ChatGPT as a Narrative Structure Interpreter

Abstract

Narrative structures define the skeleton of narratives and help at identifying common structures in stories, that then can be used to compare structures, define variations, and understand prototypical [structural] components. However, narrative structures are just one piece of the puzzle, their interpretation is what gives room to the stories seen in transmedia storytelling. In principle, a structure can be interpreted and developed with a myriad of stories, but requires some type of corpus to develop it further. Large language models such as ChatGPT could be employed for this task, if we are able to define a good narrative structure and give the tools to the algorithm to develop them further. For this paper, we use a narrative structure system called TropeTwist, which employs interconnected tropes as narrative structures, defining characters, conflicts, and plot devices in a relational graph, which gives raise to a set of trope micro- and meso-patterns. Using ChatGPT and through the web interface, we communicate all the possible elements to be used from TropeTwist and tasked ChatGPT to interpret them and generate stories. We describe our process and methodology to reach these interpretations, and present some of the generated stories based on a constructed narrative structure. Our results show the possibilities and limitations of using these systems and elaborate on future work to combine large language models for other tasks within narrative interpretation and generation.