Algorithms from "propositional" machine learning always seemed to me to be generally applicable. I could read descriptions of Support Vector Machines, Decision Trees, neural nets etc., and visualise how I might apply them to my propositional problem and what kinds of benefits I might gain.
My experiences reading about relational machine learning haven't been quite as revealing yet. I think the richness inherent in relational problem descriptions makes it more difficult to grasp the bigger picture. I'm interested in the commonalities among relational learning domains (and the pecularities of particular ones), and the descriptive power of hypothesis languages used by various relational learners (in particular, what kinds of regularity are difficult or impossible for a particular learner to find).