Systems and methods for tuning parameters in statistical machine translation

1. A computer-implemented method for statistical machine translation (MT) weight tuning, the method comprising:

executing instructions stored in memory by a processor to:

sample a pair of candidate translation units from a set of candidate translation units corresponding to a source unit, each candidate translation unit corresponding to unique numeric values assigned to one or more features;

receive an initial weighting value for each feature;

compare the pair of candidate translation units to produce a binary result; and

use the binary result to adjust the initial weighting values to produce modified weighting values using the processor.

2. The method of claim 1 , further comprising:

calculating BLEU (Bilingual Evaluation Understudy) scores for each candidate translation unit; and

determining scoring functions for the candidate translation units.

3. The method of claim 1 , further comprising ranking the initial weighting values for two or more candidate translation units, wherein the ranking is based on a ranking of results of scoring functions for one or more pairs of candidate translation units.

4. The method of claim 3 , wherein the ranking is applied to translation parameters of the sampled candidate translation units.

5. The method of claim 3 , wherein the ranking comprises one or more of the following: a pair-wise ranking, a linear binary classification, a logistic regression classification, a maximum entropy classification, and an iterative classification.

6. The method of claim 1 , wherein the sampling comprises:

generating a plurality of candidate translation unit pairs;

assigning a probability coefficient to each candidate translation unit pair; and

selecting the candidate translation unit pairs having the highest score differential.

7. The method of claim 1 , wherein the source unit and candidate translation units comprise one or more of the following: a word, a phrase, and a sentence.

8. The method of claim 6 , wherein the generating of the set of candidate translation units comprises selecting one or more candidate translation units from a translation database.

9. The method of claim 1 , wherein the translation parameters, associated with the candidate translation units, comprise one or more of the following: a candidate translation space policy, a scoring function, a BLEU score, a weight vector, and a loss function.

10. The method of claim 9 , wherein the translation parameters are ranked such that candidate translation units having higher relevancy for a correct translation of the source units are associated with one or more of the following: a minimized loss function, a highest weight vector, and a best candidate translation space policy.

11. A system for statistical machine translation (MT) parameter tuning, the system comprising:

a memory for storing executable instructions for tuning statistical MT parameters; and

a processor for executing the executable instructions stored in memory, the executable instructions comprising:

a text analyzing module to sample a pair of candidate translation units from a set of candidate translation units corresponding to a source unit, each candidate translation unit corresponding to numeric values assigned to one or more features;

a communication module to receive an initial weighting value for each feature;

a mapping module comparing the pair of candidate translation units to produce a binary result; and

a translating module to adjust the initial weighting values to produce modified weighting values based on the binary result.

12. The system of claim 11 , wherein the translating module is further configured to:

calculate BLEU (Bilingual Evaluation Understudy) scores for each candidate translation unit; and

determine scoring functions for the candidate translation units.

13. The system of claim 11 , further comprising ranking the initial weighting values for two or more candidate translation units, wherein the ranking is based on a ranking of results of scoring functions for one or more pairs of candidate translation units.

14. The system of claim 13 , wherein the translating module is further configured to sample two or more candidate translation units, wherein the ranking is applied to translation parameters of the sampled candidate translation units.

15. The system of claim 13 , wherein the ranking comprises one or more of the following: a pair-wise ranking, a linear binary classification, a logistic regression classification, a maximum entropy classification, and an iterative classification.

16. The system of claim 11 , wherein the source unit and candidate translation units comprise one or more of the following: a word, a phrase, and a sentence.

17. The system of claim 11 , wherein generating of the set of candidate translation units comprises selecting one or more candidate translation units from a translation database.

18. The system of claim 11 , wherein the translation parameters associated with the candidate translation units comprise one or more of the following: a candidate translation space policy, a scoring function, a BLEU score, a weight vector, and a loss function.

19. The system of claim 18 , wherein the translation parameters are ranked so that candidate translation units having higher relevancy for a correct translation of the source units are associated with one or more of the following: a minimized loss function, a highest weight vector, and a best candidate translation space policy.

20. The system of claim 11 , wherein the processor executes the executable instructions under direction of a cloud control system, the cloud control system managing a cloud.

21. A computer-readable medium having instructions stored thereon, which when executed by one or more computers, causes the one or more computers to:

sample a pair of candidate translation units from a set of candidate translation units corresponding to a source unit, each candidate translation unit corresponding to unique numeric values assigned to one or more features;

receive an initial weighting value for each feature;

compare the pair of candidate translation units to produce a binary result; and

use the binary result to adjust the initial weighting values to produce modified weighting values.