Loading precomputed artifact for candidate readings…parsing voynich-model.json (no inference runs in your browser)

Candidate readings

Every reading below is a hypothesis, not a translation. The structure underneath it — segmentation, class, position, illustration linkage — is measured and shown for each word. Click any word to open its evidence chain.

f1r · text-only · hand 1 · Currier Atraining folio · 6.66 bits/word surprisal

fachys ykal ar ataiin shol shory cthres y kor sholdy

Hypothesis A — literal slot reading

speculative38% confidence

in other the is said hairy thing with such half is done

Highest-scoring gloss per slot, rendered in written order. Assumes the surface order of the class sequence is also the semantic order.

  • Model surprisal on this line: 6.66 bits/word (corpus held-out average 10.93).
  • 5 of 10 words are significantly tied to a specific illustration type.

Hypothesis B — head-final reading

speculative33% confidence

in other the is said hairy with such half is done thing

Same glosses, but assumes modifiers and predicates precede their head, which is consistent with the high line-final concentration of one class.

  • Model surprisal on this line: 6.66 bits/word (corpus held-out average 10.93).
  • 5 of 10 words are significantly tied to a specific illustration type.

Hypothesis C — alternate lexicon

speculative26% confidence

for such of — is held — green item the chief double — is written —

Second-choice gloss for every slot, grouped around predicate-like words. Shows how much of the reading depends on the arbitrary semantic label rather than the structure.

  • Model surprisal on this line: 6.66 bits/word (corpus held-out average 10.93).
  • 5 of 10 words are significantly tied to a specific illustration type.

syntax trace

Class sequence for this line

Each step is compared with what independent word placement predicts. Positive z means the grammar model expects this transition; negative means the manuscript avoids it.

⟦edge⟧C4 (+1.8)C4C7 (-4.7)C7C4 (+0.0)C4C2 (+0.0)C2C7 (+0.0)C7C6 (+1.9)C6C4 (-4.3)C4C7 (-4.7)C7C5 (+0.0)C5C2 (+0.0)C2⟦edge⟧ (+15.4)

evidence

Select a word

Click a word above to unwind its reading down to raw counts. Nothing in this app is asserted without a chain like this.

No word selected.

reading the class sequence

Legend

Slot kinds are derived from the induced class, which is derived from distribution. Colour marks the slot kind, not a meaning.

entityactionqualityfunctionmarkerclasses present in this line: C4, C7, C2, C6, C5

artifact provenance

Where these numbers come from

Every figure on this page is read from a single precomputed artifact. Nothing is fitted in the browser, so the same artifact always yields the same output.

artifact v2built 2026-09-06 20:23:10 UTC44.7s inference
artifact
src/data/voynich-model.json
source corpus
src/data/voynich-corpus.json
corpus sha-256
92a4dc1c3409a3d5
corpus size
322 KB
build runtime
bun 1.3.3
generated at
2026-09-06T20:23:10.613Z
clustering seed
20260803
hybrid weights
bigram 0.6 · class 0.05 · unigram 0.35

To reproduce: run bun scripts/precompute-analysis.ts against a corpus with the same hash and these settings; the artifact, and therefore every reading and score shown here, will be byte-identical.