Agent-Based Models of Strategies for the Emergence and Evolution of Grammatical Agreement
i) Performance summary for 50 game series of 1000 games (200 per agent), average values with standard deviation. We see that variation, which is unavoidable due to the different possible ways in which re-categorization can be deployed, gets damped quickly. The number of markers is reduced compared to the plain reuse strategy (figure 9 i). ii) Development of which feature matrix becomes associated with the controller epwui. This controller started out with an empty feature matrix, i.e. no semantically grounded features. There are different ways in which it can be re-categorized with respect to the existing marker inventory. In this experimental run, a feature matrix with a positive value for has become dominant.