Semantic relations between various text units play an important role in natural languageunderstanding, as key elements of text coherence. The automatic identification of thesesemantic relationships is very important for many language processing applications. Oneof the most pervasive yet very challenging semantic relations is cause-effect. In thisthesis, an unsupervised approach to learning both direct and indirect cause-effectrelationships between inter- and intra-sentential events in web news articles is proposed.Causal relationships are leaned and tested on two large text datasets collected by crawlingthe web: one on the Hurricane Katrina, and one on Iraq War. The text collections thusobtained are further automatically split into clusters of connected events using advancedtopic models. Our hypothesis is that events contributing to one particular scenario tend tobe strongly correlated, and thus make good candidates for the causal informationidentification task. Such relationships are identified by generating appropriate candidateevent pairs. Moreover, this system identifies both the Cause and Effect roles in arelationship using a novel metric, the Effect-Control-ratio. In order to evaluate thesystem, we relied on the manipulation theory of causality
【 预 览 】
附件列表
Files
Size
Format
View
An unsupervised approach to identifying causal relations from relevant scenarios