JP7122835B2 - Machine translation device, translation trained model and judgment trained model - Google Patents

Description

The present invention relates to a machine translation device that performs machine translation, a translation-learned model, and a determination-learned model.

In recent years, attention has been focused on neural machine translation using neural networks in machine translation. A translation system based on machine learning using a neural network has been developed because it enables more natural translation than conventional statistical-based machine translation. For example, Patent Document 1 below discloses a machine translation device using a neural network.

JP-A-06-332935

However, in machine translation using a neural network or the like, there is a problem that as the number of words to be translated increases, for example, the size of the neural network increases, resulting in an increase in translation processing time.

Therefore, the present invention has been made in view of such problems, and aims to provide a machine translation device, a translation-learned model, and a determination-learned model that can reduce translation processing time in machine translation.

In order to solve the above problems, the machine translation device according to one aspect of the present invention is a translation unit for each category, and each translation unit uses a translation trained model obtained by machine learning using the corpus of the category to which it belongs. A translation unit that translates sentences in the category using a translation unit, and the category of the input sentence is judged using a judgment-learned model obtained by machine learning based on the corpus used in the machine learning of each translation unit. and a control unit that translates the input sentence using the translation unit of the category determined by the determination unit and outputs the translation result.

According to such a machine translation device, sentences are translated using a translation-learned model of the translation unit for each category. In this case, for example, the vocabulary to be translated handled by the single translation unit can be limited to the vocabulary used in the category to which it belongs, so the number of vocabulary handled by the translation trained model can be reduced. As a result, the translation processing time of a single translation unit can be reduced, and the translation processing time of the entire machine translation apparatus can be reduced. Also, the judgment-learned model is a learned model obtained by machine learning based on the corpus used in the machine learning of the translation-learned model of the translation unit. Since a category is determined using such a determination-learned model, for example, the category of sentences that the translation-learned model is good at can be determined more accurately. As a result, the translation is performed by the translation unit of the more accurate category, so that the translation accuracy can be improved.

According to the present invention, translation processing time in machine translation can be reduced.

1 is a functional block diagram of a machine translation device 1 according to an embodiment of the present invention; FIG. 4 is a diagram showing a learning example of the translation unit 10; FIG. 4 is a diagram showing a learning example of a determination unit 11; FIG. 4 is a conceptual diagram showing an example of determination processing by a determination unit 11; FIG. 4 is a flowchart showing an example of determination processing by a determination unit 11; 10 is a graph showing an example of the relationship between the number of vocabularies and translation processing time; 1 is a hardware configuration diagram of a machine translation device 1 according to an embodiment of the present invention; FIG.

Hereinafter, embodiments of the apparatus will be described in detail along with the drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and overlapping descriptions are omitted. Moreover, the embodiments in the following description are specific examples of the present invention, and the present invention is not limited to these embodiments unless otherwise specified.

FIG. 1 is a functional block diagram of the machine translation device 1. As shown in FIG. As shown in FIG. 1, the machine translation device 1 includes a translation section 10A, a translation section 10B, a translation section 10C, a determination section 11 and a control section 12. FIG. The translation unit 10A, the translation unit 10B, and the translation unit 10C are collectively referred to as the translation unit 10 as appropriate.

The machine translation device 1 is a computer device that translates a text (input text) input by a user or the like and outputs the translation result to the user or the like. Translation is to replace a sentence expressed in a certain language (first language) with another language (second language). One example is translation from Japanese text to English text, but it is not limited to this. A sentence is composed of one or more sentences. A sentence is a series of cohesive words written in letters or the like. The translation result is information based on the translation result, and is, for example, a translated sentence (translated sentence), but is not limited to this.

Each functional block of the machine translation apparatus 1 shown in FIG. 1 will be described below.

The translation unit 10 is a translation unit 10 for each category, and each translation unit 10 uses a trained model (translation trained model) obtained by machine learning using the corpus of the category to which it belongs to translate sentences of the category. translate. As shown in FIG. 1, the machine translation apparatus 1 includes a category A translation unit 10A (the translation unit 10A belongs to category A), a category B translation unit 10B (the translation unit 10B belongs to category B), and a category C translation unit. It includes a section 10C (the translation section 10C belongs to category C). Note that the number of categories is not limited to three, and may be any number of one or more.

A category is a taxonomic division of text, a text type, a text genre, a text area, a text field, or a text domain. A category corresponds to a usage scene, and includes, for example, a category that is strong against spoken language often used in hospitals, a category that is strong against written language often used in manuals in the field of IT (Information Technology), and the like. In this embodiment, it is assumed that category A is a medical category, category B is a travel category, and category C is a shopping category.

A corpus is part or all of data (database) that structures natural language sentences and accumulates them on a large scale. The corpus in this embodiment is assumed to be a corpus (parallel translation corpus, parallel translation data) in which a first language (Japanese) is associated with a second language (English) which is a translation of the first language. The process of obtaining a trained model by machine learning using a corpus and the process of translating sentences using the trained model are processes in accordance with conventional techniques.

A trained model is a combination of a computer program and parameters. A trained model is a combination of a neural network structure and a parameter (weighting coefficient) that is the strength of connection between neurons of the neural network. Also, a trained model is a command to a computer that is combined to obtain a result (perform a predetermined process), that is, a computer program that causes the computer to function.

FIG. 2 is a diagram showing a learning example of the translation unit 10. As shown in FIG. For example, the translation unit 10A belonging to category A uses the category A dedicated corpus and general-purpose corpus as learning data used when performing machine learning. The category A dedicated corpus is a corpus specialized for category A, for example, a corpus related to sentences often used in the medical field. A generic corpus is a corpus of generic phrases. The general-purpose corpus consists of sentences or words that do not include special technical terms or terms that are used only in special categories, in order to reduce the number of vocabularies. The learning of the translation units 10B and 10C is similar to the learning of the translation unit 10A. That is, the translation unit 10B belonging to category B uses a corpus dedicated to category B (for example, a corpus related to sentences often used in the field of travel) and a general-purpose corpus as learning data, and the translation unit 10C belonging to category C uses category A C-specific corpus (for example, a corpus of sentences often used in the shopping field) and a general-purpose corpus are used.

As described above, the translation units 10 belonging to each category are trained using a corpus obtained by adding a general-purpose corpus to the dedicated corpus of the category. Note that the dedicated corpus may be exclusive, in which case the same sentences will not be included in other corpora. When performing machine learning, a dedicated corpus alone cannot handle general-purpose phrases, so some or all of the translation units 10 use a common general-purpose corpus as described above. Also, one or more of the translation units 10A, 10B, and 10C may use only a dedicated corpus as learning data without using a general-purpose corpus.

The translation trained model may be a neural network-based trained model. Also, the translation trained model may be a trained model based on a recurrent neural network. Also, the translation-trained model is not limited to a neural network, and may be a trained model based on information processing that can be machine-learned.

The translation-learned model is a translation unit 10 for each category, and each translation unit 10 translates sentences of the category to which it belongs. Translation learning used when the translation unit 10 of the machine translation device 1, which includes a control unit 12 that translates the input sentence using the translation unit 10 of the category determined by the unit 11 and outputs the translation result, translates the sentence. It is a trained model, and is composed of a neural network in which weighting coefficients have been learned using a corpus (dedicated corpus) of the category to which the translation unit 10 using the translation trained model belongs. The category may be determined using a model that has undergone determination training obtained by machine learning based on the corpus (dedicated corpus) used in the learning of .

A translation-trained model is a translation-trained model for functioning a computer to translate an input sentence of a predetermined category and output a translation result, using a category corpus (dedicated corpus) The category may be the category of the input text determined using a trained model obtained by machine learning based on the corpus (dedicated corpus). .

The determination unit 11 puts the category of the input text (from the control unit 12, which will be described later) into a corpus (dedicated corpus) used in machine learning of each translation unit 10 (translation unit 10A, translation unit 10B, and translation unit 10C). A decision-learned model obtained by machine learning based on is used for decision making. The process of obtaining a trained model by machine learning based on a corpus and the process of determining the category of sentences using the trained model are processes in accordance with conventional techniques.

FIG. 3 is a diagram showing a learning example of the determination unit 11. As shown in FIG. The determination unit 11 uses, as learning data, a category A dedicated corpus used in machine learning of the translation trained model of the translation unit 10A, a category B dedicated corpus used in machine learning of the translation trained model of the translation unit 10B, and a translation unit. A corpus dedicated to category C used in machine learning of 10C translation-trained models is used. More specifically, the determination unit 11 sets the first language of the category A dedicated corpus and the information (label) indicating the category A, the first language of the category B dedicated corpus and the information indicating the category B as learning data. (label), and a set of the first language of the category C dedicated corpus and the information (label) indicating the category C are used. In the learning of the determination unit 11, learning is performed so that sentences in the first language of the dedicated corpus of a predetermined category are input, and output (correct data) is used as the label of the category.

The learning of the determination unit 11 will be described with a specific example. The determination unit 11 first labels the medical corpus (category A), the travel corpus (category B), and the shopping corpus (category C). As an example, let the sentence in the medical corpus be 1, the sentence in the travel corpus be 2, and the sentence in the shopping corpus be 3. Then, the sentences of each corpus are input to the decision-learned model (recurrent neural network, etc.) of the judgment unit 11, and the output vector whose label dimension of the input sentence is "1" and the other is "0" is learned as correct data. Let In this case, since there are three kinds of categories, the output vector becomes a three-dimensional output vector. In the shopping corpus, "(0, 0, 1)" is correct data. By learning in this way, it becomes possible to determine (identify) an appropriate category for translating an input sentence.

The decision trained model may be a neural network based trained model. Also, the judgment trained model may be a trained model based on a recurrent neural network. Further, the judgment-learned model is not limited to a neural network, and may be a learned model based on machine-learnable information processing.

The decision-learned model is the translation unit 10 for each category, and each translation unit 10 uses the translation-learned model obtained by machine learning using the corpus of the category to which it belongs (dedicated corpus) to translate the sentences of the category. Control for translating an input text using a translation unit 10 for translating, a determination unit 11 for determining the category of the input text, and a translation unit 10 for the category determined by the determination unit 11, and for outputting the translation result. A decision-learned model used when the decision unit 11 of the machine translation device 1 including the unit 12 decides the category, and a weighting factor based on the corpus (dedicated corpus) used in the machine learning of each translation unit 10 may be configured by a neural network in which is learned.

A decision-learned model is a decision-learned model for making a computer function to determine the category of an input sentence, and is a neural network in which weighting coefficients have been learned based on a predetermined corpus (dedicated corpus). The input sentence may be translated using a translation-learned model obtained by machine learning using the corpus (dedicated corpus) of the category determined based on the determination-learned model.

FIG. 4 is a conceptual diagram showing an example of determination processing by the determination unit 11. As shown in FIG. FIG. 5 is a flowchart showing an example of determination processing by the determination unit 11. As shown in FIG. An example of determination processing by the determination unit 11 will be described below based on the flowchart of FIG. 5 with reference to FIG. 4 . First, an input sentence is input from the control section 12 (step S1). Next, the input sentence input in S1 is morphologically analyzed (divided into words) to generate a word string (step S2). Next, words are extracted (in order from the front) from the word string generated in S2 (step S3). Next, the words are converted into word IDs, and input (sequentially) to the input layer of the neural network (model trained for judgment) (step S4). Next, it is determined whether or not any word remains in the word string (step S5). If it is determined in S5 that words remain (S5: YES), the process returns to S3, and the processing of S3 to S5 is repeated for the remaining words. On the other hand, if it is determined in S5 that there are no remaining words (S5: NO), that is, if all the word strings of the input sentence have been input, "<EOS>" (a symbol indicating the end of the sentence) is input to the neural network. Input to the layer (step S6). Next, an output vector whose dimension number is the number of categories to be classified is calculated from the vector of the intermediate layer of the neural network, and is used as the output layer. The likelihood of this output vector is calculated by Softmax processing or the like, and the category corresponding to the dimension with the highest likelihood is taken as the judgment result (identified as the optimum category) (step S7).

In the following, supplementary explanations regarding the above-described FIGS. 4 and 5 will be provided. The neural network described above may be a recurrent neural network. The input layer of the neural network is set in advance to have a predetermined number of dimensions. For example, if the vocabulary of the input sentence is 50,000 words, 500 dimensions are set. The output layer of the neural network described above is set to a vector whose dimension is the number of categories. The intermediate layer vectors described above are used to make decisions (identification) at non-linear boundaries. The vector of the hidden layer is also set to have a predetermined number of dimensions. For example, set the hidden layer vectors to the same number of dimensions as the input layer. The output layer vector can be calculated by linearly mapping from the intermediate layer vector to the output layer vector. Although the highest value of the vector of the output layer may be used, the vector value can be normalized and converted into a probability distribution value by performing Softmax processing. This makes it possible to use a vector of "1" only for the applicable category and "0" for the other categories as correct data used during learning.

In S4, the word is converted into a vector with a predetermined number of dimensions and input to the input layer. In S4, word ID is an ID assigned to each word, and is assigned in order from "1" in descending order of frequency. For example, an ID of "1" to "30000" is assigned to each word, and 30,000 or more types of infrequent words are discarded. When this word ID is input to the input layer, the ID is compressed (converted) into, for example, a 300-dimensional vector. The smaller the type of word ID is, the smaller the size of this vector can be, and the amount of calculation can be reduced. It should be noted that compression of 300,000 words to 300 dimensions is reversible compression. Compression uses conventional techniques such as word2vec. Compression basically constructs vectors such that the Euclidean distance between the vectors of each word is as large as possible.

Returning to the description of each functional block shown in FIG. 1, the control unit 12 inputs an input text from a user or another device or the like via a network or the like. Next, the control unit 12 outputs the input text to the determination unit 11, and translates the input text using the translation unit 10 of the category determined by the determination unit 11 according to the output. Next, the control unit 12 outputs (displays) the translation result to a user or another device or the like via a network or the like.

Next, the effects of the machine translation device 1 configured as in this embodiment will be described.

According to the machine translation device 1 of the present embodiment, sentences are translated using translation-learned models of the translation unit 10 for each category. In this case, for example, the vocabulary to be translated handled by the standalone translation unit 10 can be limited to the vocabulary used in the category to which it belongs, so the number of vocabulary handled by the translation trained model can be reduced. As a result, the translation processing time of the single translation unit 10 can be reduced, and the translation processing time of the entire machine translation apparatus 1 can be reduced. Also, the judgment-learned model is a learned model obtained by machine learning based on the corpus used in the machine learning of the translation-learned model of the translation unit 10 . Since the category is determined using such a determination-learned model, for example, the category of sentences that the translation-learned model is good at can be determined more accurately. As a result, translation is performed by the translation unit 10 of a more accurate category, so that translation accuracy can be improved. In other words, the categories are distributed to reduce the number of vocabularies in each category, thereby speeding up the processing time without lowering the translation accuracy. In addition, since the corpus used for machine learning of the translation-trained model is also used for machine learning of the decision-trained model, the best performance can be achieved by matching the learning data, and the learning data can be reused to create new It is possible to reduce the labor and cost of preparing learning data.

In addition, when the translation trained model is a trained model based on a recurrent neural network, the optimal machine translation is performed by considering not only the words of the input sentence (word level) but also the word order, phrasing or expression of the input sentence. Therefore, the accuracy of translation can be improved.

In addition, when the judgment trained model is a trained model based on a recurrent neural network, judgment of the optimal category considering not only the words of the input sentence (word level) but also the word order, phrasing or expression of the input sentence It can be performed. As a result, translation is performed by the translation unit 10 of a more accurate category, so that translation accuracy can be improved.

The machine translation apparatus 1 of the present embodiment distributes the vocabulary to be handled to a plurality of translation units 10 to shorten the translation processing time and increase the number of vocabulary that can be handled. Specifically, the machine translation apparatus 1 prepares a plurality of translation units 10 configured by, for example, a neural network, and assigns each of them a good category. By preparing the translation unit 10 for each category, the number of vocabulary handled by the single translation unit 10 can be limited to the vocabulary used in the category, so the number of vocabulary can be reduced. Therefore, the translation processing time of the translation unit 10 can be reduced. However, having the user select which category of translation section 10 should translate the input text may impair convenience. Therefore, the machine translation apparatus 1 includes a determination unit 11 that determines (identifies) a category in which an input sentence is best translated. Then, the input text is sent to the translating unit 10 of the judged optimum category, and the translation result is output as the translated text (output text).

In addition, as a method for learning the judgment-learned model of the judgment unit 11, a special corpus used when learning the translation unit 10 of each category is labeled, and sentences of the corpus for the category are input and output (correct data ) is learned to be the label for that category. This is because sentences close to the sentences that the translation unit 10 at the later stage is good at are determined to be in that category. In addition, by using, for example, a recurrent neural network similar to that used in the translation unit 10 in the latter stage as a decision-learned model, it is possible to perform optimal category classification in consideration of not only the words of the input sentence but also the word order and expressions.

With the machine translation device 1 of this embodiment, the number of vocabulary words can be distributed to a plurality of translation units 10, and the translation processing can be speeded up. In addition, since the determination unit 11 allows the user to input sentences into the machine translation apparatus 1 without being conscious of which category to translate, the machine translation apparatus 1 can handle a huge vocabulary from the user's point of view. Therefore, the machine translation device 1 that can handle a huge vocabulary can be operated in real time by a CPU (Central Processing Unit) without using a GPU (Graphics Processing Unit), and it becomes possible to reduce the computer cost. . FIG. 6 is a graph showing an example of the relationship between the number of vocabularies and the translation processing time. As the graph shows, in the conventional machine translation, as the number of vocabulary increases, the translation processing time increases in square proportion. On the other hand, the processing time of the machine translation apparatus 1 increases linearly as the number of vocabulary increases, but it is shorter than the conventional technology. Therefore, if a plurality of translation units 10 with a small number of vocabulary are prepared, and the determination unit 11 sorts the translation units 10 into appropriate categories, the processing can be performed at high speed. Moreover, by making the translation unit 10 specialize in a dedicated category, there is an effect that the expressions peculiar to the category can be well translated, and there is an advantage in terms of accuracy improvement.

Conventional techniques using neural networks have the problem that as the number of vocabularies increases, the size of the input vector and the size of the output vector increase, resulting in longer learning and translation calculations. This is because an ID is assigned to each piece of huge vocabulary data, and the input vector and output vector whose dimension is the number of IDs are used for neural network learning, so the size of the vector increases, and the size of the neural network also increases. Therefore, the calculation time becomes long. As the number of vocabularies to be handled increases, the number of vocabularies increases, and at present, the number of vocabularies of about 50,000 words can be translated in real time by the CPU. In order to handle a larger vocabulary, it is necessary to use high-performance GPUs in parallel, which increases the computer cost.

The machine translation apparatus 1 of this embodiment relates to distributed processing and accuracy improvement of a machine translation system using a neural network. The machine translation device 1 prepares a plurality of translation units 10, makes each translation unit 10 specialize in a specific category, and reduces the number of vocabularies handled by each translation unit 10, thereby reducing the calculation time. Further, by sorting the input sentences into the optimum category by the determination unit 11, it is possible to realize a machine translation system that can handle sentences of various categories from the input side.

A machine translation device 1A, which is a modification of the machine translation device 1, includes a domain classification unit that classifies an input sentence into a predetermined type of domain, and a machine translation unit for each of the domains. A dedicated corpus is used for learning of the domain classification unit, and the domain classification unit identifies the domain most suitable for the input sentence, translates the input sentence by the machine translation unit of the identified domain, and outputs the translation result. Output. The domain classification means may be a discriminator composed of a recurrent neural network. A corpus for each domain that trains the machine translation unit of each domain may be used for learning of the recurrent neural network that constitutes the domain classification unit. The machine translation apparatus 1A comprises a morphological analysis unit that divides an input sentence into words, a means for converting words into vectors in the recurrent neural network of the domain classification unit, and inputs the vectors to the recurrent neural network. It may be an input to a layer. The machine translation apparatus 1A sequentially inputs the words of the input sentence to the recurrent neural network, generates vectors of the output layer from the intermediate layer at the time when the input is completed, and calculates the likelihood of each domain from the vector value of the output layer. can be calculated.

It should be noted that the block diagrams used in the description of the above embodiments show blocks in units of functions. These functional blocks (components) are implemented by any combination of hardware and/or software. Further, means for realizing each functional block is not particularly limited. That is, each functional block may be implemented by one device physically and/or logically coupled, or may be implemented by two or more physically and/or logically separated devices directly and/or indirectly. These multiple devices may be physically connected (eg, wired and/or wirelessly).

For example, the machine translation device 1 according to one embodiment of the present invention may function as a computer that performs the discrimination method and machine translation method of the present invention. FIG. 7 is a diagram showing an example of the hardware configuration of the machine translation device 1 according to one embodiment of the present invention. The machine translation device 1 described above may be physically configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like.

Note that in the following description, the term "apparatus" can be read as a circuit, device, unit, or the like. The hardware configuration of the machine translation device 1 may be configured to include one or more of each device shown in the figure, or may be configured without some of the devices.

Each function in the machine translation device 1 is performed by loading predetermined software (program) on hardware such as the processor 1001 and the memory 1002, the processor 1001 performs calculation, communication by the communication device 1004, memory 1002 and storage It is realized by controlling reading and/or writing of data in 1003 .

The processor 1001, for example, operates an operating system to control the entire computer. The processor 1001 may be configured with a central processing unit (CPU) including interfaces with peripheral devices, a control unit, an arithmetic unit, registers, and the like. For example, the translation unit 10 , the determination unit 11 , the control unit 12 and the like described above may be implemented by the processor 1001 .

The processor 1001 also reads programs (program codes), software modules, and data from the storage 1003 and/or the communication device 1004 to the memory 1002, and executes various processes according to them. As the program, a program that causes a computer to execute at least part of the operations described in the above embodiments is used. For example, the translation unit 10, the determination unit 11, and the control unit 12 may be implemented by a control program stored in the memory 1002 and operated by the processor 1001, and other functional blocks may be similarly implemented. Although it has been described that the above-described various processes are executed by one processor 1001, they may be executed by two or more processors 1001 simultaneously or sequentially. Processor 1001 may be implemented with one or more chips. Note that the program may be transmitted from a network via an electric communication line.

The memory 1002 is a computer-readable recording medium, and is composed of at least one of, for example, ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), and RAM (Random Access Memory). may be The memory 1002 may also be called a register, cache, main memory (main storage device), or the like. The memory 1002 can store executable programs (program codes), software modules, etc. for implementing the discrimination method and the machine translation method according to an embodiment of the present invention.

The storage 1003 is a computer-readable recording medium, for example, an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disc, a magneto-optical disc (e.g., a compact disc, a digital versatile disc, a Blu-ray disk), smart card, flash memory (eg, card, stick, key drive), floppy disk, magnetic strip, and/or the like. Storage 1003 may also be called an auxiliary storage device. The storage medium described above may be, for example, a database, server, or other suitable medium including memory 1002 and/or storage 1003 .

The communication device 1004 is hardware (transmitting/receiving device) for communicating between computers via a wired and/or wireless network, and is also called a network device, network controller, network card, communication module, or the like. For example, the control unit 12 and the like may be implemented by the communication device 1004 .

The input device 1005 is an input device (for example, keyboard, mouse, microphone, switch, button, sensor, etc.) that receives input from the outside. The output device 1006 is an output device (for example, display, speaker, LED lamp, etc.) that outputs to the outside. Note that the input device 1005 and the output device 1006 may be integrated (for example, a touch panel).

Devices such as the processor 1001 and the memory 1002 are connected by a bus 1007 for communicating information. The bus 1007 may be composed of a single bus, or may be composed of different buses between devices.

Further, the machine translation device 1 includes hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array). , and part or all of each functional block may be implemented by the hardware. For example, processor 1001 may be implemented with at least one of these hardware.

Notification of information is not limited to the aspects/embodiments described herein and may be done in other ways. For example, notification of information includes physical layer signaling (e.g., DCI (Downlink Control Information), UCI (Uplink Control Information)), higher layer signaling (e.g., RRC (Radio Resource Control) signaling, MAC (Medium Access Control) signaling, It may be implemented by broadcast information (MIB (Master Information Block), SIB (System Information Block)), other signals, or a combination thereof. RRC signaling may also be called an RRC message, and may be, for example, an RRC connection setup message, an RRC connection reconfiguration message, or the like.

Aspects/embodiments described herein support Long Term Evolution (LTE), LTE-A (LTE-Advanced), SUPER 3G, IMT-Advanced, 4G, 5G, Future Radio Access (FRA), W-CDMA (registered trademark), GSM (registered trademark), CDMA2000, UMB (Ultra Mobile Broadband), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, UWB (Ultra-WideBand), It may be applied to systems utilizing Bluetooth®, other suitable systems, and/or advanced next generation systems based thereon.

The procedures, sequences, flow charts, etc. of each aspect/embodiment described herein may be interchanged so long as there is no inconsistency. For example, the methods described herein present elements of the various steps in a sample order, and are not limited to the specific order presented.

Input/output information and the like may be stored in a specific location (for example, memory), or may be managed in a management table. Input/output information and the like may be overwritten, updated, or appended. The output information and the like may be deleted. The entered information and the like may be transmitted to another device.

The determination may be made by a value represented by one bit (0 or 1), by a true/false value (Boolean: true or false), or by numerical comparison (for example, a predetermined value).

Each aspect/embodiment described herein may be used alone, in combination, or switched between implementations. In addition, the notification of predetermined information (for example, notification of “being X”) is not limited to being performed explicitly, but may be performed implicitly (for example, not notifying the predetermined information). good too.

Although the present invention has been described in detail above, it will be apparent to those skilled in the art that the present invention is not limited to the embodiments described herein. The present invention can be implemented with modifications and variations without departing from the spirit and scope of the invention defined by the claims. Accordingly, the descriptions herein are for the purpose of illustration and description, and are not intended to have any limiting meaning with respect to the present invention.

Software, whether referred to as software, firmware, middleware, microcode, hardware description language or otherwise, includes instructions, instruction sets, code, code segments, program code, programs, subprograms, and software modules. , applications, software applications, software packages, routines, subroutines, objects, executables, threads of execution, procedures, functions, and the like.

Software, instructions, etc. may also be sent and received over a transmission medium. For example, the software can be used to access websites, servers, or other When transmitted from a remote source, these wired and/or wireless technologies are included within the definition of transmission media.

Information, signals, etc. described herein may be represented using any of a variety of different technologies. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may refer to voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, light fields or photons, or any of these. may be represented by a combination of

The terms explained in this specification and/or terms necessary for understanding this specification may be replaced with terms having the same or similar meanings.

As used herein, the terms "system" and "network" are used interchangeably.

In addition, the information, parameters, etc. described in this specification may be represented by absolute values, may be represented by relative values from a predetermined value, or may be represented by corresponding other information. . For example, radio resources may be indexed.

The names used for the parameters described above are not limiting in any way. Further, the formulas, etc. using these parameters may differ from those explicitly disclosed herein. Since the various channels (e.g., PUCCH, PDCCH, etc.) and information elements (e.g., TPC, etc.) can be identified by any suitable name, the various names assigned to these various channels and information elements may be is also not limited.

As used herein, the terms "determining" and "determining" may encompass a wide variety of actions. "Judgement", "determining" are, for example, judging, calculating, computing, processing, deriving, investigating, looking up (e.g., table , searching in a database or other data structure), ascertaining as "determining" or "determining". Also, "judgment" and "decision" are used for receiving (e.g., receiving information), transmitting (e.g., transmitting information), input, output, access (accessing) (for example, accessing data in memory) may include deeming that something has been "determined" or "decided". In addition, "judgment" and "decision" are considered to be "judgment" and "decision" by resolving, selecting, choosing, establishing, comparing, etc. can contain. In other words, "judgment" and "decision" may include considering that some action is "judgment" and "decision".

The terms "connected," "coupled," or any variation thereof mean any direct or indirect connection or connection between two or more elements, It can include the presence of one or more intermediate elements between two elements being "connected" or "coupled." Couplings or connections between elements may be physical, logical, or a combination thereof. As used herein, two elements are referred to by the use of one or more wires, cables and/or printed electrical connections and, as some non-limiting and non-exhaustive examples, radio frequency They can be considered to be “connected” or “coupled” to each other through the use of electromagnetic energy, such as electromagnetic energy having wavelengths in the microwave, light (both visible and invisible) regions.

The reference signal may be abbreviated as RS (Reference Signal), or may be referred to as Pilot according to the applicable standard.

As used herein, the phrase "based on" does not mean "based only on," unless expressly specified otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

Any reference to elements using the "first," "second," etc. designations used herein does not generally limit the quantity or order of those elements. These designations may be used herein as a convenient method of distinguishing between two or more elements. Thus, references to first and second elements do not imply that only two elements may be employed therein or that the first element must precede the second element in any way.

The “means” in the configuration of each device described above may be replaced with “unit”, “circuit”, “device”, or the like.

To the extent that "including," "comprising," and variations thereof are used herein or in the claims, these terms, as well as the term "comprising," are inclusive. intended to be Furthermore, the term "or" as used in this specification or the claims is not intended to be an exclusive OR.

Throughout this disclosure, where articles have been added by translation, such as a, an, and the in English, these articles are used in the plural unless the context clearly indicates otherwise. shall include those of

DESCRIPTION OF SYMBOLS 1... Machine translation apparatus, 10... Translation part, 11... Judgment part, 12... Control part, 1001... Processor, 1002... Memory, 1003... Storage, 1004... Communication apparatus, 1005... Input device, 1006... Output device.

Claims (5)

a translation unit for each category, each translation unit translating sentences in the category using a translation-learned model obtained by machine learning using the corpus of the category to which it belongs;
a determination unit that determines the category of an input sentence that has been input using a trained model obtained by machine learning based on the corpus used in the machine learning of each of the translation units;
a control unit that translates the input sentence using the translation unit of the category determined by the determination unit and outputs a translation result;
A machine translation device with 2. The machine translation device according to claim 1, wherein said translation trained model is a trained model based on a recurrent neural network. 3. The machine translation apparatus according to claim 1, wherein said judgment-learned model is a learned model based on a recurrent neural network. a translation unit for each category, each translation unit translating sentences in the category to which it belongs; a judgment unit for judging the category of the inputted input sentence; and the translation of the category judged by the judgment unit A translation learned model used when the translation unit of a machine translation device translates a sentence, comprising a control unit that translates the input sentence using a unit and outputs a translation result,
Consists of a neural network in which weighting coefficients have been learned using the corpus of the category to which the translation unit using the translation trained model belongs,
The determination unit determines a category using a judgment-learned model obtained by machine learning based on the corpus used in learning the translation-learned model,
Translation trained model. A translation unit for each category, each translation unit translating sentences in the category using a translation-learned model obtained by machine learning using the corpus of the category to which it belongs, and an input sentence that is input. and a control unit that translates the input sentence using the translation unit of the category determined by the determination unit and outputs a translation result. A judgment trained model used when judging a category,
Consists of a neural network in which weighting coefficients are learned based on the corpus used in the machine learning of each translation unit,
Decision-trained model.

Images