Perceptual distance measures

[rsmarken]

[From Rick Marken (2007.09.25.1930)]

Martin Taylor (2007.09.25.15.18)--

Me:

>I don't think one is parsing an intensity out of a configuration when
>they judge a components of a configuration; if they are judging the
>line as part of a configuration then I think they are judging the line
>as a configuration perception.

I'm afraid I don't understand this within the conventional HPCT
hierarchic arrangement. Are you saying that the single value of the
perceptual signal in a configuration control unit IS an intensity
perception? It sure sounds like that.

No. I meant it to be a configuration (or relationship) perception.

To my amazement this is very much like what I did in my thesis
research on signal detection. One theory of tone detection held that
judgments of whether or not a tone is present in noise were based on
the total energy (intensity) in the tone band. My thesis showed that a
better model is one that assumes that judgments of tone
presence/absence are based on the spectral and temporal pattern
(configuration) of intensities at and around the temporal and spectral
location of the tone. So even though people were asked to judge the
intensity of sound in a narrow band, their judgments were better
modeled as being based on a _configuration_ of sound.

Of course, my thesis was based on an open loop model of the tone
detection task. This was 4 years before PCT for me. But the perceptual
component of the model was (as Bill once said in another context)
comfortably compatible with the PCT model of perception. Of course, at
the time I also didn't understand that there might be hierarchical
levels of perception. Basically, the model used in my thesis assumed
that people perceive tones in noise in only one way: as what I would
now call configuration perceptions. But I now think that there are
many levels at which we simultaneously perceive the world. And the
subjects in my thesis research may have been making their judgments on
the basis of different levels of the perception of the sound on
different trials.

Best regards

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin_Taylor2]

Re: Perceptual distance
measures
Martin Taylor 2007.09.26.10.12]

[From Rick Marken (2007.09.25.1930)]

Martin Taylor (2007.09.25.15.18)–

Me:

I don’t think one is parsing an intensity out of a
configuration when

they judge a components of a configuration; if they are
judging the

line as part of a configuration then I think they are judging
the line

as a configuration perception.

I’m afraid I don’t understand this within the conventional
HPCT

hierarchic arrangement. Are you saying that the single value of
the

perceptual signal in a configuration control unit IS an
intensity

perception? It sure sounds like that.
No. I meant it to be a configuration (or
relationship) perception.

I still don’t think I properly understand your thinking (that a
configuration or relationship can be perceived as an intensity), but
we’ll let that pass for the moment.

============ non-PCT digession ===============

To my amazement this is very much like
what I did in my thesis

research on signal detection. One theory of tone detection held
that

judgments of whether or not a tone is present in noise were based
on

the total energy (intensity) in the tone band. My thesis showed that
a

better model is one that assumes that judgments of tone

presence/absence are based on the spectral and temporal pattern

(configuration) of intensities at and around the temporal and
spectral

location of the tone. So even though people were asked to judge
the

intensity of sound in a narrow band, their judgments were better

modeled as being based on a configuration of sound.

I’m not sure this is a PCT thread, but it interests me (as you might
guess, knowing my long-ago background in psychoacoustics and signal
detection theory). I should have known then about your work, and maybe
I did but have forgotten. In my fading memory, I associate that kind
of work with someone famous from Bell Labs with a short Indian name
that I am blocking on right now (Atal?).

You say two different things in the quoted paragraph: that it was
a signal detection problem, and that subjects were asked to judge
intensity. These are separable questions, and I’m assuming you studied
both.

I’m not at all surprised that you found that temporal and spatial
configurations affected detection. Since you mention “tone”
as the detection target (signal), I assume the temporal and spectral
configurations you mention are of the masking noise. One theory of
magnitude perception (to which I subscribed, and maybe still do) is
that perceived intensity is determined by discriminability, so that
when the signal is hard enough to detect that the d’ is measurable,
the two should match.

I used that hypothesis in my very first solo publication, in
which I predicted the parametric form of a new (I thought) visual
illusion, and made the counterintuitive prediction that subjects who
were more precise in their estimates should show more illusion. They
did. So I still kind of like the idea, even though I now realize that
font designers usually take the illusion into account.

(Aside to the Digression, Arabian Nights style:-)Discriminability cannot be judged by adding jnds in the Weber-Fechner
manner. This fact was acknowledged in a wonderful book that my
grandfather got as a school prize, called “Popular Scientific
Recreations, a storehouse of instruction and amusement, in which the
Marvels of natural Philoosphy, Chemistry, Geology, Astronomy, etc. are
explained and illustrated, mainly by means of pleasing experiments and
attractive pastimes.” Ward, Lock, Bowden and Co, undated,
Translated and Enlarged from “Les Recreations Scientifiques”
of Gaston Tissandier. Gaston Tissandier was what was called an
“aeronaut”, who organized the balloon surveillance defences
of Paris in 1970. I seached for his book with absolutely no success,
until one day in Ann Arbor, Mich., I was waiting in the university
library stacks for a colleague, and found myself facing Gaston
Tissandier’s book. It’s about 1/3 the size of the “translated and
enlarged” book, and the important comment about the distinction
between discriminability and JND summation is not in it. I tried to
find out from the publisher when the book was written and who was the
translator, but their records were destroyed in the London Blitz. It
has to have been someone at the level of Michael Faraday, from the
quality of the text, and from internal evidence the English book has
to have been written in the late 1880s or early 1890s.

(End of Aside to the digression)

The questions arise when the detectability is high (d’ > say
3) and unmeasureable. Then you have to use theoretical approaches to
the estimation of discriminability, and that’s not always a very
secure approach. So if you found that intensity estimates well above
detectabilities of d’ = 3 (and therefore unmeasureable) were also
dependent on the spectral and temporal configurations of the masking
noise, that would be quite interesting.

=============end digression===================

Back to PCT.

… Basically, the model used in my
thesis assumed
that people perceive tones in noise in
only one way: as what I would

now call configuration perceptions. But I now think that there are

many levels at which we simultaneously perceive the
world.

Yes, I agree. And it brings us right back to a comment I made
several messages ago, that we ought to try to understand the
difference between this multi-valued conscious perception of the world
and the scalar-valued perceptual signals in the hierarchy. No one
perceptual signal can carry the different values of these “many
levels at which we simultaneously perceive the world”.

That’s been the central point of my insistence that f(s) is not
the same as f(s, context), and that signals from the intensity level
and the configuration level must be mixed, which implies a revision of
the “classical” concept of the HPCT hierarchy.

And the

subjects in my thesis research may have been making their judgments
on

the basis of different levels of the perception of the sound
on
different trials.

Yes, quite probably.

Martin

PS. As of tomorrow morning, I’ll be incommunicado (on holiday on
Manitoulin Island in Lake Huron) for a week or ten days.

[rsmarken]

[From Rick Marken (2007.09.26.1100)]

Martin Taylor (2007.09.26.10.12) --

I still don't think I properly understand your thinking (that a
configuration or relationship can be perceived as an intensity)

That may be because you are thinking that judgments of intensity must
be based on a perception of intensity. Specifically, you may be
thinking that a judgment of physical distance, s, must be based on a
perception that depends only on s. That's why I brought up my thesis
research. Even back then I knew that a judgment of a physical
intensity is not necessarily based on a perception that depends only
on that intensity.

You say two different things in the quoted paragraph: that it was a signal
detection problem, and that subjects were asked to judge intensity. These
are separable questions, and I'm assuming you studied both.

No. The subjects were asked only to judge whether or not a tone was
present in noise. This is like asking for a judgment of intensity
because the tone changes only the intensity of the noise in one
frequency band. One model of signal detection held that the judgment
of tone presence/absence was based on a perception of physical
intensity only. That is, if s is the intensity of the noise in a
narrow band around the tone frequency, then the tone presence/absence
decision was assumed to be based on p = f(s); p greater than a
criterion value was called "tone" and p less than the criterion was
called no tone. So this is a model of tone detection based on the
assumption that tone detection was based on a perception of tone
intensity, s.

My thesis research showed that a better model was that tone detection
was based on a perception, p, that was a a function of the temporal
and spectral "shape" of the noise at and near the tone frequency. That
is p = a1f1-a2f2+a3f3-a4f4+a5f5 where the fi represent the intensity
of the noise in spectral bands around the tone frequency band (say it
was f3) and the ai are the weights given to the noise in each of there
bands. So p is a configuration perception. The negative weights (for
f2 and f4 in this example) show that p cannot be considered to be
simply proportional to the total intensity in a band around the tone
frequency.

The point is that, even though the subjects are asked to say simply
whether there was a tone present in noise (which is equivalent to
saying that the intensity at the tone frequency in one noise burst --
the one with the tone -- is greater than it is in another-- the one
without the tone) they based their judgments on a configuration
perception. The subject is only asked to rate the intensity difference
between the noise bursts. But what subjects actually do, apparently,
is evaluate (at least on most of the trials) a perception of the
configuration of sound as the basis of their judgment. This is exactly
analogous to what probably happens in the Muller Lyer distance
judgments. The subjects are asked only to judge the "intensity" of the
distance between points; but apparently what the subjects judge is a
perception of a configuration (or relationship) involving the line to
be judged and the context lines (just as the context noise is included
in the judgments of tone presence/absence in my tone detection
experiment).

Discriminability cannot be
judged by adding jnds in the Weber-Fechner manner.

It's psychological magnitude, not discriminability, that is _assumed_
to result from adding up jnds. The log function that results from
integrating over jnds is _assumed_ to be a measure of the increasing
magnitude of the perception, psi, that corresponds to increasing
magnitudes of stimulus, s.

Back to PCT.

We never left;-)

No one perceptual signal can carry the
different values of these "many levels at which we simultaneously perceive
the world".

Your concept of a perceptual signal is quite different than mine. I
don't think of perceptual signals as carrying anything; they are
simply magnitudes that vary in terms of firing rate. The "meaning" of
these magnitudes is given by the perceptual functions that transform
input signals into perceptual signals. The meaning of the magnitude of
the perceptual signal depends on what the perceptual function is. If,
for example, the perceptual function computes the rate of change in
the value of the input variable, so that p = dy/dx, then the magnitude
of p is a measure of the rate of change in y relative to a change in
x. So in this case the perceptual signal is experienced as "change"
because it is the result of a perceptual function that computes the
derivative of the input, y, to that function.

That's been the central point of my insistence that f(s) is not the same as
f(s, context), and that signals from the intensity level and the
configuration level must be mixed, which implies a revision of the
"classical" concept of the HPCT hierarchy.

I don't see the problem. f(s) could be the function that gives the
lower level perception on which judgments of the intensity of s could
be based and f(s,context) could be the function that gives the high
level perception on which judgments of the intensity of s could be
based. In that case it would be better to write it as p1 = f(s) and p2
= g(p1, context), where the 1 and 2 are relative (ordinal) levels of
the perceptions.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin_Taylor2]

[Martin Taylor 2007.09.26.14.17]

[From Rick Marken (2007.09.26.1100)]

Martin Taylor (2007.09.26.10.12) --

I still don't think I properly understand your thinking (that a
configuration or relationship can be perceived as an intensity)

That may be because you are thinking that judgments of intensity must
be based on a perception of intensity.

Yes.

  Specifically, you may be
thinking that a judgment of physical distance, s, must be based on a
perception that depends only on s.

No. My problem isn't in that direction at all. One couldn't have thought about perception for more than 5 minutes and maintain that idea. My problem is with the mapping of these ideas onto the HPCT hierarchy of perceptual types.

That's why I brought up my thesis
research. Even back then I knew that a judgment of a physical
intensity is not necessarily based on a perception that depends only
on that intensity.

Well, I suppose everyone interested in perception has known that since the 19th century. I read about it (along with Donders' idea of levels of perception) in the Cassell's Children's Encylopedia that my Dad had when he was a kid.

> You say two different things in the quoted paragraph: that it was a signal

detection problem, and that subjects were asked to judge intensity. These
are separable questions, and I'm assuming you studied both.

No. The subjects were asked only to judge whether or not a tone was
present in noise.

OK. That's clear.

This is like asking for a judgment of intensity
because the tone changes only the intensity of the noise in one
frequency band.

It's not the same at all. You are right away conflating physical with psychological magnitudes. Physically the tone changes the intensity in one frequency band, but that doesn't mean the subject judges intensity. Having done a LOT of detection studies as a subject as well as an experimenter, it know very well that in a forced-choice experiment the interval that you judge to have the tone is not always the one in which the tone-frequency seems loudest, precisely because of the effect you studied. Although, I suppose if you kept the noise waveform and temporal pattern the same between the two intervals, the effective judgment would be of physical intensity.

One model of signal detection held that the judgment
of tone presence/absence was based on a perception of physical
intensity only.

That is true. It's not a good model.

My thesis research showed that a better model was that tone detection
was based on a perception, p, that was a a function of the temporal
and spectral "shape" of the noise at and near the tone frequency. That
is p = a1f1-a2f2+a3f3-a4f4+a5f5 where the fi represent the intensity
of the noise in spectral bands around the tone frequency band (say it
was f3) and the ai are the weights given to the noise in each of there
bands. So p is a configuration perception. The negative weights (for
f2 and f4 in this example) show that p cannot be considered to be
simply proportional to the total intensity in a band around the tone
frequency.

Very nice. Were the optimal weights anything like a Gaussian filter in log frequency?

The point is that they based their judgments on a configuration
perception.

Right. That is my intepretation, too.

The subject is only asked to rate the intensity difference
between the noise bursts.

That's what I don't accept. The subject was asked (I presume) to judge in which interval the signal occurred, not the intensity difference between the two intervals. Quite a different thing. If indeed they were asked to judge the intensity instead, I stand corrected.

But what subjects actually do, apparently,
is evaluate (at least on most of the trials) a perception of the
configuration of sound as the basis of their judgment.

Yes.

This is exactly
analogous to what probably happens in the Muller Lyer distance
judgments. The subjects are asked only to judge the "intensity" of the
distance between points; but apparently what the subjects judge is a
perception of a configuration (or relationship) involving the line to
be judged and the context lines (just as the context noise is included
in the judgments of tone presence/absence in my tone detection
experiment).

Yes. I think we've agreed all along on that.

Discriminability cannot be
judged by adding jnds in the Weber-Fechner manner.

It's psychological magnitude, not discriminability, that is _assumed_
to result from adding up jnds. The log function that results from
integrating over jnds is _assumed_ to be a measure of the increasing
magnitude of the perception, psi, that corresponds to increasing
magnitudes of stimulus, s.

True, but irrelevant. I was talking about discriminability, not psychological magnitude. Discriminability is what cannot be estimated by adding jnds (well, neither can psychological magnitude, probably). I _assumed_ that psychological magnitude is related to discriminability, not to the summation of jnds. I'm still not sure that assumption is false.

> Back to PCT.

We never left;-)

No one perceptual signal can carry the
different values of these "many levels at which we simultaneously perceive
the world".

Your concept of a perceptual signal is quite different than mine. I
don't think of perceptual signals as carrying anything; they are
simply magnitudes that vary in terms of firing rate.

Exactly! Hurrah, hurrah! You understand at last!

> That's been the central point of my insistence that f(s) is not the same as

f(s, context), and that signals from the intensity level and the
configuration level must be mixed, which implies a revision of the
"classical" concept of the HPCT hierarchy.

I don't see the problem.

Oh. You don't understand, after all.

f(s) could be the function that gives the
lower level perception on which judgments of the intensity of s could
be based and f(s,context) could be the function that gives the high
level perception on which judgments of the intensity of s could be
based. In that case it would be better to write it as p1 = f(s) and p2
= g(p1, context), where the 1 and 2 are relative (ordinal) levels of
the perceptions.

And what is the output? A signal at the configuration level. It's not an intensity, a distance magnitude, or anything like that. It's a configuration. At least it is in the standard HPCT hierarchy. And in the hierarchy, it isn't interpreted "as" anyting at all. It just IS.

Now we aren't really talking about the HPCT hierarchy, are we? As soon as we accept that fact, I think the problem vanishes into thin air. We are talking about conscious perception (I'll call it Cperception), which is NOT a single-valued magnitude or firing rate. We are or can be conscious of "the many levelsat which we simultaneously perceive the world". In this murky area, we may well interpret a signal from the configuration level along with one from the intensity level and get a Cperception of intensity, which we can report if we are asked.

It takes a lot of training to be able to Cperceive low-level perceptual signals. Maybe that training is what happens when illusions go away with practice, like Bill's waterfall and my rotary movement after-effects. I have no idea what the mechanism might be, but I'm quite sure it isn't accommodated by a single-valued magnitude of a perceptual signal in the "classical" HPCT hierarchy.

Martin

···

Best

Rick
--
Richard S. Marken PhD
rsmarken@gmail.com

[Bill_Powers1]

[From Bill Powers (2007.09.26.132q7 MDT)]

Rick Marken (2007.09.26.1100)

You and Martin are talking at cross-purposes. In the levels I defined,
“intensity” is the signal coming out of a first-order
perceptual input function. It doesn’t apply to any perception above the
first level. At the first level, there are no sensations, configurations,
and so on – not even an identification of what kind of perception it is.
Only HOW MUCH stimulation is present (whatever the transformation
involved). The word you should be using is magnitude; not being a formal
term in the definitions,that word applies unambiguously to a
perceptual signal at any level of perception. Magnitude means not “a
perception of how much,” but “how much perception” of
whatever it is. The term “intensity” refers specifically to the
first level of perception and not to any other.

Best.

Bill P.

[rsmarken]

[From Rick Marken (2007.09.26.1300)]

Bill Powers (2007.09.26.132q7 MDT)--

Rick Marken (2007.09.26.1100)

You and Martin are talking at cross-purposes.

Boy, are we. Could you straighten us out, please. I don't even know
what we're talking about anymore;-)

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[rsmarken]

[From Rick Marken (2007.09.26.2140)]

Martin Taylor (2007.09.26.14.17)

Me:

>f(s) could be the function that gives the
>lower level perception on which judgments of the intensity of s could
>be based and f(s,context) could be the function that gives the high
>level perception on which judgments of the intensity of s could be
>based. In that case it would be better to write it as p1 = f(s) and p2
>= g(p1, context), where the 1 and 2 are relative (ordinal) levels of
>the perceptions.

And what is the output?

The output at level 1 is p1 and the output at level 2 is p2. If you
are talking about the behavioral output, then the output, o, could be
the number said in a magnitude estimation experiment. In such an
experiment the subject is asked to keep their perception of stimulus
magnitude equal to the size of the number they say. If we call the
perception of number size p(o) then the subject is supposed to control
for keeping p(0) = p(stimulus). But we don't know which perception of
the stimulus the subject is controlling, p1 or p2. The subject might
be controlling for keeping p(o) = p1 or p(o) = p2 or p(o) = some other
perception of the stimulus. I would guess that it's usually p(o) = p2
that is controlled (as seems to have been in your Mueller Lyer study).
All I'm saying is that just because your study found evidence that
subjects control p2 = f(s,context) relative to the perception of their
output (line length estimate), that doesn't mean that there is no p1 =
f(s), a single psychophysical function that maps stimulus magnitude
into a perception of intensity.

I believe that all of this is perfectly consistent with the
hierarchical control model of perception. I don't think it has
anything to do with whether one is conscious of the perception of not.

Best

Rick

Richard S. Marken PhD
rsmarken@gmail.com

[Bill_Powers1]

All I’m saying is that
just because your study found evidence that

subjects control p2 = f(s,context) relative to the perception of
their

output (line length estimate), that doesn’t mean that there is no p1

f(s), a single psychophysical function that maps stimulus magnitude

into a perception of intensity.
[From Bill Powers (2007.09.0304 MDT)]

Rick Marken (2007.09.26.2140) –

You’re still using “intensity” in a non-PCT way. If you’re
speaking about a distance between two points, there’s no way that
perception can be classed as a perception of stimulus intensity.
Distance, yes. Intensity, no. As the distance (measured via physics)
increases, the amount or magnitude of the perception of distance also
increases in some manner.

On the other hand, if you’re perceiving the brightness of a light, the
magnitude of the first-order perception grows with the intensity of the
light impinging on the retina, where intensity is measured in units like
watts per square centimeter. Intensity is a measure of energy flow per
unit area in physics, and it’s the physical meaning of intensity that
goes with the first order of control in HPCT.

I believe this is the problem that Martin is raising. If you use the word
“magnitude” or “amount”, for the number of impulses
per second in a neural signal, avoiding “intensity” as a
measure of a neural signal, the confusion will be removed.

Best,

Bill P.

[rsmarken]

[From Rick Marken (2007.09.27.0715)]

Bill Powers (2007.09.0304 MDT)--

Rick Marken (2007.09.26.2140) --

p1 = f(s), a single psychophysical function that maps stimulus magnitude
into a perception of intensity.

You're still using "intensity" in a non-PCT way.

Yes. I noticed that after I posted it. How about "the function maps
stimulus magnitude into an intensity perception".

I believe this is the problem that Martin is raising.

Perhaps. I don't think so. But we won't find out for a while. I think
Martin is on his island.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Bill_Powers1]

Yes. I noticed that after I
posted it. How about "the function maps

stimulus magnitude into an intensity
perception".
[From Bill Powers (2007.09.27.0828 MDT)]

Rick Marken (2007.09.27.0715) –

That would be OK if the stimulus is an intensity – that is, the amount
of stimulation of a sensory nerve-ending. It wouldn’t be true of a
configuration perception.

Consider a perception of an elliptical shape. From the externalized view,
we might say a person perceives the departure from circularity, or
“eccentricity”. From the internal point of view, we would say
the perceptual signal has a variable magnitude which corresponds to the
amount of eccentricity. Inside the system, the perceptual signal is just
a train of neural impulses with a certain measure – frequency, or more
generally, magnitude.

Under the current rules, you couldn’t say that the input function maps
the eccentricity magnitude into an intensity perception. That would make
eccentricity, a configuration, into a first-order signal, and would be
asserting that a sensory nerve can respond to eccentricity, which is
certainly not true. Intensity is what individual sensory neurons respond
to, under my definitions. This is the current PCT model even if we don’t
plan to keep it forever. But we have to keep playing by the same rules
until someone persuades us to change them, and then we all have to assent
to the new rules. Anarchy is not a good idea in science. I guess that’s
my basic reason for rejecting anarchy, come to think of it.

Best,

Bill P.

[rsmarken]

[From Rick Marken (2007.09.27.0810)]

Bill Powers (2007.09.27.0828 MDT)

Rick Marken (2007.09.27.0715) --

>Yes. I noticed that after I posted it. How about "the function maps
> stimulus magnitude into an intensity perception".

That would be OK if the stimulus is an intensity -- that is, the amount of
stimulation of a sensory nerve-ending. It wouldn't be true of a
configuration perception.

OK. So is line length a configuration perception? If so, then the
perception of line length, s, which I call p1 =f(s) p1, is a
configuration perception. So p2 = g(p1,context) is a relationship
perception. How's that?

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin_Taylor2]

[Martin Taylor 2007.09.27.10.47]

[From Rick Marken (2007.09.27.0715)]

Bill Powers (2007.09.0304 MDT)--

  Rick Marken (2007.09.26.2140) --

  p1 = f(s), a single psychophysical function that maps stimulus magnitude
  into a perception of intensity.

  You're still using "intensity" in a non-PCT way.

Yes. I noticed that after I posted it. How about "the function maps
stimulus magnitude into an intensity perception".

  I believe this is the problem that Martin is raising.

Perhaps. I don't think so. But we won't find out for a while. I think
Martin is on his island.

Not for an hour or so yet.

No, that's not the issue. Here's a different slant on the same issue (though it's not the only problem I have with the idea that there exists a point in the perceiving part of the system where p = f(s). That second issue can wait until I return. I mention it at the end of this message, for reference).

Keeping this very general, and not talking about the specific illusion: Assume that at some point in the hierarchy there exists p1 = f(s), and at another point there exists p2 = g(s, context). Rick's notion, as I understand it, is that there also exists a perception p3 which is some weighted combination of p1 and p2: p3 = h(p1, p2) where h is something like a*p1 + b*p2.

If p3 exists within the HPCT hierarchy, the hierarchy is non-"classical", in that the perceptual connections jump levels. p3 is above p2 and p2 is above p1. Perceptual level-jumping isn't forbidden in HPCT, but it is rarely accepted as normal.

If the function "h" is a variably weighted combination of p1 and p2 (whatever the function), something affects the weights. The presumption has been that with continued exposure or with a different "attitude" (whatever that may be) in the observer, the weight of p1 increases with respect to that of p2. Where does that "something" come from?

If we are dealing wth a "classical" HPCT hierarchy, the "something" comes from the output of a fourth control system -- though it doesn't matter to the argument what that control system is controlling for. In classical HPCT outputs contribute to reference values at the next lower level, and nothing else. So this fourth control system, that uses its outputs to set parameters in the p3 control system's perceptual input function, is not part of the classical HPCT hierarchy.

The fourth control system is an extension of the classical hierarchy in the same way as is the suggestion that some control systems influence the gain of others. Level-jumping in the perceptual side connections can be accommodated, though it's not conventional. Controlled weight changes in computing p3 can not be accommodated within classical HPCT.

That's the issue.

(The second issue is that the stimulus -> neural signal transformation function depends on the temporal and spatial context at the most peripheral sensory level in most sense systems. It is a separate issue that argues against the existence of Rick's immutable p1 = f(s), but it's not the issue I've been trying to argue).

Martin

PS. I'm very unlikely to see any responses to this until the Thanksgiving weeked (southern North Americans, with your milder climate, may not know that Thanksgiving is the first weekend in Octaber).

[Bill_Powers1]

Keeping this very general, and
not talking about the specific illusion: Assume that at some point in the
hierarchy there exists p1 = f(s), and at another point there exists p2 =
g(s, context). Rick’s notion, as I understand it, is that there also
exists a perception p3 which is some weighted combination of p1 and p2:
p3 = h(p1, p2) where h is something like ap1 +
bp2.
[From Bill Powers (2007.09.27.0931 MDT)]

Martin Taylor 2007.09.27.10.47]

You missed a chance to simplify. What Rick should have written, and what
I think he means, is

p1 = f(s) and

p2 = g(p1, context)

The context consists of other perceptions of the same logical type as p1.
Notice that the second function is different from the first, as befits a
higher-level input function.

Perceptual
level-jumping isn’t forbidden in HPCT, but it is rarely accepted as
normal.

Well, we don’t talk about it much, anyway. A little anatomical
circuit-tracing would help here, if anyone could do it. However, my
slight revision above takes care of the level-jumping, doesn’t it?

As to the rest, I hesitate to compound the guessing, but in all
psychophysical experiments that involve verbal report, we know that the
perceptions have to get as high as the category level and probably
higher. Only methods of adjustment, wordless control, do not involve
those higher levels.

Best,

Bill P.

[Bill_Powers1]

OK. So is line length a
configuration perception? If so, then the

perception of line length, s, which I call p1 =f(s) p1, is a

configuration perception. So p2 = g(p1,context) is a relationship

perception. How’s that?
[From Bill Powers (2007.09.27.0948 MDT)]
Rick Marken (2007.09.27.0810) –
Yay! It works for me. And the other perceptions that get
classified as the “context” (so much for perceptual
level-skipping, eh? Martin?) are of the same type as p1. Notice that in
your (and my) formulation, the event level is skipped. Maybe it doesn’t
belong there, or maybe we can use a mathematical trick by defining a
“unity event” which simply transmits the input directly to the
next level without doing anything (pretty far-fetched).

Bob Clark wanted a “temporal” level in there somewhere. Maybe
that’s what we need instead of events. We do, after all, perceive
relationships between events such as before-after, so the idea has merit.
There is still the matter of relationships among configurations (rounder
than), relationships among sensations (sweeter than) and relationships
among intensities (brighter than). Perceptual level-skipping is not
that hard to find. My friend Kirk Sattley asked me once, way back, why it
had to be a hierarchy. Why not a network? I didn’t have a good answer
then, and don’t now. I found hierarchical relationships among perceptions
that make sense, but that doesn’t rule out other kinds. My old brain with
all its defunct neurons boggles; someone else will have to work all that
out. Maybe the hierarchy is simply a subset of the real network. Time
will tell; unfortunately, it hasn’t told me.

Best,

Bill P.

[Martin_Taylor2]

[Martin Taylor 2007.09.27)

[From Bill Powers (2007.09.27.0931 MDT)]

Martin Taylor 2007.09.27.10.47]

Keeping this very general, and not talking about the specific illusion: Assume that at some point in the hierarchy there exists p1 = f(s), and at another point there exists p2 = g(s, context). Rick's notion, as I understand it, is that there also exists a perception p3 which is some weighted combination of p1 and p2: p3 = h(p1, p2) where h is something like a*p1 + b*p2.

You missed a chance to simplify. What Rick should have written, and what I think he means, is

p1 = f(s) and

p2 = g(p1, context)

Yes, my typo.

The context consists of other perceptions of the same logical type as p1. Notice that the second function is different from the first, as befits a higher-level input function.

Perceptual level-jumping isn't forbidden in HPCT, but it is rarely accepted as normal.

Well, we don't talk about it much, anyway. A little anatomical circuit-tracing would help here, if anyone could do it. However, my slight revision above takes care of the level-jumping, doesn't it?

No, because Rick's argument is that the response output is derived from p3 = h(p1, p2), where the weighting function of p1 and p2 is variable.

I'm gone.

Martin

[rsmarken]

[From Rick Marken (2007.09.27.1030)]

Martin Taylor (2007.09.27)

>From Bill Powers (2007.09.27.0931 MDT)]
>

>Well, we don't talk about it much, anyway. A little anatomical
>circuit-tracing would help here, if anyone could do it. However, my
>slight revision above takes care of the level-jumping, doesn't it?

No, because Rick's argument is that the response output is derived
from p3 = h(p1, p2), where the weighting function of p1 and p2 is
variable.

You can't get ahead of Martin (to paraphrase Ring Lardner).

My argument is not that "the response output is derived from p3 =
h(p1, p2), where the weighting function of p1 and p2 is variable". But
things have been shifting around so much that it's really irrelevant.
I think Bill and I are on the same page now and that's good enough for
me. Have fun on the Island.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin_Taylor2]

[Martin Taylor 2007.10.06.11.24]

[From Rick Marken (2007.09.27.1030)]

Martin Taylor (2007.09.27)

>From Bill Powers (2007.09.27.0931 MDT)]
>

>Well, we don't talk about it much, anyway. A little anatomical
>circuit-tracing would help here, if anyone could do it. However, my
>slight revision above takes care of the level-jumping, doesn't it?

No, because Rick's argument is that the response output is derived
from p3 = h(p1, p2), where the weighting function of p1 and p2 is
variable.

You can't get ahead of Martin (to paraphrase Ring Lardner).

My argument is not that "the response output is derived from p3 =
h(p1, p2), where the weighting function of p1 and p2 is variable". But
things have been shifting around so much that it's really irrelevant.

I don't think it is irrelevant at all, if we are interested in figuring out what is happening in the illusion. If I misinterpreted your model, I'm sorry. I was going by this [From Rick Marken (2007.09.25.0915)]:

In my model of perception, the effect of context is included in perceptions at a higher level; perceptions created by perceptual functions that take the outputs of lower level perceptual functions as inputs. So the perception of line length in the context of other lines is created by a perceptual function that takes Psi values (call them Psi1 to indicate that they are level 1 perceptions) as inputs. I assume that the output of these higher level functions, which we can call Psi2, are the basis for judgments of line length when we see context affecting these judgments. So a higher level perception that is used in judging line length in the context of other line lengths (as in your Mueller Lyer study) might be Psi2 = g(Psi11, Psi12,...Psi1n). What I plotted in my graph as the Psi values for the Arrowhead and Feather contexts were actually Psi2 values; I just picked g() functions that would transform the log s (Psi1) values to give the right results. The log function that produces the Psi1 values is the same for both the Arrowhead and Feather context.

My model of perception assumes that the low and high level perceptions (Psi1s and Psi2s) are present simultaneously. Subjects who show a pronounced context effect are mainly paying attention to the higher level perception, Psi2; those who do not show pronounced context effects (or who learn not to show them) are able to focus on the lower level perception, Psi1.

What you said sounds to me very much as though to make the single judgment of inter-point distance, people use both Psi1 and Psi2, with different weights for different people (or possibly at different times?). Maybe you've altered your model, which is fine. Nobody cares what you or I or anyone else thought. What we care about is what is actually happening, so we can get the theory right.

However, for the time being, let's go with your model as expressed in the quoted passage. That's the model that I argued to be inconsistent with classical HPCT, for the reasons in my message to whioch you responded with the dismissive: "You can't get ahead of Martin (to paraphrase Ring Lardner)."

My problem right now is that the quoted passage seems to me to be incompatible with:

"My argument is not that "the response output is derived from p3 =
h(p1, p2), where the weighting function of p1 and p2 is variable"."

Maybe that is not incompatible with

"My model of perception assumes that the low and high level perceptions (Psi1s and Psi2s) are present simultaneously. Subjects who show a pronounced context effect are mainly paying attention to the higher level perception, Psi2; those who do not show pronounced context effects (or who learn not to show them) are able to focus on the lower level perception, Psi1."

But if it's not incompatible, then I understand neither.

···

----------------------

If Rick's model is correct, then strict HPCT isn't, and that's the important point. Making a dismissive comment doesn't make the issue go away.

I do have an ulterior motive for pursuing this, because Rick's multi-level model seems to be entirely consistent with a PCT model I mentioned a few years ago (and which I still rather like, though I haven't mentioned it recently). At present my model is only a sketch, though we did discuss some possible mechanisms when I originally suggested the model.

My model is strict PCT, but it differs from strict HPCT in that instead of a number of logically based levels above "category", there is one completely analogue HPCT-style hierarchy, beside which lies a "category interface" to a parallel multi-level organization of analysis and control that takes input and provides output in parallel to all the levels of the analogue hierarchy. In that model, it's perfectly reasonble to form a perception from a weighted copmbination of low and high-level analogue perceptions, and even to feed the output as a reference value to yet another control system at an arbitrary level in the analogue hierarchy.

Some day, I'd like to pursue the implications and mechanisms of that model, which is based on the BLC model of reading (I.K.Taylor and M.M. Taylor, "The Psychology of Reading" Academic Press, 1983). I haven't done so, and the idea has lain fallow in my mind for many years, but Rick's model for perceived distance brought it forward again, whether or not I misunderstood his careful explanation.

Martin

[rsmarken]

[From Rick Marken (2007.06.1700)]

Martin Taylor (2007.10.06.11.24) --

What you said sounds to me very much as though to make the single
judgment of inter-point distance, people use both Psi1 and Psi2, with
different weights for different people (or possibly at different
times?).

Sorry, I guess I wasn't clear. Maybe I'll just used English. I think
that the Mueller Lyer (or any other perception) exists as many
perceptual signals, at many different levels, simultaneously. When you
are asked to judge an inter-point distance, I think there are several
different perceptual signals that could be used as the basis for this
judgment: the signal carrying the perception of distance between the
points, the signal carrying the perception of the relationship of this
distance to other adjacent distances, etc. Some subjects may be able
to use just the distance signal; these subjects would show no "context
effect". Others seem to use the relationship signal; this shows the
context effect. My guess if that there is only one psychophysical law
relating physical distance to the distance perception. The context
effect that you observed in your experiment doesn't, therefore, mean
that there has been a change in the psychophysical relationship
between distance and distance perception. The context effect simply
represents a judgment of a different perception; not the distance
perception but a relationship perception at a higher level that
happens to depend on the distance perceptions. This is why I think you
experiment doesn't rule out the exitence of a single psychophysical
law relating physical to perceived distance.

Maybe you've altered your model, which is fine. Nobody cares

However, for the time being, let's go with your model as expressed in
the quoted passage. That's the model that I argued to be inconsistent
with classical HPCT, for the reasons in my message to whioch you
responded with the dismissive: "You can't get ahead of Martin (to
paraphrase Ring Lardner)."

Sorry, didn't mean to be dismissive. Just wanted to show that I
sometimes actually read some great literature.

If Rick's model is correct, then strict HPCT isn't

Again, I don't see how what I have described is inconsistent with the
hierarchical control model of perception. Indeed, I think that it is
consistent with one of the most important properties of that model;
that we perceive the world at all levels simultaneously. This is very
different than conventional models of perception which imply that you
either perceive something one way or the other; either context has an
effect or it doesn't, for example. Sometimes this may be true.
Mutually exclusive perceptions certainly exist, but they are usually
at the same level (the wife/mother-in-law configuration, for example).
But in your experiment I think it's possible to just line length in
terms of the lower order distance (configuration) perception or in
terms of the higher level context (configuration) perception. The fact
that some subjects "overcame" of were not affected by the context when
judging line length suggests that at least some people can learn to --
or just tend to -- ignore the relationship perception when judging the
distance between points.

But who knows.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com