Categorical perception, If you build it

CSGnet Archive CSG1993
1

[From Rick Marken (931020.0930)]

I said:

It is an incorrect representation of the PCT model of perception. A
perceptual signal (in the model) IS a perception. So a signal whose
magnitude represents, say, the degree of "d"-ness IS the perception of
"d"-ness.

Martin Taylor (931019 17:45) replies

I take this as a claim that there can be no level of control for which
the perceptual signal has a discontinuous set of values.

Not at all. A perceptual signal could be a logical (binary) variable.
So "d-ness" might be present or not. The discreetness or continuousness
of the perceptual variable was not the point of my comment. The
point was that a perceptual signal (in the model) is not perceived
differently depending on context -- because the perceptual signal
is not perceived (it is not an object of perception); it IS a
perception. Do when, say, the "d-ness" signal is "on" you experience
"d", regardless of the context. But, at a higher level (which
receives the "d-ness" and other "contextual" signals as input)
there will be another perceptual signal, the value of which depends
on the contribution of "d-ness" AND on the other (contextual) inputs.

"Categoricalness" seems to be what you consider important about
"contrast". While thinking about your emphasis on "categoricalness"
a question occurred to me that you might be able to answer: the
question is "isn't ALL perception categorical?". You seem to be
saying that a "contrast" indicates a categorical distinction. For
example, I can tell "pin" from "bin" and that's a contrast. You seem
to be saying that there are all kinds of variants in the sounds that
correspond to "pin" and "bin" which we don't hear as variants -- we
just hear "pin" or "bin".

But I was thinking -- isn't all perception categorial in this sense?
That is, any perception is "invariant" with respect to certain
transformations of its components -- just like "pin" and "bin".
So your hand, for example, looks the same (is invariant) despite
changes in its distance from you (transformations in its sensed size).
A color is known to be invariant with respect to certain tranformations
in the intensity of its spectral components. In general, given
the perceptual model of PCT -- p = f(s1, s2...sn) -- a particular
perceptual signal value will be invariant with respect to certain
transformations of the inputs, s1, s2 etc. Which transformations
leave a partcular perception invariant depend on the nature of f().
So, for perceptions of many different "types" (as defined in HPCT)
we would expect to find that a particular state of that perception
(value of p) is "categorical" in the sense that it is the same
(invariant) for many different combinations of input (s1, s2..).
This is exactly what is happening with "pin" and "bin" -- the
p value that IS "pin" is the same (we keep hearing "pin") despite
considerable variation in the acoustical signal (and, hence, the
lower level sensory inputs to the function that computes the p for
"pin").

My question is: how is this type of categorical perception (which
is true of ALL levels of perception -- even the intensity level)
different from the perception of category membership (which is
currently a perceptual type of its own in HPCT)?

Greg Williams (931020) --

What I
was looking for from Rick (and now from you) is evidence regarding
how various HF folks use the word "control." I am still looking.

My impression is that they use the word "control" in the same ambiguous
way it is used by most behavioral scientists. Sometimes they use it
to mean "influence" and sometimes they use it to mean what it really
means -- to produce a specified result regardless of prevailing
disturbances to that result.

But who cares what they mean anyway? If they really understood the
nature of control they would be testing for controlled variables and
comparing the performance of people to models that control various
perceptual variables. They ain't. They would also be devouring the
theoretical and reserch work of the few people who are doing it.
They ain't.

I think you are bending over so far backward to see PCTness in
what is already being done in the social sciences that you've
got them turned upside down -- so that their S-R approach looks
like the control approach. I say "relax" Greg, and give the ol'
back a rest. If any of those people actually are intrested in
understanding control then they'll come to us. They won't have to
be cajolled or courted or enticed. They'll learn about PCT becuase
they'll start to see behavior from a completely new perspective.
Once they learn the basics of PCT they'll give up their old beliefs
as smoothly as did the alchemists who learned to do chemistry. There
won't be any need for them to say "PCT is just like X" or "I still
think Y is true, even though it contradicts PCT" because they will
understand the new point of view and just move off in the new
direction -- dropping the confused baggage of the older point of
view. If we build it right -- carefully, slowly, precisely -- they
will come (though "they" are very likely to be my children's
children).

Best

Rick

2

[Martin Taylor 931020 14:40]
(Rick Marken 931020.0930)

Rick said

It is an incorrect representation of the PCT model of perception. A
perceptual signal (in the model) IS a perception. So a signal whose
magnitude represents, say, the degree of "d"-ness IS the perception of
"d"-ness.

Martin Taylor (931019 17:45) replies

I take this as a claim that there can be no level of control for which
the perceptual signal has a discontinuous set of values.

Not at all. A perceptual signal could be a logical (binary) variable.
So "d-ness" might be present or not. The discreetness or continuousness
of the perceptual variable was not the point of my comment. The
point was that a perceptual signal (in the model) is not perceived
differently depending on context -- because the perceptual signal
is not perceived (it is not an object of perception); it IS a
perception.

Rick's initial comment was in response to one of my vain attempts to
distinguish between "contrast" and "difference." This posting shows that
even my subsequent attempt utterly failed to get through. One more attempt,
and if that fails I will concede that an approach through simple language
is unlikely ever to succeed, and some other, as yet undiscovered, approach
must be taken.

"Categoricalness" seems to be what you consider important about
"contrast".

Well, you got that right.

While thinking about your emphasis on "categoricalness"
a question occurred to me that you might be able to answer: the
question is "isn't ALL perception categorical?".

Absolutely not. No level below the category level in the Powers
hierarchy is categorical. In lower levels, it makes sense to ask
about the magnitude of the perceptual signal. How much intensity,
how much of the configuration defined by the PIF, and so forth.
At the category level it makes sense only to ask "whether" not "how
much."

For
example, I can tell "pin" from "bin" and that's a contrast. You seem
to be saying that there are all kinds of variants in the sounds that
correspond to "pin" and "bin" which we don't hear as variants -- we
just hear "pin" or "bin".

At the category level where words as such live, you do hear only "pin"
or "bin." At the event level that feeds into the category PIF, you
can hear all sorts of variants within "pin." At the intensity level
you hear all sorts of different variants that you would not consider
to be differences at the event level. A loud "p'in" is detectably
different from a soft "p'in" at the intensity level, but this doesn't
matter to perception of the "p'in" event, which, in its turn is different
from a "pin" event. There's no contrast in any of these. Just
differences of the magnitudes of perceptual signals at different
levels.

From Bill Powers (931016.0815)

each new level introduces a new
logical type of controlled variable, not just a scaled-up or more
complex version of the types already existing at the lower
levels. The sense of motion is not just a scaled-up or more
complex configuration or sensation. An event is not just a
transition in a new direction or a bigger or more complex
configuration. And a relationship is not just a new kind of
event, transition, or configuration.

This is what I am getting at. Category is of a completely different
kind than the levels below. What is characteristic about category
perception is precisely that it segregates the world into distinct
little bits. What is happening in the world belongs to this category
or that, not more or less, but yes or no. One can indeed indentify
whether the lower-level perceptions are those that would normally
represent a good or bad example of the category, but those perceptions
ARE at lower levels. They aren't magnitudes of category.

(Debatable proposition coming up. Do not confuse with the foregoing)

If you want a process model for category, my intuition is that this is
the level (perhaps the only level) at which there is cross-connection
among the perceptual input functions within the level, in such a way
as to induce positive feedback in the perception part of the hierarchy.
The PIFs therefore produce saturated perceptual signals, and exhibit
hysteresis (or catastrophe, depending on how you want to look at it).
I know this cross-connection violates the strict upward flow of the
Powers hierarchy, but I can't help that. Take it or leave it, as a
process model.

(End intuition segment)

What you shouldn't discard is the notion that the category level is not
just another event level. It isn't. It is its own thing, a splitter
of the perceptual universe.

But I was thinking -- isn't all perception categorial in this sense?
That is, any perception is "invariant" with respect to certain
transformations of its components -- just like "pin" and "bin".
So your hand, for example, looks the same (is invariant) despite
changes in its distance from you (transformations in its sensed size).
A color is known to be invariant with respect to certain tranformations
in the intensity of its spectral components.

Any PIF, no matter what its level, produces a single-valued output, at least
if we are to believe the Powers form of HPCT in any moderately strong sense.
Any PIF, no matter what its level, receives its input from many sources.
This means that the output "invariant" of the PIF is a subspace of the
input space, one dimension smaller than the input space. Think of it,
at the lower levels, as a curved line on a plane, or a twisty sheet in
a 3-space. The "value" of the perception is its distance within this
twisty space from some sheet labelled "zero."

Bill P. has discussed this many times on CSG-L, and I don't think there
is much value in belabouring the point. But to complete it with respect
to categories, the categoric space should be imagined as discontinuous,
with only a zero-measure subspace that can be occupied. In other words,
categories can take on only predetermined places within the space, places
we as outside observers might label with the names of the categories, though
as you rightly say, all the "user" gets is a perceptual signal from the
category's PIF.

I don't know whether it bears repeating, after these many postings, but
one CAN be conscious of perceptions at many levels, sometimes simultaneously.
That one can at the same time perceive the category "pin" and know it is
not "bin" in no way precludes you from perceiving the events that feed
the category level and let you say "That was a very 'b'-like 'p' in the
'pin' you just said." There is a contrast between "pin" and "bin" but
only a difference between the ways the two may have been spoken.

Martin