Method and apparatus for machine translation using neural network and method of training the apparatus

Patent metadata

Field	Value
Publication number	US-10474758-B2
Application number	US-201815975927-A
Country	US
Kind code	B2
Filing date	May 10, 2018
Priority date	Jun 21, 2017
Publication date	Nov 12, 2019
Grant date	Nov 12, 2019

Abstract

A machine translation method and a machine translation apparatus using a neural network model are provided. The machine translation apparatus extracts information associated with a keyword from a source sentence, obtains a supplement sentence associated with the source sentence based on the extracted information associated with the keyword, acquires a first vector value from the source sentence and a second vector value from the supplement sentence using neural network model-based encoders, and outputs a target sentence corresponding to a translation of the source sentence based on any one or any combination of the first vector value and the second vector value using a neural network model-based decoder.

First claim

What is claimed is:

1. A machine translation method using a neural network model, the method comprising:
extracting information associated with a keyword from a source sentence;
obtaining a supplement sentence associated with the source sentence based on the extracted information associated with the keyword;
acquiring, using neural network model-based encoders, a first vector from the source sentence and a second vector from the supplement sentence; and
outputting, using a neural network model-based decoder, a target sentence corresponding to a translation of the source sentence based on any one or any combination of the first vector or the second vector.

2. The method of claim 1 , wherein the outputting of the target sentence comprises:
determining a third vector by combining the first vector and the second vector; and
acquiring the target sentence based on the third vector.

3. The method of claim 1 , wherein the outputting of the target sentence comprises:
outputting the target sentence based on a third vector determined by combining the first vector and the second vector in response, to the first vector satisfying a condition; and
outputting the target sentence based on the first vector, in response to the first vector not satisfying the condition.

4. The method of claim 1 , wherein the first vector is acquired from the second vector and the source sentence using the neural network model-based encoders.

5. The method of claim 1 , wherein the obtaining of the supplement sentence comprises searching for the supplement sentence based on the information associated with the keyword.

6. The method of claim 5 , wherein the extracting of the information associated with the keyword comprises extracting the information associated with the keyword based on any one or any combination of a number of documents including the keyword, a number of times the keyword appears in the documents, and a number of documents searched.

7. The method of claim 5 , wherein the searching for the supplement sentence comprises searching for the supplement sentence from any one or any combination of a database or the Internet based on the information associated with the keyword.

8. The method of claim 1 , wherein the information associated with the keyword comprises any one or any combination of the keyword, a plurality of keywords, a value indicating the keyword, a value indicating the plurality of keywords, and a value indicating a feature of an entire sentence including the keyword.

9. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, causes the processor to perform the method of claim 1 .

10. A machine translation training method using a neural network model, the method comprising:
extracting information associated with a keyword from a training source sentence;
obtaining a training supplement sentence associated with the training source sentence based on the extracted information associated with the keyword;
acquiring, using neural network model-based encoders, a first vector from the training source sentence and a second vector from the training supplement sentence;
outputting, using a neural network model-based decoder, a target sentence corresponding to translation of the training source sentence based on any one or any combination of the first vector or the second vector; and
training the neural network model-based encoders and the neural network model-based decoder by evaluating an accuracy of the target sentence.

11. The method of claim 10 , wherein the outputting of the target sentence comprises:
determining a third vector by combining the first vector and the second vector; and
acquiring the target sentence based on the third vector.

12. The method of claim 10 , wherein the outputting of the target sentence comprises:
outputting the target sentence based on a third vector determined by combining the first vector and the second vector, in response to the first vector satisfying a condition; and
outputting the target sentence based on the first vector, in response to the first vector not satisfying the condition.

13. The method of claim 10 , wherein the first vector is acquired from the second vector and the training source sentence using the neural network model-based encoders.

14. The method of claim 10 , wherein the obtaining of the training supplement sentence comprises searching for the training supplement sentence based on the information associated with the keyword.

15. The method of claim 14 , wherein the extracting of the information associated with the keyword comprises extracting the information associated with the keyword based on any one or any combination of a number of documents including the keyword, a number of times the keyword appears in the documents, and a number of documents searched.

16. The method of claim 14 , wherein the searching for the training supplement sentence comprises searching for the training supplement sentence from any one or any combination of a database or the Internet based on the information associated with the keyword.

17. The method of claim 10 , wherein the accuracy of the target sentence is evaluated based on a comparison of the target sentence to another translation of the training source sentence.

18. A machine translation apparatus using a neural network model, the apparatus comprising:
a processor configured to:
extract information associated with a keyword from a source sentence,
obtain a supplement sentence associated with the source sentence based on the extracted information associated with the keyword,
acquire, using the neural network model, a first vector from the source sentence and a second vector from the supplement sentence, and
output, using the neural network model, a target sentence corresponding to a translation of the source sentence based on any one or any combination of the first vector and the second vector.

19. The apparatus of claim 18 , wherein the processor is further configured to determine a third vector by combining the first vector and the second vector, and to output the target sentence based on the third vector using a neural network model-based decoder model of the neural network model.

20. The apparatus of claim 18 , wherein the processor is further configured to output the target sentence based on a third vector determined by combining the first vector and the second vector using a neural network model-based decoder model of the neural network model, in response to the first vector satisfying a condition, and to output the target sentence based on the first vector using the neural network model-based decoder model, in response to the first vector not satisfying the preset condition.

21. The apparatus of claim 18 , wherein the processor is further configured to acquire the first vector from the second vector and the source sentence using neural network model-based encoders model of the neural network model.

22. A translation apparatus based on a neural network model set, the apparatus comprising:
a sensor configured to receive a source sentence; and
a processor configured to
extract information associated with a keyword from the source sentence,
obtain a supplement sentence associated with the source sentence based on the extracted information,
acquire, using neural network model-based encoders, a first vector from the source sentence and a second vector from the supplement sentence, and
output, using neural network model-based

Assignees

• Samsung Electronics Co Ltd

Inventors

• Lee Min-Joong
• Choi Youngsang

Classifications

• G06N3/08 Primary

  Learning methods · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06F40/53

  Processing of non-Latin text (kana-to-kanji conversion G06F40/129; vowelisation G06F40/232) · CPC title

• G06F40/58 Primary

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