Ruth Selman reports on a 17th-century letter written in cipher:
I’m excited to report that the challenge set in my blog post of 4 August 2025 has been met. The letter sent by William Perwich on 9 April 1670 from the court of Louis XIV in France has been successfully decrypted, not once but twice.
On a murky Monday in late September, I started work to find my inbox brightened by two emails with solutions which made sense, both in the context of a columnar transposition of the text and because they included gossip at the court also reported by the Venetian ambassador on the same day. The cryptographers who cracked the cipher were Matthew Brown (working alone) and Dr George Lasry, Professor Norbert Biermann and Tomokiyo Satoshi (working together). The latter team are well known for deciphering a group of letters from Mary Queen of Scots, found in the National Library of France, in 2023.
The tricky aspects to the cipher were working out how many columns it was in (20) before randomly rearranging them until they formed recognisable words, and identifying all the letters which were ‘nulls’, i.e. should be discarded before attempting to decrypt (a number it is impossible to be certain about without Perwich’s original cipher key). The letter frequencies pushed the cryptographers in the right direction. Most letters appeared a usual number of times for the English language but, exceptionally, there were 8 Qs. On noting that 6 of these were towards the right margin, they were made aware that each line was probably completed with nulls.
The solution was reached by using codebreaking software the team had developed along with extensive manual work, in part required because Perwich had mistakenly omitted a couple of letters in his ciphertext. […] You will note that some numerical codes, standing in for names and places, remain in the deciphered letter. These were a separate system, impossible to crack without the key, especially as there were two numbers for each entity, but possible to interpret with knowledge of the historical context.
Click through for details, images, the solution in modern English, and the historical outcome; having fallen under the spell of David Kahn’s The Codebreakers almost half a century ago, I love this sort of thing.
Qbrf nalobql ernq gur pbzzragf qbja urer?
Gurl ner serdhragyl gur orfg cneg.
ontryssno ryovfarureczbpav
(switching to Double-ROT13 for even more security)
Back in the days of Usenet newsgroups over dialup, there used to be people who could read ROT13 at sight.
How is this sort of obfuscation done in the 33-letter Russian alphabet? Is one letter left intact, and the rest ROT16’d?
Excellent question — I hope someone can answer!
I have only vague memories of Usenet, but wasn’t it limited to 7-bit ASCII, anyway? I remember binary attachments were not only limited in size, but had to be transformed to 7-bit format. This was all pre-Unicode stuff, so I guess Cyrillic would be (ad hoc) transliterated. I remember how in the old times even on Russian sites some people would post Cyrillic transliterated to ASCII.
The transmission protocol (UUCP) most often used for Usenet in UNIX land was 7-bit IIRC, but there was a uuencode format that was used for parts of binary files (not unlike base64 today). There were 7-bit “code pages” for various European languages that only needed a few variant code points; ISO/IEC 646 was the relevant standard, and what was called US-ASCII in later times was actually a national profile ISO/IEC 646:US, which by 1991 was also the International Reference Version. (And guess why Unicode is ISO/IEC 10646). But there were also completely different 7-bit encodings used for non-Latin scripts based on the relevant national standards, like Shift-JIS for Japanese, you just had to make sure the recipient used the same standard. Nothing in the system checked if your mail or article body made sense as US-ASCII.
Back in the days of Usenet newsgroups over dialup, there used to be people who could read ROT13 at sight.
And here I was thinking Gurl ner serdhragyl gur orfg cneg was the Cornish national anthem.
How is this sort of obfuscation done in the 33-letter Russian alphabet? Is one letter left intact, and the rest ROT16’d?
I don’t know whether anyone did anything like that in Russian, but why not just ROT17 everything? (17 is obviously a better number than 16.) There’s no need for the cipher to be symmetrical.
…or you could merge ё back into е, end up with 32 letters, and ROT16 would be symmetrical…
@Jerry Friedman: I have probably told this before, but I had a professor (another Freedman, the last in my long line of math department advisors—several others having died or departed) who, when teaching graduate classes, would never say the word zero. If a calculation came out to have a vanishing result, he would write the digit on the board, but he would say seventeen instead.
Any ROT16or17 decoding scheme need to be equipped to deal with https://en.wikipedia.org/wiki/Faux_Cyrillic, whether as input or output. I was thinking of this while reviewing the perhaps stylistically-dated cover art of the 1970-released https://en.wikipedia.org/wiki/Charles_Lloyd_in_the_Soviet_Union. Which in turn was on my mind because Lloyd himself (still with us at age 87) had posted on Facebook a photo from the 1967 “festival” gig in occupied Estonia where it was recorded, as part of a memorial tribute to his then-drummer Jack DeJohnette who has just departed from us.
Now of course I’m trying to figure out the Estonian for “Vechnaya Pamyat” and the internet is offering both Igaveseks Mälestuseks and Mälestus Igaveseks and I can’t judge which, if either, is idiomatic. Perhaps some other case-ending entirely.
A mathematician with a zero-phobia? That’s good. Keeping up the reputation of the profession as a bunch of eccentric weirdos.
@Brett: In my undergrad class in mathematical logic, Prof. Benacerraf would often mention an arbitrary number, say, 17. Finally someone asked him why 17. He answered that it was the arbitrary number and gave the following proof. Clearly odd numbers are more arbitrary than even. An arbitrary number can’t be a multiple of 3, the number of the Trinity, or of 5, the basis of our numerals, money, etc. 7 and 11 are naturals in craps, and 13 is unlucky. Also, an arbitrary number shouldn’t be too big—less than 20, in fact. And 19 is too close to 20.
Italy disagrees: XVII is an anagram of VIXI – “I lived”, “I’m dead”.
Has anybody come up with a story about why the Romans decided that the perfect of vīvō should be vīxī instead of vīsī/vīssī which are given as alternative forms in Wiktionary. “By analogy with other verbs” is waffling. Which other verbs, and why only for vīvō?
Is is because the regular form would be vīsī but that’s also the perfect of vīdeō and just in this case (out of hundreds) the Romans noped out on a little bit of ambiguity?
(OK. de Vaan says it’s analogical to verbs like fluō (< *fluuō) which used to have the sequence *-eugʷ- so the perfect flū̆xī is expected. The same goes for the supine I/pptcp victum. I’ll buy that. And TIL why the fourth principal part is the supine I and not the pptcp: Verbs like fluō don’t have a passive and thus no pptcp. But when both exist, they are [always?] identical).
Or would the regular reflexes actually be vītī and vitum?
Sihler (A New Comparative Grammar of Greek and Latin, § 526.2.a) suggests that the model for vivo:vixi and struo:struxi is early Latin fivo:fixi (figo is a later back-formation).
XVII is an anagram of VIXI – “I lived”, “I’m dead”.
That might explain some of the grimmer 17s, as in The Book of the New Sun.
@ulr, I don’t have Sihler to hand, but de Vaan actually says that vīxī is “modelled on” verbs like fluō:flū̆xī; but s.v. fluō he quotes Ernout-Meillet 1979 to the effect that PIE *-eugʷ- would become *-eug- because of the *-u, so there should not be a labiovelar that could disappear intervocalically and then it must be analogical to struō. But s.v. struō he says that [the alternation] “must be analogical to fluō where -gʷ- disappeared regularly between vowels”.
So now I’m confused. It’s inconsistent circular references all the way down. If Ernout and Meillet are right, any dialect mixing will have to be back in PIE times. (But their objection would not seem to apply to fivo < *dʰeigʷ-e/o- so maybe that’s where it all started. As Sihler says).
That’s the boukólos rule, which seems to go all the way back, but was a frequent target for analogical undoing or hiding.