Natural Language through Abstract Memory (July 1979)

by Arthur T. Murray


ABSTRACT


A model  is proposed for the functioning of natural language in a mind.  The
model results from investigating and combining information-processing of two
types:  that of inexact patterns (e.g., visual images). and that of symbolic
code (i.e., natural language).  First an underlying  model is  presented for
the  input/output  data-flows  of  the sensorium/motorium of the human mind,
including its memory.  The underlying model is  that of  a highly orthogonal
grid, with  input and  output flowing  vertically, in  parallel but opposite
directions,  while  associative  interconnections   flow   back   and  forth
horizontally.    In  the  superstructure  model  for  natural  language,  an
"abstract" memory channel is  superimposed  to  flow  in  parallel  with the
input/output channels,  in such  a way that a spiral of habituation comes to
dominate  the  associative  crossflows   in   a   conscious,  linguistically
generative process of "transabstractivity."


          1.  INTRODUCTION

This  article  proposes  a  two-tiered  model for the functioning of natural
language in a mind.   The  two tiers  derive from  the need  to have  both a
concrete level of sensation and an abstract level of mental processing.  The
concrete and abstract  levels  correspond  to  a  duality  in  the  types of
information to  be processed  in a  mind:   inexact patterns, such as visual
images, and symbolic code, i.e., natural language.  The lower-tier  model of
pattern-recognition  is  necessary  as  an  underlying assumption before the
upper-tier model of habituated linguistic control  can be  developed.  Since
the main  endeavor in this article is to present the psycholinguistic model,
the author begs the reader's indulgence if the  model of pattern-recognition
seems far-fetched or is erroneous.


          2.  OVERVIEW OF THE ASSOCIATIVE GRID OF THE SENSORIUM/MOTORIUM

The  underlying   model  organizes   the  input/output   data-flows  of  the
sensorium/motorium of the human mind, including and emphasizing  its memory.
The rationale is to organize the data-flows as a prelude to controlling them
with a "transabstractive" grammar system.    The  model  organizes  the data
flows into  a relatively  flat, highly orthogonal grid or array.  The inputs
and outputs of the sensorium/motorium flow vertically up and down  the grid,
in respectively parallel but opposite directions.  For the remainder of this
article, the inputs  of  the  sensorium  are  to  be  visualized  as flowing
vertically downward  on the  left side of the non-abstract, "concrete" grid,
while the outputs of the motorium flow vertically upwards on  the right side
of the  grid.   The interface with the outside world is therefore at the top
of the grid, consisting of sensors on the left and motor musculature  on the
right.  After pre-processing, the input/output data-paths descend down their
respective halves of the  grid through  experiential memory  channels on the
left and  motor memory channels on the right.  There is a strict parallelism
of the various memory channels lying vertically in  the grid.   This article
is  concerned  mainly  with  interactions  among  the  memory  channels, and
accordingly the grid is to be visualized as consisting mainly  of elongated,
vertically parallel memory channels.

     Interconnecting  associative  "tags"  flow  horizontally back and forth
among the memory  channels  in  the  concrete  grid.    The  grid  is highly
orthogonal because information can change its path only at right angles.

     The   horizontal   associativity   is   vertically   analogous   to   a
time-dimension.  The gridwork  of memory  is gradually  filled from  the top
downward towards  the bottom over the psychic lifetime of the organism.  The
lavishly available memory space  is  genetically  "hardwired,"  so  that the
activity  of  mentation  has  only  to travel gradually downward through the
tabula rasa of the  memory grid.   Any  line, drawn  horizontally across the
grid so as orthogonally to intersect the various memory channels, represents
a unique moment in the experiential history of the organism.

     From this point on, the article follows a method of  circumscription by
first merely outlining a general system and then addressing those particular
elements within that system  which are  relevant to  the development  of the
"transabstractive"   superstructure   of   natural   language.    The  first
circumscription is to drop consideration of the right-hand motor side of the
underlying, concrete  associative grid.  Suffice it to say that motor memory
does figure in the genesis or development of a holistic  model of  the mind,
especially as  regards volition,  but that  this article concentrates on the
passive, experiential, "left" side of the  model  as  the  arena  over which
natural language operates.

     In the  holistic visualization, the passive left side of the mind-model
contains a separate but  parallel memory  channel for  each of  the distinct
senses which  a mind  might possess,  such as the five commonly acknowledged
human senses.  An important facet of this general model is the idea that any
conceivable sense  could be  added on  in parallel  to widen the associative
grid, as long as the principle  of  orthogonality  were  not  violated.   An
experiential memory channel here is a bidirectional data-path originally and
permanently filled in the downward direction from top  towards bottom.   The
temporally successive,  horizontal associativity  of the quasi-concrete grid
absorbs as many sensory  memory  channels  as  evolution  (or  artifice) can
provide.    A  full  complement  of  unimpaired sensory channels provides an
organism with full breadth of multi-sensory  experience.   However, a person
can  have  severe  sensory  impairments  and  still  function  as  a  highly
intelligent  mind.    This  discussion  serves   to  introduce   the  second
circumscription of  this article, namely that from this point on the article
will develop the superstructure for natural language solely in terms  of the
senses of  vision and  audition.   The aforementioned duality of pattern and
symbolic code is to be discussed using vision to treat  pattern and audition
to treat code.


          3.  THE MODEL OF THE AUDITORY CHANNEL

Now begin  descriptions firstly  of the auditory channel, next of the visual
channel, and then  of  the  "abstract"  memory  channel  which  connects and
controls the other two in "transabstractivity."

     This general  model attempts  to follow the human mind, however murkily
known, but it is directed towards a hardware  construction of  an automaton.
Therefore  the  discrete  memory  mechanisms or components are visualized as
physical hardware rather than as biological tissue.  The basic model is that
of  perhaps  an  electronic  multivibrator  or  even of an electromechanical
latching relay.  The basic requirement of a  tabula-rasa memory  "cell" here
is  that  it  shall  retain  permanently  its status of either "on" or "off"
("full"  or  "empty')  when  a  brain-wave  type  control-pulse  orders  the
unerasable "hardening" or fixation of whatever data happen momentarily to be
flowing through the cell and its associated cells.  The  cell shall  also be
associatively  "taggable,"  typically  as  part  of  an aggregate, so that a
unitary  associative  tag,  leading  orthogonally  away   from  the  channel
containing the  aggregate, can  either control and activate the aggregate or
be controlled and activated by the  aggregate.    Thus  data  are  laid down
permanently within  a channel,  but their  informational content  is free to
move either bidirectionally within the channel or orthogonally away from the
channel.

     The auditory  memory channel is designed here as the target destination
of multitudinous transmission lines carrying the  component data  of sounds.
Although  the  holistic  visualization  of  this general approach contains a
tentative scheme for clusters of short-term memory loops at the  entrance to
the  auditory  memory  channel,  this  article  requires only the long-term,
permanent memory  channel for  the discussion  of "transabstractive" natural
language.

     Within our  circumscription we  are examining the left side, or passive
experiential half, of  the  flat  associative  grid.    The  auditory memory
channel is  to be  visualized as  itself a flat cable lying both flat in the
plane of the grid and on the right side of this experiential  half.   We are
developing  a  "transabstractive"  area  bordered  on the left by the visual
memory channel and on the right by the auditory memory channel.

     Our intent with the auditory memory  channel is  to deposit  engrams of
sounds  and  phonemes.    The  channel consists of thousands of transmission
lines flowing vertically downward  in strict  registry and  in parallel with
the other  memory channels.   Each engram-aggregate is a horizontal slice of
memory "cells" as nodes upon the  transmission lines  of the  channel.  Each
slice has  (at least) two genetically-provided, tabula-rasa type associative
tags exiting the channel.  One class of associative tags  exits horizontally
and orthogonally  out towards  the visual  channel on  the left.   The other
class,  which  we  might  as  well   call  the   "habituation-class,"  exits
perpendicularly and rises orthogonally out of the associative plane en route
to interaction with a grammar habituation  system, of  which the description
comes further on as the main purpose of this article.

     This auditory  memory channel is designed to remember words or parts of
words as strings of phonemes.  A phoneme  is modelled  as a  distribution of
activated nodes  within an engram-slice.  It may actually require several or
many slices  in a  row to  comprise one  time-extended phoneme,  but for the
purpose of clear simplicity we will treat the single phoneme as if it were a
unitary engram-slice subject to unitary associative tagging.

     All sounds and words consciously heard  are deposited  automatically as
engrams  within  the  auditory  memory  channel,  but  only  those words are
"learned" to which access is gained and maintained via the associative tags.
In  our  model,  we  are  discussing  the  learning  of  language-words with
reference to visual images resting in the visual memory channel,  which lies
in parallel  with and  to the  far left of the auditory channel.  The visual
channel, too, has slices  and horizontal  associative tags,  attached to the
image-slices.   (The feature-extracted  image-slices would perhaps be rather
unrecognizable if memory-dumped from  the system.)   A  word or  morpheme is
learned, in  reference to  vision and all other senses, through a process of
associativity  based  upon  simultaneity   in  the   permanent  fixation  or
"hardening"  of  the  horizontal  associative  tags  which  interconnect the
various sensory memory channels.

     Let us summarize the  function  of  the  auditory  memory  channel with
respect to words and morphemes.  As a string of phonemes, a word or morpheme
is learned  passively when  horizontal associative  fixation enables another
sensory  channel  to  evoke  the  activation  of  the phonemic string at its
permanently fixed location within the auditory memory channel.   An incoming
(erstwhile potentially  bidirectional) tag  which starts the activation of a
phonemic string is to be described both as a "recall-tag" while en route and
as an  "onset-tag" at its point within the auditory memory channel where the
tag activates the  first  phoneme  in  the  phonemic  string  of  a  word or
morpheme.

     Here  mention  must  be  made of a tentatively proposed "string-effect"
within the auditory memory channel.  Morphemes accessed by an onset-tag will
go  into  full  serial  engram-slice string-activation either until the last
phoneme in the string shunts the  activation-flow orthogonally  out along an
exiting   associative   tag,   or   until   the  strong  activation  of  the
"string-effect" fades abruptly at the end  of the  string.   (For remembered
music,  the  string-effect  might  go  on quite long.)  The string-effect is
therefore a tendency of related and  contiguous memory  slices to  fire in a
serial string.

     When a  morpheme in  auditory memory  has been accessed by an onset-tag
laid down at some  time in  the past,  the phonemic  string of  the morpheme
instantaneously flows  serially down  through the  transmission lines within
the auditory memory channel and is deposited anew at the  freshest extremity
of the  gradually downward-travelling "front" of auditory memory of auditory
experience.  This creeping "front" of  auditory experience  is in horizontal
registry with  the other  parallel channels  that are  all moving or filling
downward into the tabula-rasa area of the associative memory grid.

     While an accessed  morpheme  is  flowing  through  the  auditory memory
channel,  each  individually  activated  transmission  line  is  momentarily
energizing or stimulating each of its  myriad cell-nodes  which have  in the
past been  positively fixated  in the  status of "on" or "full" as mentioned
above.  Thus the initial activation of one phoneme, morpheme, or word in the
channel  can  "tickle"  or  predispose  the  activation of myriad other such
auditory engram-aggregates.   We posit now a differentiation-process wherein
the relatively  largest and strongest summation of individually firing nodes
within the engram-slices of a single, aggregate memory trace can  cause that
memory trace  to win out in a sort of competitive race to be the first among
other memory traces responding to the activation of  the originally accessed
memory  trace.    (Non-activated  nodes  may act inhibitively to enhance the
differentiation.)   The  competition  for  primacy  of  response  is quickly
happening  amid  these  various  parameters,  not  the least of which is the
time-factor,  because  the  extremely  free  and   mobile  associativity  of
mentation will  quickly go  the path of least resistance without waiting for
any delayed response.

     By now the model has described  how  an  extraneous  recall-tag  from a
sense such  as vision can activate an auditory engram with two results:  the
old auditory memory is laid down  anew  at  the  freshest  extremity  of the
auditory memory  channel, and a highly similar trace in the auditory channel
can be activated in response to the activation of the  first trace.   When a
trace does respond, it outputs a signal along an associative tag which is to
be called here  an  "ultimate-tag,"  because  the  tag  is  attached  to the
ultimate phoneme  in the  phonemic string.   Of  course, that "ultimate-tag"
from audition can be a "recall-tag" going over into vision.   Thus  a visual
image can  associatively access  a morphemic  word, which  can stimulate the
response of  an identical  or similar  word at  a different  location in the
auditory channel,  with the result that a totally different visual image can
then be accessed back again in the visual memory channel.  Associativity can
loop in and out of memory channels in an untrammeled fashion.

     In  describing  how  an  associatively  accessed, auditory engram-slice
would stimulate or elicit the "recall"  of  an  identical  or  most strongly
similar  engram-slice,  the  model  has  at the same time shown how incoming
auditory data from the outside world would be processed.   External auditory
data,  to  be  laid  down  in  permanent  memory, traverse the length of the
auditory memory channel  and  are  deposited  by  fixation  at  the freshest
extremity of  the channel.  External auditory data are "recognized" if their
passage   through   the   channel   stimulates   and    elicits   responding
trace-activation as described above.  However, such responding engram-slices
do not travel to the fixation-extremity and therefore can not be redeposited
there; they are already permanently fixated in their original locations.  It
is important to understand that only two sorts of auditory data traverse the
auditory  memory  channel  into  fixation  at its freshest extremity:  fresh
external   auditory   data   on   the   one   hand,   and   old,   internal,
associative-tag-activated data  on the  other hand.   In  other words, newly
fixated data or newly re-fixated data must come  from the  active sources of
external   perception   or   internal  associativity,  respectively.    (The
mind-model remembers both its  external and  its internal  experiences.)  At
least  two  rationales  militate  against  the  notion that memory-extremity
re-fixation  might  occur  of  response-slices  "flushed  out"  by  incoming
perception-data.   Firstly, there  is no need for redeposit of such old data
if they are  highly  similar  to  the  incoming  new  data.    Secondly, the
presently  incoming  perception-data  are  already flooding the transmission
lines, making it simultaneously impossible for  the products  of "recall" to
move about within the auditory memory channel.  Of course, and in accordance
with this whole scheme of recognition  during perception,  massive data-flow
can  be  occurring  orthogonally  out  of  and away from the auditory memory
channel  while  incoming  auditory  data  are  traversing  the  channel into
fixation at its creeping extremity.

     When fresh, incoming data are deposited by fixation at the extremity of
the auditory memory channel, they also by virtue of simultaneity  enter into
permanent association  with various other memory channels lying in parallel.
The associative tags which effect this permanent association and integration
of  fresh  data  are  the  same  tags  by  which  access to the data will be
maintained in the future.

     This fixation of association through simultaneity  happens not  just to
fresh data-slices, but also to old data-slices which have been sent coursing
down through  the  memory  channel  after  activation  of  their associative
onset-tags.    Thus  an  old  memory-slice  can  enter into new associations
whenever it is redeposited.  It should be becoming apparent now that thus an
enormous body  of belief  or knowledge could gradually accrete onto some key
reference-memories.    The  very   frequency  with   which  a   memory  were
associatively  reactivated,  and  therefore redeposited, could influence the
future likelihood of subsequent reactivation.  In this  model, a  memory can
grow stronger by increasing the number of the instances of its reactivation.

     The  foregoing  model  of  the  auditory  memory  channel has contained
certain crucial highlights which will come into play when the three channels
of visual memory, "abstract" memory, and auditory memory are combined in the
development of the linguistic control system for natural language.


     4.  THE MODEL OF THE VISUAL CHANNEL

The whole genesis of this model has hinged upon distinctions  in the duality
of pattern  and symbolic  code.  It is argued here that code, because of its
exact and  facile  manipulability,  is  essential  for "transabstractivity,"
while  less   exact  but  information-rich  pattern  is  essential  for  the
underlying knowledge that culminates  in  abstractions.    In  this article,
pattern is to be treated with respect to vision, because the sense of vision
is our greatest floodgate of perception.  However, vision is  not considered
essential  in  this  model  of artificial intelligence; the autobiography of
Helen Keller [1] has  strengthened the  author's conviction  in this regard.
Vision was  merely the  most challenging  and the most enabling sense-avenue
that the author could choose as  a way  of getting  general information into
the mind-model.   The  author begs  the reader  to judge the validity of the
linguistic model regardless of any invalidity  of  the  visual  model.   The
author was  able to  advance to the development of the linguistic model only
after incorporation of the visual model into the  sensorium/motorium grid as
an underlying  assumption.  Having used the visual model as a possibly false
lemma, the  author would  like to  see the  linguistic model  prove valid no
matter what happens to the visual model.

     Articles by  Kent [2]  enabled this author to develop the visual model.
The model as finally adopted resembles one which the author long resisted as
being  probably  too  naive,  until  Kent's exposition of feature-extraction
engendered  sufficient  sophistication  in  the  model  as  to  override the
author's feelings of uneasiness.

     As mentioned  above, the  visual memory  channel flows down through the
leftmost side of the  flat associative  grid.   Whereas the  auditory memory
channel is  visualized as  flat, the  visual memory channel is visualized as
more or less round,  because it  must contain  two-dimensional image-slices.
However, the  class of horizontal associative tags exiting the visual memory
channel can be regarded as flat.

     The visual channel starts as  a  quasi-retina  and  passes  through two
stages of  pre-processing before  entering the  memory grid as an associable
memory channel.

     The first stage of pre-processing is  one in  which this  design allows
some  time  for  self-organizing  of  the  visual channel, hypothetically to
correspond to extreme infancy in humans.  This model regards accurate visual
transmission as  too great  a burden to impose upon genetic design, which is
viewed here as part of a duality or  polarity with  design-by-learning.  The
self-organizing  stage  has  been  allowed  for  and  then  left  for  later
development.  Basically, the rationale is that the  order peculiar  to light
sources in the external world shall help to create an ordered arrangement of
visual transmission lines in the internal world.

     The second stage of pre-processing is that of  feature-extraction.  The
reference  by  Kent  can  be  consulted for details, but certain notions are
forthcoming here.  A  typical extraction  might be  that all  retinal points
forming lines  at a  certain slant  will converge into single memory fibers,
each of which will now unitarily represent a slanting line  of multitudinous
points.    The  analog  of  the  retinal  optic  nerve  is  modelled here as
consisting of about a million fibers.  On  the one  hand, feature-extraction
will permit  many retinal  fibers to  be represented by fewer memory fibers.
On the other hand, a single retinal fiber can serve as an  input to multiple
features that  are then  extracted as  multiple memory fibers.  So there are
tendencies both to reduce and to multiply the number of  visual transmission
lines.

     The  final  inputs  at  the  entrance  to the visual memory channel are
visual memory lines that  represent  features  rather  than  retinal points.
(These memory  lines are  called "fibers"  above, lest they be confused with
lines in retinal geometry.)  After more description of the visual channel, a
concept  of  "virtuality"  will  be  explained as the rationale for adopting
feature-extraction.

     The model of visual memory  is  quite  analogous  to  that  of auditory
memory.   Again, moments  of perception  are pulsed into slices down through
the memory channel.   Each slice  corresponds to  a feature-extracted visual
image.   The slice  cut through  the transmission lines has hypothetically a
node for each transmission line.  Each slice is genetically  provided with a
tabula-rasa  associative  tag  potentially  integrating  that slice with the
associative grid.  The oldest visual images are laid down near the entrance,
or top,  of the  visual memory  channel, and the newest images are gradually
being laid down at  the slowly  advancing, lower  extremity of  the channel.
The visual  memory channel,  in parallel with the other sensory (and perhaps
also motor) channels, is associatively connected to them at each  instant by
virtue of simultaneity in the fixation of horizontal associative tags.

     It  is  important  to  understand  the  concept  of "virtuality" as the
author's rationale  for  adopting  feature-extraction.    The  visual memory
channel has  two purposes:   that of a comparator of patterns, and that of a
memory.  As a memory, the  visual memory  channel must  associatively retain
images by  fixating and  tagging them.   As  a comparator, the visual memory
channel has the task of announcing to the system by associative tag whenever
it finds  in memory  a highly similar counterpart to either a fresh external
image or a reactivated internal image.  For purposes of comparison, it would
certainly not  work to  store raw  visual images without feature-extraction,
because the too detailed images would be too unwieldy to  be processed  by a
non-feature-extracting   comparator.       Feature-extraction   allows   the
comparator-mechanism  to  rise  a  workable  distance  above  point-by-point
comparison.  Meanwhile, some sharpness and resolution must probably be lost,
which loss must be compensated for, but how?  And why do  we not  notice the
loss  of  fine  detail?    The  answer  proposed here lies in the concept of
"virtuality," which is the self-illusion of consciousness.

     Throughout  this   paper,  the   author  insists   upon  observing  the
orthogonality of  data-flows in  the associative  grid.   Such observance is
mainly a help for  thinking, because  it would  not matter  if the assembled
physical  grid  were  subsequently  curved  or  convoluted.    However,  the
observance of orthogonality leads  to a  consideration of  dimensionality in
general.    For  instance,  processes  which  seem  to  meander in haphazard
directions can initially be difficult to explain.  It has helped this author
to  force  emerging  facets  of  "black-box" mechanisms into such orthogonal
structures as are described herein.   Once orthogonality  was compelled here
and  there  within  a  structure,  the whole dimensionality of the structure
often  became  apparent.    So  it  is  with  vision.    The  elucidation of
"virtuality' is now attempted with regard to "dimensionality."

     According to  this model,  when we  see an  image through  an eye, that
image  briefly  floods  our  whole  visual   memory  channel.     Any  given
feature-extraction  line,  that  happens  to  be activated throughout visual
memory as part of the image, is activating  all the  myriad historical nodes
that were ever fixated as "on" or "full" all up and down the used portion of
the transmission line.  Within each memory slice of all  our previous visual
experience,  the  perchance  reactivated  nodes  are  trying in summation to
combine forces and "vote" for the selection of their own visual image as the
most similar visual memory which ought to be the first, or even only, memory
slice to send out over its efferent associative  tag a  recall-signal.  That
recall-signal  is  the  announcement  that  the  comparator, i.e., the whole
memory channel, is proffering a comparison.   Note  that that  comparison is
never exact;  it is  just the  one which  has received the most "votes" from
nodes.  (Does a Rorschach blot operate by  exact comparison?)   If  we stare
long at the same image, many associations may come to mind, perhaps aided by
a  mechanism  of  neuron-fatigue   which  would   allow  initially  dominant
associations to yield to associations receiving fewer "votes."

     The main  idea here is that this model claims that we perceive visually
through a whole visual  memory/comparison channel  operating all  at once in
extreme  parallel-processing.    It  was  claimed  above,  in the section on
audition, that the informational content of a memory channel could  move not
only within  the channel,  but also orthogonally out of the channel.  It may
seem that that claim loads quite a burden onto the  associative tag, because
even the  largest, fullest memory slice is modelled as having only a unitary
associative tag.  However, the information  escapes its  own channel through
the  "virtuality"  of  passing  over  all  the  associative  tags of all the
counterpart slices with which it can successively be compared within its own
channel.    Think  of  a  visual  image being rapidly dismantled through its
elemental  components,  each  of  which  is  perceived  so  rapidly  through
in-channel comparison  that the conscious mind is fooled into believing that
it is perceiving the image all at once.  Viewed in this light, consciousness
itself becomes  a flickering  illusion in  which we  can never quite see the
gaping voids between the flickers.

     The  idea  of  dimensionality  leads  to  an  important,  further point
involving recursiveness  within the  mind-model.   A proposal is now made to
the effect that a valuable test  of a  perceptual model  lies in determining
whether  or  not  any  information  is  lost as the information is processed
through transformations and reorganizations.   For  instance, the  reader is
invited to  ask himself or herself (as the author does), whether this visual
model suffers the loss of informational content over  the orthogonal outputs
of its memory channel.  If informational content remains rather constant, no
matter how circuitously the information flows throughout the system, then it
may turn  out that the various aggregates of knowledge within the system are
defined recursively in terms of all the other aggregates of knowledge.  When
we discuss the area of "transabstractivity," it may look as though the total
system is capable of generating new knowledge from within itself.


          5.  THE GENESIS OF THE PSYCHOLINGUISTIC MODEL

Now that the underlying models of  the visual  and auditory  memory channels
have been  explained, the  rest of  the article is devoted to presenting the
author's admittedly speculative but  perhaps  thought-provoking  model  of a
linguistic   control   system   for   natural   language.      Whereas   the
sensorium/motorium grid has been presented here  without much  discussion of
its lackluster  genesis, the  author feels that explaining how he arrived at
his linguistic model will contribute to the reader's  critical understanding
of the  model and  also to  the reader's likelihood of seeing where mistakes
were made or of devising  a  much  better  model.    If  there  is  indeed a
contribution here, it is time to get help in assessing and developing it.

     The author developed the various models and perspectives by maintaining
and sporadically writing in a "theory journal" from 1972 to  1979.   In 1977
the fruitful  decision was  made to  pursue the  aforementioned dichotomy of
(relatively  loose)   pattern   and   (relatively   exact)   symbolic  code.
Immediately  a  torrent  of  speculation  resulted  concerning  the possible
interior organization of psychic structures for processing natural language.
The author toyed with elaborate structures for habituating phonemic strings,
and then foundered in an abortive attempt to devise a  "habituated mechanism
for grammar."   The  problem was that the perceptual system for patterns had
not been elaborated, and therefore there was no base of  experiential inputs
from which  to derive control lines for mechanisms generating sentences.  As
the eventually abortive grammar system became  quite complicated,  it really
became an empty shell as more and more of its control lines became dependent
upon the nonexistent perceptual system.   A particularly  vexing problem was
that of  how the  automaton would  keep its  languages apart if it knew more
than one natural language.  The impetus of  the duality-approach temporarily
ran out in October of 1977, and the project lay dormant for five months.

     The  appearance  of  Kent's  articles  in early 1978 led to the present
formulation of the visual channel.  By November  of 1978  the author  had at
his disposal  most of  the model  of the  sensorium/motorium as described in
this article.  He wanted to integrate the grammar work of the  previous year
with the new overview of the various memory channels.  The first step was to
halt at a new impasse involving verb-recall.  The author  would now  like to
describe  the  slow  removal  of  that  impasse  and  the  rapid development
thereafter of the present linguistic control  system.   That work culminated
in  May   of  1979  and  the  author  turned  to  his  present  attempts  at
communicating his results.

     In November of 1978 the first obvious problem  was to  see how percepts
in the visual memory channel would fetch words stored in the auditory memory
channel.  It seemed simple enough for a visual memory  image representing an
object to  activate a  horizontal recall-tag  for a noun naming that object.
After all, the model regarded both the visual memory slice  and the auditory
memory string  of the  word (through  its onset-tag) as unitary memory items
between which there could  exist  a  one-to-one  correspondence, established
through  simultaneity  in  the  fixation of the horizontal associative tags.
For a well-known, thoroughly habituated, concrete noun in  the vocabulary of
the  organism,  there  would  be  many  randomly placed crossovers of recall
between the visual and auditory memory channels.  When the  organism saw the
given object,  the percept  coursing down  through the visual memory channel
would stimulate or activate the many  various instances  where the  image of
the object  had been associated over to the noun.  If the signal in just one
of many recall-tags reached the auditory memory channel,  the noun  would be
remembered.

     The author  was initially stymied with the problem of how the plurality
of nouns would be conveyed from  the perceiving  visual channel  over to the
auditory memory  channel storing  the nouns.  He quickly chose instead to be
stymied with the similar but more  significant  problem  of  how  the visual
perception of actions would lead to the recall of verbs.  For both plurality
and verb-recall, it seemed that no  single visual  percept, with  its single
associative tag,  could be  sufficient to  recall either a noun-plural, with
its added inflectional information, or a verb, with  its complex information
involving subject  and action,  plus or  minus object.   For both phenomena,
plurality and verbs, it  seemed that  a bundle  of visual  percepts would be
necessary  just  to  accumulate  the raw information behind each phenomenon.
But even if such a bundle were activated  within the  visual memory channel,
the  horizontal  associative  plane  was  too rigidly simple to transmit the
extra information gathered in the bundle.  Although the roots  of a solution
appeared right  away in  November of 1978, the tentative solution itself was
slow in coming via journal entries in January and March of 1979.

     The rain root to the proposed solution was the idea in November of 1978
(in the  theory journal)  "that maybe  there should  be an additional memory
alongside the  others (sensory  and motor).  a memory  which held perceptual
content  but  not  sensory  content,  a memory which would handle conceptual
associations beyond the linear  scope of  the purely  sensory memories:   an
abstract memory."

     However, the  person authoring these ideas could not make direct use of
them, but instead agonized until January  of 1979  and then  moved backwards
from the  basic concept of a verb, through the notion of semantic categories
as inputs to the verb, and  finally back  again to  the idea  of an abstract
memory, as a place for the semantic categories to be represented.

     The plan for verb-recall, as developed in January and March of 1979, is
that an abstract memory channel shall  permit a  process of "intermediation"
between raw percepts and stored verbs.  It is therefore time to describe the
abstract memory channel.


          6.  THE MODEL OF THE ABSTRACT MEMORY

Within our above-described model  of  the  experiential,  left  half  of the
sensorium/motorium grid, the visual memory channel flows down along the left
side of the grid, and the auditory memory channel flows down along the right
side  of  the  (experiential)  grid.    The abstract memory channel is to be
visualized not as a flat channel, but as a three-dimensional channel flowing
down  through  the  experiential  grid,  in  between  vision on the left and
audition on the right.  The abstract channel also sideways extends above and
covers  the  auditory  memory  channel  so  as  to receive the perpendicular
outputs   of   the   aforementioned   "habituation-class"   of   associative
ultimate-tags rising  from the auditory memory channel.  The abstract memory
is a channel not attached to a sensor, and it  would therefore  remain empty
if it  did not receive its inputs from the various "concrete" channels, such
as vision  and audition.   The  abstract channel  lies in  parallel with the
concrete channels  and it  consists of myriad elongated "abstract" fibers or
lines  analogous  to  the  "transmission  lines"  of  the  concrete  sensory
channels.   However, the  abstract memory channel is seen as subdivided into
"cables" or groups of lines specially organized to suit various  purposes of
design.   The dimensionality  of the abstract channel is orthogonal, just as
the rest of this model is.

     One function  of  the  abstract  memory  channel  is  to  intercept the
afferent associative  tags coming  from the  visual memory, those tags which
might otherwise proceed directly to the auditory memory  and control engrams
there.   Those afferent associative tags from the visual memory channel flow
instead into the "logicoconceptual cable" (L-C cable) of the abstract memory
channel.   That portion of the abstract channel is called "logicoconceptual"
because each of its fibers shall gather up  interrelated inputs  from vision
which  in  summation  constitute  a  concept  or  an element of logic.  (The
logical elements, associated with conjunctions, prepositions, and plurality,
may actually  require further  processing.)  The abstract fiber of a concept
receives its inputs from vision and  then in  turn controls  the recall-tags
which become  the onset-tags of words stored in the auditory memory channel.
The logicoconceptual cable is visualized as being itself flat and lying flat
within the  plane of  the associative grid.  However, the L-C cable has flat
layers within it.  The  afferent  visual  associative  tags  empty  into the
basic,  top   layer  of   the  L-C   cable,  which   is  the  layer  of  the
logicoconceptual noun-lines.  In other words, the top layer of the L-C cable
has fibers  representing concepts and their nouns.  It has direct control of
the recall-lines over to the  stored  nouns.    A  visual  image  must first
activate  a  noun-line  in  the  L-C  cable  if  the  image  is  to activate
recall-lines for a stored noun.

     The descending layers in the  L-C  cable  represent  various  levels of
abstraction.   The only  secondary level  modelled so  far is for verbs, but
conjunctions, prepositions, and plurality may also have such layers.

     The  top  level  of  the  flat  L-C  cable,  that  of  logicoconceptual
noun-lines,  is  the  level  of  least abstraction.  Beneath that level, but
moved sideways to the  right  and  out  of  the  way,  is  a  flat  level of
verb-lines  controlling  recall-lines  heading  to  the right towards stored
verbs.  Associative input-lines, or "feelers," move left from the flat level
of  abstract  verb-lines  to  flow  directly underneath the logicoconceptual
noun-lines and receive inputs  from them.   The  logicoconceptual noun-lines
serve as  the above-mentioned  "semantic categories"  that are inputs in the
selection of a verb.  Verbs  are thus  at least  doubly abstracted  from raw
visual perception.  Each abstract verb-line sends its "feeler" out leftwards
underneath the flat, primary  level  of  the  abstract  noun-lines.   (Other
simple  percepts  may  be  mingled  in  with  the  noun-lines.)    Thus  the
input-feeler of each abstract  verb-line  has  access  to  all  the abstract
noun-lines, or  "semantic categories."   When  a verb  is to  be recalled to
describe a visual  percept,  an  extremely  parallel differentiation-process
operates over  all the  verb-input-feelers that  are sampling  the status of
each relevant semantic category expressed  among  the  noun-lines.    At the
junctures  of  the  feelers  and  the  abstract  noun-lines, there are nodes
fixated by learning.   A (complex)  verb which  has both  many nodes  on its
feeler and  most of them activated will win the recall-race over other verbs
with many feeler-nodes but few of them activated.   The  differentiation may
be enhanced  if available  but unactivated feeler-nodes serve to inhibit the
input-feeler to a verb-line.  A (simple) verb with few  feeler-nodes but all
of  them  activated  could  thus  win  out  over  a (complex) verb with many
activated and many unactivated feeler-nodes.  (The feeler  described here is
actually one  of very  many feelers for each verb-line.)  The sentient being
that is bringing verbs to mind may not realize that the whole  vocabulary of
verbs  is  competing.    The  differentiation-process  makes  possible  both
malapropisms and flights of fancy.

     The mentally proleptic reader may be realizing here  that such abstract
cables will  be used to control the parts of speech in syntax, but first the
function of the abstract logicoconceptual cable must be  examined in greater
detail.

     It  is  important  to  realize  that the abstract memory channel always
operates upon the basis of old, rather than fresh, memory data.  This notion
of reliance  upon past experience makes sense if the abstract memory channel
is a  habituated mechanism.   This  dependency upon  past experience becomes
clear if  we examine  how the  "abstract fibers"  operate.  For consistency,
only vertical fibers flowing  within  the  abstract  memory  channel  and in
parallel with all the other memory channels shall be called "abstract."  Any
horizontal or perpendicular associative lines or fibers making  contact with
an abstract  fiber shall  be called "concrete," to highlight the distinction
between the "abstract"  fibers  of  extended  experience  and  the momentary
associative fibers  that represent  a "concrete" influence upon, or function
of, the abstract fibers.

     When an  intelligent organism  sees the  bundle of  visual images which
constitute the raw information capable of ultimately fetching or recalling a
stored verb, the process of  abstraction  begins  immediately.    The images
within  the  bundle  are  deposited  in  their  proper order at the freshest
extremity of the visual channel, but these images at the  freshest extremity
of the  channel can  not send out signals sideways to activate the necessary
abstract fibers within the  logicoconceptual cable.   As  may be  clear from
previous  portions  of  this  article,  fresh  association can occur through
simultaneity,  but  instances  of  recall  must  filter  through associative
pathways laid  down in  the past.   Since  the fetching  of a stored verb is
recall rather than fresh  association, the  images in  the verb-input bundle
must all  do their  best to  activate highly similar counterpart-slices down
through the visual memory  channel.   These counterpart-slices,  and not the
fresh  images,  will  then  associatively  access  abstract  fibers  in  the
logicoconceptual cable.

     Notice that such spread-out routing of information as inputs for access
to   verbs   permits   the   process   of   verb-recall   to   be  a  rather
catch-as-catch-can,  labile,  loosely  configurational  process.   Remember,
practically all  verbs in  the active  vocabulary end up competing to be the
verb which is consciously recalled within the auditory memory channel.   The
process described  here is  not meant as faulty or hit-and-miss; rather, the
idea is to let such a  variety and  wealth of  inputs serve  in selection of
verbs   that   fine   differences   and   subtle   nuances   can   sway  the
selection-process into vectoring towards the most  apt and  descriptive verb
to name an action or a state of being.

     An  abstract  fiber,  or  cable  of  such  fibers, stands in remarkable
isolation.  The  abstract  fiber  flows  down  through  the  abstract memory
channel,  hypothetically  perhaps  for  a  distance  as  long as the channel
itself.  Although some provision is to be made for the control  of cables of
abstract   fibers   in   syntax,   the  individual  abstract  fiber  of  the
logicoconceptual  cable  unitarily  accepts  myriad  "concrete"  associative
inputs and governs myriad "concrete" associative outputs.

     The   abstract   fiber   accomplishes   its  feat  of  logicoconceptual
abstraction by allowing a great leeway in its inputs while  insisting upon a
rather restricted  target for its outputs:  a stored word or morpheme in the
auditory memory channel.  For  instance,  an  English-speaking  organism may
permit many  sorts of  images of a dog to access the stored word, "dog."  Of
course, a single image stored in  the visual  memory channel  can perhaps be
associated over  to several  or many abstract lines, so that varying degrees
of differentiation or specificity can occur.  Thus a speaker  might refer to
a "dog" or to a particular species of dog.

     The important  consideration here  is that these processes of selection
are forms  of  extremely  broad  and  parallel  competition.    Not  only do
practically all the verbs in an active vocabulary compete, but, from a given
abstract line that is going to  access  the  winning  verb,  there  are many
competing concrete  associative recall-lines going over as the onset-tags of
various historical instances of the recording of that verb here and there in
the auditory  memory channel.  It may be that only the more recent instances
are likely to win out, so that verbal memory can  gradually shift  or mature
over  time,  but  the  abstract  fiber  is not limited to a single, possibly
unreliable,  concrete  recall-line  as  an  output.    Such  availability of
multiple, but  essentially identical,  output-lines can perhaps be viewed as
an automatic error-reducing mechanism of redundancy.

     As was mentioned above, a  bundle  of  input-images  for verb-selection
will be  deposited at  the freshest  extremity of the visual memory channel,
after using the internal comparison-mechanism of  the visual  memory channel
to  gain  access,  via  horizontal  associative  lines, to relevant abstract
fibers in the logicoconceptual cable.    Only  after  such "supratraversial"
associative activation  of the  abstract fibers  can each fresh image-slice,
being newly deposited, enter into a  fresh associative  interconnection with
its relevant abstract fiber (or perhaps fibers) at the freshest extremity of
both the visual and abstract memory  channels.   Thus, although  language is
always used  as a  habit from  the past, there are mechanisms, all along the
advancing  front  between  experience  and  tabula  rasa,  for  the constant
reaffirmation, updating, and  perhaps gradual  change of  the way in which a
mind perceives and interprets its world.


           7.  THE BASIC MODEL OF SYNTAX

The foregoing sections of this paper  have modelled  systems of information-
flow in basic automatic routing where one flow of information did not govern
another.  A fresh, unitary flow  of  information  within  a  sensory channel
might  associatively  scatter  eddies  and  ripples  throughout  much of the
associative memory grid, but such propagation of signals can still be viewed
as the single dispersal of a single flow.  Syntax, however, involves letting
one  (habituated,  permanent)  flow  of  information  control  other (freely
associated, transitory) flows of information.

     The above  described layers of the abstract logicoconceptual cable will
serve to contain some of the  transitory flows  of information  which syntax
will control.   Whereas  the logicoconceptual cable lies in the plane of the
flat associative memory grid, we now posit an abstract "syntax  cable" lying
in a second tier of the abstract memory channel, above and covering both the
logicoconceptual cable and the  auditory memory  channel.   The syntax cable
will control  flows of information within the logicoconceptual cable and the
auditory memory channel.

     The auditory memory channel is unique  among memory  channels, because,
according to  this model,  it is where conscious verbal thought occurs.  The
auditory memory channel is  a  self-perceiving  channel.    Syntax generates
sentences by stringing together morphemes and words residing directly within
the auditory memory channel.

     The basic function of the syntax cable is  now described,  although the
development of  the syntax  cable is left for the section on the habituation
of grammar.

     A textbook by Liles  [3] provided  this author  with some  knowledge of
Chomskyan transformational  grammar.  This paper confines itself to a simple
example of syntax and does not treat the selection of  transformations.  The
following discussion of how syntax strings a sentence together is based upon
a notion of a very simple English sentence consisting of a noun  as subject,
a verb,  and another  noun as  direct object of the verb.  Almost everything
but word-order is disregarded  here.   The sentence  is probably  a slightly
ungrammatical one,  such as  a baby  might utter, e.g., "Daddy drink water,"
without the inflectional ending on the verb.   Inflection is  treated in the
next section of this paper.

     Although transformational  grammar expresses  such a simple sentence as
the tripartite one above as a tree with nodes, this  section of  the article
is  concerned  with  syntactic  nodes  only  in  a  row representing surface
structure.  In other words, somehow (explained further on in this paper) the
syntax cable  will have  assembled a sequential series of nodes representing
and controlling sentence-elements that are to be activated in the same order
as that of the nodes, so as to string together a sentence.

     The volitional  aspect of why the sentence is thought or uttered is not
treated deeply here.    As  for  the  mere  thinking  or  generation  of the
sentence, without regard to motor utterance, it does not matter much whether
we imagine that syntactic structures are always waiting  for the opportunity
automatically  to  assert  themselves,  or whether an activated mechanism of
attention initiates the operation of a syntactic structure (i.e., one of the
possible  transformations,   including  what   could  be   regarded  as  the
untransformed, original one).  Here we are merely examining how  a syntactic
structure,  once  activated,  would  generate  a rather concrete sentence to
describe what is being observed through the visual memory channel.

     Further to eliminate volition from our discussion, let the following be
considered.   Once the  components of a sentence have been assembled, or, to
express the  idea more  carefully, are  being assembled  within the auditory
memory  channel,  that  sentence  is  free  to move down within the auditory
memory channel and to deposit itself  in  a  quasi-capsule  at  the freshest
extremity of  the channel.   As  was stated  in the  section on the auditory
memory  channel,   "old,  internal,   associative-tag-activated  data"  will
"traverse  the  auditory  memory  channel  into  fixation  at  its  freshest
extremity."  Thus the elements of the sentence may be gathered from randomly
and widely  varying locations  up and  down the auditory memory channel, but
the  sentence  coalesces  into  an  integral  whole  when  it  is  deposited
automatically at the freshest extremity of the channel.  Now, motor volition
is modelled here as  a sort  of shifting  valence topography  of integrating
associative  potentials  dynamically  moving,  at  the elongated associative
interface between experiential and motor memory, either towards or away from
threshold firing-levels at which positive volitional motor activation occurs
automatically, but premeditatively.  Physical, motor utterance of a sentence
can occur either as the sentence is being generated up and down the auditory
memory channel, or after  the  sentence-capsule  has  been  gathered  at the
freshest  extremity  serving  as  a  kind  of "pre-elocution register."  The
reader will  please  forgive  the  author  for  brushing  aside  the perhaps
tantalizing  topic  of  motor  volition,  but  it  is  not  essential to the
linguistic model which the author is eager to communicate first.

     We now examine the generation of an English sentence under  the control
of three  serial syntactic  nodes in the syntax cable of the abstract memory
channel.  Let the sentence be  the slightly  incorrect one  mentioned above,
"Daddy drink water."  Therefore the three syntactic nodes are one for a noun
as subject, one for a verb, and one for a noun as direct object of the verb.

     Each syntactic node within  the syntax  cable is  actually an elongated
abstract memory  fiber.  Viewed in a cross-slice, it has the unitary quality
of being a node, but its elongation through the channel allows it to operate
in myriad instances extended over time.  The syntactic node-fiber enjoys the
isolation  attributed  to  abstract  fibers  above,  in  that  it  is itself
unitarily abstract,  while access  to and  from it must occur via the myriad
concrete associative lines.

     Each of the three syntactic node-fibers is associated with the  part of
speech which  it will activate within the auditory memory channel during the
generation of a sentence.  Although  the syntactic  node is  associated with
all members  of the class of its part of speech, it must select and activate
within the auditory memory channel only one member of the  class.   In other
words,  each  syntactic  node  must  propose  only  one candidate-word for a
particular spot in  the  sequence  of  a  sentence  being  generated.   Each
syntactic  node  has  concrete  associative  lines  descending to access and
govern a whole part-of-speech layer within the logicoconceptual cable.

     Thus the first syntactic node for our sentence, representing a  noun as
subject  of  the  verb,  accesses  and  controls the whole noun-layer of the
logicoconceptual cable.   At  the transitory  moment of  its activation, the
syntactic node-fiber  "flushes out" from the whole noun-layer whichever noun
happens to be "voted for" most strongly on the basis of present  inputs from
visual perception.   The  noun-layer, while awaiting a flush-signal from the
syntactic node, is to be viewed as being normally inhibited  from activating
its  myriad  recall-lines  that  are  the  onset-tags of morphemes and words
stored in the auditory memory channel.   A  flush-signal from  the syntactic
node for  subject nouns  briefly disinhibits the whole noun-layer to let the
perceptually dominant noun  escape  over  to  the  auditory  memory channel.
Activation of  the onset-tag  of that  noun causes its phonemic string to be
consciously "heard" or "perceived" within the auditory memory  channel.  The
phonemic string  travels to  the freshest extremity, where it is redeposited
by fixation.

     The activation of the phonemic string is terminated by the ultimate-tag
which rises  perpendicularly from  the string up into the syntax cable.  The
ultimate-tag from the stored word  signals  to  the  syntax  cable  that the
firing of the first syntactic node has produced its results.  Control is now
passed sequentially to the  next syntactic  node, that  of the  verb for our
sentence.

     The syntactic  node-fiber for verbs flushes out a perceptually dominant
verb from the verb-layer  of the  logicoconceptual cable.   At  the speed of
thought, the  verb is  now activated,  is now "thought," within the auditory
memory channel.  A signal rises along the ultimate-tag of  the verb  up into
the syntax  cable to pass control from the verb-node on to the noun-node for
the direct object of the verb.

     Neuron-fatigue, plus a slight shift or  alteration in  perception, will
prevent  the  originally  dominant  subject-noun ("Daddy") from firing again
when the syntactic node-fiber for a noun as direct object is trying to flush
out a  noun as  the final  word in  the sentence.   The  noun which actually
represents the direct object should be  perceptually dominant  and should be
flushed out  for recall-activation in the auditory memory channel.  When the
ultimate-tag of this final word fires,  control can  be passed  perhaps to a
new sentence or to a new focus of attention.

     The syntactic  control-structure causes  control to loop quickly in and
out of the self-perceiving auditory memory channel at the speed  of thought.
Although the  process is  happening over a much wider area than the auditory
memory channel alone, the thinking organism is conscious only of the results
surfacing  within  the  auditory  memory  channel.    The words seem to flow
together  naturally  and  without  pause,  because   the  syntactic  control
structure  operates   so  quickly  and  automatically.    Thoughts  are  not
especially "willed," they just occur to a person's  mind and  help formulate
the person's will.

     So-called "Freudian  slips" of  the tongue  may occur when a residually
dominant concept in the logicoconceptual cable is originally present because
of one instance of perhaps strong mentation, and then the concept interferes
in a new instance of perhaps weak  mentation.    If  a  silent  but strongly
dominant concept  within the  logicoconceptual cable  does not  have time to
subside before an unrelated sentence is spoken, a "Freudian slip"  can occur
involuntarily.    In  this  vein,  the  author  has  not  yet  analyzed  how
spoonerisms might occur.

     The problem of how to keep  multiple natural  languages apart  within a
fluently multilingual  mind may be solved by certain features of this model.
Each syntactic  structure is  probably learned  as pertaining  to a specific
language.   Then the  syntactic nodes  of the  structure can probably access
only classes of parts of speech containing words that have  customarily been
included within  the vocabulary of the specific language.  There probably do
not have to be multiple  noun-layers  and  multiple  verb-layers  within the
logicoconceptual cable.  Instead, the descending concrete associative fibers
from syntactic nodes of  a particular  language probably  access (in blanket
fashion) only  words of  that particular  language.  Thus a person fluent in
several  languages  can  speak   continuously   in   one   language  without
interference from  the grammar  or vocabulary  of any  other language.  Note
that persons who are not truly fluent will still have problems, because they
are frequently  stopping and thinking, that is, associating every which way,
even into divergent languages.


          8.  AN EXAMPLE OF THE MODEL OF INFLECTION

Although English was kept in mind during the development of  the basic model
of syntax,  the model of inflection was developed with a view to such highly
inflected languages as Latin and Russian.  This paper now  presents a rather
circumscribed example  of how  the linguistic  control system for inflection
would operate in just one instance.  The author developed this model  in May
of 1979,  and is  eager to  communicate it  before entering a more leisurely
phase of examining many specific or universal aspects of inflection.

     Suppose that a cybernetic organism is operating in the Russian language
and must  discriminate between  the nominative  and accusative  forms in the
singular number  of the  Russian noun  for "Moscow,"  which we transliterate
here as  "Moskva."  The nominative form will accordingly be "Moskva" and the
accusative form "Moskvu."  In the  sentence being  generated, the  mind must
use "Moskv-"  as a  stem and  automatically decide whether to add the ending
"a" or "u."

     Within the model of syntax in  the foregoing  section of  this paper, a
noun such  as "Moskva"  is controlled by a syntactic node-fiber which guides
the process of recall resulting in the activation of the  phonemic string of
the noun  down through  the auditory  memory channel.   Two  aspects of that
model must now be changed.

     Firstly, the syntactic node-fiber will not access the complete phonemic
string, "Moskva."   Instead,  only the  stem "Moskv-"  will be accessed.  Of
course, such a stem has an  ultimate-tag rising  perpendicularly out  of the
auditory  memory  channel  and  into  the  syntax  cable.    We will use the
ultimate-tag as one of two converging inputs necessary for the  selection of
the proper  case-ending within the paradigmatic engrams of the declension of
Russian nouns like "Moskva" in the  singular number.   (English-speakers are
familiar  with  many  such  Russian  nouns:    "vodka,"  "beluga," "tundra,"
"Pravda" - to name a few.)

     Secondly, the syntactic node-fiber governing the noun will  now have an
additional output  beyond that  which merely  flushes the recall-line of the
noun-stem out of  the  logicoconceptual  cable.    This  new  output  of the
syntactic node-fiber  is to  be called  a "function-vector"  and it is based
upon the grammatical position and function of the noun within  the syntactic
string of  the sentence.   Although we speak of a single function-vector, we
really  mean  numerous  concrete  associative  lines  exiting  the  unitary,
abstract syntactic node-fiber.  Since these associative vector-lines are all
going to the same destination over the short run, we  can speak  as if there
were a  single function-vector.   Of course, over the long run, a neophyte's
use of inflectional endings might change by improving or maturing.

     In our Russian example, one syntactic  node-fiber is  perhaps trying to
flush out  a noun  that will  be in  the nominative case as the subject of a
verb.  Another syntactic node-fiber is trying to flush out a noun  that will
be in  the accusative  case as  the direct object of a verb.  Let us examine
how the system generates the  form  of  the  noun  in  the  accusative case-
"Moskvu" instead of "Moskva."

     The flush-line, or "recall-vector," of the syntactic node-fiber flushes
out the recall-line for the stem  "Moskv-" from  the logicoconceptual cable.
Meanwhile,  the  function-vector  of the syntactic node-fiber simultaneously
sends a signal into a portion  of  the  syntax  cable  called  the "function
cable."    This  function  cable  is  a  group of abstract control-fibers or
"bars."   Each specific  control-bar in  the function  cable receives inputs
only from  syntactic node-fibers  representing certain grammatical functions
that require always the same grammatical case.  For instance, in our Russian
example, possibly  several syntactic node-fibers representing direct objects
of verbs  and accusative  objects of  prepositions can  send their function-
vectors to  the same  accusative-case control-bar within the function cable.
Thus the function cable  serves as  a collecting-point  for function-vectors
and,  as  we  shall  see,  as a distribution-point for case-activation lines
leading to all declensions.

     The  signal  from  our  Russian  example  of  a   syntactic  node-fiber
generating an  accusative direct-object  form comes  into the function cable
and activates a control-bar representing the search for an accusative ending
to  the  noun-stem  "Moskv-"  presently  being activated within the auditory
memory channel.   The  accusative-case control-bar  in turn  sends out semi-
activating signals to all accusative-case endings in all Russian declensions
available within the system.

     The syntactic node-fiber does  not  "know"  in  advance  the particular
declension  (among  several  Russian  declensions)  to which the flushed-out
noun-stem will chance to belong.  Only the noun-stem itself is associatively
privy to that information.  Therefore the function cable must blanket-access
and pre-poise  all appropriate  case-endings in  all declensions.   We shall
name  as  the  "inflection  cable"  that  portion  of the syntax cable which
contains  the  various  Russian   declensions   as   clusters   of  abstract
inflectional  lines   controlling  concrete   associative  recall-lines  for
specific case-endings stored as morphemes in the auditory memory channel.

     Now the two separate inputs must  converge within  the inflection cable
to  select  the  accusative  singular  ending  "u"  for  the activated stem
"Moskv-."    The  ending  "u"  is  pre-poised  or  semi-activated  when  the
accusative-case  control-bar  blanket-accesses  all  declensions.  Final and
full activation of the ending "u" occurs when the ultimate-tag  from the end
of  the  fetched  stem  "Moskv-"  carries  a signal from the auditory memory
channel into the proper declension-cluster within the inflection cable.

     Thus an intersection of two  fan-outs  occurs.    Syntax distributively
accesses  all  accusative  endings,  and  the fetched noun-stem collectively
accesses all the possible endings peculiar to its own declension.   Thus the
stem  "Moskv-"  and  the  ending  "u"  are  combined to yield the accusative
singular form "Moskvu" of  the Russian  noun for  "Moscow."   Of course, the
Russians, a  nation of poets, sometimes have their choice between equivalent
forms of  an inflectional  ending, but  there are  additional factors easily
governing the  exercise of  that choice.   Typically, a set, habituated form
will come  to mind,  and then  conscious purposes  of poesy  or archaism can
cause the  substitution of  an equivalent  form for the originally occurring
form.


          9.  THE MODEL OF LINGUISTIC HABITUATION

Now that examples of  basic  syntax  and  rudimentary  inflection  have been
presented, the author offers a general insight into how syntactic structures
might come to be habituated in a mind.  These thoughts on the habituation or
"learning" of  language cover  how a mind shall learn to generate sentences.
The comprehension of sentences, although  expected  to  be  more  or  less a
reversal of  these processes  of generation,  is an  area of further enquiry
from  which  the  author  has  temporarily  withdrawn  in  his  eagerness to
communicate  what  tentative  results  are  already at hand.  Of course, the
author realizes that comprehension  must  both  precede  and  accompany that
learning by  which either a human infant or a newly assembled automaton will
habituate syntactic mechanisms for the generation of sentences.

     The whole abstract memory channel, proposed  here as  the essential and
enabling  medium   for  the  linguistic  control  of  data-flows,  has  been
hypothetically divided into several cables in  this  paper.    First  it was
divided  into   a  "syntax   cable,"  which   operated  upon  an  underlying
"logicoconceptual cable."  Then the syntax cable was described as containing
the special  mechanisms of a "function cable" and an "inflection cable."  In
general, the author has found that  it  is  easier  and  more  productive to
imagine a  superfluous cable  that can  be eliminated  if found unnecessary,
than to suffer the frustration of overcaution when a lacking  cable would be
really  valuable  if  dreamed  up  for  incorporation  into  the design.  An
abstract cable can be valuable even if it serves temporarily  only to buffer
and isolate  two mechanisms  in the  plan of the designer.  Accordingly, the
author freely designed various cables in  May of  1979 in  such numbers that
the presently described possibility for habituation suddenly appeared.

     The genesis  of this  model of habituation followed the idea that first
nouns would be learned,  and  then  verbs.    First  a  syntactic node-fiber
controlling nouns would be used and practiced so repeatedly as to constitute
habituation.   Then  the  organism  would  become  aware  that  it  was also
encountering verbs in its environment of language-speaking entities.  In our
model here, which is now English, verbs are to be associated with  the nouns
which  are  their  subjects.   However, the verb-controlling syntactic node-
fiber becomes  habitually operative  only after  the operation  of the noun-
controlling  syntactic   node-fiber  has  been  thoroughly  and  permanently
habituated.  The key element in the sequenced  habituation of  the syntactic
nodes is  the aforementioned  "habituation-class" of  the ultimate-tags from
the auditory memory channel.

     An ultimate-tag serves as a signal to return  control of  the sentence-
generating process  to the  syntax cable.  Once the control of English nouns
is habituated, it is  easy for  the signals  in ultimate-tags  from nouns to
serve  as  stimuli  to  trigger  the activation of an erstwhile unhabituated
syntactic node-fiber controlling verbs.  Over time, any  naturally occurring
syntactic  node-fiber  can  be  added  to the syntactic string, because each
syntactic node-fiber becomes habituationally  active when  the morphological
phenomenon behind  it is  being dealt with by the young or curious mind, and
because the same syntactic  node-fiber  becomes  habituationally  dormant or
"transparent" when  it has  been "embedded"  in the syntactic string and the
mind is dealing with a  new,  typically  more  subtle  phenomenon.    Thus a
syntactic node-fiber  for nouns as direct objects can be added to the string
after the habituation of verbs.

     The habituation process forms a gradual spiral over lengthy time in the
following  manner.    Suppose  that  we  start with a positionally superior,
syntactic node-fiber for nouns, controlling the  positionally inferior noun-
layer  in  the  logicoconceptual  cable.    Thus  we have the first downward
movement in  our spiral,  which, by  the way,  will actually  be composed of
square-like coils strung together.

     The logicoconceptual  cable sends signals out horizontally to the right
to access nouns stored in the auditory memory  channel.   Then ultimate-tags
from accessed  nouns rise  perpendicularly from  the auditory memory channel
into the upper tier of the syntax cable.  There  the ultimate-tags  meet and
connect with  horizontal concrete  associative lines  that flow leftwards to
begin activation, not of the already  habituated, dormant  noun-node, but of
the  newly  active  verb-node,  or  syntactic  node-fiber for the control of
verbs.  Thus the  loop of  habituation has  gone full  circle.   The process
slowly, over  many months  in the  case of humans, continues to loop around,
adding various syntactic nodes in a spiral of habituation.

     The model of linguistic habituation presented here is new to the author
and  rather  simple,  but  it  may  serve  as  a  basis  for  quite  complex
elaborations.  The author advises that  orthogonality should  be observed in
any elaborations, even those of syntactic "trees."  In other words, branches
drawn in syntactic trees should contain  no angles  other than ninety-degree
right angles,  so that  each tree  will mesh and conform with the structures
around it in the orthogonal model.


          10.  THE NOTION OF "TRANSABSTRACTIVITY"

Within this general model, since the aggregates of conceptual knowledge join
pyramidally  to  culminate  punctiformly  in  highly  manipulable  words  of
symbolic code, the conscious,  thinking mind  can almost  endlessly generate
thought-sentences by skimming through the uppermost levels of conceptuality,
constantly descending only to that  depth  automatically  determined  by the
free interassociativity  of the  interacting conceptual aggregates.  What we
call an "abstraction" or "abstract concept" is modelled here as  a pyramidal
aggregate of conceptual knowledge, which becomes abstract through the highly
manipulable "punctiformity" of its  associable tip:   the  word.   Since the
mind is  modelled here  as skimming  rapidly across  countless such tips, it
seems apt  and useful  to call  this uppermost  abstract level  the level of
"transabstractivity."


          REFERENCES

1.   Keller, Helen,  "The Story  of My  Life" (Grosset and Dunlap, New York,
1903).

2.  Kent, Ernest W., The brains of men and machines, "BYTE", Vol. 3, Nos. 1-
3 (January, February, March, 1978).

3.   Liles, Bruce  L., "An Introductory Transformational Grammar" (Prentice-
Hall, Englewood Cliffs, 1971).


SourceForge Logo /projects/mind
Last updated: 27 July 2003
Return to top; or to the
weblog.html Weblog-Sitemap.