SSRN Preprint Available · Under Review at Consciousness and Cognition
Narrative-derived and questionnaire-derived reports of the same NDE correspond systematically but incompletely, revealing structured representational mismatch rather than simple extraction failure.
Suggested citation and DOI correspond to the SSRN preprint posted on August 12, 2026. The manuscript is also under review at Consciousness and Cognition.
Two valid representations of the same NDE can diverge: free-text narratives preserve local expression, sequence, and ambiguity, whereas questionnaires organize reports into predefined dimensions and retrospective evaluations.
Core Idea
Abstract
Example 1 · Local worry
"It was very bright... I was worried because I wanted to go downstairs and wake up, but I couldn't."
A globally positive report can still include locally distressing narrative segments.
Example 2 · Ambiguous language
"monochrome beige fixed image, in an attic, mushroom and wooden case. Don't calm down."
Fragmented expression weakens alignment between text tone and questionnaire labels.
Example 3 · Muted expression
"No pain, no voice, no reaction, total nothing."
Low-affect wording can appear neutral despite a positive retrospective appraisal.
This study tests whether narrative-derived labels and structured questionnaires provide interchangeable representations of near-death experiences. The main result is structured partial correspondence rather than full equivalence.
Question
Main Findings
NDE-C: phenomenological content features reported during the experience.
LCI-R: perceived post-experience life changes reported after the event.
Macro F1: class-balanced metric that weights each class equally.
Cohen's kappa: agreement beyond chance between paired representations.
Paired questionnaire data and sectioned first-person narratives
Screen, translate, validate, and resegment narrative sections
Derive tone, NDE-C, and LCI-R labels with 12 LLMs and human reference
Compare narrative-derived and questionnaire-derived paired labels
Estimate where correspondence holds and where interchangeability breaks down
190
Semantically valid narratives in the final corpus
NDE-C highest
Strongest family-level correspondence
LCI-R lowest
Weakest family-level correspondence
What Mismatch Is Not
Two Representational Layers
Family-level agreement occupies distinct regions of the macro F1-kappa space: NDE-C combines the strongest overall correspondence, tone overlaps in kappa but stays lower in macro F1, and LCI-R clusters lower in chance-corrected agreement.
How to read: each point represents one model-family combination; upper-right indicates stronger cross-representational correspondence.
Item-level agreement is higher for many NDE-C features than for most LCI-R items, suggesting that more concrete experiential content is more directly available in narrative text than longer-term evaluative change.
How to read: each point is an item-model combination; NDE-C items more often occupy the higher-agreement region than LCI-R items.
Confusion matrices show that disagreement is concentrated in mixed and neutral narrative-tone assignments rather than in direct positive-negative reversals, indicating partial affective correspondence without equivalence.
How to read: rows are questionnaire valence and columns are narrative-tone assignments; off-diagonal mass is concentrated near ambiguous categories.
Standard sentiment benchmarks cluster well above NDE tone results in macro F1 and kappa, making a purely general model-failure explanation less plausible.
How to read: benchmark points cluster in the upper-right, while the shaded NDE tone region remains distinctly lower.
Human annotations show the highest agreement for segmentation and heterogeneous downstream agreement across NDE-C, tone, and LCI-R, with human-LLM agreement generally exceeding human-questionnaire agreement.
How to read: segmentation agreement is highest, but downstream construct correspondence remains family-dependent, indicating that mismatch is not explained by section assignment alone.
Translation and semantic resegmentation do not uniformly improve downstream agreement; instead, they restructure narrative evidence with heterogeneous effects across NDE-C, tone, and LCI-R.
How to read: each dumbbell links the complete pipeline to an alternative preprocessing configuration, showing how agreement shifts by family.
What The Gap Suggests
Why It Matters
Try the interactive demo to explore narrative sections, structured coding outputs, and cross-representational alignment behavior.
Open Demo in New TabA short visual summary of the updated findings, including family-level correspondence profiles, human-reference comparisons, and preprocessing effects.
Open Video SourcePreprint available on SSRN. Manuscript currently under review at Consciousness and Cognition.
Suggested citation: Pulido, Cristian and Gómez, Francisco and Tewarie, Prejaas K.B. and Hoyer, Roxane S. and Laureys, Steven, Quantifying Representational Mismatch between Narratives and Structured Reports in Near-Death Experiences (August 12, 2026). Available at SSRN: https://ssrn.com/abstract=7272084 or http://dx.doi.org/10.2139/ssrn.7272084.
@article{pulido2026representationalmismatch,
title = {Quantifying representational mismatch between narratives and structured reports in near-death experiences},
author = {Pulido, Cristian and Gómez, Francisco and Tewarie, Prejaas K.B. and Hoyer, Roxane S. and Laureys, Steven},
journal = {SSRN Electronic Journal},
year = {2026},
month = {August},
doi = {10.2139/ssrn.7272084},
url = {https://ssrn.com/abstract=7272084},
note = {Preprint; manuscript under review at Consciousness and Cognition}
}