Building a Chinese Dependency GraphBank

How to represent the structure of a sentence is a key issue in linguistic and NLP fields. Dependency Grammar (DG) has been widely used as it directly describes the relations between words in a sentence. However, it always follows the tree structure that does not fit the argument sharing phenomenon. On the other hand, the Semantic Role Labeling (SRL) annotation does not give a full structure for a sentence. In this paper, we combine DG and SRL together to build a unified dependency graph for Chinese sentences. Then we introduce the tagging sets for dependency graph as well as the toolkit designed for manual annotation. Annotation problems are discussed at the end.