Word alignment with bridge languages

• US 8,185,375 B1
• Filed: 07/23/2007
• Issued: 05/22/2012
• Est. Priority Date: 03/26/2007
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Kumar, Shankar, Och, Franz J., Macherey, Wolfgang
• Primary Examiner(s): Saint Cyr, Leonard
• Application Number:
  US11/781,824
• Time in Patent Office: 1,765 Days
• Field of Search: None
• US Class Current: 704/4
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/51 Translation evaluation

Abstract

Systems, methods, and computer program products are provided for statistical machine translation. In some implementations a method is provided. The method includes receiving multi-lingual parallel text associating a source language, a target language, and one or more bridge languages, determining an alignment between the source language and the target language using a first bridge language that is distinct from the source language and the target language, and using the determined alignment to generate a candidate translation of an input text in the source language to the target language.

Claims

1. A method comprising:

receiving multi-lingual parallel text associating a source language, a target language, and one or more bridge languages;

determining a direct alignment between the source language and the target language, wherein the direct alignment represents a word-for-word connection between source language and target language words;

determining, using one or more computing devices, a first bridge-based alignment between the source language and the target language using a first bridge language that is distinct from the source language and the target language, wherein the first bridge-based alignment is determined based in part on a first posterior probability matrix for a first alignment from the source language to the first bridge language and a second posterior probability matrix for a second alignment from the first bridge language to the target language;

using the direct alignment between the source language and the target language to generate a first candidate translation of an input text in the source language to the target language;

using the first bridge-based alignment to generate a second candidate translation of the input text in the source language to the target language; and

determining a consensus translation using the candidate translations.

2. The method of claim 1, where determining the first bridge-based alignment using the first bridge language includes:

determining a first alignment between the source language and the first bridge language;

determining a second alignment between the first bridge language and the target language; and

determining the first bridge-based alignment between the source language and the target language using the first and second alignments.

3. The method of claim 2, further comprising:

determining a combined alignment using two or more bridge-based alignments determined using two or more bridge languages; and

using the combined alignment to generate a candidate translation of the input text.

4. The method of claim 3, wherein determining a combined alignment includes using the combining posterior probability matrices for each bridge language according to:

5. The method of claim 2, where determining the first bridge-based alignment using the first bridge language further comprises:

determining a first posterior probability matrix associated with an alignment from the source language to the first bridge language;

determining a second posterior probability matrix associated with an alignment from the first bridge language to the target language;

determining a third posterior probability matrix associated with an alignment from the source language to the target language as a function of the first posterior probability matrix and the second posterior probability matrix; and

identifying a maximum alignment probability using the third posterior probability matrix.

6. The method of claim 1, further comprising:

determining a plurality of bridge-based alignments between the source language and the target language using a plurality of respective bridge languages.

7. The method of claim 6, further comprising:

using each of the determined plurality of bridge-based, alignments to generate respective candidate translations of the input text.

8. The method of claim 1, where determining a consensus translation using the candidate translations includes evaluating a score for each candidate translation.

9. The method of claim 1, where determining the alignment using the first bridge language includes calculating:

10. A computer program product, encoded on a non-transitory computer-readable medium, operable to cause data processing apparatus to perform operations comprising:

identifying multi-lingual parallel text associating a source language, a target language, and one or more bridge languages;

determining a direct alignment between the source language and the target language, wherein the direct alignment represents a word-for-word connection between source language and target language words;

determining a first bridge-based alignment between the source language and the target language using a first bridge language that is distinct from the source language and the target language, wherein the first bridge-based alignment is determined based in part on a first posterior probability matrix for a first alignment from the source language to the first bridge language and a second posterior probability matrix for a second alignment from the first bridge language to the target language;

using the direct alignment between the source language and the target language to generate a first candidate translation of an input text in the source language to the target language;

using the first bridge-based alignment to generate a second candidate translation of the input text in the source language to the target language; and

determining a consensus translation using the candidate translations.

11. The computer program product of claim 10, where determining the first bridge-based alignment using the first bridge language includes:

determining a first alignment between the source language and the first bridge language;

determining a second alignment between the first bridge language and the target language; and

determining the first bridge-based alignment between the source language and the target language using the first and second alignments.

12. The computer program product of claim 11, further comprising:

determining a combined alignment using two or more bridge-based alignments determined using two or more bridge languages; and

using the combined alignment to generate a candidate translation of the input text.

13. The computer program product of claim 11, where determining the first bridge-based alignment using the first bridge language further comprises:

determining a first posterior probability matrix associated with an alignment from the source language to the first bridge language;

determining a second posterior probability matrix associated with an alignment from the first bridge language to the target language;

determining a third posterior probability matrix associated with an alignment from the source language to the target language as a function of the first posterior probability matrix and the second posterior probability matrix; and

identifying a maximum alignment probability using the third posterior probability matrix.

14. The computer program product of claim 10, further comprising:

determining a plurality of bridge-based alignments between the source language and the target language using a plurality of respective bridge languages.

15. The computer program product of claim 14, further comprising:

using each of the determined plurality of bridge-based alignments to generate respective candidate translations of the input text.

16. The computer program product of claim 10,

where determining a consensus translation using the candidate translations includes evaluating a score for each candidate translation.

17. A system comprising:

one or more computing devices operable to perform operations including;

identifying multi-lingual parallel text associating a source language, a target language, and one or more bridge languages;

determining a direct alignment between the source language and the target language wherein the direct alignment represents a word-for-word connection between source language and target language words;

determining a first bridge-based alignment between the source language and the target language using a first bridge language that is distinct from the source language and the target language, wherein the first bridge-based alignment is determined based in part on a first posterior probability matrix for a first alignment from the source language to the first bridge language and a second posterior probability matrix for a second alignment from the first bridge language to the target language;

using the direct alignment between the source language and the target language to generate a first candidate translation of an input text in the source language to the target language;

using the first bridge-based alignment to generate a second candidate translation of the input text in the source language to the target language; and

determining a consensus translation using the candidate translations.

18. The system of claim 17, where determining the first bridge-based alignment using the first bridge language includes:

determining a first alignment between the source language and the first bridge language;

determining a second alignment between the first bridge language and the target language; and

determining the first bridge-based alignment between the source language and the target language using the first and second alignments.

19. The system of claim 18, further operable to perform operations comprising:

determining a combined alignment using two or more bridge-based alignments determined using two or more bridge languages; and

using the combined alignment to generate a candidate translation of the input text.

20. The system of claim 18, where determining the first bridge-based alignment using the first bridge language further comprises:

determining a first posterior probability matrix associated with an alignment from the source language to the first bridge language;

determining a second posterior probability matrix associated with an alignment from the first bridge language to the target language;

determining a third posterior probability matrix associated with an alignment from the source language to the target language as a function of the first posterior probability matrix and the second posterior probability matrix; and

identifying a maximum alignment probability using the third posterior probability matrix.

21. The system of claim 17, further operable to perform operations comprising:

determining a plurality of bridge-based alignments between the source language and the target language using a plurality of respective bridge languages.

22. The system of claim 21, further operable to perform operations comprising:

using each of the determined plurality of bridge-based alignments to generate respective candidate translations of the input text.

23. The system of claim 17, where determining a consensus translation using the candidate translations includes evaluating a score for each candidate translation.