Research On Annotation And Extraction Of Chinese Event Factuality

Event factuality is understood here as the level of information expressing the factualnature of event mentioned in the text. It plays an important role in understanding discoursesemantics. At present, most researches focus on event factuality and certainty informationcorpus in English and no literature concerns Chinese. Therefore, the research on Chineseevent factuality is beneficial to many natural language understanding applications.This dissertation studies on the representation, annotation and extraction method ofChinese event factuality as follows:Firstly, this dissertation proposes five kinds of Chinese event factual information fromtwo aspects, vocabulary and sentence structure, according to the characteristics of Chineselanguage, and annotates the Chinese event factual information based on the ACE2005Chinese Corpus.Secondly, it proposes a3D representation of Chinese event factuality based on theannotated factual information. The3D representation breaks Chinese event factuality into atriple including polarity, degree and tense. It also presents the transformation rules betweenthe factual information and3D representation and those between the3D representation andevent factuality. The experimental results demonstrate that our3D representation cansignificantly improve the performance of event factuality analysis.Finally, it takes event selecting predicate as an example, propose a supervised factualinformation extraction approach using effective syntactic features and a semi-supervisedfactual related information annotation approach based on the co-training algorithm. Theformer borrows ideas from the extraction method of English hedge cues and uses BoW(Bag of Words) and syntactic features to extract Chinese event selecting predicates. Thelater takes advantage of the classifier view and pattern view to form a co-training strategy to annotate unlabeled samples based on a small set of annotated samples. Experimentsshow these two methods have achieved good results.This dissertation focuses on Chinese event factuality and its results, i.e. the Chineseevent factuality corpus and the extraction methods, will boost the researches oninformation factuality of Chinese language.