Artificial Intelligence May Have Helped Us Decipher Unknown Code In A 600-Year-Old Manuscript

Since it was first discovered more than a century ago, the Voynich manuscript has excited and confounded historians, linguists, and cryptographers.

Images courtesy: Beinecke Rare Book & Manuscript Library, Yale University

With its 240 pages of coded script and puzzling illustrations, the manuscript¡¯s meaning has remained hidden all these. In fact, no one is even sure of where it was written and by whom or what language it¡¯s in. Now, we¡¯ve got our first real lead on the document¡¯s history thanks to the power of artificial intelligence.

The manuscript gets its name from Wilfrid Voynich, the Polish book dealer who first obtained it in 1912. Not only is it coded in an unfamiliar pattern, it¡¯s also a language no one recognises, making decrypting it doubly hard. And while some illustrations suggest astronomical symbols, or biology diagrams, they¡¯re not much help either. It also doesn't help that there's no known history about the book, because it dates all the way back to the fifteenth century.

Because of this, the Voynich manuscript is considered the Holy Grail of cryptography, an un-crackable code that has resisted efforts to decipher it for decades, going back to during the Second World War. Some have even gone so far as to suggest it¡¯s indecipherable because it¡¯s all an elaborate hoax.

That¡¯s when Greg Kondrak, an expert in natural language processing at the University of Alberta, comes in. To him, this was the perfect task to delegate to an artificial intelligence. After all, if there was a pattern to discover underneath all the confusion, machine learning would be able to find it. With the help of his grad student Bradley Hauer, Kondrak began the first step, that of figuring out the language of the text, before it was encoded.

The two trained a neural network on the text of the ¡°Universal Declaration of Human Rights¡±, seeing as it has translations in 380 different languages. After that, they set the AI to examining patterns in the Voynich manuscript, and it turned up something they didn¡¯t expect. Though they had wondered if the text was encrypted from Arabic, it turns out the document is actually written in Hebrew.

¡°That was surprising,¡± said Kondrak in a statement. ¡°And just saying ¡®this is Hebrew¡¯ is the first step. The next step is how do we decipher it.¡±

Next, the two decided to try out one hypothesis mentioned by previous researchers, that the document was encoded by writing it with alphagrams. An alphagram is an anagram of a word, but with the letters put in alphabetical order. For instance, INDIATIMES, would be written as ADEIIIMNST. Of course, this was much easier to test now they knew the manuscript was in Hebrew and could test the supposed alphagrams against real Hebrew words.

¡°It turned out that over 80 percent of the words were in a Hebrew dictionary, but we didn¡¯t know if they made sense together,¡± Kondrak said.

Finally deciphering the first line in the manuscript, kondrak and Hauer took it to Moshe Koppel, a colleague with computer science expertise as well as a native Hebrew speaker. Unfortunately, the sentence still didn¡¯t make any sense. But after a few spelling corrections, Google translate was able to convert the sentence into something recognisable in English.

¡°She made recommendations to the priest, man of the house and me and people.¡±

Yeah, that¡¯s a strange way to start an even stranger 240-page manuscript. However, the important bit is that we finally have something understandable from the most intriguing puzzle of recent times. It¡¯ll be a long while before the researchers manage to fully translate the Voynich manuscript, after which they¡¯ll still need Hebrew experts, and maybe even historians, to weigh in on the discoveries to really understand what they mean, However, it¡¯s a start.

Additionally, the team is also planning to apply its algorithm to other ancient texts, showing us yet another way artificial intelligence can be used to solve puzzles and handle tasks far too complex for humans.