Co-evolving Language and Social Structure Using a Genetic Algorithm

It is interesting how we can take a train of thought and transfer this into an other person's mind by pushing the air around us. Human language, this complex medium that distinctly separates humans from animals, has baffled scientists for centuries. But as it lacks of historical data, researchers wish to benefit from computer science and the field of artificial life to understand the origin of language. This thesis illuminates the potential for using agent-based models to investigate the relationship between biology, culture and behavior on an individual level. This is done in two parts. First, different theories and computational models experimenting with language evolution are presented. This includes a thorough implementation of and elaborations on one recent paper, where language acquisition is illustrated favorable over multiple evolutionary time scales in an agent-based model. In the second part, a more bio-inspired methodology is proposed to make the former model more robust and better suited for extensions. This is demonstrated by letting the agents evolving some social biases, while they are conducting a naming game in a social structure. A naming game is an abstraction, often used in the research field, to model the spreading and diversity of language. Through pair-wise dialogs, the goal of the game is to reach self-organized agreement on naming an arbitrary object in their environment. Given the assumption that communication is beneficial for social structure and that social structure is beneficial for reproduction, the experimental work demonstrates that agents are able to build social structures that resembles real life social topologies, although the naming game might happen too rapid in respect to the evolving social structure. Hopefully, with support from other disciplines, the presented model is suited for further investigation of social, or other functional, traits that can influence language evolution