A Study on the Sources Identification in the Dependency Structure Annotation for Modality in Chinese News Articles

Abstract

This paper discusses source identification in the dependency structure annotation for modality in Chinese news articles. We updated the annotation scheme and decomposed the modal dependency structure annotation into four subtasks: event identification, source identification, modal superstructure construction and modal strength annotation. One-hundred-thirty documents containing 544 sources have been annotated by three annotators with high levels of agreement. The disagreements that arose between annotators in the past include mislabeling, missing labeling, labeling scope and sources that are not easy to determine. Given the four types of issues, this paper analyzes and concludes the main factors that affect the identification of sources, including the type of sources, the location of sources, and the type of source introducing predicates.