Rbosaghz writeup of Frietag 2000

This is a review of Frietag 2000 Maximum Entropy Markov Models for Information Extraction and Segmentation by user:Rbosaghz

This paper introduced a variant of HMMs called, Maximum Entropy Markov Models (MEMMS), in which the HMM transition and observation functions are replaced by a single function p(s|s',o) that provides the probability of current state s given the previous state s', and the current observation o. This is different from regular HMMs in that the current observation can depend on the current state as well as the previous state. Then it becomes intuitive to think about observations as being associated with transition arcs instead of states. The authors use these new MEMMs to segment FAQs with good results.