Put comments here for "Research on Statistical Relational Learning at the University of Washington" by Pedro Domingos and coauthors.
From James Cussens: bottom of your 1st page: I think that saying the "issue of sample dependency ... is still very much in its infancy" needs to be qualified since there's stuff on eg temporally and spatially correlated observations in statistics. I think this qualification is implicit in your paper, since you go on to talk about modelling arbitrary dependencies, which is indeed the big deal.
Section 7: It's interesting that you're doing PBD. This was a big deal in early ILP, and indeed contributed to it being called ILP. There seems to be a lot less work in it now, I think it was difficult to get it to work without heavy constraints.