System and method for capitalizing machine translated text

• US 8,886,518 B1
• Filed: 08/07/2006
• Issued: 11/11/2014
• Est. Priority Date: 08/07/2006
• Status: Active Grant

• Current Assignee: SDL PLC (RWS Holdings Plc)
• Original Assignee: Language Weaver, Inc. (RWS Holdings Plc)
• Inventors: Wang, Wei, Marcu, Daniel, Knight, Kevin
• Primary Examiner(s): Shah, Paras D
• Application Number:
  US11/501,189
• Time in Patent Office: 3,018 Days
• Field of Search: 704/1, 704/2, 704/4, 704/5, 704/6, 704/9, 715/256, 715/264, 715/265
• US Class Current: 704/9
• CPC Class Codes: G06F 40/232 Orthographic correction, e.g. spell checking or vowelisation
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models

Abstract

A system and method for capitalizing translated text is provided. A capitalized source text is automatically translated to a target text. The target text is capitalized according to information in the capitalized source text.

Claims

1. A method for capitalizing translated text comprising:

executing a translator module stored on a device to automatically translate a capitalized source text to a target text, wherein prior to translation the capitalized source text is converted to lower case and then translated; and

capitalizing the target text according to capitalization information in the capitalized source text and the target text, wherein the step of capitalizing the target text according to capitalization information in the capitalized source text includes;

generating one or more capitalization configurations for the target text;

computing a configuration probability for each of the one or more capitalization configurations, the configuration probability computed from capitalization information in the capitalized source text and at least one capitalization model feature function based on an alignment between the capitalized source text and the target text or the capitalized source text and the capitalization configuration; and

selecting the best capitalization configuration based on the highest configuration probability.

2. The method of claim 1 further comprising capitalizing the target text according to the translated target text.

3. The method of claim 1 further comprising capitalizing the target text using conditional random fields.

4. The method of claim 1 wherein the step of capitalizing the target text further comprises:

assigning the computed configuration probability to each respective one or more capitalization configurations,

wherein the configuration probability is computed for the one or more capitalization configurations.

5. The method of claim 4 wherein the at least one capitalization model feature function includes a capitalized translation model feature function.

6. The method of claim 4 wherein the at least one capitalization model feature function includes a capitalization tag model feature function.

7. The method of claim 4 wherein the at least one capitalization model feature function includes an uppercase model feature function.

8. The method of claim 4 wherein the at least one capitalization model feature function includes a monolingual language model feature function.

9. The method of claim 4 wherein the at least one capitalization model feature function includes an initial position model feature function.

10. The method of claim 4 wherein the at least one capitalization model feature function includes a punctuation model feature function.

11. The method of claim 4, further including:

selecting a source phrase from the capitalized source text;

selecting a target phrase from the target text;

determining an alignment between the source phrase and the target phrase;

computing a word probability from capitalization information in the source phrase, the alignment, and the at least one capitalization model feature function for the one or more capitalization configurations; and

applying the word probability to the computing of the configuration probability.

12. The method of claim 11 wherein the at least one capitalization model feature function includes a capitalized translation model feature function.

13. The method of claim 11 wherein the at least one capitalization model feature function includes a capitalization tag model feature function.

14. The method of claim 11 wherein the at least one capitalization model feature function includes an uppercase model feature function.

15. A translation system comprising:

a device;

an automatic translator module executable and stored on the device and configured to automatically convert a capitalized source text to lower case text and translate the lower case text to a target text; and

a capitalization module configured to recover a capitalized text from the target text according to capitalization information in the capitalized source text and the target text, and capitalize the target text, the capitalization of the target text including;

generating a plurality of capitalization configurations for the target text;

for each capitalization configuration, computing a feature probability for each of a plurality of capitalization model feature functions;

associating a feature weight with each capitalization model feature function;

applying the associated feature weight to the respective computed feature probability for each of the plurality of capitalization model feature functions;

for each capitalization configuration, calculating a capitalization configuration probability based on a weighted sum of the computed feature probabilities and applied feature weights, and based on an alignment between the capitalized source text and the target text or the capitalized source text and the capitalization configuration; and

selecting the best capitalization configuration from the plurality of capitalization configurations based on the highest calculated capitalization configuration probability.

16. The translation system of claim 15, wherein the capitalization module is further configured to recover the capitalized text from the source text.

17. A translation system comprising:

a device;

an automatic translator module executable and stored on the device and configured to automatically convert a capitalized source text to lower case text and translate the lower case text to a target text;

an aligner configured to determine an alignment between one or more phrases in the capitalized source text and one or more respective phrases in the target text of a capitalization configuration; and

a capitalization module configured to recover a capitalized text from the target text according to capitalization information in the capitalized source text and the target text and the alignment determined by the aligner, and to capitalize the target text,

the capitalization of the target text including;

generating a plurality of capitalization configurations for the target text;

for each capitalization configuration, computing a feature probability for each of a plurality of capitalization model feature functions;

associating a feature weight with each capitalization model feature function;

applying the associated feature weight to the respective computed feature probability for each of the plurality of capitalization model feature functions;

for each capitalization configuration, calculating a capitalization configuration probability based on a weighted sum of the computed feature probabilities and applied feature weights, and based on the alignment between the one or more phrases in the capitalized source text and the one or more phrases in the target text or between the capitalized source text and the capitalization configuration;

assigning the calculated capitalization configuration probability to each respective capitalization configuration; and

selecting the best capitalization configuration from the plurality of capitalization configurations based on the highest calculated capitalization configuration probability.

18. The translation system of claim 17 wherein the capitalization module further includes a capitalized translation model feature function.

19. The translation system of claim 17 wherein the capitalization module further includes a capitalization tag model feature function.

20. The translation system of claim 17 wherein the capitalization module further includes an uppercase model feature function.

21. The translation system of claim 17 wherein the capitalization module further includes a monolingual language model feature function.

22. The translation system of claim 17 wherein the capitalization module further includes an initial position model feature function.

23. The translation system of claim 17 wherein the capitalization module further includes a punctuation model feature function.

24. A computer program embodied on a non-transitory computer readable medium having instructions for capitalizing translated text, comprising:

executing a translator module stored on a device to automatically translate a capitalized source text to a target text, the translation of the capitalized source text including converting source text to lower case and translating the lower case source text; and

capitalizing the target text according to capitalization information in the capitalized source text, the step of capitalizing the target text according to the capitalized source text including;

generating one or more capitalization configurations for the target text;

computing a configuration probability for each of the one or more capitalization configurations, the configuration probability computed from capitalization information in the capitalized source text and at least one capitalization model feature function based on an alignment between the capitalized source text and the target text or the capitalized source text and the capitalization configuration; and

selecting the best capitalization configuration based on the highest computed configuration probability.