Narrative World Model: Narratology-Grounded Writer Memory for Long-Form Fiction

Jul 16, 2026

Upload by Asher

Long-form fiction writers need memory that answers multi-hop questions about evolving story state: who knows a secret and when they learned it, whether an event occurred before the narration that revealed it, whether setups pay off, and how relationships shift. General retrieval and agent-memory systems represent entities and facts but not the narratological structure required. The Narrative World Model (NWM) pairs a narratology-grounded typed temporal-state graph with query-conditioned hybrid retrieval and outperforms strong temporal KG baselines on multi-hop narratological QA.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages23

Reading Time58 min

Introduction

Related Work

Long-form story generation and revision

Narrative World Model (NWM)

NWM pipeline

Publish/update flow and typed records

Method and Evaluation

Multi-hop benchmark and scoring protocol

Results and comparisons

Contributions

Protocol, system, and benchmark

arXiv :2607 .05577v 1 [ cs .AI] 6 Jul 2026 Narrative World Model: Narratology-Grounded Writer Memory for Long-Form Fiction Mohammad Saifullah, Thomas Kornmaier, Taaha Kazi, Vasu Sharma, Aditya Sanjiv Kanade, Aanand Kumar Yadav PocketFM ullahsaif1[3407@gmail.com](3407@gmail.com) , [mohammad.saifullah@pocketfm.com](mohammad.saifullah@pocketfm.com) Abstract Long-form fiction writers need memory that answers multi-hop questions about evolving story state: who knows a secret and when they learned it, whether an event preceded the narration that revealed it, whether a setup paid off, and how a relationship shifted. General-purpose retrieval and agent-memory systems represent entities and facts but not the narratological structure these questions turn on, so they surface the wrong evidence or none at all. We introduce the Narrative World Model (NWM), a writer-memory system that pairs a narratology-grounded typed temporalstate graph with query-conditioned hybrid retrieval. To measure memory rather than the answerer, we read every system through a single heldconstant Opus 4.8 reader over only that system’s chapter-safe evidence, on a reproducible public corpus and a validated multi-hop benchmark, and we compare against the strongest existing temporal-knowledge-graph agent-memory framework, Graphiti/Zep (Rasmussen et al., 2025) . NWM substantially and significantly outperforms this baseline on multi-hop narratological QA across both corpora, and far exceeds GraphRAG and flat retrieval. The advantage is representational rather than an artifact of extraction: it survives rebuilding the baseline with NWM’s own extractor, and traces to its narratology-grounded structure and query-conditioned retrieval, not to graph size or extractor quality. 1 Introduction Long-form fiction generation exposes a gap between local fluency and durable narrative state. A model can write an appealing scene while forgetting who knows a secret, where an object is, whether a promise paid off, or how a relationship changed. Chapter N+1 must respect finalized chapters 1 . . . N while advancing the story. A writer-facing narrative memory system must do more than surface nearby text. Rolling summaries discard details that later become plot-critical. RAG (Lewis et al., 2020; Guu et al., 2020) retrieves relevant prose, but a passage may say where an object was before it moved. Graph retrieval can expose entities and events, but generic graphs do not necessarily track who knows what, the difference between event order and reveal order, relationship deltas, unresolved promises, or dramatic function. These approaches can miss the current, typed narrative state a writer needs even when the source prose exists. These failures concentrate on multi-hop narratological queries: who knew a fact at chapter nand when they learned it, whether an event preceded the narration that revealed it, whether a setup planted earlier paid off, and what dramatic function a beat serves. Answering such a query requires evidence from two or more distinct chapters and a representation that types narratological structure rather than generic entities. We therefore ask: does a narratology-grounded memory system provide better chaptersafe evidence for multi-hop story-state questions than generic search, source-chunk RAG, GraphRAG, or the strongest existing temporal-knowledge-graph agent-memory framework, Graphiti/Zep (Rasmussen et al., 2025)? We propose Narrative World Model (NWM), which Publish Figure 1: NWM pipeline. Finalized chapters are extracted into narratology-grounded typed records (focalization/epistemic state, event-vs-reveal order, dramatic function, promise/payoff) and a temporal knowledge graph with validity intervals; queryconditioned hybrid retrieval (BM25, vector, and one-hop graph expansion) assembles a chapter-safe evidence packet from chapters up to the checkpoint only, and a held-constant reader answers or abstains. publishes finalized chapters into typed

FAQs

Why do general retrieval and agent-memory systems struggle with long-form fiction writer questions?

They represent entities and facts but not the narratological structure required for multi-hop questions about reveal order, epistemic state, dramatic function, and relationship changes. As a result, they retrieve wrong or incomplete evidence.

What is the core idea behind the Narrative World Model (NWM)?

NWM builds writer memory from a narratology-grounded typed temporal-state graph and combines it with query-conditioned hybrid retrieval to assemble chapter-safe evidence for multi-hop story-state QA.

How is NWM evaluated to measure memory versus generation behavior?

Systems are assessed using a held-constant reader (Opus 4.8) over only each system’s chapter-filtered evidence. A reproducible protocol and multi-hop benchmark compare NWM against temporal-knowledge-graph and retrieval baselines.

Jul 16, 2026

Upload by Asher