clarification

CSGnet Archive CSG1992
1

[From: Bruce Nevin (Fri 920710 13:42:31)]

(Bill Powers (920709.2000) ) --

Neuropeptides: rightly or wrongly, I am using this term in a generic
sense. I posted some stuff about work of Candace Pert and others a
while back, but this is certainly not my field.

Even so, I apparently included too much detail so that you missed the
main point, which is that control systems of one order (cells),
controlling for coordination, cooperation, or simply for a more stable
and predictable environment (which happens to include numerous of their
fellows), together can constitute what we recognize as control systems
of another order (ECSs within a complex living control system).

Each constituent cell of an ECS is in itself blind to the functioning of
the ECS. Nor does it in any direct sense control for helping to
constitute an ECS. Nor does it have any means for perceiving the ECS of
which it is a constituent.

Now let me restate the question about one cell controlling another, if I
can. Consider two neural fiber cells connected by a synapse. Does the
state of one control the state of the other, with respect to neural
impulses (modulo the value of the synapse)?

Consider an ECS with a sensory input function I, a reference input
function R, a comparator C, and an output function O. In this simple
and perhaps merely schematic example, I has coming into it a number of
neural fibers bearing sensory input signals, and it has leaving it one
neural fiber bearing a unified sensory input signal to the comparator.
Ditto for R, with respect to reference signals. Conversely, O has one
fiber entering and a number exiting. C has two entering from I and R,
and one exiting to O. Each function (I, R, O, and C) comprises a number
of cells, and each neural fiber is at least one cell. We want to say
that the input signal and the reference signal together determine the
error signal in the comparator, C. In this really rather complex chain
of inter-cellular relationships, does one cell in the chain control the
next, with respect to the transmitted neural current?

Perhaps with this context, what I said earlier turns out to say
something a little different than it seemed to?

If it does, please consider again the analogy to the human situation.

  Bruce
  bn@bbn.com

2

[Martin Taylor 920710 14:15]
(Bruce Nevin 920710 13:42;31)

I don't want to obscure Bruce's main point, but he provides a wonderful
opening for me to mention a reference I was planning to introduce to the
discussion: "Evidence for a computational distinction between proximal and
distal neuronal inhibition," E.T.Vu and F.B.Krasne, Science, March 27 1992,
255, 1710-1712.

Bruce:

Consider an ECS with a sensory input function I, a reference input
function R, a comparator C, and an output function O. In this simple
and perhaps merely schematic example, I has coming into it a number of
neural fibers bearing sensory input signals, and it has leaving it one
neural fiber bearing a unified sensory input signal to the comparator.
Ditto for R, with respect to reference signals. Conversely, O has one
fiber entering and a number exiting. C has two entering from I and R,
and one exiting to O. Each function (I, R, O, and C) comprises a number
of cells, and each neural fiber is at least one cell. We want to say
that the input signal and the reference signal together determine the
error signal in the comparator, C. In this really rather complex chain
of inter-cellular relationships, does one cell in the chain control the
next, with respect to the transmitted neural current?

If I read Vu and Krasne correctly, all of these functions of an ECS can be
executed in a single neuron, plus the control of gain from another neuron
(interestingly, in their example, this gain control comes from the motor
area of the cortex). They were studying "inhibitory control of the lateral
giant command neurons for crayfish tail-flip escape behaviour," but I doubt
that the specifics matter. They do claim more generality. Their claim is
that inhibitory connections to the outer portions of dendrites have a
subtractive relation to the excitatory connections, whereas the inhibitory
connections to the roots of the dendrites have a multiplicative (I guess
divisive would be a better term) effect, including total inhibition.
I can see this in PCT terms, that the distal connections might, depending
on where the input comes from, contribute either to I or to R,and the additive
(subtractive) relationship has the function of C. The normal firing function
of the neuron represents G, and the proximal inhibitory connection is a control
on G, including switching it off. Multiple neurons of this type working in
parallel could be seen as generating the neural current in an abstract ECS.

This view is not exactly the same as the standard model, in that there is
no clear distinction between I and R inputs to the ECS. But then, functionally
there is no distinction other than that the I inputs are subject to a
transformation providing a single value before being linked to R. I think
that this conventional distinction has no formal effect. Even if I and R
inputs converge on the same dendrite, nature might well maintain their
connections in the way demanded by the standard model.

Let this not stand in the way of an answer to Bruce's original question, please.

Martin

3

[From: Bruce Nevin (Fri 93108 15:55:48 EDT)]

Looking over my (Fri 93108 14:54:04 EDT) as it returned to me from the
server, I see at least one paragraph where I may have invited further
misunderstanding. I said

I have tried repeatedly over a couple of years to work out the relation
between phonemes and the actual sounds that people pronounce.

I'm not talking about my private efforts to work this out. This refers
to working out this relation in a PCT account of language, and it refers
to my attempts to work this out in communication with others on the net.

I said

It cannot be accounted for by saying
that the appearance of phonemic contrast is a byproduct of controlling a
perception of a particular phonetic output called (for example) the
phoneme p or the phoneme b.

I should make clear that in my view identifying a phoneme as a particular
phonetic output is an error, much as it would be an error to say that the
category perception "dog" was nothing more than the particular
perceptions that trigger a dog detector ECS. A given phoneme can have
rather different phonetic alternant forms, and some alternant of one
phoneme may be phonetically indistinguishable from an alternant of a
second phoneme, what is called phonemic overlapping.

Thus, for example, the flap of the tongue tip that occurs in many
American dialects as an alternant of t between vowels (Betty) also occurs
between th and vowel (throw) as an alternant of r. (Historically, this
is a relic of the pronunciation of r that still prevails intervocalically
in many British and Canadian dialects, where intervocalic t in turn is
always aspirated t, never a tongue flap.) Like the p in spin, this is
something the speaker of English is extremely unlikely to notice without
training. The hyperarticulated target is an aspirated t just like the t
in top, and indeed that sound is approximated in contrastive situations,
e.g. "I said `tell BeTTy bye', not `go beDDy bye!" To sort this out one
must take contrast as primitive (as shown by the pair test) rather than
sounds or phonetic productions; a paper I wrote this summer for a
conference deals with this.

I said

The consonant after the s in "spin" is a p phoneme of course, identical
to the p in pin and different from the b in bin.

Here, I am putting into words what I believe has hitherto been the
opinion of many readers on the net, and of you and Bill in particular.
This is not my opinion, it is what I have been inveighing against.
Irony is awfully treacherous in email, and I shouldn't have
used it.

Probably I have made other blunders inviting misunderstanding. Please
read charitably, on the assumption that I am trying to say something
reasonable, though perhaps not doing such a good job of it.

    Bruce
    bn@bbn.com