Machine translation system

Abstract

To improve accuracy of machine translation, obtain a translated document with high accuracy, and faithfully translate a translation target document.SOLUTION: A system includes the steps of: generating a first translated document by translating a translation target document using a neural network; determining a phrase to be corrected among phrases in the translation target document on the basis of an analysis result of the first translated document; revising the translation target document by replacing the phrase to be corrected with a high-frequency word in training data used to train the neural network;.and generating a second translated document by translating the revised translation target document using a neural network.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] One aspect of the present invention is a machine translation method, a machine translation system, a program, and a non-transitory computer. The present invention relates to a computer-readable storage medium.

[0002] Note that one embodiment of the present invention is not limited to the above technical field. Examples of the semiconductor device include a semiconductor device, a display device, a light-emitting device, a power storage device, a memory device, an electronic device, a lighting device, Input devices (e.g., touch sensors, etc.), input / output devices (e.g., touch panels, etc.), These driving methods and manufacturing methods can be given as examples. [Background technology]

[0003] Research and development of machine translation, which uses computers to translate one natural language into another, is underway. Machine translation is widely used. There are rule-based machines that translate based on rules. Translation, statistical machine translation using language models and translation models, etc., and news Examples include neural machine translation, which uses neural networks to translate.

[0004] Statistical machine translation uses various tools and undergoes multiple steps of training. Therefore, it is necessary to prepare training data for each tool. Since training is done with one neural network, the required training data is the translated document. Therefore, the training data is easy to obtain and the This reduces the effort required to create training data. And neural machine translation uses rules. These tend to be more accurate than base machine translation and statistical machine translation. As a result, efforts are underway to put neural machine translation into practical use.

[0005] However, machine translation is still not perfect, and mistranslations occur. A translation method for prevention is disclosed. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2019-20950 A Summary of the Invention [Problem to be solved by the invention]

[0007] An object of one embodiment of the present invention is to improve the accuracy of machine translation. One of the objects of the present invention is to obtain a highly accurate translation document by machine translation. One of the objectives of the present invention is to obtain a translation result that is faithful to the original document using machine translation. .

[0008] An object of one embodiment of the present invention is to provide a highly accurate machine translation system. An object of one embodiment of the present invention is to provide a highly accurate machine translation method.

[0009] One aspect of the present invention is a machine translation system that can faithfully translate a document to be translated. One aspect of the present invention is to provide a method for faithfully translating a document to be translated. It is an object of the present invention to provide a machine translation method capable of

[0010] Note that the description of these problems does not preclude the existence of other problems. However, it is not necessary to solve all of these problems. It is possible to extract other issues from the description in the section. [Means for solving the problem]

[0011] In one aspect of the present invention, a neural network is used to translate a document to be translated, thereby obtaining a first Based on the analysis result of the first translation document, a translation of the first translation document is generated. The correction target phrase is determined from the list, and the correction target phrase is used for training the neural network. The translated text is corrected by replacing it with a high-frequency word from the training data, and the neural network generating a second translated document by translating the corrected to-be-translated document using the network; A machine translation method.

[0012] In the step of determining a phrase to be corrected, a sentence to be corrected is determined from the first translation document; The phrase to be corrected is determined from the phrases contained in the sentence in the translation document corresponding to the sentence to be corrected. It is preferable that

[0013] One aspect of the present invention is to translate a document to be translated using a neural network. By generating a document and analyzing the translated document, a score based on the translation accuracy of the translated document is obtained. The translation process continues until the end condition is met, and the translation is generated based on the corrections. The method repeats the steps of: obtaining a score for the translated document based on the linguistic feature; and correcting the translated document. is generated based on the results of the analysis of the translation document generated before this step. A step of determining a correction target phrase from among the phrases that are to be corrected; By replacing the words with high frequency words in the training data used to train the network, the and modifying the translation document based on the modification. By using the search function, the translation of the document that reflects the latest revisions is possible, and the translation is based on the latest revisions. The method further comprises the step of generating a translated document.

[0014] In the step of determining the correction target phrase, a correction target sentence is determined from the translation document, and the correction target phrase is Determining words to be corrected from among words contained in sentences in the translation document that correspond to the target sentence is preferred.

[0015] The high frequency words are preferably selected from among synonyms and similar words of the phrase to be amended.

[0016] One aspect of the present invention is a text processing apparatus having a function of performing at least one of the above machine translation methods. It is a book search system.

[0017] One aspect of the present invention is a machine translation system having a processing unit, the processing unit comprising: a function of translating a document to be translated using the network to generate a first translated document; Based on the analysis of the translated document, the words to be corrected are selected from the words contained in the original document. The function to determine the target phrases and the correction target phrases are based on the training data used to train the neural network. The function of correcting the translated text by replacing it with a high-frequency word that can be translated by a neural network is also included. A function of translating the corrected translation document using the translation function to generate a second translation document; Yes.

[0018] One aspect of the present invention is a method for performing at least one of the above machine translation methods on a processor. According to one aspect of the present invention, the program is a program having a function of storing the program. A temporary computer-readable storage medium.

[0019] The program can be read by any type of temporary computer-readable storage medium. Examples of temporary computer-readable storage media include electrical signals, optical signals, and Transient computer-readable storage media include electrical signals, optical fibers, and the like. The program can be supplied to the computer via a wired communication path or a wireless communication path.

[0020] In one aspect of the present invention, a neural network is used to translate a document to be translated, thereby obtaining a first generating a translation document of the first translation document; and determining a correction target phrase from among the phrases included in the classification; By replacing the words with high frequency words in the training data used to train the network, the and translating the corrected translation document using a neural network. and generating a second translated document by One aspect of the present invention is a non-transitory computer-readable storage device that stores the program. It is a medium.

[0021] As non-transitory computer-readable storage media, various types of tangible storage media are used. The non-transitory computer-readable storage medium may be, for example, a RAM (Random Access Memory). volatile memory such as ROM (Read Only Access Memory), Non-volatile memory such as hard disk drives (HDDs) are also included. (Hard Disk Drive: HDD) and Solid State Drive (Soli d State Drive (SSD) and other recording media drives, optical magnetic disks, Examples include D-ROMs and CD-Rs. Effect of the Invention

[0022] According to one aspect of the present invention, the accuracy of machine translation can be improved. According to one aspect of the present invention, a translation document with high translation accuracy can be obtained. This makes it possible to obtain translation results that are faithful to the original document.

[0023] According to one aspect of the present invention, a highly accurate machine translation system can be provided. This provides a highly accurate machine translation method.

[0024] According to one aspect of the present invention, there is provided a machine translation system capable of faithfully translating a document to be translated. According to one aspect of the present invention, a translation system that can faithfully translate a document to be translated can be provided. This provides a machine translation method.

[0025] The description of these effects does not preclude the existence of other effects. However, it is not necessary to have all of these effects. However, it is possible to extract effects other than these. [Brief description of the drawings]

[0026] [Figure 1] FIG. 1 is a flow diagram showing an example of a machine translation method. [Diagram 2] FIG. 2 is a schematic diagram showing an example of a machine translation method. [Diagram 3] FIG. 3 is a flow diagram illustrating an example of a machine translation method. [Figure 4] FIG. 4 is a block diagram showing an example of a machine translation system. [Diagram 5] FIG. 5 is a block diagram illustrating an example of a machine translation system. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] The embodiment will be described in detail with reference to the drawings. However, the present invention is not limited to the following description. The present invention is not limited to the above embodiment, and various changes and modifications may be made in the form and details without departing from the spirit and scope of the present invention. It will be easily understood by those skilled in the art that the above-mentioned invention can be obtained by the following embodiments. It should not be construed as being limited to the contents described.

[0028] In the configuration of the invention described below, the same parts or parts having similar functions are referred to as the same parts. The same reference numerals are used in common among different drawings, and the repeated explanations are omitted. When referring to a function, the same hatch pattern may be used and no particular reference number may be given.

[0029] In addition, the position, size, range, etc. of each component shown in the drawings are not necessarily the same as in the actual embodiment, in order to facilitate understanding. Therefore, the disclosed invention may not necessarily represent the position, size, range, etc. Furthermore, the present invention is not limited to the position, size, range, etc. disclosed in the drawings.

[0030] (Embodiment 1) In this embodiment, a machine translation method according to one embodiment of the present invention will be described with reference to FIGS. .

[0031] In a machine translation method according to one aspect of the present invention, when a document to be translated is translated to generate a translated document, Use artificial intelligence (AI). Specifically, In one embodiment of the machine translation method of the present invention, an artificial neural network (ANN) is used. ificial Neural Network, hereafter simply referred to as neural network A neural network is a circuit (hardware) or a program. This is achieved through (software).

[0032] In this specification, the term "neural network" refers to a network that imitates the neural circuitry of living organisms and that learns by learning. This refers to a general model in which the strength of connections between neurons is determined by the neural network, giving the brain problem-solving capabilities. A neural network has an input layer, an intermediate layer (hidden layer), and an output layer.

[0033] In this specification, when discussing neural networks, Determining the connection strengths (also called weight coefficients) between neurons is called "learning." There are cases.

[0034] In this specification, the neural network is constructed using the connection weights obtained by learning. In one aspect of the present invention, the construction of a new theory and the drawing of a new conclusion from it are sometimes called "inferences." In this case, translating a document using a neural network corresponds to inference. It can be said that.

[0035] In this specification, translation using a neural network is referred to as "neural machine translation." It is sometimes called.

[0036] Here, neural machine translation can sometimes result in mistranslations.

[0037] To train a neural network, a large set of documents to be translated and translated is required. Training data (also called a training corpus) is used. If the source document contains words or phrases that are not translated properly, neural machine translation will not be able to translate them properly. There are some things I can't do.

[0038] In addition, in a neural network, the higher the dimensions of the input layer and output layer, the The amount of calculations becomes enormous. Therefore, it is necessary to consider the languages ​​used during learning (languages ​​of documents to be translated, languages ​​of translated documents, etc.) The vocabulary of the language of the text is limited to a certain number. The training data is limited to words that are frequently used in the data (hereinafter referred to as high-frequency words). Words that are not often used in the language (hereinafter referred to as low-frequency words) are specially marked as unknown words. The word replaced as an unknown word in this way is Even in such cases, it is difficult for neural machine translation to translate the content appropriately.

[0039] Furthermore, even if a sentence uses a word selected as a high-frequency word in the training data, the word When a word sequence or combination is unusual, neural machine translation is prone to mistranslations.

[0040] There are also errors that are characteristic of neural machine translation. The translated document may contain errors, such as repeated use of the same word (or phrase) or An error occurs when a word (or phrase) appears that does not correspond to a word (or phrase) in the translation document. - etc. are likely to occur.

[0041] As a method to reduce mistranslations in translated documents, the translated documents are processed by neural networks. One option is to modify the expression to match the trained model and then perform neural machine translation. However, if the original document is modified before neural machine translation, the original document may not be translated. There is a risk of over-correcting the translation, which may result in a translation that is not faithful to the original document. be.

[0042] In the machine translation method according to one aspect of the present invention, a neural network is used to generate a target sentence. A first translated document is generated by translating the translated document. Next, the analysis result of the first translated document is Based on the results, the words to be corrected are selected from the words contained in the translation document. The target phrase is replaced with a high-frequency word in the training data used to train the neural network. Then, the neural network is used to correct the translated text. A second translated document is generated by translating the to-be-translated document.

[0043] In the machine translation method, neural machine translation is performed once, and the translation result is analyzed. Based on the analysis results, the words to be corrected in the translation document are determined. This makes it possible to prevent excessive document correction and generate a translation that is faithful to the original document. In addition, when correcting a phrase in a translation document, the phrase is processed by a neural network. This replaces the words with high frequency words in the training data used to train the neural machine. The accuracy of translation can be improved.

[0044] Alternatively, in a machine translation method according to one aspect of the present invention, a neural network is used to generate a translation text. By translating the document, a translated document is generated, and the translated document is checked for errors. The system obtains a score based on the translation accuracy of the document, and corrects the translated document until it meets the termination criteria. Repeatedly generating a translation document based on the corrections and obtaining a score for the translation document based on the corrections. return.

[0045] A mistranslation in a translation document is not limited to one sentence, but may contain multiple sentences containing mistranslations. If you modify many parts of the document to be translated, it becomes difficult to know which modifications are valid and which are not. It becomes difficult to judge whether a correction is desirable or not. Also, there is not always one correction candidate for the translation document. There may be multiple corrections, each of which may result in a different correction. By creating multiple translation documents, each of which is translated, For example, if the score is greater than the standard value, multiple translated documents may be generated based on the corrections. The translation document may be modified repeatedly until a desired translation result is reached. The plurality of corrected translation documents and the translation documents corresponding to the plurality of translation documents are processed in a predetermined manner. A number of translation documents may be created, and the translation document with the highest score may be output as the translation result.

[0046] <Machine translation method example 1> FIG. 1 shows a flowchart of a machine translation method. As shown in FIG. The machine translation method includes seven steps, Step A1 to Step A7. FIG. 2(B) shows a schematic diagram of step A2, FIG. 2(B) shows schematic diagrams of steps A3, A4, and A5, and FIG. 2(C) shows a schematic diagram of step A6.

[0047] [Step A1: Obtain the translation document SD1] First, obtain the translation document SD1. The translation document SD1 contains multiple sentences even if it is only one. It may be a number.

[0048] For example, the document SD1 to be translated may be, as text data, audio data, or image data, It is supplied from the outside. When audio data or image data is supplied, the Then, create text data.

[0049] The document to be translated is not particularly limited, and may be, for example, a document related to intellectual property. The documents related to the property include, specifically, the specification, claims, and Furthermore, intellectual property documents include patent documents (publications of published patents). These include publications such as patent gazettes, utility model gazettes, design gazettes, and papers. Not limited to publications published in Japan, but also publications published in countries around the world, It can be used as:

[0050] Other documents to be translated include emails, books, newspapers, papers, reports, columns, etc. The translation may include various works including other texts. Medical documents, etc. may also be used.

[0051] There are also no particular limitations on the language of the document to be translated. For example, Japanese, English, Chinese, Korean Able to use documents written in Japanese, etc.

[0052] [Step A2: Neural machine translation is performed on the to-be-translated document SD1 to generate a translated document TD1 ] Next, as shown in FIG. 2(A), the neural network NN of the processing unit 103 is used. Then, the to-be-translated document SD1 is translated to generate a translated document TD1.

[0053] Neural networks (NN) handle variable-length data such as text data. Recurrent neural networks can network (RNN) can be used.

[0054] There is no particular limitation on the model or mechanism used for neural machine translation. For example, Sequence-to-Sequence model, Transformer model, Atten tion mechanism, etc. can be used.

[0055] The language of the translation document is not particularly limited as long as it is different from the language of the document to be translated. For example, Japanese, English, Chinese, Korean, etc. can be used.

[0056] [Step A3: Determine the correction target sentence CTS from the translated document TD1] Next, the translated document TD1 is analyzed to determine sentences that may be mistranslated (sentences to be corrected CTS). In Figure 2(B), the possible mistranslations in the translation document TD1 are used as the correction target sentence CTS. Here is an example where a valid sentence has been determined.

[0057] Step A3 can be considered as an error check of the entire translated document TD1. It is preferable to be able to calculate a score that serves as an evaluation index for the translation accuracy of the translated document TD1.

[0058] In the analysis of the translation document TD1, sentences with the same word (or phrase) repeated, Sentences containing words (or phrases) not contained in document SD1, grammatically incorrect It is preferable to detect at least one of the following: a sentence containing a special symbol, a sentence containing a special symbol, etc. I wish.

[0059] One or more types of processing can be used to detect sentences that may be mistranslated. For example, Various types of natural language processing can be used to detect sentences that may be mistranslated.

[0060] For example, by using a word dictionary prepared in advance, the nouns in the translation document SD1 and the translation document It may also be compared with the nouns contained in TD1.

[0061] In addition, the result of back-translating the translated document TD1 into the language of the translated document SD1 and the result of back-translating the translated document SD1 You can also compare sentences between the two.

[0062] To detect potentially mistranslated sentences, we use the probability of the output of the neural machine translation model. In neural machine translation, it is possible to analyze the value of the The weight coefficients of the intermediate layer are taken into account to output the most probable word in the translation language. If the probability of a word is low, the sentence containing that word can be determined to be a possible mistranslation. do.

[0063] In order to detect sentences that may be mistranslated, the system analyzes past translation results stored on a server, etc. For example, the information on the strength of the word relationship between the original sentence and the translation sentence ( The weight coefficients of the intermediate layer in the neural network (NN) are stored and are different from the past. If the association with the word is strong, it can be determined that there is a possibility of a mistranslation.

[0064] The number of sentences to be corrected may be one or more. If there are multiple sentences to be corrected, For example, after performing steps A4 and A5 for each sentence, step A6 Alternatively, for a sentence, perform steps A4 through A6, Then, go back to step A3 or A4 and repeat the same process for other sentences. You may do so.

[0065] [Step A4: Words included in sentence S1 in the translation document SD1 that correspond to the correction target sentence CTS Select the phrase to be corrected from the list. Next, among the words contained in the translation document SD1, the words ( In Figure 2(B), the translation document SD1 is determined as the correction target sentence CTS. An example in which a corresponding sentence S1 is included and a correction target phrase is selected from among the phrases contained in the corresponding sentence S1 The correction target term may be a word or a phrase.

[0066] Among the words in sentence S1, the words that are predicted to cause mistranslation are, for example, The words used in training the neural network NN (i.e., unknown words with special symbols) Among the words that have not been replaced, there are words that are used relatively infrequently. ,Information on the frequency of use of words in the training data used to train the neural network NN By looking at the least frequently used words, or frames that contain the least frequently used words, The phrases can be determined as the phrases to be corrected.

[0067] The frequency of a word can be calculated, for example, by calculating the TF (Term Frequency) value of the word. In addition, the TF-IDF (Ter m Frequency (Inverse Document Frequency) value may also be used.

[0068] In order to detect words that are predicted to cause mistranslations, we use a neural network (NN). In neural machine translation, the weighting coefficients of the intermediate layer can be analyzed. By referring to this, you can see which words in the translated sentence the output word is closely related to. For example, words that are weakly related to any words in the translation can be determined as correction targets. Cut.

[0069] The method for determining the words and phrases to be corrected is not limited to the above, and various methods can be used. .

[0070] [Step A5: Replace the correction target phrase with a high-frequency word in the training data, and Fix SD1] Next, the correction target phrase is the training data used to train the neural network NN. Correct the translated document SD1 by replacing it with high-frequency words, and create the translated document SD2. In the following, the corrected translation document SD2 shown in Figure 2(B) is a sentence S1 whose correction target phrase is different from the other sentences. The sentence S1' has been replaced with the phrase:

[0071] The target phrases to be corrected are those with high frequency in the training data used to train the neural network NN. By replacing it with a new word, the translation accuracy in neural machine translation can be improved.

[0072] However, as a result of correcting the translated document SD1 to reduce mistranslations, the translated document S It is not desirable to be unable to translate D1 faithfully. Therefore, the words to be corrected are as follows: It is preferable that the target phrase is replaced with a synonym or a similar word. Even if you modify the translated document SD1, the contents of the translated document SD1 will change significantly. Therefore, the translation will be faithful to the original text SD1 before the correction. Furthermore, mistranslations can be reduced.

[0073] Specifically, multiple synonyms and similar words of the target phrase are extracted, and then the neural network is used to select them. Select words that are frequently used in the training data used to train the network NN, and It is preferable to replace the correction target phrase with

[0074] Synonyms and similar words for the words to be corrected can be obtained, for example, by using a dictionary prepared in advance. In addition, based on the similarity or distance of the distributed representation vectors of words, Synonyms and related words of the phrase to be amended may be extracted.

[0075] The high frequency words in the training data used to train the neural network NN are The selection can be based on information about the frequency of use of words in the data.

[0076] The frequency of use of a word can be calculated, for example, by calculating the TF value of the word. The TF-IDF value of the word may be used.

[0077] Please note that there may be one or more candidates for the word (or phrase) to replace the target word or phrase. For example, the most frequently used words in the training data may be used as the replacement word. You can use multiple frequently used words as candidates for replacement words. If there are multiple replacement word candidates, the target phrase is replaced with a different word. In step A6, multiple modified translation documents are created. In this case, the translation of the translation document may be generated as described above in step A3. By using the analysis, it is possible to calculate a score based on the translation accuracy, and it is possible to determine which of multiple translated documents is the most accurate. This is preferable because it makes it easier to judge whether the translation result is satisfactory.

[0078] In addition, when the phrase to be corrected includes multiple words, each word may be replaced with another word. Only some words may be replaced with other words.

[0079] [Step A6: The corrected translation document SD2 is subjected to neural machine translation, and the translated document TD2 is Generate Next, as shown in FIG. 2(C), the neural network NN of the processing unit 103 is used. Then, the corrected to-be-translated document SD2 is translated to generate a translated document TD2.

[0080] [Step A7: Output the translated document TD2] Then, the translated document TD2 is output.

[0081] For example, the translation document TD2 may be externally transmitted as text data, audio data, or image data. When audio data or image data is to be output, the data generated in step A6 is Based on the text data, audio data or image data is created.

[0082] In the machine translation method of the present embodiment, an example in which the process is automatically performed by a machine has been described. However, some human processing may be included. For example, the correction target in step A3 The specification of the sentence and the specification of the word or phrase to be corrected in step A4 may be performed manually.

[0083] In addition, the data used in the machine translation method of this embodiment and the generated data are appropriately For example, the translation document acquired in step A1, The translation document generated in step A2 and step A6, the modified translation document generated in step A5, At least one of the translated documents may be stored in a server, a memory, or the like.

[0084] In addition, when retraining the neural network, the translated document SD1 and the translated document TD1 and either or both of the set of the translated document SD2 and the translated document TD2. It may be added to the training data. This will improve the translation accuracy of neural machine translation. It is possible to achieve this.

[0085] <Machine translation method example 2> FIG. 3 is a flowchart showing a machine translation method according to an embodiment of the present invention. The machine translation method includes nine steps, namely, steps A11 to A19. Detailed description of steps similar to steps A1 to A7 will be omitted.

[0086] [Step A11: Obtain the document to be translated] First, as in step A1, the document to be translated is obtained.

[0087] [Step A12: Neural machine translation is performed on the to-be-translated document to generate a translated document] Next, similarly to step A2, the document to be translated is translated using a neural network. This generates a translated document.

[0088] [Step A13: Check for errors] Next, an error check is performed. In the error check, Alternatively, the translation document generated in step A18) described later is analyzed to determine the translation accuracy of the translation document. The score is calculated based on the result. The score is an evaluation index of the translation accuracy. -When multiple revisions are made and multiple revision proposals (translated documents) are obtained, the score is used to evaluate the translation accuracy. It is possible to determine which translated documents have the highest quality.

[0089] [Step A14: Determine whether the error correction termination condition is met] Next, it is determined whether the error correction end condition is met, and if so, the process proceeds to step A19. If not, proceed to step A15.

[0090] The error correction end condition is not particularly limited. For example, if the score calculated in step A13 is When a threshold value is reached (mistranslations are sufficiently reduced), error correction may be terminated. Alternatively, the error correction may be terminated when the error correction has been performed a predetermined number of times.

[0091] [Step A15: Determine the sentence to be corrected] Next, similar to step A3, sentences that may be mistranslated (sentences to be corrected) are selected from the translation document. The sentence to be corrected is determined using the result of the error check in step A13. You may do so.

[0092] [Step A16: Select the words and phrases included in the sentence in the translation source document that correspond to the sentence to be corrected. Determine the correct target phrase Next, as in step A4, select words and phrases contained in the translation document that are believed to be the cause of the mistranslation. Determine predicted phrases (phrases to be corrected).

[0093] [Step A17: Replace the correction target phrase with a high-frequency word in the training data, and Modify the document. Next, similarly to step A5, the correction target phrase is used for training the neural network. The translated document is corrected by replacing it with high frequency words in the training data.

[0094] [Step A18: Neural machine translation is performed on the corrected translation document to generate a translated document] Next, as in step A6, the corrected translation document is generated using a neural network. After step A18, the process proceeds to step A13.

[0095] [Step A19: Output the translated document] If the error correction end condition is met in step A14, the translated document is output. For example, If the score calculated in step A13 reaches a reference value, the error correction is terminated. It is preferable to output a translation document whose score reaches a certain threshold. When the error correction is completed after performing the above a predetermined number of times, the scores of a predetermined number of translated documents are compared. It is preferable to output the translated document with the highest translation accuracy. The number of translated documents to be output is , may be one or more.

[0096] As described above, in the machine translation method of the present embodiment, neural machine translation is performed once, and then Based on the translation result, the words to be corrected in the translated document are determined. This makes it possible to prevent excessive document correction and generate a translation that is faithful to the original document. In addition, when correcting a phrase in a translation document, the phrase is processed by a neural network. This replaces the words with high frequency words in the training data used to train the neural machine. The accuracy of translation can be improved.

[0097] This embodiment mode can be appropriately combined with other embodiment modes. In the case where a plurality of configuration examples are shown in one embodiment, the configuration examples may be appropriately combined. It is possible to do this.

[0098] (Embodiment 2) In this embodiment, a machine translation system according to one embodiment of the present invention will be described with reference to FIGS. Reveal.

[0099] The machine translation system of this embodiment uses the machine translation method shown in the first embodiment to translate a certain It is possible to translate one natural language into another. It is possible to translate, and the translation accuracy is high.

[0100] <Machine translation system configuration example 1> FIG. 4 shows a block diagram of a machine translation system 100. The block diagram shows components classified by function as independent blocks. However, in reality, it is difficult to completely separate components by function, and one component may be complex. A function may involve multiple components. For example, the processing performed by the processing unit 103 may be executed by different servers depending on the processing. There are times when this happens.

[0101] The machine translation system 100 includes at least a processing unit 103. The system 100 further includes an input unit 101, a transmission path 102, a storage unit 105, a database 106, and a 07, and output unit 109.

[0102] [Input section 101] The input unit 101 is supplied with a document to be translated from outside the machine translation system 100. The document to be translated supplied to 101 is transmitted via transmission line 102 to processing unit 103, storage unit 105, and , or provided to database 107.

[0103] The document to be translated is input as, for example, text data, audio data, or image data. do.

[0104] The method of inputting the document to be translated can be, for example, key input using a keyboard or touch panel. Input from a microphone, voice input, reading from a recording medium, or using a scanner or camera Examples include image input and acquisition via communication.

[0105] The machine translation system 100 has a function of converting voice data into text data. For example, the processing unit 103 may have this function. The system 100 may further include a voice conversion unit having this function.

[0106] The machine translation system 100 preferably has optical character recognition (OCR) capabilities. This makes it possible to recognize characters contained in image data and create text data. For example, the processing unit 103 may have this function. However, it may further include a character recognition unit having this function.

[0107] [Transmission path 102] The transmission path 102 has a function of transmitting various data. Data is transmitted and received between the memory unit 105, the database 107, and the output unit 109 via a transmission line 1. For example, data such as documents to be translated and documents to be translated can be transmitted through a transmission line. The data is transmitted and received via 102.

[0108] [Processing section 103] The processing unit 103 receives information provided from the input unit 101, the storage unit 105, the database 107, etc. The processing unit 103 has a function of performing calculations using the data. , can be supplied to the database 107, the output unit 109, etc.

[0109] The processing unit 103 has a function of performing neural machine translation and generating a translated document. The document to be translated input to the input unit 101 or the document to be translated corrected by the processing unit 103 is It can be translated.

[0110] The processing unit 103 has a function of determining a sentence to be corrected from the translation document. 103 is a word that is selected from the words in the sentence in the translation document that corresponds to the sentence to be corrected. The processing unit 103 has a function of determining the correction target phrase based on the frequency of the correction target phrase in the learning data. It has the function of replacing the translated text with a new word, correcting the translated text, and generating a corrected translated text.

[0111] The processing section 103 may use a transistor having a metal oxide in a channel forming region. Since the off-state current of the transistor is extremely low, the transistor can be preferably used as a memory element. It is used as a switch to hold the charge (data) that has flowed into the capacitance element that functions as a By doing so, data can be retained for a long period of time. 103 for at least one of the register and the cache memory. The processing unit 103 is operated only when necessary, and in other cases, the information of the immediately preceding processing is stored in the storage element. By retracting the processor 103, the processor 103 can be turned off. This will enable fast computing and reduce the power consumption of machine translation systems. do.

[0112] In this specification and the like, a semiconductor device using an oxide semiconductor or a metal oxide for a channel formation region The transistor is called an Oxide Semiconductor transistor, or OST. The channel formation region of an OS transistor often contains a metal oxide. preferable.

[0113] In this specification, the term "metal oxide" refers to a metal oxide in a broad sense. Metal oxides are oxide insulators and oxide conductors (including transparent oxide conductors). , oxide semiconductor (also called OS) For example, when a metal oxide is used for the semiconductor layer of a transistor, the metal Oxides are sometimes called oxide semiconductors. In other words, metal oxides have amplifying and rectifying properties. and switching action, the metal oxide is Metal oxide semiconductor, abbreviated as OS. This can be done.

[0114] The metal oxide of the channel formation region preferably contains indium (In). When the metal oxide in the panel formation region contains indium, the OS transistor The carrier mobility (electron mobility) of the semiconductor is increased. The oxide is preferably an oxide semiconductor containing an element M. The element M is aluminum (Al). , gallium (Ga), or tin (Sn). The elements are boron (B), silicon (Si), titanium (Ti), iron (Fe), Ni, Germanium, Yttrium, Zirconium, Mo Libdenum (Mo), Lanthanum (La), Cerium (Ce), Neodymium (Nd), Hafnium Hf, tantalum (Ta), tungsten (W), etc. However, the element M is In some cases, a combination of the above elements may be used. The element M may be, for example, bonded with oxygen. For example, the bond energy with oxygen is higher than that with indium. In addition, the metal oxide of the channel formation region contains zinc (Zn). It is preferable that the zinc-containing metal oxide is easily crystallized in some cases.

[0115] The metal oxide contained in the channel formation region is not limited to a metal oxide containing indium. The semiconductor layer may be an indium-free material such as zinc tin oxide or gallium tin oxide. , metal oxides containing zinc, metal oxides containing gallium, metal oxides containing tin, etc. It's okay.

[0116] The processing section 103 may also include a transistor containing silicon in a channel forming region. stomach.

[0117] The processing section 103 includes a transistor having an oxide semiconductor in a channel formation region and a It is preferable to use a transistor including silicon in a panel formation region in combination with the transistor.

[0118] The processing unit 103 is, for example, an arithmetic circuit or a central processing unit (CPU: Central Processing Unit). ocessing Unit) etc.

[0119] The processing unit 103 includes a DSP (Digital Signal Processor), a GP U (Graphics Processing Unit) and other microprocessors The microprocessor may be a Field Programmable Gate Array (FPGA). Gate Array), FPAA (Field Programmable A PLD (Programmable Logic Device) The processing unit 103 may be realized by a processor. It performs various data processing and program control by interpreting and executing instructions from various programs. The program that can be executed by the processor can be executed by the processor. The information is stored in at least one of the memory area and the storage unit 105 .

[0120] The processing unit 103 may have a main memory. The main memory may be a volatile memory such as a RAM. The input / output unit has at least one of a volatile memory such as a read-only memory (ROM) and a non-volatile memory such as a read-only memory (ROM).

[0121] As an example of RAM, DRAM (Dynamic Random Access Memory) mory), SRAM (Static Random Access Memory), etc. is used, and a memory space is virtually allocated and used as a working space for the processing unit 103. The operating system and application programs stored in the storage unit 105 , program modules, program data, and lookup tables, etc. These data, programs, and processes loaded into RAM are Each of the program modules is directly accessed and operated by the processing unit 103.

[0122] The ROM contains a BIOS (Basic Input / Output) that does not require rewriting. It can store the ROM, ... ROM, OTPROM (One Time Programmable Read Only Memory), EPROM (Erasable Programmable EPROM is a type of memory that can be read and written by ultraviolet light. UV-EPROM (Ultra-Violet) Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmability) e Read Only Memory, flash memory, etc.

[0123] [Storage section 105] The storage unit 105 has a function of storing the program executed by the processing unit 103. The memory unit 105 stores, for example, the calculation results (translated documents, error check results) generated by the processing unit 103. The results of the translation are displayed on the input section. The input data (such as the document to be translated) may be stored in the input data 101 .

[0124] The storage unit 105 includes at least one of a volatile memory and a non-volatile memory. The unit 105 may include a volatile memory such as a DRAM or an SRAM. The unit 105 is, for example, a resistive random access memory (ReRAM). Memory, also called resistive memory), PRAM (Phase change RAM) andom Access Memory), FeRAM (Ferroelectric Random Access Memory), MRAM (Magnetoresis tive Random Access Memory, also known as magnetoresistive memory), Alternatively, the storage unit 105 may have a non-volatile memory such as a flash memory. Hard Disk Drive (HDD) and solid Recording media drives such as solid state drives (SSDs) It may have a live.

[0125] [Database 107] The database 107 includes at least the learning data used for training the neural network NN. The database 107 also has the function of storing information on the frequency of use of words in the data. For example, the calculation result generated by the processing unit 103 (translated document, error check result, correction candidate A list of complements, a list of words to be corrected, a corrected translation document, etc.) and The storage unit 1 may have a function of storing input data (such as a document to be translated). The input data 105 and the database 107 do not have to be separate from each other. For example, in a machine translation system The system has a storage unit having the functions of both the storage unit 105 and the database 107. It may be possible.

[0126] The memories included in the processing unit 103, the storage unit 105, and the database 107 are This can be considered an example of a non-transitory computer-readable storage medium.

[0127] [Output section 109] The output unit 109 has a function of supplying data to the outside of the machine translation system 100. For example, the calculation result in the processing unit 103 can be supplied to the outside. A translated document corresponding to a document to be translated can be provided to an external party.

[0128] The machine translation system 100 uses text data of a translation document to translate speech data and It may have the capability of generating one or both of the image data.

[0129] <Machine translation system configuration example 2> FIG. 5 shows a block diagram of a machine translation system 150. The machine translation system 150 The system includes a server 151 and a terminal 152 (such as a personal computer).

[0130] The server 151 includes a communication unit 161a, a transmission path 162, a processing unit 163a, and a database 1 67. Although not shown in FIG. 5, the server 151 further includes a storage unit, an input / output unit, etc.

[0043]

[0131] The terminal 152 includes a communication unit 161b, a transmission path 168, a processing unit 163b, a storage unit 165, and an input The terminal 152 further includes an output unit 169. Although not shown in FIG. 5, the terminal 152 further includes a database, etc.

[0132] A user of the machine translation system 150 inputs a document to be translated into the server 151 from a terminal 152. The document to be translated is sent from communication section 161b to communication section 161a.

[0133] The document to be translated received by the communication unit 161a is transmitted to the database 167 via the transmission line 162. Or the document to be translated is stored in a storage unit (not shown). It may also be supplied directly to the processing unit 163a.

[0134] The generation of a translated document, the determination of a sentence to be corrected, the determination of a phrase to be corrected, and The translation and correction of documents to be translated require high processing power. The processing unit 163a has a higher processing capacity than the processing unit 163b of the terminal 152. Therefore, it is preferable that these processes are each performed by the processing unit 163a.

[0135] Then, the processing unit 163a generates a translated document. The translated document is transmitted via the transmission path 162. The translated document is stored in the database 167 or a storage unit (not shown). The processing unit 163a may directly supply the information to the communication unit 161a. The translated document is output from communication unit 161a to terminal 152. It will be sent to 161b.

[0136] [Input / output section 169] Data is supplied to the input / output unit 169 from outside the machine translation system 150. 169 has a function of supplying data to the outside of the machine translation system 150. As in the translation system 100, the input section and the output section may be separate.

[0137] [Transmission path 162 and transmission path 168] The transmission path 162 and the transmission path 168 have a function of transmitting data. Data is transmitted and received between the management unit 163a and the database 167 via a transmission line 162. The communication unit 161b, the processing unit 163b, the storage unit 165, and the input / output unit 16 Data transmission and reception between the devices 9 can be performed via a transmission line 168.

[0138] [Processing unit 163a and processing unit 163b] The processing unit 163a receives data from the communication unit 161a and the database 167. The processing unit 163b has a function of performing calculations using the communication unit 161b, the storage unit 165, It also has a function of performing calculations using data supplied from the input / output unit 169, etc. The processing unit 163a and the processing unit 163b can refer to the description of the processing unit 103. It is preferable that the processing capacity of the processing unit 163b is higher than that of the processing unit 163b.

[0139] [Storage section 165] The storage unit 165 has a function of storing the program executed by the processing unit 163b. The storage unit 165 stores the calculation results generated by the processing unit 163b and the data input to the communication unit 161b. The input / output unit 169 has a function of storing data input thereto, and data input to the input / output unit 169.

[0140] [Database 167] The database 167 includes at least the learning data used to train the neural network NN. The database also has the function of storing information on the frequency of use of words in the data. The processing unit 163a generates a calculation result, and the data input to the communication unit 161a is input to the communication unit 161b. Alternatively, the server 151 may have a function of storing the information separately from the database 167. The storage unit stores the calculation results generated by the processing unit 163a and the a may have a function of storing data inputted thereto.

[0141] [Communication Unit 161a and Communication Unit 161b] Using the communication units 161a and 161b, data is exchanged between the server 151 and the terminal 152. The communication unit 161a and the communication unit 161b can be a hub, a router, or the like. Data can be sent and received using wired or wireless (e.g. For example, radio waves, infrared rays, etc. may be used.

[0142] This embodiment mode can be combined with other embodiment modes as appropriate. [Explanation of symbols]

[0143] CTS Correction Target Sentence S1 sentence TD1 Translation Documents TD2 Translation Documents SD1 Translated document SD2 Translated documents 100 Machine Translation Systems 101 Input section 102 Transmission Line 103 Processing section 105 Storage section 107 Database 109 Output section 150 Machine Translation System 151 Server 152 terminals 161a Communications Department 161b Communications Department 162 Transmission Line 163a Processing section 163b Processing section 165 Storage section 167 Database 168 Transmission Line 169 Input / output section

Claims

[Claim 1] A processing unit is provided. The processing unit includes: A function of generating a first translated document by translating the to-be-translated document using a neural network; determining a sentence from the first translation document that is suspected to be a mistranslation; A function of determining a phrase to be corrected from among phrases contained in the translation document corresponding to the sentence suspected to be mistranslated; A function of correcting the translation document by replacing the correction target phrase with a high frequency word in training data used for training the neural network; and a function of translating the corrected to-be-translated document using the neural network to generate a second translated document, The sentence suspected of being mistranslated is at least one of a sentence in which the same word is repeated, a sentence containing a word not contained in the translated document, a sentence containing a grammatical error, and a sentence containing a special symbol, The correction target phrase is a word that is weakly related to the weight coefficient of an intermediate layer in a neural network, A machine translation system, wherein the high frequency words are selected from among synonyms and similar words of the correction target phrase.

Images