Modelling Protagonist Goals and Desires in First-Person Narrative

摘要

Many genres of natural language text are narratively structured, a testamentto our predilection for organizing our experiences as narratives. There isbroad consensus that understanding a narrative requires identifying andtracking the goals and desires of the characters and their narrative outcomes.However, to date, there has been limited work on computational models for thisproblem. We introduce a new dataset, DesireDB, which includes gold-standardlabels for identifying statements of desire, textual evidence for desirefulfillment, and annotations for whether the stated desire is fulfilled giventhe evidence in the narrative context. We report experiments on tracking desirefulfillment using different methods, and show that LSTM Skip-Thought modelachieves F-measure of 0.7 on our corpus.