Liuliu writeup of Klein and Manning

This is a review of the paper Klein_2002_conditional_structure_versus_conditional_estimation_in_nlp_models by user:Liuliu.

I think this is a very nice paper in illustrating some fundamental problems in NLP. It's not about a new model or new estimation method of a particular NLP problem. In contrast, it clarifies the conflicting thoughts existed in NLP area with experiments on different NLP problems. They argued that discriminative objectives helps improve accuracy while conditional model structure suffers helps decrease accuracy because of the label bias and observation bias it brings.

I like the way they separate conditional model structures and conditional parameter estimation. I also like the way they qualify the benefits or badness of "models", "parameter estimations"(and "features").