Can artificial intelligence be used to replace artificial translation? _Artificial intelligence, machine learning, technology

When a new technology, no, it should be said that when a technology has a little bit of joy, now the technology circle immediately put a variety of "subversion" labels on it.

Can artificial intelligence be used to replace artificial translation?

If you are a reader of tech news, you must have felt a little numb or even disgusting about the word "subversion"? The change in the ground that day seems to be coming tomorrow.

The science and technology media is good at attracting people's attention. This is certainly a must-have for practitioners. But now everyone is looking around and looking like an ant on a hot pot. Is this really normal? In the eyes of technology media. Technological innovation has turned the world world upside down. Nothing is three days, one innovation, five days and one subversion. Where is the focus of our future standing in this world?

Among them, especially the voice surrounding "artificial intelligence" is the most surging. Not long ago, Google launched a new Google Translate based on neural network, which caused an exclamation in the industry. Many people even suggested that there is no need for the country to open a foreign language major, and the translation has completely disappeared under the guidance of machine learning!

However, in fact, if we stand a little farther to see it, if we are still confused, then put an ice on the forehead, you will find that any statement that "the machine will replace the human translation" is nonsense. . Today, 50 degrees of silicon will be a good way to talk to you about why the machine can't replace the artificial translation, and we can talk a little bit about how people and machines should maintain such a relationship.

What is Google's neural network translation?

In fact, online translation services were already available on the Internet a decade ago, but in the long run, the argument that "the machine is about to replace humans for translation" is endless. And now everyone's eyes are back to this again, because Google has upgraded its online translation service and launched a machine translation system called "Neural Network".

Let's talk about what is a "neural network"?

Since the beginning of the computer, efforts to make machine intelligence gradually transition to the human brain has become a dream of many researchers. But until now, progress in this area can almost be described as moving. It is clear that the focus of machines and humans is different. In the eyes of the machine, the operation of the effortlessly, the logical judgment, which is a human being, requires a lot of brain power to complete, and for humans, things that can be easily done, such as identifying objects, people's identity, freedom. Smooth language communication, rapid learning and adapt to a certain dialect, these tasks are simply impossible for the machine.

After several unsuccessful attempts to launch the "human brain", the researchers proposed that we don't build a neural network similar to the human brain. Let the relationship between computer computing units be built and designed according to the human brain. although,

Humans still don't understand how the human brain works, but it doesn't matter. We can first copy a human brain's computational model: one thing is not the input of the command output as a computer program, but the multiple neurons in the human brain. Each neuron processes its own information and outputs the result. In the same way, computer scientists add a lot of "nodes" between input and output. Each node processes the data from the previous node according to a weight coefficient it owns. Sometimes the node also Will be layered. This is the "Arithmetic Neural Networks" (ANN).

So, what is Google doing with this neural network-based machine translation system?

The translation services we used before were actually based on "phrases". That is to say, after a piece of text appears, the system will directly split the text. The smallest unit is the phrase, and then each word will be found in the corresponding translated version, then the order will be adjusted, and finally a translation will be produced. . This is a translation work with a clear sequence, namely: breaking up, finding the corresponding phrase, sorting, including re-segmentation of punctuation, finishing and other links.

It is reasonable to say that it is not very easy to make mistakes, but the accuracy of each link is quite high, but there are too many links. Once there is a deviation in one place, the back is all like the fallen blocks. The final translation is of course not tolerated.

The neural network-based translation system skips so many cumbersome steps directly, it directly relies on “massive data” to win, that is, it directly establishes two corresponding parallel language libraries, and then lets the machine itself Learn how to respond. This method actually appeared very early, but it also has certain technical problems that have not been broken. For example, the speed of machine learning is very slow, and if you encounter something that you haven’t seen before, it doesn’t know what to do. However, this time, through technological innovation, Google has greatly improved its accuracy and speed in translation.

Therefore, translation is no longer a simple word and word, phrase and phrase correspondence, but can grasp the full text from the perspective of an article, so that the translation is more smooth and natural.

So, does this mean that human translation is laid off here?

Now that we know the significant progress made by Google Translate, it is time to put the machines and humans on the scale of translation and weigh them together.