Learning to tell stories from the crowd (updated)
Automatically Learning to Tell Stories about Social Situations from the Crowd, accepted to the LREC 2012 Workshop on Computational Models of Narrative. Story generators to date have been dependent on an author-defined domain (i.e. micro world). These micro worlds are cumbersome to build, especially when involving highly specific social and cultural phenomena, limiting the applicability of story generation in general. Read more below the break...
However, computational systems can now live in the complex information ecosystem of the Internet: rich sources of knowledge online and human knowledge obtained through crowd sourcing. Specifically, we investigate whether a computational system can learn to tell stories by querying script-like knowledge structures from a crowd. We have discovered that we can instruct a crowd to give narrative examples from which to extract general understanding. The result is a "plot graph" that describes the space of possible stories about a given topic.
Below are two plot graphs that our system learned: how to go on a date to a movie theatre (left), and how to go to a fast-food restaurant (right). Nodes are events and arcs are temporal ordering constraints that approximate causal relations.