Hierarchical reinforced urn processes

We define a class of reinforced urn processes, based on Hoppe's urn scheme, that are Markov exchangeable, with a countable and possibly unknown state space. This construction extends the reinforced urn processes widely used in Bayesian nonparametric inference and survival analysis. We also shed light on the connections with apparently unrelated constructions, recently proposed in the machine learning literature, such as the infinite hidden Markov model, offering a general framework for a deeper study of their theoretical properties.