I'm just working through the math in the paper right now. To me, it looks like all of the formulas could be expressed in a probabilistic PL and MCMC could be used to estimate the probabilities of both latent state and parameters. Would there be some particular roadblocks to this approach?