What are the main approaches researchers use to try to decode the Voynich Manuscript, and how do they differ from each other?

Question

Accepted Answer

The sharpest modern test is not “Can you make a Voynich line sound meaningful?” It is: take an ordinary text from the claimed source language and period, encode it with the proposed method, then compare its symbol frequencies, word lengths, repetitions, positional glyph behavior, entropy, and page-level patterns against the manuscript. Most theories die there. The Voynich vellum dates to roughly 1404–1438, the text has no accepted decipherment, simple substitution is effectively ruled out, and the writing behaves strangely: language-like in Zipfian word frequency and topical clustering, but abnormal in low entropy, weak word-order signals, high repetition, and rigid glyph positions.

Here are the main research approaches and how they differ.

## 1. The forensic-codicological approach: “What is this object before we ask what it says?”

This approach starts with the book as a physical artifact.

Researchers examine:

- radiocarbon dating of vellum;
- ink and pigment chemistry;
- stitching, folding, bifolia, rebinding, and page order;
- handwriting and scribal hands;
- provenance marks, marginalia, ownership notes;
- multispectral imaging to reveal faded, erased, or hidden writing.

This approach has produced some of the firmest ground: the parchment dates to the early 15th century, commonly given as 1404–1438, with samples from several folios clustering tightly. The inks and pigments are consistent with medieval materials, including iron-gall ink and pigments such as azurite, cuprite, red ochre, and glair-like proteinaceous binder. This makes a modern forgery highly unlikely.

But this approach does not decode the text. It narrows the battlefield. It tells us Roger Bacon is too early, modern forgery is implausible, post-Columbian New World theories struggle against the dating, and any proposed cipher must be historically plausible for the early 1400s.

Its hidden power is in page order. The current binding may not preserve the original sequence; forensic study of bifolia, ink transfer, stitching, and illustration continuity may change how the manuscript should be read.

## 2. The statistical-computational approach: “Does Voynichese behave like language, cipher, or generated noise?”

This is the approach that treats the manuscript as a corpus.

Researchers transliterate the script, often using EVA, then analyze:

- character frequency;
- word frequency and Zipf’s law;
- word length distributions;
- entropy;
- repeated strings;
- glyph position within words;
- line and paragraph patterns;
- bigram and trigram predictability;
- topic clustering by page or section;
- differences between Voynich A and Voynich B.

This approach has revealed the paradox at the heart of the manuscript. Voynichese looks organized, but not like ordinary prose.

It has language-like features:

- word frequency distributions resemble Zipf’s law;
- vocabulary shifts by section;
- herbal and pharmaceutical sections show related terminology;
- words appear to have prefix–stem–suffix-like structure;
- pages show topicality.

But it also has features that disturb normal linguistic models:

- many words differ by only one or two glyphs;
- repeated or near-repeated words occur in immediate succession;
- word bigrams and trigrams are surprisingly weak;
- long-distance word correlations are minimal;
- many glyphs occur mainly at beginnings, middles, ends, paragraph starts, or line edges;
- letter predictability is unusually high: bigram letter predictability has been measured around 40.02%, compared with about 22.62% for English and 24.78% for Arabic.

This approach differs from decipherment claims because it does not begin by assigning meanings. It asks what kind of machine could have produced the text.

## 3. The cryptographic approach: “What kind of cipher could produce these patterns?”

Older attempts often assumed simple substitution: one Voynich glyph equals one plaintext letter. Experienced researchers largely reject that. The alphabet is too small, the glyph positions are too rigid, the entropy is too low, and the repetition patterns are wrong.

More serious cryptographic work now asks whether the text could involve:

- homophonic substitution;
- nulls or filler;
- abbreviation systems;
- positional variants;
- ligatures;
- anagramming;
- word-level encoding;
- algorithmic generation;
- hand-executable cipher tables;
- dice or playing-card randomness;
- abjad-like vowel omission;
- constructed language.

A crucial modern rule is this: a proposed cipher must reproduce the manuscript’s statistics, not merely generate readable snippets. If a system claims the manuscript encodes Latin, Italian, Hebrew, Turkic, Arabic, or anything else, researchers should encode a comparable historical text with that system and see whether the output resembles Voynichese.

This is where most proposed “solutions” collapse. They can translate a line, but they cannot reproduce the whole manuscript’s behavior.

## 4. The linguistic/philological approach: “Can we find anchors—names, labels, grammar?”

This approach tries to do what helped decipher Egyptian hieroglyphs and Linear B: find proper names or labeled objects.

Stephen

Approach	Core idea	How it differs
Cryptographic analysis	Treats the manuscript as an encrypted text. Researchers look for substitution ciphers, transposition systems, nulls, abbreviations, or more complex encipherment methods.	Assumes there is a meaningful plaintext hidden by a deliberate cipher. Focus is on breaking a code.
Linguistic/statistical analysis	Studies the structure of the “Voynichese” text: word frequencies, repeated patterns, entropy, line structure, prefixes/suffixes, and similarities to natural languages.	Does not necessarily assume encryption. It asks whether the text behaves like a real language, an artificial language, or gibberish.
**Language-identification...”

The Voynich Manuscript: Evidence, Symbols, and Origins

About this Capsule

2 unique approaches explored here

Generic AI vs This Capsule

1. The forensic-codicological approach: “What is this object before we ask what it says?”

Tags

Related in History

Columbus and the Flat Earth Myth: Origins and Afterlife

The Bermuda Triangle Myth: Risk, Evidence, and Ocean Reality

Dyatlov Pass Theories: Avalanche, Injuries, and Evidence Handling

A Critical Thinker's Guide to Testing Stories and Claims