2009 Note: When I wrote the following paper, I had not yet discovered my final definition for "word," as given by the following theorem:

THEOREM: Every word of every coherent phrase ever expressed by man is simply a semlink and a lexlink emanating from the same node.

If I had known this definition for "word" at the time (1997), I might have done a better job of understanding the true meaning of Shaumyan's "functional and lexical characterization" for each word, which fits so neatly into the final theory.  In fact without realizing it, by specifying a "dual characterization" for each word, Shaumyan had hit exactly upon the roles words assume in the corpus versus in the lexicon/ontology.

Some other parts of this document are also out of keeping with the final form of my theory, for example the idea of an ontology containing much more information than just hypernymy, holonymy, synonymy, and antonymy.  Also the naming of link type and the final definition of the links between words and their meanings.  In its final form my theory uses the following definitions and terminologies which differ from those found within this paper:

1. The links between semantic nodes are called "radical links," or radlinks.

2. The links that run from semantic nodes to words in the lexicon are lexlinks.

3. The links that run from words in a phrase or sentence to their semantic nodes are semantic links, or semlinks.

4. Syntactic links (synlinks) are the same as in this paper.

Original paper follows:

Language: A New Integrated Working Hypothesis.

By Chaumont Devin,
Honolulu, March 23, 1997.

 1. Foreword.

In January and February of last year I was a member of the Internet mailing
list, dg-list@UGA.CC.UGA.EDU (Michael Covington, University of Georgia).
During that time an inspired discussion developed between Dr. Sebastian
Shaumyan, Dr. Richard Hudson, and Mr.  Jean-Loup Komarower--all living on
different continents and persuing different lives.  Other people were also
involved in this dialogue, but the most important insights came from these
three men, whose words I will employ in the following discussion to develop a
new working hypothesis for linguistics.  In some cases I have applied their
words in the exact way they were intended.  In other cases, I have attempted
to abstract from them what I perceive to be subliminal inferences which they
did not consciously intend.  In every case I have given the time it was in
Honolulu when their messages were received.  I trust this paper will be a
fitting tribute to the scientific potential of the Internet, without which its
very existence would have been impossible.

2. Linguistical Planes.

Sebastian Shaumyan <shaumyan@MINERVA.CIS.YALE.EDU>, Wed Jan 17 19:28:23 1996:
"... we distinguish two planes of language: 1) the PHONIC PLANE and 2) the
SEMANTIC PLANE."

Richard Hudson <r.hudson@linguistics.ucl.ac.uk>, date unknown: "The text we
see and hear in the world around us has at least three structures:

- a phonological (or graphological) structure, involving in vowels,
consonants, etc etc

- a syntactic structure, involving words

- a semantic structure, involving the meanings of words."

Imagine two parallel planes in three dimensional space, each plane containing
thousands of points.  Imagine further that each point in either plane can be
linked by a binary relation to any other point in either plane.  Now call one
of these planes the "syntactic" plane, and the other the "semantic" plane.

Notice that Shaumyan goes to the heart of the matter by singling out two
planes, while Hudson reminds us that there exist other associated planes.
Hudson then goes on to separate out the "phonological (or graphological)
structure" from the syntactic plane.  The scope of this paper will be confined
to the syntactic and the semantic planes, recognizing that they are bounded
"above" by the phonic plane and "below" by an as yet undefined plane involving
the inner workings of the mind.

Richard Hudson, Mon Jan 15 23:43:08 1996: "... I would argue very strongly
that it's essential to recognise a semantic structure, and to have a clear
idea about what it looks like. I would also argue that the semantic structure
is different from the (surface) syntactic structure, ..."

In order to further simplify our three-dimensional model, let us now orient
our two planes horizontally with the semantic plane below and the syntactic
plane on top.  In this way, our model will lend itself nicely to references to
a "surface" syntactic structure and a "deep" semantic structure.  Let us then
call all points in the syntactic plane "words," and all points in the semantic
plane "semantic nodes," or "semnods."

3. Linguistical Links.

Jean-Loup Komarower <jlk@SATURN.CSSE.SWIN.EDU.AU, Sun Feb 11 14:37:23 1996:
"In set theory the 'element' is the building block, and the relation simply an
ordered n'tuple of elements. Here we take an opposite point of view, and the
association is the building block."

Now let us do a very unusual thing.  Instead of ascribing significance to the
points in our two planes, let us instead assume that the points, be they words
or semnods, are utterly meaningless in and of themselves. They function merely
as connecting points for links--in other words for binary relations between
nodes.  It is in these links between nodes, then, that meaning resides.

Exactly three kinds of binary relation or link are possible in our
three-dimensional model.  These are:

Syntactic links, or links between words and other words.

Lexical links, or links between words and semnods.

Semantic links, or links between semnods and other semnods.

Syntactic links lie entirely within the syntactic plane, and semantic links
lie entirely within the semantic plane; but lexical links bridge the gap
between the two planes, and lie outside both planes.  Syntactic links
establish such things as word dependencies, case role and part of speech.
Semantic links reflect such things as hypernymy, holonymy, synonymy, antonymy,
etc.  Lexical links determine aspect, intensity, and other factors related to
state of being.

Sebastian Shaumyan, Wed Jan 17 19:28:23 1996: "Syntax is rooted in semantics
and cannot be separated from semantics. Syntax and semantics constitute a
unity, they constitute a whole."

Sebastian Shaumyan, Sun Jan 21 17:21:13 1996: "I wonder what it means to say
that syntactic structure involves words, but semantic structure involves the
meaning of words. Does it mean that syntactic structure has no meaning? A word
without any kind of meaning is no word at all; it is nothing more than a
sequence of sounds."

Sebastian Shaumyan, Wed Jan 31 23:34:15 1996: "words are only blocks the
sentence is assembled from. The sentence is a whole, and words serve as its
parts; words a sentence is assembled from are subordinated to a sentence as
parts are subordinated to a whole."

A great deal of confusion has arisen over the seemingly artificial distinction
that has been drawn between syntax and semantics.  Assuming our definitions of
syntax and semantics approximate those held by students of language at large,
which (judging by our quotations from Hudson and Shaumyan, above, they appear
to do), it is clear that syntax and semantics are, indeed, just "two sides of
the same coin," and cannot be separated.
Another oft-repeated source of linguistical consternation is the fact that
isolated words, without being related to other words, seem to have no meaning
at all.  And yet, in some strange way, they do seem to have meaning after all.
This phenomenon is nicely explained by our model, in which words are
meaningless points within the syntactic plane, gaining meaning only by virtue
of their links to other nodes in the model.  The seeming "meaninglessness" of
"isolated" words stems from the fact that they indeed ARE linked to other
words, but only by indirect links through semnods.  They lack the strength of
meaning characteristic of words occupying appropriate positions in phrases,
where their meaning is fully determined by direct links to other words as
well.

4. The Meaning of Meaning.

Sebastian Shaumyan, Sun Jan 21 17:21:13 1996: "We must assume two kinds of
meaning: 1) grammatical (or structural) meaning and 2) non-grammatical (or
lexical) meaning."

Sebastian Shaumyan, Wed Jan 31 23:34:15 1996: "No adequate theory of grammar
is possible without a rigorous distinction between grammatical and lexical
meanings of words and word complexes."

Sebastian Shaumyan, Wed Jan 31 23:34:15 1996: "Every autonomous word has a
grammatical and a lexical meaning. So every word has a dual characterization:
a functional characterization and a lexical characterization."

In our three-dimensional model, words are nothing but meaningless points to
which links are anchored.  Meaning is determined by binary relationships
(links) between the points (nodes) of our model.  When we speak of the
"meaning of a word," therefore, we are speaking of the meaning residing in the
links (binary relations) that terminate at the point we call a "word."

***** The meaning of any word in any language is fully defined by its links to
other words (Shaumyan's grammatical meaning or functional characterization),
and its link to a semnod (Shaumyan's lexical meaning or lexical
characterization).

It should be noted that only one link exists between a word and its semnod.
Within the syntactic plane, on the other hand, each word (unless it is a verb
of the first rank) is linked by a single dependency link to a regent node, and
by 0, 1, or more dependency links to dependent nodes.  And although dependency
links are directional, both regent and dependent links must be evaluated in
order to establish the full meaning of a word.

The identity of the semnod to which a word is linked is critical to its
meaning, not because the semnod carries any meaning in or of itself, but
because of the significance of the links from said semnod to other words and
yet other semnods, etc.

Note also that by the definitions for "word" and "meaning" developed above,
there may be certain purely functional "words" which do not qualify as real
"words," but which, in conjunction with other words, make up an atom in the
universal subsurface language.

5. Syntactic Links.

Sebastian Shaumyan, Fri Feb  2 00:16:15 1996: "... distinction between classes
of words is based not on the nature of objects to which words refer but upon
the mode of their presentation."
***** The case role and part of speech classifications of words and groups of
words are determined purely by syntactic links.  In other words, these are
fully determined within the syntactic plane.

6. Lexical Links.

As already stated, lexical links link a word to a semnod.  All such links
associate a word with a particular state of being.  The type (sometimes called
"aspect") of this state of being is determined by the type of the lexical
link.  Lexical link types are mutually exclusive.  That is, no lexical link
can be of more than one type.  So far the following lexical link types have
been tentatively defined:

Type            Example
genetive       Hawaiian
processable    breakable
unprocessed    unbroken
in process     breaking
repeating      breaks (repeatedly)
processed      broken
consisting of  gaseous
similar to     stoney
abounding in   rusty
action         breakage, breaking
doer           breaker
doee           breakee

These lexical link types can be modified, or "colored," by the following
classifications, which are not necessarily mutually exclusive:

comparative
superlative
diminutive
intensive
property
negative

7. Semantic Links.

Jean-Loup Komarower, Wed Jan 24 20:30:34 1996: "Syntactic analysis restricts
itself to word distinctions and features (rules) in the syntactic layers in
order to determine word position and form. A dependency syntactic analysis
will happily generate links for sentences like "Colourless green ideas sleep
furiously.". We then have to resort to features stored in the semantic layers
to determine that the links in such sentences are in fact not permissible."

"So, what are the fundamental differences between the constraints that apply
to the syntactic layers and those that apply to the semantic layers? Do those
differences justify a separate processing and separate fields of study?"

Jean-Loup Komarower, Sun Feb 11 14:37:23 1996: "Humans notice regularities in
their perceptions and connect groups of these regularities into the conceptual
building blocks of their internal model of the world. I believe that words,
the building blocks of language, reflect some of these groupings (and in fact
guide the formation of these concepts during our formative years).

"The grouping of perception regularities is not done randomly, but follows
some principles of economy that make models more powerful: Conceptual entities
are constructed and perfected culturally through the use of language until
they acquire good modular qualities -high internal cohesion, stability through
time, and low coupling between concepts. Low coupling is achieved when there
are few, clearly defined types of associations between entity types, so that
any event which affects one entity will affect only few other entities through
these associations."

Richard Hudson, Tue Feb 13 07:21:55 1996: "An alternative is to say that verbs
are also markers for entity types, but the entities tend to be more abstract
than those marked by verbs.

Please correct me if I am wrong, but I believe Hudson must have intended: "An
alternative is to say that verbs are also markers for entity types, but the
entities tend to be more abstract than those marked by nouns.

We call the thing Komarower has so eloquently defined an "ontology."  It is
essentially the links and nodes of the semantic plane.  But Komarower's
description raises a problem in Hudson's mind.  It seems to imply an ontology
consisting only of nouns.  Then what are verbs?  Shouldn't verbs be considered
conceptual entities as well? His reservations appear to be entirely justified.

In our model, the nodes of the semantic plain are not marked as nouns, verbs,
or anything else.  They are simply the connecting points to which links are
attached.  They may be thought of as "conceptual entities," but only with the
understanding that their real significances lie only in the links to which
they are attached.
Thus it is possible to know, for example, that "red," "redness," "reddish,"
"redden," etc., are all linked to the same semnod, but it is never possible to
know what that node really is, or what part of speech it should have, etc.  It
only exists in that it links these and a host of other words to one another in
various ways.

And since semantic nodes are nothing but connecting points, there is not
really much about them to understand.  The same is far from true, however, of
semantic links.  These include such relations as antonymy, in which one node
is related to another as opposite. In such a case, all the states of being
associated with one node are also the opposites of the corresponding states of
being associated with the antonym node.  Recall that these states of being are
encoded as lexical link types.
Another important relation is that of hypernymy, also known as the "is a"
relation.  For example:

a male (thing) is a sexual (thing) is a living (thing) is a physical (thing)
is a tangible (thing) is a visible (thing) is a material (thing) is a thing.

Note that although an expression like the above implies conceptual entities
for male (thing), sexual (thing), living (thing), physical (thing), tangible
(thing), visible (thing), material (thing), and thing, in fact no such
conceptual entities actually exist.  All we are really saying is that there
exist semnods (points in the semantic plane) to which the words, male, sexual,
living, physical, tangible, visible, etc., are linked by lexical links.

In fact the nodes and links of the semantic plane form a lattice--a hierarchy
in which each node can have multiple parents.  This can be seen, for example,
in relations like those of the node linked to the word, "crocodile":

a crocodile (thing) is a reptile (thing)
a crocodile (thing) is a dangerous (thing)

Through multiple parent links, an ontology (collection of semantic links and
nodes) can thus determine that a crocodile is a dangerous reptile.

8. The Ontology.

Jean-Loup Komarower, Sun Jan 28 12:18:36 1996: "It seems that the coherence of
a sentence depends upon the existence of a semantic link between the pairs of
words in every predicate/argument binary association -verb/object,
verb/subject, verb/adverb, noun/adjective...- in a dependency analysis of the
sentence. I think that the necessity for these semantic links maps cleanly,
into the collocation constraints of sublanguage theory (correct me Stephen if
I am mistaken).  For example, the object of "eat" must be edible, and a
"colour" adjective must modify nouns of entities with a surface (ignoring
metaphor, metonymy and other tropes for a first approximation).

"By 'coherence' of a sentence, I mean the quality the lack of which makes a
sentence like "Colourless green ideas sleep furiously" appear incoherent to
English speakers.

"This existential constraint on binary semantic relations between
predicate/argument seems to be very close in nature to the syntactic
constraints on syntactic dependency relations. No inference is involved.  They
are collocation constraints between pairs of words, dependent only on word
type.

"In natural language understanding however, these binary coherence
associations are not considered sufficient for some types of applications. A
sentence like "My buxom neighbour is a father of two." is coherent (all
predicate-argument pairs are accompanied by permissible semantic binary
associations). It is however quite incongruous.

...

"It seems to me therefore that there are three grades of quality of a
sentence: it can be syntactically correct, it can be coherent, and it can be
congruous, in that order. A syntactically incorrect sentence cannot be
coherent or congruous, a syntactically correct but incoherent sentence cannot
be congruous.  It may well be this ordering of sentence quality which explains
the perception that many have of the precedence of syntax."

In fact both coherence and congruity can be readily checked using the semantic
plane of our model.  As a free demonstration let us answer that central
question of 20th-Century linguistics:

Can colorless green ideas sleep furiously?

It should be noted that ontologies (the way the links and nodes of the
semantic plane are configured) may differ from language to language, and even
from individual to individual.  For example, in some Southeast Asian
languages, green things and blue things employ the same semnod. Thus for a
native speaker of such a language learning English,he/she may be crossing the
street when the light is "blue" for many years.  I imagine, therefore, that
for some people ideas may indeed be colorless, while for others colorless
things must be material.  So I will not attempt to ascertain whether or not
ideas may be colorless from our ontology.

But green ideas?  I should hope not!  A check of our ontology reveals the
following links:

an idea (thing) is an abstract (thing) is a thing
a material (thing) is a thing
a material (thing) has a potential patient relationship to green

To understand the significance of the "potential patient" relationship it may
be necessary to re-examine our understanding of verbs.  There are only two
kinds of verbs: stasis verbs and transition verbs.  Stasis verbs indicate the
maintainance of a state through time, and  transition verbs indicate
transition to a state through time.  Thus, for example, to "wait" is a stasis
verb, since it does not indicate a transition from unwaited to waited, whereas
to "open" is a transition verb because it indicates a change of state from
unopened to opened.
The important thing here is that in both cases some semnod posing as patient
uses another semnod as a state.  Thus for example in, "The girl waited," the
"girl" patient adopts "wait" as a state.  Similarly in "He opened the door,"
the "door" adopts "open" as a state.

Please note again that semnods in and of themselves have no meaning, and
cannot be either "wait" or "open."  When we speak of a semnod as being "girl,"
"wait," "door," or "open," all we are saying is that we are refering to
semnods connected through appropriate links to the words, "girl," "wait,"
"door," and "open."

What we want to know is whether an idea can be green (adopt green as a state
of being), and for this to be possible it would be necessary for "idea" (or
else something that "idea" is) to have a "potential patient" relationship
(link) to green.  Our ontology tells us that material things have a potential
patient relationship to green, but that an idea (thing) is not a material
(thing).  The hypernymy chain from idea (thing) to thing bypasses material
(thing) completely.  We need go no further--it is clear that ideas CANNOT be
green.

But what about sleeping furiously?  Can sleeping things be furious?

a sleeping (thing) is an animal (thing) is a ... is a thing
a conscious (thing) is an animal (thing) is a ... is a thing
a conscious (thing) has a potential patient relationship to furious

These ontological entries tell us that conscious things can be furious but
that sleeping things cannot.

Once and for all, colorless green ideas CANNOT sleep furiously!

This kind of word-pair analysis is what Komarower might call "coherence
checking."  Sentences with word pairs such as "green ideas" are said to be
incoherent.

But Komarower also speaks of another kind of compatibility which jumps across
words.  He calls this "congruity."

Can my buxom neighbour be the father of two?

If only coherence were involved, then obviously he/she could.  Neighbors can
be buxom, and fathers can be "of two."  However, a deeper analysis reveals
that:

a woman is a human is a ... is a thing
woman has a potential patient relationship to buxom
a father is a man is a human is a ... is a thing

The "is," or "to be," verb equates father with something buxom, but a father
is not a woman, and in all the ontology only "woman" has a potential patient
relationship to buxom.  Sorry:

My buxom neighbor CANNOT be the father of two.

Without going into further detail, these examples tell us that a correctly
formed ontology can provide accurate coherence and congruity checking using
limited computational resources.

9. Ramifications.
Have you ever wondered why, for sentences like, "Roses are red," the "are" is
necessary in languages like English, and redundant in other languages? Simple.
Other languages simply employ the link:

red, present tense verb + pre-existing state of  being.

The meaning is exactly the same.  The syntactic difference arises merely from
the fact that English, for whatever reason, has no "verb + pre-existing state"
link to the appropriate semnod.

There might be many other such questions this model could resolve, but its
most important applications probably lie in the field of computational
linguistics, in areas such as automatic parsing and machine translation where
great simplicity and precision are desired.  Coincidentally, it may also
provide insights into the workings of the human mind.

10. The Biological Model.

Do these observations have universal significance?  Does this hypothesis have
biological implications?  Are the various linguistic layers, two of which we
have examined here, reflected in the layers of the cerebral cortex of the
human brain?  A reasonable next step would seem to be to probe another layer
down in an attempt to see what the semnods of the semantic plane might link
to, and how.  Could such links be established to sensations and emotions?
Does the answer to the perplexing question, "Why are the nodes of the system
meaningless in and of themselves?" lie within some lower stratum? Only time
will tell.

November 22, 1998:

I am not an academically trained linguist, and have 0 published work in the
field.  I was brought up by missionary parents in Indonesia, where I learned
early to slip between various languages and various ways of looking at the
world.  My education was spotty.  In adult life I became a missionary in
Indonesia, Papua New Guinea, and the Solomon Islands.  Then (after losing a
boat in the Solomon Islands) I turned my hand to computer programming, and
became interested in artificial intelligence.  After a cursory examination of
Chomsky and various contacts with some of the foremost linguistic scholars of
the current time (Shaumyan, Starosta, Hovey, McCovington, Stampe, Hudson) I
determined that no one had yet come up with a theory of language that could do
anything on a computer, and so I created a theory of my own, and this is
Panlingua.

September 24, 1999:

And I am also aware that the beginnings of most of the things I have
discovered and refined were not first expressed by me, but result from the
synthesis of many inputs.  I have made many important discoveries on my
own, some of which I just enumerated in a recent posting on this
NewsGroup.  But the raw material I synthesized came from people like
Jean-Loup Komeror, Sebastian Shaumyan, Hovey, Hudson, and others.  All of
these people had part of the puzzle, but various limitations kept them
from ever putting it all together.  Perhaps I should say that I was lucky
to have been blind, because this forced my mind, over a process of years,
to develop powerful internal modeling capabilities that I would probably
never have gained had I been able to see sheets of paper.  The advantage
this seems to have given me has been the ability to see language as a
three-dimensional structure instead of always trying to work out
everything on a not-so-real two-dimensional plane.  The other place I got
lucky was because I was born an American, and so, as a blind person, I was
entitled to receive a government stipend to keep me alive.  This was like
grant money for which others have to worry and struggle.  Because of this
I was able to carry out my work virtually unimpeded by monitary
constraints.  All I needed was a computer and plenty of time.  Sometimes
my computers broke down, and cost me small fortunes (from my perspective)
to fix, but they all held out just long enough to let me squeek by.

January 1, 2003:

JLK wrote:

>it seems to me much more
>promising than WordNet or Cyc,

Yes, I was aware of both these systems (I got a copy of WordNet, and I
talked to Doug Lenat on the telephone).  I saw that Dr. Lenat had a few
problems.  Among other things: (1) He was clearly not as smart as he was
cracked up to be, (2) he had the academia sickness and so was unable to
take me seriously, and (3) he was not a man of his word.  The early CYC
system used frames, an option I quickly discarded.  They later changed
this, and I don't know how things came out in the end.  I was also in
contact with Dr. Hovey, who was developing an ontology at some university
in California.  All these systems seemed to have serious problems, and
their common denominator was firstly the fact that their creators lacked
maturity in computer programming and  Secondly that they had no adequate
theoretical framework into which they could neatly integrate their various
system components.  Nor, I felt, did they really know precisely what
components the systems they were building should have, which might also
have been determined by a comprehensive theoretical framework.  This
elusive theoretical framework then became my obsession, because I realized
that without it I would only be batting my head against a wall.

Before going on, I should say that I perceive WordNet as a very valuable
educational tool, and that educators should find more means such as this
for getting their ideas across--something that will never really work very
well, but which nevertheless showcases the general idea in detail.

The reason SEMLEX shines is that it was not developed in isolation, but as
part of a comprehensive greater system that would really have to work in
real time.  As an example, this thrashing out of the theoretical framework
by means not only of careful thought but also of relentless
experimentation resulted quickly in an ontology that was not broken up by
part-of-speech as in the case of WordNet, but was able to employ a single
system (array or file) to hold information for ANY word in ANY language.

January 8, 2003:

JLK wrote:

>(Semantic) Networks are viewed as knowledge representation. No one before
>you thought they had any relationship with language universals.

Really?  To me they are a most integral and inalienable part of all
language.  Sometimes I try to figure out just how I came up with this idea
of the ontology.  You (JLK) sent me a large file by somebody or other that
seemed to cover the subject but which I failed to understand.  I think the
person who did the most to put me onto it was a South African linguist
named Dr. Hovey, who lives in California.  I talked to him for some time
over the telephone in order to get a handle on what he was doing, and this
is how the idea began to take real form.  Then you or someone else or both
recommended WordNet, and I managed to telnet in and download it.  It
proved of very great value as an educational tool,but I couldn't
immediately see what was wrong with it or all that I would have to do.  I
think I may take some time someday and try to get this chronology a little
bit straighter in my mind.

I remember Hovey saying this on the telephone: "We can't understand it yet
now, but someday somebody will come along and put it all together and make
it work."  These, of course, must not be his exact words, but their deep
meaning got stuck immediately in my mind, and seemed to foreshadow my own
destiny.  Unfortunately, whenever I have put anything together for them,
the very first thing any selfrespecting linguist wants to do is to tear me
apart and stomp the results into the earth!  In general linguists are, I
fear, a very violent people!

But now I really HAVE put it all together, perhaps beyond old Hovey's
wildest dreams, and I cannot help but feel excited for the future.  They
really did have this thing all in little bits and pieces with nobody who
could see the big picture, just like Hovey said.  But my mind positively
revels in the elegance of the theory I have discovered, which seems to
explain EVERYTHING without exception.  I have yet to hear of any
linguistic phenomenon this theory fails to touch on or explain.

>I have seen you at work, and I understand that the unability to read
>forces you to attempt to get people to spill out what they know using
>various techniques.

Yes, but see how far and how quickly I have traveled.  No one has ever
done this by reading linguistics papers, because there is simply too much
junk out there to read.  I don't think it was humanly possible to get
where I am by reading through the material, and this is precisely why
nobody ever got to where I am.  So this blindness was the determining
factor of my destiny.