Prototype categorisation and the emergence of a lexicon in an infinite world

Abstract

One of the least understood issues in language evolution is how hominins were able to ground and establish a shared lexicon. Recently, researchers have explored this issue using a variety of computational models, whose results have suggested that a shared lexicon could have emerged spontaneously through a process of self-organisation. However, these models have used psychologically unrecognised concept representations and an oversimplified environment. In this dissertation, I present a new computational model in an attempt to address these problems. Agents' category representations are inspired by prototype theory, having central members and graded membership. The environment consists of an infinite number of objects, and has a probabilistic structure which can be easily manipulated through model parameters. Despite the relatively complex model, simulation results are generally in line with previous ones and add further support to the self-organisation hypothesis. In addition, the speed and level of lexical convergence depend on the world structure, confirming that this is an aspect of past models which has seen too little attention. Future work should investigate the vast parameter space in further detail, and extend the simulations in various new directions.