Ksuravar writeup of Sha and Pereira 2003

This is review of Sha_2003_shallow_parsing_with_conditional_random_fields by user:ksuravar.

Summary:

1. The paper is about using conditional random fields for shallow parsing.
2. The paper starts of with explaining two current methods for shallow parsing and discusses the weaknesses in both the systems and how CRFs address both the weaknesses.

Comments:

1. I did not understand the data preparation part. Why is the paper using pre chunked (or pre shallow parsed) data? Doesn't this make their method dependent on the preparation part? What is the reason behind doing this?
2. I did not understand their NP chunking F scores table. The paper said they used a variety of training methods but did not mention whether the F score presented in the table is the average of F scores or F-score of one particular training method.