PRISM: A Framework for Authoring Interactive Narratives

Yun-Gyung Cheong, Yeo-Jin Kim, Wookhee Min, Eok-Soo Shim, Jin-Young KimICIDS 2008 · DOI

TL;DR

This paper introduces PRISM, a novel framework for authoring interactive narratives that merges traditional branching story structures with AI planning techniques. This hybrid approach aims to simplify the creation of complex, dynamic stories for both novice and expert users, enabling more engaging and personalized narrative experiences.

What Problem Does This Paper Address?

Creating compelling interactive narratives is a significant challenge. While advances in computing allow users to create and share stories, authoring systems that adapt dynamically to user interaction remain complex and laborious. Existing approaches, primarily branching narratives (like directed graphs) and AI planning, each have limitations. Branching narratives are intuitive but can lead to an explosion of conditional branches and limit interactivity to pre-anticipated user actions. AI planning, on the other hand, can generate more dynamic stories but often results in less immersive experiences for the user. This paper addresses the gap of effectively combining these two powerful approaches to create a more robust and user-friendly authoring system.

Methods

PRISM employs a hybrid approach to authoring interactive narratives. It utilizes a branching narrative structure that is enhanced with planning annotations.

  • Branching Narrative Structure: This forms the core of the story map, represented as a directed graph where nodes are "situations" and arcs represent user choices or transitions. This provides an intuitive way for authors to visualize and construct the story flow.
  • Planning Formalism: To overcome the limitations of purely branching narratives, PRISM integrates AI planning. "Situations" in the story graph can have associated preconditions (conditions that must be met for the situation to occur) and effects (states that change as a result of the situation). When there are no explicit branches defined by the author, a planner can automatically create links between situations by finding those whose preconditions are met.
  • Condition Data Structure: A unified data structure called "condition" is used to represent user interactions (text, speech, gesture), external information (web data, GPS), and virtual environment states. This structure serves as both a branching decision point in the graph and as preconditions/effects for planning.
  • TVML (TV Program Making Language): For content creation and visualization, PRISM uses TVML, a text-based language that describes all elements needed for 3D animation (stage, camera, characters, etc.). This allows for the realization of stories into 3D animations across various platforms.

The system architecture includes a PRISM Writer (for authoring) and a PRISM Player (for presenting the story). The writer uses a graphical interface to build story maps, and the player decodes the PRISM container (an XML document representing the story structure) to render the narrative.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The Synergy of Branching and Planning: The core insight is that combining the intuitive, author-driven nature of branching narratives with the dynamic, adaptive capabilities of AI planning creates a more powerful authoring tool. This hybrid approach allows authors to guide the story explicitly where they want, while letting AI handle more complex or emergent situations.
  • Prioritizing Author Intent: PRISM prioritizes the author's explicit branching choices over AI-generated plans when conflicts arise, ensuring that the author's direct intentions are always respected. AI planning is primarily used when no author-defined branches are available.
  • Simplifying Authoring for Non-Experts: By providing a graphical interface and integrating planning, PRISM aims to lower the barrier to entry for creating interactive stories, making it accessible to users without deep AI or programming expertise.
  • Ubiquitous Storytelling with TVML: The use of TVML for story realization enables stories to be played on a wide range of devices, promoting a more ubiquitous interactive storytelling experience.

Key Results

The paper describes the framework and its components, focusing on the architecture and methodology rather than presenting quantitative results from user studies or evaluations. The key "results" are the successful design and conceptualization of the PRISM framework itself, which integrates branching narratives with AI planning. The authors state that the content authoring tool and their TVML player have been implemented, and the XML schema for representing interactive story structures has been designed. The PRISM player and writer are under development.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that while planning can generate stories, the system might require heuristic functions to ensure "dramatic experiences" for the user, as purely precondition-based selection might not be sufficient.
  • The current planning approach is described as a "simple forward planning at a depth of one," which might be a limitation for more complex narrative structures.
  • The system is still under development, with the full PRISM player and writer yet to be implemented.
  • Future Work:
    • Implementing the full PRISM player and writer.
  • Developing intelligent assistance for professional tasks like virtual camera control.
  • Exploring statistical user models, such as Bayesian networks, to further personalize stories.
  • Integrating novel user input devices (e.g., neuroheadsets) to enhance immersive experiences.
  • Formal evaluation of the system.

Summary updated

BibTeX
@inproceedings{cheong2008prism,
  author = {Yun-Gyung Cheong and Yeo-Jin Kim and Wookhee Min and Eok-Soo Shim and Jin-Young Kim},
  title = {{PRISM: A Framework for Authoring Interactive Narratives}},
  booktitle = {ICIDS},
  year = {2008},
  doi = {10.1007/978-3-540-89454-4\_37},
}
Figure