This paper discusses successes and failures of computational linguistics techniques in the study of how inter-event time intervals in a story affect the narrator's use of different types of referring expressions. The success story shows that a conditional frequency distribution analysis of proper nouns and pronouns yields results that are consistent with our previous results - based on manual coding - that the narrator's choice of referring expression depends on the amount of time that elapsed between events in a story. Unfortunately, the less successful story indicates that state-of-the-art coreference resolution systems fail to achieve high accuracy for this genre of discourse. Fine-grained analyses of these failures provide insight into the limitations of current coreference resolution systems, and ways of improving them.
展开▼
