Automatic identifying of maximal length noun phrase

The automatic recognition of the maximal-length noun phrase (MNP) helps to the shallow parsing. In this paper, automatic labeling of Chinese MNP is regarded as a sequential labeling task and Support Vector Machine model (SVM) is employed in the model. We propose a method which takes 2-phase hybrid approach which first identifies base chunk and then identifies MNP. Furthermore, the base chunk features can be exploited to improve performance of MNP recognition. In addition, both left-right and right-left sequential labeling were employed to identify Chinese MNP by bidirectional sequence labeling merging. The data set in the experiments is selected from Penn Chinese Treebank 5.0 Corpus, and split into train set, development set and test set according to the proportion of 4:4:1. Experimental result shows a high quality performance of 90.13% in F1-measure.