Show, Don't Tell: Uncovering Implicit Character Portrayal using LLMs

Tools for understanding character portrayal in literary fiction are valuable for writers and literary scholars to develop better story drafts and characters. Existing tools, such as visualization tools for analysing fictional characters, primarily rely on explicit indicators of character portrayal. However, portrayal is often implicit, revealed through actions and behaviours rather than explicit statements. We address this gap with a new approach, powered by large-language models (LLMs) to uncover and evaluate implicit character portrayals in narrative texts. Our approach uses LLMs for dataset curation and to evaluate existing implicit portrayal detection pipelines. We also design prompts for using LLMs to uncover implicit portrayal. We therefore aim to provide a more holistic view of character analysis in creative writing. This approach has the potential to empower writers in crafting more complex characters and to contribute to our understanding of how implicit portrayal biases may manifest in literature.