Welcome Osmo;

[From Bill Powers (931007.1115 MDT)]

Tomorrow Mary and I will be off on a four-day camping trip into
Canyonlands in Utah, so between tomorrow and Tuesday don't start
imagining things when you don't hear from me.

Osmo Eerola (931007) --

(Note format convention on CSGnet: hand-enter a line identifying
yourself and date-time, then precede comments on other people's
posts with identification of person and date or date-time)

Welcome, Osmo! I've very glad that you're interested in PCT and
especially that you're an engineer in neuroscience, a wonderful
combination. Your best source for general information about PCT
is in the books and publications listed in the "introduction to
PCT" that is transmitted at the start of every month. But the
basic material was not written for a technical audience, and it
will simply get across basic PCT ideas. The real explorations of
technical questions take place on this net, and are (as you have
seen) in a continual state of change as we try to work out
complicated ideas. Every now and then the subject of language
comes up and starts off an intensive discussion that may go on
for weeks and months, and then everyone gets exhausted and some
other topic takes over for a while. Your best bet for following
the exchange on language is to start downloading the weekly
archives and just search through them for key words like
"phoneme" or "speech" and so forth. If you become a member of the
Control Systems Group (CSG) you can receive Closed Loop, which is
a very nicely edited set of extracts from the net on particular
topics. I wouldn't be surprised if Greg Williams, the editor, did
an issue on language topics pretty soon (but that's up to him).

Of course another way to find out about PCT ideas on topics that
interest you is to start up a conversation of your own!

ยทยทยท

---------------------------------------------------------------
Joel Judd (931007) --

Interesting scenario by Lakoff on anger. For a closer fit to my
proposals about emotion, I would modify it somewhat;:

        STAGE 1: offending event
                  this may be an event or a wrongdoer that
                  displeases the Self (S) [i.e., disturbs a
                  variable that is under control].

        STAGE [2] attempt at control [opposition to effects of
                  disturbance]

        STAGE [3]: [threatened] loss of control [disturbance
                   continues, leading to increases of output
                   efforts and marked mobilization of
                   biochemical systems]

        STAGE [4]: anger
                this is associated with physiological effects
                such as increase in body temp, blood pressure
                rise, etc. [as well as with the preparation to
                correct the situation with motor behavior]

        STAGE 5: act of retribution
                this is carried out in terms of the perceived
                "intensity of offense" for the purpose of
                "balancing the scales." [ this applies if there
                is no inner conflict to prevent actually taking
                action and correcting the error].

As you can see I put the attempt at control as the initiating
event, with the (somatically felt) emotion being a consequence of
preparing to act.
---------------------------------------------------------------
Bruce Nevin (93107.1125 EDT) --

Your post contains one pregnant statement that we must explore in
detail:

A word is an event perception as a whole, not broken down into
phonemic elements.

On this statement may hinge a number of the difficulties we have
been having.

In HPCT, perceptions do not "break down" into component
perceptions. They are _synthesized from_ component perceptions,
and can't exist unless the component perceptions already exist.

Consider the relation between the sensation level and the
intensity level. If you look at a sensation like an edge, you can
"break it down" into different intensities on either side of the
edge. The intensities are of a different class from edgeness. But
this "breaking down" is a matter of redirecting your attention
among existing perceptions, not of creating perceptions. In the
model of the nervous system, the intensity signals are created
not by edges, but by external stimuli acting on sensory endings.
If there were no intensities, there could be no way that a
perception of edgeness could exist. As the intensities differ
more where they are adjacent, the sensation of edgeness becomes
more pronounced. As the intensities move toward equality, the
sensation of edgeness fades and disappears, even though the
sensations of intensity are not zero. It is the sensation of
edgeness that is synthesized from a collection of intensity
signals, not the other way around.

The same relation holds at every level. Configurations depend for
their existence on a set of sensation-signals and perhaps
intensity signals. Transition perceptions depend on a set of
configuration signals and lower. And _event perceptions depend
for their existence on the existence of signals representing
transitions, configurations, etc._.

To speak of pre-existing "event perceptions as a whole" suggests
strongly that you are projecting event perceptions into an
imaginary mirror world in the environment. The "real" event-as-a-
whole breaks down into constituent "real" syllables and/or
phonemes, which break down into "real" sounds, which break down
into "real" stimulations of individual acoustic sensory nerves.
This says that the word is the objective primary whole event, and
that all the more detailed aspects of it are derived from the
whole event.

This is exactly the mirror image of the perceptual processes in
HPCT, with the direction of dependence and causality exactly
reversed and the whole thing externalized.

HPCT says that the perceptual process begins with detailed
sensory stimulation that leads to a large, maximally
differentiated, set of intensity signals. A second level of input
functions receives subsets of these intensity signals, weights
them and adds the results together, and produces a new kind of
signals, sensation signals, that each represent the collective
state of intensity signals in some specific respect.

Without the initial intensity signals, there are simply no
sensation signals: there is no input to any second-order input
function, so there is no output. Furthermore, the only sensation
signals that exist are those that are derived by specifically
organized input functions from sets of intensity signals.
Although there may be other possible ways of deriving sensations,
those other sensations do not exist in experience, or at all.

Then, at least as the model is organized today, a new layer of
input functions receives signals standing for sensations (and
possibly intensities) and construct invariants represented by a
new level of signals: configuration signals. Now objects,
orientations, sizes, positions, and other patterns come into
existence. Again, only those configurations that are specifically
computed by an input function (of the third level) are
represented as signals. Again, a configuration signal can't exist
unless sensation and/or intensity signals are already present,
for the configuration signal is computed from those lower-level
signals. ONLY those configurations are perceived which are
derived from existing lower-level signals by a specific input
function. Other possible configurations which are not so computed
and represented are simply not part of experience.

Transitions likewise are derived by specific input functions from
signals standing for configurations, sensations, and intensities,
and depend for their existence entirely on the existence of the
lower-level signals and the appropriate input functions.

And finally, event perceptions are derived by specific input
functions from sets of transition signals, configuration signals,
and so forth. Without the lower-level signals, there could be no
event signal, and thus no experienced event.

So the perception of an event is _synthesized_ from sets of
lower-level signals, which are _synthesized_ from still lower
signals, until we reach the level of sensory stimulation. Without
the lower signals, the event perception could not exist. And only
those event perceptions exist which are derived from lower-level
signals by specific input functions.

When we analyze a perception into "its components" we are really
backtracking the perceptual process to see where a specific
perception is coming from. As we attend to lower and lower levels
of perception, more and more detail, we are moving the locus of
awareness against the flow of information, toward the source. We
are looking toward the causes of perceptions, not the effects.
Higher perceptions are caused by lower ones, in a manner dictated
by the organization of the intervening input functions.

This is the physical model of perception in HPCT. The flow is
exclusively upward, from lower levels to higher. Under this
model, it is not possible that a higher-level perception could
exist before lower-level perceptions exist. It is not possible
that the perception of a "whole word event" could exist before
perceptions of "its components" exist, because the event is a
function of the components.

You are having difficulties in reconciling contrast and other
phenomena with the idea that words are derived from syllables
and/or phonemes. This is because you are looking at the
perceptual process as one of analysis instead of synthesis. If
you think of an event as a function of sets of syllables and
phonemes, which in turn are functions of sets of auditory
sensations, your problem becomes that of proposing input
functions which would create the same dependencies (and lacks of
dependencies) that are observed.

If a person perceives spin and sbin as the same word, this is
simply because the input function that synthesizes that
particular word-event treats the 'b' and 'p' inputs as equivalent
-- gives them about equal weight in the structure of the
perceived event. Thus either spin or sbin will produce the same
word-signal in perception. A slightly-different input function
might require that p' rather than p` be heard, and create a given
event-signal when the one is present but not when the other is
present. Or it might weight these two inputs equally and produce
the same word-signal in either case.

This is a matter of how the input function is organized. If you
find a strong tendency in a population to make a certain
distinction, you have simply found that the corresponding input
functions strongly tend to be similarly organized. You might
conjecture that this preference results from greater ease in
controlling one perception than the other, perhaps because of co-
articulation problems or problems in others' understanding what
you're saying. Further investigation of lower-level aspects of
the situation might support one or the other proposal.

My point is that if you think of perception as an upgoing process
of synthesis rather than a downgoing process of analysis, you can
still account for the phenomena of speech recognition by
proposing suitable input functions. In fact, it seems to me that
this approach would be both simpler and more amenable to testing
than the rather mixed approach you're taking.

Over to you -- but not until next Tuesday (or possibly Monday).
------------------------------------------------------------
Best to all,

Bill P.