Procedural generation of narrative worlds

A Narrative World typically consists of several interrelated locations that, all together, fully support enacting a given story. For this, each location in a narrative world features all the objects as required there by the narrative, as well as a variety of other objects that plausibly describe or decorate the location. Procedural generation of narrative worlds poses many challenges, including that (i) it cannot lean only on domain knowledge (e.g. patterns of objects commonly found in typical locations), and (ii) it involves a temporal dimension, which introduces dynamic fluctuations of objects between locations. We present a novel approach for the procedural generation of narrative worlds, following two stages: first, a Narrative World Mold is generated (only once) for a given story; second, the narrative world mold is used to create one (or more) possible narrative worlds for that story. For each story, its narrative world mold integrates spatio-temporal descriptions of its locations with the object semantics and the domain knowledge previously acquired on typical locations. We describe how a narrative world mold can be generated, as well as how it can be fed to existing procedural generation methods, to create a variety of narrative worlds that fit that narrative. We evaluate our own implementation of this approach, performing a number of experiments that illustrate both the expressive power of narrative world molds and their ability to steer the generation of narrative worlds.

@Article { BB23,
  author       = "Balint, J. Timothy and Bidarra, Rafael",
  title        = "Procedural generation of narrative worlds",
  journal      = "IEEE Transactions on Games",
  number       = "2",
  volume       = "15",
  pages        = "262--272",
  year         = "2023",
  doi          = "10.1109/TG.2022.3216582",
  note         = "doi 10.1109/TG.2022.3216582",
  url          = "http://graphics.tudelft.nl/Publications-new/2023/BB23"
}