POST-EDITING APPARATUS AND METHOD FOR CORRECTING TRANSLATION ERRORS

• US 20100138210A1
• Filed: 11/19/2009
• Published: 06/03/2010
• Est. Priority Date: 12/02/2008
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Kim, Yong Kil, Park, Sang Kyu, Lee, Ki Young, Huang, Jinxia, Kwon, Oh Woog, Kim, Chang Hyun, Jin, Yun, Yang, Seong II, Yin, Changhao, Park, Eun Jin, Roh, Yoon Hyung, Choi, Sung Kwon, SEO, Young Ae, Wu, Ying Shun
• Granted Patent: US 8,494,835 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 40/232 Orthographic correction, e.g. spell checking or vowelisation
• CPC Class Codes: G06F 40/51 Translation evaluation

Abstract

A post-editing apparatus for correcting translation errors, includes: a translation error search unit for estimating translation errors using an error-specific language model suitable for a type of error desired to be estimated from translation result obtained using a translation system, and determining an order of correction of the translation errors; and a corrected word candidate generator for sequentially generating error-corrected word candidates for respective estimated translation errors on a basis of analysis of an original text of the translation system. The post-editing apparatus further includes a corrected word selector for selecting a final corrected word from among the error-corrected word candidates by using the error-specific language model suitable for the type of error desired to be corrected, and incorporating the final corrected word in the translation result, thus correcting the translation errors.

Claims

1. A post-editing apparatus for correcting translation errors, comprising:

a translation error search unit for estimating translation errors using an error-specific language model suitable for a type of error desired to be estimated from translation result obtained using a translation system, and determining an order of correction of the translation errors;

a corrected word candidate generator for sequentially generating error-corrected word candidates for respective estimated translation errors on a basis of analysis of an original text of the translation system; and

a corrected word selector for selecting a final corrected word from among the error-corrected word candidates by using the error-specific language model suitable for the type of error desired to be corrected, and incorporating the final corrected word in the translation result, thus correcting the translation errors.

2. The post-editing apparatus of claim 1, wherein the error-specific language model is built from a target language corpus in a form specified for types of translation errors.

3. The post-editing apparatus of claim 2, wherein the error-specific language model is built in such a way that factors required for correction of errors for respective error types are separately defined for word choice error, word order error based on the corpus for which dependency was analyzed.

4. The post-editing apparatus of claim 3, wherein the word choice error is at least one of an error in selection of a translated word such as a noun, a verb, an adjective, an adverb, an article, a preposition, and an auxiliary verb, an error in singular/plural agreement, and an error in determining whether singular or plural.

5. The post-editing apparatus of claim 3, wherein the word order error is an error in an order of adjective sequence and an order in nominal compound.

6. The post-editing apparatus of claim 3, wherein the word presence error is an error related to presence of an article, a preposition and an auxiliary verb or adjective.

7. The post-editing apparatus of claim 1, wherein the translation error search unit sets priorities for correction of found translation errors in conformity with error correction priority determination rules.

8. The post-editing apparatus of claim 7, wherein the error correction priority determination rules are made in such a way that content words have higher correction priority than function words, modifier has higher correction priority than modified, and the word choice error has higher correction priority than word order error, and that, among the content words and among the modifiers, words having a high probability of errors have higher correction priority.

9. The post-editing apparatus of claim 7, wherein the corrected word candidate generator sequentially corrects the errors based on the error correction priorities set by the translation error search unit.

10. The post-editing apparatus of claim 1, wherein the corrected word selector calculates probabilities of sentences in which erroneous word in each erroneous sentence is replaced with relevant error-corrected word candidate by using the error-specific language model, and selects a word having a highest probability, from among the error-corrected word candidates, as a corrected word.

11. A post-editing method for correcting translation errors, comprising:

estimating translation errors using an error-specific language model suitable for a type of error desired to be estimated from translation result obtained using a translation system;

generating error-corrected word candidates for respective estimated translation errors on a basis of an analysis of original text by the translation system; and

selecting a final corrected word from among the error-corrected word candidates using the error-specific language model suitable for the type of error desired to be corrected, and incorporating the final corrected word in the translation result, thus correcting the translation errors.

12. The post-editing method of claim 11, wherein the error-specific language model is built from a target language corpus in a form specified for types of translation errors.

13. The post-editing method of claim 12, wherein the error-specific language model is built in such a way that factors required for correction of errors for respective error types are separately defined for word choice error, word order error and word presence error, based on the corpus for which dependencies were analyzed.

14. The post-editing method of claim 13, wherein the word choice error is at least one of an error in selection of a translated word such as a noun, a verb, an adjective, an adverb, an article, a preposition, and an auxiliary verb, an error in singular/plural agreement, and an error in determining whether singular or plural.

15. The post-editing method of claim 13, wherein the word order error is an error in an order of adjective sequence and an order in nominal compound.

16. The post-editing method of claim 13, wherein the word presence error is an error related to presence of an article, a preposition and an auxiliary verb or adjective.

17. The post-editing method of claim 11, wherein the estimating errors is performed to set priorities for correction of found translation errors in conformity with error correction priority determination rules.

18. The post-editing method of claim 17, wherein the error correction priority determination rules are made in such a way that content words have higher correction priority than function words, modifier has higher correction priority than modified, and the word choice error has higher correction priority than word order error, and that, among the content words and among the modifiers, words having a high probability of errors have higher correction priority.

19. The post-editing method of claim 17, wherein said generating the error-corrected word candidates is performed to sequentially correct the errors based on the set error correction priorities.

20. The post-editing method of claim 11, wherein said correcting the errors includes:

calculating probabilities of sentences in which an erroneous word in each erroneous sentence is replaced with relevant error-corrected word candidate by using the error-specific language model; and

selecting a word having a highest probability, from among the corrected word candidates, as a corrected word.