Author-Created Story Schemas

• Author-created story schemas are explicit narrative blueprints that specify structural, causal, and functional components in storytelling.
• They employ formal representations like plot graphs, context-free grammars, and state machines to enable AI-driven narrative planning and simulation.
• These schemas serve as generative templates and operational constraints, ensuring story coherence and alignment with authorial intent.

Author-created story schemas are explicit, formalized narrative blueprints constructed by writers to specify the structural, causal, and functional components of a class of stories. These artifacts—central to automated, semi-automated, and collaborative narrative generation systems—encode authorial intent at varying levels of abstraction, from high-level narrative function lists and directed plot graphs to machine-readable DSLs and data-centric objects for LLM-driven simulation. In contemporary research and practice, author-created story schemas function both as generative templates (defining the design space of permissible narratives) and as operational constraints for AI systems, ensuring generated stories preserve coherence, role consistency, and desired progression.

1. Formal and Algorithmic Characterizations

Story schemas admit multiple formalisms, depending on narrative abstraction, domain specificity, and system integration requirements:

• Linear Narrative Sequences (Proppian scripts): A schema is an ordered vector of narrative functions or plot blocks, each optionally labeled as mandatory or optional. Propp's morphology encodes fairy tale structures as a set $\langle f_1, f_2, ..., f_n \rangle$ with rules about repetition and omission (Botelho, 2021).
• Plot Graphs: A schema $G = (V, E)$ with $V$ as abstract event nodes and edges $E$ of types $B(e_1,e_2)$ (ordering) and $M(e_1,e_2)$ (mutual exclusion), determining valid partial orders (Botelho, 2021).
• Context-Free Grammars: Schemas as CFGs: $story \rightarrow setting\,Episodes$, $$Episodes \rightarrow Episode\,Episodes \mid \epsilon$$. Semantic and causal constraints may be attached as side conditions (Botelho, 2021).
• Role-Relation Graphs (Schema Discovery): $S = (C, R)$ with $C$ as components (Protagonist, Climax, etc.), $R \subseteq C\times C$ as labeled relations (motivates, leads_to), often induced via maximal common subgraph from a set of example story graphs. Each $C_j$ can carry attributes $A_j$ (Wang, 7 Aug 2025).
• State Transition Systems (SAGA DSL): $M = (S, E, T, s_0, F)$ with $S$ as story states, $E$ as event labels, $T \subseteq S \times \wp(E) \times S$, and $F$ as terminal states (Beyak et al., 2011).
• Structured Scene Graphs (Dramamancer): Directed graphs where nodes are scenes; edges are event-triggered transitions. Each scene contains explicit fields for style, character set, setting, opening line, and a collection of events, each modeled as a conditional effect (Wang et al., 26 Jan 2026).
• Abstract Act Tuples (StoryVerse): $a = \langle G, P, H \rangle$ where $G$ is a narrative goal (possibly parameterized by placeholders $H$), and $P$ is a logical formula capturing preconditions (world state, player actions, act dependencies). These are evaluated and expanded by iterative LLM-based planning (Wang et al., 2024).
• Storyforms with Intent Constraints (NCP): $S = (P, F, D, Q, B, C)$ where $P$ (perspectives), $F$ (forms of conflict), $D$ (dynamics), $Q$ (perspective-form quad), $B$ (ordered storybeats), $C$ (logical constraints). Enums and bijective mappings enforce narrative alignment and validate outputs in downstream generation (Gerba, 5 Mar 2025).
• Habitual Event Schemas for Persona-Based Dialogue: $S = (H, Pr, S, Po, G, E)$ capturing header, preconditions, static conditions, postconditions, goals, and episodes; used to anchor consistent persona-driven conversational outputs (Kane et al., 2023).

2. Schema Construction and Authoring Workflows

The process of creating and operationalizing story schemas spans manual abstraction, collaborative discovery, and machine-assisted workflows:

• Manual Encoding: Pioneered by Propp, Colby, and Lang, this involves expert-driven abstraction of narrative structures from corpora, encoding function sequences, dependencies, or grammar rules based on close reading (Botelho, 2021).
• Crowdsourcing and Statistical Induction: Protocols like Scheherazade collect annotated short stories with strict templates; clustering and frequency analysis yield statistically grounded plot graphs (Botelho, 2021).
• AI-Assisted Discovery (Schemex): Interactive clustering, component abstraction, and contrastive refinement over LLM-derived event tuples leads to inductive schema identification, supporting iterative refinement through side-by-side output comparison and user correction (Wang, 7 Aug 2025).
• Direct Specification (DSLs, JSON, Visual Tools): Languages like SAGA allow writers to codify transitions, nodes, and event triggers in a formal syntax automatically compiled to executable code for integration in game engines. JSON-based standards (e.g., NCP) offer portable, versioned schemas for interoperability (Beyak et al., 2011, Gerba, 5 Mar 2025).
• LLM-Powered Co-Authoring (StoryVerse, Dramamancer): Authors supply abstract, high-level acts or scene-structuring directives; LLMs operationalize these into concrete action plans or scene transitions, with feedback and revision loops enforcing coherency and intent alignment (Wang et al., 2024, Wang et al., 26 Jan 2026).

Table: Representative Schema Formalisms

System/Approach	Formalism	Key Schema Elements
Propp, INES, etc.	Function sequence (list/CFG)	Named plot slots, ordering, optionals
Plot Graphs	DAG or graph (V,E; B, M edges)	Partial order, mutual exclusion
SAGA	FSM (nodes, events, transitions)	States, triggers, sections
StoryVerse	Tuple $\langle G, P, H \rangle$	Goal, logical prerequisites, placeholders
Schemex	Graph $(C, R)$	Dimensions/components, relations, attribute sets
NCP (Storyform)	Tuple $(P,F,D,Q,B,C)$	Perspectives, forms, dynamics, constraints, beats
Persona Schemas	Tuple $(H,Pr,S,Po,G,E)$	Event header, pre/post/static/goal/episode fields

3. Application in Story Generation and Simulation Workflows

Author-created schemas are applied as generative scaffolding and operational constraints in both interactive and non-interactive narrative systems:

• Schema-Driven Planning: SAGA interprets the schema as a state machine, triggering narrative transitions on event satisfaction, and outputs code for RPG engines (Beyak et al., 2011). In StoryVerse, the Act Director evaluates pending abstract acts against current world state, invoking LLM planners to concretize narrative plans that map high-level schema goals to executable character actions (Wang et al., 2024).
• Constraint-Based Generation: NCP’s Storyforms encode perspectives, forms, and dynamics as constraints (Boolean logic, predicates over candidate story segments), enabling both filtering and prompt construction for LLMs to enforce thematic and structural fidelity (Gerba, 5 Mar 2025).
• LLM Conditioning and Persona Realization: Persona schemas are retrieved and matched to dialogue turns via embedding similarity, forming the basis for context-rich, episode-driven conversational outputs. The schema’s fields contribute knowledge, event structure, and goals as direct prompt features for dialogue models (Kane et al., 2023).
• Interactive and Co-Creative Authoring: Systems like Schemex/SchemaBuilder and Dramamancer provide iterative, panel-based interfaces for schema construction and instantiation, including exemplars, scenario-based suggestion, diverge-converge variation cycles, scene progression, and player/author interleaving for personalized or branching narrative paths (Wang, 7 Aug 2025, Wang et al., 26 Jan 2026).
• Macrostructural Annotation and Analysis: High-level macro-structural schemas (e.g., from Freytag, Labov-Waletzky) operationalize tension arcs and event salience via annotation categories for computational modeling, training, and downstream generation (Li et al., 2017).

4. Typical Schema Components and Exemplar Structures

Schematization covers granular and abstract story constituents, often hierarchically or relationally organized:

• Abstract Acts (StoryVerse):
  • Narrative goal $G$, e.g., "A character gets into a life-threatening accident"
  • Prerequisites $P$: logical formulas over world state, player actions, act outcomes
  • Placeholders $H$: identifiers with semantic descriptions, e.g., "X" bound to character entities
• Universal Narrative Model (NCP):
  • Perspectives $P$ (MC, CP, RS, OS), Forms $F$, Dynamics $D$, Quad mapping $Q$, Beat sequence $B$, Constraints $C$
  • Example for Star Wars: MC → EF, CP → IP, with Dynamics: Resolve=Maintained, Outcome=Success, etc. (Gerba, 5 Mar 2025)
• Role Event Graphs (Schemex):
  • Components {Protagonist, Antagonist, Inciting Incident, ...}
  • Relations: "motivates", "opposes", "leads_to" among components
  • Each dimension $d_j$ (e.g., Climax) has candidate attribute sets $A_j$
• Habitual Persona Schemas:
  • Header ("I work in a bookstore"), preconditions ("My shift has started"), static ("Customers visit the bookstore"), postconditions, goals, and episode list ("I help customers find books...") (Kane et al., 2023)
• Plot Functions (Propp): "Absentation", "Interdiction", "Violation", ..., "Wedding", each with slot rules and co-occurrence constraints (Botelho, 2021).

5. Evaluation, Portability, and Limitations

Contemporary research emphasizes both operational validation and human factors, alongside technical and theoretical constraints:

• Evaluation and Metrics: Systems report diversity (D-1, D-2), n-gram entropy (ENTR), BLEU/ROUGE similarity, human judgments of engagement/relevance, author metrics (structural fidelity, constraint satisfaction), and schema quality via attribute-typicality ratings (Kane et al., 2023, Wang, 7 Aug 2025, Wang et al., 26 Jan 2026). Systematic studies for large-scale story generation are ongoing (Wang et al., 2024).
• Interoperability: Formats such as NCP’s JSON-based schema files, or DSLs like SAGA, ensure schemas are tool- and platform-agnostic, supporting reuse across web frontends, Unity plugins, and custom engines. Versioned schemas facilitate long-term archival and reproducibility (Gerba, 5 Mar 2025, Beyak et al., 2011).
• Limits and Challenges:
  • LLM limitations in long-range coherence, context tracking, and runtime latency under heavy planning pipelines (Wang et al., 2024).
  • Expressiveness vs. manageability: complex, highly branching, or large-scale schemas introduce cognitive load for authoring and debugging—not always mitigated by abstraction tooling (Wang et al., 2024, Wang, 7 Aug 2025).
  • Genre, domain, and structure specificity: While macrostructural annotations (Labov, Freytag) generalize, complex role-/action-based schemas often require careful adaptation for novel genres or interactives (Li et al., 2017).
  • Player/author agency tradeoffs: Relative rigidity of symbolic schemas versus the unpredictability and richness of LLM-driven, character-centric approaches (Wang et al., 2024, Wang et al., 26 Jan 2026).

6. Historical Trajectory and Research Synthesis

The lineage of author-created story schemas traces foundational narratological scholarship through symbolic AI, crowdsourcing, and, most recently, LLM-enabled co-creative pipelines:

• Early schema systems (Propp, Colby, Lang) focused on fixed, expert-defined lists and tree/graph representations, operationalized via classic grammar and state machine machinery (Botelho, 2021).
• Crowdsourced and statistically induced plot graphs democratized schema construction, enhancing coverage of everyday or domain-diverse narratives (Botelho, 2021).
• Interactive tools and schema-induction frameworks using LLMs (Schemex, StoryVerse) blend human abstraction with AI pattern recognition and suggestion, making authoring scalable and accessible without losing structure (Wang, 7 Aug 2025, Wang et al., 2024).
• Standardized, machine-actionable schema formats (NCP, SAGA, Dramamancer objects) enable cross-platform, constraint-driven generativity, integrating authorial control into dynamic procedural or interactive storytelling pipelines (Gerba, 5 Mar 2025, Beyak et al., 2011, Wang et al., 26 Jan 2026).

Author-created story schemas thus encapsulate the hybridization of narratological theory, formal language/graph methods, and computational creativity frameworks, serving as the infrastructure for next-generation narrative generation, analysis, and personalization across literary, game, and conversational domains.