Newsgroups: rec.arts.int-fiction
Path: gmd.de!Germany.EU.net!news.dfn.de!darwin.sura.net!paladin.american.edu!howland.reston.ans.net!usc!cs.utexas.edu!uunet!ddsw1!chinet!jorn
From: jorn@chinet.chi.il.us (Jorn Barger)
Subject: Fractal thickets for IF
Message-ID: <C49wMu.1vq@chinet.chi.il.us>
Organization: Chinet - Public Access UNIX
Date: Mon, 22 Mar 1993 04:32:54 GMT
Lines: 45

For non-readers of comp.ai, here's a precis of my new indexing topology:

I propose a basic vocabulary with a structure that approximates 7 * 7 * 7 
orthogonality, with about 7 *relationship-verbs* each for any *pair of 
subject-and-object types*, with 7-or-so *types* including at least person, 
place, thing, and motive.

Examples:
person-thing: makes-acquires-uses-breaks-disposes
  (person makes thing, person acquires thing...)
person-place: born-arrives-lives-defends-leaves
person-motive: suffers-seeks-enjoys
motive-motive: tradeoff, substitute, exchange, etc
person-skill: discovers-learns-uses-teaches
person-person: over, communicates, kin, etc

person-person-thing: gives, etc
etc

Further building blocks to explore:
person *knows* (etc) relationship
relationship *then* relationship


The fractal 'thicket' structure for story data:

In this system, one may build (or index) *relationships* by selecting 
*elements* repeatedly from the person-place-thing menu.

The person-place-thing menu is the top level of a simple abstraction 
hierarchy, whose *specializations* will also be useful in specifying 
relationships. (Person may specialize by age and gender, thing into tool-
food-bodypart-waste, motive into food-safety-family-esteem-sex etc.)

The only possible data structure for repeated selections from a tree-menu is 
a fractal topology of trees within trees.

This stands in the same relationship to a single tree as an N-dimensional 
array does to a 1-d array.

The resulting structure is combinatorically explosive, but can't be bettered 
for semantic richness.


Jorn
