Emergence and evolution of language in multi-agent systems

Abstract

Human language provides a highly efficient communication system, which emerged due to cultural interactions. To understand this still mysterious process, various interdisciplinary efforts are being made, which refer to evolutionary linguistics, evolutionary game theory or cognitive sciences. A promising approach considers language as a collective phenomenon, which emerges in a population of communicating agents. Numerous papers based on computational modelling demonstrate ubiquity of spontaneous linguistic synchronization among such agents. The approach got considerable impetus with the introduction of the so-called language game models. Their important feature is the horizontal nature of interactions between agents, which interplay within one generation only and do not create offspring, to whom they would transfer their linguistic skills. Recently, more sophisticated approaches are being developed with agents equipped with some cognitive abilities, employing, for example, reinforcement learning. Such a framework originated from Lewis signaling game, which was adapted to language evolution with subsequent extensions implementing, for example, Bayesian inference, neural networks or deep learning. Models with the reinforcement learning may combine single-generation language games with intergenerational learning, and certainly deserve further studies. The present paper provides a brief review of this interesting and rapidly developing research field.