Home > News content

Google was professionally translated diss: does not speak human

via:博客园     time:2018/2/19 10:34:03     readed:873

Sina Technology Jiang Yi group

Spring Festival holidays have been a few days, do not know if you have any good to eat? Friends circle these days is almost food, cute baby videos and sights photos brush screen, people envy envy.


Circle of friends (to protect the privacy of some information code)

I do not know if you noticed that the message from the four friends above was sent out in Russia. In fact, quite a number of people chose to travel abroad as a whole to go back to their hometown during the traditional days of returning to their hometowns. According to the China Travel Research Institute and Ctrip, the number of Chinese outbound tourists is expected to reach 6.5 million this Spring Festival. Chinese people travel abroad during the Spring Festival holiday almost "sweeping the globe."

But traveling abroad has a problem: how to communicate locally? This is not a problem that can be solved by a few words of English in the optical system. If you are going to a country that is not a native speaker of English, you may be in trouble. Not every native speaker can speak English. At this time, the translation software on the phone is useful.

The most popular travel abroad is google translate. For users in mainland China, the most important update recently is that on March 29 last year, Google's updated translation App can be used in the Chinese mainland without any problems. After a wave of publicity, more and more people are using it. In the know "on the travel abroad what good portable translator or App available? "In this question, the most reputed answer is Google Translate App.


Knowing what's going on? "What's the best portable translator or app available? "In this question, praise the most answer

But machine translation has always been despised by professional translators, even though Google has not changed the perception of many human translators as a translational blessing neural network. In an article recently published in the Atlantic Monthly in the United States, Douglas Hofstadter wrote a diss (insult) to Google Translate. He translates into four languages: English, French, German and Chinese, demonstrating one result: Google Translate does not really understand the language even with AI.

In the article, he cited an example of Chinese translation of English, the text is a section of Yang Jiang's "We 仨".


After one year of working in Tsinghua University, Zhong Shu was transferred to Mao's translation committee to live in the city and back to school on weekends. He still holds the post of graduate student.

The leader of the Mao Selected Translation Committee is Comrade Xu Yonglian. Introducing Zhong Shu to do this job is Tsinghua classmate Qiao Guanghua.

On the appointed day, after dinner, an old friend hired a rickshaw to come from the city to congratulate. After the guests go, Zhong book said to me in fear:

He thought I had to do "Southern study walk". This is not a good thing to do.

Google Translate version:

After a year of work at Tsinghua, he was transferred to the Mao Translating Committee to live in the city and back to school on weekends. He is still a graduate student.

The leadership of the Mao Tse Translation Committee is Comrade Xu Yongjian. Introduction to the book to do this work is Tsinghua students Qiao Guanhua comrades.

On the day of the event, after dinner, an old friend hired a rickshaw from the city to congratulate. Guest to go, the book of fear in the book said to me:

He thought I had to do & ldquo; South study walking. & rdquo; This is not a good thing to do, not for meritorious service, but for nothing.

Author translation version:

After Zhongshu had worked at Tsinghua University for a year, he was transferred to the committee that was translating selected works of Chairman Mao. He lived in the city, but each weekend he would return to school. He also was still supervising his graduate students.

The leader of the translation committee of Mao & rsquo; s works was Comrade Xu Yongying, and the person who had arranged for Zhongshu to do this work was his old Tsinghua schoolmate, Comrade Qiao Guanhua.

On the day this appointment was decided, after dinner, an old friend specially hired a rickshaw and came all the way from the city just to congratulate Zhongshu. After our guest had left, Zhongshu turned to me uneasily and said:

& ldquo; He thought I was going to become a & lsquo; South Study special aide. & rsquo; This kind of work is not easy. You can & rsquo; t hope for glory; all you can hope for is to do it without errors. & rdquo;

The author pointed out here a lot of problems Google version:

The translation once again referred to the names of "Chung Shu", but were translated into "he", "the book" and "the book of fear in the book". Obviously, Google Translate does not understand that this is a personal name;

The original is "He still holds the post of graduate student", but the translator becomes "He is still a graduate student" (he is still a graduate student);

The phrase "Mao Tse Translation Committee" in the Google translation missed the last word in the name of Chairman Mao Tse Tung, and "it was obviously not translated;

The original "Xu Yongying" was translated into "Xu Yongjian".

"Guest go after" was translated into "guest to go" (guest to go).

"Southern study walk" in "walk" was translated into "walking", which is why Google did not understand the original meaning.

Not only in this paragraph, the author selected a number of paragraphs from his book to test, almost every paragraph will appear in different forms, the size of the translation errors, including the incomprehensible sentence above.

Why Google Translate in the above example is still not satisfactory? In the author's opinion, Google Translate is just like dealing with the same password processing statements, but not as human as "read" statement, it handles the symbols and the real world is separate, there is no realistic experience, it will not understand.

Google Translate is less than satisfactory, but is it really useless?

Since Google Translate is so bad in the previous example, is Google Translate useful to overseas travelers?

Hofstadter, the author of a previous Google translation of diss, pointed out that Google Translate is not yet able to accurately understand the sentence; but Hofstadter only analyzed a type of text, that is, literature class. Google translation is faced with more than just want to understand literary works, for different needs, its performance will not change? In other words, is Google Translate better for overseas travelers? A simple test may be able to see it.

Sina science and technology in Baidu "out of English XXX sentences" randomly selected in the 10 common sentences abroad as a sample, enter the iOS version of the Google App App test (test statements here in addition to the common types of questions, as well as Relatively complex), and then divided the 10 original texts and translation into three categories, namely "ambiguous", "infrequently used" and "correct" types, with the following results:


Classification results

It should be noted here that the classification here is based on the degree of comprehension, not grammar. This means that even with obvious grammatical errors, as long as they are understood without any ambiguity, they can be assigned to the "infrequently" category, such as "How to get to the subway station" The sentence, but also through word understanding, it is assigned to "not commonly used" category.

But even if classified by such less stringent criteria, the ratio of understandable to misleading is still 6: 4,Here are four "ambiguous" sentences may appear misunderstood:

"I want to withdraw this ticket" translated into "I want to withdraw this ticket", may be misunderstood as "I want to take this ticket" because "withdraw" has "taken away" and "rdquo; ; The meaning of, and no "refund" meaning;

"I want to put this $ 50 in exchange for" in exchange for "can not be translated into" swap ", because this English word can not be used for" change "in this sense;

"Can you give me a rebate form? & rdquo; Google Translate & "can you give me a tax return? "The meaning of" form "is lost in translation, and" tax return "means" tax return "rather than" tax rebate "

"Do you have duty-free shops? "Google Translate mistranslated" Is this duty-free shop? "(Is this shop a duty-free shop?) This is a mistake that anyone with a little knowledge of English can see.

From the random test results, the accuracy of Google translation is still a little worse than meeting the needs of tourists.

Why artificial intelligence blessing, Google translation is not good enough?

As mentioned earlier, Google is more important recent update is accessible in mainland China. But in fact, a more significant update from the previous year was Google's move to translate the translation system from its original "phrase-based machine translation" (PBMT) to "neuromechanical translation" (NMT) Neural network technology is introduced into machine translation. Google claims that after this update, the translation will be more streamlined and more usability.

According to Baidu Encyclopedia's theory, neural network "is a simulation of animal behavior characteristics of neural networks, distributed parallel processing of information mathematical mathematical model". This definition is not easy to understand non-professionals, it does not matter, as long as the neural network to know the basic technology to achieve artificial intelligence just fine.

Artificial intelligence is the hot news key word and the most popular technology in recent years. Even this year's Spring Festival Gala presented an autonomous vehicle driven by artificial intelligence technology. But how often does artificial intelligence in the news that speaks poorly in language translation? Hofstadter, the author of the google google translation in the previous article, gave us an explanation.

Hofstadter believes that many people in fact the expectations of artificial intelligence is too high, and thus "excessive brain recovery", which is psychologically referred to as "Eliza effect", probably means that people will interpret The result of the machine Eliza "is an early AI project in the 1960s that understands simple natural languages ​​through scripts and can interact like humans, and many people who interact with it think Eliza really understands their heart a feeling of. To Hofstadter, even some researchers in the field of artificial intelligence will inevitably fall into the trap of "Eliza" effect.

According to Hofstadter, human translation is a process of "reading - internalization - expression". After human being reads the text, he or she will associate with the relevant realistic scene and then give another language based on the scene Kind of scene expression. But Google Translate is just a "decoding" process. Even artificial AI blessings, Google Translate does not really understand the text, it just treats the original as an encrypted translation, as long as the decoding can be translated. The lack of realistic experience of such a process can not be understood, of course, inevitable mistakes.

Someone may ask, can not neural network technology imitate human thinking to translate it? Hofstadter also gave his own answer, he believes that such machine translation is not beyond the word and phrase level, this artificial intelligence in the "read" to the text, just think of the corresponding language words, Rather than think of the reality, despite the complex statistical word clustering algorithm, but this is still not the same as human translation thinking. This technology plus the Google translation, although the original and translated words can be associated, there is a certain ability to organize into sentences, but still difficult to understand.

Hofstadter's argument has also been tested in the Sina technology test. Previously used to translate the test statement abroad, translated from "refund ticket" and "swap" & rdquo; translated into "swap" These two examples, we can see that Google translation is still stuck in Word-to-word or phrase-to-phrase levels, rather than associating them with the actual scene, are difficult to understand even if organized according to the correct grammar. To borrow a professor of translation studies: "You can understand every single word, but it is not what people say. & rdquo;

From this we can see that there is still a certain distance from the completely barrier-free application of Google translation. In this case, in addition to a few translation App mobile phone installed on the tourist, it is best to learn more commonly used English, and then bad also practiced a drawing skills, to prepare for a rainy day. Of course, AI researchers are still going to explore human thinking more deeply and develop a better translation app; if there is such a day, professional translators will also be astonished at the machine.

Finally, Sina Technology wish you a happy Spring Festival holiday, whether you are in a foreign or domestic.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments