HomeNewsIndustry News > Whether artificial intelligence translation can replace manual translation, the answer is not.

Whether artificial intelligence translation can replace manual translation, the answer is not.

Date:2018-09-26

Whether artificial intelligence translation can replace manual translation, the answer is not

 
In the era of artificial intelligence, the collaborative work of people and machines is especially important. The existing AI framework is still based on big data and deep learning. No matter how much data there is, there are always some corner cases that will be missed. Without a good human-machine coupling approach, AI will fail. Sometimes these failures are fatal. Many accidents caused by manual driving are because some scenes are not encountered during training, and the machine does not know how to judge. Many people now deify the artificial intelligence. All the foundations of artificial intelligence, until today, are based on whether there is enough data and there is not enough resources to label this data. As some people say, the artificial component of artificial intelligence is really big.
artificial intelligence translation can replace manual translation
 
Under the existing framework, one thing is especially important: how people get involved when artificial intelligence can't be handled.
 
The ideal state of human-machine coupling is that if a machine translation makes a mistake, the same translator will correct and correct the error, and the machine will not make this mistake again. The man solved the crisis and at the same time gave the machine a training data. In this process, AI will get better and better.
 
Human-machine coupling in the same interpretation, currently only wishful thinking.
 
After the general speaker starts speaking for two to three seconds, the interpreter will open the translation. The interval between them is called "Essence Time Difference" (EVS, Ear-Voice Span). The translators cut the sentences into groups of meaning and information units without interruption, according to the original order of the words, and then connected these units to try to translate the original intention of the speaker. This is called "sentence drive." According to Zhang Wei, who has been working on interpreters for eight years, interpreters generally hear one or two Italian groups and they will open their translations. While adjusting the logic to speak out, the ears should listen to the third and fourth group of meanings below.
 
 
Regardless of the stage of translation, once the speaker is heard on another topic, the translator needs to give a little more energy to the beginning of the speaker without affecting the current translation. I understood the general topic, and then recovered more energy from the past and continued to translate.
 
Please note that this process only takes place in a few seconds.
 
In order to cultivate the "distraction" ability, the interpreter will do "Shadowing". The practitioner listens to the English recording, half-sentence and then reads the English while handwriting numbers. At the beginning is 12345... I am writing, and after becoming proficient, I become 999, 998, 997... Write backwards, or just write double/singular. While listening, while talking, while writing, one heart and three.
 
For a skilled translator, the translation is almost instinctive: you can flip it when you sit down and forget it when you are finished. After a meeting, things turned over are almost forgotten. Skilled interpreters can even listen while listening and checking words.
 
Throughout the process, the current machine does not play any supporting role.
 
Even with the machine, the link of the interpreter can not be omitted. Because only after listening, the translator can find the error of machine voice transfer. As soon as the interpreter hears the information, the translator begins to process the information and translate it “instinctively”. Translators, like the work of the machine, are completely repetitive.
 
Even if the machine voice transfer can guarantee 100% correct rate, it is interference for the translator to read the text. When listening, people can grasp the focus of a paragraph more concisely, but when looking at the text, people will try to translate every detail, but not keep up with the speed of the speaker. An interpreter said that there is a lot of effort to read subtitles. "Close your eyes, listen clearly, open your mouth and say that you are all right!"
 
More importantly, the same interpreters who participated in the conference where AI translators were present and did not send headphones said that in the same box, they could not see the subtitles given by the machine. How can this be called human-machine coupling?
 
Artificial intelligence translation to replace humans?
 
Although the outsiders were immobile, they exclaimed that "the translation of this business has to disappear" and "the artificial intelligence translation has replaced humanity." However, in the translation circle, everyone’s mentality is quite optimistic. It is agreed that the translator’s job is very stable and will never Was taken away by the machine.
 
In the eyes of the industry, even if you understand every word, there are still three major obstacles to artificial intelligence translation.
 
One is the word polysemy. The term Mission basically means “an important task for a person or group of people, usually requiring travel to be done abroad” (Oxford English Dictionary). On the basis of this meaning, the specific meanings such as "mission", "task", "special mission", "business trip", "consultation", "delegation", "mission", "missionary mission" and "visiting group" are derived. The translator must judge whether it is translated into “mission” or “consultation” according to the context, whether it is “missionary group” or “mission”. This judgment is not clear at a glance. If you give machine translation, the machine usually chooses the most common meaning based on the probability of statistics, and this often goes wrong.
 
The second is a complex sentence structure. You don't even need complicated sentences. It's just a simple functional indicator posted on the hotel's electric kettle. "Please drink the tap water and drink it after drinking." The machine translation is not satisfactory.
 
The third is the changing scene.
Some scholars have found that English speakers have a speed of 120 words per minute, and simultaneous interpretation is the most accurate. Please note that the most accurate here does not mean that the accuracy is 100%, but 80% to 95%. Even at the most appropriate speed, human translators lose information. If the speaker's speech rate reaches 180 words per minute, the interpreter will miss the words in a large area; if the speed reaches 200 words per minute, the human translator will not be able to keep up.
 
The machine is different. Even if the speaker speaks fast, the machine can grab every word and not miss any information points.
 
This should be the advantage of the machine, but at present, it has become a disadvantage. The translator should not only translate the information on the spot, but also make the length of the translated Chinese and the length of the original English text as close as possible. An English abbreviation proper noun, Chinese translation may have six or seven words long. At this time, the interpreter will temporarily adjust and compress to simplify the information behind. In general, after two to three seconds of the speaker's speech, the translator will also end the speech. If it takes five seconds to end, "It's over. These five seconds are as long as eternal," said translator Zhang Wei.
 
The machine has no ability to adapt, and will only listen to each word with due diligence and then translate. This leads to an excessive amount of information translated by the machine, which makes the audience feel hard. Zhang Wei said: "Translated word by word can be translated, can be understood and digested, is another matter. In order to make the audience sound comfortable, the translator is always making choices."
 
The randomness of the interpreters can even be said to be "casual". If the speaker greets the participants with full energy, the interpreter will adjust his tone and call out good morning. The translator can imitate the speaker's tone, emphasizing when he emphasizes, humor when he speaks.
 
This is also why the simultaneous transmission needs to sit in a "box" (simultaneous box) behind the venue to see the venue: they need to feel the atmosphere of the venue.
 
This is all that the cold machine can't do.
 
The current AI translation can only be an auxiliary tool, at least until the new framework of artificial intelligence arrives. "If there is a breakthrough in artificial intelligence translation, it must be a breakthrough in the framework: not based on the existing framework of big data plus deep learning, but from the cognitive level of people, to find a new framework of AI that does not rely on data.

TypeInfo: Industry News

Keywords for the information:AI translation  artificial intelligence translation  AI