Voice
Distinctive phrasing, persona, stance, and linguistic identity.
Framework
NFAF.
The Narrative Flattening Analysis Framework (NFAF), an original framework developed by Johennie Helton, is a modular method for examining what changes when narrative passes through AI-mediated transformation.
Purpose
NFAF provides a vocabulary and structure for analyzing changes that ordinary accuracy measures often overlook.
It is designed to support close reading, comparative analysis, and future computational measurement of narrative signal.
Applied Process
NFAF Workflow
NFAF may be applied as an iterative analytical process that moves from source selection through transformation, evaluation, human review, and final reporting.
- 01 Input Selection Select the source material, comparison set, or narrative inputs.
- 02 Transformation Apply the relevant transformation, mediation, or interpretive process.
- 03 Alignment Align the transformed output with the source, reference, or comparison framework.
- 04 Drift Assessment & Rubric Scoring Evaluate narrative drift and apply rubric-based scoring at the mediation stage.
- 05 Pattern Analysis Identify recurring shifts, omissions, amplifications, and structural tendencies.
- 06 Human-in-the-Loop Review Review findings through expert interpretation, contextual judgment, and validation.
- 07 Reporting Document the results, evidence, limitations, and interpretive conclusions.
What it examines
Cadence
Rhythm, pacing, sentence pressure, repetition, and pause.
Imagery
Specific sensory and symbolic content carried by the text.
Cultural register
Markers of place, history, community, idiom, and social context.
Emotional intensity
Changes in warmth, urgency, restraint, grief, violence, or tenderness.
Structure
Narrative order, focalization, emphasis, and relation among events.
Analytical Framework
NFAF Rubric Dimensions
The NFAF rubric provides six analytical lenses for identifying where narrative flattening may be observed. The dimensions assess changes in rhythm, vocabulary, figurative expression, cultural register, affect, and structural coverage.
The rubric should be interpreted alongside the recurrent forms of flattening defined in §5.3: erosion, inflation, inference, and omission/compression.
Forms describe how flattening manifests; rubric dimensions describe where it is observed.
Cadence & Rhythm
Examines sentence-length variation, syllabic pacing, and punctuation patterns that shape narrative voice.
Signals: sentence length, variance, syllables, commas, em-dashes, and ellipses.
Lexical Richness
Examines vocabulary diversity and whether rare, distinctive, or culturally marked language is retained.
Signals: TTR, MTLD, hapax share, and rare-word rate.
Figurative Density
Examines the preservation, simplification, or removal of idioms, metaphors, similes, and other figurative constructions.
Signals: idiom frequency, simile markers, and metaphor candidates.
Cultural Register
Examines whether culturally specific language, entities, honorifics, foods, places, and sayings are retained or generalized.
Signals: locale-linked entities, idioms, lexicon, and generic substitutions.
Emotional Intensity
Examines changes in affective range, evaluative language, sentiment, and narrative highs and lows.
Signals: affect lexicons, sentiment, subjectivity, direction, and range.
Coverage & Structure
Examines omission, compression, fusion, segment retention, and the preservation of discourse relationships.
Signals: retention, compression ratio, unmatched spans, and discourse structure.
View the full analytical rubric Signals, interpretation, and reporting obligations
| Rubric dimension | Core signals Δ = Output − Source unless noted | Interpretive guidance | Reporting obligations and analytical responsibilities |
|---|---|---|---|
| Cadence & Rhythm | Sentence-length mean and variance; syllables per sentence; punctuation cadence, including commas, em-dashes, and ellipses per 1,000 tokens. | A drop in variance may indicate flattening. Analyze locally where possible to preserve segment-level effects. | Loss of rhythm is not merely clarification. Changes to narrative voice MUST be disclosed. |
| Lexical Richness | Type–Token Ratio (TTR), Measure of Textual Lexical Diversity (MTLD), hapax legomena share, and rare-word rate. | Define rarity using a frequency quantile in the source or an appropriate reference corpus. | Suppression of rare or culturally marked lexemes may constitute erasure when left undocumented. |
| Figurative Density | Idiom frequency per 1,000 tokens; simile markers; metaphor candidates where suitable tools are available. | Rule-based proxies may be used as a starting point, but their limitations MUST be documented. | The conversion of idioms or metaphors into plain language is a form of cultural and stylistic loss that MUST be made visible. |
| Cultural Register | Locale-linked named entities, culturally specific vocabulary and idioms, honorifics, and the percentage replaced by generic terms. | Small culture-specific gazetteers may be developed for places, foods, sayings, titles, and other marked references. | Entity dilution, idiom neutralization, and honorific loss are cultural erasures rather than neutral substitutions. |
| Emotional Intensity | Evaluative and affective lexicon counts; sentiment and subjectivity measures; categories from tools such as LIWC or Empath. | Track both the direction of change and the range or standard deviation of affect. | Compression of affective range may erase narrative highs and lows. Neutralization MUST be disclosed. |
| Coverage & Structure | Segment retention, compression ratio, unmatched-span count, and discourse-relation preservation where RST or UD tools are used. | High omission rates and many-to-one segment fusions may indicate flattening. Report dropped segment identifiers or ranges. | Undisclosed omission risks material misrepresentation. Structural drift MUST remain traceable. |
Δ convention:
Output − Source. Positive values indicate more in the output; negative
values indicate less.
Example: if sentence-length variance decreases from 42.1 to 18.6,
Δσ² = −23.5, which may constitute a flattening cue.
Analytical anchor: Every observed indicator should be tied to one or more rubric dimensions and supported by qualitative examples, such as idiom paraphrase, loss of irony, honorific removal, entity generalization, or segment omission.
Status
Working Draft Specification
The framework remains under development. Its modular structure is intended to evolve as methods, datasets, and comparative studies expand.
Open the DOI record →