Computer Science · Computer Networks and Communications
Where Innovation Meets Evolution Signaling Game Dynamics and the Vowel Space
See on the mapAbstract
With the objective to explore the evolution of language conventions, signaling games (Lewis, 1969) recently became a leading model for this purpose. In line with this trend researchers used simula-tions to explore the behavior of agents playing repeated signaling games combined with learning dynamics like reinforcement learning on social network structures or at least multi-agents accounts (c.f. Mühlenbernd and Franke, 2012). With this paper we want to make a contribution to this line of research by extending the account of signaling games and reinforcement learning with the possi-bility of innovation, i.e. agents can invent new messages and furthermore unused messages can get extinct (c.f. Mühlenbernd, Nick, and Adam, 2012). This new account constitutes a form of agree-ment dynamics (c.f. Steels, 1998). We’ll show that with the property of innovation the emergence of society-wide signaling systems is guaranteed even for complex signaling games. Additionally, inspired by a study of Jäger (2008), who used repeated signaling games with a metric message space applied on multi-agent populations to simulate the emergence of categorization systems of the human vowel space, we extended our account with a metric message space to simulate exactly the same phenomenon. As opposed to Jäger’s results our experiments show that the segments of the vowel space that constitute phoneme categories cannot emerge simultaneously, but