CS229 Final Project Using WordNet and Clustering for Semantic Role Labeling

In this paper, we compare the performance of several clustering algorithms on the task of semantic role labeling. We use a baseline system based on logistic regression classifiers, and also a distributional clustering algorithm based on word association lists. We use a Latent Semantic Analysis system to compare to two previously implemented clustering algorithms, k-means and a more comprehensive discriminative clustering algorithm. We also focus on features we can extract through specialized corpuses such as WordNet. Overall, we demonstrate that semantically clustering text leads to significant improvements on semantic role labeling tasks.