Reproduce “A Skeleton-Based Model for Promoting Coherence Among Sentences in Narrative Story Generation” 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
Jingjing Xu, Xuancheng Ren, Yi Zhang, Qi Zeng, Xiaoyan Cai, Xu Sun
Researchers from Peking University and Northwestern Polytechnical University developed a skeleton-based model that utilizes reinforcement learning to automatically learn adaptive intermediate "skeletons" for promoting coherence in narrative story generation. This approach significantly improved human-evaluated coherence and overall story quality, achieving a 20.1% increase in G-score over baselines.
Key insights
- Mimicking the human writing process, where core ideas are formed before full sentences, can enhance machine-generated narrative coherence.
- Adaptive intermediate representations ("skeletons") learned automatically via reinforcement learning are more effective for promoting coherence than rigid, manually defined rules.
- Reinforcement learning, with a carefully designed reward function, can successfully optimize the learning of discrete, high-level structural properties in text generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.