Statistical noun phrase translation

• US 20050038643A1
• Filed: 07/02/2004
• Published: 02/17/2005
• Est. Priority Date: 07/02/2003
• Status: Active Grant

• Current Assignee: University of Southern California
• Original Assignee: University of Southern California
• Inventors: Koehn, Philipp
• Granted Patent: US 8,548,794 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models

Abstract

A statistical machine translation (MT) system may include a noun phrase/prepositional phrase (NP/PP) translation subsystem to translation NP/PPs as a subtask in an MT operation. The NP/PP translation subsystem may use a model trained on an NP/PP corpus and a decoder to generate an n-best list of candidate translations and a re-ranker to re-rank the candidate translations based on a machine learning method and additional features based on known properties of NP/PPs.

Claims

1. A method comprising:

receiving an input string in a source language;

detecting at least one of a noun phrase and a prepositional phrase in the input string;

generating a translation in a target language for said at least one of the noun phrase and the prepositional phrase; and

generating a translation of the input string in the target language including the translation for said at least one of the noun phrase and the prepositional phrase.

2. The method of claim 1, wherein said generating the translation for said at least one of the noun phrase and the prepositional phrase comprises generating the translation using a first model;

and

wherein said generating the translation of the input string comprises generating the translation using a second model different than the first model.

3. The method of claim 2, wherein the first model comprises a model trained using a corpus consisting of noun phrases and prepositional phrases.

4. The method of claim 2, wherein the first model comprises a phrase-based model.

5. The method of claim 1, wherein said generating the translation for said at least one of the noun phrase and the prepositional phrase comprises:

generating a ranked n-best list of candidate translations; and

re-ranking the n-best list using a machine learning method.

6. The method of claim 5, wherein said machine learning method comprises a maximum entropy learning method.

7. The method of claim 5, wherein said re-ranking comprises re-ranking using one or more additional features based on known properties of noun phrases and prepositional phrases.

8. The method of claim 7, wherein the one or more additional features comprise syntactic features.

9. The method of claim 7, wherein the one or more additional features comprises using the web as a language model.

10. A machine translation system comprising:

a parser to detect at least one of a noun phrase and a prepositional phrase in an input string in a source language;

a translation subsystem including

a first model to generate possible translations in a target language for said at least one of the noun phrase and the prepositional phrase,

a first decoder to generate a ranked n-best list including the possible translations, and

a selector to select a translation for said at least one of the noun phrase and the prepositional phrase;

a second model to generate possible translations in the target language for the input string in the source language, each of the possible translations comprising the selected translation for said at least one of the noun phrase and the prepositional phrase; and

a second decoder to select one of the possible translations for the input string.

11. The system of claim 10, wherein the first model comprises a model trained using a corpus consisting of noun phrases and prepositional phrases.

12. The system of claim 10, wherein the first model comprises a phrase-based model.

13. The system of claim 10, wherein the translation subsystem further comprises:

a re-ranker to re-rank the n-best list using a machine learning method.

14. The system of claim 13, wherein the machine learning method comprises a maximum entropy learning method.

15. The system of claim 13, wherein the re-ranker is operative to re-rank the n-best list using one or more additional features based on known properties of noun phrases and prepositional phrases.

16. The system of claim 15, wherein the one or more additional features comprise syntactic features.

17. The system of claim 15, wherein the one or more additional features comprises using the web as a language model.

18. An article comprising a machine-readable medium including machine-executable instructions, the instruction operative to cause a machine to:

detect at least one of a noun phrase and a prepositional phrase in an input string in a source language;

generate a translation in a target language for said at least one of the noun phrase and the prepositional phrase; and

generate a translation of the input string in the target language including the translation for said at least one of the noun phrase and the prepositional phrase.

19. The article of claim 18, wherein the instructions for generating the translation for said at least one of the noun phrase and the prepositional phrase comprise instructions operative to cause the machine to generate the translation using a first model;

and

wherein the instructions for generating the translation of the input string comprise instructions operative to cause the machine to generate the translation using a second model.

20. The article of claim 19, wherein the first model comprises a model trained using a corpus consisting of noun phrases and prepositional phrases.

21. The article of claim 19, wherein the first model comprises a phrase-based model.

22. The article of claim 18, wherein the instructions for generating the translation for said at least one of the noun phrase and the prepositional phrase comprise instructions operative to cause the machine to:

generate a ranked n-best list of candidate translations; and

re-rank the n-best list using a machine learning article.

23. The article of claim 22, wherein said machine learning article comprises a maximum entropy learning article.

24. The article of claim 22, wherein the instructions for re-ranking comprise instructions operative to cause the machine to re-rank using one or more additional features based on known properties of noun phrases and prepositional phrases.

25. The article of claim 24, wherein the one or more additional features comprise syntactic features.

26. The article of claim 24, wherein the one or more additional features comprises using the web as a language model.