Rbalasub writeup of Bellare and McCullum

A review of Bellare_2009_generalized_expectation_criteria_for_bootstrapping_extractors_using_record_text_alignment by user:rbalasub

This paper approaches the task of aligning fields from database entries to unstructured text. The model uses is similar to IBM Model 1 and uses General Expectation Criteria to express beliefs about the target alignments. Two models are proposed - a) the supervised alignCRF model and b) the extrCRF

Experimental show improvements over baseline methods. The use of GE criteria is an interesting aspect about the paper. It is a very appealing way to introduce external domain knowledge into a model without complicating the model.