The paper "Hierarchical Neural Story Generation" by Angela Fan, Mike Lewis, and Yann Dauphin explores the challenge of generating coherent and fluent stories from writing prompts. They collect a large dataset of 300K human-written stories paired with prompts from an online forum, enabling hierarchical story generation where the model first generates a premise and then transforms it into a story. The authors introduce a novel fusion mechanism to improve the relevance of the generated story to the prompt and a gated multi-scale self-attention mechanism to model long-range context. Experiments show significant improvements over strong baselines on both automated and human evaluations, with human judges preferring stories generated by their approach twice as often as those from a non-hierarchical baseline. The paper also discusses the challenges of modeling long-range dependencies and the benefits of hierarchical generation, highlighting the effectiveness of their proposed methods in generating creative and fluent text.The paper "Hierarchical Neural Story Generation" by Angela Fan, Mike Lewis, and Yann Dauphin explores the challenge of generating coherent and fluent stories from writing prompts. They collect a large dataset of 300K human-written stories paired with prompts from an online forum, enabling hierarchical story generation where the model first generates a premise and then transforms it into a story. The authors introduce a novel fusion mechanism to improve the relevance of the generated story to the prompt and a gated multi-scale self-attention mechanism to model long-range context. Experiments show significant improvements over strong baselines on both automated and human evaluations, with human judges preferring stories generated by their approach twice as often as those from a non-hierarchical baseline. The paper also discusses the challenges of modeling long-range dependencies and the benefits of hierarchical generation, highlighting the effectiveness of their proposed methods in generating creative and fluent text.