Estimating the entropy of an unknown (ergodic, stationary) stochastic source with unknown memory, only relying on finite samples, poses interesting and challenging mathematical questions. In this talk we aim first to review some recent and less recent relative entropy indicators, such as the zipper's based methods. Then we will recall the main definitions and results on non-sequential recursive pair substitutions and their use as relative entropy indicators. Finally, we will discuss some recent applications to authorship attribution of literary texts.