An Incremental Model of Lexicon Consensus in a Population of Agents by Means of Grammatical Evolution, Reinforcement Learning and Semantic Rules - Foundations on Natural and Artificial Computation - page 40

An Incremental Model of Lexicon Consensus in
a Population of Agents by Means of
Grammatical Evolution, Reinforcement Learning
and Semantic Rules

Jack Mario Mingo 1 and Ricardo Aler 2

1 Computer Science Department, Autonomous University of Madrid
[e-mail]
2 Computer Science Department, Carlos III University of Madrid
[e-mail]

Abstract. We present an incremental model of lexicon consensus in a
population of simulated agents. The emergent lexicon is evolved with a
hybrid algorithm which is based on grammatical evolution with semantic
rules and reinforcement learning. The incremental model allows to add
subsequently new agents and objects to the environment when a consen-
sual language has emerged for a steady set of agents and objects. The
main goal in the proposed system is to test whether the emergent lexi-
con can be maintained during the execution when new agents and object
are added. The proposed system is completely based on grammars and
the results achieved in the experiments show how building a language
starting from a grammar can be a promising method in order to develop
artificial languages.

Keywords: Swarm Intelligence, Grammatical Evolution, Language
Acquisition and Language Development.

1 Introduction

Traditionally the study of language has been promoted by psychologists, biolo-
gists, linguists, philosophers and other researchers of human and social sciences.
With the increasing power and flexibility of computers a new via in the study
of the language raised. Mostly, these computational studies try to analyse the
human language using simulations (see [1] or [2] for a detailed review). Fewer
studies have been proposed only to develop a language for a population of robots
or simulated agents. The system we present here corresponds to this approach.
The first attempts in elaborating complex communication for artificial agents
is being devoted to create a basic vocabulary or lexicon that it is shared by all
the agents. In the context of human communication, grammars have been used
to explain the language acquisition and development process since Chomsky
popularized them in the fifties [3]. Curiously, grammars are not extensively used
to build artificial languages for artificial organisms. A pioneer work in this line