Nlao writeup of Poon 2007

This is a review of Poon_2007_joint_inference_in_information_extraction by user:Nlao.

This paper discribes an interesting problem with interesting solution. Beware that there is significant amount of domain knowledege used in the solution: the special treatment of punctuation, the construction of JointInferenceCandidate feature. MLN becomes a handy tool to express these knowledges, freeing developers from implementation of learning joint model.

It seems very hard but interesting to have a more general algorithm that can discover and utilize these domain knowledge from just the labeled data.

[minor points]