Build a Large-Scale Syntactically Annotated Chinese Corpus

This paper reports on our research to build a large-scale Tsinghua Chinese Treebank (TCT). We propose a two-stage approach to reduce manual proofreading labors as much as possible. The insertion of an intermediate functional chunk level creates a good information bridge to link simple chunk annotation with detailed syntactic tree annotation. We describe our chunk and tree annotation schemes, focus on two grammatical relation tag sets designed to give more detailed description for most of the special language phenomena in the Chinese language. We also briefly introduce our current progress in building a Chinese chunk bank with 2,000,000 Chinese characters, developing an efficient Chinese chunk-based parser and building a 1,000,000 words Chinese treebank. All this work lays good foundations for further research project to build a good Chinese parser.