Machine Translation Assistant Based on Internet Corpus"

If machine translation is regarded as a line, the line is similar to Kingsoft's dictionary method. This method only provides translation at the word level. For an unknown English word, you can find the corresponding Chinese Character in the dictionary. Similarly, you can find many possible English words by searching for a Chinese word. One of the major disadvantages of this translation method is the lack of context information in translation. It can only translate words and words in isolation, the meaning of words in different contexts is not considered to be different. For example, the word "eat" in "lunch" should be "eat", but the phrase "lunch" generally corresponds to "have lunch ".
The other end of the line is the machine translation technology represented by traditional machine learning. It includes three main schools: rbmt (Rule-based machine translation), ebmt (instance-based machine translation), and SMT (Statistical-based machine translation ). Google's Translation System is such a system. However, no matter which school or method is used, these methods aim to provide a complete translation solution, that is to say, the whole sentence or even the whole article is translated without any manual intervention. It should be said that this is the ultimate goal of machine translation, but unfortunately, their performance is far from practical (unless in some small specific fields ).

Most of the above two solutions are very familiar, but in fact, there has recently been a new attempt, somewhere between the two ends of the line. I am trying to give an example to illustrate this method.
Many people have used. It belongs to the dictionary-based method. However, its dictionary structure is somewhat special. This dictionary does not come from human orchestration, such as the niujin dictionary, but from the network. There are a large number of bilingual corpus on the network. For example, many web pages have this noun followed by a bracket, which contains the corresponding English words, such as "machine learning) is .... "youdao dictionary captures the bilingual corpus in these networks by some method (which should be a rule method) and automatically forms a dictionary. Such a dictionary has a major advantage: it will contain translations of many proprietary terms or the latest popular word phrases, such as "subprime mortgage crisis-subprime mortgage crisis ", this is beyond the reach of previous dictionaries.
After talking about this, I want to mention another feature of youdao dictionary, that is, its example sentence function. If you search for any word or phrase in the youdao dictionary, it returns a bilingual example containing the relevant word/phrase, which is also from the Internet, it can be considered that their translations are basically correct. For example:
Subprime mortgage crisisIt is moving towards a worse situation.

The subprime mortgage crisis is wrongly ed to get much worse.
This function is very useful to me, because the most troublesome thing for me to write in English is that the sentences I write are not authentic enough, the sentence structure is not beautiful enough. But now, I can solve this problem through the youdao dictionary. For example, I want to translate "Google's Translation System is such a system ", then I searched for "that is the case" and found that most of them were translated into "is such ", then I will know that this sentence can be translated into "Google's translate is such a sytem ".
Search for the sentence trunk or sentence to find the corresponding translation example, and then replace the other part of the sentence based on the example. Instead of traditional machine translation methods, manual intervention is required. The key point here is that the examples you have found are correct and true. You can trust the results you have found and correct them on this basis. However, if you use traditional machine translation systems such as Google translation, the results do not require manual intervention, but they are not reassuring (and indeed inaccurate ), because the current level of machine translation has not reached this level.

The reason for writing this article is that I saw this article on csdn:Translation and writing in Chinese and English based on the example sentence search
What this system provides is what I described, a role similar to a machine translation assistant. As mentioned in the article,
"
The application prospects of this system include Chinese and English Translation and writing for users, mainly reflected in the following three aspects:

1. the user only knows chinese? Such as "Looking forward to the future and complying with standards;

2. You can write a sentence in English, but are not sure whether the statement is correct? In this case, the instance-based machine-assisted translation can help users write to accuracy ";

3. the user wants to know whether there is a more elegant expression. At this time, the instance-based machine-assisted translation and writing can help the user to "write better". For example, the user wrote the phrase "the story
Started with a storm. (The story begins with a storm .)", The system prompts the user to consider changing it to "The Story opened
With a storm ."
"
I think this will be a brand new idea in the Machine Translation field.