Common Language Acquisition by Multi-Agents

Abstract

We propose a multi-agent model, where each agent learns a common language in the community, modifying his/her own initial grammar through a number of exchanges of sentences with other agents. As the initial step toward the realization of this model, we presuppose that a child agent who has a primitive grammar set is thrown into a community consisting of adult agents. The child agent modifies his/her grammar and learns a refined one , imitating adult’s sentences, while the adult agents also loosen their grammar in order to accept what the child agent says tolerantly. In this paper, we report the result of experimentation with this model, together with grammar representation and learning methods. We showed that our model could realize adaptability, which is one of the important features of natural language.