Method and apparatus for processing natural language, method and apparatus for training natural language processing model

Patent metadata

Field	Value
Publication number	US-10540964-B2
Application number	US-201715598966-A
Country	US
Kind code	B2
Filing date	May 18, 2017
Priority date	Nov 16, 2016
Publication date	Jan 21, 2020
Grant date	Jan 21, 2020

Abstract

A natural language processing method and corresponding apparatus are disclosed. The natural language processing method may include converting words in sentence data, recognized through voice recognition, to corresponding word vectors, and converting characters in the sentence data to corresponding character vectors. The natural language processing method also may include generating a sentence vector based on the word vectors and the character vectors, and determining intent information of the sentence data based on the sentence vector.

First claim

What is claimed is:

1. A natural language processing method, comprising:
receiving a voice signal;
extracting features from the received voice signal;
recognizing a phoneme sequence from the extracted features through an acoustic model;
generating sentence data by recognizing words from the phoneme sequence through a language model;
converting words in the sentence data, to corresponding word vectors;
converting characters in the sentence data to corresponding character vectors;
generating a sentence vector based on the word vectors and the character vectors;
determining intent information of the sentence data based on the sentence vector; and
outputting a control signal corresponding to the intent information to perform an operation by an electronic device.

2. The natural language processing method of claim 1 , wherein the generating of the sentence vector comprises:
generating a concatenation vector of each of the words by concatenating a word vector of each of the words and a character vector of each of characters in each of the words; and
generating the sentence vector by concatenating corresponding concatenation vectors of the words.

3. The natural language processing method of claim 1 , wherein the converting of the words to the word vectors comprises:
in response to an unlabeled word being among the words, converting a word corresponding to the unlabeled word to a word vector corresponding to an unknown word vector.

4. The natural language processing method of claim 1 , wherein the determining of the intent information comprises:
calculating a probability of each of pieces of candidate intent information determined from the sentence vector; and
determining the intent information among the pieces of candidate intent information based on the calculated probability of each of the pieces of candidate intent information.

5. The natural language processing method of claim 1 , wherein the determining of the intent information comprises:
determining a target word corresponding to an error word in the sentence data based on a concatenation vector of the error word; and
recognizing the error word as the target word,
wherein the concatenation vector of the error word is generated by concatenating a word vector of the error word and a character vector of each of characters in the error word.

6. The natural language processing method of claim 5 , wherein, in response to the error word being an unlabeled word, the word vector of the error word corresponds to an unknown word vector.

7. The natural language processing method of claim 1 , wherein the intent information is generated by a natural language processing model receiving the sentence vector.

8. The natural language processing method of claim 7 , wherein, in response to the natural language processing model receiving training data including a second word generated by applying noise to a first word, the natural language processing model is trained to recognize the second word as the first word.

9. The natural language processing method of claim 8 , wherein the second word is generated by changing a portion of characters in the first word to other characters, or adding another character to the first word.

10. A natural language processing apparatus comprising:
a controller; and
a memory including at least one instruction executable by the controller,
wherein, in response to the instruction being executed by the controller, the controller is configured to:
receive a voice signal,
extract features from the received voice signal,
recognize a phoneme sequence from the extracted features through an acoustic model,
generate sentence data by recognizing words from the phoneme sequence through a language model,
convert words in the sentence data to corresponding word vectors,
convert characters in the sentence data to corresponding character vectors,
generate a sentence vector based on the word vectors and the character vectors,
determine intent information of the sentence data based on the sentence vector, and
output a control signal corresponding to the intent information to perform an operation by an electronic device.

11. The natural language processing apparatus of claim 10 , wherein the controller is further configured to:
generate a concatenation vector of each of the words by concatenating a word vector of each of the words and a character vector of each of characters in each of the words; and
generate the sentence vector by concatenating corresponding concatenation vectors of the words.

12. The natural language processing apparatus of claim 10 , wherein the controller is further configured to:
in response to an unlabeled word being among the words, convert a word corresponding to the unlabeled word to a word vector corresponding to an unknown word vector.

13. The natural language processing apparatus of claim 10 , wherein the controller is further configured to:
calculate a probability of each of pieces of candidate intent information determined from the sentence vector; and
determine the intent information among the pieces of candidate intent information based on the calculated probability of each of the pieces of candidate intent information.

14. The natural language processing apparatus of claim 10 , wherein the controller is further configured to:
determine a target word corresponding to an error word in the sentence data based on a concatenation vector of the error word; and
recognize the error word as the target word,
wherein the concatenation vector of the error word is generated by concatenating a word vector of the error word and a character vector of each of characters in the error word.

15. The natural language processing apparatus of claim 14 , wherein, in response to the error word being an unlabeled word, the word vector of the error word corresponds to an unknown word vector.

16. The natural language processing apparatus of claim 10 , wherein the intent information is generated by a natural language processing model receiving the sentence vector.

17. The natural language processing apparatus of claim 16 , wherein, in response to the natural language processing model receiving training data including a second word generated by applying noise to a first word, the natural language processing model is trained to recognize the second word as the first word.

18. The natural language processing apparatus of claim 17 , wherein the second word is generated by changing a portion of characters in the first word to other characters, or adding another character to the first word.

19. The natural language processing apparatus of claim 10 , wherein the controller is further configured to:
receive a voice signal;
extract features from the received voice signal;
recognize a phoneme sequence from the extracted features through an acoustic model; and
generate the sentence data by recognizing words from the phoneme sequence through a language model.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Lee Sangho
• Park Hyoungmin

Classifications

• G06F40/35

  Discourse or dialogue representation · CPC title

• G10L15/02

  Feature extraction for speech recognition; Selection of recognition unit · CPC title

• G10L15/20

  Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence) · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G10L15/063

  Training · CPC title