Engineers from Georgia Institute of Technology, with US Defense Advanced Research Projects Agency support, have created a story telling AI algorithm Scheherazade that do not rely on “a priori defined domain models about fictional worlds” like characters, their actions, or places, but creates stories “in an unknown domain”.
In the research findings published in the article “Story Generation with Growdsourced Plot Graphs” engineers have presented an open story generator The Scheherazade System:
“In this paper, we explore a new narrative intelligence challenge: open story generation, the ability to autonomously create narratives about any topic without relying on a priori engineered domain models”
The Scheherazade System uses crowdsourcing to acquire from humans examples of linear narrative examples on a chosen topic, based on which it creates a generalized model, or a plot graph:
“By learning plot graphs from the collected narratives, our system can generate a story about any topic for which a crowd of people can generally agree on the main events that should occur, although not necessarily on the specific sequence of events”
When the system choses plot options , it adds only “executable” events, in parallel excluding mutually exclusive events.
To evaluate the quality and coherence of Scheherazade’s stories, researchers invited 360 participants for editing both human and AI-created texts. As a result, Scheherazade’s stories proved to be comparable to human creative writing:
“We find the difference between human-authored and computer-generated stories to be insignificant in many measures”
Scientists concluded that open story generation helps to overcome “knowledge engineering bottlenecks” and thus paths the way for further progress:
“We believe that this is a crucial step toward achieving human level computational narrative intelligence”