Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Creative Convergence or Imitation? Genre-Specific Homogeneity in LLM-Generated Chinese Literature

Yuanchi Ma, Kaize Shi, Hui He, Zhihua Zhang, Zhongxiang Lei, Ziliang Qiu, Renfen Hu, Jiamou Liu

Abstract

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.

Creative Convergence or Imitation? Genre-Specific Homogeneity in LLM-Generated Chinese Literature

Abstract

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.
Paper Structure (19 sections, 5 figures, 15 tables)

This paper contains 19 sections, 5 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Propp's Narratology Theory in NLP.
  4. Creative writing
  5. Proppian Taxonomy for Chinese Web Fiction
  6. Theoretical Induction Method
  7. Annotated Corpus Construction
  8. Experimental
  9. Experimental Setup
  10. Experimental Result
  11. Case Study
  12. Conclusion
  13. Limitation
  14. Appendix A
  15. Homogeneous region experiment and experts
  16. ...and 4 more sections

Figures (5)

  • Figure 1: The working framework consists of theoretical, dataset and evaluation. The theoretical section presents our two principles for dividing narrative texts, including the fundamental division principle and the new function improvement principle.
  • Figure 2: An overview of the Cinderella. This picture analyzes the entire storyline and the related narrative functions.
  • Figure 3: Narrative function frequency sampling
  • Figure 4: The narrative text generated by Deepseek in Ep.1 contains six narrative functions.
  • Figure 5: Our experimental approach is shown in (a). Based on the pretexts of the novel with a unified theme, different LLMs are used to continue the following text five times. Experts classified the composition of the five continuation results of the same model respectively, breaking down the continuation text into the three elements that make up the novel: environmental description, character development and storyline summary. Then bert-score is used to calculate the similarity among each element to preliminarily evaluate the homogeneity phenomenon of the novel text generated by LLMs. The experimental results are shown in Figure (b).