Large Language Models as Narrative Planning Search Guides

Abstract

Symbolic planning algorithms and large language models have different strengths and weaknesses for story generation, suggesting hybrid models might leverage advantages from both. Others have proposed using a language model in combination with a partial order planning style algorithm to avoid the need for a hand-written symbolic domain of actions, or generating these domains from natural language input. This article offers a complementary approach. We propose to use a state space planning algorithm to plan coherent multiagent stories using hand-written symbolic domains, but with a language model acting as a guide to estimate, which events are worth exploring first. We present an initial evaluation of this approach on a set of benchmark narrative planning problems.