Several extensions of bayesian belief networks to the first order
  logic or relational framework have been proposed.  Many of these
  have in common that they are embedded in some kind of probabilistic
  or other extension of logic programming.  In this paper we take yet
  another approach, which could be called a meta-interpreter approach.
  We discuss the representation of ``first order'' bayesian belief
  networks in standard Prolog.  The representation formalism we
  propose is very simple, does not make use of any extensions to logic
  programming, allows inference using a simple interpreter written in
  Prolog, and the formalism has an expressiveness similar to other
  relational variants of bayesian belief networks.  Due to the
  simplicity of the framework, we believe it may be a suitable
  reference point to compare other approaches to.