A Study On Mongolian Statistical Parsing

Statistical parsing has been a research focus in parsing fields. In recentyears, researchers made a lot of achievements in Mongolian parsing research.But there is a distance comparing with the development of Chinese andEnglish. It mainly reflected in two aspects, in the aspects of Treebank,comparing with English and Chinese, Mongolian Treebank’s scale is not verylarge. In parsing aspect, the development of Mongolian parsing is in theprimary stage.Mongolian is a language of rich morphology, its word form basedon the stem with connecting suffix. One word has different meaning afterconnecting different kinds of cases. We could improve the accuracy ofMongolian PCFG based parser by using Mongolian case.This paper has done the following research efforts. First, wedeveloped a Mongolian phrase structure grammar tagset. Second,According to the tagset, we built a Mongolian Treebank that contains3645sentences, and developed a Mongolian Treebank auxilia ryprocessing system, which including auxiliary tagging function,proofreading function, statistical function, acquiring rules function andacquiring probability of rules function. Third, we develope d a Mongolianstatistical parser by referring domestic and foreign relevant studies; thenimproved the probabilistic context-free grammar based Mongolian parserby recognizing Mongolian case. Experimental results show that underclosed test, precision and recall of the improved PCFG based Parser is65.1041%and65.5000%. Under open test, precision and recall is61.2903%and64.0000%. Compared with PCFG based parser, precisionand recall has improved5.7461%and10%under closed test, precision and recall has improved4.6059%and10.50%under open test. Thisshowed the improved Mongolian parser got better performance.