Research On Chinese Negation And Speculation Identification

Negative and speculative expressions are popular in natural language. Negation andspeculation identification has become an important task in Information Extraction. Mostprevious researches focus on English negation and speculation identification. However,there are no publicly researches on Chinese. Lack of Chinese corpus hiders thedevelopment of negation and speculation identification. Besides, there exists somedifference between Chinese and English, such as in grammar, so it is necessary topropose the specific model for Chinese negation and speculation identification.In this paper, we focus on the construction of Chinese negation and speculationcorpus and the identification of Chinese negative and speculative cue and scope. Themain content includes the following three parts:First, we design guidelines for Chinese negation and speculation annotation. ChineseJournals of Computer and hotel reviews from Ctrip website are chose to annotate negativeand speculative cue and scope. Then, we make statistics and an analysis on the corpus.This corpus provides the source for research on Chinese negation and speculationidentification.Second, we implement baseline system under the character-based and word-basedframework. The statistics and baseline system results indicate the ambiguity is the mainquestion of cue identification. In order to solve this question, we propose a new modelcombining CRF-based (conditional random fields) and probability statistics method forChinese negative and speculative cue. The experimental results show this model enables agreat improvement on the identification of Chinese negative and speculative cue.Third, we put forward Chinese-oriented features and combinated features to identifyChinese negative and speculative scope while learning word-based features, syntactic features and cue-concerned features used in English. Then, we propose a multi-classifiersmodel to identify Chinese negative and speculative scope. The experimental results showthat our model improves the performance of the identification of Chinese negative andspeculative scope.This paper presents an annotation and identification method for Chinese negation andspeculation. It is conducive the development of Chinese negation and speculationidentification, and provides the service for applications based on semantic knowledge suchas Natural Language Understanding.