The Evolution of Shared Concepts in Changing Populations

The evolution of color categorization systems is investigated by simulating categorization games played by a population of artificial agents. The constraints placed on individual agent’s perception and cognition are minimal and involve limited color discriminability and learning through reinforcement. The main dynamic mechanism for population evolution is pragmatic in nature: There is a pragmatic need for communication between agents, and if the results of such communications have positive consequences in their shared world then the agents involved are positively rewarded, whereas if the results have negative consequences, then involved agents are punished. A mechanism for changing the composition of the population due to agents’ birth and death is also investigated. This birth-death mechanism is found to effectively move an established population color naming system toward a theoretically more optimal one. Specifically, our use of birth-death dynamics suggests that (i) agent populations do reach stable, consistent color categorization solutions, (ii) population categorization solutions derived from categorization game encounters can change even though each individual agent’s solutions tend to remain unchanged, (iii) the number of color categories in observed solutions varies with fluctuations of agents’ lifespans, and (iv) categorization solutions are subject to change until they reach theoretical optimality. The simulation results of this article provide insights regarding mechanisms that may constrain universal tendencies in human color categorization systems observed in the linguistic and anthropological literatures.