Event-Based and Graph-Based Approaches to Skeleton Selection for Goal-Driven Storytelling

Anonymous

Abstract: This study addresses the task of selecting a skeleton for narrative story generation from closely-associated Story Plan Graphs (SPGs). While advanced language models, such as Large Language Models (LLMs), demonstrate potential, they often fall short in manifesting semantic consistency. Utilizing the SPGs generated by Neural Story Planning, which ensures the logical soundness of symbolic planning, we introduce two novel methodologies for skeleton selection: event-based and graph-based approaches. These methods discern salience events within the fabula, helpful the selection of skeletons that are engaging, coherent, and logically consistent. Evaluated against the GPT-3.5 using the ROCStories dataset, our approach evidences enhanced skeleton selection capabilities, offering an efficient and cost-effective solution for skeleton selection.

Paper Type: long

Research Area: Generation

Contribution Types: NLP engineering experiment, Approaches low compute settings-efficiency

Languages Studied: english

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.