Once again, it is being claimed that researchers have cracked the mysterious Voynich manuscript – the infamous, unreadable, possibly 15th-century botany book – only this time with technology.
Computer scientists have apparently trained artificial intelligence (AI) to decipher the codex that has perplexed historians for over a century. By feeding in over 400 different languages into the computer program, researchers Greg Kondrak and Bradley Hauer were able to use the AI to ascertain which language it most closely correlated with.
Initially, they hypothesized that it was most likely written in Arabic, but it turned out that the manuscript is probably penned in Hebrew instead. “That was surprising,” said Kondrak, whose paper has been published in Transactions of the Association for Computational Linguistics. “And just saying 'this is Hebrew' is the first step. The next step is how do we decipher it.”
Following this, the pair suspected that the words were, in fact, alphagrams; anagrams where the letters have been ordered alphabetically. Developing a program to translate these, they claim that 80 percent of the words are found in the Hebrew dictionary, but as neither of them could find someone to translate the words from Hebrew to see if they make sense, they turned to Google Translate.
This apparently showed that the text was made up of Hebrew words that sort of forms sentences, but doesn’t exactly make perfect sense, such as “She made recommendations to the priest, man of the house and me and people.”
Since its discovery in 1912 by a Polish book dealer – after whom the manuscript is now named – the codex has baffled and intrigued people in equal measure. Handwritten in an unknown language or cipher on vellum that dates to between 1404 and 1438, it consists of 240 surviving pages. These are covered with the text, which is accompanied by detailed and colorful illustrations and charts, many of which are seemingly botanical in inspiration.
Over the years, there have been many high profile declarations that the mystery has been solved, the most recent being only last year, although that was rapidly debunked. Equal skepticism has been cast on this latest “translation” of the Voynich manuscript.
The authors themselves say that their algorithm is limited, and might even be wrong. It is only managing to pick up on individual words within the text, which is too “noisy” to result in “a fluent output”.
They note that in one of the sections seemingly about herbs, the program picks out the words “narrow”, “farmer”, “light”, “air” and “fire”, which would not be out of place in such a chapter. But at the same time, the authors write that “the results presented in this section could be interpreted either as tantalizing clues for Hebrew as the source language of the [Voynich manuscript], or simply as artifacts of the combinatorial power of anagramming and language models.”
Whatever the truth, it seems that the true meaning of the codex will continue to be a mystery.