Semantic Role Labeling With Markov Logic Networks

Recently, with the development of machine learning, more attention has been paid to the Semantic Parsing of sentence. Semantic Role Labeling(SRL) is a feasible proposal of Shallow Semantic Parsing, and it is valuable for many applications, such as Question&Answering and Information Extraction.The purpose of the paper is to integrate Word Sense Disambiguation and Semantic Role Labeling as one task. In the CoNLL2009 Shared Task, Word Sense Disambiguation for predicate is a subtask of Semantic Role Labeling. After automatically determining the word sense disambiguation of predicate, the best result was selected as the input for Semantic Role Labeling. This pipeline method cannot get the global optimal solution. In this paper, Word Sense Disambiguation and Semantic Role Labeling are jointed as one task with Markov Logic Networks (MLNs). They can help each other in the prediction stage, and a global optimal solution is got with this jointly model.This paper is divided into two stages. The first stage is to verify the information of all word sense on the Semantic Role Labeling. Predicate sense on Semantic Role Labeling is useful, but whether all word sense information on Semantic Role Labeling is useful has been inconclusive. The second stage is to joint the Word Sense Disambiguation and Semantic Role Labeling with Markov Logic Networks based on that all word sense is useful for Semantic Role Labeling, and analyze the experimental results.