A probabilistic graphic model for recognizing np chunks from texts

Abstract. We present a probabilistic graphical model for identifying noun phrase patterns in texts. This model is derived from mathemat-ical processes under two reasonable conditional independence assump-tions with different perspectives compared with other graphical models, such as CRFs or MEMMs. Empirical results shown our model is effec-tive. Experiments on WSJ data from the Penn Treebank, our method achieves an average of precision 97.7 % and an average of recall 98.7%. Further experiments on the CoNLL-2000 shared task data set show our method achieves the best performance compared to competing meth-ods that other researchers have published on this data set. Our average precision is 95.15 % and an average recall is 96.05%. Key words: NP chunking, graphical models, cliques, separators