| The research question for this project is: under what conditions can a group of simulated agents develop a compositional language from scratch, which they can use to perform task-oriented behaviour? This question is raised to investigate two important issues in cognitive science. The first involves the fundamental question how human language evolved. As language is one of the most complex forms of cognitive behaviour in the animal kingdom and because it is unique in its compositional structure and extensive use of symbols, understanding how human language evolved will gain our understanding of the nature of language and cognition. The second issue comes from an engineering viewpoint and investigates how a multi-agent system can be designed that can acquire linguistic knowledge through interactions with their environment. The objective is to design a simulation in which a group of agents bootstrap a limited compositional language with which they can transmit knowledge that is relevant for them in a particular task-environment. The adopted assumptions are that language can develop through cultural evolution for which interactions, learning and self-organisation are the key mechanisms. In addition, mechanisms are required to discover linguistic categories from the agents' sensorimotor activity to enable the development of a compositional semantics that relates to their task-oriented behaviour. Both language and task-oriented behaviours develop from adaptive interactions between the agents and are passed over to subsequent generations. At the start of the simulation, no agent has linguistic knowledge and only a few agents have skills to solve their tasks. Simulation results should reveal whether the adopted assumptions are viable or not, and if so, they indicate under what physical conditions human language might have emerged. In addition, successful design contributes to the development of agents that are able to learn language, which is interesting from an engineering point of view. |
