AI Seminar
-------------------------------
Tuesday, October 11th, 2005
4:00 pm - 5:30 pm
175 ATL (Large Conference Room)

"An Agent-Based Model of Language Acquisition and Evolution"
Prof John H. Holland
Psychology & EECS
University of Michigan

===============================

The contemplation in natural science of a wider domain than the actual 
leads to a far better understanding of the actual. 
                              – A. S. Eddington	

Is it possible to model language acquisition and evolution using simple 
cognitive mechanisms not tied to language?  The model proposed here 
involves the interaction of multiple agents trying to acquire spatially 
distributed resources in order to survive and reproduce.  In the 
parlance of linguistics – the agents are situated.  Language has value 
to the agents only if it increases their ability to collect resources. 
The model centers on inter-agent communication that combines elements of 
a limited vocabulary to describe novel or complex situations.  In 
linguistic terms – the agents are structured.

The model is constrained by well-established concepts from cognitive 
psychology, but it does not employ parameters from language acquisition 
studies.  That is, the model is exploratory: It is intended to outline 
possibilities, not actualities.  Because the model can be run with and 
without language acquisition capabilities, control experiments are 
possible:

Each agent’s survival depends upon its ability to collect resources that 
are distributed spatially in its environment.

Language has value to the agent only if it increases the agent’s ability 
to collect these resources.

There’s no explicit a priori value assigned to language.

Control experiments are possible:  The model can be run with and without 
language acquisition.

Note:  This is an exploratory model.

The model is meant to demonstrate that simple cognitive mechanisms not 
tied to language are adequate to the acquisition and evolution of grammars.

The model is constrained by well-established concepts from cognitive 
psychology, but it does not employ parameters based on language 
acquisition studies.