The Real ABCs of Language

By Chaumont Devin
Ka'u, Hawaii, May 21, 2006.
Last updated December 26, 2009.

Your feedback is valuable to me.  Please send comments to
joedevin@witchit.com.

Section 0.  What this paper can do for you.

The modern study of language and intelligence covers so many scientific
fields that it would probably be impossible for any one person to master
them all in a lifetime, and by then they would all have changed beyond
recognition.  This paper provides an alternative.  It is a shortcut that
shows how the internal workings of language can be determined without
knowing anything about brain structure, FMRI, neurotransmitters, or
dissecting a single human brain.

How is this possible?  By a rigorous analysis of the LOGICAL rather than
the PHYSICAL workings of language.  In this paper I will assume that
language is a logical rather than a physical phenomenon, so that the
details of the physical host supporting the logical structures of language
are irrelevant whenever the requisite hardware is present.  I will also go
ahead and assume that the modern digital computer is capable of supporting
language--in other words no quantum or other futuristic kind of computer
is required, the problem being one of software design rather than any
hardware limitation.

And how dare I say this?  In this paper I will show how ALL of the
internal
data structures of language can be readily modeled using nothing but
directed links and nodes that work on our simplest computers.
 Furthermore,
modern brain science agrees that the human linguistic apparatus is a
logical rather than physical entity because it cannot be localized to any
particular area of the
human brain.

And why are these things important to you?  If you have read this far, you
are obviously interested in language.  If you are interested as a matter
of
intellectual curiosity, then this paper is important to you because it
will
enable you to see how language works without a great deal of investment in
effort or time; but if you are a researcher, then it will be important as
a
guide to give you a clearer idea of what to look for in the physical brain
and why.  The dissection and analysis of brain structure and brain
chemistry are an essential part of brain science, but so far they have
failed to reveal any clear logical pattern, while the study of language
provides a part of this logical pattern without explaining the underlying
brain structure.  So the information in this paper should be seen as
complimentary to any study of brain biology--at least in those areas of
the
brain found to be responding to language and thought.

Section 1.  How Children Learn To Speak.

In the following paragraphs I will attempt to tell the story of infant
language acquisition in the first person.  This is not the only way people
can learn language, but I believe it is the best.

After I have learned to bawl very well, and proven my ability to do so at
all hours of day and night, I make a discovery.  I learn that besides just
bawling, I can also gurgle and coo.  So I lie on my back gurgling and
cooing for hours at a time.  And I discover that I can get better and
better control of these gurglings and cooings, and even make them sound a
little bit like the sounds my mother makes when she holds me in her arms.

And besides just holding me in her arms, little by little my mother begins
to carry me outside and show me the various things that are in the world.
She shows me pretty things, like flowers, and slowly and carefully repeats
some certain sound that I have come to associate with flowers.  She also
does this for feathered and furry things, and a host of other things.

And after these excursions, she lays me down again and I gurgle and coo,
but the things she has shown me keep running through my mind, and with
them the sounds she made when she showed them to me.  I see them again and
again in my imagination and keep hearing the sounds, and start trying to
make the same sounds, but I can't.

Yet slowly, ever so slowly, my body becomes stronger, and I get so I can
roll over on my stomach and creep around, and she has to put me in a
playpen to keep me from tumbling off onto the floor.  I learn to clutch
the bars in my hands, and finally to pull my whole body upright and stand
while holding onto them.

My ability to make different sounds also keeps getting better and better,
and slowly, ever so slowly, I begin to be able to utter sounds that
approximate the sounds my mother makes when she shows me various things.
And now I am learning the names of many other things--things that she
calls "toys."

She still picks me up and carries me outside, but now instead of making
just one sound, she often makes two.  I now know that these sounds have
special meanings, so I will call them "words."  She will say the word for
something I already know, for example, "bird," but along with the word I
already know, she will say another word that I don't know, and when she
does so, she will raise the pitch of her voice, evidently for emphasis.
And by some strange means I will know that this other, new word, is
telling me the way the thing I know about IS.  For example, she will say
things like, "BIG bird," "LITTLE bird," "PRETTY bird," and suchlike on and
on about birds.  I may not know exactly what these new words mean right
away, but I know they are telling me things about the way birds ARE.

Another thing she does is something like this, but a little different.
She will say a word I know, for example, "bird," and then, without any
change in the pitch of her voice, she will immediately say another word,
like, "fly."  And somehow, I don't know how, I know that this is the word
for something the bird is doing.  And sure enough, when I look, I see the
bird flapping its wings and flying through the air.

And as I keep getting bigger and bigger and stronger and stronger, I start
crawling around the house, and one day she brings me this thing called a
"kitty."  It looks really cute, and all I want to do is to touch it and
pet it.  "NICE kitty," she says.  Then she pours out some milk in a dish
and sets it on the kitchen floor, and I watch as the kitty drinks.  It
doesn't drink like I do, but I know it is drinking because I see it
lapping the milk up with its tongue, and pretty soon the milk is gone.
"Kitty drinks MILK," she says, and I don't know why, but somehow I know
that the kitty is doing something, and that something is to drink, and the
thing the kitty is drinking is milk.

And this is essentially how I learned language.

Section 2.  What Is A Pattern?

Since earliest times, The most perplexing of all mysteries to confront the
human mind has surely been the problem of the pattern.  In fact the ways
in which people have dealt with the fundamental idea of the pattern covers
a broad spectrum from the perspective of the primitive tribesman who sees
invisible "spiritual" patterns in just about everything around him  to
that of the modern conspiracy theorist who sees the darkest conspiracies
behind every action of those higher in the social hierarchyÀthan himself
to that of the Christian who believes in the survival of his own soul
beyond the grave and in the existence of an all-powerful, invisible God to
that of the materialist marxist who would do away with everything
"supernatural" altogether by disbelieving anything that cannot be detected
through empirical means.  And disagreements regarding the ways in which
people perceive and adhere to patterns have sparked the bloodiest wars and
the greatest of all inhumanities ever perpetrated by man against man.

In recent times people have come to understand the nature of patterns much
better by means of the computer.  They go to the nearest computer outlet,
buy a CD with the pattern of a program they want already recorded on it,
take it home, copy it to their home computer, and thence to their hard
drive.  And during all of this copying and installation, were they to keep
weighing their computer and their hard drive, they would find that this
piece of software, which they may have purchased for hundreds of dollars,
adds nothing whatsoever to the mass of either.  Why?  Because patterns are
just as real as anything else, but quite massless.

So on the one hand we have matter, while on the other hand we have the
patterns formed in matter, the "spooky" part of it being that patterns
have no mass whatsoever--not even a single atom or molecule--and yet they
are just as real as anything else in existence.

it is surprising that no real "pattern science" has ever emerged, since
the universe itself and all matter would immediately fall apart without
patterns.  Yes, there are branches of learning that may call themselves
"pattern science," but they deal with pattern recognition and pattern
comparison rather than with the nature and properties of patterns
themselves, the difference being pretty much the same as the difference
between being able to compare girls on a scale of from one to ten and
being able to explain what girls are made of and how.   Because of this
apparent scientific oversight and deficiency, I will attempt to nail down
some basic facts about patterns as follows:

1. Patterns are perfectly massless.  No smallest part of any pattern is
ever an atom or molecule or a subatomic particle.

2. Patterns are perfectly real, and just as real as anything else in the
universe.

3. Patterns are completely separate from all of the laws of the
three-dimensional universe, but are not free from the laws of time.  They
can never be said to be here or there or in motion, but they can be said
to exist or not exist at some time and may change over  time.

4. Patterns are perfectly scalable.  A pattern can be just as large or as
small as deemed practical.

5. All material things including the molecules of gases are the outward
manifestations of patterns.   Why?  For one thing, because electrons form
specific "cloud patterns" about their nuclei without which the kind of
matter we know would be quite impossible.

6. No pattern can exist without at least one physical host.

7. A pattern can be hosted in more than one physical host at
the same time.

8. Patterns come in two flavors: static and dynamic.  A static pattern
never changes, like the pattern of print on a typed page.  A dynamic
pattern, on the other hand, is self modifying and always changing.  The
human mind is precisely such an entity.  It is no more a human brain than
Microsoft Word is a computer, and yet it is just as real as Microsoft
Word.

All words are nought but massless patterns.  "But surely the ink on a page
has mass," you may argue, Yet the ink on a page is not really words.  It
is only matter that has been configured to encode a pattern of words.  The
real pattern, that thing which can be copyrighted in Washington, D.C., is
not paper and ink but the actual pattern that paper and ink can be made to
represent.  The paper and ink are visible and material, but the
copyrighted pattern itself is invisible and massless, yet no less real.

Section 3.  What Is A Linguistic Link?

Before we continue, let us take a moment to consider the basic building
blocks of language.  What if we could find some basic building block from
which we could quickly build sidewalks, houses, skyscrapers, bridges, and
entire cities?  No longer would it be necessary to erect great steel
frames and scaffolds.  No longer would it be necessary to bring in
truckloads of concrete.  No longer would it be necessary to order timber,
roofing materials, and other supplies.  Everything would just snap
together effortlessly and seamlessly, and nothing else but this universal
building block would ever be needed again.  Needless to say, such a
building block would revolutionalize the construction industry and turn
out to be one of the most valuable discoveries ever made.

Now as it happens, there exists just such a universal building block for
language, from which all of the internal data structures of the human
linguistic apparatus can be quickly assembled.  It is called the
"linguistic link," and it consists of nothing but a source, a destination,
and a type.  Such links are directional, and their sources and
destinations
are called "nodes."

Such linguistic links are at once the smallest common denominator of all
linguistic structures and the largest component that all linguistic
structures share in common.  And such links are nonmaterial patterns.  We
can represent them in all kinds of ways, but we can never really see them.
 Nevertheless they are quite real, as we shall soon see.

In the system of classification and terminology used in this paper, we
define four generic kinds of linguistic link:

1. SYNLINK, or SYNTACTIC LINK.

2. SEMLINK, or SEMANTIC LINK.

3. RADLINK, or RADICAL LINK.

4. LEXLINK, or LEXICAL LINK.

We also define three kind of node:

1. WORD, or SYNTACTIC NODE.  Any word appearing in coherent text.

2. SEMNOD, or SEMANTIC NODE.  A node in the ontology.  Each such node is
associated with just one and only one meaning, and can be thought of as
equivelant to a word-sense definition in a dictionary.

3. LEXNOD, or LEXICAL NODE.  A symbol, often called a "word" appearing as
keyword in a dictionary or lexicon or expressed in a question like, "What
does 'FISH' mean?" where "fish" is not a part of the syntactical structure
of the phrase.

In order to simplify the text, we will employ the short forms of these
terms from now on.

Link Direction.

Since every linguistic link has a source and a destination, it is clear
that all linguistic links must have DIRECTION, as follows:

1. SYNLINKS generally link one word (a dependent) to another word (a
regent).

2. SEMLINKS link words to their meanings.

3. RADLINKS link meanings (semnods) to one another within the ontology.

4. LEXLINKS link meanings (semnods) to external symbols.  These may be the
sounds associated with words, written "words," etc.

Recall that all of the internal structures of language are patterns
consisting of invisible links and nodes, so that the only visible parts of
this system are the external symbols associated with lexnods, for example
English "words."

Sometimes it is convenient to visualize all of the internal structures of
language as a single structure consisting of three planes occupying
positions one above the other in space.  In this model, the top plane is
called the PHONOLOGICAL PLANE, the second the SYNTACTIC PLANE, and the
third the SEMANTIC PLANE.  The phonological plane is populated by nothing
but lexnods, the syntactic plane is populated by words and synlinks, and
the semantic plane is populated by semnods and radlinks.  The phonological
plane corresponds to a vocabulary list such as a dictionary or lexicon in
which each entry (lexnod) appears in isolation from all the other entries
(lexnods) in the list (the whole plane).  Thus there exist no links of any
kind lying completely inside the phonological plane.  The syntactic plane,
on the other hand, is made up of synlinks linking one word to another,
each word being a node within the plane.  Thus it can be seen that ALL
synlinks lie within the syntactic plane, and no synlink ever strays
outside this plane.  The semantic plane is similar in that it consists of
semnods linked one to another by radlinks with no radlink ever straying
from within the semantic plane.  The gap between the syntactic and
semantic planes is bridged by semlinks, which link words downwards to
their meanings; whereas lexlinks go upwards from semnods in the semantic
plane all the way to terminate at lexnods in the phonological plane.

Section 4.  What Is An Ontology?

Once the linguistic link has been thus carefully defined, it immediately
becomes possible to build something called an ONTOLOGY.

An ontology is a collection of nodes each of which is linked to a meaning
roughly corresponding to what is called a "word-sense definition" in
dictionary parlance or a "concept" among linguists.  These nodes, called
semnods, are also linked in various ways to one
another, and to the nodes of a lexicon (lexnods).

The most important types of links between semnods are links to hypernyms
and links to holonyms.  Hypernyms and holonyms are simply other semnods
within the same ontology.

A hypernym is roughly "something that something else IS."  For example, an
automobile is a conveyance, thus "conveyance" is a hypernym of
"automobile."  A holonym is roughly "something that something else is a
part of."  For example, a rubber tire is part of a wheel, thus "wheel" is
a holonym of "tire."

Besides hypernymy and holonymy, a good ontology should also contain
synonymy and antonymy.  But although it is possible to include many other
kinds of relationships between semnods in an ontology, it is not necessary
to do so in order to create an ontology sufficient for the parsing of
natural languages.  This is because it will be possible to add linkages
encoding such things as potential agency, potential patiency and various
other types of relationships between semnods almost instantaneously at a
later time, as we shall see.

It seems clear that animals have ontologies whose nodes they associate
with various sights, sounds, smells, and sensation patterns.  They are
aware, for example, of moving and non-moving classes of objects, which
information can be best represented by means of hypernymy.  They recognize
individual members of various species as identical objects, which could
best be done using either hypernymy or synonymy.  They recognize the
various parts of things, which might best be accomplished using holonymy.
 If so, then the linguistic link must be part of the animalÀmind just as
much as it is of the human.

How to build your own ontology.

If you are using C code, begin with a structure like the following:

typedef struct {
	unsigned char typ;
	unsigned short dst;
	} lnk_t;

This selfsame structure can be used to encode both nodes and links.  As a
node, the "typ" field serves as "number of elements in this entry," and
the "dst" field serves as node number.  Number your nodes sequentially
beginning with 1.  If, as an example, a node has five links emanating from
it, then the "typ" field for the element being used to represent the node
should hold a value of 6--that is, the element being used for the node
itself plus five elements representing links.  The elements representing
the links, always emanating from the node, then immediately follow the
element representing the node.  Thus it can be seen that an entire
ontology can be built from nothing but a long array of identical elements.
 The entity traversing the ontology in search of information can then
search for any node by going to the first element of the array, checking
to see whether the number of this node is what is being searched for, and
if not, then adding the value of the "typ" field to get to the start of
the next node and checking it, and so on.

It is important to notice that links other than semlinks (links to other
semnods) also emanate from most semnods.  The most common of
these are lexlinks to symbols (written words) in the lexicon.  Another
common link type is that indexing a line in a file of word-sense
definitions.

Section 5.  What Is A Lexicon?

A lexicon is a collection of nodes called "lexnods," each of which is
linked to the mechanism used to generate a word symbol (spoken or written
word) in a natural language.  The mechanism that recognizes the spoken or
written form of the same word is linked back to the same lexnod.  There is
thus a one-on-one correspondence between the nodes of the lexicon and each
separate sound or text segment representing a word in the outside world.
 This means, for example, that there exists a separate lexnod for each of
"talk," "talks," "talking," and "talked," etc., but that there is no
separate lexnod for "-s," "-ing," or "-ed."

Each node in the lexicon is also linked TO by one or more links from the
ontology.  Recall that every English word symbol has at least one meaning,
and may in fact have several very different meanings.  Each node in the
ontology is associated with just one and only one meaning, as we have
already seen, thus for all intents and purposes each node of the ontology
can be thought of as just a meaning which can be defined by a single
word-sense definition.

So in order to summarize what I have written about linkages thus far, each
semnod in the ontology is generally linked to one or more other nodes in
the ontology and to one or more nodes in the lexicon (lexnods), whereas
each node in the lexicon can have the following kinds of linkages:

1.  An outgoing link through some kind of generating device to a spoken
and/or written word.

2.  An incoming link through some kind of recognition device from a spoken
and/or written word.

3.  One or more incoming links from semnods in the ontology, a separate
one for each potential meaning corresponding to a word-sense definition.

It appears that every human being has a built-in lexicon and ontology for
every language he/she knows.  As we have already seen, each node in an
ontology is linked to a meaning, which is roughly the same as a word-sense
definition of the kind one might find in a dictionary.  The owner of the
ontology is therefore able to know the exact meaning of every semnod.  In
addition, most semnods link to one or more lexnods, each of which is
linked by pattern recognition and pattern generating units to word symbols
(written or spoken words) in the natural language by means of links of
various types.  For example, for the semnod associated with the meaning,
"to go," we might have the following linkages to lexnods connected (as
described) to English word symbolss:

1.  Present tense verb, "go."

2.  Third person present tense verb, "goes."

3.  Past tense verb, "went."

4.  Present participle, "going."

5.  Past participle, "gone."

And in addition to these links to English word symbols, the same semnod
may link to other semnods as follows:

1.  Hypernym: The semnod whose meaning is "to move."

2.  Antonym: The semnod whose meaning is "to come."

Notice that in the case of links from semnods to lexnods linked to English
word symbols, the link types are just parts of speech.

It is apparent that animals are equipped with lexicons because the sensory
patterns they recognize must be reduced to lexnods in order to receive
their appropriate lexlinks from the semnods of the ontology.  For example
a primate may resolve some pattern it sees to the lexnod for "panther,"
which is linked to by a lexlink from the semnod associated with panthers
in its ontology.  Yet animals are not equipped to generate identical sound
patterns for ALL of the sounds they recognize, so that only a limited
number of these lexnods are linked to mechanisms for generating sounds.
 So the primate may be able to produce an alarm call for "panther," but it
may not be able to say, "I saw a panther in that tree," or even generate a
call representing "to see."

Section 6.  What Is A Word?

Every word in every language known to man, when part of a coherent thought
or sentence, is nothing but two links emanating from a single node.  One
of these links connects the word to its meaning, and the other link
connects the word to another word within the same thought or sentence
called its REGENT.  The node of the word is the source of these two links,
and may serve as the destination for links from other words.  The only
exception to this rule is the top word of the thought or sentence, which
has a synlink that would ordinarily link to another word but does not.
 Instead it has a link type but no link destination--in other words it
goes NOWHERE.

In general whenever we speak of a word, we are referring specifically to
the node from which its two links emanate as described above.  Thus for
example, in the sentence, "John loves Mary," we may say that, "The word,
'John,' links to the word, 'loves,' by means of a synlink of type
'subject,' and 'loves' links to the semnod associated with the meaning,
'to love.'" Etc.

Every word in every sentence has a syntactic and a semantic role.  The
syntactic role of a word will be something such as "doer of the action,"
"object of the action," "time of the action," etc., whereas its semantic
role will usually just be what is commonly called "part of speech."  In
Panlingua these syntactic and semantic roles are encoded in LINK TYPE.

As an example, let us consider the sentence:

John loves Mary.

Let us then number these words as follows:

1.  John.

2.  loves.

3.  Mary.

Then the syntactic roles for this sentence will be:

1.  Doer, or subject.

2.  Action being performed.

3.  Thing affected, or object.

And the semantic roles will be as follows:

1.  Name (proper noun).

2.  Present tense, third person use of the meaning, "to love."

3.  Name (proper noun).

Since these syntactic and semantic roles are encoded in link type, all
that remains to be explained about this sentence are the link
destinations.

The three synlinks ( links that ordinarily link one word to another) have
destinations as follows:

1.  "John" links to "loves."

2.  "Loves" links to nothing.

3.  "Mary" links to "loves."

And the destination of each semlink is just a semnod in the ontology
associated with a particular meaning.

So combining both kinds of linkage, we obtain:

1. John. Synlink type: subject.  Synlink destination: loves.  Semlink
type: proper noun.  Semlink destination: the semnod associated with this
particular John.

2. Loves.  Synlink type: declarative.  Synlink destination: nowhere.
 Semlink type: 3rd person singular present-tense verb.  Synlink
destination: the semnod associated with the meaning "to love."

3. Mary.  Synlink type: object.  Synlink destination: loves.  Semlink
type: proper noun.  Semlink destination: the meaning associated with this
particular Mary.

Notice that the type of the synlink emanating from the top word of the
sentence and going nowhere is the type of the whole sentence (declarative
sentence).  Also notice that by "meaning" is meant "semnod."

A great deal of confusion arises from the various uses of the word, WORD.
 Here the sounds and groupings of written characters commonly construed as
"words" are not what is intended.  What we mean are rather those invisible
internal structures consisting of two links emanating from the same node
that exist only within our minds.  From this point of view, the sounds and
scribblings of the external world are NOT words, but mere symbols used to
represent words in communications.  Once the external symbol for a word
has been shed, it consists of nothing but two linguistic links emanating
from the same node, and since all linguistic links are invisible, massless
patterns, all words are also nothing but invisible and massless patterns
existing in the human mind.  The external sounds and written characters
representing words, which we often call "words," are then not really words
at all, but only the symbols representing words in the outside world, the
real words being internal and invisible.

Section 7.  What Is A Panlingua Array?

A Panlingua array is the internal representation of a phrase, sentence, or
string of sentences whose words are not external symbols such as spoken or
written words, but  units consisting of two links emanating from a common
node as described in Section 6.  The word, PANLINGUA, is derived from the
Greek word, "pan," meaning "all," and the Latin word, "lingua," meaning
language or tongue.  Its intended meaning is "universal language."

Of course a thought may be the thought of a single thing, for example the
thought of a cup of coffee.  Or it may be of something ordinarily
represented by a phrase, such as, "What a pretty flower!"  Or it may be
any other constituent part of a complete sentence up to and including the
entire sentence itself that can be represented as a Panlingua array
consisting of a regent and its dependents.

Since every word except the top word of a Panlingua thought representation
must have a regent, that is, just one and only one regent, it will be seen
that Panlingua arrays naturally assume patterns similar to those of the
file structures of computer disk operating systems.  Each word is like a
computer-disk file or folder, and each regent is like the folder in which
a particular file or folder is found.

But the human mind is not flat like the surface of a computer screen.
Instead it is modeled in three dimensions.  What this means is that
although we may know that "John" and "Mary" are dependents of "loves"
(have "loves" for their common regent) in the example sentence, there is
no way to tell just by looking at the structure of links and nodes whether
"John" or "loves" or "Mary" should appear first in the linear form.  If we
were to pick the tree representation up by the top verb, "loves," and
allow "John" and "Mary" to hang free, then it would be possible but
difficult for us to know which word should come first, which second, and
which third in the linear, textual representation.

Words like "John" and "Mary" in the above example are said to be of the
same "grammatical rank."  Words are of the same grammatical rank if they
have the same regent.

If we study many languages, we will learn a curious fact.  In noun phrases
like "the big red ball," it will be found that no matter whether the
adjective comes before or after the noun in that language, "red" must
always be closer to "ball" than "big."

Another reason why it is necessary for internal thought representations to
retain information about linear sequencing is the fact that in narratives,
one thought must occur after another in chronological order.

So although the internal representation appears to be a tree in three
dimensions, it must also retain information about the closeness of words
to each other and how the thought can be output in linear form.  This may
seem to be difficult at first, but after some thought it will be seen that
it can be done by linking the leftmost word directly upwards to its
regent, linking the second word of the same grammatical rank indirectly to
the first word, linking the third word to the second, etc.  The immediate
dependent (dependent of the same grammatical rank as the one linking
directly upwards to its regent) before which the regent should appear
during text generation must then somehow be marked by the system, and if
no direct dependent is marked in this way, then it will be assumed that
the regent must come at the end of the phrase.

This kind of linkage can be represented on a piece of paper in patterns
that tend to take 'L' shapes with links always running leftwards or
upwards.

Just how such thought representations are set up in our minds is not
completely clear.  Are memories encoded in brain cells?  In some kind of
molecule that floats around in some part of the brain or is pumped around
the whole body in the blood stream?  No one really knows.  Sometimes we
know things but cannot immediately remember them.  We tell our own minds
to search for them, and after some hours, we are able to recover them.  It
is as if these memories consisted of tiny snippits being pumped around in
our blood.  Our minds watch for them and snag them as they pass by.  Then
we make many fresh copies and release these back into the blood stream,
and for awhile it is easy to remember what we had forgotten again.  But
there must be some way we are able to access millions of thoughts in split
seconds for linguistic analysis because the process of understanding
(which is really just parsing) is very fast indeed.

Section 8.  What Is An Interlinguish Array?

Panlingua arrays are not material objects but patterns.  It is therefore
possible to model Panlingua arrays in many materials.  Some of these
models will be functional (able to do things), while others will not.  For
example a Panlingua array drawn with pen and ink on a piece of paper will
be capable of representing a snapshot of the Panlingua array it
represents, but it will be incapable of doing anything or having anything
done to it except for having a human interpret it and perhaps scribble
some alterations or notes.

But when properly represented in an automated system, a panlingua array
can cause any number of events to occur and be acted upon to produce any
number of results.  As a command it can trigger a process and control that
process by means of the arguments it contains.  As a question it can
trigger a search for an answer.  As a statement of fact, it can be added
to a knowledge base.  And no matter what it is, it can be used to generate
text in a natural language, translated into other natural languages,
analyzed to create links in an ontology, and probably a lot of things
nobody has even ever thought of.

An Interlinguish array is a special kind of Panlingua array--one modeled
using computer memory instead of neurons as the medium.

WARNING: Here this text becomes somewhat technically involved.  The casual
reader is therefore encouraged to skip ahead to Section 9.

The Interlinguish representation of a word contains the following fields:

1.  The number of words in the subtree headed by the current word (whose
regent is the current word).  This number is always 1 or > 1 unless the
current word is an array element beyond the end of the Interlinguish
array, in which case its value is 0.  A value of 1 means that the current
word has no dependents, 2 means that the current word has one dependent,
etc.

2.  A numeric code representing synlink type.

3.  A numeric code representing the type of the link to the meaning of the
word.

4.  A "flags" field whose bits are used to mark various things about the
word--is it capitalized, is it the word immediately following its regent
in linear texts, is it a quoted word, etc.).

5.  An integer value identifying the semnod associated with the meaning of
the word.

The reader will notice that things do not immediately appear to be as they
should be.  A Panlingua word is defined as two links emanating from the
same node, one leading to a semnod in the ontology, and the other leading
to another word in the same sentence, namely the regent of the current
word, unless the current word is the top word, in which case this link may
lead nowhere.  Yes, we have a field for the type of the link from the
current word to its regent.  Yes, we have a field for the type of the link
from the current word to its semnod.  And yes, we even have an integer
identifying this semnod.  But there is no identifier for the destination
of the link from the current word to its regent, and we have these two
extraneous fields, one for the number of words in the current subtree and
one to be used as a "flags" field.  Why?

The words of an Interlinguish array are elements of a computer array
indexed 0, 1, 2, ..., which are instantiations of the word data structure
just specified.  But unlike the words in a linear text, in an
Interlinguish array, the regent always comes first (has the lowest index
in the subtree) followed by its dependents.  The direct dependent before
which the regent must appear in a linear text is marked in its "flags"
field, and this is the most important reason for the existence of the
"flags" field in the word data structure.  And since each word structure
has a field holding the number of words in the subtree, it is not
necessary to specify the identity of the regent in the word structure
because when the "number of elements in the subtree" field is added to the
current computer array index, the result is the index of the first
position beyond the end of the current subtree.  Thus any word whose array
index is greater than that of the current word and less than this result
is a dependent of the current word and the current word is its regent.

The Interlinguish specification given above may seem cumbersome at first,
but in fact it lends itself very elegantly to all kinds of automated
manipulations and makes possible some very fast search-match operations.

Section 9.  What Is Parsing and What Is Understanding?

Parsing is the conversion of spoken or written phrases into Panlingua
arrays.

In general, all that is necessary to convert a sentence from English or
some other spoken or written language into a Panlingua representation is
to determine the two links for each word in the sentence and then abandon
the written or spoken symbol for the word.  Once all the internal linkages
have been established, the meaning is precise and unequivocal and no other
kind of internal structure is necessary for the representation of
knowledge.

Caveat: This one-on-one mapping holds for languages like English, but will
not work for languages like Malay, which contain invisible "be" verbs.  In
such languages the subsurface form (the internal representation gotten by
determining internal linkages and dropping surface symbols) will sometimes
have more words than the surface form for a particular sentence.

This can be seen in sentences that transliterate to constructions such as
the following:

John Sunday School teacher last Sunday.

We might be tempted to try to make "Sunday School teacher" the predicate
with "John" as subject, but we quickly find that in that case "last
Sunday" must also be made a dependent of "Sunday School teacher," and this
will not do because we know that "last Sunday" is a temporal modifier
indicating when something WAS.

And because it is easy to find many more similar examples, we are forced
to the conclusion that such constructions do indeed contain "hidden"
auxiliary "be" verbs which do not appear in the surface structure.

When the words of an incoming sentence are compared against the internal
representations of sentences we have already parsed, this comparison
occurs at several levels:

1.  A simple, one-on-one match may be found.

2.  A match may be found to a hypernym or synonym.

3.  There may be no match in meanings at all, but a part of speech match
may be found (the human brain can automatically handle part of speech).

In any case, a "best match" is determined after comparing between the
incoming sentence and thousands of successfully parsed sentences, and this
best match is used to select the correct pair of links for each unparsed
word.  In many cases it will not be possible to parse the entire sentence
based upon any one internal thought representation, but the incoming
sentence will have to be parsed piecemeal using subtrees from different
parsed sentences.

A spoken or written phrase that has been successfully parsed in this
fashion can be said to have been "understood" by the automated system.
Thus understanding is really just correct parsing.

Section 10.  What Is Text Generation?

Text generation is the conversion of Panlingua representations into spoken
or written words.

Text generation involves going to the head (regent) of a subtree and
recursively outputting its dependents taking care to position this regent
in its proper place in the text.  The recursive function looks only at the
regent and its direct dependents, re-invoking itself using each direct
dependent as head.

If the position of the regent has been marked (if there is always a mark
on the word that should come right after the regent in the linear text)
this is a very straightforward and a very high-speed operation.  But if
there is no such mark, the text-generation function will be forced to
determine regent position by an examination of the Panlingua
representations of previously parsed texts, and this may turn out to be a
very tedious and lengthy process.

Whether or not regent positions are marked in Panlingua representations
existing within the human mind is a fascinating question.

Section 11.  On The Fallacy of Grammatical Rules.

For centuries scholars have labored over grammatical rules, and many
people have attempted to build computer models of language based upon
them, but these never work for computers OR for people.  This is because
language does not work by means of grammatical rules but by immitation.

In fact it now appears that, at least for purposes of parsing and text
generation, there are essentially no such things as grammatical rules,
that these are only a figment of human imagination, and that it will never
be possible to deal with the complexities of any real natural language
using a set of grammatical rules.

One way to see this is by noting how many rules are necessary to do even
the simplest parsing, and then multiplying even these rules by all there
many exceptions.  The number of grammatical rules and their exceptions
quickly explodes into an intractable "spaghetti bowl" problem.

Another way to see this is by the fact that if people were learning
language using grammatical rules it would be impossible for a child born
into one language group to learn any other language.  He would have to
have too many ad hoc rules in the programming of his brain in order to
handle even his own native language--if indeed this were even physically
possible, which it is probably not.

And as a last reason for the impossibility of learning languages by means
of grammatical rules, I will evoke Ockham's razor.  I have given the
simple solution above, which requires only a minimum number of items of
inherited preprogramming plus one or two more unknown items.  All of the
major parts of this process have been successfully modeled on computers
and proven to work.

The great temptation and deception of rule-based systems is that they
appear at first to work well and do work well over very limited domains.
This result initially gives the experimenter a feeling of achievement and
power, and suggests the belief that if the rule-based system can work over
a limited domain, then it should be expandable to work over any domain.
But what the experimenter eventually discovers (or sometimes never does
quite discover) is that there are simply too many rules to handle all of
any language, and that these rules may even conflict with one another
requiring special ad hoc code to make them work.  So the experimenter is
led willy-nilly down a long and slippery path of creating kludge upon
kludge upon kludge.

Section 12.  Back to the Baby Brain.

Armed with these new facts about language, let us now reconsider the
process of infant language acquisition described in Section 1.

If what I have just written is correct, then every healthy human child
must be born with the following structures, functions, and capabilities
already built in at some subconscious level:

1.  A working knowledge of the linguistic link, which is the fundamental
building block of all linguistic structures.

2.  A working knowledge of the two-link, one node word, which is the
fundamental building block from which all thoughts and sentences are
constructed.

3.  Space for an ontology to be modeled from radlinks and lexlinks.  The
ontology for a natural language may require enough space for ten to twenty
thousand semnods plus the radlinks between them plus the lexlinks to
lexnods for external symbols.

4.  Space for the creation of thousands of functions to recognize and
reproduce spoken and/or written word symbols.

5.  Space to model Panlingua arrays using words.  Enough space will be
required to hold the panlingua representations of many thousands of
sentences.

6.  The ability to generate functions to identify different speech sounds
or other speech symbols.

7.  The ability to generate functions to accurately reproduce various
speech sounds or other speech symbols.

8.  The ability to associate semnods in the ontology with objects in the
real world and with sounds or other symbols representing those objects in
the real world.

9.  The ability to deduce the parts of speech of other word symbols by
means of speaker intonation and word-symbol position within the linear
text stream.

10.  The ability to learn the meanings of new word symbols for which only
the parts of speech are known and to link semnods in the ontology to the
appropriate lexnods for these word symbols.

11.  The ability to readily convert between Panlingua and textual
representations of sentences or thoughts.

Armed with these essential capabilities and one or two more that we don't
yet understand, the infant mind is ready to learn any language known to
man.  At first he will learn to recognize and reproduce speech sounds.
Then he will learn to link particular sounds to semnods reflecting objects
in the real world in an ontology.  And finally he will learn to manipulate
words into dependency relationships in internal structures and into linear
positions in arrays of spoken words.

The reader will notice that many animals have the ability to associate
sounds with objects in the real world, and some may even have the ability
to generate symbolic sounds representing things in the real world.  It
would also seem true that animals are able to think in terms of Panlingua
arrays and their two-link, one-node words.  But what they are unable to do
is to translate these internal representations into linear strings of
spoken or written words and vice versa.

In the example of infant language acquisition given in section 1, the
language, of course, was English.  The infant heard his mother pronounce
two words.  One of them, the one he already knew, was unaccented, while
the word he did NOT know was accented.  In English, the accented word came
first.  In Malay it would also have been accented, but it would have come
second.  As a general rule, it is found that modifying words are accented
whereas modified words are not.

When the infant finally understands (which means "can successfully parse")
the result of his parsing (the Panlingua structure representing "Big
bird!") is retained, and this enables the infant to parse a host of other
adjective-noun pairs.  No grammatical rules are involved.  It simply
happens that within his baby brain the infant holds a Panlingua
representation indicating that an adjective is followed by a noun in the
case of "Big Bird!" which indicates that there may or may not exist a host
of other similar adjective-noun pairs just waiting to be parsed in the
same fashion.

If the infant had instead parsed the Malay representation, which is
"Burung Besar!" the Panlingua representation would have indicated that the
pair was noun-adjective with synlinking in the opposite direction,
and all would have worked just the same, once again with not a single
grammatical rule involved.

The even tone of the mother's voice when pronouncing "Bird flies!" without
emphasis on either word would also have occurred in Malay if the mother
had
been a native speaker of that language and said, "Burung terbang!"

The rules for deducing object or patient nouns also works exactly the same
for every language I know of.

I will not attempt to delve into the processes involved in deducing
morphology here because I find them difficult to determine and of little
importance to the overall picture of language as a whole.

It has been noticed that children make grammatical errors but either will
not listen to those who correct them or else never get corrected, and yet
somehow mysteriously manage to correct these errors automatically on their
own.  This apparently happens because the infant brain does not retain the
Panlingua representations that it generates itself, but only those
Panlingua representations of incoming sentences that it has successfully
parsed (understood).  Once an incoming sentence has been understood, it
becomes part of the Panlingua archive of the child, and thereafter it is
available as a template to be used for future thinking, parsing, and text
generation.  In other words the infant mind appears to generate thoughts
in Panlingua using the Panlingua representations of previous successful
parsings as templates.  These thoughts then may or may not be expressed in
spoken words, but their Panlingua representations are soon discarded
because they were self-generated and not the result of parsing.  But when
a child successfully parses a sentence he believes to be spoken by a
reliable native speaker of the natural language he is learning, the
parsing results are held for the long haul.

Section 13.  What Is A Language?

It is often convenient to call Panlingua a language.  Actually it is not.
Instead it is the universal subsurface structure built of links and nodes
which underlies ALL languages.

A man named Xul Solar once created an artificial language and named it
Panlingua.  I was not aware of this when I named the universal
underpinnings of all languages "Panlingua."  The Panlingua that I have
defined has nothing to do with artificial languages and is not itself even
a real language.

In general, a language is a set of external symbols that assume linear
patterns determined by precedent in order to represent meanings.  Such
linear arrangements of symbols are often referred to as "texts," while the
symbols themselves are referred to as "words."

The linearity of language can be expressed as a line of symbols written on
some medium, scribed into clay tablets, or carved into wood or stone.  It
might be a line running parallel to the X or Y axis in a Cartesian
coordinate system, or it may appear as a sequence of symbols appearing
through time, as in spoken words, the symbols of sign languages, etc.

The meaning of each symbol in a string of coherent text is fully
determined
by two invisible/inaudible links: (1) a synlink, usually to the
regent of the current word, and (2) a semlink to the actual meaning of the
word.

Having thus armed ourselves with this rigorous definition of "language,"
let us proceed to identify some things that are clearly languages.

1. All natural languages spoken/written by humans are languages.

2. All computer programs are written in some language.  Computer languages
are much simpler than natural languages, each sentence or "thought"
consisting of a top word (an "operator") and 0 or more dependents of this
top word (its "operands").

3. Math.  Mathematical expressions are much more complex than computer
programming languages, but still far less complex than natural languages.

In many cases it is easy to see that a mathematical expression is just a
meaningful linear arrangement of symbols as in any other kind of language,
and can be easily translated into a linear arrangement of symbols in some
natural language.  For example, "1+1=2" is just "one plus one equals two."
And even such an expression as "E=MC^2"Àis simply "energy equals mass
times
the velocity of light squared."

But what about expressions such as"( (1 - sqrt(2) ) / 3 ) + 4?"  Oops,
still no problem.  Simply, "The quantity, one minus the square root of
two,
end quantity, this quantity divided by three, to which entire result is
added four."  Or:

Subtract the square root of two from one.
Divide the result by three.
Add four to the result.

All of which looks suspiciously like computer code, which is just another,
albeit very simplified, human language.  And if we keep on going, we will
find that it is impossible for us to come up with any mathematical
expression that cannot be readily translated into linear strings of words.
Therefore it is obvious that all of mathematics, no matter how surprising
or elegant, is just more human language and operations based on human
language.

The Two Great Fallacies of Modern Science.

To the uninitiated, mathematics may seem utterly magical and spookey, and
because the human mind is inherently superstitious, this has opened
science
up to two major fallacies that smack of "urban myth."

1. The universe is governed by mathematical laws.

But as we have seen, mathematics is just another form of human language.
So saying that the universe is governed by mathematical laws is just like
saying the universe is governed by English grammar!

2. Mathematics is the mother of all sciences.

Not only mathematics, but also plain English can be used to describe any
science, but this does not make English the mother of all sciences.

It is indeed sad to see how quickly genuine scientists who have been
trained for the merciless application of scientific rigor, can be so
easily
blindsided by such assumptions that have snuck in unawares without the
slightest proof to back them up.

There must be many examples of failure attributable to the two above
fallacies, but I will mention only two that I perceive to be real:

1. Albert Einstein did poorly in math, but happened upon the greatest
physics discoveries of the 20th century by daydreaming about such things
as
watches and elevators and trains, and fitting these daydreams into the
framework of what he knew.  Yet when it came time to mathematically
formalize and communicate his discoveries, he had to rely upon friends.
 In
the end, he realized the importance of the math he had been skipping, went
back to his books, and became a "great mathematician," and modern physics
was reduced to the working out of mathematical details describing quantum
phenomena.  According to those who knew Einstein personally, he became
convinced that "the universe is governed by mathematical laws."  So
instead
of continuing to go for the substance of science, it appears that Einstein
was unwittingly seduced by the language describing this substance.  He
died
trying to come up with a unified field theory for physics by means of
math.

2. Noam Chomsky told the world that linguistics was based upon mathematics
in the early days of computational cognitive science, the world believed
him, and (at least for the time being) that was the end of computational
linguistics and AI.  The obvious fallacy of course being (for those who
still don't quite get it) that linguistics CANNOT be based upon math since
math is just more language.

So mathematics is the mother of nothing and never was the mother of
anything nor does it "govern" anything, no matter how much the idea of
being at the controls of some governing principle may fill men with a
false
sense of power.  Mathematics is just another language with which to
describe the free and untrammeled universe, which can be described in any
language man cares to dream up.

But then how did Einstein figure out relativity anyway?  This is a
question
I can't resist toying with.  The human mind is very complex, and exhibits
knowledge and abilities that far exceed what we understand at a conscious
level.  Yet it appears that this powerful subconscious mind is either
mostly cut off from our conscious awareness or else works so differently
from our conscious mind that the two exist more or less "on different
wavelengths."  It ought to be easy to see this just from what I have
written in this paper.  How many people you know, for example, would be
aware that they are walking around with a built-in ontology?  The key
would
therefore seemÀto be to give full reign to the imagination, where the
subconscious and conscious minds can somehow interact and integrate what
we
know from outside with what we know from inside and come up with pictures.
I am writing these things because I consider them to be of critical
importance to the future of mankind.  This is in fact how I discovered the
details of Panlingua.  I imagined them in three dimensions, trusted to the
wisdom of my subconscious mind, and gave full reign to my own imagination.
But these things do not happen immediately.  Instead the various elements
form slowly and become elucidated and integrated into the imagined whole.
So the greatest discoveries of all time are still waiting to be
apprehended
by those who learn how to think, and this goes far beyond any knowledge of
math.  I could say much, much more on this subject, but must leave it in
order to get on with the central theme of this paper.

Section 14.  The Answers to some Hitherto Very Puzzling Questions.

Based on the theory described above, here are my answers to some questions
of the type put forward at one time or another by such well known
researchers as Noam Chomsky, Steven Pinker, and others:

How do children know so much by the time they are four years old when it
would seem like they have not been exposed to enough information for them
to know what they know?

The answer lies in the fact that in the child's mind is an ontology that
can provide all kinds of answers.

Suppose that the child has created an ontology containing the following
kinds of information:

hypernymy synonymy antonymy holonymy of inalienable possession holonymy of
alienable possession potential agency potential patiency potential state
assumption.

I have given just eight kinds of relationship between semnods.  In fact
there are more, but these are the main ones.

Now suppose that in the ontology of the child there are 5,000 semnods,
some for things, others for states and actions.  The child can then query
his ontology about the relationships between any of 5,000 semnods and
4,999 other semnods.  Now 5,000 x 4,999 = 24,995,000 possible
source-destination pairs.  But for each of these he is able to test  not
just for one kind of relationship, say hypernymy, but for eight.  So even
with this limited ontology he is able to get YES or NO answers for 8 x
24,995,000 = 199,960,000 possible connections.

For many of the queries he makes, of course, he will get NO answers
because his ontology will not be able to find linkages.  On the other hand
ontologies work such that just one new link established can result in the
validation of thousands of new linkages.  This is because hypernymy and
synonymy work transitively to provide access to other kinds of
relationships.

For example, suppose that the ontology of the child contained the
following information:

birds can fly (potential agency)
ducks are birds
geese are birds
starlings
are birds
cardinals are birds
parrots are birds
cockatoos are parrots
lories are parrots
Etc., etc., etc.  for many kinds of birds.

Then for any query such as:

Can geese fly?

the answer will be YES, because a goose is a bird and birds can fly.

And not only will the "birds can fly" linkage be multiplied over all
birds: it will also extend to anything that is something else that is a
bird.  So for a query such as:

Can lories fly?

the answer will be YES because a lory is a parrot and a parrot is a bird
and birds can fly.

So by simply taking note of various kinds of relations between various
meanings associated with semnods in his ontology, the child can quickly
gain a perfectly amazing grasp of the essentials of his environment.

Now as I have already mentioned, there are only about four kinds of
relationships between semnods necessary for purposes of parsing.  But once
the child has accumulated a corpus of Panlingua representations, in other
words a large corpus of parsed sentences, it is possible to add entries
for the other four kinds of relationships I have mentioned with amazing
speed.  This is because all of the kinds of relationships that are not
essential for parsing will be found embedded somewhere or other in these
Panlingua thought representations, and it is a very simple matter to
extract them and add them to an ontology if required.

So the first answer to the question put forth by the "poverty of stimulus"
argument would seem to be the ontology.  The other answer is the fact that
the child is parsing sentences not using rules but by scanning through
precedents and examples, as I have already shown.  Just as they contain
information from which the child can build an ontology, the accumulated
corpus of Panlingua arrays can also provide grammatical information, and
this is why the child has a grasp of both semantic and grammatical
information that at first glance would seem to be nearly miraculous.

Caveat: This is not to say that an ontology can be created in this way
from scratch.  Except in rare cases, It appears to be impossible to infer
such things as hypernymy, synonymy, antonymy, or holonymy by examining
Panlingua arrays, whereas these are the very types of relationships that
are critical in parsing.  How a knowledge of these fundamental
relationships between semnods is acquired automatically by the infant mind
remains a mystery.  It may be that in every case the infant is being
specifically told without anyone realizing that this is happening, and
that this is what is going on at a high rate during the "question box"
stage of child development.  More study is needed.

Another question that has hitherto eluded a satisfactory answer is why,
when confronted with linear strings of words in a spoken or written text,
a child will always attempt to analyze their relationships to one another
in terms of phrases and not at a word level.  For example, in the
sentences:

The man in the moon is ugly.

and

The moon is ugly.

Why doesn't the child ever infer that "the moon is ugly" from the first
sentence, seeing that"moon" always appears right before "is ugly" in both
cases?  It would appear to be much more difficult to group "moon" into the
phrase, "The man in the moon," than to make it part of "moon is ugly," and
yet children never seem to make such mistakes in understanding.

This would seem to be the result of two things.  In the first place, the
child will probably reject any sentence he cannot parse completely out of
hand.  Thus if he were to break the sentence up as follows: The man in |
the moon is ugly, although he might be able to parse the second half
successfully, the first part would remain unparseable and the child would
reject the whole sentence out of hand.

In the second place, from the very beginning the child has been parsing
groups of words and setting them up in Panlingua arrays.  His goal, from
the outset, has never been to remember strings of words, but only to set
words up internally in Panlingua arrays.  His mind has been programmed to
do this, and this is why although it may be hard for him to learn nursery
rhymes, he can remember all kinds of things about the relationships
between semnods and words.  Now in a Panlingua representation every phrase
is a subtree of words, and every such subtree has just one and only one
head, which means that once this head has been established, all its
dependents can safely be ignored in further parsing.  A quick scan of his
internal corpus of Panlingua arrays will probably reveal a match to "the
man in the moon," which will tell the child how to parse the phrase and
leave him with only "man" to deal with in order to parse the remainder of
the sentence.  He will also find a match for "man is ugly," "or person is
ugly," or just "thing is ugly," and this will tell him how to parse the
rest of the sentence.

Such fast conversion and comparison may seem impossible at first, but in
fact it can now already be done using modern computers, which are known to
be slower than human brains.  It is therefore not at all unreasonable to
assume that this is precisely what is going on inside the child's head.

Why does it appear that people think in languages while it is well known
that animals can think without being able to speak, people who have never
learned to speak or understand words can think, infants can think before
they can use language to communicate, and normal adults can think without
speaking (although many do talk to themselves)?

Using an ontology it is obviously possible to set up thoughts, which are
Panlingua structures made up of words which are each just two links and a
common node, without any reference to external symbols (spoken or written
words).  But as a child develops his linguistic abilities by parsing
incoming sentences and building his corpus of Panlingua arrays, it is only
natural that he should come to rely more and more upon this growing corpus
of Panlingua arrays to guide him in setting up thoughts and to add
sophistication to his thoughts by drawing upon the common oral heritage of
his people.  This is probably one reason why, although animals can think,
they cannot think as well as humans--they do not have the oral heritage of
a people upon which to draw in order to refine their thoughts.

Feral children (children raised in isolation from all human contact) are
evidently the brightest of all non-human animals.  They have the most
sophisticated brains in the animal kingdom, and yet, just as this theory
predicts, they never become fully human.  We read of example after example
of feral children being mentored by highly trained persons, yet except for
the Romulus and Remus of Roman mythology, none of them ever becomes fully
human.  And this, according to our theory, is to be expected because it is
the corpus of parsed sentences, instituted at the inception of human
speech development and built upon ever thereafter, that makes humans
really human.

Two more questions are often asked by those who ponder what it is to
remember.  One is, "Why can't I remember anything that happened before I
learned to speak?" and another is, "Is it possible for any human being to
remember anything from the time before he/she began to speak?"  As we have
seen, it is the corpus of parsed sentences within us, which I have called
Panlingua arrays, which defines all of the essential ingredients of our
humanity beyond a purely animal existence.  So, although this may be a
very hard statement to accept, before we begin to speak, we are not fully
human, and our humanity is a function of the richness of our speech
experiences.  One piece of smoking-gun evidence for this has been the
acceptance of the custom of male circumcision without anesthesia.  Yes,
anesthesia may be used during male circumcision in modern nations today,
but this was not the case in just the very recent past.  This is because
there was an unspoken agreement among doctors and parents that newborn
infants are not fully human, therefore it is okay to circumcise them
without anesthesia.  Many doctors are able to slice the foreskin from a
bawling baby, but I think it would be hard to find many to do this if the
infant were saying, "Don't do that, doctor--you're hurting me."  So it is
probable that most of us cannot remember anything before the time we began
to speak because it is the corpus of parsed sentences within us that truly
defines us as full-fledged human beings.  This is how our humanity begins,
and this is how it continues, and when we stop parsing sentences, either
through death, Alzheimer's disease, or else some other process, that is
where our humanity will end.  All ideas of reincarnation, past lives,
etc., must therefore be delusional at best.

But is there life after death?  This would certainly seem possible.  If
the true essence of a human being is a corpus of parsed sentences he/she
has developed over a lifetime and an associated ontology, then all that is
needed in order to preserve a human life beyond death is simply to copy
this corpus and ontology onto some medium and store it some place for
later re-use.  This may seem fantastic to us at present, but just think
what scanning a package at a supermarket would have seemed like in 1900.
How can we be sure there are not other, more advanced intelligences in the
universe with whose technology it might be possible to scan a whole human
mind by just passing a wand over their head like a clerk passes a wand
over an item at a supermarket?  And in fact there may be much more
efficient ways of reading the contents of human brains than even by
passing wands over their heads.

Do languages play a role in shaping personality and determining patterns
of thought?  Almost certainly.  A child begins to think in some language
or some group of languages for which he has developed an ontology and a
set of Panlingua arrays, and the kind of thinking normal in some language
or other may pervade his thinking so strongly that it is almost impossible
to think outside its constraints.  As an example it is nearly impossible
for some Vietnamese immigrants to bring themselves to differentiate
between the colors, "green" and "blue."  This is not because they are
physically unable to tell the difference between green and blue, but
because they are using a Panlingua corpus built by parsing sentences in
Vietnamese, in which no distinction has ever been made between these two
colors.  Because of this and other similar phenomena, therefore, it must
be assumed that in some sense acquiring a language is not only learning to
talk in certain ways but learning to think in certain ways as well.

Are human languages too ambiguous and unweildy to be used for thinking?

People will often argue that language cannot be the basis for human
intelligence because it is too ambiguous and slow.  "When a person in the
street is about to be hit by a car," they may say, "does he have time to
say, 'Oops, looks like trouble!  Now if I just move my right foot sideways
really fast and get my body over it, maybe he will miss me.  Now let me
see...  According to Newton's third law of motion...' etc.? Obviously not,
right?"  The problem with this argument is that although vocalizing words
in the real world may take time, the internal workings of language are
lightening quick.  Recall that in order to parse a sentence it may be
necessary to scan through many hundreds or thousands of parsed thoughts,
yet this process takes place so quickly as to seem almost instantaneous.

The other part of the argument is that language is too ambiguous.  "It
would be impossible for anyone to think using language," they may say,
"because thinking needs to be free and unencumbered and clear.  But how
can linguistic thinking be clear when the sentence is something like, say,
'We will need a tight fit,' when a fit can mean either how tight something
is or an episode in which someone lies convulsing and drooling at the
mouth on the floor?"

Once again the argument is spurious because it tacitly assumes that all
language is only the linear surface representation.  As we have seen,
understanding or parsing is the process of determining some correct
meaning and some correct regent for each word.  Thus although the words of
spoken and written texts are ambiguous before parsing, once they have been
parsed this ambiguity is removed.  Thus there can be no ambiguity in
thoughts set up as Panlingua arrays.  So thinking is apparently a
linguistic process, and may not even be possible outside the linguistic
apparatus, but thinking does not require spoken or written words.

Besides thinking in internal words, is it not true that people also think
in terms of images and abstract logical propositions?

It is easier to answer the second part of this question than the first.
"Just show me an abstract logical proposition without using symbols or
words," will be my answer.

But the matter of images is much harder to explain.  We may claim that we
think in images, but what is an image?  Do we really see things
photographically in our minds?  Of course not.  We imagine things in our
minds by digging up what we know about them, and this is why the images
generated by our minds are not picture perfect.

Many mental images can be generated from a good ontology.  As an example
consider the image of an eagle.  The ontology may give us the information
that the overall color is brown.  It may yield the information that the
parts of an eagle are eyes, skin, beak, wings, feathers, legs, three-toed
feet, talons, etc., and using this information we can quickly cobble
together the image of an eagle in our minds.

But other images will be more like moving pictures, and for these it will
be necessary to draw upon panlingua arrays containing the kinds of
information one might expect to hear over the radio during the broadcast
of a ball game.  At one time people listened to ball games over their
radios, and this was almost as good as watching them on television because
the running commentaries of the sports announcers were so calculated as to
create moving-picture images of the ball games in our minds.

In fact there seems to be a one-on-one correspondence between Panlingua
arrays and the images that we "see" in our minds, and this is why, by the
process of generating spoken or written texts which can be heard or read
by other people from those internal Panlingua arrays, we are able to very
accurately transfer such images to other minds.  Apparently the image is
generated by a Panlingua array in the first place, and we know that
Panlingua arrays can be converted into strings of spoken or written words.
These words are then heard or read and parsed, thus reproducing the
original Panlingua arrays, or else very close approximations.  And once
these Panlingua arrays have been archived in the mind of the reader or
hearer, they are able to evoke the same internal images because the two
systems, both human, are the same.  This is just like recording video
footage on tape, then inserting that tape into another machine and playing
it on its screen.

So I would argue that until we have evidence to prove otherwise it is
safest to assume that even the images in our minds are linguistic in
character.  Of course we do not understand all the details of these
processes at this time, and there remain many questions to answer about
how, for example, we might be able to store the pattern of a familiar face
in an ontology, but this does not mean it is not being done.

What is it in the seemingly simple mind of a child that enables it to
develop into the complex mind of an adult?

From what we have observed, it would appear that the only thing necessary
to build an adult mind from a child mind is just more and more of the
same.  However it must be added that there are many facets of intelligence
that remain to be discovered and explored.  For example we have seen how
children apparently discard many thoughts that they generate on their own
while retaining the thoughts they have parsed from the speech of others.
Is there a slow process by which the child mind finds a way to evaluate
its own inferences well enough to trust them and retain them in its corpus
of Panlingua arrays?  And by what means are unused or inaccurate thoughts
being purged?

What is a "grammatical phrase?"

We may surprise even ourselves with this question.  How is it that we can
hear one phrase and find it "grammatically acceptable," then listen to a
very similar phrase and determine for no apparent reason that it is
decidedly not?

The answer, once again, lies in the Panlingua corpus that we accumulate
for each natural language over time.  If some phrase is "grammatically
correct," a normal adult human being will be able to find some precedent
for it in the Panlingua corpus he has accumulated during his lifetime.  If
it is NOT "grammatically correct," he will not.  He is not determining
grammatical correctness by any set of grammatical rules, but only by
examining the Panlingua corpus for precedents in his mind.

The function of the fireside tale and the role of traditional oral
literature are most clearly explained by this theory.  These are the
essential ingredients that make us human.  As the child lies back and
listens, his mind expands, and he "sees" incredible scenes out of the
remembered past of his people, both real and unreal, in lurid detail.
This is because there exists a one-on-one correspondence between the
images we see in our minds and the internal representations of language.
All that is required to transfer the same images to the mind of a listener
is therefore to output the Panlingua structures that trigger them in a
stream of words.  For these words, when correctly parsed, will trigger the
same images in the mind of the hearer.  The child learns every detail of
the life of his tribe--the technical names of all the objects of
importance in his life and the terms that describe the various states they
can assume.  But more importantly still, as he listens by the fireside, he
is building that internal corpus of parsed sentences which will forever
define his very character as a human being.

Much research is needed to establish the relationship of this theory to
dreams.  Nevertheless it can be said here that because there exists a
one-on-one correspondence between the images we see in our minds and
Panlingua structures, then dreams must result from the traversal and
manipulation of parsed sentences within our minds.  The master system
shuts down the processes of conscious thought and control and begins
processing and consolidating the load of fresh data that have assaulted
our senses while we were awake.  If we have just thought or heard about
monsters, then we are apt to dream monsters.  The precise nature of the
processes of information analysis and knowledge consolidation that go on
during sleep are, of course, unknown at this time.  But it is clear that
they happen, because upon waking we often discover that we have
mysteriously solved some logical problem.  We are able to understand
things that were once unclear, and our minds are different than they were
the night before.  This is because new linkages have somehow been
established in our personal ontologies.  It is also probable that new
Panlingua structures have been generated using information from the old
corpus.

This theory also explains a lot about humor.  It is apparently true that
traversing linkages in our minds gives us pleasure, especially if those
linkages turn out to be new or unexpected.  But this is only natural if as
a species we are programmed to consolidate information and learn.  (At
this point I shall leave those who doggedly maintain that "No, we are not
programmed but have happened by accident," to fend for themselves and come
up with their own explanations).  It turns out that things that are
humorous tend to involve finding unexpected or un-looked-for connections
in our ontologies.  The pun is a perfect example, but a careful analysis
will show the same thing happening in many other kinds of humor if not
every other kind of humor altogether.

Why do many people enjoy talking to themselves or reading out loud when
they are alone?  Thought is apparently generated by the traversal and
activation of neurological pathways, therefore thought should give us
pleasure, especially when it results from the activation of unexpected
connections.  And since the activation of neural pathways gives us
pleasure, then actually speaking the words we are thinking should give us
even more satisfaction and pleasure.

Poetry is a manipulation of various components of the human linguistic
apparatus, some of which are covered by this theory and others of which
are not--at least not directly.  Poetry is intentionally designed to
remain ambiguous so that our minds can leap from pleasure to pleasure by
examining and re-examining its lines for fresh meaning.  Its cadence,
evidently a phonological component, is almost certainly related to that of
music.

Another great question is, "What is the faculty of music, and why do only
humans seem to possess it?"  Although this theory was not designed to
answer this question directly, I think it may be able to shed some light
on the matter.  Like language, the fully functioning faculty for music,
wherever it is found, seems only to be found among humans.  It is natural,
therefore, to at least toy with the hypothesis that music and language
must be somehow related.  But what could be the connection?  First of all,
as I have already pointed out above, finding pathways along linkages
inside our heads gives us pleasure.  It is as if we have a built-in need
for the stimulation and activation of pathways and neurological structures
inside our heads.  Two elements directly related to music play a critical
role in language, and these are (1) rhythm and (2) pitch.  The rate of
human speech is governed in almost metronome-like fashion by some rate of
syllables per minute or maybe syllables per second.  Were it not so, then
human speech would either come out too fast or else two slow, or else
(worse yet) at some wild and unpredictably changing rate, and human speech
would be impossible to understand.  There seems to be some kind of
metronome inside our heads which matches itself roughly to the cadences of
the words we hear spoken around us and determines the rate at which we
express the syllables of our own spoken words.  So I would conjecture that
we enjoy rhythm because it stimulates this process of matching our inner
metronome to an external cadence, but one that is different from that of
speech, which bores us as a result of overstimulation.  The other key
element of music, the variation and subtle manipulation of pitch,
activates that part of the phonological apparatus designed to pick up
intonation, which is probably a critical element in all human languages.
I will not attempt to elaborate further because an explanation of the
human phonological apparatus lies outside the scope of this theory.
Nevertheless I would point out that what we have learned by an examination
of the areas covered above can help to explain the workings of musical
phenomena.

Since writing the preceding paragraph, I have learned that researchers
have determined that music influences animal behavior from as far back as
Ibn Al Haytham (born 965ad).  Scientists recently became aware of dancing
birds from youtube and went to see an actual bird named Snowball, whom
they determined to be dancing to the beat of a song because Snowball
altered his dance rhythm when they altered the rhythm of the song.  They
found that various other animals were also capable of dancing in time to
music.  More study is needed to determine precisely what kinds of animals
can do this and what kinds cannot.

And what about religion?  What animal other than the human species has
religion?  Then is not religion related to the linguistic phenomenon as
well?  Surely.  In fact there are so many aspects of religion that are
touched by this theory that it would be impossible to explain them all
here.  But let us look at a few common cases.  A young girl gets into bad
society, starts using drugs, has no money to pay for them, becomes a
prostitute, and begins the slippery journey to Hell.  Somebody tells her
about Jesus and puts a pamphlet in her hand.  She is bored, so she reads
it.  As soon as she begins parsing the words, her heart is filled with a
freshness and light that is in total contrast to the steady diet of words
about lust and hopelessness upon which her mind has been feeding.  She
goes to church to hear more, and the result is very positive.  Somebody
gives her a Bible, and tells her to read it.  She does, and at first it is
hard to understand the King James, but slowly yet surely she begins to
succeed in parsing the words.  Being able to do this will already give her
pleasure because it means establishing and activating new linkages in her
linguistic apparatus, so she reads more and more.  Then steadily but
surely these sentences she is parsing are incorporated into the very
fabric of her being, and she becomes a changed person.  The life she has
been living as a prostitute is now completely out of keeping with what she
has become, and suddenly she is "reborn."  How did it happen?  It was a
miracle, and she knows it, and she knows it happened by reading the
"Word," so this is what she then does for the rest of her life, and as
long as she keeps doing it, it is impossible for the powers of darkness to
get back into her life ever again.  In other words, she is "saved."  To
the person familiar with this theory it is obvious that what she is doing
is building and maintaining a strong, new corpus of parsed sentences based
on the Bible, and this has changed her life.

Another thing that happens with religion is that the ontology gets
changed.  In the mind of the converted thief, for example, the hypernym
for stealing is shifted from "something good to do" to "something I should
never do."  So a reworking of the ontology is also definitely a part of
rebirth.

The phenomenon of "speaking in tongues" is harder to explain, but it
apparently involves jumping right out of the groove of the corpus to
experience some kind of altered reality impossible as long as we are
tracking the corpus.  It is some kind of temporary capitulation of the
corpus which IS ourselves.  But what it is beyond this, and whence the
accompanying charge of ecstasy I cannot tell.  In ordinary prayer, the
Christian is formulating words, which means generating and expressing
thoughts using the corpus, but when someone shifts to speaking in tongues
this stops happening, and there is a release from the "tyranny" of the
corpus into an altered state of mind.

Can a machine become more human than a human being?

If our humanity and existence as a human being depend upon our memories
and the size of the personal corpus of parsed sentences we are able to
maintain, then the answer would seem to be a definite YES.  As people get
older, their memories fail them, and it would seem to be true that instead
of growing, the internal corpus actually shrinks.  People forget things
that happened in the distant past, and also even in the recent past, and
they may also lose the ability to process what information they have left
using various human algorithms.  In this way they change and become
something less than the human beings they once were.  In fact they
actually become different people.  It should be possible to build machines
that will not do this but instead will continue to grow over hundreds or
thousands or even millions of years without losing any of their original
identity.  According to the insights provided by this theory, therefore,
such machines may actually become more human than the flesh and blood
human beings of today.

Section 15.  How To Build The Human Machine.

Based on the above information I will now define thinking as "traversing
the corpus and ontology in order to generate thoughts."  The thoughts thus
generated may be brand new thoughts that no one has ever thought of
before, as the thoughts that spill from the pen of an original writer, or
they may be thoughts that have been generated before, as when one parses a
familiar sentence.

And I will say that where dependency grammar is concerned a single thought
can have only one top word, be it the top verb of a sentence or a single
exclamation.

I have already shown by my own work that sentences can be parsed
(understood by machines) using the ontology and the corpus, but the
details of the particular process I am using remain secret for proprietary
reasons.  As you may already have gathered, I consider the building of the
parsed corpus to be the very cornerstone and foundation of our humanity.
It is when we begin to parse sentences that we begin to understand and to
remember.  It is by parsing these sentences I am writing now that you
understand me.  The infant mind learns to parse sentences, at first one
word at a time, then two, then three.  And each time he parses a new
sentence, he adds to his internal corpus of parsed sentences.  And using
this corpus he is building, he is able to parse other new sentences, and
so on until he becomes a mature speaker of the language and an adult
member of society.

So I have shown that a crucial component of the phenomenon we call being
human is produced by traversing the corpus and ontology, and I have
speculated that many other aspects of being human are produced in the same
way, and given strong evidence for my speculations.

At this point in time, I feel that I have learned a little bit about what
it means to be human, but there remain many mysteries I do not understand.
Still, let me list what I think I know:

1.  Understanding or parsing.  This faculty is at the heart of all truly
human intelligence.  I have done it with computers using the corpus and
ontology.

2.  Translation.  This is having a semantic structure that has not been
put into words in some target language and being able to transform it into
an appropriate structure to do so.  A corpus developed for the target
language is traversed in order to find appropriate precedents that have
been used to express the same or similar thoughts.  I have successfully
modeled this process on computers.

3.  The spontaneous generation of thought.  The human mind is constantly
traversing its internal corpus of parsed sentences in order to synthesize
thoughts.  These thoughts may be questions to ask or new ideas to express,
but they may also just be endless repetitions of old thoughts like some
nursery rhymes that keep on going round and round inside our heads.  This
process has never yet been modeled on computers.

4.  The deliberate study of new knowledge in order to have it available
for future reference, the ability to incorporate this new knowledge into
an existing corpus and ontology, and the ability to implement it and to
augment it by traversing it along with all the thoughts that were already
in the original corpus.  This has also never yet been modeled on machines.

5. (And most important of all) the ability of an automated system to
program and reprogram its own internal algorithms based upon the contents
of its corpus and ontology.  As far as I can tell, the man or woman who
builds a machine with this faculty will be the first human in history to
create true artificial intelligence, and once this has happened, it will
only be a matter of time until machines become more intelligent than
humans.

during this beginning phase of artificial intelligence design, all such
human faculties will come only at the cost of massive investments in
coding, debugging, and experimentation.  But once it becomes possible to
build machines capable of auto-programming, these efforts will be taken
over by the automated systems themselves, just as in the case of real
human
beings.  Then we may begin to see the emergence of other algorithms that
we
have never even dared imagine, such as those that have governed the minds
of individuals like Einstein, Newton, Shakespeare and others.  And this
time, instead of being left with only the dead remains of a human brain in
a jar of alcohol, we will be able to access copies of the actual code
itself.

So the way to build human machines is to equip them with corpora,
ontologies, and the algorithms to traverse them as i have described.

Section 16.  What Is Personality?

Suppose that we were to build a machine having the bare minimum of human
cognitive ability.  Then the following interaction might take place
between user and machine:

User: Dave put the car in the garage.

User: Did dave put the car in the garage?

Machine: Yes.

User: Where did Dave put the car?

Machine: In the garage.

User: Where is the car?

Machine: I don't know.

Although this answer ("I don't know") is quite correct, it is not the
answer that a human being would expect.  What a real human machine ought
to have answered would be something like:

I am not sure, but I know Dave put the car in the garage.

The stupid machine can be made to answer that the car is in the garage,
but only by the following kind of interaction:

User: The car is in the garage.

User: Where is the car?

Machine: In the garage.

So what if we then did some more work on the code and made the machine
capable of coming up with the right answer the first time round--would
this make the machine human?  No, but it would bring the machine a little
bit closer to being human.

This is only one tiny example of the many kinds of algorithms that must be
present in a real human being.  Patients with various kinds of brain
damage have been observed to lose such capabilities, so we know that they
must constitute small parts of the human mind.

The implication is that were we to continue adding tiny algorithms that
are an essential part of being human to the basic ontology-corpus machine,
this machine would ultimately be able to think just like a human being.

It appears that we are born with the basic data structures of language
(the ontology and corpus) built into our heads and primed to devour
information from the world.  We are also born with the basic algorithms
necessary to make this possible.  But the human mind has yet one other
faculty that we have never understood:  It can program itself, or
automatically "write" its own algorithms.  And this ability would seem to
be the final barrier separating us from the Holy Grail of true artificial
intelligence.

If we could but learn to build machines capable of programming themselves,
then all that would be necessary to build real AI would be just to get all
of the basics right, provide a minimum of code to jump-start the system,
and let it go.  Until then we shall have to content ourselves with
manually building the human machine one tiny algorithm at a time.

Before leaving this matter of self programming, let us consider a little
further just how it is done.  Take as an example the aspiring classical
guitarist of age 9.  He carefully learns exactly where to place his
fingers on the keyboard and how to hold them.  He also learns precisely
how he must move them, how much pressure to apply, and at what time.  In
fact the whole process is a very complex one, and he can hardly move a
finger without making some mistake.  But if we chance to meet him again at
40, we may find him sitting in his house over a cup of coffee with some
friends.  They are discussing the finer points of philosophy and his mind
is focused hard, but all the time his hands are going through the complex
movements of the "Themas de Farucca" on the strings of a guitar."  So the
process is careful thought, the determination to be able to do something,
and practise.  Using just these inputs, the human mind can somehow take
over and program itself into the full automatization of the task.

It is also important to note that all such algorithms (for example the
algorithms that handle the driving of cars), no matter how autonomous they
may become, in fact remain under the ultimate control of the human
linguistic apparatus.  This can easily be seen by the fact that they are
prone to malfunction under linguistic overload, which is why various
states have enacted laws against driving while talking on the telephone.
And the operation of these autonomous algorithms may continue to involve
the traversal of the corpus or the ontology or both.  We can infer this
latter supposition because the generation of these algorithms was
triggered by linguistic information in the first place and all of their
functionality can immediately be restated in terms of words and phrases
on demand.  For example, "I'm slowing down, I'm hitting the breaks, I'm
turning into the driveway," etc.  I have written down these observations
because it is imperative that we understand just how these algorithms can
be spontaneously generated if we are ever to build true AI.  I cannot tell
you how it is done, but these observations I have made may provide some
beginning of insight into the matter.

But no matter how we add new algorithms, whether automatically or
laboriously programmed line by line, as we proceed in this endeavor, a
strange phenomenon will emerge, and this is that thing known as
"personality."  We know that individual animals have different
personalities.  Extraordinary features of body and mind lend an organism
its personality.  For example, a wooden leg might lend character to the
personality of an old man named "Pegleg," and sweet lips may constitute an
important part of the personality of some young girl.  But even more
important to personality are the differences that we observe from one mind
to another.  If a human being were to behave as the machine I have
described above, and not be able to answer where the car was when he/she
knew that Dave had put it in the garage, we would say that this person had
a psychological quirk or defect, and this would become recognized as a
feature of his/her personality, just like sweet lips or a peg leg.

So as some kind of personality emerges in our human machine, at first it
will be barely perceptible.  But as we add the algorithms required for
this human trait or that one-at-a-time, the personality of the human
machine will become ever more apparent, and will change and shift before
us as we add this feature and that.

Thus at a psychological level what we call personality is the totality of
algorithms that have been built upon the bare-bones human substrate--what
they are and how they function--and the data that has been amassed in the
ontology and corpus of the human mind.

Section 17.  Conclusions.

The most important conclusion that can be drawn from this theory is that
all human intelligence revolves around an ontology and a corpus of parsed
sentences or Panlingua arrays.  To be a human being is to be an automated
system possessing such an ontology and  corpus and able to access and
manipulate that ontology and corpus in various human ways.  This ontology
and corpus and their interaction with their physical host are what we call
the human heart and soul, and the essential characteristics of this
ontology and corpus are what we call the human spirit.  All of the most
human aspects of intelligence and psychology are firmly based in the
totality of the human linguistic apparatus including the algorithms that
traverse the ontology and corpus.

To have said this clearly at last is something new, but in fact these
things are things many people have always known.  As the great Solomon
once pointed out, "As a man thinketh in his heart, so is he."  That idea
was written down about 3,000 years ago.  In modern language we often hear
people say things like: "You are what you think."  "For better or for
worse, you ARE the sum total of your memories and nothing less or more."
etc., all of which allude directly to the parsed corpus and the ontology I
have just described.

And if we truly understand these things, then the path to individual and
social improvement will be along lines determined by this knowledge.  If
we wish to take charge of our lives, we will start paying close attention
to what kinds of words and ideas we are parsing, because we will realize
that whatever we parse becomes part of ourselves.

Where computer science is concerned, the most important implication of
this theory is that it is probably 100% possible to build a human
computer.  Add the parsed corpus and the means to use it to an animal and
what you get is a human animal, or a human being if you please.  In the
same way, therefore, the following must be true: Add the parsed corpus and
the ability to use it to a machine and you will have a human machine.

What I have described may or may not fit the bill for a "universal
grammar," because it is not a set of grammatical rules.  Nor is it a body
of grammatical rules parts of which can be activated or deactivated by
throwing sets of switches as Chomsky imagined.  What it represents is a
universal language machine built into the brain of every normal human
being, a machine that has been carefully designed to learn and use
languages, and out of which all of our most fundamentally human
characteristics will emerge.

Because of my many years of interaction with computers, upon which I have
at one time or another successfully modeled almost everything I have
described (with the exception of the unproven speculations I have made in
Section 14 of this paper and the algorithms I have proposed in Sections 15
and 16), my approach to language is hard and rigorous and highly
intolerant of conjectures put forward without working models running on
automated systems or other strong proofs.  Others write on and on about
unfounded or poorly founded suppositions.  I write about what I have
learned and what I have been able to prove.  Others write in fuzzy
generalizations.  I write in terms of independently verifiable details.

In these pages I have clearly shown what the fundamental building blocks
of language are--the linguistic link and the word--and described them in
exhaustive detail.  I have shown how they are used to build all of the
internal data structures of language, and how these linguistic structures
are the underpinnings of all human language and thought--how Panlingua
exists in all coherent human utterances and how it serves as the bedrock
upon which all human languages are formed.

I have shown how all of the internal data structures of language can be
modeled upon computers because all of them can be and apparently ARE
modeled from nothing but simple links and nodes--both of which are an
everyday part of computer software design.

And finally, I have shown  how the theory I have presented can answer many
questions that have seemed intractable to psychologists and linguists in
the recent past.

I have written humbly and honestly and as clearly as I can, and I have not
attempted to fatten these pages by dwelling upon any one topic longer than
necessary in order to get my message across.  My purpose was to move
forward quickly in order to keep the reader's full attention.  It should
not therefore be assumed that I have here even begun to explore all of the
implications and ramifications of the theory I have described.  My hope is
that I shall have provided just enough detail to make it possible for my
readers to understand the (in my opinion) suppressed truth about language
without bothering them with unnecessary details.

Human intelligence is linguistic, therefore the path to better human
intelligence, be it for men or machines, passes straight through a clearer
understanding of the human linguistic apparatus, because what we THINK and
say, and HOW we think and say it, are what we ARE.