Research Summary: Incipit: Axiom-Grounded Scaffolding for Human-AI Literary Creation

A new research prototype, Incipit, introduces an explicit planning layer to improve Human-AI literary creation. It allows writers to guide Large Language Models (LLMs) more effectively by grounding narrative generation in 'literary axioms' – curated propositions about human experience and narrative craft. This system moves beyond simple prompt-to-text interaction, enabling more detailed control over creative premises, story beats, character arcs, and chapter outlines through a structured dialogue and a comprehensive knowledge base.

Why it matters

This development is significant because it addresses a key limitation in current AI-assisted creative processes: the lack of explicit control over underlying narrative assumptions. By introducing structured planning and axiom-grounded guidance, it allows for more sophisticated and intentional human-AI collaboration in content generation, potentially expanding the scope and quality of AI-assisted creative output across various domains.

Key insights

• Large Language Models (LLMs) can produce fluent prose but lack direct mechanisms for writers to access and influence underlying narrative assumptions.
• Incipit introduces an explicit planning layer that mediates between a writer's intent and AI-generated prose, enhancing control over narrative development.
• The system is grounded in 'literary axioms,' which are curated, reusable propositions about human experience and narrative craft, stored in a knowledge base of 1455 axioms and 472 typed relationships.
• Incipit employs a five-round directional dialogue, a retrieval-and-selection pipeline, and a three-level blueprint to structure creative premises, story beats, character arcs, and chapter outlines.
• The prototype allows writers to inspect and edit the planning layer, offering greater transparency and collaborative interaction with the AI.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.31007