Method and device for machine translation

Patent metadata

Field	Value
Publication number	US-10108607-B2
Application number	US-201615223950-A
Country	US
Kind code	B2
Filing date	Jul 29, 2016
Priority date	Jul 30, 2015
Publication date	Oct 23, 2018
Grant date	Oct 23, 2018

Abstract

A machine translation method includes determining source language text to be translated and obtaining a translation rule table, which has been trained in advance, that includes multiple translation rules associated with the target language text and the source language text in multiple languages; determining candidate results of the target language text; and determine the target language text to be output based on the candidate results. During the translation, a specific language of the source language text need not to be specified by a user. The implementations improve accuracy of the translation, and avoid errors introduced from the process of language identification during recognizing unknown languages. The implementations avoid developing a translation engine for an individual source language of text to be translated for a certain target language, and therefore save development costs and computing resources.

First claim

What is claimed is:

1. A method for machine translation comprising:
determining, by one or more processors of a computing device, a source language text to be translated;
obtaining, by the one or more processors, a translation rule table comprising one or more translation rules, wherein the one or more translation rules is associated with at least a first bilingual parallel corpus and a second bilingual parallel corpus, the first bilingual parallel corpus forming a first language pair between a first source language and a target language, and the second bilingual parallel corpus forming a second language pair between a second source language and the target language;
determining, by the one or more processors, a candidate result of a target language text based at least in part on the translation rule table; and
determining, by the one or more processors, the target language text to be output based on the candidate result.

2. The method of claim 1 , wherein the determining the candidate result of the target language text comprises:
dividing the source language text using a phrase as a unit to obtain one or more phrases;
obtaining the one or more translation rules associated with the target language text and the source language text in the multiple languages to determine a translation result corresponding to each of the phrases; and
splicing translation results corresponding to the phrases, or splicing the phrases and searching the one or more translation rules associated with the target language text and the source language text in the multiple languages, to obtain at least one candidate result of the target language text.

3. The method of claim 2 , wherein the determining the target language text to be output based on the candidate result comprises:
determining that the source language text comprises phrases in a single language;
performing language model scoring, translation rule table scoring, and reordering model scoring to each of candidate results; and
selecting one or more candidate results as the target language text to be output based on scoring results.

4. The method of claim 3 , wherein the selecting the one or more candidate results as the target language text to be output based on the scoring results comprises:
multiplying a score of the language model scoring by a weight of the language model, multiplying a score of the translation rule table scoring by a weight of the translation rule table, and multiplying a score of the reordering model scoring by a weight of the reordering model to obtain products for each candidate result of the one or more candidate results;
adding the products to obtain a value corresponding to each candidate result of the one or more candidate results; and
selecting at least one candidate result having a maximum value as the target language text to be output.

5. The method of claim 2 , wherein the determining the target language text to be output based on the candidate result comprises:
determining that the source language text comprises phrases in multiple languages;
performing language model scoring and translation rule table scoring to each of candidate results; and
selecting one or more candidate results as the target language text to be output based on scoring results.

6. The method of claim 5 , wherein the selecting the one or more candidate results as the target language text to be output based on the scoring results comprises:
multiplying a score of the language model scoring by a weight of the language model and multiplying a score of the translation rule table scoring by a weight of the translation rule table to obtain products for each candidate result of the one or more candidate results;
adding the products to obtain a value corresponding to each candidate result of the one or more candidate results; and
selecting at least one candidate result having a maximum value as the target language text to be output.

7. A device for machine translation comprising:
one or more processors; and
memory to maintain a plurality of components executable by the one or more processors, the plurality of components comprising:
a first module configured to:
determine a source language text to be translated,
obtain a translation rule table comprising one or more translation rules associated with a target language text and the source language text in multiple languages, the target language text and the source language text in the multiple languages forming multiple language pairs, the translation rule table trained in advance, and
determine a candidate result of the target language text, and

a second module configured to determine the target language text to be output based on the candidate result.

8. The device of claim 7 , wherein the first module is further configured to:
divide the source language text using a phrase as a unit to obtain phrases;
obtain the one or more translation rules associated with the target language text and the source language text in the multiple languages to determine a translation result corresponding to each of the phrases; and
splice the translation result corresponding to the each of the phrases, or splicing the phrases and searching the one or more translation rules associated with the target language text and the source language text in the multiple languages, to obtain at least one candidate result of the target language text.

9. The device of claim 8 , wherein when the source language text comprises phrases in a single language, the second module is further configured to further:
perform language model scoring, translation rule table scoring, and reordering model scoring to each of candidate results; and
select one or more candidate results as the target language text to be output based on scoring results.

10. The device of claim 9 , wherein the second module is further configured to:
multiply a score of the language model scoring by a weight of the language model, multiply a score of the translation rule table scoring by a weight of the translation rule table, and multiply a score of the reordering model scoring by a weight of the reordering model to obtain products for each candidate result of the one or more candidate results;
add the products to obtain a value corresponding to each candidate result of the one or more candidate results; and
select at least one candidate result having a maximum value as the target language text to be output.

11. The device of claim 7 , wherein the second module is further configured to:
perform language model scoring and translation rule table scoring to each of candidate results when the source language text comprises phrases in multiple languages; and
select one or more candidate results as the target language text to be output based on scoring results.

12. The device of claim 11 , wherein the second module is further configured to:
multiply a score of the language model scoring by a weight of the language model and multiply a score of the translation rule table scoring by a weight of the translation rule table to obtain products for each candidate result of the one or more candidate results;
add the products to obtain a value corresponding to each candidate result of the one or more candidate results; and
select at least one candidate result having a maximum value as the target language text to be output.

13. One or more computer-readable media storing computer-executable instructions that, when executed by one or more processors, instruct the one or more processors to perform acts comprising:
determining a source language text to be translated;
obtaining a translation rule table comprising one o

Assignees

• Alibaba Group Holding Ltd

Inventors

• Song Kai
• Luo Weihua
• Lin Feng

Classifications

• G06F40/45

  Example-based machine translation; Alignment · CPC title

• G06F40/58

  Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation · CPC title

• G06F40/55 Primary

  Rule-based translation · CPC title

• G06F40/289

  Phrasal analysis, e.g. finite state techniques or chunking · CPC title

• G06F40/44 Primary

  Statistical methods, e.g. probability models · CPC title