Working memory, short term memory, long term memory

[Luis.Botelho at iscte.pt]

96/12/18
Dear ACT-R Users,
my name is Luis and I work in the Computer Science Dept. of 
ISCTE, Lisbon, Portugal. I am not an ACT-R user, although I 
recently became a member of this mailing list.

This message is not directly related to ACT but it also 
concerns cognitive science, namely working memory, short 
term memory and long term memory. If you feel this is not the 
appropriate place to this kind of discussion, please ignore this 
message and accept my apologies.

I've developed a model of memory for autonomous artificial 
agents taking inspiration in cognition, namely in ACT and in 
the works of Robert Wyer and Thomas Srull, Allan Collins 
and Elizabeth Loftus, and Gordon Bower. The model is an 
hybrid model called SALT (Schema-Associative Long Term 
memory). SALT has two main components besides some 
temporary buffers: working memory and long term memory. 
Some time ago, Katia Sycara said that I'd better call short 
term memory to what I was calling long term memory. I 
would very much appreciate your comments on this matter. 
Please find bellow a short description of the SALT model.

According to SALT, long term memory is an associative 
network with spreading activation. Each node contains a set of 
symbolic structures representing cognitive structures. These 
cognitive structures may be procedures, frames, condition-
action rules, declarations, and the like. In the current 
implementation of the model, each of the mentioned cognitive 
structure is a Prolog clause. A node also contains a header that 
describes its contents. Headers are used in pattern-matching.

Each time an external stimulus is put to the agent, the currently 
most activated node whose header matches the external 
stimulus receives some activation that spreads to the nodes to 
which it is associated.

When the agent faces some problem, it searches its long term 
memory in decreasing order of activation and selects one or 
more nodes appropriate to solve the problem. The cognitive 
structures contained in the nodes selected are copied from long 
term memory to working memory. There, some form of 
symbolic processing takes place and a solution to the current 
problem is tried.

If I understood correctly, Katia Sycara suggested that the 
component of SALT currently named Long Term Memory 
should be called Short Term Memory instead. What is your 
opinion? If you agree with her, then what is Long Term 
Memory?

You may reply either to the ACT-R mailing list or to my own 
email address (see bellow). I thank you for your attention,
Sincerely

-- Luis
======================
Luis Miguel Botelho
luis at iscte.pt
Fax:		351-1-7964710
Phone:	351-1-7935000	Ext. 12411
Dept. de Ciencias e Tecnologias de Informacao do ISCTE
Av. das Forcas Armadas, Edificio ISCTE
1600 Lisboa, Portugal