Has AI finally cracked the mysterious Voynich manuscript?
Two computer scientists from the University of Alberta claim to have created a series of algorithms that can decipher unknown alphabetic scripts, and to test their system they have targeted the infamously impenetrable Voynich manuscript.
The Voynich manuscript, named after the Polish book dealer who purchased the codex in 1912, has been the source of enormous controversy over the past century. Dated back to the early 15th century, this manuscript was written in an unknown language that many have struggled to decipher over the years. The mysterious codex has been the source of dozens of different hypotheses, from it being either a hoax or gibberish to the suggestion it is written in a complex cipher yet to be cracked by anyone.
Every year it seems someone comes along with a new Voynich hypothesis. Last year, a history researcher made international news by saying he had finally cracked the code. Nicholas Gibbs claimed the manuscript was actually written in an abbreviated version of Latin and translated it as a women's health manual. Critics of Gibbs' interpretation pretty quickly piled on the critiques suggesting his work combined elements of information we already knew with translations that were fundamentally grammatically incorrect.
The latest attempt to decode the mysterious manuscript comes from Greg Kondrak and Bradley Hauer at the University of Alberta. The duo began by using samples from 400 different languages to algorithmically identify the underlying language of the manuscript. Despite initially suspecting the manuscript was written in Arabic, it turned out the algorithms concluded Hebrew was the most likely language.
"That was surprising," says Kondrak. "And just saying 'this is Hebrew' is the first step. The next step is how do we decipher it."
Hypothesizing the manuscript was encoded using alphagrams (alphabetically ordered anagrams), the duo then developed an algorithm that could decipher the text.
"It turned out that over 80 per cent of the words were in a Hebrew dictionary, but we didn't know if they made sense together," says Kondrak.
Taking a closer look at the system's output the duo concluded that the first line of the Voynich manuscript, translated into English after a couple of spelling corrections, reads as "She made recommendations to the priest, man of the house and me and people."
Kondrak suggests that ancient Hebrew historians would still need to work to interpret these translations further as the syntax is quite clearly strange and unusual. Early responses to the duo's work from Voynich specialists haven't been positive according to Kondrak.
"I don't think they are friendly to this kind of research," he recently said in an interview with CTVNews.
It perhaps isn't a huge surprise that Kondrak and Hauer's research is being met with a degree of skepticism. The researchers admit that the Voynich text, as an input ciphertext for their algorithms, is too noisy to generate a fluent output. This means the ultimate value of the work is essentially limited to single word translations. One short section analyzed in the study reveals the Hebrew words for 'narrow', 'farmer', 'light', 'air', and 'fire', leading the duo to suggest that hypotheses the manuscript is a medieval herbal guide could be accurate.
Still, these are far from definitive translations, and the authors reasonably conclude in the study that these results "could be interpreted either as tantalizing clues for Hebrew as the source language of the VMS, or simply as artifacts of the combinatorial power of anagramming and language models."
This new study adds yet another hypothesis to the scores of Voynich claims out there. Kondrak and Hauer plan to continue refining their algorithm and hope to apply it to other ancient manuscripts.
The study was published in journal Transactions of the Association for Computational Linguistics.
Source: University of Alberta