On Scaling and Models of Language Evolution


Computational models of language evolution offer important insights for explaining the emergence and evolution of human languages. However, such models have recently been criticized for being computationally intractable. The goal of this paper is to show that this criticism is misleading because it reduces all models of language evolution to only a specific subset of models that assume that the basic unit of cultural transmission is the language itself, which leads to astronomically large hypothesis spaces. In fact, there is already decades worth of computational modelling using the Language Game paradigm that has successfully addressed the issue of scaling by treating language as a complex adaptive system that spontaneously evolves as the side-effect of local communicative interactions. This paper explains why the Language Game method scales so well, and how it incorporates insights from constructivist usage-based learning and Relevance theory.