Hierarchical CRF

I have a NLP task which I current model with a Bi-LSTM CRF model in Tensorflow. I want to add some hierarchy to the CRF, i.e. add another linear-CRF to the current one with some parent-child edges (the objective is still to maximize the log likelihood of the CRF). Is this something I can do in Edward?

Undirected models have very limited support. I wouldn’t recommend it unfortunately.