The big idea
The Voynich Manuscript is a 600-year-old book that nobody can read. So how do researchers study it? They look for patterns — the fingerprints that real writing leaves behind — without needing to know what any word means. In this activity, learners become “repetition detectives” and do the same thing.
Set-up
Give each group three short samples, unlabeled (A, B, C):
- A — a paragraph of ordinary English.
- B — random gibberish (mash the keyboard, or scramble letters).
- C — a short passage of Voynich transliteration (e.g.
daiin daiin qokedy qokedy chol chol shedy).
Don’t tell them which is which yet.
Steps
- Hunt for repeats (10 min). In each sample, highlight any word that appears more than once. Which sample repeats words the most? Which the least?
- Measure word length (5 min). Are the words mostly short, mostly long, or a mix? Real languages usually have lots of short common words (the, a, of) and fewer long ones.
- Look at word beginnings/endings (5 min). Do many words start or end the same way? (The Voynich text does — lots of words look like cousins of each other.)
- Make your call (5 min). Which sample do you think is a real language? Which is nonsense? Which is the mystery? Rate your confidence: A, B, C, or D (see below).
- Reveal & discuss (10 min). The surprise: the mystery text (C) has strong patterns — it repeats and has word-families like a language — but that still doesn’t prove it means anything. Nonsense can be patterned too.
Confidence grades (borrowed from the real research)
Researchers on this project grade every claim: A = very sure (checked many ways) · B = fairly sure · C = a guess with some evidence · D = just a hunch. Ask learners to grade their own guesses — and to say why.
The takeaway
You can learn a lot about writing by counting patterns, even in a language nobody understands. But a pattern is a clue, not a conclusion — the honest answer about the Voynich Manuscript is still “we don’t know.”
Extend it
Older learners can try the real tool: an adult can run each sample through
python -m ms408 and compare the “verdict” to the class’s guesses.