Rbosaghz Han writeup of Cohn and Blunsom

This is a review of Cohn_2005_semantic_role_labelling_with_tree_conditional_random_fields by user:Rbosaghz.

This paper uses CRFs to solve the semantic role labeling task. For each sentence, they parse the sentence using the Collins parser, then for each node of the tree, the model predicts a semantic role label, which is interpreted as the labelling for the corresponding syntactic constituent.

This was the first time CRFs have been used for SRL. I particularly liked how they used CRFs on a non-traditional structure (i.e. not just a boring sequence).