Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

What is Wrong with Language Models that Can Not Tell a Story?

Ivan P. Yamshchikov, Alexey Tikhonov

TL;DR

This position paper argues that narrative processing is a fundamental bottleneck in NLP and AI, essential for long-form, engaging text generation and for enabling causal reasoning and human–machine interaction. It analyzes three core bottlenecks—data, evaluation, and concepts—claiming that while data on narratives is growing, long-form, diverse, and multilingual resources remain scarce, and that evaluation frameworks and narrative representations are underdeveloped. Through a critique of existing datasets and evaluation practices, the authors demonstrate the lack of robust, automatable metrics for coherence and novelty and the absence of a unified theory of storytelling. They advocate elevating narrative processing as a core sub-field within NLP/AI, urging the community to develop shared benchmarks, theoretical foundations, and multilingual resources to advance narrative generation.

Abstract

This paper argues that a deeper understanding of narrative and the successful generation of longer subjectively interesting texts is a vital bottleneck that hinders the progress in modern Natural Language Processing (NLP) and may even be in the whole field of Artificial Intelligence. We demonstrate that there are no adequate datasets, evaluation methods, and even operational concepts that could be used to start working on narrative processing.

What is Wrong with Language Models that Can Not Tell a Story?

TL;DR

This position paper argues that narrative processing is a fundamental bottleneck in NLP and AI, essential for long-form, engaging text generation and for enabling causal reasoning and human–machine interaction. It analyzes three core bottlenecks—data, evaluation, and concepts—claiming that while data on narratives is growing, long-form, diverse, and multilingual resources remain scarce, and that evaluation frameworks and narrative representations are underdeveloped. Through a critique of existing datasets and evaluation practices, the authors demonstrate the lack of robust, automatable metrics for coherence and novelty and the absence of a unified theory of storytelling. They advocate elevating narrative processing as a core sub-field within NLP/AI, urging the community to develop shared benchmarks, theoretical foundations, and multilingual resources to advance narrative generation.

Abstract

This paper argues that a deeper understanding of narrative and the successful generation of longer subjectively interesting texts is a vital bottleneck that hinders the progress in modern Natural Language Processing (NLP) and may even be in the whole field of Artificial Intelligence. We demonstrate that there are no adequate datasets, evaluation methods, and even operational concepts that could be used to start working on narrative processing.
Paper Structure (7 sections, 1 figure)

This paper contains 7 sections, 1 figure.

Table of Contents

  1. Introduction
  2. On the Importance of Narrative
  3. Where Do We Fail?
  4. Data
  5. Evaluation
  6. Concepts
  7. Conclusion

Figures (1)

  • Figure 1: "Copyrighted protein fiction may be deemed speculative propaganda" — a line from a generative art project "Paranoid Transformer — a diary of an artificial neural network", agafonova2020paranoid. The diary was generated end-to-end without any human post-processing and published as a hardcover book.