Apparatus and method for generating translation model, apparatus and method for automatic translation

• US 10,198,435 B2
• Filed: 08/24/2016
• Issued: 02/05/2019
• Est. Priority Date: 11/17/2015
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Na, Hwi Dong
• Primary Examiner(s): Ky, Kevin
• Application Number:
  US15/245,474
• Publication Number:
  US 20170139905A1
• Time in Patent Office: 895 Days
• Field of Search: None
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/47 Machine-assisted translation, e.g. using translation memory
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G10L 15/005 Language recognition
• CPC Class Codes: G10L 15/16 using artificial neural networks
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

Disclosed herein are an apparatus and method for generating a translation model and an apparatus and method for automatic translation. A translation model generation apparatus and corresponding method include a learning data collector configured to collect an original text, an original text containing an error, and a translation of the original text as learning data, and a model generator that generates a translation model to simultaneously perform error correction and translation generation by enabling a neural network to learn based on the collected learning data.

Claims

1. An apparatus to generate a trained translation model using a processor, the apparatus comprising:

the processor comprising;

a learning data collector configured to collect an original text, an original text containing an error, and a translation of the original text as learning data; and

a model generator configured to generate the trained translation model to simultaneously perform an automated correcting of an error in a speech recognition and perform translation generation of a recognized speech, the trained translation model being generated by enabling a neural network of the trained translation model to learn based on the collected learning data.

2. The apparatus of claim 1, wherein the neutral network of the trained translation model comprises a first neutral network configured to convert the original text including an error into abstract information, a second neural network configured to generate the translation of the original text from the abstract information, and a third neural network configured to generate the original text from the abstract information.

3. The apparatus of claim 2, wherein the first neural network, the second neural network, and the third neural network are configured as one deep neural network (DNN) to enable the one DNN to learn through one learning process.

4. The apparatus of claim 2, wherein the abstract information is an embedding vector disposed on a multi-dimensional vector space.

5. The apparatus of claim 1, wherein the model generator uses a multi-task learning technique to enable the neural network to learn.

6. The apparatus of claim 5, wherein the model generator enables the neural network to learn by setting the original text including an error as an input and setting the original text and the translation of the original text as targets.

7. A processor-implemented method of generating a trained translation model, the method comprising:

collecting an original text, an original text containing an error, and a translation of the original text as learning data; and

generating the trained translation model by enabling a neural network of the trained translation model to learn based on the collected learning data, the trained translation model being used to simultaneously perform an automated correcting of an error occurring in a speech recognition and perform translation of a recognized speech.

8. The method of claim 7, wherein the neutral network of the trained translation model comprises a first neutral network configured to convert the original text including an error into abstract information, a second neural network configured to generate the translation of the original text from the abstract information, and a third neural network configured to generate the original text from the abstract information.

9. The method of claim 8, wherein the generating of the trained translation model configures the first neural network, the second neural network, and the third neural network as one deep neural network (DNN) to enable the one DNN to learn through one learning process.

10. The method of claim 8, wherein the abstract information is an embedding vector disposed on a multi-dimensional vector space.

11. The method of claim 7, wherein the generating of the trained translation model uses a multi-task learning technique to enable the neural network to learn.

12. The method of claim 11, wherein the generating of the trained translation model enables the neural network to learn by setting the original text including an error as an input and setting the original text and the translation of the original text as targets.

13. An apparatus for automatic translation, comprising:

a voice recognition processor configured to recognize an input voice;

a translation processor configured to simultaneously perform an automated correcting of an error in the recognizing and perform translation of the input voice to a voice recognition result using a pre-trained translation model; and

a processor configured to indicate a result of the automated correcting of the error and a result of the translation.

14. The apparatus of claim 13, wherein the pre-trained translation model is in advance generated by enabling a neural network to learn based on an original text, an original text including an error, and a translation of the original text.

15. The apparatus of claim 14, wherein the pre-trained translation model is in advance generated by enabling the neutral network to learn using a multi-task learning technique.

16. The apparatus of claim 15, wherein the pre-trained translation model is in advance generated by enabling the neural network to learn by setting the original text including an error as an input and setting the original text and the translation of the original text as targets.

17. A processor-implemented method for automatic translation, the method comprising:

recognizing an input voice;

simultaneously performing an automated correcting of an error in the recognizing and performing translation of the input voice to a voice recognition result using a pre-trained translation model; and

indicating a result of the automated correcting of the error and a result of the translation.

18. The method of claim 17, wherein the pre-trained translation model is in advance generated by enabling a neural network to learn based on an original text, an original text including an error, and a translation of the original text.

19. The method of claim 18, wherein the pre-trained translation model is in advance generated by enabling the neutral network to learn using a multi-task learning technique.

20. The method of claim 18, wherein the pre-trained translation model is in advance generated by enabling the neural network to learn by setting the original text including an error as an input and setting the original text and the translation of the original text as targets.

21. An automatic translation apparatus, comprising:

a processor configured to

receive a voice from a user in a first language,

perform voice recognition of the voice in the first language and generate a text in the first language as a voice recognition result,

translate the voice recognition result to a second language,

performing an automated correcting of an error occurring in the voice recognition result using a pre-trained translation model, and

generate a translation in the second language, using the pre-trained translation model, simultaneously with the automated correcting of the error.

22. The automatic translation apparatus of claim 21, wherein the processor comprises:

a voice receiver configured to receive the voice from the user in the first language;

a voice recognition processor configured to perform the voice recognition of the voice in the first language and generate the text in the first language as the voice recognition result; and

a translation processor configured to;

translate the voice recognition result to the second language,

perform the automated correcting of the error occurring in the voice recognition result using the pre-trained translation model, and

generate the translation in the second language, using the pre-trained translation model, simultaneously with the automated correcting of the error.

23. The automatic translation apparatus of claim 22, wherein the pre-trained translation model is in advance generated by teaching a neural network learning data, the learning data including an original text, an incorrect original text, and a translation of the original text, the incorrect original text being the original text including the error.

24. The automatic translation apparatus of claim 23, wherein the neural network comprises a first neutral network that converts the incorrect original text into abstract information, a second neural network that generates the translation of the original text based on the abstract information, and a third neural network that generates the original text by correcting an error based on the abstract information.

25. The automatic translation apparatus of claim 21, further comprising:

a storage configured to store a sound model, a language model, and a pronunciation dictionary used in the voice recognition, and configured to store the pre-trained translation model.

26. The automatic translation apparatus of claim 25, wherein the voice recognition processor recognizes the input voice in the first language based on the sound model, the language model, and the pronunciation dictionary.