The problem we need to solve is that we don’t know the alignment between states and observations.
What is “hidden” ?
The model can use many possible state sequences to generate any given observation sequence. We don’t know which one it used, and we don’t care!
This is the new version. Still under construction.