Narrative Pathways: A Player-Driven Approach to Interactive Fiction Text Game Generation
Author(s)
Zhou, Eric
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
We detail the development of Story2Game, an approach to dynamic text game generation through Large Language Models within interactive fiction games. The approach involves a multi-stage process: firstly by writing a story, populating the elements of a representative game world, and code-building actions that emulate the events in the story. Outside of a short and rudimentary list of default actions, there are no constraints on valid actions, removing a major restriction of the diversity of story generation and translation into a game-based setting. Action generation lies at the key of this process, and is built upon identifying key preconditions and effects to connect items and characters through causal effect reasoning, which can then be interpreted into game state changes whenever a user acts. Additionally, we introduce a technique to allow for dynamic action generation upon a player’s desire to perform an action not recognized by the game state. This dynamic action generation process may require novel adjustments to the representation of certain items in the game state and also checks for logical consistency within other relevant actions and objects. Our evaluation hinges on the measurement of accurate and successful generations of user playthroughs of an entire story.
Sponsor
Date
Extent
Resource Type
Text
Resource Subtype
Undergraduate Research Option Thesis