Empirical Insights Into Short Story Draft Construction

Existing cognitive models of narrative creation provide accounts for story invention that, while useful, are too high level to be directly applied to formal systems like computational models of narrative generation. Inversely, existing automatic story generation systems that try to implement cognitive models can only rely on approximations to the general concepts these models describe. In order to provide insight to fill the gap between these two approaches, we have conducted a study in which human participants would invent and write short stories while reflecting on their thoughts out loud. The sessions and the analysis of the recordings was designed so that we could observe which specific modifications the participants apply to their story drafts, with the intention to inform the process of creating computational systems based on cognitive descriptions of the narrative creation process. After running the experiments, annotating the videos and analysing the output, we have concluded that there are a number of common modifications that humans tend to apply to a newly created draft, and that this information can be used to the development of storytelling systems.