Ok everyone, In this chapter, we will mainly discuss the ways of the machine translation, which is very difficult, because it's related to a lot of knowledge of math and linguistic, especially machine translation nowadays. If you don't know well about math tools. Even the best programmer can't do the machine translation well. In this lesson, you are not expected to research on how the programing of machind translation is written. We'll discuss some key points of the ways of the machine translation to help you to form a direct recognition which enables you to know the assistive technology machine translation better. How does the machine translation work? We can see it like the process of human learning English. Say,we learn a language, firstly we recite the words then the patterns and sentences. Remember your first English lesson? This is a desk This is a chare Put a chicken a dog on a desk and it changes into there's a dog. When you know this suite of method, the most important is to practice, which's a process of using accepting and putting out language. The listening, speaking, reading,and writing, among which listening and reading is the passive process of receiving language while speaking and writing is the process of producing. In the traditional language education, too much emphasis is put into the training. But in the modern language teaching theory, the output of language is more valued. For a computer, how can it work? Just the same. Firstly,let the computer learn the vocabulary. And then it can find a translation in the dictionary. Then the pattern, the so-called usage of the rules. Next is some processing which is related to the language features. Just like we make mistakes and then learn this language constantly. We polish the mistakes. Computer is largely like this. The basic rule is formed and in practice, problems are found then supplement the rules improve the rules, at last let the machine translate it. And then the machine can work better. Next question is human tends to learn more. The longer you learn a language, with a correct way, then your language ability is going to be better and better. Is machine the same? A bigger data base,a more capable machine? Let's discuss the process of a translator's translation. Just now, we mentioned the process of lauguage translation. The translator's process of lauguage translation. is almost the same as that of the machine's. A translator read the OT and know the meaning, if not,he may ask a native speaker about it, or the expert. Finally,the TT comes out. And the TT should be accepted by the readers. How about the machine translation? Firstly is the analysis of the sentence then is about the language, also the transformation about the words ranks. then, Like what has been mentioned above, learn the language constantly, and process the rule of the language. Finally,TL comes out. The work process of machine's learning a language or it's translating a language is almost the same. But there are many doubts. Though they are almost the same, but we can go on to see the small stuffs. Let's think, what is needed in machine translation? language experts we are not going to talk about it. Traditionally,machine translation can't go on only by programmer without a language expert. But now it tends more to be handled by the mathematicians and the programmers. The language expert isn't now playing a very clear role in the research of machine translation. In the future, we can see whether the experts are more and more important or less and less important. What supports machine translation is mainly the technical resources. The first is the language database. Whether it's the Statistical machine translation or the Rule-based machine translation. language database is a must. A basis for language resources Corpus balance Monolingual corpora balance and monolingual corpora field are mainly used to form the language. That is to learn how to produce this language. Monolingual corpora balance can provide the rule of translation and change and some basic resources used by translation training. Comparable corpus is less important than those two. But it's very useful for some research on machine translation. Besides language database, the dictionary,which is the most frequently used by the translators is useful to the machine translation. We mentioned that machine translation uses some simple dictionaries. That's the dual List. OL is here TL is there. But there are some complicated structures. such as Thesaurus, which is a restrain and make sure about the meaning of the language. If you are not very clear about this, in the Terminology Management, we discussed about it and you can review. Besides Thesaurus, in the Linguistic computing research, there's a famous WordNet, raised up by Psychological language lab of Princeton University. It is a framework for a semantic description. So it's words are connected by ThinkSet, and portrayed by the meaning links between words. It's an important reference resource for Semantic computing. Others like FrameNet HowNet we'll not say in details. If you are interested, you can find some relevant materials. Besides these dead resources, we also need the research of natural language provide some tools for us to do the machine translation well. Basically, the tools founded during its research is useful to the machine translation. Such as the basic, tokenizer means let words become a unit. Then the Segmentation like Segmentation in Chinese. Actually,English also has Segmentation. then the cut of sentences, which we made some discussion in translation memory database. Then it's about the Morphological analysis Vocabulary speech tagging then Parsing dependency analysis Chunk of extracted sentences then the relationship between the various control in Bilingual library, all these can influence greatly the machine translation. Also extraction and discovery of the so-called named entities is also included. Usually Named Entity have set rules and we use them to translate the result is good. When the Named Entity is drawn out the whole sentence will be simpler and the process will be easier. pattern translation rules this resource is not born to exist. They are researched specially for the machine translation. Pattern and translation rules are usually made by the Linguists. They think that English is SVO Japanese is SOV. He set the rule for transformation. When English is transformed into Japanese, Verb should be transformed with objective. There are many this kind of rules. We see it ahead many this kind of challenges. Actually each one is related to a big set rule. Traditionally, the linguists and the computer engineers will work together to do this rule database. But now, Statistical Methods are invented the rule is more complicated that only machine can read it instead of the people. Then it's totally related to possibility, Do more language resources mean more accurate translation? Basically it's. Of course there is a problem about the quality. it means if you use more the same kind of language resources which is of the same quality the translation quality will be better. If it's not, some are good while some are bad only to increase the number not raise the quality. The realization of some large scale machine translation say use tens or hundreds of computers to do this experiment, the outcome is negative. That's to say, to get a good translation you need to pay attention to the quality of the language when you are enlarging the resources.
