Context

What Problem Does TNGS Solve?

Literary narratives have structure — causal chains, recurring patterns, focalization choices, tonal registers — but most digital text tools treat narrative as an undifferentiated string. When a writer asks "what happens if I shift this scene to an external narrator?" or "how does the chronotope change if I compress the time frame?", they have no computational tool for exploring the answer systematically.

TNGS makes narrative structure explicit as a property graph and provides auditable operations for transforming that structure along defined literary axes. The output is not final prose — it is a graph state that can be rendered into prose, diffed against a prior state, exported, or further transformed.

Theoretical Foundations

Genettean Narratology

Gérard Genette's Narrative Discourse (1972) provides the vocabulary for focalization: zero (omniscient narrator), internal (filtered through one character's consciousness), and external (recording behaviour without interiority). TNGS encodes these as the FocalizationDistance enumeration on Perspective nodes.

Bakhtinian Chronotope

Mikhail Bakhtin's concept of the chronotope (in The Dialogic Imagination, 1975) describes the intrinsic connection between time and space in narrative. A road novel has an open/linear chronotope; a fairy tale has a bounded/cyclical one. TNGS's Chronotope node captures time_mode and space_mode.

Barthesian Codes

Roland Barthes's S/Z (1970) identified five codes operating on narrative text: - Hermeneutic — creates mystery and anticipation - Proairetic — action implying a consequent action - Semic — connotative detail building character or atmosphere - Symbolic — binary or antithetical thematic opposition - Cultural — reference to shared knowledge or convention

TNGS attaches these as CodeTag nodes on individual atoms via the code_overlay transformation axis.

Design Philosophy

1. Ambiguity is data. Human annotation varies. Confidence fields and review flags represent ambiguity rather than forcing false certainty. An atom with confidence=0.45 and needs_review=True is a valid graph node that awaits human resolution.

2. Transformations are non-destructive. Every transform creates a new state node and detaches (but does not delete) the previous edge. The full lineage is always traversable in the graph.

3. The graph controls structure; prose style is downstream. TNGS determines what is said, in what order, from whose perspective, in what mood. How it is phrased is the responsibility of a rendering layer that may use templates, rule-based generation, or LLM-assisted writing.

4. Provider neutrality. No mandatory dependency on any LLM vendor. The ingest pipeline is rule-based; the rendering layer is pluggable.

Who Is This For?

User	Use Case
Experimental writers	Exploring alternative narrative paths, perspectives, or genres for the same story material
Computational narratologists	Studying pattern families, transformation sequences, and inter-text comparison at corpus scale
Digital humanists	Encoding and querying literary structure in a graph database
Knowledge-graph engineers	Building narrative-aware knowledge representations

Relationship to the White Paper

This system is the reference implementation of the theory described in the companion white paper "Transformable Narrative Graph: A Formal Model for Literary Narrative as Mutable Graph Structure." The SRS is the operational specification; the white paper provides the theoretical justification for the design choices.