Request for data/help

mr k.h. geppert ins749r at
Tue Oct 22 10:18:34 EST 1991

A program that can simulate people has long been an interest of computer
scientists. So far, an unfulfilled interest... However, there is a fairly
small field in which a simulation is possible, and if not useful, or insightful
then at least amusing. This is a simulation of the free associations
produced by people. This is easier because there is less structure to be
interpreted, and the raw content is more visible.

I am trying to do the simulation by using a simple neural network based
on some classical conditioning rules established by Grossberg in the late
60's and in his book "Studies of Mind and Brain", and updated by Klopf in
a recent paper. It is a system for learning temporal sequences, where each node
represents an item (or word, or concept), instead of a vector representing an
item. I am presently attempting to get context to work, so that
the next item to be output by the simulation will depend on what went before,
and not just the immediate predecessor. Grossberg mentions the idea of context,
but doesn't say much useful about it. I've found one article that does say
something about it : "Neural networks that learn temporal sequences by
selection", Stanislas Dehaene et. al., Proc. Nat. Acad. Sci. USA, Vol. 84, 
pp. 2727-2731, May 1987
If anyone is doing research in this area, please drop me a line:
bren at

If you are not, you might still like to contribute free-associations to help 
train the network. Here's a blurb:

Free-associations are words that follow each other with no set relationship
between them, apart from that a person has said them one after the other. An
example is:

middle east
kentucky fried

It's good to visualize scenes as you go along, and it's also very relaxing.
If you would like to contribute free-associations to help train this network,
please post them to:

bren at


More information about the Biomatrx mailing list