Electronic apparatus and method for controlling thereof

Patent metadata

Field	Value
Publication number	US-12288030-B2
Application number	US-202217739827-A
Country	US
Kind code	B2
Filing date	May 9, 2022
Priority date	Feb 18, 2021
Publication date	Apr 29, 2025
Grant date	Apr 29, 2025

Abstract

An electronic apparatus and a control method thereof are provided. The control method includes acquiring a plurality of words, acquire a plurality of word combinations by combining the plurality of words, and acquiring first scores indicating the sentence completeness of each of the plurality of word combinations by inputting each of the plurality of word combinations to a first a language model for identifying a sentence completeness, based on all the first scores with respect to each of the plurality of word combinations being identified to be less than a predetermined value, acquire second scores for each of the plurality of word combinations by inputting each of the plurality of word combinations to a sentence separation model for identifying an accuracy of a word combination, identifying one word combination among the plurality of word combinations based on the second scores, and providing the identified one word combination to a translation model which performs translation.

First claim

What is claimed is:

1. A method for controlling an electronic apparatus comprising:
acquiring a plurality of words based on an image including a plurality of words;
acquiring a plurality of word combinations by combining the plurality of words, by inputting at least one word group constituting a first word combination among the plurality of word combinations to a first language model for identifying a sentence completeness, respectively, acquiring scores corresponding to each of at least one word group, and acquiring first scores indicating a sentence completeness of each of the plurality of word combinations based on the scores corresponding to each of the at least one word group;
based on all the first scores with respect to each of the plurality of word combinations being identified to be less than a predetermined value, by inputting the at least one word group constituting the first word combination to a sentence separation model for identifying an accuracy of a word combination, respectively, acquiring scores corresponding to each of the at least one word group, and acquiring second scores for each of the plurality of word combinations based on the scores corresponding to each of the at least one word group;
identifying one word combination among the plurality of word combinations based on the second scores; and
providing the identified one word combination to a translation model which performs translation.

2. The method of claim 1 , wherein the identifying the one word combination further comprises:
based on a first word combination in which the first scores are equal to or greater than a predetermined value being identified to be existed among the plurality of word combinations, identifying one word combination among the plurality of word combinations based on a first score.

3. The method of claim 1 , wherein the acquiring the plurality of words comprises:
acquiring the image including a plurality of words; and
acquire the plurality of words included in the image by performing an image recognition for the image.

4. The method of claim 1 , wherein the acquiring a first score comprises:
acquire a sentence score indicating a sentence completeness of all the plurality of words by inputting the plurality of words to a second language model; and
based on the sentence score being identified to be equal to or greater than the predetermined value, acquire a plurality of word combinations by combining the plurality of words, and acquire the first score by inputting the plurality of word combinations, respectively, to the first language model.

5. The method of claim 4 , further comprising:
based on the sentence score being identified to be less than the predetermined value, acquire a word combination obtained by separating the plurality of words into sentence units by inputting the plurality of words to the sentence separation model; and
providing the word combination to the translation model.

6. The method of claim 1 , wherein the acquiring a first score comprises:
based on scores corresponding to each of the at least one word group, acquiring a first score corresponding to the first word combination.

7. The method of claim 6 , wherein the acquiring a second score comprises:
based on the scores corresponding to each of the at least one word group, acquiring the second score corresponding to the first word combination.

8. The method of claim 1 ,
wherein the acquiring a second score comprises:
based on the first scores with respect to each of the plurality of word combinations being identified to be less than the predetermined value, identifying at least one word combination among the plurality of word combinations based on the first scores, and
acquire second scores with respect to each of the at least one word combination by inputting each of the at least one word combination to the sentence separation model, and

wherein the identifying the one word combination comprises:
based on the second scores, identifying at least one word combination among the at least one word combination.

9. The method of claim 1 , wherein the sentence separation model is configured to be learned based on target data including a plurality of sentences and learning data in which sentence separation is removed from the target data.

10. The method of claim 9 , wherein the sentence separation model is configured to be learned by using a pre-learned neural network model that separates spacing between words.

11. An electronic apparatus comprising:
a memory configured to store at least one instruction; and
a processor configured to execute at least one instruction stored in the memory to control the electronic apparatus,
wherein the processor is configured to:
acquire a plurality of words based on an image including a plurality of words,
acquire a plurality of word combinations by combining the plurality of words, by inputting at least one word group constituting a first word combination among the plurality of word combinations to a first language model for identifying a sentence completeness, respectively, acquiring scores corresponding to each of at least one word group,
acquire first scores indicating sentence completeness of each of the plurality of word combinations based on the scores corresponding to each of the at least one word group,
based on all the first scores with respect to each of the plurality of word combinations being identified to be less than a predetermined value, by inputting the at least one word group constituting the first word combination to a sentence separation model for identifying an accuracy of a word combination, respectively, acquiring scores corresponding to each of the at least one word group, and acquire second scores for each of the plurality of word combinations based on the scores corresponding to each of the at least one word group,
identify one word combination among the plurality of word combinations based on the second scores, and
provide the identified one word combination to a translation model which performs translation.

12. The apparatus of claim 11 , wherein the processor is configured to, based on a first word combination in which the first scores are equal to or greater than a predetermined value being identified to be existed among the plurality of word combinations, identify one word combination among the plurality of word combinations based on a first score.

13. The apparatus of claim 11 , wherein the processor is configured to:
acquire the image including a plurality of words; and
acquire the plurality of words included in the image by performing an image recognition for the image.

14. The apparatus of claim 11 , wherein the processor is configured to:
acquire a sentence score indicating a sentence completeness of all the plurality of words by inputting the plurality of words to a second language model;
based on the sentence score being identified to be equal to or greater than the predetermined value, acquire a plurality of word combinations by combining the plurality of words; and
acquire a first score by inputting the plurality of word combinations, respectively, to the first language model.

15. The apparatus of claim 14 , wherein the processor is configured to:
based on the sentence score being identified to be less than the predetermined value, acquire a word combination obtained by separating the plurality of words into sentence units by inputting the plurality of words to the sentence separation model; and
provide the word combination to the translation model.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Kim Jiwan
• Kim Jonghyun
• Park Soyoon
• Lee Indong
• Lim Sungjun

Classifications

• 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

• G06V30/148

  Segmentation of character regions · CPC title

• G06V30/1916

  Validation; Performance evaluation · CPC title

• G06V10/82

  using neural networks · CPC title

• G06F40/289 Primary

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