Just a wild guess, but I wonder if there could be a nice connection between probabilistic programming and logic languages like prolog, if only due to the fact that both Bayes graphs and unification both have this "run forwards and backwards" characteristic that is so unlike imperative programming