Panlingua, by Chaumont Devin, May 7, 1998.

Chapter 2, The Foundations of Knowledge.

Knowledge can be represented in Panlingua, as we have seen. In the last chapter we had a detailed look at how "A lion killed this zebra" could be stored. And we know that Panlingua structures are built from various kinds of synlinks and lexlinks. And as we shall see later, there is no thought that cannot be represented in Panlingua.

But is Panlingua all there is? Is there any other way in which knowledge can be stored in the thre-dimensional model we have designed? The answer is "yes." Knowledge can also be stored in semlinks, which we have not implemented thus far. Semlinks link semnods to each other, and lie completely within the semantic plane.

Another name for the collection of all semlinks and semnods is "the ontology." The word, "ontology," comes from the Greek words, "ontos" and "logos," and means roughly, "the study of what IS," or "words about what IS." As it turns out this is a very good name, because the major link types of the ontology (semlink types) correspond closely to English auxiliary verbs (recall that the main such verb is "to be," which in Greek is "ontos").

Now imagine with me that an African woman has walked down to the stream at the edge of the village with her child. The child sees a crab scurry away and hide among the rocks. "What's that?" she asks her mother, pointing.

"A crab," her mother responds.

"What is a crab," her daughter persists.

"A kind of animal."

"Do crabs bite?"

"Yes."

"Do they hurt?"

"A little bit."

"Can we eat them?"

"Yes."

Etc.

What is happening here is the fundamental learning of fundamental knowledge by a child. Almost all of this information can be stored in an ontology, where it is evidently more permanent and can be accessed more rapidly than if it were stored as Panlingua.

I have created a computer program that serves as a lexicon and ontology. Using such a program one might enter the following:

crab sng
crabs plr * (asterisk means "of same")
crab isa animal
crab can bite
crab can not injure crabs are edible etc.

In this mechanized example the word in the middle is a mnemonic. In my program it is always three letters chosen to help humans remember. When refering to the ontology such link types are also sometimes called operators, and the two words they link are sometimes called operands.

Whenever such ontological entries are made, either (1) A word is defined and assigned a semnod, as in:

crab sng

or else (2) a semlink is forged between two previously-established semnods, as in:

crabs can bite

The properties of such ontologies have only begun to be explored, but have already proven to be quite amazing. One of the greatest strengths of ontological structures is their ability to "grasp" the ramifications of new information. For example, suppose an ontology is up and running, and the following entry is made:

Charley is a man

Then the following interactions should immediately be possible:

Can Charley talk Yes

can Charley menstruate No

can Charley give_birth No

has Charley testicles YES

has Charley vagina No

Etc.

This power to subsume and know many things about new items without ever being told enables massive amounts of information to be stored using minimal computational resources. In fact an ontology can ideally return relational information on all semlink types for n(n-1) word pairs, where n is the number of words in the lexicon. Thus for only 10,000 words and 10 semlink types, an ontoloty can return information about 99,990,000 * 10 = 999,900,000 combinations. The ontology is thus by far the most powerful knowledge storage medium ever known to man.

But ontological knowledge, although very fast and very massive, is limited strictly to knowledge about binary relations. Thus it is possible for an ontology to know that "Roses are red," but it is not possible for an ontology to know that "John can dance the jitterbug." In the ontology knowledge is organized into a sort of rough-and-ready instant reflection of the real world, but an ontology cannot be used to store specific knowledge about people, places, and events.

So ontological knowledge might be viewed as the foundation of all knowledge, because it exists within the linguistic planes but at a sublinguistic level. In other words, Panlingua is a language--indeed, the mother of all languages, but ontological information is not.

Do animals possess ontologies? Almost certainly. In fact it may be that ALL animals possess minimal ontologies of some sort or another. Otherwise how could animals know that:

item_a is edible
item_b is not edible
item_c is edible
etc.

And many animals can probably construct and manipulate Panlingua structures as well. As a matter of fact it has been shown that predatory insects such as spiders can carefully plan and execute their attacks. It may be that they employ something other than Panlingua, but this would not seem likely for reasons we shall later explore.

But it is important to notice that not all ontological operators (semlinks) are created equal. Certain semlink types are of crucial importance, and must be present in nearly all ontologies. The first of these is probably the hypernym link (something is a kind of something). The second is probably the holonym link (something is a part of something). Then antonym, synonym, and others. And some semlinks must be handled in special ways. For example, X is a hypernym of A if A is a B and B is a C and ... is an X. And A is a holonym of B if A has a B, A is something that has a B, or A has something that is a B.

Among other indispensable things, the hypernym link enables a great deal to be said about many words using only a relatively few entries. The holonym link also enables many questions to be answered using minimal data.

To understand how ontologies can return such powerful results, let us examine briefly how they work. Suppose we have the entries:

man isa human
human isa ape
ape isa simian
simian isa primate
primate isa mammal
mammal isa vertebrate
vertebrate isa animal
animal isa organism
organism isa material_object
material_object isa material_thing
material_thing isa thing
man isa male_mammal
male_mammal has testicle
woman isa human
woman isa female_mammal
female_mammal can give_birth
animal can eat
Charley isa man

Such an ontology would "know" that Charley can eat because: 

Charley isa human
human isa ape
ape isa simian
simian isa primate
primate isa mammal
mammal isa vertebrate
vertebrate isa animal
animal can eat

And it would know that Charley can't give birth because there is nothing above Charley in the hypernym hierarchy that can give birth. But it would know that Charley has testicle because:

Charley isa man
man isa male_mammal
male_mammal has testicle

>From these examples it can be seen thatif A is a hypernym of B, and A can do something, then B can do that same thing also unless specifically stated otherwise. And if A is a hypernym of B, and an A has something, then a B has that something also. But if A is a holonym of B, and B is a holonym of C, then an A has a C and a B has a C. Etc.

The hypernymy hierarchy serves mainly to scramble up and down within an ontology looking for other relations. A is a hypernym of B if B is a kind of an A, and B is a hyponym of A if b is a kind of an A.

For holonymy and meronymy, on the other hand, A is a holonym of B if an A has a B, and A is a meronym of B if a B has an A. Thus "brake" is a meronym of "automobile," and "automobile" is a holonym of "brake" (cars have brakes).

So a hypernym is the opposite of a hyponym, and a holonym is the opposite of a meronym.

An important use of the holonym/meronym relation is to discover the topic of discourse for a segment of text. Thus for a sentence like:

He adjusted the brakes and set the timing.

a check of the ontology might reveal the following semlinks:

a brake is part of a wheel
a wheel is part of a conveyance
timing is part of an engine
an engine is part of an automobile
an automobile is a conveyance

Thus from the ontology we might find that automobiles have both brakes and timing, and surmise correctly that the topic of discourse of this sentence is an automobile.

Besides its role as a repository of world knowledge, the ontology plays a crucial role in parsing without which certain kinds of disambiguation would be virtually impossible. As an example, take the word, "had," in "Martha had a baby in the hospital." Does this mean that Martha possessed or that martha gave birth? Well, if the ontology "knows" that Martha is a woman, and if in the general Panlingua reference there exists an entry like, "Women give birth in hospitals," then by traversing the general reference using the ontology the system should be able to figure out that "Martha had a baby in the hospital" probably means that Martha gave birth.

The ontology also plays an important role in search-match operations in which identical words are not used. For example, if a Panlingua reference says that "Automobiles travel on roads," and it is searched to learn whether a chevrolet can travel on Highway 99, such a search should return "Yes" because the ontology says that a Chevrolet is a kind of automobile and Highway 99 is a particular road, etc.

Ontologies tend to differ from person to person, and multiple ontologies may exist for certain people. To illustrate what I mean, I was once fascinated by a Philippino woman who lived in California. She is surely one of the most brilliant people I have ever known. But I observed three basic personalities in her. She was a smart Philippino with a perfect command of Tagalog, and when she spoke Tagalog, her whole being went into Philippino gear, and she became a talkative Philippino lady. But in English she had two personalities. One was the very shrewd, very professional businesswoman, and another was plain Miss Everyday American. If I had to guess, I would quickly say that she must have had a separate ontology for each of these three modes. As another example, suppose that one man is a criminal. He believes that stealing is good because it brings him satisfaction. The semnod linked to "steal" in his ontology is linked by hypernymy to "good thing." There is another man who is honest, and believes that stealing is a crime. In his ontology, the semnod that links to "steal" is linked by hypernymy to the semnod linked to "bad thing." The two ontologies are not the same primarily because their link patterns differ.

Ontologies also differ from language to language. 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.

But once reliable ontologies have been built for two languages, bridging relations can be forged between the semnods of the two ontologies for use in machine translation.

Lastly ontologies can be made to self economize by the following rules:

If A has a C and B has a C and a B is a kind of an A, then break the "B has a C" link.

If an A has a C and an A has a B and a B has a C, then break the "A has a C" link.

Etc.

So to summarize, ontologies contain massive amounts of the most basic kinds of knowledge, and it is a pretty sure bet that no human language would be possible without them. However ontologies tend to differ from culture to culture and from individual to individual.