Exploring Various Features in Semantic Role Labeling
Hongling Wang, Guodong Zhou, Qiaoming Zhu, Peide Qian · 2008
This paper explores the contributions of various features in semantic role labeling. Moreover, an optimal set of features is selected using a greedy strategy. Finally, an effective headword-driven pruning algorithm is proposed to filter out irrelavant instances. Evaluation on the CoNLL'2005 SRL benchmark corpus shows that our method achieved comparable performance with the best-reported ones on a single automatic parse tree.