Nschneid writeup of Cohen 2000 WHIRL

This is Nschneid's review of Cohen_2000_whirl_a_word_based_information_representation_language

• Is it fair to say that this is broadly similar to Markov Logic, only (a) specific to a representation in which documents consist of sets of short text segments and (b) using pairwise TFIDF similarities rather than a complex graphical model?
• Has this sort of formalism for entities been used with feature-based representations? i.e. documents and/or their segments are feature-valued rather than string-valued?