[Top] [All Lists]

Re: [ontac-forum] Semantics and Ontology (former What should be in an u

To: Azamat <abdoul@xxxxxxxxxxxxxx>
Cc: semantic-web@xxxxxx, ONTAC-WG General Discussion <ontac-forum@xxxxxxxxxxxxxx>
From: Ken Ewell <mitioke@xxxxxxxxxxxx>
Date: Tue, 23 May 2006 16:52:17 -0400
Message-id: <44737601.9010405@xxxxxxxxxxxx>
AA>.... As an example, consider the class of relationship, which can expressed by as many names as ‘connection’, ‘association’, ‘link’, ‘reference’, ‘regard’, ‘tie’, ‘bond’; or indicated by as many verbs as ‘to relate’, ‘associate’, link’, ‘link up’, ‘connect’, ‘tie-in’, ‘colligate’, ‘refer’, pertain’, ‘concern’, ‘bear on’, etc. Or, take the class of events expressed by as many words as ‘happening’, ‘occurrence’, ‘occurrent’, ‘contingency’, ‘outcome’, ‘effect’, ‘issue’, ‘upshot’, ‘result’, etc....

I don't know what class of relationship that is. Those terms do not seem to make it clear to me.  
For what reason is reference and regards in the same group? Some reason may be at hand but it may not be agreeable or pertinent.

There is a certain problem that issues from matters of agreement.  We call it disagreement.  That is what motivates us all towards more reliable methods of reaching suitable agreements using more reliable distinctions and more pertinent knowledge.

In this regard, we have extensive research into many languages and terminologies over more than twenty years and we have developed and tested very rich semantics.  From the start we always considered how to validate our understandings of certain semantic features and characteristics, and moreover we were directly concerned in its usefulness in advanced information systems. 

We discovered that our understandings could not be validated by linguists. Most other kinds of researchers and scientists
in computational linguistics and artificial intelligence, had little time for something this far from their traditional focus.

We believe that each thought reaction (every interpretation of a pattern) can be seen as a (regular, repeating) relationship (emerging, obtaining) between the regular forces and objects in the field of perception (including psychophysiolgical impulses) and a corresponding (universal) system of abstract objects and forces in the mind of the interpreter. This correspondence would be observable as something like fidelity and relevance or pertinence combined.

The conjecture is that repetitive sounds and visual clues organize our world as it were.  We determined to discover how that organizational mechanism works.   In terms of language, over the objections of many linguists, we believe that the phonemes of any word are signs that refer to abstract objects that are somehow related to the properties of the object to which the word refers.

   word X refers to object A
   each phoneme P of word X refers to an abstract object BP
   abstract object BP is related to property T of object A

Let me just show you how the natural concept of two-sidedness in nature can be articulated with single distinctive and repetitive sounds/clues.  This can be demonstrated using English language personal pronouns (
I, you, we, us and he) that represent a universal abstract category we call assignment: 

Assignment can be seen as the abstract process necessary to distinguishing identity (i.e., A=A).
Polarity is here used to distinguish between the different persons in a sensory manner.

The bipole(p,n) polarizes the first side (first person, I) and inversely polarizes the second side (targets the second person, you (n,p)).

(p,p) places equal polarity on both sides (first and second person) and gives us the concept we and us.

Placing inverse polarity (no focus) on both sides (n,n) (neither first nor second) expresses the third person, he.

Thereby, incredible as it may sound, we found a logically complete way of distinguishing all things distinguished by such process-polarity pairs.  Next, we ask, is that understanding of abstraction valid, is it really universal,  and if so, is it useful? 

We believe that the human mind constantly interprets such abstract objects and that the resulting interpretations also can be abstract objects that may in turn be reinterpreted. Both the original abstract objects and their successive interpretations are related to the properties of the object to which the word refers.

  abstract object BP is interpreted as abstract object B’P
  abstract object B’P is related to property T’ of object A

In addition, we believe that the morphology of a word, its structure, is also a sign that refers to an abstract object structure that is somehow related to the structure of the object to which the word refers. The human mind also constantly interprets and reinterprets this abstract object structure.

  structure of word X refers to an abstract object structure S
  abstract object structure S is related to structural property TS of object A
  abstract object structure S is interpreted as abstract object structure S’
  abstract object structure S’ is related to structural property TS’ of object A

The repeated interpretation of the abstract objects to which the phonemes of a word refer, in light of the repeated interpretation of the abstract structure to which the morphology of that word refers, will establish more and more relationships in the human mind to the properties of the object to which that word refers. That is our story of cognitive growth by reinterpretation. 

We found that thousands of three-consonant word roots of Old Arabic are in fact structured signs that refer to triples of process-polarity pairs. Higher-order process control precedence rules dictate control structures within each triple, giving us root interpretation mappings. We find that these process-polarity pairs seem to organize the terms of modern languages as well as the ancient semitic languages where we first noticed them.

We found several rules that we implemented as analytical tools for computing these relationships by parsing them from texts or messages, or by measuring relations between a specific question and its specific answer; between a query or text.  In order to provide examples, let us examine one rule, we found that we will call:

  Inward at Interface.

  • If we have two bipoles of the same category  in a word stem, such as the case represented by the consonants m and d in mold, middle and model, their affinity causes them to combine into a dual-action tool with specific semantic relations.

There are four bipoles of this type that can be paired in six different ways (without repetition and without regard for sequence).  If we examine English terminology with a single pattern designated as (p,n)+(p,p) We have six types of dual-action rules or tools indicated by these letters in a stem structure (j+w,j+v,i+v,r+b,m+d,s+c).

Using the pattern, we have derived the taxonomy and analyzed below eighteen (18) distinct semantic classifications from all the English word stems of this type that we extracted from a spelling dictionary of more than thirty-thousand popular and often used words.  Terms of this type that are related to mental activity include "brain", "mind", "remind", "admonish", "meditate", "medulla", "dementia", "demented", and "cerebral".

Here is the rest of the analysis. As many examples as could be found for each of the abstract concepts are given. As you may notice many words are assigned to many  of the semantic classifications.  Where we found Arabic word roots representing a variation as the English terms do, they are included in variation titles (triple letters in parenthesis represent 3500 year old Arabic roots, capital letters represent Arabic consonants not available in the Roman alphabet):

1. Join and Repeat--Multiple, Multiply, Coherent Group (rbO, srb)

Breed, birth, branch, brigade, brother (multiple), scion (breed), tandem (join two), tremendous (a lot), modulus (multiples), democracy (group rule), demography, demagogue (group leader), comrade (of one's group), academy (assembly), decimal (multiple, 10), December (10), endemic (of certain group), algebra, rabies (repeats joining=biting).

2a. Reverse Connection--Discontinue, Break, Pieces (qTO, zbr)

Break, rubble, debris, secede, sect (break away), slice, modicum, dime (piece), abrupt, abrogate.

2b. Reverse Connection--Hurt, Cut, Scrape (grH, kST, sHg)

Damage, bruise, scar, scrape, abrade, abrasive, scratch, sick (hurt), scare (damage pending), scissors, incision, indemnity, malady.

2c. Reverse Connection--Banish

Damn, curse, condemn, malediction, demon.

2d. Reverse Connection--Destroy (hdm, dmr)

Demolish, demise, doom, decimate, succumb, murder, homicide, armageddon.

3a. Reverse Possession--Take Away, Deprive

Rob, burglar, bare, barren, demote, discriminate (both sc and dm).

3b. Reverse Possession--Dump

Dump, garbage, rubbish.

4. Restore Connection--Repair, Heal, Bridge (gbr)

Mend, medicine, remedy, redeem, bridge, medium (bridge), mediate, arbitrate.

5. Repeated Interaction--Reaction, Dynamics, Social

Social, society, association, burn (reaction with free energy n), brand, rub, rubber (keeps responding), scan (repeated sensing), dynamic, drama, verb, adverb, acrobat, robot.

6. Inward to Interface--Middle

Mid, middle, medium, mediate, meddle, median, intermediate, moderate (middle), abdomen (middle), meridian, demi (half, middle).

7. Repeat Combining -- Structure, Construct, Building (rkb)

Structure, brick, module, rib (structural part), modulate (combine in a regular manner), model, mode, modus, mold, modify (redo combining), made, amend, dome (building), domicile, domestic, condominium, diagram, scheme, sculpture (structure), dummy (constructed figure), melody (repeated combination), madrigal, commode (structure).

8a. Repeat and Continue--Pile up, Excess, Persist (brg, bVr, gbr, Cbr)

Burst (excessive stream s), brutal (excessive attack t), burden (pile-up), bear (persevere, pile up on oneself), burgeon (overflow), bright, brilliant, -berg (pile up), barricade (pile-up), dominate (excess in applying force n), adamant (persist), rebel (persistent negation l), mad (excess), dam (pile-up), scream (excessive sound), screech, robust (durable, persists), mound, dumpy, boredom (excess), doldrums, bedlam, acerbic.

8b. Repeat and Continue--Smooth, Fine, Flow, Fluid (mrd)

Breeze, brook, humid, damp, mud, meander (flow), emerald (fine), dolomite, diamond, jewel, dame (fine), damsel, mild.

9a. Defined Convergence--Near, New (Qrb)

Close, modern (near time), juvenile (new), precise (close).

9b. Defined Convergence--Brief, Shrink (brd)

Brief, concise, midget, timid (shrink), abridge, adsorb.

10a. Back off from Interface--Border, Limit, Restraint (Hjz)

Barring, barrier, border, brim, brow (border), brace (restraint with structure c) and brake (restraint by applying force k), dam (barring water), dampen, smolder, dumb (limited speech or intelligence), dummy (dumb), modest (moral restraint), delimit, dimension (limit), demure (restrained), demarcate, remand (jail).

10b. Back off from Interface--Stay Aloof, Fly, Flee (hrb)

Bird, albatross, nomad (aloof from towns), seclude, escape, abscond, timid (flees), abroad.

10c: Back off from Interface--Before, Front, Ahead (Qdm, Qbl)

Before, breast, bra, brave (goes ahead).

10d. Back off from Interface--Stand out, Emerge, Show (brz)

Barb, demonstrate, syndrome (what shows), prodrome, dream (vision).

11. Charged Contact--Hit, Knock, Fight (Drb, Hrb, QrO, brQ)

Burst, dynamite, bruise, dilemma, drum, amber (chargeable by rubbing).

12. Enter Junction--Go in, Take in, Drink (Srb, Qbl)

Bore (go in), breathe (take in), bury (put in), beer (drink), brew (make a drink), bar (place for drinks), admit (take in), dimple (goes in), absorb, suck, approbation.

13. Repeat Connection--Extend, Fabric (zrb, mhd)

Fabric, robe (fabric), ribbon (fabric), broad (extended), ivy (repeats connection), branch (repeat connection), modem, derm- (skin), drum (skin), denim, damask, dimension (extent), diameter (extent), domain (extent), demur (extend), commodious.

14. Return Equal--Reciprocate, Praise (Hmd)

Barter, commodity (bartered), accommodate, scale (balance), medal (reward), (re)commend, admire.

15. Define Commitment--Prescribe, Command, Contract, Hire (Oqd)

Prescribe, command, dominate, demand, mandate, mandatory, administer, maid (hired), commodore, admiral, baron.

16. Closed Interface--Cover, Curtain, Protect (Hjb)

Condom, diaphragm, demon (hidden), secret, sacred (protected), secure, dim (veil), sack (cover, closure), bark (cover), barn.

17. Limited Presence--Scarce, Little

Barely, diminish, scarce, scanty, seldom, demean, diminutive.

18. Grab and Return--Bring


Using this kind of analysis over our complete set of 32 process-polarity pairs, we are able to make a complete analysis of the morphological and stem structures of the words from many languages.  A lot of work was needed to adjust for language change and vague language when we began testing algorithms that could take some concepts represented with English words (a question in English) and identify possible answers in the Russian or Swedish languages for example..

KE> Just the knowledge of the upper level made things in the
> lower and middle layers fit -- that, in my mind, may not have
> fit before; I learned.  I did not alter my way of thinking
> in that I adapted to new facts.

JS>  That is an important point:  It's necessary to have guidance
on how to organize the categories of an ontology and how to
associate axioms with those categories.  But that kind of
guidance could be obtained from a textbook, a set of design
tools, or a collection of examples.
If only it were that easy! Choose any dozen non-trivial words and look up their definition in three dictionaries and you get confusing accounts.  Put a dozen expert and respected linguists in a room and ask them to agree on the roots and origins of the dozen words you choose.  You will get a dozen perhaps conflicting accounts. 

The design tools for this case are parsers, stemmers, lexicon, thesauri, NLP, etc.. I have found C-MAPS to be excellent design tools, particularly in defining subsumption and part-whole relations but it takes guidance and due consideration to achieve the correct harmony of design criteria with the significant features and characteristics to be specified.

Because pertinence of access to organized knowledge, and the entire utility of such access and such knowledge, is a kind of situational performance, there are measures used to attest to that performance.  Indexing, search and ranked relevance retrieval demonstrate this kind of performance. 

Determining what is relevant requires a capacity for recognizing semiotic patterns that (tend to, or probably) distinguish the features and characteristics considered relevant.  Determining what will or should constitute that relevance to others is not a trivial task.

There are ways of indexing keywords and there are ways of indexing patterns and also relations between patterns.  Using a collection of examples is how Bayesian-based indexing and retrieval systems are trained to learn the patterns of the texts well enough to find hit documents for queries or to perform classification.  The performance of such systems is measured by recall and precision tests. 

<AA> ... The goal of ontology is to formulate the overall patterns and fundamental laws of the universe, while its role is to set the world models, rules, and reasoning algorithms for advanced information technology. <snip>

To test our findings and ideas, we created automatic indexing methods that accept undefined text as input and output a latent semantic table. No training is done. 

The inputs are compiled into compact signatures using a special meta-language made up of our logically complete set of process-polarity pairs, and a language reference assigning about 12,000 English language words to a finite set of a little more than 2000 root linguistic forms and constants we chose to more completely define for use as situational indexicals in our propositional methods.

The result is essentially a binary table that allows for the fast and dynamic computations of relations of concepts and terms (columns) to documents (rows), all according to our methods. That makes this a form of latent semantic indexing, combined with analytical modeling, which is commonly defined as "advanced information technology:".

In addition to deterministic concept and word indexing methods, based on our studies of terminologies and vocabulary, we developed a model of knowledge extraction for the reliable identification of the instances of specified type/subtypes in the records forming a collection.  We also added a logic framework for using Boolean and Horn logic in the specification of types/subtypes. 

In independent relevance tests, designed and hosted by the National Institutes of Standards and Technologies (NIST) at the annual Text REtrieval Conference, we showed that our rather deterministic model could outperform systems based on Bayesian, keyword or those using any other methods. The results of our performance at TREC-8 (1999) is available at:  ftp://ftp.www.readware.com/Software/Support/T8MITi.pdf

It is six pages that speaks volumes about the computational model we developed for organizing the knowledge (compiled from texts). The knowledge was organized for the purpose of addressing various requests (compiled from plain queries specified with a small set of commands).  The objective is to determine which few of about two million candidate documents of various subjects and kinds, are pertinent to requests formulated by the judges themselves.  They should know best what they intended with their request on what they deem relevant.

It is useful also to look at the NIST publications over the entire conference.  There you will find that in terms of relevance, these methods captured five hundred (500) times more  pertinent documents to detailed and sophisticated queries than all other participating systems.  I would say, statistically speaking, that is significant.

Proceedings of the 8th annual Text Retrieval Conference (TREC-8)

I am not saying that search, indexing and retrieval give us all the answers, I am just showing how our model of relevance was developed and validated in a text retrieval context and suggesting therefore it would be useful for other reasoning tasks as well.

-Ken Ewell

Message Archives: http://colab.cim3.net/forum/ontac-forum/
To Post: mailto:ontac-forum@xxxxxxxxxxxxxx
Shared Files: http://colab.cim3.net/file/work/SICoP/ontac/
Community Wiki: 
http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/OntologyTaxonomyCoordinatingWG    (01)
<Prev in Thread] Current Thread [Next in Thread>