* is based on statistical method to need large-scale bilingual corpora, how many what the accuracy of parameter of its interpreter model, linguistic model relies on corpora directly, its interpreter quality basically depends on the covering power of the stand or fall of probability model and corpus. At the same time interpreter model, linguistic model is in simplify a few blemish also are brought in the process, in simplify and feasible between put in problem of a balance. The method that is based on statistic does not need the dependence of a large number of knowledge, rely on statistical result to have the choice of processing of different meanings clear up and translation directly, kept away from a lot of difficult problem that the language understands. But the choice of corpora and processing not only project amount is large, and need handles problem likeness actually together. Because this is general,the machine translation system of the domain is given priority to with statistical method rarely.

* is based on the methodological EBMT of example: The machine translation that is based on example (Example-Based Machine Translation) the thought is the earliest it is by famous Japanese machine translation the expert grows end true (Nagao, m. ) put forward. Its are imagined basically is the analysis that does not pass depth, pass merely already some experience knowledge, undertake an interpreter through analogy principle. Interpreter process of the mankind is decompose input sentence correctly above all, decompose for phrase fragment, catch a these phrase fragment interpret to become other language phrase, incorporate these phrase finally grow sentence. Every phrase fragment adopts the principle of analogy to undertake an interpreter. The fundamental of law of this one party is Baconian rise very simple, main knowledge source of the system is bilingual contrasting interpreter example library, when inputting S of sentence of language of a source every time, the system is found out and the sentential S with most similar S ' , imitate S ' translation T ' the translation T that forms S is outputted next. This kind of method needs a very big corpus to regard as prop up, build a corpus to need to throw vast labor power and material resources for compose. The machine translation method that is based on example has a few more the following advantage:

It is easy that * system safeguards: The knowledge in the system exists with translating the form such as example and semantic dictionary, can very easy use the means augment system that adds example and vocabulary.

* produces the translation of high quality easily, used older interpreter example warehouse especially, when the input can match accurately with example, perhaps be such more.

* can avoid the is based on regular machine translation to must undertake deep-seated paralinguistics of a few traditions to analyse.

* is the same as a language relevant knowledge is very few. Appearance of the existence in wanting to remember a library only inputs similar sentence together, can undertake matching.
Previous 1 2 3 4 56 Next