Thursday, August 19, 2004
In this message Ken continues to develop some description of the
techniques that were used in developing the Readware stratified ontology.
Paul,
Okay, for completeness and focus, let's see how law and court are defined as concepts in a (data)base of concepts in the ReadWare system.
We call the term 'law" a concept. It is a "bucket" in the structure of the conceptbase, for one or more lexical representatives. The term law is the leader of a group of lexical representatives of the concept named by the group-leader 'law', including: bylaw, cannon, decree, edict, ordinance, sharia, regs, regulation, and statute. All the terms are given the same vector. In practice each of the representatives is linked to the (memorized) location of the law-bucket. In addition to this, no lexical representative of any known concept is independent of its place in the taxonomy (schemata) of the entire domain of reference and expression.
The conceptbase is not a dictionary or thesaurus as much as it is a referential database, though it provides the function of processing synonymous and polysemous relations. WordNet can provide that same kind of functionality. WordNet goes further to define words and their usage and linguistic relations.
But the use of WordNet in retrieval systems does not improve retrieval results beyond a certain point. [1]
Through the inherent taxonomy of the ReadWare conceptbase schemata, the law reference also extends to other buckets of concepts. For example: Legislation(lawmaking), bill(account), legality(lawfulness), and code(symbol).
We do not specify constraints nor limits to the possible relations. The objective is to remain open to all points of view and all possible interpretations at this stage of reference.
The context itself, or rather the strength of the coherency of term occurrences in a given context, delimits the validity of the connections. It is the harmonious and indirect extension of the reference that connects it to another node of the concept taxonomy-- that of court. It is not pre-specified, it can only come from the measure in a context or situation characterized by its high variability in size, forms and structure.
Nonetheless, we know also that the concept of court is polysemous: there is a court-house, a court-yard, a court-ship, and a court and the courtier of the palace. It is the connection to lawfulness and the presence of the judge or the tribunal in the courtroom of the courthouse where the law is decided. Courthouse is called a superconcept (a superconcept is a group of concepts, while a concept is a group of words) in readware terminology. Not only the law, also its litigation, adjudication and practioners belong to this group. The courtier and the electric bill do not.
Well, people need to understand that these sorts of things are not fixed. We people cannot fix anything for certain. Even our satellites eventually fall out of the sky and disintegrate in flames. We are, or should be, looking for the beauty, balance, harmony, and fidelity that makes us happiest and most secure. Where there is insecurity and conflict we should be looking to bring the opposing forces into a balanced and peaceful relationship. But that last part is just my opinion and maybe there is a little topic drift there as well.
At any rate, it is at this point, at this place where our attention can be focused onto "things we know". Operations involving < a, r, b > type constructs can be applied liberally to support other reasoning techniques and processes or for verification and validation purposes.
Studying the Arabic language and looking at the triplets of consonants in the Arabic language roots were the inspiration for the taxonomy, for the measures of relatedness and for the structure of the conceptbase, so we do not say we invented it at all. We just happened to see it; to discover it. I say we, though it was completely the keen and studious vision of Dr. Tom Adi.
Until then, we did not know how to tell some nonsense, "plexore the drowl with dingabid spectre" from something verifiable and coherent "explore the world with abiding respect".
--/Ken Ewell
[1] A principled discussion can be made as to why one cannot expect performance from WordNet beyond a certain point. Evidence that the Readware conceptual fidelity is better than machine ontology with WordNet can be shown, but this evidence is not accepted within the mainstream professional community. Part of the issue here has to do with the degree to which economic and academic success is conditions by highly conservative forces. There is history here to consider.