What Is in the Voynich Manuscript?/info


Introduction

The Voynich Manuscript (Beinecke MS 408) is a 240-page illustrated codex written entirely in an unknown script — now conventionally termed "Voynichese" — and accompanied by illustrations of unidentifiable plants, astronomical diagrams, human figures in apparent bathing scenes, and elaborate cosmological schemata. Radiocarbon dating of the vellum, conducted at the University of Arizona in 2009, places its production between 1404 and 1438 CE. Despite more than a century of analysis by professional cryptographers, linguists, historians, and computer scientists — including senior codebreakers from both World Wars — no decipherment has achieved independent verification. The manuscript remains one of the most studied and least understood documents in the history of writing.

The codex is currently held in the Beinecke Rare Book and Manuscript Library at Yale University, which acquired it in 1969 from the estate of rare-book dealer Hans P. Kraus. Its earlier provenance includes the collection of Wilfrid Voynich, who purchased it in 1912 from the Jesuits at the Villa Mondragone near Rome, and a likely chain of ownership connecting it to the court of Rudolf II of Bohemia, the Jesuit polymath Athanasius Kircher, and possibly the English polymath Roger Bacon (though the Bacon attribution is now considered improbable given the radiocarbon dates).

What we know

The manuscript comprises approximately 170,000 characters of text in the Voynich script, which contains roughly 20–25 distinct glyph types (depending on analytical decisions about character boundaries). The text is written left to right with clear word boundaries and line spacing, and shows consistent page layouts with text wrapping around illustrations. There are no corrections, deletions, or marginal annotations in the primary hand — an unusual feature for a manuscript of this period.

Statistical analysis of the text reveals properties consistent with natural language. Zipf's law (the inverse relationship between word frequency and rank) holds for Voynichese, as it does for known languages. The conditional entropy — a measure of the predictability of the next character given the preceding context — falls within the range observed for natural languages, as demonstrated by Rao et al. (2009) for the Indus script and applied to Voynichese by subsequent researchers. Montemurro and Zanette (2013) demonstrated that the text exhibits long-range statistical patterns consistent with meaningful semantic content, ruling out simple random or purely mechanical generation.

The illustrations are organised into several thematic sections: a "herbal" section depicting plants (none of which have been convincingly identified with known species), an "astronomical" section with zodiacal and cosmological diagrams, a "biological" or "balneological" section depicting nude human figures in plumbing-like systems, a "pharmaceutical" section with drawings of jars and plant roots, and a final "recipes" section of dense text with marginal stars. The illustrations' relationship to the text remains unknown.

Stylistic analysis places the manuscript's production in northern Italy, possibly the Veneto region, based on the drawing style, pigment palette, and quire construction. This attribution, proposed by art historian Rene Zandbergen and supported by several codicologists, is consistent with the radiocarbon dating but does not resolve the question of the script's identity.

Multiple decipherment claims have been published. Among the more scholarly: Newbold (1928) proposed an anagrammatic cipher based on microscopic shorthand; the Friedmans (William and Elizebeth, both eminent cryptanalysts) worked on the manuscript for decades without reaching a definitive conclusion; Cheshire (2019) claimed identification with proto-Romance, but this was widely rejected by medievalists and linguists. No claim has produced independently verifiable results — that is, no proposed key has been shown to generate meaningful text from untested portions of the manuscript.

Classification of theories

A. Plausible explanations (supported by evidence, not refuted):

  • Enciphered natural language: The text encodes a known language using an unknown or partially known cipher. Statistical properties consistent with natural language, combined with the codex's evident functional structure, support this interpretation. The specific language remains unknown — candidates include Latin, Italian, a Semitic language, or a Turkic language.
  • Unknown or constructed language: The text may represent a natural language not previously recorded in writing, or a constructed language devised for a specific scholarly or esoteric purpose. The consistent statistical properties argue against pure randomness.

B. Possible explanations (raised by researchers, less well supported):

  • Glossolalia or inspired writing: The text may have been produced in an altered state of consciousness, generating language-like patterns without semantic content. This would account for the statistical properties while explaining the failure of decipherment.
  • Steganographic text: The meaningful content may be encoded in visual elements (illustration details, letter positioning) rather than the text itself, with the text serving as camouflage.

C. Highly unlikely but argued by some:

  • Elaborate hoax: The hypothesis that the manuscript is a meaningless fabrication, possibly by Edward Kelley or Wilfrid Voynich himself, has been argued but is undermined by the radiocarbon dating (ruling out Voynich-era fabrication), the statistical complexity of the text (difficult to achieve manually without computational tools), and the manuscript's evident material authenticity.
  • Modern forgery using old vellum: The consistent integration of ink, pigment, and vellum, confirmed by spectroscopic analysis, argues against post-medieval production.

---

Research Papers

Landmark Studies

1. Montemurro, M.A. & Zanette, D.H. (2013). "Keywords and Co-Occurrence Patterns in the Voynich Manuscript: An Information-Theoretic Analysis." PLoS ONE, 8(6), e66344. Full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC3689824/ [open access] [Publisher abstract] The Voynich manuscript has remained so far as a mystery for linguists and cryptologists. While the text written on medieval parchment — using an unknown script system — shows basic statistical patterns that bear resemblance to those from real languages, no successful decipherment has been achieved. The authors demonstrate that the text possesses semantic-like keyword patterns at the section level. Full-text notes: Montemurro and Zanette apply information-theoretic methods to demonstrate that the Voynich text exhibits long-range statistical regularities inconsistent with random generation or simple encoding schemes. They identify "keywords" — words whose frequency varies significantly between manuscript sections — arguing that this pattern is characteristic of texts with meaningful semantic content organised by topic. The study does not identify the language or cipher but significantly strengthens the case that the text is not a hoax. The methodology is drawn from computational linguistics and is rigorously applied.

2. Currier, P.H. (1976). "Some Important New Statistical Findings." Presented at the New Research on the Voynich Manuscript seminar, Washington, D.C. Full text: Not available online [link unverified — may require institutional access] [Agent-generated summary] Currier, a Navy cryptanalyst, identified at least two distinct "hands" or "languages" within the Voynich text (termed Currier A and Currier B), based on statistically significant differences in character frequency distributions between manuscript sections. This finding has been confirmed by subsequent researchers. Full-text notes: Currier's identification of two distinct statistical populations within the text is one of the most robust empirical findings about the manuscript. The distinction correlates roughly with the herbal sections (A) and the pharmaceutical/recipe sections (B), suggesting either two scribes using somewhat different conventions or two dialects/registers of the same system. This finding constrains decipherment attempts: any valid solution must account for the A/B distinction. Currier's work was conducted using methods standard in military signals intelligence.

3. Bowern, C. (2024). "The Linguistics of the Voynich Manuscript." Annual Review of Linguistics, 10, 227–247. Full text: https://www.annualreviews.org/content/journals/10.1146/annurev-linguistics-011619-030613 [publisher paywall] [Agent-generated summary] Bowern reviews the linguistic evidence bearing on the Voynich Manuscript, assessing claims about the text's script type, morphological structure, and possible language family. She concludes that the text shows properties consistent with a natural language but that no specific identification has been established. Full-text notes: Bowern's review is the most authoritative recent linguistic assessment. She systematically evaluates prior claims — including Cheshire's proto-Romance hypothesis, various Turkic and Semitic proposals, and the constructed-language hypothesis — and finds none meeting the evidentiary standards expected in historical linguistics. The paper is particularly valuable for its clear articulation of what would constitute a valid decipherment: the proposed key must generate meaningful text from passages not used to develop the key, and the results must be independently reproducible. This criterion has not been met by any proposal.

4. Rugg, G. (2004). "An elegant hoax? A possible solution to the Voynich manuscript." Cryptologia, 28(1), 31–46. Full text: https://www.tandfonline.com/doi/abs/10.1080/0161-110491892755 [publisher paywall] [Agent-generated summary] Rugg demonstrates that a Cardan grille method — a Renaissance-era cipher technique — could generate text with statistical properties closely resembling those of the Voynich manuscript, arguing this supports the hoax hypothesis. Full-text notes: Rugg's paper is the most technically rigorous articulation of the hoax hypothesis. He shows that a table of syllable fragments, read through a Cardan grille with multiple cutout positions, can produce text that is locally patterned but globally meaningless — matching several statistical properties of Voynichese. The demonstration establishes that the hoax hypothesis is mechanistically feasible but does not prove it is correct. Critics note that Rugg's method does not replicate all statistical properties of the text, particularly the long-range semantic patterns identified by Montemurro and Zanette.

5. Zandbergen, R. (2020). "The Voynich Manuscript." Online resource. Full text: http://www.voynich.nu/ [open access] [Agent-generated summary] Zandbergen maintains the most comprehensive scholarly resource on the Voynich Manuscript, including detailed codicological analysis, provenance research, and critical assessment of decipherment claims. His stylistic analysis supports northern Italian production. Full-text notes: Though not a peer-reviewed paper, Zandbergen's website is widely cited in the scholarly literature as the most reliable single source for Voynich studies. His codicological work — examining quire structure, pigment analysis, and binding features — provides essential context that published papers often assume. The site includes critical assessments of all major decipherment claims, consistently applying the standards of independent verification and reproducibility. Zandbergen's attribution of the manuscript to northern Italy, c. 1420–1440, is the current scholarly consensus on provenance.

Recent Studies

6. Lindemann, L. & Bowern, C. (2020). "Script and language in the Voynich Manuscript." Proceedings of the Royal Society A, 476(2240). Full text: https://doi.org/10.1098/rspa.2020.0208 [link unverified — may require institutional access] [Agent-generated summary] The authors apply computational methods to analyse the structural properties of the Voynich script, comparing character clustering and positional distributions with known writing systems. They find that Voynichese shows properties intermediate between alphabetic and abjad systems. Full-text notes: Lindemann and Bowern provide quantitative evidence that the Voynich script has structural properties distinct from pure cipher systems and more consistent with a writing system adapted to a natural language. Their analysis of glyph positional constraints — certain characters appear only word-initially, others only word-finally — mirrors patterns seen in Semitic abjads and some abugidas. The paper does not identify the language but narrows the structural possibilities, potentially guiding future decipherment efforts.

7. Zelinka, I., Zmeskal, O., Windsor, L., & Cai, Z. (2019). "Unconventional methods in Voynich manuscript analysis." MENDEL, 25(1), 1–14. Full text: https://doi.org/10.13164/mendel.2019.1.001 [open access] [Agent-generated summary] The authors apply fractal dimension analysis, complex network analysis, and evolutionary computation to the Voynich text, identifying structural regularities that distinguish it from both random text and known ciphers. Full-text notes: This paper exemplifies the application of computational complexity methods to the Voynich problem. The fractal dimension analysis reveals self-similar patterns at multiple text scales, a feature characteristic of natural language but absent from simple ciphers. The network analysis of character co-occurrence identifies clustering patterns that are language-like. The paper contributes to the growing body of evidence that the text encodes meaningful content but does not achieve decipherment.

---

Current Discussions

  1. Beinecke Rare Book & Manuscript Library — Voynich Manuscript: Yale's digital facsimile and curatorial description.

URL: https://beinecke.library.yale.edu/collections/highlights/voynich-manuscript

  1. Undark Magazine — "The Strange Quest to Crack the Voynich Code" (2020): A long-form journalistic account of the scholarly community and its debates.

URL: https://undark.org/2020/02/12/decoding-bizarre-voynich-manuscript/

  1. voynich.nu — René Zandbergen's Research Pages: The most comprehensive scholarly resource online, regularly updated with new findings and critical assessments.

URL: http://www.voynich.nu/

---

Future Research Directions

Multispectral imaging of the manuscript — which has been partially conducted but not fully published — may reveal erased or faded text, under-drawing, or palimpsest features that could provide additional data for decipherment. Full publication of the imaging results is a priority.

Machine learning approaches to decipherment, particularly neural language models trained on known cipher systems and historical languages, represent a promising frontier. However, the short text length, unknown script-to-language mapping, and possible use of nulls or homophones present significant technical challenges.

Comparative analysis of the manuscript's illustrations with specific regional botanical, astronomical, and medical traditions of 15th-century northern Italy may narrow the cultural context of production and provide clues to the text's content. Art-historical analysis has been relatively underutilised compared to cryptographic and statistical approaches.

Provenance research, particularly in Jesuit archives in Rome, may recover documents relating to the manuscript's history between the 16th and 19th centuries, potentially including descriptions of its content or authorship.

---

Summary of Existing Research and Public Opinion

The Voynich Manuscript is among the most intensively studied undeciphered documents in existence, and the scholarly consensus remains that no decipherment has been achieved. The text's statistical properties are consistent with natural language, and the manuscript's material authenticity has been confirmed by radiocarbon dating and spectroscopic analysis. The hoax hypothesis, while mechanistically possible, struggles to account for the full range of statistical regularities identified in the text.

Public fascination with the manuscript is intense and has produced a large popular literature, much of it uncritical. Claimed decipherments appear regularly in media, few of which meet the basic scholarly criterion of independent verification. The scholarly community maintains rigorous standards for evaluating claims, centred on the requirement that a proposed key must generate meaningful text from untested passages.

The manuscript occupies an unusual position: it is not "mysterious" in the sense that its material properties are unknown, but it remains genuinely undeciphered, and the possibility that it encodes meaningful content in an unrecognised language or cipher system ensures continued scholarly interest.

---

Where Do I Come In?

This episode invites engagement with a problem that sits at the intersection of linguistics, cryptography, art history, and the history of science. The Voynich Manuscript is a genuine unsolved problem — not a pseudo-mystery inflated by popular media, but a document that has resisted the efforts of professional codebreakers and linguists for over a century.

Readers with backgrounds in computational linguistics, information theory, or machine learning may find productive approaches through the statistical literature cited above. Those with expertise in medieval Italian codicology, pharmacopoeia illustration, or 15th-century astronomical traditions may contribute through the art-historical pathway.

Observatory.wiki's community can contribute by maintaining a critical bibliography of decipherment claims, evaluating each against the independent-verification criterion articulated by Bowern. The field would benefit from a clear, publicly accessible scorecard distinguishing demonstrated findings (radiocarbon date, statistical properties, codicological features) from unverified claims.