[Martin Taylor 960115 18:40]
Since Shannon brought up the topic of Language in her "How to make a Turing
machine" posting of Dec 22, I've been hoping to be able to be able to make
a comprehensive posting on language. To do so, I saved some 15 message from
different posters. But it's clear that I don't have time to do it, and
since I'm going to be away for a couple of days, probably, I want to post
a shorter note before the memory of the topic dies away completely.
Shannon originally talked about a system that would understand language
input and produce contextually appropriate language output (Searle's
"Chinese Room" problem). Someone else quite rightly mentioned that there
is a significant "grounding problem" with what Shannon wanted to do. (I'm
not convinced that the problem is insoluble, but it's not an issue here).
Bill P. talked about the control of perceptions involved in generating the
language output in the form of acoustic waveforms, gestures, keystrokes,
or whatever. A different problem, but, like Shannon's problem, part of
what is needed for language.
It seems to me that a "proper PCT" view of language has to start with
language as an output that influences a perception that is to be controlled.
In other words, the focus should be on what is being controlled that has
language as one of its possible effective actions.
In controlling our perceptions, we act on the world. Other people are part
of the world on which we act. If the actions of someone else can affect
some perception (P) we are presently controlling, then to influence
those actions appropriately will be one way of controlling perception P.
Language is one way to influence the actions of another person.
Communication through language is the control of perception, where the
perception being controlled is of some state of another person. If I want
to send you a message, I do so because I want you to do something or to
know something. Only by your actions can I perceive what you are doing or
(problematically) what you know. Your actions are the inputs to my perceptual
functions whose outputs correspond to your doing or knowing what I want
you to do or know (or not, as the case may be). In using language, I
affect the values of those perceptions because you respond (and I can use
no other term) as I would wish (or not).
The originator of a message wants to achieve a certain state in the recipient.
What does the recipient want? If the recipient has a reference to be
cooperative (which includes that the originator should perceive a state of
cooperativeness) then the recipient has a reference perception to see a
certain state in the originator--that the originator is satisfied that
the recipient has correctly interpreted (and perhaps acted upon) the
message sent; the message sent being the desired state of the recipient.
That's a pretty complicated statement. Let me rephrase it so say that the
recipient has two important perceptual signals: a controlled perception
of the originator's state of satisfaction, and an uncontrolled perception
of what it is that the originator wants. To control the first of these
two perceptions, the recipient must affect the originator's state of
satisfaction, by doing something the originator can perceive--typically
by originating what is sometimes called a "feedback message."
The recipient therefore must have a reference to actually determine what
state the originator wants to see (or what the originator wants the recipient
to do). So the result of a successful communication is that the originator
has satisfied the reference perception that the recipient should be in a
particular state, and the recipient has satisfied a reference perception
that the originator is satisfied. To do that latter may not mean that the
recipient actually IS in the desired state, but at least the recipient
must ordinarily have determined with reasonable correctness what the
desired state is.
Each partner is controlling a perception (or set of perceptions) for which
the inputs are solely the actions of the partner, language actions or
otherwise. Now the problem can be reduced to the mechanisms whereby those
perceptions are controlled. And these mechanisms are what Bill P. addressed
very well for the originator of the message, in one posting, and perhaps
what Shannon was originally getting at for the recipient.
My way of treating the mechanisms is called "Layered Protocol" theory.
Briefly (code word for "watch out":-), the state in which either partner
wants to see the other is termed a "primal message". It is a possibly
complex set (pattern, Shannon) of reference perceptions relating to the
partner. The current state of the partner is compared with the primal
message, and the difference encoded (PCT output function) into zero or
more "virtual messages" which are are reference perceptions for states
of the partner which, if achieved, would bring the partner closer to the
"primal message" reference state. These "virtual messages" are "primal
messages" for lower-level control units. And so on, down to a level
at which only the perceptual control internal to the actor for the purpose
of making the right noises or gestures is of interest.
The recipient's "feedback messages" are likewise "Primal messages" for
lower-level control units whose outputs affect the originator (in other
words, the recipient at one level may well become the originator at a
lower level).
We call these language-specific control units "protocol nodes" rather than
control units, because the signals in them are more complex than simple
scalars, even though we assume that they are actually composed of many
ordinary scalar ECUs (or the equivalent). Protocol nodes are arranged
in a hierarchy just like the HPCT hierarchy of ECUs, and I presume that
if it exists, it is actually only a simplified description of a part of
the HPCT hierarchy.
Bottom line: control of language perception is not only control of the
noises and gestures of language. Much more importantly, it is control of
the perception of the states or actions of another person who is (in a
cooperative situation) controlling for seeing the satisfaction of the
originator of the communication. Control of these complementary perceptions
has been eased by the evolution of a range of protocols (such as agreed
meanings of particular words, or of conventional forms of argument). The
normal processes of reorganization stabilize these protocols, since they
get changed when they don't work, but remain stable when they allow the
partners effective control of their language-related perceptions.
Too short, too complicated, but all I have time for. Wanted to get it out.
Martin