In today’s world, encryption plays a vital role in all aspects of data protection and cybersecurity, ensuring that unauthorized individuals cannot access the information you transmit or store. However, the practice of sending messages that can only be interpreted by their intended recipients predates the age of computers by centuries.
Since antiquity, it has been common to use what is called a cipher system to send secret messages.
“A cipher system operates on a kind of formula described in a key,” explains Michelle Waldispühl, a professor of German at the University of Oslo.
“The key is shared only among the people who should be able to read it. One person encrypts, and the other decrypts using the key.”
In a cipher system, for example, each letter is replaced with a number. To make it more complicated, each letter can be substituted with several different numbers, making it harder to identify a pattern. Unless you have the key.
Discovered conspiracy letters from Mary Stuart
Without the key, you have a monumental task ahead of you. Unless you do as Waldispühl and her collaborators did. Historians, linguists, and computer scientists have joined forces to use artificial intelligence to uncover the secrets.
In the French National Library, one of the researchers, George Lasry, found materials that would bring international recognition to the team.
“He discovered over 50 letters in the same cipher system that turned out to be written in Mary Stuart’s handwriting. No one had understood what they were, so they were archived in a very peculiar way,” Waldispühl explains.
The letters revealed how the former Scottish queen conspired against her cousin, Elizabeth I of England. In the years leading up to her execution, she was imprisoned, yet she managed to send numerous encrypted letters from her cell to the French ambassador in England. The Guardian described the decryption as the most significant new discovery about Mary Stuart in over a century.
Peace negotiations and love letters
Waldispühl and her colleagues have scoured old archives in search of cipher scripts to compile everything into a database.
“The material we now have in the database is primarily from the 18th and 19th centuries, and it mostly deals with diplomatic letters,” she says.
She has personally examined 15 letters received by Axel Oxenstierna, the Chancellor of Sweden, which were sent by his ambassador in Germany during the Thirty Years’ War, spanning from 1618 to 1648.
“The letters contain reports from the war, details about peace talks, other ongoing negotiations and the parties involved. They also contain a wealth of personal information,” she notes.
There does not appear to be a consistent system governing what parts are encoded and what parts are written in plain text.
“When they were pressed for time, it’s evident that not much was encrypted. Mainly, place names, personal names, and highly sensitive information were encoded.”
Not all matters revolve around war and peace; love too can demand a veil of secrecy.
“From a private collector, we have received more than 400 postcards containing cipher script. They are from the late 1800s and early 1900s, and he found them at flea markets across Europe. Among them are love letters,” Waldispühl says.
For instance, there is a postcard from 1908 in which the message begins with “=voevoeeoggvkov/l” and continues just as incomprehensibly. So far, only the first couple of lines have been decoded. They turn out to be in German, and the message is “Meine innig geliebte einzige herzensgute Miezefrau,” which in English roughly translates to “My dearly beloved, only kind-hearted kitty.”
Deciphering cannot be left to machines alone
Throughout history, codes have evolved to become increasingly sophisticated. Each time a key is exposed, conspirators are compelled to develop a more challenging code to crack. Around the era of World War II, machines started playing a significant role in execution and breaking codes. One of the most renowned coding devices from that time is the Nazi Enigma machine. Alan Turing and the British needed machines to crack the code.
Thus, the prospect of employing artificial intelligence and modern language models to unravel these centuries-old cipher puzzles is undeniably enticing. However, there is one significant challenge:
“Conventional language models require vast amounts of material for training, which isn’t feasible in this case. Sometimes we have as little as half a page of text to work with,” Waldispühl explains.
This is why the human element, often referred to as the “human-in-the-loop,” becomes especially crucial. For instance, when a computer scientist encountered a letter where every part was encoded, the initial step was to transcribe the text—translating the letter’s characters into a format the machine could process. The transcription was then sent to the historical linguist at the University of Oslo for further analysis.
“I immediately saw that he missed a comma and a couple of dots over some of the characters. So, I corrected it with my philological eye,” Waldispühl says.
Then they used their computer tools to match the code against texts in multiple languages. Despite these efforts, the colleague had to confess that the contents of the letter remained elusive to him.
“So, I did a manual analysis to check what the machine might have gotten wrong. It’s like a puzzle where we continuously update the key the machine is working with.”
It turned out the letter was campaign propaganda from the Holy Roman Emperor Maximilian II, who in the 1570s was campaigning to become king of Poland-Lithuania as well. It contained promises of what recipients would receive if they chose him and threats of military force if they did not.
Developing tools for more unsolved mysteries
Waldispühl and her colleagues have compiled thousands of coded documents into a database (de-crypt.org), but their ambitions extend beyond just deciphering old, coded letters and postcards.
“We now aim to take a step further, not only by examining cipher scripts but also by expanding our focus to other writing and symbol systems, even those with limited data,” says Waldispühl.
She mentions examples such as the early Greek language known as “Linear B” and the 4,000-year-old Phaistos Disc from Crete, which remains undeciphered to this day. However, the primary goal is to simplify everyday tasks when encountering a document of unknown content.
“The main aim is to develop models for transcribing and deciphering, and then to develop tools that can benefit everyone,” Waldispühl explains.
If Waldispühl’s dreams come true, in a few years, you will be able to simply upload an image on your phone and get a complete decryption in seconds.
University of Oslo
Citation:
AI assists researchers in decoding old secret letters (2025, February 25)
retrieved 25 February 2025
from https://techxplore.com/news/2025-02-ai-decoding-secret-letters.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
