Chinese Semantic Role Labeling Based on Conditional Random Fields

In the paper, we choose Chinese propbank (CPB) as the experimental data to do the semantic role labeling task. In Chinese propbank, there are many sentences which contain multiple verbs, but how to deal with these sentences is seldom referred. In our experiment, we propose a method to deal with these sentences and add some new features. In order to get a feature, we take two-phase classification: binary-classification and multiple-classification. At last, we get a promising result: Precision: 80.40% and F-score: 79.92. From the last results, we can see that this method is feasible.