A Statistical Approach Based on Boundary POS Feature to Prosodic Phrasing
Zheng Niu · Zhongwen xinxi xuebao · 2001
It is often difficult to construct a rule based parser and adapt it to largescale real text.So we tried a statistical approach to prosodic phrasing.At first the text was segmented into Chinese words,then word sequences are tagged automatically by POS tagger.The boundary pattern and boundary distribution probabilities are used in the algorithm to predict phrase breaks.The boundary distribution probabilities are derived from hand annotated corpus.The errors caused by statistical method are corrected by rules.Through close testing and open testing on about 1000 sentences,the correct POS tagging rate is about 95%,the recalling rate of prosodic phrasing is around 60%,and the correct rate of prosodic phrasing is about 80%.