Get best of the week

Voynich Code: Imaginary Triumph of Artificial Intelligence

The area of ​​interest for employees and teachers of the EnglishDom online school of English is much broader than just English. The mysteries of linguistics are also interesting to us. Recently, a controversy broke out in our office about the Voynich code, and we decided to make an article on this topic.



The Voynich manuscript is one of the most burning mysteries of linguistics and cryptography, which has not been solved to this day. For 600 years, even the best minds in the world cannot come close to unraveling this mysterious text.

In 2016, researchers connected a neural network to the solution. The result was unexpected - the computer analyzed the text and made a mistake. Read more about this.

The Voynich manuscript is an illustrated handwritten code that is written in an unknown language or code.

According to the results of carbon analysis, the book was written in the first half of the 15th century. 240 pages of parchment covered with strange letters that look like text. But the difficulty of deciphering it is that the book uses an unknown alphabet that does not correspond with any existing or studied existing language.

A detailed analysis of the text allows us to determine that the letters obey certain grammar rules, but the rules themselves cannot be determined. There are practically no one- or two-letter words in the text, which are many in Latin-based languages; individual principles of writing words remotely resemble Arabic script or Hebrew. Individual words are generally repeated several times in a row. In general, the structure of a language or cipher cannot even be roughly determined - it is too different from all the principles of writing written speech that are familiar to us.

The only thing that linguistic experts have been able to determine for almost 600 years is that the informational entropy of the code is approximately equal to the entropy of English and Latin. This means that the text is definitely not a set of random characters, but carries a certain meaning.

In theory, it may even be encrypted English, but how can one find out if the researchers still cannot determine if the manuscript itself is a cipher or just some strange language?

Even with a key, deciphering the principles of a language requires tremendous effort on the part of linguists. Deciphering the Rosetta stone took researchers 20 years. And this provided that they knew one of the three languages ​​in which the text was written in stone.

Just imagine, even knowing the translation of the ancient Greek text, it took researchers more than two decades to decipher the same text written in hieroglyphic writing. The demotic letter was deciphered earlier, but it is striking that the fact that having the key, the essence of the language was unraveled for so long.


The Voynich manuscript also contains short fragments of the text, which are knocked out of the total. Separate words written in Latin letters with combinations of unknown characters.

However, these inscriptions are either encrypted or written according to the rules of an unknown language. Because it is impossible to translate them. In any case, the researchers say so.

Theories about decoding Voynich manuscript


For 600 years, researchers have heaped up a whole bunch of theories of the origin of the language and alphabet of the book. There are quite strange ones, there are noteworthy ones.

Most scholars until the 20th century believed that Voynich’s manuscript hides just one of the European languages ​​in a special way .

But the text does not correspond to the ciphers that existed in the 15th century. Substitution, polyalphabetic, nomenclator, and homophonic ciphers are not suitable.

It is possible that the text was encrypted with one of the above ciphers, and then complicated with the help of false characters and spaces or another level of encryption, but this hypothesis is extremely difficult to verify - because in this case it is impossible to track which characters are false and which are true .

The second popular hypothesis states that the Voynich code is a common codebook cipher . That is, a separate combination of characters is a separate word in an existing language. Indeed, the form of the manuscript suggests that the text has a very definite meaning. But today it is impossible to confirm or refute this hypothesis - to crack such a cipher is possible only with the help of a dictionary.

Some researchers believe that the manuscript is written in a real exotic language with a unique alphabet . For example, in one of the eastern or American dialects. Some stylistic features of the text hint at this, but evidence for this is still not enough.

There are still many considerations: artificially created unique language, multilingual encrypted text, proto-language, which preceded all languages ​​of the Romanesque group. There were even thoughts that the manuscript was written by a madman and did not make sense at all. Researchers also tried to prove that the manuscript is a hoax, but radiocarbon analysis still shows that the book was really written in the 15th century.

None of the hypotheses has yet received sufficient evidence of their innocence. Therefore, the Voynich code has not yet been solved.

Neural network is trying to crack the Voynich code


So, after a voluminous and broad introduction, we turn to the essence of the article. In 2016, they tried to hack Enigma from the world of literature using a neural network. Yes, it was in 2016 - the media learned about these attempts only in 2018, because of this date they are often confused. Here is a link to the original study . The text is in English, so you need at least a little understanding of scientific terminology.

Canadian scientists have "trained" the neural network to recognize individual elements of the alphabet and tokens from 380 existing or previously existing world languages. According to the researchers, the accuracy of the analysis of the neural network was within 97%.

The system showed that the most likely manuscript language is Hebrew. Of course, not simple Hebrew, but with a sub-subscript. Researchers have suggested that the book has a fairly simple cipher, in which vowels are omitted or encrypted with other characters, and consonants are placed in alphabetical or random order.

It is also worth clarifying that the system also provided other possible sources: Mazatec (the native language of the modern south of Mexico), Mozarabic (Arabicized language of the Iberian Peninsula), Italian and Ladino (the language of the Jews of the Iberian Peninsula). Also, the neural network found elements of the standard Arabic and Amharic language (the territory of modern Ethiopia, part of the Semitic group).

Such an approach suddenly yielded results and the neural network was able to translate part of the text of the book. The first phrase was translated as:

She made recommendations to the priest, man of the house and me and people.
She gave advice to the priest, the owner of the house, me and the people.

It would seem, here it is, the triumph of artificial intelligence! Based on this interpretation and illustrations, the researchers even made the assumption that the Voynich manuscript was a kind of pharmacopeia - a medical book that described the healing value of herbs, methods of manufacturing and use of drugs, and the structure of the human body.

In total, the algorithm “recognized” approximately 80% of the words from the entire manuscript. The analysis was based on the same assumption about the absence of vocalizations and the arbitrary order of letters in words.

But repeated checks of the first test phrase showed a different result:

And the priest made a man for him to his house, and to his men.
, .

Unleavened bread and made her the priest, and one which leaves his home.
, , .

The phrases make less sense than the original version, but in theory this can be attributed to the imperfection of the translation algorithms of the system. In general, the lexical foundations in all versions of the translation remained unchanged: “priest” and “house”.

One could claim success, but there are a couple of serious “buts” that do not make the results of the study sensational.

Firstly, the settings of the neural network allowed a certain liberty in the interpretation of words, because even if you take into account that the alphabet is just a changed type of Hebrew letters, there are quite a few variants of words that can be made up by rearranging the letters.

If we assume that the language of the manuscript is not Hebrew, but simply belonging to a Semitic group or related to it, then a perfect analysis will not make sense - there are too many options for analyzing even those characters whose value seems to have already been determined. And there are even more unknowns.

In this situation, I want to recall the theorem on endless monkeys. If anyone has not heard, here it is:

Suppose we have an infinite number of monkeys with typewriters, each of which randomly taps the keys for an unlimited amount of time.

Sooner or later, one of the monkeys will be able to “trick” any arbitrary text: be it a short note or “War and Peace”.

This theory can be applied if the text is interpreted by a neural network. Initially, the neural network itself creates a pool of variants of the meaning of each word, and then from the entire pool of variants it selects the most possible interpretations based on combinations with neighboring variants.

As a result, in a sentence of 5-8 putative words we get several tens of thousands of options, of which the neural network chooses the one that carries the most meaning.

That is, there is a very high probability that among these disparate options there will accidentally be one or more that will really make sense. Moreover, if there is a more complex cipher or other lexical structure of sentences or words, then the method turns out to be false positive.

In fact, there is some result, it can be "felt" and presented to the public, but there is no sense from it, because it does not take a step closer to the real solution of the cipher.

And objectively combining the style of the letters of the alphabet with Hebrew is a rather unusual solution. However, most scholars of the manuscript doubt that the original language of the manuscript is Hebrew. The lexical structure does not coincide quite strongly, and it is still not possible to analyze the degree of encryption, if any.

Moreover, some believe that linguists with a neural network did not conduct an objective analysis, but sought confirmation of a separate theory. The hypothesis that the book is a pharmacopeia can be made based on drawings of herbs, people, and star bodies, even without analyzing the text.

As a result, the research results were not accepted in the scientific community. Because they do not reveal the specific features and principles of the language, as required for a full-fledged linguistic study of adverbs. In order for the research results to be recognized, there is a corny lack of evidence. It is impossible to trace a clear logical chain that guided the neural network during the analysis, so the results cannot be considered scientifically sound - there is a non-zero chance that the chain will turn out to be erroneous.

However, there were no more adequate hypotheses about the Voynich manuscript.

Linguists tried, but they all look more like farce. For example, in 2019, one British scientist stated that he had unraveled the Voynich code. But the theory of the “Protoromanian language” or vulgar Latin was sharply criticized by scholars who accused the British of artificially choosing words without defining the principles of writing and without convincing arguments about the lexical connections between meanings.

Now it’s already 2020 and the hype around the “only and correct decoding of the Voynich manuscript” has ceased. He still continues to be considered one of the main linguistic and cryptological puzzles of our time.

Of course, I would like to believe that someday they will solve it all the same. If this is some kind of language, then it is entirely possible. But if this is still a cipher with a lost key, then the manuscript risks forever remaining just a beautiful ancient book with a mysterious history.

In general, linguistic puzzles are a very cool topic. Crosswords and puzzles - this is just the tip of the iceberg - there are so many ways to simultaneously learn English and pump logic and thinking. EnglishDom teachers often use them in class to diversify the learning process and make it more interesting.

EnglishDom.com online school - inspire you to learn English through technology and human care




Only for readers of Habr the first lesson with the teacher on Skype for free ! And when you purchase classes, get up to 3 lessons as a gift!

Get a whole month of premium subscription to the ED Words app for free .
Enter the Voynich promotional code on this page or directly in the ED Words app . The promotional code is valid until 01/30/2021.

Our products:

Source: https://habr.com/ru/post/undefined/

More articles:


All Articles