Philgoo Han writeup of Cohn and Blunsom

This is a review of Cohn_2005_semantic_role_labelling_with_tree_conditional_random_fields by user:Ironfoot.

• CRF on tree structure
• Data representation
  • Pruned tree: reduce size
  • Constituents children labeling (new inside or parent's argument label)
    • new inside label is better
  • What is 'argument'? I don't understand this part very well
• Model
  • 1-node, 2-node feature
  • Use only 1-node feature to mimic maxent
• Result
  • There are no comparisons
  • Where is the result for the preliminary case?