10.5281/zenodo.22101179
NIROUYAR, R. (2026). Morphemes vs. Manifolds: Diagnosing Structural Blindness and Morphological Hijacking in Large Language Models via Finite Affine Orbits (Version v1.0.0). Zenodo.
Morphemes vs. Manifolds: Diagnosing Structural Blindness and Morphological Hijacking in Large Language Models via Finite Affine Orbits
Frozen-model audit and negative result record; interpret within the tested models and protocol.
Evidence boundary
A Zenodo deposit is a citable public record; it does not by itself imply peer review, independent validation, or broader empirical generalization.
What this record documents.
This preprint is the public record for the VARZIN study titled “Morphemes vs. Manifolds,” which audits whether structure recovered from finite-affine and constructed-language representations should be interpreted as semantic understanding. The current VARZIN reading treats detectable morphology, clustering, and geometric regularity as distinct from independent recovery of the intended semantic axes. The record is therefore presented as part of the project’s falsification-oriented evidence trail rather than as evidence of semantic emergence.
Research question
How much of the apparent structure recovered by language models reflects morphemic or representational regularity, and how much supports the stronger claim that intended semantic structure has been independently recovered?
This question is interpreted within the documented model, dataset, implementation, and evaluation procedure; it is not generalized beyond that evidence.
Canonical record metadata.
Repository metadata, checked against Zenodo.
NIROUYAR, REZA
Normalized VARZIN researcher identity: Reza Nirouyar · ORCID 0009-0000-4690-6842.
2026-08-25
Version v1.0.0 · Preprint · open access.
Not supplied in current Zenodo API metadata
Reported only when present in verified Zenodo metadata.
The Zenodo REST API description retrieved on 2026-09-14 appears internally inconsistent with this record’s deposited title and references DOI 10.5281/zenodo.22036769. VARZIN therefore does not use that description to characterize this record; the title, DOI, version, date, creator identity, and resource type were verified independently from the record metadata.
What this record adds to the research trail.
The record anchors the negative-result side of the VARZIN language-model program: structural regularity can be detectable while stronger semantic interpretations remain unsupported under the tested procedures.
Version v1.0.0 is retained as a fixed, citable public deposit. Later VARZIN pages may refine interpretation, but they do not silently rewrite this versioned record.
Keep the record distinct from later interpretation.
Frozen-model audit and negative result record; interpret within the tested models and protocol.
The VARZIN portal preserves the version-specific DOI, deposited creator string, publication date, and repository identity while allowing later project pages to state narrower interpretations, corrections, or negative results explicitly.