For what and for whom machine translation is suitable?
Ilon Mask does not spend the time of his children on learning foreign languages. He believes that they will surely live to a fantastically reliable and useful machine translation. Indeed, in front of our eyes, science fiction is becoming a reality: smart homes meet us from work with a hot dinner, voice assistants joke with us in chat rooms, and anthropomorphic robots support dialogue in several languages. So, when will a translator equivalent to a person appear in every smartphone?
Never! Or very soon - here is the first disappointing news. The fact is that people who speak languages are able to convey the meaning of what is written in their own words, without being tied to the structure of the source code. Machines, however, are translated word by word or by phrase and teaching them to operate not with words, but with images is the same thing as inventing artificial intelligence. What does it mean “to operate with images?” It means - to understand the translated text, to interpret it. That is, neither more nor less - to have consciousness.
The good news is that in more than 70 years of the existence of machine translation, we have already come a long way from statistical methods to artificial neural networks.
Networks can read sentences from left to right and from right to left, literally transliterate their own names and, instead of memorizing many translation options, operate on the semantics of the whole text, breaking it up into segments, and then analyze and synthesize them. The result is worthy, and, in some cases, the system translates even phraseological units.
')
Language is a very flexible system with an unlimited set of vague rules. And although neural networks already seize semantic and syntactic connections in sentences and even recognize the speaker’s accent, they do not know how (and, most likely, do not learn) to take into account cultural, cognitive, literary and other aspects of translation. In other words, the context can fundamentally disrupt communication, because:
The computer can not understand the culture:
Computer poorly translates idioms:
The computer is not able to convey the emotions of songs and poems:
Emotionally colored texts, phraseological turns, cultural overtones are also not about the car:
Contracts, letters of guarantee, marketing materials, medical documents, the error in which can cost someone's life - this is not to the car:
Advertising slogans, any literary texts - this is too hard, polysemous and not formalized for machine translation. Yes, and the term "translation" in relation to the literary text is probably not quite correct. Here, the task of the translator is not to decode the text, but to find the equivalent vocabulary in the target language. The translator works with meanings, but not with words and in the work relies on literary flair:
Trusting the car, we cut off from the texts all the amazing linguistic game, on which the humor of famous shows, films and TV shows is built, from which a sentimental aftertaste of your favorite songs springs:
For not knowing the language of people who need in a very general way to understand the content of a text. For translators who need a “template” for editing. Well, of course, for a business that needs to accelerate the process of intercultural communication.
Another thing is that machine translation will still have to be edited by a person, and for this you need to be able to notice and correct errors made by the machine. This is a separate time-consuming process that requires a specific skill. This skill is basic for a philologist, but it is almost as labor-intensive for an ordinary student to teach him as ... English in the degree necessary for understanding most texts.
Machine translation can serve as a good fighter in formal business correspondence, but will betray you in live communication. Relying on machine translation, we generally deprive ourselves of the elementary joy of communication, because no one wants to talk with a smartphone - at least, until it sticks out in the place of your own head. But even science fiction did not predict this to us.
Making a bet on machine translation, we actually put on the imminent emergence of consciousness in computers, similar to human. That is, self-consciousness, which would allow the machine to understand what exactly it “reads” and translate it humanly. Are all the processes of the human brain can be reduced to algorithms? It is unlikely that this issue will be resolved soon. But the study of English with the use of all the achievements of scientific progress is a quick and effective thing.
For those who are not ready to exchange warm live communication with wonderful people all over the world for heartless machine translation, we have something to cook. Under the link - sign up for a free English lesson at Skyeng School. Enter the registration code HABR2 when registering: a gift will add 2 lessons when you first pay.
Never! Or very soon - here is the first disappointing news. The fact is that people who speak languages are able to convey the meaning of what is written in their own words, without being tied to the structure of the source code. Machines, however, are translated word by word or by phrase and teaching them to operate not with words, but with images is the same thing as inventing artificial intelligence. What does it mean “to operate with images?” It means - to understand the translated text, to interpret it. That is, neither more nor less - to have consciousness.
The good news is that in more than 70 years of the existence of machine translation, we have already come a long way from statistical methods to artificial neural networks.
Networks can read sentences from left to right and from right to left, literally transliterate their own names and, instead of memorizing many translation options, operate on the semantics of the whole text, breaking it up into segments, and then analyze and synthesize them. The result is worthy, and, in some cases, the system translates even phraseological units.
')
Neural networks fail
Language is a very flexible system with an unlimited set of vague rules. And although neural networks already seize semantic and syntactic connections in sentences and even recognize the speaker’s accent, they do not know how (and, most likely, do not learn) to take into account cultural, cognitive, literary and other aspects of translation. In other words, the context can fundamentally disrupt communication, because:
The computer can not understand the culture:
Computer poorly translates idioms:
The computer is not able to convey the emotions of songs and poems:
Emotionally colored texts, phraseological turns, cultural overtones are also not about the car:
Contracts, letters of guarantee, marketing materials, medical documents, the error in which can cost someone's life - this is not to the car:
Advertising slogans, any literary texts - this is too hard, polysemous and not formalized for machine translation. Yes, and the term "translation" in relation to the literary text is probably not quite correct. Here, the task of the translator is not to decode the text, but to find the equivalent vocabulary in the target language. The translator works with meanings, but not with words and in the work relies on literary flair:
After all, the earth makes a revolution in twenty-four hours ...
- Turnover? - repeated the Duchess thoughtfully.
And, turning to the cook, she added:
“Take her into circulation!” For a start, chop off her head!
Trusting the car, we cut off from the texts all the amazing linguistic game, on which the humor of famous shows, films and TV shows is built, from which a sentimental aftertaste of your favorite songs springs:
For what and for whom machine translation is suitable?
For not knowing the language of people who need in a very general way to understand the content of a text. For translators who need a “template” for editing. Well, of course, for a business that needs to accelerate the process of intercultural communication.
Another thing is that machine translation will still have to be edited by a person, and for this you need to be able to notice and correct errors made by the machine. This is a separate time-consuming process that requires a specific skill. This skill is basic for a philologist, but it is almost as labor-intensive for an ordinary student to teach him as ... English in the degree necessary for understanding most texts.
What are the conclusions?
Machine translation can serve as a good fighter in formal business correspondence, but will betray you in live communication. Relying on machine translation, we generally deprive ourselves of the elementary joy of communication, because no one wants to talk with a smartphone - at least, until it sticks out in the place of your own head. But even science fiction did not predict this to us.
Making a bet on machine translation, we actually put on the imminent emergence of consciousness in computers, similar to human. That is, self-consciousness, which would allow the machine to understand what exactly it “reads” and translate it humanly. Are all the processes of the human brain can be reduced to algorithms? It is unlikely that this issue will be resolved soon. But the study of English with the use of all the achievements of scientific progress is a quick and effective thing.
For those who are not ready to exchange warm live communication with wonderful people all over the world for heartless machine translation, we have something to cook. Under the link - sign up for a free English lesson at Skyeng School. Enter the registration code HABR2 when registering: a gift will add 2 lessons when you first pay.
Source: https://habr.com/ru/post/430360/
All Articles