Reproduce “DOC: Improving Long Story Coherence With Detailed Outline Control” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kevin Yang, Dan Klein, Nanyun Peng, Yuandong Tian
Researchers from UC Berkeley, UCLA, and Meta AI developed Detailed Outline Control (DOC), a framework for generating long stories up to 3500 words with improved plot coherence. Human evaluations showed DOC achieved 67.6% preference for coherence and 65.3% for outline relevance over previous methods, along with a 20.7% absolute gain in interestingness.
Key insights
- Detailed, hierarchical planning significantly improves long-range coherence and relevance in generated stories.
- A strong, dynamically adjusted control mechanism is crucial for strictly enforcing natural language-based outline constraints throughout the generation process.
- Shifting the creative burden to a comprehensive planning phase, akin to human writing processes, enhances overall story quality and facilitates more effective human-AI collaboration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.