categorical perception of categories

[Marken_Richard_S1]

[From Rick Marken (931020.1500)]

I asked "isn't ALL perception categorical?"

Martin Taylor (931020 14:40) replies --

Absolutely not. No level below the category level in the Powers
hierarchy is categorical.

Bill called one type of perception "category"; I think of the
perceptual signals at this level as being representations of the
degree to which any lower level perception is an instance of the
category (the "degree" can be binary if you like -- the instance
either is a member or it's not, but my perception of category seems
a bit smoother -- I am looking right now, for example, at something
that I perceive as "sort of" a book and "sort of" a pamphlet;
perception of categories is not necessarily categorical).

In lower levels, it makes sense to ask
about the magnitude of the perceptual signal.

Why does it not make sense to represent category perception
as a magnitude? I am scanning across a bookshelf where most
objects are perceived as books -- all or none. But there are
some intermediate cases; I don't perceive a real strong sense
(perception) of "book" category when I look at them.

At the category level it makes sense only to ask "whether" not "how
much."

Obviously, I disagree with this assertion. Why do think it is not
"sensible". Is this based on your experience with modelling control
of category perceptions?

Category is of a completely different
kind than the levels below.

This is true -- but that does not mean that other perceptions are
not categorical. This was the reason for my question -- how does
a category perceptual signal differ from the perceptual signal
at any other level of the hierarchy.

For example, imagine that there is a perceptual function (probably
at the sensation level) that produces a perceptual signal that
is proportional to the sum of its two lower level perceptual
inputs, s1 and s2. So p = s1+s2. Now the p signal "categorizes"
its inputs in the sense that it puts out the SAME value for many
different combinations of s1 and s2. For example, p is EXACTLY
the same when the inputs are 3 and 1, 2 and 2, 4 and 0, etc. So
when p is 4, we can say nothing about the inputs that produced
it; many different inputs result in the same output. This
is categorization.

The same kind of categorization happens when I look at my bookshelf;
in this case I see different configurations (again call them s1,
s2 ... sn) at different times (as I scan across the bookshelf).
Apparently, there is a function is my brain that produces a perception,
p, of "bookness" when I look at s1, s3 and s5 but NOT when I look at
s8 -- and I get an intermediate sense of "bookness" when I look at
s2. So I get a category perception from the same PROCESS that results
in a sum perception -- from a function that maps some lower level
perceptual inputs into one p value and maps other lower level perceptual
inputs into other p values -- a process that I would call
"categorization". But the function that computes the category
is of a different TYPE then the function that computes the sum.

In the process of trying to explain this to you I have moved myself
a bit closer to understanding the difference between perceptual
categorization (which I think is just a way of describing how perception
works -- ie. p = f(s1, s2,sn) )-- and the perception of category.

Best

Rick

[_Martin_Taylor]

[Martin Taylor 931020 20:10:39] -- nice time, that
(Rick Marken 931020.1500)

imagine that there is a perceptual function (probably
at the sensation level) that produces a perceptual signal that
is proportional to the sum of its two lower level perceptual
inputs, s1 and s2. So p = s1+s2. Now the p signal "categorizes"
its inputs in the sense that it puts out the SAME value for many
different combinations of s1 and s2.

What you are calling "categorization" is what I would call "measurement"
or "evaluation." If the measurement is to determine in which bin (there's
that word) the perception p (= s1 + s2) fits, then the result is a
categorization. p is in Bin 136. It is not in Bin 135 or 137. Its
value may be 136.7317854....

In this simple context, all p with the same value will be found also
to belong in the same category. But one can easily cook up contexts
in which that is not so. Remembering that the category level is higher
than the level at which the sum occurs, suppose that the category level
PIF has another input, called clock time. Before noon, all category
boundaries are shifted 0.5 units down as compared to their position
after noon. Now the same p is in bin 137 before noon and in bin 136
after noon.

This is kind of like the category boundary shifts we propose occur when
the contexts of language perception include perceptual signals derived
from the partner state. More generally, they occur in (almost?) all
category perception because of hysteresis. The existence of hysteresis
has been demonstrated in many situations (it was the topic of a thread
here a year or two ago). The mechanism for hysteresis is unknown, but
I proposed one that seems plausible to me (within-level recursion among
PIFs in the category level).

when I look at my bookshelf;
[...] I see different configurations (again call them s1,
s2 ... sn) at different times (as I scan across the bookshelf).
Apparently, there is a function is my brain that produces a perception,
p, of "bookness" when I look at s1, s3 and s5 but NOT when I look at
s8 -- and I get an intermediate sense of "bookness" when I look at
s2.

I claim that you are perceiving multi-level perceptual signals.
Configuration signals that fall near the edge of the "book" category
PIF boundary (as it is at that moment) are not so "booky" as ones that
are like prototypical books. How can that perception of "bookiness"
occur? Without having worked it out fully, I think it has to do with
a reference provided to the lower ECS by the output of the book category.

When a poor example of "book" is providing the input to that lower level,
there will be a substantial error signal in its ECS. Do you remember my
Durango talk, in which I had the Syntax Recognizer generating syntactically
correct output (without categories, by the way)? When it was not producing
output visible from outside, it was still controlling in imagination,
keeping synchronized with the input, so far as possible. That's a bit like
what I am talking about. The book category controls in imagination what
a book should in lower level perceptions, and in some way the discrepancies
are perceived as poor bookiness (as that is poor wordifaction).

I know that's pretty vague. Can't help that. It may get refined, or it
may get lost, over time.

Martin