Using word sense disambiguation for semantic role labeling

Wanxiang Che,Ting Liu
DOI: https://doi.org/10.1109/IUCS.2010.5666646
2010-01-01
Abstract:Word sense disambiguation (WSD) is the process of identifying the correct meaning, or sense of a word in a given context. Semantic role labeling (SRL) aims at identifying the relations between predicates in a sentence and their associated arguments. They are two fundamental tasks in natural language processing to find a sentence-level semantic representation. To date, they have mostly been modeled in isolation. However, this approach neglects logical constraints between them. In this work, we present some novel word sense features for SRL and find that they can improve the performance significantly. Later, we exploit pipeline strategies which verify the automatic all word sense disambiguation could help the semantic role labeling and vice versa. We further propose a Markov logic model that jointly labels semantic roles and disambiguates all word senses. We show that this joint approach leads to a higher performance for WSD and SRL than those pipeline approaches. ©2010 IEEE.
What problem does this paper attempt to address?