Scott Wen-Tau Yih will give a talk titled "Learning with Integer Linear Programming Inference for Constrained Output". The talk will first demonstrate how constraints can be incorporated into conditional random fields using a novel inference approach based on integer linear programming. This allows CRF models to efficiently support general constraint structures. Experimental results will be provided for semantic role labeling. The second part will compare simple learning plus inference to inference based training, finding the latter is superior when local classifiers are difficult but requires more examples to show differences.