Is language learning associationist or symbolic or both?
What language learning mechanisms do babies have?
What features would a neural network need to learn such rules?
Sequential time can be handled by delay connections.
Segmentation can be handled by synchronization on a binding dimension.
Can a neural network learn the rule task?
Playpen incorporates delay connections,
a binding dimension, and micro-relation units.
What can this tell us about variables,
variable-like behavior, and
neural networks?
http://www.cs.indiana.edu/~gasser/Talks/MAICS99/