Scaling Long-Form Story Generation via Narrative State Tracking
The paper claims its training-free agent keeps 100K-word generated stories coherent by tracking narrative state.
NstAgent maintains structured records of characters, past events, and future requirements while an LLM writes. The authors extend a benchmark to test narrative consistency at different lengths, then evaluate stories from 10K to 100K words alongside writing quality. They report that consistency and quality improve relative to baselines as stories get longer, without noticeable degradation from length alone. The paper is under review, with code and data listed as available. ArXiv · AI/CL/LG's note
NstAgent maintains structured records of characters, past events, and future requirements while an LLM writes. The authors extend a benchmark to test narrative consistency at different lengths, then evaluate stories from 10K to 100K words alongside writing quality. They report that consistency and quality improve relative to baselines as stories get longer, without noticeable degradation from length alone. The paper is under review, with code and data listed as available. ArXiv · AI/CL/LG's note
score 4