DOC: Improving Long Story Coherence With Detailed Outline Control

By Kevin Yang, Daniel Klein, Nanyun Peng, Yuandong Tian. 2023

141 citations recorded by the source corpus.

We propose the Detailed Outline Control (DOC) framework for improving long-range plot coherence when automatically generating several-thousand-word-long stories. DOC consists of two complementary components: a detailed outliner and a detailed controller. The detailed outliner creates a more detailed, hierarchically structured outline, shifting creative burden from the main drafting procedure to the planning stage. The detailed controller ensures the more detailed outline is still respected during generation by controlling story passages to align with outline details. In human evaluations of automatically generated stories, DOC substantially outperforms a strong Re3 baseline (Yang et al., 2022) on plot coherence (22.5% absolute gain), outline relevance (28.2%), and interestingness (20.7%). Humans also judged DOC to be much more controllable in an interactive generation setting.

Subfields: Ocean Engineering, Large Language Models, Natural Language Generation

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Ocean Engineering

1. Ruslan Salakhutdinov
2. Andrew Gordon Wilson
3. Yann LeCun
4. Ari S. Morcos

In Large Language Models

1. Avi Schwarzschild
2. Micah Goldblum
3. Ruslan Salakhutdinov
4. Andrew Gordon Wilson
5. Yann LeCun
6. Tom Goldstein
7. Hamed Pirsiavash
8. Jing Yu Koh

In Natural Language Generation

1. Ruslan Salakhutdinov

Citations in the corpus

• Unleashing the potential of prompt engineering in Large Language Models: a comprehensive review (2023, background): “For another example, Detailed Outline Control (DOC) [120], aimed at preserving plot coherence across extensive texts generated with the assistance of GPT-3.”
• A Cookbook of Self-Supervised Learning (2023, methodology): “Similar approaches are also used in guided language generation [Yang et al., 2023].”
• Generating Images with Multimodal Language Models (2023, background): “When trained at large enough scale, LLMs exhibit compelling properties, such as the ability to learn from few-shot in-context examples [9, 11] and generate and process long text inputs [61, 59, 53, 7].”
• Deep Literature Survey Automation with an Iterative Workflow (2025, unknown): “Second, the survey structure can be incoherent (Fabbri et al., 2019; Gidiotis & Tsoumakas, 2020; Yang et al., 2023a).”