Paper: A Memory-Based Approach For Semantic Role Labeling

ACL ID W04-2418
Title A Memory-Based Approach For Semantic Role Labeling
Venue International Conference on Computational Natural Language Learning
Session Main Conference
Year 2004
Authors

New instances are then assigned the most frequent class within a set of k most similar examples (k-nearest neighbors). Memory-based learning algorithms have proven to be effective for several NLP tasks, including named entity recognition (Hendrickx and van den Bosch, 2003), clause identification (Tjong Kim Sang, 2001) and most relevantly, grammatical relation finding (Buchholz, 2002). As testing all possible distance metrics in combination with different values for k is not feasible, we have limited the experiment to the Overlap and Modified Value Difference (MVDM) metrics. The values for k tested each metric were 1, 3, 5, 7, and 9. Even values were omitted in order to avoid ties. The Overlap metric computes the distance between two instances by adding up the differences between the featur...