Infusing Commonsense World Models with Graph Knowledge
This work addresses the challenge of improving narrative consistency in language models for interactive storytelling applications, representing an incremental advancement in domain-specific methods.
The paper tackles the problem of maintaining consistency in language models when generating narratives for dynamically changing worlds, specifically in open-world text adventure games, and finds that training models on graph representations improves consistency in action narration, as shown by both automatic metrics and human evaluations.
While language models have become more capable of producing compelling language, we find there are still gaps in maintaining consistency, especially when describing events in a dynamically changing world. We study the setting of generating narratives in an open world text adventure game, where a graph representation of the underlying game state can be used to train models that consume and output both grounded graph representations and natural language descriptions and actions. We build a large set of tasks by combining crowdsourced and simulated gameplays with a novel dataset of complex actions in order to to construct such models. We find it is possible to improve the consistency of action narration models by training on graph contexts and targets, even if graphs are not present at test time. This is shown both in automatic metrics and human evaluations. We plan to release our code, the new set of tasks, and best performing models.