[From Bill Powers (920318.0800)]
Rick Marken: WHAT'S HAPPENING WITH THE SAN DIEGO MEETING???
···
-------------------------------------------------------------------
Bruce Nevin (920317) --
Emotion:
The term "emotional" isn't used much in any technical sense -- in science
it's usually a pejorative meaning "you're letting considerations other than
scientific truth into the argument." It also means "impulsive", or acting
without higher-level reflection. In neither case do I think that anything
is running open loop. Even the cockroach's escape "response" is an act that
removes the danger. If it didn't it wouldn't have been learned or wouldn't
have evolved.
------------------------------------------------------------------
Semantics:
One comment on your semantics remarks to Avery. The idea that words somehow
refer to things "out there" is a SUBSTITUTE for the epistemology of control
theory, in which the only knowable referents of words are perceptions IN
HERE, which, of course are attributed to an objective external reality.
----------------------------------------------------------------------
Correlations:
... correlation of an outcome with a purpose or goal smells >unscientific
from a conventional perspective because the goal is a >perception hidden
within the black box of the behaving system and the >controlled outcome is
that subset of the possible sensory inputs to the >behaving system that are
relevant to that goal.
True. But such opinions are based on a misunderstanding of science that
assumes we mustn't guess about things we can't see. If that were how
science really works, we wouldn't be able to explain why holding a magnet
near a television set distorts the picture. If you guess carefully and
quantitatively, you get a science like physics. If you guess sloppily and
without high standards for accepting models, you get ... well, what you
get.
[I'll return to "missing the point"]
In your reiteration of basics, you say:
Ordinarily, we would expect action and disturbance to correlate
with the outcome.
Do you mean the algebraic sum of action and disturbance correlates with
the outcome, while neither correlates well, taken separately?
I meant that ordinarily we expect effects to correlate statistically with
their causes. If there are multiple causes, like action and disturbance, we
would expect each one to "contribute to the variance" of the effect. This
is the bread and butter of statistical analysis. Conventional behavioral
science is not prepared to recognize the case in which adding two causes
together produces less effect than when either alone is used.
What am I misunderstanding here?
Not much. I'm using the term "variance" in the statistical sense, not
simply to mean "variation." In the statistical sense, variances don't add
"algebraically." They add "in quadrature", meaning they add as the sums of
the squares of the individual average variabilities. Statistical measures
don't give significance to single data points.
In a tracking experiment, if you add the individual disturbance magnitudes
to the corresponding action magnitudes point by point, you get the
successive cursor positions. That's the algebraic sum you're talking about.
To characterize the quality of control for an entire run, thousands of sets
of data points, you can use statistical measures, like mean squared
amplitude of the variables. The entire data set is used to obtain
variances. You can compute the variance of the entire set of thousands of
handle positions, the variance of all the disturbance magnitudes, and the
variance of all the cursor positions. If there were no control, you would
expect the sum of the handle and disturbance variances, the mass measures,
to equal the cursor variance, another mass measure using all the cursor
data points. When there is control, the actual cursor variance is much less
than the expected variance -- the expected variance is 10 to 15 times the
standard deviation of the actual cursor variance in a normal tracking
experiment.
The only reason I have used these statistical concepts is to demonstrate to
those who use statistics as the preferred way of analyzing data that
something very strange is going on here. We've gotten a bit off the track
by talking about how you can detect controlled variables when the reference
level is randomly changing, using statistical methods like Marken's "mind-
reading" demo. That's really doing things the hard way. While we may
eventually have to do things the hard way, right now there are plenty of
aspects of behavior we can study in which there aren't large unpredictable
changes in reference signals -- especially if we design experiments to
avoid them. In that kind of experiment, the statistical treatment is
unnecessary, except to satisfy those who don't believe data unless they can
see a correlation and a confidence level.
---------------------------------------------------------------------
Missing the point:
When I use ill-advised expressions like that, I deserve the
misunderstanding they generate. I think you and Avery know enormously more
about control theory than any of your colleagues do, or for that matter
than 99.99% of behavioral scientists do.
I was referring to something much narrower, namely, the point that I was
trying to make about the difference between talking FROM the point of view
of a theory and talking ABOUT the point of view. Bringing up the fact that
your analysis and Avery's differ was an attempt to point toward a point of
view superordinate to both. No takers yet. There's a hint of taking me up,
however, in the following:
Recourse to nonverbal perceptions in the perceptual hierarchy as the
universe of "meanings" promises a way out of this circularity [of
metalanguage]. The question remains: if the work can be done within
language (with slight extensions to enhance regularity), why have
recourse to a metalanguage purportedly outside the familiar language of
words organized in sentences and texts?
I'm guessing that you're referring here, obliquely, to Avery's structural
diagrams, which are not expressed in natural language. But can the work
really be done within language? Doesn't it ALL have to be done within
meaning? And isn't a structural diagram as good a way to refer to a meaning
as a lineal sequence of words -- and in some cases, better? Diagrams in two
dimensions would be best for referring to meanings in which more than one
dimension of relationship exists at the same time. And lineal ordering is
best for dealing with meanings in which time or sequence is a dimension. I
don't know of any good way to refer explicitly to the structure of a
program (other than just laying out the source code) except with a block
diagram or some diagrammatic way of showing alternate paths at the same
time.
Conventions and universals:
There are some aspects of language that appear to be universal. There
are other aspects that are defined by social convention. These
conventional aspects of language are not universal.
The universal aspects, I presume, are universal because they arise from
human properties common to all people, not statistically but without
exception. If you have discovered universal aspects of language, then they
are universal, period. Is this what linguistics has discovered? If so,
there's no room for argument, is there?
... having reached the desired "state where linguistic universals
are trivially deduced from first principles," PCT still has to describe
or account for those remaining aspects of language that are socially
inherited and which do NOT follow from first principles.
PCT has no specific behavioral content, which is why it applies to all
behavior. It doesn't tell you what variables people will control for, what
perceptions they will become organized to experience, how stable their
control will be, or what environments their actions will occur in. After
you fill in the unknowns, it will tell you something about the consequences
of your choices of variables and constants. So control theory has nothing
to say about social conventions, linguistic or other.
<The means in the perceptual control hierarchy for learning, >maintaining,
and orienting one's behavioral outputs to social >conventions or norms are
presumably universal.
They are also the means for flouting, changing, and ignoring social
conventions. Social conventions are descriptive, not prescriptive. They
become prescriptive only when an individual accepts them as a means of
achieving something, and what they prescribe is only what the individual
takes them to prescribe. What I'm trying to emphasize is that the
underlying machinery of the brain does more than language, and exactly what
it does with language is optional (even though, through interactions,
people tend to converge toward similar uses of language).
--------------------------------------------------------------------
Expansions:
[Expansion] uses at each point of word entry (operator entry) only the
relatively small stock of words and word relations that are attested in
many perfectly ordinary sentences of the language.
It uses at each such point only reductions of word shape that are
attested as minimal sentence-differences between pairs of perfectly
ordinary sentences, such that the sentence-pairs informally meet my
judgment of saying the same thing ...
This informal judgment brings in meaning: "the same thing" is "the same
meaning." The terms, in addition to satisfying the mechanical or
statistical word-relationship requirements, point to the same meaning as
the reduced form. But if you can make this judgment, the implication is
that the reduced form has the same meaning as the first expansion, the
first has the same meaning as the second, and so on. Otherwise each
expanded form would be adding meanings, not having the SAME meaning. This
supports my scenario, not yours.
Yours:
Heard (reduced) form ----> expansion rules ----> Expanded form
> >
> >
initial meaning <-- The same? --> expanded meaning
Mine:
Heard (reduced) form Expanded form
> >
> >
initial meaning ---> imagination ---------> expanded meaning
So in your scenario, a rule is applied BY THE HEARER to the reduced verbal
form, which has limited meaning, to produce an expanded verbal form, which
has a more complete or explicit meaning. In mine, no such rule is applied
BY THE HEARER.
What I'm suggesting is not that the transformation rules you propose are
incorrect or unobservable, but that they may not be related to speech
comprehension as you imagine them to be related. The situation I imagine
could be represented this way:
Linguist--> Expansion rules
/ \
/ \
/ \
/ \
/ -------------------
/ \
/ \
Heard (reduced) form Expanded form
> >
> >
initial meaning ---> imagination ---------> expanded meaning
The linguist can certainly derive objectively defensible expansion rules
which, applied to the reduced verbal form, will yield expanded verbal
forms. But the observer can only imagine that the hearer is actually
applying those rules. If the hearer is doing the expansion as I suggest,
then the rules seen by the observer analogize but do not describe what is
going on perceptually in the imagination of the hearer. The processes I
label "imagination" explain why the reduced form and the expanded form are
related as they are observed to be related. They APPEAR to be related by a
rule in which words, not perceptions, are the arguments. But in fact, in
the hearer, words are not involved in the expansion until it is finished. I
say "in fact," but of course this is just a proposal.
In order to distinguish your scenario from mine, and show that the hearer
is actually applying the rules, it's necessary to do more than show that
the rules fit what is observed. It's possible to devise a set of rules that
will describe any process in words or symbols, even if that process uses
neither words nor symbols. Every computer simulation of a physical process
illustrates this fact. Nobody believes that the computing steps are
actually employed in the physical process. They are simply equivalent in
their final effects to the processes actually at work.
I'm not picking on you: the same alternative applies to Avery's diagrams.
By seeing these approaches to representing language structure as analogies,
I think one can get a better sense of the machinery underlying language
phenomena (and behavioral phenomena in general). There must be a
relationship between the linguistic analyses and that underlying machinery.
But I doubt very much that it's a one-to-one correspondence -- that
linguistic processes are specifically about words, or that they are
actually carried out in words.
I'm not trying to put down anyone's approach -- just to get a higher-level
look at what's going on.
------------------------------------------------------------------
Best,
Bill P.