Welcome toShandong translation network
About us | Add to Favorites | Contact us | 中文版
Translate IT and development
From;    Author:Stand originally

The machine translation that is based on regulation develops today, opposite for more mature already. Although pass long-term effort, people has been built contain by tens of thousands regular library, enclothed quite big language phenomenon, but from theoretic tell, this kind of process is finite. Because the language is of the chiliad on course of a nation,accumulate, it is established by usage and of dynamic development. As ceaseless development of the society, new vocabulary and language phenomenon appear ceaselessly. The regulation of existing machine translation system is again much, also be specific language phenomenon only generalize and sum up. Accordingly, the machine translation method that is based on regulation is in the advantage that draws lessons from other method, produce a lot of change. These change basically behave in:

* is in get a respect regularly: Traditional regular method basically relies on linguist to sum up regulation to undertake debugging, and take seriously more now get regulation from inside the corpus (if use the study algorithm of wrong drive) .

The regular method of * tradition: The knowledge of regulation of big range linguistics that often slants to describe thick granuality, overall situation to change again, and the lingustic knowledge that values descriptive fine granuality, local little scope more now, appear piece " small regular library, big dictionary " trend.

* shows a side in knowledge: To have a description to translating knowledge with more smaller size, accurate ground, have nothing to do to wanting pair of pure context commonly regulation is done a few improve.

The regular method of * tradition: Use often dispute this namely the determinism principle of those. Rash club sex of the system is poorer, and multiform probability or grading function introduce commonly in regular now system, the gender raises rash baton of the system somewhat.

(2) the machine translation method that is based on a corpus: The machine translation method that is based on statistic and the machine translation method that are based on example are the origin that use corpus regards an interpreter as knowledge. This both distinction depends on: In the machine translation method that is based on statistic, intellectual expression is statistical data, is not corpus itself. Those who translate knowledge getting is finish before the interpreter, the corpus is used no longer in the interpreter's process; And in the machine translation method that is based on example, the form of a kind of expression that bilingual corpus itself translates knowledge namely (not be exclusive certainly) , those who translate knowledge get did not finish entirely before the interpreter, inquire even in the interpreter's process and use a corpus.

The mathematical model that statistic translates is offerred by the person such as the Brown of IBM company. The basic idea of statistical machine translation is, the process that machine translation regards an information as to transmit, undertake explanatory to machine translation with model of a kind of channel. Suppose S of text of language of a paragraph of source, the course turns target language into T after some noise channel, that is to say, t of text of hypothesis target language passes some kind of strange encode to get by S of text of language of a paragraph of source, so the interpreter's target wants namely T reductive into S, this is the process of a decipher namely. Problem of statistical machine translation is OK and wraparound decompose for 3 problems: The parameter estimation of the linguistic model, parameter estimation that translates a model and search problem (search best translation) .
Previous 1 2 3 45 6 Next