Computerized statistical machine translation with phrasal decoder

• US 9,176,952 B2
• Filed: 09/25/2008
• Issued: 11/03/2015
• Est. Priority Date: 09/25/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Aikawa, Takako, Ruopp, Achim
• Primary Examiner(s): BAKER, MATTHEW H
• Application Number:
  US12/238,207
• Publication Number:
  US 20100076746A1
• Time in Patent Office: 2,595 Days
• Field of Search: 704 1- 10
• US Class Current: 1/1
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/45 Example-based machine translation; Alignment
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation

Abstract

A computerized system for performing statistical machine translation with a phrasal decoder is provided. The system may include a phrasal decoder trained prior to run-time on a monolingual parallel corpus, the monolingual parallel corpus including a machine translation output of source language documents of a bilingual parallel corpus and a corresponding target human translation output of the source language documents, to thereby learn mappings between the machine translation output and the target human translation output. The system may further include a statistical machine translation engine configured to receive a translation input and to produce a raw machine translation output, at run-time. The phrasal decoder may be configured to process the raw machine translation output, and to produce a corrected translation output based on the learned mappings for display on a display associated with the system.

Claims

1. A computerized system for performing statistical machine translation, the system comprising:

a statistical machine translation engine executed on a user computing device, the statistical machine translation engine trained on a bilingual parallel corpus including source language documents and a corresponding target human translation of the source language documents, and configured to receive a translation input and to produce a raw machine translation output, at run-time;

a phrasal decoder, separate and distinct from the statistical machine translation engine, executed on the user computing device, the phrasal decoder being trained prior to run-time on a monolingual parallel corpus, the monolingual parallel corpus including a machine translation output of the source language documents of the bilingual parallel corpus and the corresponding target human translation output of the source language documents of the bilingual parallel corpus, to thereby learn mappings and build a phrase table by establishing phrase pairs between the machine translation output and the target human translation output, wherein the machine translation output is unedited by human translators, and wherein the phrasal decoder is trained prior to run-time on a developer computing device on which the bilingual parallel corpus is stored, assigning to each phrase pair a statistical score representing a utility of each phrase pair; and

wherein at run-time on the user computing device the phrasal decoder is configured to process the raw machine translation output, and to produce a corrected translation output based on the learned mappings and the phrase table, programmatically correcting the raw machine translation output if a statistical score for correspondence of the phrase pair is above a predetermined threshold.

2. The computerized system of claim 1, wherein the phrasal decoder is configured to determine the learned mappings by applying a word alignment algorithm.

3. The computerized system of claim 2, wherein the word alignment algorithm is selected from the group consisting of a hidden markov model (HMM), an expectation-maximization (EM) model, a discriminative model, and a syntax-based model.

4. The computerized system of claim 1, wherein the bilingual parallel corpus includes bi-text training data for one or more language pairs, the bi-text training data including source language documents and target human translation output for each language pair.

5. The computerized system of claim 1, wherein the statistical machine translation engine is configured to translate between each of a plurality of language pairs, each language pair having a source language and a target language;

wherein the monolingual parallel corpus is one of a plurality of monolingual parallel corpora that the phrasal decoder is trained on, each of the monolingual parallel corpora being for a target language in one of the language pairs, and each of the monolingual parallel corpora including a machine translation output and a corresponding target human translation output; and

wherein the target human translation output for each monolingual parallel corpora is from a corresponding bilingual parallel corpus for one of the language pairs.

6. The computerized system of claim 5, wherein the language pairs are typologically different language pairs.

7. The computerized system of claim 1, wherein the statistical machine translation engine includes a syntax-based statistical machine translation engine;

and

wherein the phrasal decoder includes a phrase-based statistical machine translation engine.

8. The computerized system of claim 1, wherein the phrasal decoder is configured to produce the corrected translation output without prompting a user for a correction input.

9. A computerized method of statistical machine translation, the method comprising:

training a statistical machine translation engine on a bilingual parallel corpus including source language documents and a corresponding target human translation of the source language documents;

training a phrasal decoder, separate and distinct from the statistical machine translation engine, on a monolingual parallel corpus, the monolingual parallel corpus including a machine translation output of the source language documents of the bilingual parallel corpus and the corresponding target human translation output of the source language documents of the bilingual parallel corpus, to thereby learn mappings and build a phrase table by establishing phrase pairs between the machine translation output and the target human translation output, wherein the machine translation output is unedited by human translators, assigning to each phrase pair a statistical score representing a utility of each phrase pair;

performing statistical machine translation via the statistical machine translation engine trained on the bilingual parallel corpus of a translation input to thereby produce a raw machine translation output; and

processing the raw machine translation output to thereby produce a corrected translation output based on the learned mappings and the phrase table, programmatically correcting the raw machine translation output if a statistical score for correspondence of the phrase pair is above a predetermined threshold.

10. The computerized method of claim 9, wherein learning mappings between the machine translation output and the target human translation output includes applying a word alignment algorithm.

11. The computerized method of claim 10, wherein the word alignment algorithm is selected from the group consisting of a hidden markov model (HMM), an expectation-maximization (EM) model, a discriminative model, and a syntax-based model.

12. The computerized method of claim 9, wherein the bilingual parallel corpus includes bi-text training data for one or more language pairs, the bi-text training data including source language documents and target human translation output for each language pair.

13. The computerized method of claim 9, wherein performing statistical machine translation of a translation input includes translating between each of a plurality of language pairs, each language pair having a source language and a target language;

wherein the monolingual parallel corpus is one of a plurality of monolingual parallel corpora that the phrasal decoder is trained on, each of the monolingual parallel corpora being for a target language in one of the language pairs, and each of the monolingual parallel corpora including a machine translation output and a corresponding target human translation output; and

wherein the target human translation output for each monolingual parallel corpora is from a corresponding bilingual parallel corpus for one of the language pairs.

14. The computerized method of claim 13, wherein translating between each of the language pairs includes translating between each of a plurality of typologically different language pairs.

15. The computerized method of claim 9, wherein training the phrasal decoder occurs on a developer computing device on which the bilingual parallel corpus is stored;

wherein performing statistical machine translation of a translation input to thereby produce a raw machine translation output and processing the raw machine translation output to thereby produce a corrected translation output occurs on a user computing device.

16. The computerized method of claim 9, wherein performing statistical machine translation of a translation input is accomplished at least in part by a syntax-based statistical machine translation engine;

and

wherein processing the raw machine translation output to thereby produce a corrected translation output is accomplished at least in part by a phrase-based statistical machine translation engine.

17. The computerized method of claim 9, wherein processing the raw machine translation output to thereby produce a corrected translation output occurs without prompting a user for a correction input.

18. A computerized system for performing statistical machine translation, the system comprising:

a user computing device configured to execute;

a statistical machine translation engine trained on a bilingual parallel corpus including source language documents and a corresponding target human translation of the source language documents and configured to receive a translation input and to produce a raw machine translation output, at run-time; and

a phrasal decoder, separate and distinct from the statistical machine translation engine, configured at run-time to process the raw machine translation output, and to produce a corrected translation output for display on a display associated with the system;

wherein the phrasal decoder, separate and distinct from the statistical machine translation engine, is trained on a developer computing device prior to run-time based on a monolingual parallel corpus, the monolingual parallel corpus including a machine translation output of the bilingual parallel corpus and the corresponding target human translation output included within the bilingual parallel corpus, the machine translation output being unedited by human translators, to thereby learn mappings and build a phrase table by establishing phrase pairs between the machine translation output and the target human translation output, assigning to each phrase pair a statistical score representing a utility of each phrase pair, and wherein at run time on the user computing device the corrected translation output is produced based on a plurality of the learned mappings and the phrase table, programmatically correcting the raw machine translation output if a statistical score for correspondence of the phrase pair is above a predetermined threshold.