Agent-Based Models of Strategies for the Emergence and Evolution of Grammatical Agreement
i) Computer simulations for 50 game series involving 10 agents playing 1000 language games (200 per agent). Average values and standard deviations are shown. At each time point only two agents play a game, although the model works just as well with parallel interactions. Usage inventory reaches a peak after 50 games (10 per agent) after which it gets damped to an optimum of three markers (because the maximum number of objects chosen as topic is 3) due to the lateral inhibition dynamics, before increasing slightly when new agents come into the population. Variation gets damped quickly and efficiency is close to 0.3. The inventory is maintained despite population turnover () although new inventions may arise and in rare cases occur where a new invention overtakes existing markers. ii) The average preference scores for all invented markers in the memories of all agents for a single experiment. There is one marker with the highest score and two others with lower scores. When a new agent comes in, the average scores go down (see circles) but move back up as the new agent acquires the existing preferences.