Developing Probabilistic Models for Identifying Semantic Patterns in Texts

We present a probabilistic graphical model that finds a sequence of optimal categories for a sequence of input sym-bols. Based on this mode, three algorithms are developed for identifying semantic patterns in texts. They are the al-gorithm for extracting semantic arguments of a verb, the algorithm for classifying the sense of an ambiguous word, and the algorithm for identifying noun phrases from a sen-tence. Experiments conducted on standard data sets show good results. For example, our method achieves an average precision of 92.96 % and an average recall of 94.94 % for extracting semantic argument boundaries of verbs on WSJ data from Penn Treebank and PropBank; an average accu-racy of 81.12 % for recognizing the six sense word ′line′; and an average precision of 97.7 % and an average recall of 98.8 % for recognizing noun phrases on WSJ data from Penn Treebank.