1.       Show how you can extend the methodology discussed in Exercise 1 to relationship extraction. 3. Suppose you have a corpus in which words and phrases indicating positive/negative sentiment are tagged. You want to tag corresponding words and phrases with various sentiments in an unmarked corpus. Discuss the relationship of this problem to entity extraction.

2.       The Nymble approach uses the features of the tokens in a relatively rudimentary way. Discuss an approach based on HMMs in which each state of the model can output a multidimensional feature vector.

3.       Show that the running time of the convolutional kernel is at most proportional to the product of the number of nodes in the two trees between which the kernel similarity is being computed.

4.       Implement the string kernel discussed in this chapter.