Combining Data-Driven Constituent and Dependency Parsers for CIPS-ParsEval-2009

Rui Wang, Yi Zhang, Wei Qiu, Mosha Chen, Tingyu Li, Wenbo Zhang, Tianfang Yao

Chinese Information Processing Society 2009.


In this paper, we describe our Chinese parsing system participated in the CIPS-ParsEval-2009 evaluation task 5. Two open source data driven parsing systems are used: constituent tree parsing with Berkeley's Parser, and dependency parsing with MSTParser. The results are combined to produce the phrase structure trees with head indices. Results show that the Berkeley parser achieved close to state-of-the-art performance on constituent tree parsing without head indices. And the combination gives us good partial-head identification accuracy. However the poor performance on full head identification indicates that the combination cannot handle the cases well where multiple heads exist.

Weitere Links

05-Task5-C+O-12-WangRui.pdf (pdf, 188 KB )

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence