The recent google translate

The program uses advanced AI-technology, but the simplest tests suggest that it is still far from real understanding

The author of the article is Douglas Richard Hofstadter - American physicist and computer scientist; son of the Nobel Prize in Physics Robert Hofstadter. Gained worldwide fame thanks to the book “Godel, Escher, Bach: This Endless Festoon,” published in 1979 and in 1980, won the Pulitzer Prize in the category of “non-fiction”.

One Sunday, at a weekly salsa class, my friend Frank brought a guest from Denmark with him. I knew that Frank spoke Danish well, since his mother was from there and he lived in Denmark as a child. His friend spoke fluent English, which is considered the norm for Scandinavian countries. However, to my surprise, in the process of communication, it turned out that this couple usually exchanges emails translated by Google Translate (GT). Frank writes a message in English, drives it through GT to get the text in Danish; she, on the contrary, writes in Danish, and then allows GT to translate the text into English. How strange! Why would two clever people who speak each other’s language do such things? My experience with machine translation software has always led me to be skeptical about its capabilities. But these two clearly did not share my skepticism. Many intelligent people are fascinated by the possibilities of translators, and find few reasons to criticize them. It amazes me.

I love languages ​​and passionately translate. I am an expert in cognitive science and have been interested in the subtleties of the work of the human mind all my life, therefore I have been following the mechanization of translations for decades. I first became interested in this question in the mid-1970s when I came across a letter of 1947 written by mathematician Warren Weaver , an early supporter of machine translations addressed to cybernetics founder Norbert Wiener , in which Weaver made an interesting statement, quite famous these days:

When I look at a Russian article, I say: “It’s actually written in English, just encrypted with strange characters. And now I will move on to deciphering. ”

After a few years, he suggested a different point of view: “No intelligent person will ever think that machine translation will ever be able to achieve elegance and style. Pushkin should not be afraid. I recall one unforgettable stressful year of my life spent on translating the brilliant novel in verse by Alexander Pushkin “Eugene Onegin” into my native language (I radically reworked the greatest Russian work, turning it into an English novel in verse), and I find this remark of Weaver much more true than the first, which demonstrably strangely simplified view of languages. Nevertheless, his view of translation as decoding from 1947 became the motto for a long time directing the field of machine translation.

Since then, machine translations have gradually improved, and recently the use of so-called. “Deep neural networks” even led some observers to the idea (see the articles “The Great Awakening of AI ” and “ Machine Translation: Outside of Babylon ”) that human translators are an endangered species. With such a development of events, people who work as translators will in a few years turn into simple quality controllers and proofreaders of minor mistakes, and will not issue fresh texts in their entirety.

Such a scenario would have caused a crushing spiritual revolution in me. Although I understand the desire to teach machines to translate texts well, I don’t want to see how people-translators will replace dead machines. This idea scares me and causes disgust. In my opinion, translation is an incredibly sophisticated art, constantly requiring many years of experience and creative imagination. If, on one “wonderful” day, the translators become relics of the past, my respect for the human mind will be shaken, and this shock will leave me in incredible confusion and sadness.

Each time, reading an article about how the guild of people-translators will soon be forced to bow their heads to the frightening sword of a particular technology, I feel the need to check these statements on my own, in particular because of the fear that this nightmare has already come very close, and also because of hope and desire to convince himself that he is not so close, and, finally, because of my long-held belief in the need to struggle with exaggeration of the achievements of AI. Therefore, after reading about how the old idea of ​​artificial neural networks, recently adopted by a division of Google called Google Brain, and improved with the help of “depth learning”, led to the emergence of a new type of program that supposedly produced a “revolution” in machine translations, I I realized that I need to check the latest version of the GT. Did she really change the field of translations in the way that Deep Blue and AlphaGo managed to do in such venerable games as chess and go?

I found out that although the old GT version worked with a very large assortment of languages, the new one first worked with only nine - true, then expanded the set to 96. I limited my research to such supported languages ​​as English, French, German and Chinese [ add to interest and Russian / approx. trans. ]

Before the discovery of discoveries worth mentioning the ambiguity of the word "deep." When a person hears about how Google bought the company DeepMind, which produced “deep neural networks” with “deep learning”, he immediately perceives these names in terms of values ​​like “outstanding”, “capable of much”, “astute”, “wise”. At the same time, the word “deep” means only the fact that these neural networks have more layers (say, 12) than the older versions, which could have two or three. Does it follow from such a depth that such a network possesses some kind of wisdom? Unlikely. This is a verbal PR.

I am very suspicious of GT, especially because of the hype surrounding it. But, despite the hostility, I recognize the existence of amazing facts related to this product I dislike. It is available to anyone on Earth completely free of charge, and can convert text in any of nearly hundreds of languages ​​into text in any other language. It is worthy of respect. If I proudly call myself “pi-lingual” (that is, the total number of languages ​​I know is a little more than three - this is my playful way of answering the question “How many languages ​​do you know”), then how much more proud can GT be? if he can call himself "buy-in" ("buy" in Mandarin means 100). Piyazychnogo man very impressive bayyazychnost. Moreover, if I copy a page in language A to GT, then it will take just a few seconds to get a page from words in language B. And all this happens all the time, on screens across the planet, in dozens of languages.

The practical benefits of GT and similar technologies cannot be denied, and in principle, they are beneficial for us, but in this approach, something is clearly not enough, and this can be described in one word: understanding. Machine translation never concentrated on understanding the language. Instead, in this area they have always tried to “decipher” - that is, to cope with the task, without bothering about what understanding and meaning are. Could it be that understanding doesn’t need understanding for a good translation? Can someone, a person or a car, give out a high-quality translation without paying attention to the meaning of the language? To shed light on this question, we turn to the experiments I have done.

* * *

I will begin my research with modest examples - with a short remark that immediately causes a clear idea in a person:

In their house, everything comes in pairs. There is a car.

GT Translation:
In their house everything comes in pairs. There is his car and her car, towels and towels, as well as his library and hers.

Human Translation:
In their house, everything has its own pair. There is his car, and her car, his towels and her towels, his library and hers.

The task of translation seems to be unequivocal, but in French (and other Romance languages ) the words “his” and “her”, denoting things, are put not in the owner's kind, but in the kind of thing itself. This is what the GT gave me:

Dans leur maison, tout vient en paires. Il ya sa voiture et sa voiture, ses serviettes et ses serviettes, sa bibliothèque et les siennes.

The program fell into my trap, not understanding how any person would understand that I described a couple who had her thing for each of his things. For example, the neural network with in-depth training used the word sa for both its machine and its machine, so we cannot say anything about the field of the machine owner. Similarly, she used the asexual word ses for the plural of “his towels” and “her towels”, and in the latter case, with libraries, she was confused by the last letter s in hers, and she somehow decided that s here the plural (les siennes). The sentence, translated by GT into French, has lost its meaning.

Then I translated this phrase myself into French, so that the original meaning remained in it. Here is my version:

Chez eux, ils ont tout en double. Il ya sa voiture à elle et sa voirie lui, ses serviettes à elle et ses serviettes à lui, sa bibliothèque à elle et sa bibliothèque à lui.

The phrase sa voiture à elle means “her car”, and sa voiture à lui - his car. After that, I decided that for GT it would be a trivial task to translate my French translation back into English and get the correct English version - but I was cruelly mistaken. Here is what he gave me:

At home, they have everything in double. There is no evidence of that.

I'm sorry, what? Even in spite of the fact that the input sentence unequivocally declares the field of the owners, the translating machine ignored these statements and attributed everything to the male sex. Why did she throw away the most important information from the proposal?

People know a lot about couples, houses, personal belongings, pride, rivalry, jealousy, personal life, and many other intangible things, leading to such whims as towels with embroidery "him" and "her." GT such subtleties are unknown. GT is generally unknown subtleties. He knows only lines consisting of words consisting of letters. He is engaged in ultra-fast processing of pieces of text, he does not think, does not imagine, does not remember, does not understand. He doesn't even know that words mean things. I hasten to say that, in principle, a computer program could find out why a language is needed, and it could have ideas, memory and experience, and she could use them - but the GT was not designed for that. Such goals were not in the plans of its developers.

In general, I laughed at the results, relieved to see that we had not yet come close to replacing translators with machines. But I still thought that it was worth exploring the car in more detail. One sip of thirst is not quench.

And by the way, how about this phrase - “do not quench thirst in one gulp” (hinting, of course, to the saying “one swallow doesn’t make spring”? [ English swallow doesn’t mean a word make ” and as “swallow.” In the English version, one swallow does not make summer / approx. transl. ] Could not resist not to check. And this is what the GT gave me: “Une hirondelle n'aspire pas la soif” [ Russian GT issued "one swallow does not quench thirst" / approx. transl. ]. Grammatically this sentence is French, but difficult to understand. At first there is a bird mentioned (une hirondelle - swallow a), then it says that the bird does not breathe, or does not absorb (n'aspire pas), and finally, it is indicated that what the bird does not absorb is thirst (la soif). GT clearly did not understand the meaning of the phrase - just issued a bunch of bull shit. “Il sortait simplement avec un tas de taureau.” “He just went out with a pile of bulls.” “Il vient de sortir avec un tas de taureaux.” Forgive my French - or, more precisely, pseudo-French GT.

* * *

From the fire of the French we jump into the fire of German. Recently, I have been immersed in the book Sie nannten sich der Wiener Kreis (They called themselves the Vienna Circle) of the Austrian mathematician Carl Siegmund. She describes a group of idealist intellectuals from Vienna in the 1920s and 1930s who seriously influenced philosophy and science in the second half of the 20th century. I chose a small sentence from Sigmund’s book and gave it to GT. Here it is, first in German, then my translation goes, then GT. (By the way, I checked my translation with two German speakers, including Karl Siegmund - so that we can consider it accurate).

Sigmund:

Nach dem verlorenen Krieg sahen es viele deutschnationale Professoren, inzwischen die Mehrheit in der Fakultät, gewissermaßen als ihre Pflicht an, die Hochschulen vor den “Ungeraden” zu bewahren; am schutzlosesten waren junge Wissenschaftler vor ihrer Habilitation. Und Wissenschaftlerinnen kamen sowieso nicht in frage; über wenig war man sich einiger.

Hofstadter:

It has been taken into account that they have been young scholars. who had not yet earned university classes. As for female scholars, well, they had no place at all; nothing was clearer than that.

Google translate:

After the loss of the war Young scientists were most vulnerable before their habilitation. And scientists did not question anyway; There were few of them.

GT, Russian:

After the lost war, many German-national professors, now the majority in the faculty, considered it their duty to protect universities from the "strange"; the most vulnerable were young scientists before their habilitation. And women scientists still did not doubt; there were several of them.

Human Russian:

After the defeat, many professors who were prone to German nationalism (and by that time the majority of them were in the leadership of the faculty), considered it their duty to protect higher educational institutions from “undesirable elements”. Most likely, young researchers, who have not yet earned the right to teach, could be refused. For female scientists, there was no place in this system at all; it was clearer than that.

GT translation consists of English words (for some unknown reason, for some reason, a couple of them begin with a capital letter). So far so good! But soon the whole translation begins to blur, and the further it goes, the more blurry it becomes.

Take first the “odd” [eng. odd - strange, odd, unpaired]. This corresponds to the German die 'Ungeraden', meaning "people who are undesirable for political reasons." But GT had a reason - pure statistics - to use the word "odd". Namely: in its huge bilingual database, the word ungerade is almost always translated as odd. Although the machine does not understand why, I can explain. This is because ungerade - literally meaning "indirect" or "uneven" - almost always means "odd." And my choice of the word undesirables [unwanted elements] has nothing to do with the statistics of words, but comes from my understanding of the situation - understanding of an idea that is not mentioned in the text directly and not present in the list of options for translating the word ungerade in any of the German dictionaries.

Let us turn to Habilitation, denoting the status of a teacher, resembling a staff member of the institute. English tracing paper habilitation [qualified] exists, but is used very rarely, and certainly does not remind of full-time employees. Therefore, I briefly explained this idea, and did not simply use the rare word, because such a mechanical approach would have given nothing to English-speaking readers. Of course, GT will never do that, it doesn’t have the knowledge model of its readers.

The last two sentences clearly demonstrate the critical importance of understanding for proper translation. The 15-letter German word Wissenschaftler means “scientist” or “scientist” [“scientist” / “scholar”]. I chose the latter option, as it denotes all intellectuals in general. GT did not recognize these subtleties. The 17-letter related word Wissenschaftlerin in the final sentence, in the plural, Wissenschaftlerinnen is a consequence of the use of childbirth in German nouns. A short noun grammatically refers to a male gender, therefore it means a male scientist, and a long one refers to a female, and applies only to women. Therefore, I wrote "female scientists." GT, meanwhile, did not understand that the main suggestion was the female suffix -in. Since he did not understand that women were excluded from consideration, the machine simply once again used the word "scientist", completely missing the whole point of the sentence [ interestingly, the GT translated it into Russian more true / approx. trans. ]. As in the case of the French, the GT had no idea that the sole purpose of the German proposal was to reveal the contrast between men and women.

Well, apart from this error, the remainder of the last sentence is a complete nightmare. Take the first part. Will “scientists not question anyway” really be a translation of “Wissenschaftlerinnen kamen sowieso nicht in frage”? They are completely different in meaning. The sentence simply consists of English words inspired by the German. And what, is this enough to describe some kind of output text, as a “translation”? [ in the translation version of the Russian nightmare remains / approx. trans. ]

The second part is just as wrong. The last six words in German literally mean “little over what was more united”, or, more smoothly, “little existed to those about which people's agreement was stronger” [“over little is one more united” / “there was little about which people were more in agreement ”] - but the GT turned this clear idea into“ there were several of them ”[ in English and Russian are about the same / approx. trans. ]. Perplexed people may ask “who was several?”, But for a mechanical listener this question would have no meaning. GT has no idea what's going on behind the scenes, and he could not have answered this seemingly simple question. The translation program did not imagine large or small quantities or numbers or things. She was simply thrown by symbols, without the slightest idea that they could symbolize something.

* * *

It is very difficult for a person who has gained experience, understanding, and used words intelligently, to understand how much the contents of words that are thrown on the screen by the GT machine are lacking. It is almost impossible for people to resist the assumption that software that works so well with words must understand their meaning. This classic illusion associated with AI is called the " Eliza effect ", since one of the first programs that made people believe that she seemed to understand English, back in the 1960s, the useless manipulator with the phrases "ELIZA" pretended to be a psychotherapist. Many people who interacted with the program had a supernatural feeling that she understood their deepest feelings.

For decades, intelligent people — even AI researchers — fell under the influence of the Eliza effect. So that my readers will not fall into this trap, let me quote some phrases from the text above - “GT did not understand,” “did not realize,” “GT did not have the slightest idea.” It is ironic that, although these phrases reiterate a lack of understanding, they almost hint that a GT can, though sometimes, understand the meaning of a word, phrase or sentence. But it is not. GT simply bypasses the question of understanding the language.

For me, the word "translation" has a mysterious aura, evoking memories. It denotes an extremely human form of art, gracefully transforming the clear ideas of language A into the clear ideas of language B. This networking not only should maintain clarity, but also convey the nuances, nuances and distinctive features of the original author’s writing style. When I do the translation, I first carefully read the original text, assimilate ideas as clearly as possible, let them shake in my mind. It’s not the words that are floundering in my mind - ideas causing all sorts of related ideas, creating a rich halo from the accompanying scenarios in my head. Most of this halo, of course, is in the unconscious. Only when the halo in my mind is sufficiently awakened, do I begin to try to express it - “squeeze” - in a second language. I try to say in language B that which seems natural to language B to talk about such situations that make up the halo of the meaning in question.

In general, I don’t turn from words and phrases of language A to words and phrases of language B. Instead, I unconsciously summon images, scenes, ideas, experiences that I have (or experience that I read about, saw in movies, heard from friends ), and only when it is non-verbal, imaginative, experienced, mental “halo” - only then I start the process of formulating words and phrases in the target language, and then I edit, edit and edit. This process, the mediator of which is the meaning, may seem slow - and, of course, compared to two seconds per page at GT, it is the same - but it is used by any serious translator. Something like that I imagine when I hear the phrase "deep mind."

* * *

Taking all this into account, I’ll switch to Chinese, a language that is much more intense with deep-seated learning software than the two European languages ​​do. To test, I chose the touching memoirs of the Women Sa (“We Are Three”) - a Chinese playwright and translator Yang Jiang , who recently died at the age of 104 years. In the book, she recalls the interweaving of her life, her husband Qian Zhongshu (also a writer and translator) and their daughters. It is written not particularly intricate, but uses a lively and highly educated Chinese. I chose a small passage and set GT on him. Here are the results, along with my translation:

Jan:

书 到 清华 工作 一年 后 ， 调任 毛 选 翻译 委员会 的 工作 ， 住 在 城 里 ， 周末 回 校。 他 仍 兼管 研究生。
选 翻译 委员会 的 领导 是 徐永 煐 同志。 介绍 锺 书 做 这份 工作 的 是 清华 同学 乔冠华 同志。
定 之 日 ， 晚饭 晚饭 后 ， 位 一位 旧友 特 雇 黄包车 从 城 里 客 锺 书 惶恐 说 ：

以为 我 要做 “南 书房 行走” 了。 这 件事 不是 好 做 的 ， 不 求 有功 ， 但 求 无 过。

Hofstadter:

He has been working for the Mao. He lived in the city He also was still supervising his graduate students.
It has been noted that it has been a school year, it has been a school of commerce, comrade Qiao Guanhua.
He was a little girl, and he was on his own. After our guest had left, Zhongshu turned to me uneasily and said:
“He thought I was going to become a 'South Study special aide.' This kind of work is not easy. You can't hope for glory; errors. ”

Google translate:

It was transferred to him. He is still a graduate student.
Mao Xu Yongjian. Tsinghua students Qiao Guanhua comrades.
After eating a night out from the city to congratulate you. The book of fear of the go
He thought I had to “South study walking.” It’s not a meritorious service, but for nothing.

GT, Russian:

After one year of work at Tsinghua University, Zhong Shu was transferred to the Mao Translation Commission to live in the city and return to school on weekends. He still holds the post of graduate student.
The leader of Mao’s elected Translation Committee is Comrade Xu Yongliang. Introducing Zhong Shu, to perform this work, is Qinghua's classmate Qiao Guanghua.
On the appointed day, after lunch, an old friend hired a rickshaw to come from the city to congratulate him. After the guests left, the book Zhong told me in fear:
He thought that I needed to make a “south walk”. This is not good.

Human:

When Zhongshu worked for a year at Tsinghua University, he was transferred to a committee engaged in translating the work of Chairman Mao. He lived in the city, but every weekend he returned to school. In addition, he still supervised his graduate students.
The leader of the Mao translation committee was Comrade Xu Yunliang, and Zhongshu was appointed to this position by his old classmate in Tsinghua, Comrade Tsinghua Qiao Guanghua.
On the day when this appointment was approved, after dinner, an old friend specifically hired a rickshaw and came from the city only to personally congratulate Zhongshu. After the guest left, Zhongshu turned to me, and said uneasily:
He thought I would become a “special assistant from the Southern Cabinet.” This is a tough job. Hope for glory is not necessary; one can only hope for the absence of errors.

I will briefly describe the strangeness. First, the GT never mentioned Zhongshu by name, although his name (“锺 书”) is found three times in the original. First, the machine uses the pronoun "he", the second time it writes the book [book], and the third - the book of fear in the book. Go guess! [ in the Russian version of the mysterious "book" appears once / approx. trans. ]

The second oddity - in the first paragraph it is clearly stated that Zhunshu supervises graduate students, but GT made him a graduate student.

The third oddity is that in the phrase Mao Tse Translation Committee [the Mao Zedong Translation Committee] a third of the name of Chairman Mao [Mao Tse Tung] has fallen off somewhere.

The fourth oddity - the name Yongying is replaced by Yongjian.

The fifth oddity - after our guest had left [after the departure of our guest] was simply replaced by guest to go.

The sixth oddity - the last sentence is completely meaningless.

Six of these oddities are enough to shame the GT, but we'll forgive them and forget. We better focus on the confusing phrase I met in the text - a five-character phrase enclosed in quotes from the last paragraph (“书房 行走”). Symbolically, it can be translated as "the southern room of books to go for a walk," but this is clearly unacceptable, especially since it must be a noun in context. GT invented the “walking south office”, and it did not help.

I admit that the Chinese phrase was completely incomprehensible to me.Literally, it meant something like walking in an office on the south side of a certain building, but I knew that it was wrong - in context it did not make sense. To translate, I needed to find something unknown to me in Chinese culture. And where did I go for help? Go to google! (But not to google translate). I typed Chinese characters, surrounded with quotes, ran the search by exact match. And a bunch of web pages in Chinese jumped out onto the screen headlong, after which I painfully smashed through the first paragraphs of a couple of first sites, trying to understand the meaning of the phrase.

I discovered that this term dates back to the Qing Dynasty(1644–1911), and designates a knowledge worker who helped the emperor compile stylish official communications, and who worked in the southern office of the imperial palace. Two characters, ostensibly denoting "go for a walk," in fact, mean helper. Therefore, using the information from Google search, I came up with the phrase "special assistant from the southern cabinet."

It is a pity that GT itself can not use the Google Search service, as I did. But again, GT is not able to understand web pages, although it is able to translate them into an eye. Or maybe? Then I bring an amazing piece of text that GT instantly threw to me on the screen after I fed him the beginning of the website, from which I got the necessary information:

“South study walking” is not an official position, before the Qing era this is just a “messenger,” generally by the then imperial intellectuals Hanlin to serve as. South study in the Hanlin officials in the “select chencai only goods and excellent” into the value, called “South study walking.” Because of the close to the emperor, the emperor's decision to have a certain influence. Yongzheng later set up “military aircraft,” the Minister of the military machine, full-time, although the study is still Hanlin into the value, but has no participation in government affairs. Scholars in the Qing Dynasty into the value of the South study proud. Many scholars and scholars in the early Qing Dynasty into the south through the study.

Is it generally in English? Of course, we can all agree that the text consists of English words (for the most part), but does it follow from this that this is an English text? In my opinion, since this paragraph makes no sense, it is not English; it's just randomly arranged English ingredients - a random salad of words, an incoherent mess.

In case you are interested, here is my version of this passage (I poured over it for several hours):

Должность «нан шу фан сяон зу» (особый помощник из Южного кабинета) была не официальной, но в ранней династии Цин это была особая роль, которую обычно играл текущий придворный учёный. Группа учёных, работавших в южном кабинете имперского дворца, выбирала из своей среды человека великих талантов и хорошей репутации, дабы он писал речи для императора и всегда был у него на посылках; поэтому эта роль называлась «особый помощник из Южного кабинета». Помощник, будучи приближен к императору, очевидно был способен влиять на его политические решения. Однако, после того, как император Юнчжэнfounded the war ministry, with the minister and his various subordinates, an assistant from the Southern Cabinet, despite serving the emperor, ceased to play a major role in government decisions. Nevertheless, scholars of the Qing dynasty sought with all their might for the glory of this position, and in the early years of the dynasty several famous scholars were special assistants.

Some readers may suspect that, for the purpose of harsh criticism, GT specifically selected such passages on which it actively stumbled, and that in fact it copes with most texts much better. It sounds true, but it is not. Almost every paragraph I chose from the books I am reading now led to translation errors of all kinds and kinds, including such meaningless and incomprehensible phrases as mentioned above.

Of course, I admit that sometimes GT gives out a few sentences that sound quite good (although they can be misleading or just make mistakes). It may even happen that a paragraph or two will turn out great, creating the illusion that the GT understands what it does, understands what reading means. In such cases, the GT looks impressive - almost human! And all praise definitely relate to its creators and their hard work. But at the same time, do not forget what GT did with these two Chinese passages, and earlier with both French and German passages. To understand these mistakes, you should not forget about the "Eliza effect." Byyazychnaya machine does not read anything - not in the human sense of the verb "read." It processes the text. The characters to be processed are not related to worldly experience. She has no memory from which to draw anything, no imagination,neither understanding nor meaning behind the words with which she operates so quickly.

* * *

A friend asked me if GT skills could be considered a simple database function. He estimated that if the database was increased by a million or a billion times, in the end, it would in principle be able to translate everything they give, and in fact is perfect. I do not think so. Increasing the amount of “big data” will not bring you closer to understanding, because understanding depends on the availability of ideas, and the lack of ideas is the root of all the problems of modern machine translation. So I am sure that larger databases - even much larger ones - will not help here.

Another natural question is whether the use of neural networks in GT - imitations of the brain - will bring us closer to a true understanding of the machines of the language. At first it sounds true, but so far no one has attempted to get beyond the surface level of words and phrases. All sorts of statistics about the huge databases are embedded in the neural network, but these statistics simply associate some words with others, and not with ideas. There are no attempts to create internal structures that represent ideas, images, memories or experiences. Such mental constructions are too elusive for computational methods, and therefore fast and complex algorithms for the statistical accumulation of words are used instead. But the results of such techniques can not be compared with in order to actually have ideas that appear when someone reads, understands, creates,modifies and judges the text.

But despite my negativity, GT provides a service appreciated by many: it quickly converts meaningful sentences written in natural language A into not necessarily meaningful lines of words in B. And while the text in B is more or less understandable, many People result is fully satisfied. If they can get the “main idea” of a sentence written in a language they don’t know, they’re excited. For me, this is not at all what the word "translation" means, but for some people it is a great service, and for them it is a translation. Well, I can understand what they need, and that they are happy with it. Lucky for them!

Recently, I have watched bar charts drawn by technophiles who claim that they represent the quality of the translation made by people and computers, and that these graphs show how closely machine translation came close to human. From my point of view, such a quantification of concepts that cannot be quantified smacks of pseudoscience, or, if you like, neryds trying to bring to mathematics those things whose intangible, subtle, artistic nature eludes them. In my opinion, today's conclusion GT ranges from excellent to grotesque, but I cannot evaluate my feelings about this numerically. Remember my first example, using the concepts of "his" and "her." The thoughtless program translated almost all the words correctly, but despite this success, I did not understand the meaning at all.And how in this case to express the quality of work numerically? The use of scientific bar graphs to represent the quality of translation is simply an abuse of the external features of science.

Let us return to the sad picture, in which people-translators will soon be left behind, become old-fashioned, and eventually will be engaged in proofreading. In the best case, the result will be a certain mediocrity. A serious artist does not begin to work with kitsch garbage, full of errors, so that, after making a couple of changes, to issue a work of art. This is not the nature of art. And translation is an art.

In my works written over many years, I have always supported the point of view on which the human brain is a machine; very complicated car; and I vigorously argued with those who said that machines are inherently incapable of understanding. There is even a school of thought that states that computers will never “have semantics”, since they are composed of “not from that material” (silicon). As for me, this is superficial nonsense. Here I will not indulge in debate, but I do not want to impress the reader that I believe that intelligence and understanding will never be available to computers. If this essay pushes such thoughts, then it is only because the technologies I discuss make no attempt to reproduce human intellect. On the contrary: they are trying to circumvent the intellect,and the resulting passages show these gaping gaps.

From my point of view, there is no fundamental prohibition on cars ever starting to think, becoming creative, funny, nostalgic, joyful, scared, enthusiastic, submissive, full of hope, and as a result, could make good translations between languages. There are no fundamental reasons why someday they could not successfully translate jokes, wordplay, scripts, novels, poems, and essays like this. But all this will appear only when the machines are filled with ideas, emotions and experiences, like people. And this is not yet visible. I believe that this is still very far away. At least, this person is fervently hoping for this, who has admired his whole life the depth of the human mind.

When, one day, a mechanical translator will compose a masterful novel in verses in English, using a precise rhythmic four-stop iambus, rich in thoughts, sensuality and lively lines, I understand that it is time for me to bow out.