night:
2. As far as a speaker providing intent without a
listener that is true. But the CA is not *effective*
without a listener.
This is true; but on reflection, it is beside the
point.
The critical point is that there can be no act of
communication -- no application of intent -- unless there is
both a listener and speaker. So, if I talk to an empty room,
I am talking but not communicating.
On the other hand, even when we are communicating, it
does not mean that it is successful communication: I can
listen to you, and still fail to understand/act what you are
saying.
Frank
On Jun 9, 2008, at 7:43 PM, Laskey, Ken wrote:
I agree that the speaker and listener are actively
involved in communication. Our question is what is
their involvement. When does listening include the
activities that the speaker wishes to follow from not only
the listening but hearing and understanding.
In the Action Model, the service identifies the
messages it understands when it is a listener. It
does not guarantee it will do anything for any other
messages. It does not need a speaker present to
still understand those messages.
So the disconnects are:
1. There seems to need more than a speaker and a
listener to have a useful service interaction, i.e. the
listener has to commit to initiated activity. Our
discussion doesn't include that.
2. The speaker can apply intent without the listener
receiving the message or responding. There is action
on the part of the speaker but no interaction.
3. Action, or at least potential action, exists on
the part of the listener without any speaker. There
is potential for interaction, there are prescribed steps
in interaction, but there is no interaction until until
there is a speaker, an exchange of information, and an
understanding of that exchange.
Ken
P.S. We've been offlist for a while.
-----Original Message-----
From: Francis McCabe [mailto:]
Sent: Mon 6/9/2008 9:55 PM
To: Laskey, Ken
Subject: Re: [soa-rm-ra] reaching closure on
Action
Ken
There is a book that I recommend that you read. It is
called "Using
Language" by Herbert Clark. It is pretty informal but
gives an
excellent account of the concepts involved in human
communication, and
by extension computer communication.
Essentially, the bottom line is that both speaker and
listener are
actively involved, and that the communication has not
happened without
both participating. And he also addresses (not in the
same language)
the counts-as relationship.
As for denial of service, etc., I agree that
willingness is an
essential part of what is going on. hence the active
role of both
parties!
Frank
On Jun 9, 2008, at 6:08 PM, Laskey, Ken wrote:
> Arghh!!!
>
> In email , I'll buy that the speaker creates and
sends the message,
> but the listener only becomes aware that the
message exists. The
> speaker assumes whatever is listening will
initiate the activity of
> opening and reading. As with a denial of
service attack where it is
> appropriate to withhold/withdraw willingness,
whatever has the
> listener may not process the email if they
suspect embedded malware.
>
> Ken
>
>
> -----Original Message-----
> From: Francis McCabe [mailto:]
> Sent: Mon 6/9/2008 7:54 PM
> To: Laskey, Ken
> Subject: Re: [soa-rm-ra] reaching closure on
Action
>
> No, this does not get the join action
aspect.
>
> I admit that I thought some about the Patient in
a CA. I believe that
> the Patient in a CA is the medium of
communication. We jointly act on
> the email medium when we communicate by email.
It is a little tricky
> because there is some danger of infinite
regress:
>
> I act on an email to compose it and to push it
into the Internet. You
> act on the email to open it and read it. But
these actions are the
> actions of Speaking and Listening respectively.
The Joint CA is the
> combination of the two. In that world, we are
using the Internet
> (actually SMTP) as a means of communicating and
we act on it by
> sending and receiving messages (the messages
become the Instruments of
> our CAs).
>
>
> On Jun 9, 2008, at 3:59 PM, Ken Laskey
wrote:
>
> > Wouldn't
> >
> > CA -> Agent ->
Speaker=Initiator
> > CA -> Instrument -> Message[Do
ServiceActionPerformative]
> > CA -> Patient ->
Listener=Service
> > CA -> Verb -> CA_Performative
> >
> > where I assume CA_Performative is pass
message. I can't see having
> > two Agents and no Patient.
> >
> >
> > For a ServiceAction SA, we get
> >
> > SA -> Agent -> Initiator
> > SA -> Instrument -> CA
> > SA -> Patient -> Service
> > SA -> Verb ->
ServiceActionPerformative
> >
> > Is ServiceActionPerformative what I have
called Initiating Activity?
> >
> > Ken
> >
> > On Jun 9, 2008, at 6:48 PM, Francis McCabe
wrote:
> >
> >> There is nothing about intent that
denies join intent. A joint
> >> action necessarily implies joint intent
-- both speak and listener
> >> intend that there be a
communication.
> >>
> >> And yes, the communicative action
involves *both* the sender and
> >> the receiver.
> >>
> >> And no, the service action is *not*
singular: it is the actor
> >> acting on the acted.
> >>
> >> If we expand the ontology of action a
little bit:
> >>
> >> Action -> Agent
> >> Action -> Instrument
> >> Action -> Patient
> >> Action -> Verb
> >>
> >> where Agent is the entity performing
the action, Instrument is the
> >> tool with which the action is
performed, Patient is the target of
> >> the action and Verb is the action being
performed.
> >>
> >> Then, for a CA, we get
> >>
> >> CA -> Agent ->
[Speaker=Initiator, Listener=Service]
> >> CA -> Instrument -> Message[Do
ServiceActionPerformative]
> >> CA -> Patient -> None
> >> CA -> Verb ->
CA_Performative
> >>
> >> For a ServiceAction SA, we get
> >>
> >> SA -> Agent -> Initiator
> >> SA -> Instrument -> CA
> >> SA -> Patient -> Service
> >> SA -> Verb ->
ServiceActionPerformative
> >>
> >> The counts-as relation has to map the
two actions, probably as here
> >> by linking the Instrument of the CA to
different parts of the SA,
> >> as well as some implied linking between
Listener/Service etc..
> >>
> >> This is probably a whole lot more
detailed than we should go into
> >> in the spec; but if *we* need to to
convince ourselves, so be it :)
> >>
> >> Frank
> >>
> >> On Jun 9, 2008, at 2:24 PM, Ken Laskey
wrote:
> >>
> >>> I still have the question of
whether Action as the application of
> >>> intent requires a receipt of that
intent. This is back to the
> >>> singular vs. communicative nature
of the Action.
> >>>
> >>> If the message is the Action, then
the Action has to be both the
> >>> sending AND receiving of the
message in order for it to be a
> >>> communicative action. Intent
sounds like one way; it is my
> >>> motivation and the action is my
acting on that motivation, but
> >>> that is all separate from the
receiver.
> >>>
> >>> The Service Action, OTOH, is
singular on the side of the service/
> >>> receiver. The service Action
Model delineates what messages need
> >>> to be sent in order for certain
"activities" to be carried out,
> >>> leading to certain RWE. The
Action Model exists independent of a
> >>> speaker.
> >>>
> >>> The Communicative Action CANNOT
count-as the Service Action
> >>> because one requires a speaker and
the other does not.
> >>>
> >>> Ken
> >>>
> >>> On Jun 9, 2008, at 1:13 PM, Francis
McCabe wrote:
> >>>
> >>>> I believe that there are 4
'concepts' of action involved:
> >>>>
> >>>> 1. The abstract sense of
Action. Application of intent etc.
> >>>> 2. Abstract Joint Action (which
is either a subclass of Action or
> >>>> a particular use of Action; not
sure of the right relationship).
> >>>> 3. Communicative Action (which
is a subclass of Abstract Joint
> >>>> Action)
> >>>> 4. Service Action which is an
Action against a Service (which is
> >>>> described in the Action Model
and the Process Model)
> >>>>
> >>>> 3. and 4. are connected via the
counts-as relationship:
> >>>>
> >>>> A valid Communicative Action
counts as a Service Action
> >>>>
> >>>> At some level, all of these
should be introduced and explained in
> >>>> Section 3.
> >>>>
> >>>> Frank
> >>>>
> >>>>
> >>>>
> >>>> On Jun 6, 2008, at 12:02 PM,
Ken Laskey wrote:
> >>>>
> >>>>> Dear Fellow
Explorers,
> >>>>>
> >>>>> We've had some very
stimulating discussions over the past few
> >>>>> weeks but I feel there are
other things caught in limbo until we
> >>>>> reach some consensus.
I don't think we are plagued by major
> >>>>> disagreements but rather
the different facets of complexity for
> >>>>> the range of things we want
to capture and make understandable
> >>>>> to a wider audience.
> >>>>>
> >>>>> So I think we need a plan
for how to proceed. The elements of
> >>>>> such a plan would
cover
> >>>>> 1. capturing the different
facets;
> >>>>> 2. capturing where in the
document these facets currently live;
> >>>>> 3. work a consistent
understanding that covers all the facets.
> >>>>>
> >>>>> Unfortunately, this is not
an 80-20 situation because a standard
> >>>>> that only covers 80% of the
scope is looking for trouble.
> >>>>>
> >>>>> Now I would suggest an
extended call (all day?) but I realize we
> >>>>> are all busy and that may
not be feasible. What's more is it
> >>>>> may not be productive
unless we have all the background material
> >>>>> together going in.
> >>>>>
> >>>>> As a precursor to an
extended meeting (or even a regular
> >>>>> meeting), is it possible
for us to have a short list of
> >>>>> questions and for the
author of each section to satisfy items 1
> >>>>> and 2 above through the
answers? Would that be enough to help
> >>>>> structure a productive (and
hopefully not too long) call?
> >>>>>
> >>>>> I haven't yet considered
the questions, but figured I'd float
> >>>>> the idea and see if someone
came up with something better.
> >>>>>
> >>>>> Ken
> >>>>>
> >>>>>
>
>>>>>
-----------------------------------------------------------------------------
Ken Laskey
MITRE Corporation, M/S H305 phone: 703-983-7934
7515 Colshire Drive
fax: 703-983-1379
McLean VA 22102-7508
