Building a trajectory syntax through language evolution

Abstract

If we are to understand the innately human ability to solve complex problems, we must first understand the cognitive processes that allow us to combine different kinds of knowledge, to learn new things and to communicate with other people. I have built a computer simulation, based on the work of Simon Kirby, in which I show that a population of induction agents, capable of perceiving their environment and producing utterances, can develop a compositional grammar to describe the world they observe with no prior linguistic knowledge. This system expands the semantic domain proposed by Kirby which expressed meanings such as "John knows Pete" to a physical world of trajectories such as "The boy ran from the tree to the pole". In this new simulation, I demonstrate that a compositional syntax still develops if the level of semantic complexity increases over time. I then argue that using multiple representations decreases the time necessary for a compositional grammar to emerge.