Narrative Plan Generation with Self-Supervised Learning

Narrative Generation has attracted significant interest as a novel application of Automated Planning techniques. However, the vast amount narrative material available opens way to use Deep Learning In this paper, we explore feasibility generation through self-supervised learning, using sequence embedding techniques or auto-encoders produce sequences. We datasets well-formed plots generated by planning approach, pre-existing, published, domains, train generative models. Our experiments demonstrate ability models with similar structure those obtained techniques, but plot novelty in comparison training set. Most importantly, share structural properties associated quality measures used Planning-based methods. As plan-based structures account for higher level causality and consistency, suggests that our approach is able extend set narratives sequences display same high-level properties. Unlike methods developed sets textual narratives, ours operates at structure. Thus, it potential be across various media complexity, being initially limited operating genre.