METHOD OF AND A SYSTEM FOR TRANSLATION

• US 20120016657A1
• Filed: 07/13/2011
• Published: 01/19/2012
• Est. Priority Date: 07/13/2010
• Status: Active Grant

• Current Assignee: Dublin City University
• Original Assignee: Dublin City University
• Inventors: He, Yifan, Ma, Yanjun, Way, Andrew, van Genabith, Josef
• Granted Patent: US 8,805,669 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 40/47 Machine-assisted translation, e.g. using translation memory

Abstract

A translation system for translating source text from a first language to target text in a second language. The system comprises a translation memory (TM) module that stores translation segments. The TM module is operable to generate a TM target text output in response to source text. A statistical translation machine (SMT) module is configured to generate translations on the basis of statistical models whose parameters are derived from the analysis of bilingual text corpora. The SMT module is operable to generate a SMT target text output in response to source text. An extractor is configured to extract features from the TM target text output and the SMT target text output. A vector generator is configured to generate a vector with a unified feature set derived from the extracted features and features associated with the SMT module and the TM module. A recommender is operable to read the vector and determine whether the TM target text output or the SMT target text output is optimum for post editing.

Claims

1. A translation system for translating source text from a first language to target text in a second language, the system comprising:

a translation memory (TM) module storing translation segments;

the TM module being operable to generate a TM target text output in response to source text,

a statistical machine translation (SMT) module configured to generate translations on the basis of statistical models, the statistical models each having a respective number of parameters derived from the analysis of bilingual text corpora;

the SMT module being operable to generate a SMT target text output in response to source text,

an extractor configured to extract features from the TM target text output and the SMT target text output,

a vector generator configured to generate a vector with a unified feature set derived from the extracted features and features associated with each of the SMT module and the TM module, and

a recommender operable to read the vector and determine whether the TM target text output or the SMT target text output is optimum for post editing.

2. A system as claimed in claim 1, wherein the extractor is configured to encode the TM target text output into numerical values.

3. A system as claimed in claim 1, wherein the extractor is configured to encode the SMT target text output into numerical values.

4. A system as claimed in claim 1, wherein the extracted features are encoded into numerical values.

5. A system as claimed in claim 1, wherein the extracted features are calculated.

6. A system as claimed in claim 5 wherein the TM target text output and the SMT target text output are accessible.

7. A system as claimed in claim 1, wherein the extracted features include at least one of:

a number of source and target side LM scores, a number of pseudo source fuzzy match scores, or a number of IBM model 1 scores.

8. A system as claimed in claim 1, wherein the vector generator receives a fuzzy match score associated with the TM module.

9. A system as claimed in claim 8, wherein the fuzzy match score provides one of the features in the generated vector.

10. A system as claimed in claim 1, wherein the vector generator receives at least one of the following features from the SMT module:

a phrase-translation model score, a language model probability score, a distance-based reordering score, a lexicalized reordering model score, or a word penalty.

11. A system as claimed in claim 1, wherein the vector generator is operable to label the TM target text output and the SMT target text output.

12. A system as claimed in claim 11, wherein the label includes a numerical value.

13. A system as claimed in claim 12, wherein the vector generator labels the TM target text output and the SMT target text output based on a measure of the post-editing effort.

14. A system as claimed in claim 12, wherein the measure includes a number of translation edit rate (TER) scores.

15. A system as claimed in claim 1 wherein the system is a Computer Assisted Translation (CAT) environment.

16. A method of for translating source text from a first language to target text in a second language, the method comprising:

generating a translation memory (TM) target text output in response to a source text input,

generating a statistical machine translation (SMT) target text output in response to a source text input,

extracting features from the TM target text output and the SMT target text output,

generating a vector with a unified feature set derived from the extracted features and features associated with the SMT module and the TM module, and

reading the vector for determining whether the TM target text output or the SMT target text output is optimum for post editing.

17. An article of manufacture storing machine readable instructions which, when executed, cause a machine to:

generate a translation memory (TM) target text output in response to a source text input,

generate a statistical machine translation (SMT) target text output in response to a source text input,

extract features from the TM target text output and the SMT target text output,

generate a vector with a unified feature set derived from the extracted features and features associated with the SMT module and the TM module, and

read the vector for determining whether the TM target text output or the SMT target text output is optimum for post editing.

18. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to encode the TM target text output into numerical values.

19. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to encode the SMT target text output into numerical values.

20. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to encode the extracted features into numerical values.

21. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to calculate the extracted features.

22. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to access the TM target text output and the SMT target text output.

23. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to extract features which include at least one of:

a number of source and target side LM scores, a number of pseudo source fuzzy match scores, or a number or IBM model 1 scores.

24. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to receive a fuzzy match score associated with the TM module.

25. An article of manufacture as defined in claim 24, wherein the machine readable instructions further cause the machine to include the fuzzy match score in the generated vector.

26. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to receive at least one of the following features from the SMT module:

a number of phrase-translation model scores, a number of language model probability score, a number of distance-based reordering scores, a number of lexicalized reordering model scores, or a word penalty.

27. An article of manufacture as defined in claim 17, wherein the machine readable instructions further cause the machine to label the TM target text output and the SMT target text output.

28. An article of manufacture as defined in claim 27, wherein the machine readable instructions further cause the machine to label using a numerical value.

29. An article of manufacture as defined in claim 28, wherein the machine readable instructions further causes the machine to label the TM target text output and the SMT target text output based on a measure of a post-editing effort.

30. An article of manufacture as defined in claim 29, wherein the machine readable instructions further cause the machine to measure the post-editing effort using a number of translation edit rate (TER) scores.