Learning Vowels Using Self-Organization

This paper describes research that investigates the structure of human vowel systems through individual learning and self-organization. A population of agents that have to learn to imitate each other has been implemented. The agents can interact with each other through so-called imitation games: structured exchanges of sounds from which they can learn how to improve their repertoire of sounds. The architecture of the agents, the way they produce and perceive sounds and the workings of the imitation game are described in some detail. The results of experiments with artificial agents show that realistic vowel systems can be learned. Then it is described how the system can be extended to work with real signals and it is indicated that the results obtained in this way are similar to the ones obtained with the first system. It is also demonstrated that it is possible to learn vowels from a human speaker. Finally the relation to other linguistic work into the structure of vowel systems is described.