A two-step approach for event factuality identification

This paper focuses on identifying event factuality. Different from pervious rule-based method, this paper proposes a novel two-step approach of combining machine learning and rule-based approach. Firstly, a maximum entropy model is constructed to determine whether the informant's degree of certainty of events is expressed. Then, a set of rules containing cue and scope detection is introduced to further identify various event factuality values. Experimental results manifest that our two-step approach achieves a higher performance than that of the state-of-the-art rule-based system.