Literary Reasoning in Large Language Models

<a href="https://embed.stanford.edu/iframe/?url=https%3A%2F%2Fpurl.stanford.edu%2Fyw452qc5753" class="su-underline">Show Content</a>

Abstract/Contents

Abstract: Humans are capable of recognizing similarities not only between single concepts but also between intricate real-world situations that involve multiple entities and relationships. Such analogical reasoning has been considered the hallmark of human cognition. Humans are also able to draw affective inferences from such real-world situations, such as which situations inspire attention and empathy, and to what degree. Large Language Models (LLMs) have shown a variety of human-like content effects in logical reasoning and human-like performance in analogical reasoning in zero-shot and few-shot conditions. This study examines the interaction of LLMs with narratives, with a focus on inferred affect, fictionality, and embedding. In particular, I investigate LLMs’ estimations of affective impact and empathy in response to narrative stimuli presented in one of four different conditions: a fictional story with no embedding, a fictional story with embedding, a non-fictional story with no embedding, and a non-fictional story with embedding. This study builds upon prior work in narrative psychology and natural language understanding, proposing that LLMs as few-shot and zero-shot learners can differentiate the affective impact of narratives presented in different ways, and demonstrate literary reasoning, in particular, the abilities to 1) draw affective inferences from a story and 2) generate stories analogous to the story in question in mood, form and content.

Author	Koul, Radhika

DOI	https://doi.org/10.25740/yw452qc5753
Location	https://purl.stanford.edu/yw452qc5753

Use and reproduction: User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor.
License: This work is licensed under a Creative Commons Zero v1.0 Universal license (CC0).

Preferred citation: Koul, R. (2023). Literary Reasoning in Large Language Models. Stanford Digital Repository. Available at https://purl.stanford.edu/yw452qc5753. https://doi.org/10.25740/yw452qc5753.

Master's Theses, Symbolic Systems Program, Stanford University

Loading usage metrics...