[From Rick Marken (951227.1000)]
Shannon Williams (951222) --
At long last, here are my comments on your "Turing" post.
You say that your goal is to:
Write a program that would cause a computer to respond to questions
like a human would.
I take this to mean that your goal is to write a program that responds
to questions for the same reason that people respond to them; in order
to control perceptions.
Sentences input to the computer will be considered perceptual inputs
from reality.
OK.
Computer responses will be based on previous sentences input to the
computer.
If the computer is simulating a control system, then its response
should be based on the difference between reference, r, and actual
perception,p. Previous sentences (S[t-N]....St[t-1]) could be the
input to the perceptual function that computes the controlled perception
(p = f(S[t-N]....St[t-1]) but computer responses must based on the
difference, r-p.
If sentence associations conflict, then error condition.
I think of conflict as the attempt by two control systems to maintain the
same perception at different reference levels. I don't see how there can
be this kind of (error producing) conflict between sentence associations.
I think you are using "conflict" in a sense that I don't understand.
Sentence structure would be determined by words that stay constant
in regular sentence positions.
This seems to be a description of a perceptual function; one which takes
a sentence as input and produces an indication of "structure" as output;
a structure perceiving function. This suggests that your question
answering Turing machine is controlling a perception of sentence
structure. This would be interesting.
Sentence concepts must be converted into blocks of associated
concepts.
This is another possible controlled perception; sentence concept. Even
more interesting.
I would expect sentences like 'The kid walked home' to be associated
with 'person', 'young', 'school' (maybe), 'strolled', 'daydreamed',
'building', 'house', etc.
Now you seem to be describing a perceptual function that maps one
input (sentence) into several outputs (concepts). I don't see how a
control system can be organized around this type of perceptual
function. What we need are perceptual functions that take a
sentence (or several sentences) as input and produce perceptions that
vary on a single dimension as outputs; these perceptions can then be
compared to reference inputs that can vary on the same dimension as
the perceptions.
IN PCT TERMS:
loop 1:
controlled variable = The input sentence structure is like a sentence
structure in memory.
I would say "sentence structure" is the controlled variable.
perception = sentence input.
The perception and the controlled variable are the same thing. The
sentence input, to the extent that it contributes to the perception of
sentence structure, is a disturbance; an independent (of system
output) contribution to the output of the perceptual function.
output = if input structure cannot be found in memory, then add
this to memory, and process according to error condition.
The output should counteract disturbances (such as the sentence input)
that contribute to the value of the controlled perception. If the
loop is designed to control a perception of sentence structure, system
output should influence sentence input in such a way that the
perception of structure of that sentence is moved toward the reference
structure. For example, the output might change "I" to "me" in the
sentence "Give the money to John and I" ; this sentence might be
perceived as "subject as object", compared to a reference ("object as
object") and the output generated to remove the error.
In principle, the most difficult part of building a system to control
complex variables, like sentence structure, is building perceptual
functions that can detect these aspects of the sentence. According
to PCT, these complex perceptual functions take the outputs of lower
level perceptual functions as input. That's why "jumping into the
middle of the hierarchy" like this is a big problem; we don't know
much about how people control "simple" aspects of sentences (like
word configuration) so it's difficult to just jump right in and develop
models that control percetions that are probably functions of these
lower level perceptions.
Nevertheless, the sentence structure control system would be worthwhile,
I think, just as a demonstration of principle. I am working on
experiments that demonstrate control of complex perceptions, like
sequences and programs. And I can simulate the behavior of people in these
experiments; I don't actually build the functions that can perceive a
particular sequence or program; I just take advantage of the fact that
I know when a particular sequence or program is occurring (because I
know the program code) and I just insert this information into the model
as though it were the output of a perceptual function.
So you can finesse the perceptual function problem just to get a working
model of a system that can control an "interesting" perception, like
"sentence structure".
You also seem to be interested in building a system that "learns" to
control some perception. I think it is important to first understand the
behavior that is being learned (control of sentence structure, for example)
before we start trying to develop learning algorithms. Learning is
also a control process; it's not particularly hard to add learning
to a control model; but we have to start out with _some_ structure.
So I suggest you try to build a very simple sentence structure control
systems first -- then you can incorporate various learning algorithms.
For example, you can have learning algorithms that alter the perceptual
function and/or the output function in order to achieve better control of
sentence structure ("better" as determined by the "learning" control
system). But first we have to have a perceptual and output function so
that the learning system has something on which to operate.
Best
Rick