Search papers, labs, and topics across Lattice.
This paper introduces a narratology-based framework, extending Propp's theory with 34 narrative functions tailored for Chinese web literature, to analyze narrative structures in LLM-generated text. They construct a human-annotated corpus and use it to analyze stories generated by LLMs. The analysis reveals that LLMs exhibit a singular narrative logic and homogenization due to their inability to comprehend narrative functions, instead relying on rigid generation paradigms.
LLMs' creative writing may be more imitation than innovation, as they struggle to grasp narrative function and instead produce homogenized stories by adhering to rigid, stereotypical plot structures.
Large Language Models (LLMs) have demonstrated remarkable capabilities in narrative generation. However, they often produce structurally homogenized stories, frequently following repetitive arrangements and combinations of plot events along with stereotypical resolutions. In this paper, we propose a novel theoretical framework for analysis by incorporating Proppian narratology and narrative functions. This framework is used to analyze the composition of narrative texts generated by LLMs to uncover their underlying narrative logic. Taking Chinese web literature as our research focus, we extend Propp's narrative theory, defining 34 narrative functions suited to modern web narrative structures. We further construct a human-annotated corpus to support the analysis of narrative structures within LLM-generated text. Experiments reveal that the primary reasons for the singular narrative logic and severe homogenization in generated texts are that current LLMs are unable to correctly comprehend the meanings of narrative functions and instead adhere to rigid narrative generation paradigms.
