Getting a machine to understand humannarratives has been a classic challenge forNLP and AI. This paper proposes a newrepresentation for the temporal structure ofnarratives. The representation is parsimonious,using temporal relations as surrogates fordiscourse relations. The narrative models,called Temporal Discourse Models, are treestructured,where nodes include abstractevents interpreted as pairs of time points andwhere the dominance relation is expressed bytemporal inclusion. Annotation examples andchallenges are discussed, along with a reporton progress to date in creating annotatedcorpora.
展开▼