1. Live PHOIBLE data
PHOIBLE supplies the real language inventories and distinctive-feature vectors used by the experiment. The 4×4 classifier is our experimental compression layer, not something imposed by PHOIBLE.
2. The sixteen-state bridge
Rows encode broad manner/openness; columns encode broad place/front-back territory. Tone remains outside the sixteen cells as prosody.
Select a cell
Once PHOIBLE loads this box shows its real segment occupancy.
3. Test one language against another
Choose two actual PHOIBLE language records and inventories, then a source sound. The page compares the unconstrained nearest PHOIBLE-feature match with the answer obtained when the 4×4 bridge is enforced.
Source
Target
Feature distance retains detailed residual features rather than pretending the sixteen cells contain all phonetic information.
4. Visual sound-to-sound correspondence
This uses the same two language selections above. Every source phoneme is paired with its nearest sound in the target inventory. Expand any row to see why the mapping happens: shared articulatory features, bridge cells, residual differences and feature distance.
Source → Target
These are phoneme/sound correspondences, not naïve written-letter substitutions. Orthographies can represent the same sound differently, and the same letter can represent different sounds.
5. The 256 directed gates
Sixteen states yield 16² = 256 directed state-pairs. PHOIBLE establishes the segment states; WikiPron lexical pronunciations let us observe which transitions actually occur in word-level IPA trajectories.
6. Full WikiPron cross-language benchmark
The batch benchmark runs the same bridge over every indexed WikiPron dataset, then selects one canonical dataset per language for controlled cross-language analysis. Equal-size sampling prevents larger corpora from automatically appearing to have more gates.
80% universal backbone
Stable gates observed in at least 80% of the 348 canonical languages.
Powers-of-3 transition profile
For a gate, the source and target magic-square cells define an exponent difference Δ, giving the ratio 3^Δ. These are the most widely recurring exponent deltas.
Gate prevalence across languages
Each percentage is the fraction of canonical languages in which that gate is stably observed. Hover a cell to see its gate and powers-of-three exponent delta.
7. Does the magic-square ordering beat chance?
Independently of the corpus benchmark, this compares the Dürer 4×4 numerical path against 5,000 random cell orderings using PHOIBLE feature-centroid distance. The mathematical overlay is not allowed to define the phonetic result.
Dürer 4×4 overlay
Rows, columns and principal diagonals sum to 34. The optional power overlay maps these values to powers of 3.
Permutation experiment
Lower path cost means consecutive magic-square values traverse more similar PHOIBLE feature centroids.
8. What the experiment can and cannot establish
| Statement | Status |
|---|---|
| PHOIBLE supplies cross-linguistic inventories and distinctive-feature data. | Data layer |
| WikiPron supplies real lexical IPA sequences for gate-frequency testing. | Corpus layer |
| The present rules assign non-prosodic segments to sixteen coarse bridge cells. | Implemented hypothesis |
| A high-prevalence cross-language gate backbone exists in this benchmark. | Observed in current corpus run |
| The 16-cell compression is globally optimal. | Not established |
| WikiPron lexical transitions equal continuous connected-speech articulation. | No — acoustic validation remains separate |
| The magic-square/powers-of-3 ordering has a phonetic relationship beyond chance. | Open hypothesis / controlled testing |