METHOD FOR CONTROLLING IDENTIFICATION AND IDENTIFICATION CONTROL APPARATUS

• US 20180046619A1
• Filed: 07/28/2017
• Published: 02/15/2018
• Est. Priority Date: 08/09/2016
• Status: Active Grant

• Current Assignee: Panasonic Intellectual Property Management Co., Ltd. (Panasonic Holdings Corporation)
• Original Assignee: Panasonic Intellectual Property Management Co., Ltd. (Panasonic Holdings Corporation)
• Inventors: SHI, HONGJIE, USHIO, TAKASHI, ENDO, MITSURU, YAMAGAMI, KATSUYOSHI
• Granted Patent: US 10,460,039 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 3/02 Input arrangements using manually operated switches, e.g. using keyboards or dials
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/45 Example-based machine translation; Alignment
• CPC Class Codes: G06F 40/47 Machine-assisted translation, e.g. using translation memory
• CPC Class Codes: G06N 3/042 Knowledge-based neural networks; Logical representations of neural networks
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G10L 15/16 using artificial neural networks

Abstract

A method for controlling identification includes obtaining first text, which is text in a first language, obtaining second text, which is text in a second language obtained by translating the first text into the second language, obtaining correct labels, which indicate content of the first text, inputting the first text and the second text to an identification model common to the first and second languages, and updating the common identification model such that labels identified by the common identification model from the first text and the second text match the correct labels.

Claims

1. A method for controlling identification, the method comprising:

obtaining first text, which is text in a first language;

obtaining second text, which is text in a second language obtained by translating the first text into the second language;

obtaining correct labels, which represent content of the first text;

inputting the first text and the second text to an identification model common to the first and second languages; and

updating the common identification model such that labels identified by the common identification model from the first text and the second text match the correct labels.

2. The method according to claim 1, further comprising:

converting the first text into a first value string using a first conversion model, which converts a plurality of elements expressed in the first language into a plurality of close values when the plurality of elements are semantically close to one another; and

converting the second text into a second value string using a second conversion model, which converts a plurality of elements expressed in the second language into a plurality of close values when the plurality of elements are semantically close to one another,

wherein, in the inputting the first text and the second text to the identification model common to the first and second languages, the first text converted into the first value string and the second text converted into the second value string are input to the common identification model.

3. The method according to claim 2,

wherein the first conversion model converts the plurality of elements expressed in the first language into a plurality of close vectors when the plurality of elements are semantically close to one another,

wherein the second conversion model converts the plurality of elements expressed in the second language into a plurality of close vectors when the plurality of elements are semantically close to one another,

wherein the first value string is a first vector string, and

wherein the second value string is a second vector string.

4. The method according to claim 1,

wherein the common identification model is a convolutional neural network model.

5. The method according to claim 4,

wherein the convolutional neural network model includes a plurality of convolutional layers and a plurality of pooling layers alternately stacked on one another.

6. The method according to claim 4,

wherein the convolutional neural network model separately includes at least one convolutional layer corresponding to the first language and at least one convolutional layer corresponding to the second language.

7. The method according to claim 4,

wherein the convolutional neural network model separately includes at least one pooling layer corresponding to the first language and at least one pooling layer corresponding to the second language.

8. The method according to claim 4,

wherein the convolutional neural network model includes at least one pooling layer common to the first and second languages.

9. The method according to claim 4,

wherein the convolutional neural network model includes at least one convolutional layer common to the first and second languages.

10. The method according to claim 1,

wherein the common identification model is a recurrent neural network model.

11. The method according to claim 1,

wherein the common identification model includes

a first extraction layer in which a first feature, which is a feature of text in the first language, is extracted,

a second extraction layer in which a second feature, which is a feature of text in the second language, is extracted, and

a combining layer in which the first feature and the second feature are combined with each other.

12. The method according to claim 11,

wherein, in the combining layer, a plurality of values indicating the first and second features are connected with each other.

13. The method according to claim 11,

wherein, in the combining layer, a plurality of values indicating the first and second features are subjected to linear connection.

14. The method according to claim 11,

wherein, in the combining layer, a probability of each of a plurality of labels is calculated through a process including combining and weighted addition of a plurality of values indicating the first and second features, and

wherein the labels identified by the common identification model are identified on the basis of the probabilities calculated in the combining layer.

15. The method according to claim 1, further comprising:

obtaining third text, which is text in the first language;

obtaining fourth text, which is text in the second language, either the third text or the fourth text being text obtained by translating the other text into the language corresponding to the foregoing text;

inputting the third text and the fourth text to the updated common identification model; and

identifying labels identified by the common identification model from the third text and the fourth text as labels representing the third text or the fourth text.

16. The method according to claim 15, further comprising:

converting the third text into a third value string using a first conversion model, which converts a plurality of elements expressed in the first language into a plurality of close values when the plurality of elements are semantically close to one another; and

converting the fourth text into a fourth value string using a second conversion model, which converts a plurality of elements expressed in the second language into a plurality of close values when the plurality of elements are semantically close to one another,

wherein, in the inputting the third text and the fourth text to the updated common identification model, the third text converted into the third value string and the fourth text converted into the fourth value string are input to the common identification model.

17. The method according to claim 16,

wherein the first conversion model converts the plurality of elements expressed in the first language into a plurality of close vectors when the plurality of elements are semantically close to one another,

wherein the second conversion model converts the plurality of elements expressed in the second language into a plurality of close vectors when the plurality of elements are semantically close to one another,

wherein the third value string is a third vector string, and

wherein the fourth value string is a fourth vector string.

18. An identification control apparatus comprising:

a control circuit that executes a method for controlling identification; and

a storage circuit that stores information used by the control circuit to execute the method for controlling identification,

wherein the method for controlling identification includes

obtaining first text, which is text in a first language;

obtaining second text, which is text in a second language obtained by translating the first text into the second language;

obtaining correct labels, which indicate content of the first text;

inputting the first text and the second text to an identification model common to the first and second languages; and

updating the common identification model such that labels identified by the common identification model from the first text and the second text match the correct labels.