neural nets with internal classes

Hi !

I'm looking for keywords (or papers) about NN that :
- have sorts of internal states (or classes, maybe by unsupervised learning)
- could predict (by linkage, not by interpolation) the next internal 
state (sequentiality, so takes time into account) : that's sounds a bit 
like markov models, but I'm not sure.
This seems to be linked with natural language learning and processing, 
but not specialized in that.

My searches with Google weren't very helpful.
I'm aware I'm not quite clear, that's why I'm looking for information...

Thanks,
Laurent