Cognitive scientists have shown that, in order to learn efficiently, human infants are selective in sampling information from their environment. Children apparently focus their attention on material that is neither too simple or too complex: the so-called Goldilocks effect. Experimentation has shown that there is a complexity of around 1.25 bits corresponding to infants' preferred information rate.
The task in this project would be to model the complexity of a simple language domain (it can be pseudo-language); and then produce a classifier which can predict whether an item from this domain would be focused on by a human child given their current knowledge base. An extension to this would use the classifier's output in conjunction with the linguistic items as input to a model that attempts to learn the whole domain.
The Goldilocks Effect: Human Infants Allocate Attention to Visual Sequences That Are Neither Too Simple Nor Too Complex. Celeste Kidd , Steven T. Piantadosi, Richard N. Aslin