Neural Story Planning

TL;DR

The paper tackles the challenge of generating causally coherent story plots by unifying partial-order causal-link planning with neural language models in an open-world, ending-guided framework. It backward-chains from a given ending, using a GPT-J-6B-based module to infer preconditions and events and to establish explicit causal links that form a directed plan. The approach, enhanced by PMI_DC rescoring and careful precondition/event management, produces plots with higher coherence as measured by enablement-question performance against ROCStories baselines. This work demonstrates that large language models can serve as a commonsense knowledge base for planning, yielding more controllable and interpretable narrative skeletons than purely stochastic generation. Overall, it advances open-world narrative generation by providing a transparent planning backbone that guides neural generation toward coherent, ending-aware plots.

Abstract

Automated plot generation is the challenge of generating a sequence of events that will be perceived by readers as the plot of a coherent story. Traditional symbolic planners plan a story from a goal state and guarantee logical causal plot coherence but rely on a library of hand-crafted actions with their preconditions and effects. This closed world setting limits the length and diversity of what symbolic planners can generate. On the other hand, pre-trained neural language models can generate stories with great diversity, while being generally incapable of ending a story in a specified manner and can have trouble maintaining coherence. In this paper, we present an approach to story plot generation that unifies causal planning with neural language models. We propose to use commonsense knowledge extracted from large language models to recursively expand a story plot in a backward chaining fashion. Specifically, our system infers the preconditions for events in the story and then events that will cause those conditions to become true. We performed automatic evaluation to measure narrative coherence as indicated by the ability to answer questions about whether different events in the story are causally related to other events. Results indicate that our proposed method produces more coherent plotlines than several strong baselines.

Figures (1)

• Figure 1: Visualization of a partial order plan for a plotline. The final event, which is given as input, is green. Events are blue. Preconditions are white and red; red preconditions are those that matched against given initial conditions. Arcs from events to preconditions are causal links, indicating that the event makes the condition true. Arcs from precondition to event indicate that the former is a necessary condition of the event. The plan is generated from right to left, starting with the final event. The plan can be totally-ordered, as shown in Table \ref{['tab:topological']}.