KR20210049478A - Apparatus and method for providing user customized translation - Google Patents

Abstract

The present invention provides a method for providing a user-customized translation. The method for providing a user-customized translation comprises: a step of receiving a sentence to be translated from a user; a step of extracting a multisense word appearing at a high frequency by the user from the inputted sentence based on log data in which the sentence inputted by the user is stored; a step of considering an association with the multisense word and one or more words adjacent to the multisense word to predict a domain in which the user mainly uses the multisense word; a step of checking whether the multisense word is translated to match the meaning used in the domain to correct an error to complete a translation sentence; and a step of storing the completed translation sentence and the inputted sentence in the log data to train a translation model.

Description

Apparatus and method for providing customized translation {APPARATUS AND METHOD FOR PROVIDING USER CUSTOMIZED TRANSLATION}

The present invention relates to a method and apparatus for providing a user-customized translation, and more particularly, to a method and apparatus for providing a user-customized translation in which performance is improved as a user uses it.

Neural Machine Translation is a type of machine translation method, and refers to a technology in which artificial intelligence translates language in sentence units through data learning. Here, in the neural network machine translation method, when a user inputs a sentence into a translation engine, the context is identified and then converted into a vector containing information such as words, phrases, and word order, and the translation is performed.

In addition, the performance of neural network translation has improved significantly compared to the past due to the research and development of major neural network machine translation technologies. Therefore, more natural and accurate translation is possible than statistical machine translation, which is mainly used for conventional machine translation.

However, even though the major technologies of neural network machine translation have improved compared to the past, the problem of multilingual translation is still not solved. In the case of multilingual translation, the meaning of the word is determined by the word vector value containing the relationship with the words and the characteristics of the word, and the word vector value tends to vary depending on the shape appearing in the corpus. There is a problem in determining the meaning of the multilingual word for.

In addition, although the translation performance of neural network machine translation is improved by the corpus, it is difficult to collect the corpus, and the collected corpus also has a problem in that it cannot accurately translate multiwords according to the existing neural network machine translation due to biased embedding information.

An object of the present invention for solving the above problems is to predict a domain in which a user mainly uses multiple words by using accumulated log data.

Another object of the present invention for solving the above-described problems is to accurately translate a multi-word to be translated in accordance with a specific environment.

Another object of the present invention for solving the above problems is to provide an improved translation model based on log data that is accumulated as a user uses it.

A method for providing customized translation for a user according to an embodiment of the present invention for achieving the above object includes the steps of: receiving a sentence to be translated from a user; Extracting polysynthetic words appearing at a high frequency by a user from the input sentences based on log data in which the sentences input by the user are stored; Predicting a domain in which the user mainly uses the multi-term in consideration of the association between the multi-term and one or more words adjacent to the multi-term; Completing a translated sentence by checking whether the multi-word is translated according to the meaning used in the domain and correcting the error; And storing the completed translation sentence and the input sentence in the log data to train a translation model.

According to an embodiment of the present invention, the translation result is additionally stored in log data and the translation model is trained with the added data, thereby eliminating the need to collect external data.

According to an embodiment of the present invention, since log data of a user is used, it has an advantage of efficiently providing customized translation according to a domain mainly used by the user.

1 is a block diagram of an apparatus for providing customized translation according to an embodiment of the present invention.
2 is a detailed block diagram of a high frequency word extraction unit according to an embodiment of the present invention.
3 is a detailed block diagram of a domain analysis unit according to an embodiment of the present invention.
4 is a detailed block diagram of an error correction unit according to an embodiment of the present invention.
5 is a detailed block diagram of a model learning unit according to an embodiment of the present invention.
6 is a flowchart illustrating an operation of a method for providing customized translation according to an embodiment of the present invention.
7 is a block diagram of an apparatus for providing customized translation according to another embodiment of the present invention.

In the present invention, various modifications may be made and various embodiments may be provided, and specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention. In describing each drawing, similar reference numerals have been used for similar elements.

Terms such as first, second, A, and B may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another component. For example, without departing from the scope of the present invention, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. The term "and/or" includes a combination of a plurality of related described items or any of a plurality of related described items.

When a component is referred to as being "connected" or "connected" to another component, it is understood that it may be directly connected or connected to the other component, but other components may exist in the middle. It should be. On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there is no other component in the middle.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present application, terms such as "comprise" or "have" are intended to designate the presence of features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, but one or more other features. It is to be understood that the presence or addition of elements or numbers, steps, actions, components, parts, or combinations thereof does not preclude in advance.

Unless otherwise defined, all terms used herein including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms as defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in the present application. Does not.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of an apparatus for providing customized translation according to an embodiment of the present invention.

Referring to FIG. 1, a user-customized translation providing apparatus 1000 includes a data storage unit 1100, a high frequency word extraction unit 1200, a domain analysis unit 1300, an error correction unit 1400, and a model learning unit 1500. ) Can be included.

Here, the data storage unit 1100 may store a sentence input when a user uses the translation providing device and a sentence output after using the translation providing device as log data.

In addition, the high frequency word extractor 1200 may find and extract a multiword that appears at a high frequency by a user based on log data stored in the data storage unit 1100.

In addition, the domain analysis unit 1300 may predict a domain in which a user mainly uses a specific multi-word by analyzing the polyunction extracted by the high-frequency word extracting unit 1200.

In addition, the error correction unit 1400 may check whether there is a translation error based on the domain predicted by the domain analysis unit 1300, and if there is a translation error, correct the translation to generate a complete translated sentence.

In addition, the model learning unit 1500 may store the completed translation sentence and the sentence input from the user as log data in the data storage unit 1100 and collect the data to train the translation model.

Therefore, if the user-customized translation providing device according to an embodiment of the present invention is used, it is possible to select and translate the correct meaning of multiple words by grasping the situations and environments frequently used by the user using log data that is accumulated as the user uses it. Knowledge-enhanced translation is possible.

In addition, if a user-customized translation providing device is used, a vast amount of log data can be easily collected from a translation model, and words mainly used by the user can be stored as log data. Accordingly, the apparatus for providing customized translation for a user may predict a domain in which a user mainly uses a word by using log data and accurately grasp the meaning of a multilingual word to provide a correct translation result.

In addition, if there is an error in the translation of a multilingual language, the user-customized translation providing device can provide a customized translation model by correcting the error and retraining the translation model by collecting the corrected data, and can generate a model specialized for a specific domain. There is no need, so it is also cost effective.

2 is a detailed block diagram of a high frequency word extraction unit according to an embodiment of the present invention.

Referring to FIG. 2, the high frequency word extraction unit 1200 may include a morpheme analysis module 1210, a stop word removal module 1220, a frequency calculation module 1230, and a word extraction module 1240.

The high frequency word extracting unit 1200 may extract a multiterm word that appears at a high frequency by a user in order to predict an environment in which the user mainly uses the translation providing device. Here, the polyunction may be extracted using log data stored in the data storage unit 1100.

Also, sentences of the stored log data may be separated into words and morpheme units by the morpheme analysis module 1210 of the high frequency word extracting unit 1200. Here, the separated sentence may be removed by the stop word removal module 1220 from a stop word such as an investigation or an article that is not related to the interpretation of a polyunism. Here, the morpheme analyzer used in the morpheme analysis module 1210 may be based on the TTA standard morpheme tag set (TTAK.K0-11.0010/R1) based on the Sejong tag set.

For example, a sentence such as'The black level value is adjusted from 0 to 255 in 8-bit resolution' is displayed by the morpheme analysis module 1210 as'black/NNG color/NNG level/NNG value/NNG silver/JX 0. /SN ~/S0 up to 255/SN/JX 8/SN bit/NNB resolution/NNG/JKB adjustment/NNG/XFV/EF./SF' can be separated.

Also, the frequency calculation module 1230 may calculate the number of times a word appears in a sentence stored in log data.

In addition, the word extracting module 1240 can extract the polysynthetic words that appear at high frequency by the user by arranging the results in descending order based on the results calculated by the frequency calculation module 1239 and comparing them with a pre-built multi-word dictionary. have.

3 is a detailed block diagram of a domain analysis unit according to an embodiment of the present invention.

Referring to FIG. 3, the domain analysis unit 1300 identifies a domain in which a user mainly uses a multi-term based on the association between a multi-term extracted by a user at a high frequency from an input sentence, and a multi-word existing in the same sentence and adjacent words. It is predictable.

For example, in a sentence such as'The black level value is adjusted from 0 to 255 in 8-bit resolution', the word'beat' is used as a polyphonic word, in the music field it is used as'beat', which means the beat, and in the computer field, the computer Since it is used as'bit', which is the smallest unit of the amount of information used, and is used as'beet' meaning sugar beet in the plant field, the exact meaning of'bit' in the sentence can be grasped by adjacent words.

Therefore, the exact meaning of'bit' in the sentence can be grasped from adjacent words such as'black','color','level', and'resolution' existing in the same sentence. In the end, since'bit' can be inferred that it is a word related to color, computer, monitor, etc. in light of adjacent words, the translated word of'bit' can be selected according to the meaning used in fields related to color, computer, monitor, etc. .

In addition, since the word'bit' accumulated in the log data stored in the data storage unit 1100 can be identified as a word belonging to the'computer' domain, the domain in which the user mainly uses the multi-term'bit' is the'computer' field. It is predictable.

4 is a detailed block diagram of an error correction unit according to an embodiment of the present invention.

Referring to FIG. 4, the error correction unit 1400 may appropriately translate a multi-term that has not been translated according to a meaning in which the multi-term is used in the domain predicted by the domain analysis unit 1300. Accordingly, the domain analysis unit 1300 predicts a domain in which the multi-word is mainly used by the user, and inputs the multi-word into the clustering model trained for each domain to check the correct translation word of the multi-word.

Here, in order to check whether the multi-word is translated according to the domain, the vector distance calculation module 1410 may calculate a vector distance of a word vector value between the multi-word and the translated word existing in the cluster model.

In addition, it may be confirmed by the word verification module 1420 whether or not the multi-word is translated according to the domain based on the vector distance between the multi-word and the translated word calculated by the vector distance calculation module 1410.

In addition, the error correction module 1430 may check whether the word verification module 1420 has translated the multi-language correctly for the domain, and if the multi-language is not translated for the domain, correct the translated word so that the multi-language can be translated to the domain. .

5 is a detailed block diagram of a model learning unit according to an embodiment of the present invention.

Referring to FIG. 5, the model learning unit 1500 may include a log data extraction module 1510 and a translation model training module 1520.

Here, the log data extraction module 1510 may store a sentence input from a user in the data storage unit 1100, a translated sentence, and a complete sentence translated by the user-customized translation providing device as log data, and store the stored log data. It may be extracted and transmitted to the translation model learning module 1520 to train a user-customized translation model.

In addition, the translation model training module 1520 may receive log data from the log data extraction module 1510 and train a user-customized translation model.

6 is a flowchart illustrating an operation of a method for providing customized translation according to an embodiment of the present invention.

Referring to FIG. 6, a method of providing customized translation for a user according to an embodiment of the present invention may include receiving a sentence to be translated from a user (S610).

In addition, it may include a step (S620) of extracting a polyunism that appears at a high frequency by the user from the input sentence based on the log data in which the sentence input by the user is stored.

In addition, it may include a step (S630) of predicting a domain in which the user mainly uses the multi-term in consideration of the association between the multi-term and one or more words adjacent to the multi-term.

In addition, it may include a step (S640) of completing a translated sentence by checking whether the multi-word is translated according to the meaning used in the domain and correcting the error.

In addition, it may include a step (S650) of learning a translation model by storing the completed translation sentence and the input sentence in the log data.

7 is a block diagram of an apparatus for providing customized translation according to another embodiment of the present invention.

The apparatus 1000 for providing customized translation according to an embodiment of the present invention is connected to a processor 1010 and a memory 1020 for storing at least one instruction executed through the processor and a result of execution of the instruction, and communicates with a network. It may include a transmitting and receiving device 1030 to perform.

The apparatus 1000 for providing customized translations may further include an input interface device 1040, an output interface device 1050, and a storage device 1060. Each of the constituent elements included in the apparatus 1000 for providing customized translations may be connected by a bus 1070 to communicate with each other.

The processor 1010 may execute a program command stored in at least one of the memory 1020 and the storage device 1060. The processor 1010 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor in which methods according to an embodiment of the present invention are performed. Each of the memory 1020 and the storage device 1060 may be configured with at least one of a volatile storage medium and a nonvolatile storage medium. For example, the memory 1020 may be composed of at least one of read only memory (ROM) and random access memory (RAM).

The storage device 1060 may also store a sentence input from a user, a translated sentence of the input sentence, and a sentence in which the sentence input from the user by the user-customized translation providing device is completely translated, and the sentence stored in the storage device It is possible to store the data of the translation model learned from them.

Here, the at least one command may include: a command to receive a sentence to be translated from a user; A command for extracting a polymorphism that appears at a high frequency by a user from the input sentence based on log data in which the sentence input by the user is stored; An instruction for causing the user to predict a domain in which the polyunism is mainly used in consideration of the association between the multi-term and one or more adjacent words; A command for completing a translated sentence by checking whether the multi-word is translated according to the meaning used in the domain and correcting an error; And a command for learning a translation model by storing the completed translation sentence and the input sentence in the log data.

The operation of the method according to the embodiment of the present invention can be implemented as a computer-readable program or code on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices that store data that can be read by a computer system. In addition, a computer-readable recording medium may be distributed over a network-connected computer system to store and execute a computer-readable program or code in a distributed manner.

Further, the computer-readable recording medium may include a hardware device specially configured to store and execute program commands, such as ROM, RAM, and flash memory. The program instructions may include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

While some aspects of the invention have been described in the context of an apparatus, it may also represent a description according to a corresponding method, where a block or apparatus corresponds to a method step or characteristic of a method step. Similarly, aspects described in the context of a method can also be represented by a corresponding block or item or a feature of a corresponding device. Some or all of the method steps may be performed by (or using) a hardware device such as, for example, a microprocessor, a programmable computer or electronic circuit. In some embodiments, one or more of the most important method steps may be performed by such an apparatus.

In embodiments, a programmable logic device (eg, a field programmable gate array) may be used to perform some or all of the functionality of the methods described herein. In embodiments, the field programmable gate array may work with a microprocessor to perform one of the methods described herein. In general, the methods are preferably performed by some hardware device.

Although the above has been described with reference to preferred embodiments of the present invention, those skilled in the art will be able to variously modify and change the present invention within the scope not departing from the spirit and scope of the present invention described in the following claims. You will understand that you can.

Claims (1)

Receiving a sentence to be translated from a user;
Extracting polysynthetic words appearing at a high frequency by the user from the input sentences based on log data in which the sentences input by the user are stored;
Predicting a domain in which the user mainly uses the multi-term in consideration of the association between the multi-term and one or more words adjacent to the multi-term;
Completing a translated sentence by checking whether the multi-word is translated according to the meaning used in the domain and correcting the error; And
And learning a translation model by storing the completed translation sentence and the input sentence in the log data.

Images