Infusing Commonsense World Models with Graph Knowledge

TL;DR

This paper introduces a method to improve narrative consistency in language models for text adventure games by integrating graph-based world knowledge, enhancing grounded understanding and action narration.

Contribution

The paper presents a novel approach that combines graph representations with language models to improve consistency in dynamic world narratives, along with a new dataset and tasks.

Findings

Training on graph contexts improves narration consistency.

Models perform well even without graphs at test time.

Human evaluations confirm improved narrative quality.

Abstract

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.