Semantic Role Labeling Based on Maximum Entropy Model

A method based on maximum entropy model is proposed for Semantic Role Labeling (SRL) . This method takes shallow syntactic parsing as base, and takes phrase or named entity as the labeled units, and maximum entropy model is trained to label the predicates’ semantic roles in a sentence. The key of the method is parameter estimation and feature selection for maximum entropy model. In this paper, the IIS algorithm was employed for parameter estimation, and four categories features: features based on sentence constituents, features based on predicate, predicate-constituent features and semantic features as features set of the model were selected. The method is used to label semantic roles in an event mention sentence for information extraction. We got F=76.3% and F=72.2% results on different test set for “management succession” and “meeting”.