Reproduce “Plug-and-Play Dramaturge: A Divide-and-Conquer Approach for Iterative Narrative Script Refinement via Collaborative LLM Agents” 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
Wenda Xie, Chao Guo, Yanqing Jing, Junle Wang, Yisheng Lv, Fei-Yue Wang
Dramaturge introduces a hierarchical multi-agent LLM system that iteratively refines long-form narrative scripts using a divide-and-conquer strategy for task and feature decomposition. This approach yielded a 53.2% improvement in overall script quality compared to original scripts and outperformed the strongest LLM baseline by 8.2% in script-level evaluation.
Key insights
- Iterative, multi-granularity refinement with specialized collaborative agents significantly enhances long-form narrative quality beyond single-pass LLM generation.
- A top-down, coarse-to-fine coordination strategy, where high-level directives guide local modifications, is crucial for maintaining global narrative consistency during refinement.
- Structured architectural design and a systematic refinement workflow contribute more to substantial quality improvement than merely increasing the capacity of the underlying language model.
- The system successfully emulates human scriptwriting practices by progressively enhancing coherence, character depth, and scene immersion through iterative evaluation and revision.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.