Machine translation method and machine translation system

1. A machine translation method, including:

Utilizing a plurality of machine translation apparatus, the text of the source language are translated into the target language, in order to get a plurality of candidate translation;

Utilize the language model, a plurality of candidate-oriented language model score is respectively calculated MR;

Respectively given plurality of machine translation apparatus on a plurality of candidate of the apparatus of the MR score;

Candidate based on the original length and the length of the MR, a plurality of candidate to the length of the respectively calculated MR score;

Based on the language model score, score equipment, the length of at least one of the scores, a plurality of candidate respectively calculating the total score of the MR; and

Selection of the highest total score candidate translation as a result of machine translation.

2. Machine translation method according to Claim 1, wherein the respectively computing language model score comprises:

Utilize the language model, candidate-based smoothness degree of MR, or semantic grammar structure for at least one of the structure, each of a number of candidate calculation of the MR score language model.

3. Machine translation method according to Claim 1, wherein the respectively for obtaining equipment score comprises:

Given according to the machine translation apparatus of the characteristics and weight, computing the output of the MR score candidate of the apparatus.

4. Machine translation method according to Claim 1, wherein the length of the respectively calculating the score comprises:

Candidate according to the original length and the length of the MR ratio of the comparison with a predetermined value, each of a number of candidate to calculate the length of the MR score.

5. Machine translation method according to Claim 1,

Wherein the machine translation apparatus including a plurality of: training corpus based on the expansion of 1st translation device;

Wherein the extended corpus obtained by the following steps:

The source language and intermediate language sentences in the corpus of 1st-to, in the double-sentence pair of intermediate language into a target language, in order to obtain the source language and the bilingual sentence of the target language, new bilingual sentence as the 1st;

The intermediate language and of the target language sentence in the corpus of 2nd-the, in the double-sentence pair of intermediate language translated into the source language, in order to obtain the source language and the bilingual sentence of the target language, new bilingual sentence as the 2nd; and

The new double-sentence pair based on the 1st and 2nd the new bilingual sentence, obtain an extended corpus.

6. Machine translation method according to Claim 5, wherein the new double-sentence based on the 1st and 2nd to the new bilingual sentence, obtain an extended corpus comprises:

Removing the following conditions is unsatisfied with the 1st and 2nd new double-sentence pair of the new double-sentence pair: of the new double-sentence pair in the length of the sentence of the source language to the target language sentence is greater than the sum of the length of 1st and 2nd threshold value is less than the threshold value; and

The rest of the 1st and 2nd new double-sentence pair of the new double-sentence pair and with the existing source language to the target language sentence of a joint and to remove and repeat the, in order to obtain an extended corpus.

7. Machine translation method according to Claim 1,

Wherein the machine translation apparatus including a plurality of: 2nd translation device, the 2nd translation device comprises a cascade of and intermediate language in language translation between the 1st translation sub-apparatus and can be in the intermediate language and target language translation between the 2nd translation sub-device;

Wherein the translation sub-device using 1st, the text of the source language to the intermediate language translation of a plurality of intermediate result; using 2nd translation sub-apparatus, the plurality of intermediate results translated into each of a plurality of target language candidate MR; and from a plurality of candidate MR selection of optimum as a candidate translation;

Wherein the selecting step includes:

For a plurality of candidate MR each, given translation sub-apparatus according to the 1st of the characteristics and weight, calculate the score translation sub-device 1st, and 2nd is given according to the characteristic of the translation sub-device and a weight, the calculation of score 2nd translation sub-device; and

The score and 1st 2nd translation sub-apparatus sub-device and the translation of the largest score candidate MR, as candidate translation.

8. Machine translation method according to Claim 1,

Wherein the machine translation apparatus including a plurality of: 3rd of the expansion rules-based translation device;

Wherein the expansion rules obtained by the following steps:

And intermediate language based on the language corpus 1st, decimation on language rules 1st and the intermediate language;

Based on the intermediate language and of the target language corpus 2nd, decimation on intermediate language and 2nd rules of the target language;

Selection rules and 1st 1st 2nd the rules of the rules with the target end of the source is the same as the rules of 2nd; and

Rules based on the selected source end of the 1st and 2nd target of the rules, generating the extended rules.

9. Machine translation method according to Claim 8, wherein the generating the expansion rules:

The selected source end of the 1st and 2nd the rules of the rules of the rules of as an expansion of the target end of the source and the target; and

Based on the selected rules of the rules and 1st 2nd forward translation probability, reverse translation probability, forward lexicalized probability, reverse lexicalized probability, of the expansion rules respectively calculating the forward translation probability, reverse translation probability, forward lexicalized probability, reverse lexicalized probability;

Wherein with a source end for having a plurality of expansion rules, only retain its forward translation probability, reverse translation probability, forward lexicalized probability, reverse lexicalized probability K and the maximum of the expansion rules before, K is a predetermined natural number.

10. A machine translation system, comprising:

Plurality of machine translation apparatus, is used for the source language of the original translated into the target language, in order to get a plurality of candidate translation;

Language model, a plurality of candidate for respectively calculating the MR score language model;

Equipment shall be sub-acquisition device, is configured to respectively given plurality of machine translation apparatus on a plurality of candidate of the apparatus of the MR score;

Length score calculating device, is configured to the length of the original and candidate based on the length of the MR, a plurality of candidate to the length of the respectively calculated MR score;

total score calculating device, is configured based on the language model score, score equipment, the length of at least one of the scores, a plurality of candidate respectively calculating the total score of the MR; and

Said selection device, is configured to select the highest total score candidate translation as a result of machine translation.