30 reference words
Canonical public reference subset used in several early clustering/audit tasks; it remains a reference set rather than the whole LUXVAR system.
LUXVAR: A Constructed Language with Distinctive Morphemic Structure and AI-Recoverable Morpheme Families
The strongest supported reading is morphemic and phonotactic structure, not independent recovery of the designed semantic axes. The August addendum explicitly rejects the “AI-recoverable” axis reading.
Evidence status
Mixed: reproducible morphology/phonotactics results plus negative semantic-axis tests; the historical “AI-recoverable” title claim is rejected by the paper’s later addendum. Publication status: Preprint; explicitly marked not peer-reviewed in the deposited manuscript.
Research question and current reading.
Research question
Which parts of LUXVAR’s designed structure are recoverable computationally, and do those recoverable regularities extend from morphology and clustering to the intended semantic axes?
The strongest supported reading is morphemic and phonotactic structure, not independent recovery of the designed semantic axes. The August addendum explicitly rejects the “AI-recoverable” axis reading.
What was actually tested.
- Morphological clustering over 801 roots with 36 phonotactic features and shuffle baselines.
- GEN-001 random-forest phonotactic classification using 34 features and five-fold cross-validation.
- GEN-003R+ cross-language benchmark against six curated lexical corpora with balanced n = 107 and bootstrap uncertainty.
- SEM-001 string-level and SEM-002 graph-level semantic-axis tests.
- VPE-001B+ independent AI clustering of Core-30, scored against both inter-model agreement and the designed axes.
Strongest supported findings.
- Non-random morphology: z = 4.36, p < 0.001; silhouette 0.406 versus shuffled mean 0.174.
- Generator balance at n = 2,000,000: H/Hmax = 0.9995 and Cramér’s V < 0.02 for the stated first-order associations.
- Phonotactic distinctiveness: GEN-001 CV accuracy 98.9%; uniqueness = 0.738, 95% CI [0.703, 0.781], versus six curated reference language lists.
- Claude↔Gemini inter-model cluster ARI is reported as 0.812 in the corrected manuscript, but mean ARI versus the designed axes is 0.150; the later addendum rejects interpreting inter-model agreement as axis recovery.
How this experiment relates to Core-30 and later scaled benchmarks.
801 stable roots
The v2.2 manuscript reports 801 stable roots across six semantic fields and other generator dimensions. Core-30 separately uses five designed axes; the relationship is retained as a source-level distinction, not silently harmonized.
≈1.3 billion generated forms
Exhaustive combinations of the manuscript’s design dimensions. The rows are generated by construction, not observations collected from natural language use.
Scale boundary: Core-30 remains a canonical reference subset. Larger generated corpora and later benchmark datasets are distinct experimental artifacts and should not be silently relabeled as a single enlarged canonical lexicon.
Negative findings are part of the result.
- Hexacore k = 6 was rejected; the reported natural elbow was k = 3.
- SEM-001 found no semantic-axis string signal; SEM-002 reached only 1/4 positive tests against a ≥2/4 criterion.
- Root→field and root→frequency first-order signals were not found under the stated thresholds.
- The August 2026 addendum explicitly marks the “AI-Recoverable” claim as REJECTED after direct axis-label testing in the companion Track D audit.
What this record does not establish.
- Reference-language comparisons use curated lexical lists rather than large natural-text corpora.
- AI clustering is not a human perception study; VPE-001A remained unperformed in this manuscript.
- Frequency values are symbolic design parameters, not experimentally validated physical or cognitive categories.
- LUXVAR is a single-creator constructed system; semantic-axis inter-rater validity had not been established.
- Taxonomy provenance: Core-30 is described with five designed axes, while the wider v2.2 layer is described with six semantic fields; the public site preserves this documentary distinction rather than assuming equivalence.
Boundary: a Zenodo deposit is a citable public record; it does not by itself imply peer review, independent validation, causal inference, or generalization beyond the documented experiment.
Full text, code, data, and provenance.
- The uploaded PDF supplied for this website audit is byte-identical to the Zenodo file VARZIN_LUXVAR_preprint_v2.2.pdf (verified by MD5).
- The verified Zenodo record currently exposes the PDF only; no ZIP/code file is present in that record’s public file inventory as of 2026-09-14.
- The PDF itself contains July and August post-publication addenda that narrow and then reject the earlier AI-axis interpretation.
VARZIN_LUXVAR_preprint_v2.2.pdf
paper · 87.40 KB · md5:131b44e2c769ed483e0287978abf0447
Zenodo REST file inventory and checksum re-verified 2026-09-14; uploaded paper copy exactly matches the public Zenodo PDF. Repository description still carries an older 0.718 inter-model figure, while the PDF correction standardizes the documented project figure to 0.812 and later rejects it as evidence of axis recovery.
Canonical version-specific record.
10.5281/zenodo.22115483
Reza, N. (2026). LUXVAR: A Constructed Language with Distinctive Morphemic Structure and AI-Recoverable Morpheme Families (Version v2.2). Zenodo.
Verified repository identity.
Reza, Nirouyar
Normalized project identity: Reza Nirouyar · ORCID 0009-0000-4690-6842.
2026-08-26
Version v2.2 · Preprint · open access.
CC BY 4.0
Reported only when present in verified Zenodo metadata.
Zenodo metadata identifies this item as an open-access preprint published on 2026-08-26 under CC BY 4.0.
Keep this deposit distinct from later interpretation.
The record provides a clear separation between designed structure, computationally detectable morpheme-family organization, and the stronger semantic claim that the intended axes are independently recoverable. The latter is not treated as established.
Version v2.2 is the current public LUXVAR record listed by VARZIN. The historical title wording is retained for citation continuity even where later interpretation is more conservative.