Abstract: Narrative is central to how humans reason and make sense of their experiences and communicate to each other. There are key narrative elements that contribute to a good story such as plot, characters, their emotions and theme of the story. Understanding, and generating narratives have long been a challenging AI problem. In this proposal, we provide methods for automatically generating narratives by modeling different narrative elements.
While previous approaches in automatic storytelling focused mostly on one-shot generation, we instead propose to take story keywords as narrative elements which are provided by a user and enables building the plot of the story interactively. We present two content-inducing approaches to incorporate this additional information.
Moreover, different from objective texts such as news articles, what makes stories interesting and engaging to readers is the fact that most of them depict evolving characteristics such as changing of emotions in their plot as they progress through. Motivated by this, we present the emotional trajectory of the protagonist as a narrative element, and generate stories that follow desired emotion arcs. We propose two Emotion-Consistency rewards designed to enforce the desired emotion arcs using reinforcement learning.
While a good story will have both a plot and characters, a few works focused most primarily on plot development. Lastly, as a future direction, we propose to explore characters as another key narrative element to study the causal relation between character and the plot of the story.