Stochasticity as a Source of Innovation in Language Games

Recent work on viewing language as a complex adaptive system has shown that self-organisation can explain how a group of distributed agents can reach a coherent set of linguistic conventions and how such a set can be preserved from one generation to the next based on cultural transmission. The paper continues these investigations by exploring the presence of stochasticity in the various aspects of lexical communication: stochasticity in the non-linguistic communication constraining meaning, the transmission of the message, and the retrieval from memory. We show that there is an upperbound on the amount of stochasticity which can be tolerated and that stochasticity causes and maintains language variation. Results are based on the further exploration of a minimal computational model of language interaction in a group of distributed agents, called the naming game.