Learning Biases for the Evolution of Linguistic Structure: An Associative Network Model