Search papers, labs, and topics across Lattice.
The paper introduces BiT-MCTS, a novel theme-driven framework for long-form Chinese fiction generation that leverages a "climax-first, bidirectional expansion" strategy. It extracts a core dramatic conflict from a given theme and generates an explicit climax, then employs bidirectional Monte Carlo Tree Search (MCTS) to expand the plot backward and forward. Experiments using three LLM backbones on a newly constructed Chinese theme corpus demonstrate that BiT-MCTS enhances narrative coherence, plot structure, and thematic depth compared to existing methods.
Forget linear outlines: this new method generates longer, more coherent stories by starting with the climax and expanding bidirectionally using Monte Carlo Tree Search.
Generating long-form linear fiction from open-ended themes remains a major challenge for large language models, which frequently fail to guarantee global structure and narrative diversity when using premise-based or linear outlining approaches. We present BiT-MCTS, a theme-driven framework that operationalizes a "climax-first, bidirectional expansion" strategy motivated by Freytag's Pyramid. Given a theme, our method extracts a core dramatic conflict and generates an explicit climax, then employs a bidirectional Monte Carlo Tree Search (MCTS) to expand the plot backward (rising action, exposition) and forward (falling action, resolution) to produce a structured outline. A final generation stage realizes a complete narrative from the refined outline. We construct a Chinese theme corpus for evaluation and conduct extensive experiments across three contemporary LLM backbones. Results show that BiT-MCTS improves narrative coherence, plot structure, and thematic depth relative to strong baselines, while enabling substantially longer, more coherent stories according to automatic metrics and human judgments.
