Аннотация:We study conditions under which a sequence of statistical experiments can be approximated in a certain sense by experiments generated by exponential families with a convex canonical parameter space or weakly converges to such an experiment.