Suranah writeup for Frietag 2000

This is a review of the paper Frietag_2000_Maximum_Entropy_Markov_Models_for_Information_Extraction_and_Segmentation by user:Suranah.

This paper explores a new type of Markovian sequence model which allows us to model overlapping features instead of just a fixed vocabulary. The transition probabilities are learned through GIS. I like the modification that tries to model the transitions more as features which are informed by the structure of the extraction task. This method has worked very well for the task at hand, and this combination of graphical properties to MEMs seems novel to me.