JP6518981B1 - Translation apparatus, control program for translation apparatus, and translation method using translation apparatus - Google Patents

Abstract

An object of the present invention is to provide a translation apparatus for translating a highly confidential document safely while using external computer resources. A translation apparatus is an external computer used by an unspecified number of people, and inputs a sentence described in a first language through the Internet and translates it into a sentence described in a second language Communicate with an external computer. The translation apparatus is provided with a first substitution means for replacing a word part in a set consisting of a word which is a noun and a reference code immediately following it with a character string corresponding to the word, and a line And transmitting means for transmitting the transmitted text to the external computer through the Internet, and receiving means for receiving the text described in the second language after translation from the external computer through the Internet. [Selected figure] Figure 1

Description

  The present invention relates to a translation device, a control program for the translation device, and a translation method using the translation device.

  Research on technologies underlying machine translation (MT: Machine Translation), such as morphological analysis and dependency analysis, began in the 1950s AD. In the 1980's, with the improvement of computer performance, rule-based machine translation technology came to have some practicality.

  In the 1990s, statistical machine translation (STM) methods were developed. This is to statistically process the correspondence between words in different languages and sentence structures, and select words having a high probability of being translated, and translated sentences from translation candidates. In statistical machine translation, the translation model determines the word corresponding to the original word and the word order of the word according to the probability. Furthermore, the language model determines the correct arrangement of translated words according to the probability. In statistical machine translation, a large amount of parallel translation data (training data) is required to build a translation engine (translation model and language model). The translation engine is appropriately tuned by comparing the translated sentence using the translation engine with the correct translation.

  In statistical machine translation, in order to determine the word order correctly, the syntactic analysis of the sentences of the original language is performed, and the word order of the words of the original language is previously translated into the word order of the translation target language using the dependency relation. Techniques for converting are known.

  In the 2010s AD, neural machine translation (NMT: Neural Machine Translation) using deep learning by neural network appeared. It has been rapidly commercialized since around 2015 and is replacing statistical machine translation. Neural machine translation generally translates a sentence directly (end to end) with a neural network without preprocessing (translation of word order, etc.) of the translated text.

  In neural machine translation, words are treated as real values of vectors, and are treated as several-hundred dimensional distributed representations. That is, the vector expresses word sense and syntactic information (Word Embedding). Since the semantic relation between another original word and its translated word can be analogized to a certain extent from the semantic relation between one original word and its translated word, flexible translation is possible.

  In machine translation, the length of the input data (text length) is not constant, and it is necessary to use data input in the past. For this reason, in neural machine translation, a recurrent neural network (RNN) is generally used. When a word (or character) of the original text is sequentially input to the recurrent neural network and a code indicating the end of the sentence (EOS: End of String) is output, the translation of one sentence is completed.

  In neural machine translation, since weighting coefficients of each neuron (node) become translation knowledge, a table for translation (a table for correlating original words and translated words such as a phrase table) is not required. Neural machine translation, like statistical machine translation, does not perform translation by replacing component words of the original text and rearranging, but creates a new translation from the input text so as to match the learned language model. It can be said that it is a thing.

JP 2018-120584

  Compared to rule-based machine translation and statistical machine translation, neural machine translation can be translated flexibly (by Word Embedding etc.) even if the sentence is not learned and the translated sentence is fluent It has the advantage of being

  However, neural machine translation is not a technique for translation by substitution of the original text like rule-based machine translation and statistical machine translation, so it is difficult to exactly translate the information contained in the input sentence without excess or deficiency There is a problem that (translational omission may occur). In addition, duplicate parts of the translation may be output.

  Furthermore, in neural machine translation, it is not possible to use the method of converting the word order of original words, which has been used in statistical machine translation, in advance into the word order of the language to be translated. There is a problem that it is difficult to use data (dictionary data, word order data, etc.).

  Furthermore, in neural machine translation, while flexible translation can be performed by Word Embedding, sometimes a completely different translated sentence is created by selecting a word having a completely different meaning (in particular, a word with a low frequency of occurrence, For example, infrequent proper nouns). In addition, in the case of long sentences or complex sentences, translated sentences having meaningless were sometimes created.

  Furthermore, in general, in neural machine translation, unlike rule-based machine translation or statistical machine translation, it is difficult for a human to understand the process of translation, and it is predicted whether or not an accurate translation will be output. There was a problem that it was difficult to do. As a result, in neural machine translation, mistranslation tends to occur and there is a problem that mistranslations are easily provided as a final result.

  The fluency and flexibility of the generated translation, rather than the accuracy of the content, for a document for everyday conversation or for the transmission of less important information that only needs an overview even if there are a few mistranslations and translation omissions. May be emphasized. Neural machine translation is very useful for such documents.

  On the other hand, mistranslations and missing translations are fatal for documents that require strict accuracy, such as laws and patents. For this reason, even when using neural machine translation, it has been necessary to check whether there is a mistranslation or omission of translation taking considerable time and effort. In addition, due to the lack of checks, fatal mistranslations and missing translation may occur, which may cause property damage.

  In addition, as other tasks related to neural machine translation, there may be a problem in the handling of character decoration and style, such as superscript, subscript, italic, bold and underlined characters. In a sentence, for example, in the case of superscript, a case in which a normal character is enclosed in a tag of <sup> to </ sup> in the HTML indicates that it is a superscript and a character code such as Unicode Superscript may be expressed directly by. If these are mixed and used for machine learning, or if the handling of text decoration and style in the document at the time of translation differs from the handling at the time of learning, an incorrect translation result may be output.

  Superscripts and subscripts are used to describe chemical symbols, chemical formulas, formulas, variables, etc. In neural machine translation, seemingly similar but completely different chemical symbols, chemical formulas, formulas, variables, depending on the effect of Word Embedding May be output as a translation result. Chemical symbols, chemical formulas, formulas, and variables are often found as mistranslations in translation products because it is difficult to find errors at first glance.

  Furthermore, in neural machine translation, there is a problem that the same word in the original text is translated into different plural kinds of words. For example, although the word "address" contained in the original document has the same meaning, it may be translated as "address", translated as "address" or translated as "address" depending on the place of appearance. May be Also, for example, although the word "display" contained in the original document indicates the same thing, it may be translated as "display unit", translated as "display portion", or " It may be translated as "display means". Such notational fluctuations and the output of words having different meanings are not particularly desirable in the translation of legal documents and patent documents that require strictness.

  Furthermore, in neural machine translation, it is necessary to prepare a large number of bilingual sentences for learning. In addition, since the amount of calculation for learning is large, there is a problem in that a large amount of time and computer resources (GPU etc.) for learning are required.

  In order to reduce the cost of neural machine translation, it is also considered to perform neural machine translation using a computer that has not been dedicated to its own company but has already been learned (Internet-connected sites, servers, etc. operated by other companies and others). Be One business operator provides a learned neural machine translation system over the Internet, and a plurality of users (unspecified many users) use the system over the Internet. In addition, use of many unspecified means that the site (server) is open to the public and accessible from various places.

  When such a system is used, the user of the system sends document data to a learned neural machine translation computer operated by another company via the Internet, and the result of machine translation performed by the server is sent via the Internet. Receive. Transmission and reception can be performed using an API (Application Programming Interface) provided by a system provider, and can also be performed using a general communication protocol such as HTTP or HTTPS.

  Moreover, transmission and reception can also be performed via the web browser which is a user interface. In this case, the user pastes a sentence (text) to be translated by copy and paste into a form displayed on the WEB browser, and transmits it to an external neural machine translation computer. The translation result, which is the response from the neural machine translation computer, is displayed on the web browser. As a data exchange format in these communications, a JSON (JavaScript Object Notation) format or the like is used.

  In the above-described jointly available neural machine translation system, a system user does not need to prepare a large amount of bilingual sentences for learning, a great amount of time for learning, and maintenance. Also, the system user does not have to possess computer resources (server, GPU, etc.) for system construction. That is, it is convenient that the user can substantially reduce the time required for learning the neural network, the cost of hardware, the cost of maintenance and the labor while using the neural machine translation service.

  On the other hand, when translating highly confidential documents (for example, legal documents such as contracts, especially patent documents before publication, confidential documents in company, documents of research and development, etc.), those secrets to the outside through the Internet It should be avoided that high quality documents are sent.

  Even if the communication path is encrypted by using HTTPS communication or the like, the transmitted document is naturally converted to plain text at the communication destination. Therefore, if you use services provided by other companies, you should avoid sending highly confidential documents over the Internet. Ensuring security by enciphering the communication path and security risk by passing information that should be kept secret to the outside (such as another company) are different dimensions and need to be dealt with separately There is.

  In this regard, it is also conceivable to enter into an agreement regarding confidentiality obligation with the destination of the document (a provider of neural machine translation service on the Internet). However, since there is a problem that information is easily leaked in the Internet society and it is difficult to reveal the leak, it is scarce to realize that concealing the information by the confidentiality agreement is unrealistic. Also, there is no guarantee that such non-disclosure agreements will be fulfilled. Furthermore, it is difficult to regularly supervise and manage the ongoing performance of confidentiality contracts.

  Today, personal computers and smartphones have become familiar tools, and the Internet and web browsers have become everyday necessities that anyone can easily use. On the other hand, the lack of understanding of Internet systems, communication protocols, and information transmission paths has helped to continuously transmit highly confidential documents to machine translation services on the Internet. Also, the risk of information leakage from using machine translation services on the Internet is often overlooked.

  In the translation service and dictionary service via the web, the user inputs a sentence or a word to be translated from a web browser or the like (this includes input methods such as typing, voice input, copy and paste, etc.). The input sentence or word is sent to a server owned (or managed) by a translation service or dictionary service provider, and is stored together with the sender's information.

  Providers of translation services and dictionary services can define usage policies (terms of use) of stored information, but there are also providers that do not. Also, depending on the terms of use, “When the user uploads, provides, or transmits content, the user uses the content for the service provider (and a third party who cooperates with the service provider). There are also provisions that "provide a global license for storage, reproduction, modification, creation of derivative, (public) transmission, publication, public display, and distribution."

  Confidentiality by a translator of a translation company, or a personal translator who receives an order from a translation company, easily entering highly confidential sentences (or partial sentences or words thereof) into a web browser of a personal computer or smartphone There are many cases in which high texts leak to the outside (especially overseas). There is also the problem that such leaks have continued for many years without the knowledge of the translator, the manager of the translator, the translation company, and the source of the translation.

  In particular, leakage of highly confidential documents relating to contracts, R & D documents, patents before publication, etc. to the outside world (especially foreign countries) is not only from the viewpoint of profit protection of one company, but also from the viewpoint of national interest protection. It is absolutely necessary to avoid.

  For example, the following information is described in the material of “Awareness of unintended information leakage of the content entered in“ Attention ”cloud service” which is a press announcement of February 20, 2015 of Information processing promotion mechanism of independent administrative corporation It is done. (Quoted from https://www.ipa.go.jp/about/press/20150220.html.)

  “IPA (Independent Administrative Institution, Information Processing Promotion Organization, Chairman: Kazumasa Fujie) has been aware of the fact that the text entered in the translation service on the Internet has been disclosed as it is on the Internet, For the purpose of raising the user's awareness of cloud service usage, we decided to issue a new alert on the user's attitude about usage. In recent years, various cloud services have been enhanced, and the use of services for individuals as well as for companies has progressed. For example, by storing data such as photos and documents on the Internet, there are services that can be used anytime, anywhere, and translation can be done easily. However, while cloud services have spread rapidly due to their convenience, they have repeatedly pointed out the problem of unintended information leaks because users used the services without properly recognizing the content and risks of the services ( * 1).

  In 2013, the users of the Google group, one of the services provided by Google, did not correctly recognize the setting of the information disclosure range, and it was possible for other parties as well to view the exchange. There is a problem that In addition, in the IPA survey (* 2), they answered that they “do not mind at all” or “do not mind very much” that “information input to the browser, search history, etc. are collected by the company that provided the browser”. Even from 31.3% of the total, users need to improve their awareness and knowledge of cloud services.

  On the other hand, service providers are also required to explain in an easy-to-understand manner the contents of services and the handling of information on the service side in order to prevent information leaks due to lack of explanation to users and lack of user awareness. "

  In addition, the following matters have been described in the February 2014 Call for Information Processing Promotion Agency of the Independent Administrative Agency. (Quoted from https://www.ipa.go.jp/security/txt/2014/02outline.html.)

  "" Does the information be leaked outside without knowing? "-The key to using the cloud service-(omitted)

  While cloud services can be used conveniently, it is inevitable to pass some information to the service provider side.

  The three cloud services shown in Table 1 below are services that are used casually on a daily basis. In particular, "Online Translation Service" allows you to translate sentences without having to install translation software on your PC, but it can also be said that you are translating "the sentences themselves" to the business operator.

  When dealing with confidential information in business, please be aware of this and use it with caution.

  Table 1: How to use cloud services that have the risk of leading to information leaks

  Online translation service

  Overview When you copy and paste a document you want to translate into a web page, the translation program on the service provider side automatically translates.

  Usage risk

  The contents of the copied and pasted source document will be transmitted to the outside, and if the contents correspond to confidential information, it will go against internal regulations while unknown.

  Expected damage example

  An employee exchanged encrypted negotiations with a foreign company via encrypted email. Because the decrypted email was written in English, it was translated into Japanese using an online translation service. Contract information was leaked by that thing. Although I was using encrypted email to prevent information leakage, information was leaked through another route called online translation. "

  Although attention is drawn as described above, the security awareness of those who use an external (such as cloud) translation system remains low because of its high convenience.

  The present invention has been made to solve at least one of the above-mentioned problems.

  To achieve the above object, according to an aspect of the present invention, there is provided a computer used by an unspecified number of persons, wherein a sentence described in a first language is inputted through the Internet and described in a second language. In a translation apparatus for translating into a sentence and communicating with a computer that transmits the translated sentence through the Internet, the sentence described in the first language includes, in part, a word that is a noun and the word The translation apparatus replaces the word portion in the set consisting of the word that is the noun and the reference code that immediately follows it with the character string that corresponds to the word. A first replacement means, a sending means for sending the text replaced by the first replacement means to the computer through the Internet, and the computer From comprises receiving means for receiving via the Internet a sentence written in the second language after translation.

  Preferably, the translation apparatus further comprises second substitution means for replacing the character string replaced by the first substitution means in the text received by the reception means with the corresponding word of the second language. .

  Preferably, the character string after replacement by the first replacement means is a word of the second language corresponding to the word which is the noun, or an encrypted character string.

  Preferably, the noun word includes superscripts or subscripts.

  Preferably, the first replacement means is a read means for reading out the words in one set and the replacement character string from a plurality of sets which are registered in the database and which associate the words with the replacement character strings; Processing for searching one word read out by means from a sentence described in the first language, and replacing the word with a replacement character string associated with the word if it is found; And first processing means for repeatedly performing in a sentence described in a first language, and the first processing means performs all replacements in a sentence described in the first language, The reading means reads out the words in the next set and the replacement character string, and the processing in the first processing means is performed.

  Preferably, the transmitting unit transmits one unit portion among the plurality of unit portions obtained by dividing the sentence described in the first language to the computer, and the receiving unit is configured to Receiving a sentence described in the second language after translation for one unit part, processing the transmission unit and the processing in the reception unit for each of the plurality of unit parts Do.

  Preferably, the unit portion is a portion obtained by dividing a sentence described in the first language at at least one position of a phrase, a reading point, and a line feed code.

  Preferably, the unit portion is a portion obtained by dividing the sentence described in the first language at the position of a phrase, and in the unit portion, a line feed code follows at least a portion of phrases or clauses included in the unit portion. It is inserted.

  Preferably, the receiving means sequentially receives the sentences described in the second language after the translation for the unit part, and combines the received documents by inserting a space between them.

  Preferably, the translation apparatus further comprises post-processing means for post-processing the sentence received by the receiving means, the post-processing means deleting spaces in the sentence, unifying words, upper / lower case in the text Perform at least one of conversion, spell check, and processing of correct notation of plural words.

  Preferably, the sentence described in the first language includes a paragraph number in a part thereof, and the transmission means does not transmit the paragraph number included in the sentence.

  Preferably, the translation device converts a paragraph number included in the sentence from full-width to half-width, and inserts it into the corresponding position of the text received by the receiving means.

  Preferably, the first replacement means replaces a specific word in a sentence described in the first language with a corresponding character string, and at least one of a specific word before and / or after the character string. Insert character code

  Preferably, the specific character code is a code representing a space character or a mark for distinguishing parentheses, quotation marks, and other enclosed parts from others.

  According to another aspect of the present invention, there is provided an external computer used by an unspecified number of persons, wherein a sentence described in a first language is input through the Internet and translated into a sentence described in a second language A control program of a translation device for communicating with an external computer that transmits the translated text through the Internet, the translation device includes a computer, and the text described in the first language is a part of And the control program of the translation apparatus includes a word portion in a set consisting of the word which is the noun and the reference which immediately follows it. , The first replacement step of replacing the word with the character string corresponding to the word, and the text replaced in the first replacement step in the Internet A transmission step of transmitting to the external computer through the external computer, a sentence written in the second language after translation to execute a receiving step of receiving via the Internet to the computer.

  According to still another aspect of the present invention, an external computer used by an unspecified number of people, wherein a sentence described in a first language is input through the Internet and translated into a sentence described in a second language And a translation apparatus using a translation apparatus that communicates with an external computer that transmits the translated text through the Internet, the translation apparatus includes a computer, and the text described in the first language is The part includes a set of words which are nouns and a reference code immediately following them, and the translation method comprises the words which are nouns and a reference code which follows immediately after the translating device. A first replacement step of replacing a word portion in a set with a character string corresponding to the word; and replacing in the first replacement step by the translation device Transmitting the transmitted sentence to the external computer through the Internet, and receiving the translated sentence written in the second language from the external computer through the Internet. including.

  According to the present invention, at least one of the above problems can be solved.

It is a block diagram which shows the structure of the translation system in the 1st Embodiment of this invention. It is a functional block diagram showing composition of a translation system in a 1st embodiment of the present invention. It is a figure which shows the specific example of the data structure stored in the database 207. FIG. It is a flowchart which shows the Japanese-English translation process of the computer program contained in the translation system in the 1st Embodiment of this invention. It is a flowchart following FIG. It is a flowchart which shows the word of the computer program contained in the translation system in the 1st Embodiment of this invention, an encryption registration process. It is a flowchart which shows the word of the computer program contained in the translation system in the 2nd Embodiment of this invention, an encryption registration process. It is a flowchart which shows the Japanese-English translation process of the computer program contained in the translation system in the 3rd Embodiment of this invention. It is a flowchart which shows the Japanese-English translation process of the computer program contained in the translation system in the 4th Embodiment of this invention.

  First Embodiment

  FIG. 1 is a block diagram showing the configuration of a translation system in the first embodiment of the present invention.

  Referring to the figure, the translation system includes computer 100 operated by a user (system user), LAN (Local Area Network) 300, Internet 400, other company computer resource 200, and own company computer resource 500. Be done.

  The computer 100 operated by the user includes a CPU 101, a communication unit 103, an I / O 105, a ROM 107, a RAM 109, a storage device 111, an audio input / output unit 113, a display 115, and a graphic unit 117. A keyboard 119 and a mouse 121 are included.

  The computer 100 is connected to the LAN 300, and the LAN 300 is connected to the Internet 400. The company's computer resource 500 is connected to the LAN 300, and another company's computer resource 200 is connected to the Internet 400.

  Here, computer resources include servers, personal computers, storage devices (storage in a computer, NAS (Network Attached Storage), etc.), information communication paths (networks, load balancers, switches, routers, etc.), and these. CPU, memory, storage device (hard disk, optical or magnetic storage device, semiconductor device other than SSD), ROM, motherboard, keyboard, input device such as keyboard, mouse and microphone, output device such as display and speaker, and storage device Generally indicates all or part of software such as a computer program or data operating on a memory.

  The in-house computer resource 500 is a computer resource owned by the user (or an organization to which the user belongs). The organization may be an organization consisting of one user or an organization consisting of a plurality of personnel.

  The in-house computer resource 500 and the computer 100 are not directly connected to the Internet 400, but are connected to the Internet 400 via a router (not shown) of the LAN 300. Further, by providing the router, the company's own computer resource 500 and the computer 100 with a firewall function, access to the company's computer resource 500 and the computer 100 from the Internet 400 side is prohibited. As a result, the in-house computer resource 500 and the computer 100 are not operated from the outside via the Internet 400, and external access to data stored in the in-house computer resource 500 and the computer 100 is prohibited. .

  The third-party computer resource 200 is a computer resource not owned by the user (or the organization to which the user belongs). By accessing from the computer 100, the user can temporarily borrow the computing capacity and storage of another company computer resource 200. The form of use may be free or chargeable.

  The original text data of the translation, which is a highly confidential sentence, is described in HTML or plain text, and is recorded in the in-house computer resource 500 or the computer 100. If the original text data of the translation is an image, it is converted into an HTML document or text document by the software of the optical character recognition / reader (OCR) recorded in the in-house computer resource 500 or the computer 100 and processed. Ru.

  A communication protocol between the in-house computer resource 500 and the computer 100 and the other-company computer resource 200 is performed by HTTPS (Hypertext Transfer Protocol Secure) or the like. That is, since HTTP communication is performed based on secure encryption (a secret key and a data encryption method using a public key) provided by a Secure Sockets Layer (SSL) / Transport Layer Security (TLS) protocol, the in-house computer The contents of communication between the resource 500 and the computer 100 and the computer resource 200 of another company can be concealed, and leakage of the contents of communication to a third party is prevented. If communication contents can be prevented from leaking to a third party, encrypted communication other than HTTPS may be performed.

  The communication content encrypted by HTTPS or the like is naturally decrypted by the computer resource 200 of another company. For this reason, even if a communication protocol such as HTTPS is used, it is not desirable to transmit the plaintext of the document to be kept secret in the user or the organization to which the user belongs to the other company computer resource 200. In addition, there may be a case where the operator of another company's computer resource 200 reserves the right of disclosure, use, etc., of transmitted data as a usage rule. In such a case, it is technically possible, but not business-friendly, to send the plaintext of the document to be kept secret in the above-mentioned user or in the organization to which the above-mentioned user belongs. It is possible.

  The details of how the communication contents returned to plaintext by the other-company computer resource 200 are stored and used by the other company are practically unknown to the user. Also, in general, the user can not know the configuration and connection of another company's computer resource 200, and can not know the specific handling of received data in the other company.

  Actually, in view of the frequent occurrence of the leakage of confidential information, and it is difficult to prevent it, even if a confidentiality agreement is concluded with the operator of another computer resource 200, it is also considered as confidential. It may not be desirable to send the plaintext of the document to be stored to another company's computer resource 200.

  That is, even if encrypted communication such as HTTPS is used, the text to be translated, which is a highly confidential text, or part of the text to be translated should not be transmitted to the external computer resource 200 of another company.

  Therefore, the computer 100 in the present embodiment encrypts a part of the sentence to be translated, which is a highly confidential sentence, by replacing it with a character string that is difficult to decipher (hereinafter referred to as “first encryption”. Then, when the text is sent to another company's computer resource 200 outside, it is further encrypted (hereinafter referred to as “second encryption”) by a protocol such as HTTPS (SSL encrypted communication). .

  The communication content encrypted by the second encryption is returned to the plaintext by a protocol such as HTTPS in another company's computer resource 200 (it corresponds to “second encryption”, so Called “second decryption”). However, the first encryption is not decrypted, and the document as it is encrypted by the first encryption is translated by a translation program in another computer resource 200. The translated text (text document or HTML document) created by the translation is encrypted by a protocol such as HTTPS (that is, encrypted by the “second encryption” method) and transmitted to the computer 100 through the Internet 400. Be done. The computer 100 returns the received text to plain text by a protocol such as HTTPS, which is the “second decryption” method. In this state, the first encryption has not been decrypted yet.

  The computer 100 performs “first decryption” which is decryption processing for “first encryption” on the received translated data. In addition, processing is performed as necessary. This allows the user of the computer 100 to obtain a complete translated text.

  Even if communication data leaks to other than the company itself and other companies shown in FIG. 1, it is double-encrypted by the first encryption and the second encryption. This prevents leakage of information to a third party other than the communicating party.

  Also, while the other companies shown in FIG. 1 can obtain the decrypted information for the second encryption, they can not decrypt the first encryption. Therefore, leakage of information from the company shown in FIG. 1 to other companies shown in FIG. 1 is also prevented. That is, the other company (person on the operator side of the computer resource 200) shown in FIG. 1 can obtain only the first encrypted text. Also, the post-translational sentences are still subjected to the first encryption. For this reason, the other companies shown in FIG. 1 can not obtain either the full untranslated data before translation or the full untranslated data after translation.

  The first encryption needs to be an encryption that allows machine translation of a text after encryption for the other-company computer resource 200, and an encryption that is strong enough that no one other than the company can understand the contents. There is.

  FIG. 2 is a functional block diagram showing the configuration of the translation system in the first embodiment of the present invention.

  A plurality of computer-readable executable programs that cause the computer 100 to operate as a translation device are recorded in the storage device 111 of FIG. 1 by being sequentially executed by the CPU 101. The program is read from the storage device 111 and expanded on the RAM 109. A program is executed by the CPU 101 on the RAM 109.

  By execution of the program, as shown in FIG. 2, the computer 100 includes a control unit 201, a communication unit 103, a search / replacement unit 203, a document editing / word registration unit 205, a database 207, a memory 209, a storage device 111, and a display. The functions as the output unit 211 and the input unit 213 are exhibited.

  The control unit 201 is a functional block that performs various controls of the entire translation apparatus. The communication unit 103 is a functional block for communicating with internal and external computer resources. The search / replacement unit 203 searches for a specific element (word, clause, etc.) for at least one of the document to be translated and the document after machine translation, and the searched element is represented by a character string or encryption code It is a functional block to replace with. The document editing / word registration unit 205 is a functional block that edits at least one of a document to be translated and a document after machine translation and registers a dictionary for word conversion.

  The database 207 is a database that mainly registers a word dictionary. Software of a database server may be adopted as the database 207, or a simple text file such as a CSV file or a table of spreadsheet software (such as EXCEL) may be recorded, searched, and read out corresponding data. It may be a database.

  The memory 209 is a working area for temporarily recording data. The storage device 111 is a non-volatile storage device, and stores data before translation, data after translation, data recorded in a database, data in the memory 209, and the like. The display / output unit 211 is a user interface such as a display and a speaker. The input unit 213 is a user interface such as a mouse, a keyboard, and a microphone.

  FIG. 3 is a view showing a specific example of the data structure stored in the database 207. As shown in FIG.

  Here, the database 207 is a table having column numbers (column names, field names) as numbers (record numbers and serial numbers of data), words, ciphers, bilingual words, and the like.

  As words, nouns are registered, but verbs, adverbs, adjectives, etc. may be registered. The nouns also include compound nouns composed of a plurality of nouns (e.g., "electric car" which is electric + car).

  The column of numbers records the serial number of one record. The word column registers the translated original word. The encryption column registers the encryption corresponding to the word. The parallel word column registers the translated word corresponding to the word (and the code).

  For example, in FIG. 3, translation data used for translation of Japanese and English are registered. That is, Japanese words, corresponding English words (translational words), and corresponding ciphers are registered. By using the table shown in FIG. 3, it is possible to convert Japanese words into their corresponding code, convert Japanese words into their corresponding English words, convert codes into their corresponding Japanese words, and Or, it is possible to convert a code into its corresponding English word, convert an English word into its corresponding code, and convert an English word into its corresponding Japanese word.

  Further, by adding a column and registering parallel words in other languages in the table of FIG. 3, translation between three or more languages is possible. For example, by adding a column of bilingual Chinese words, Japanese-English, English-Japanese, Japanese-Japanese, Chinese-Japanese, Chinese-English, Chinese-English, and Chinese-English translations (and conversion between words in each language and encryption) are possible.

  As an example, as a record of the number “0”, the database records the original word “electric car”, the encrypted character string “AAA”, and the bilingual word “electric vehicle” in association with each other.

  Such a database may be created in advance by the user, or may be created at the time of translation work or other than translation work by repeating the process shown in the flowchart of FIG. 6 or 7. Also, a vendor can create a database and provide it to the user.

  By using such a database, it is possible to convert words into ciphers, and also to convert words into parallel words. In addition, it is also possible to convert the code into parallel words. Furthermore, it is also possible to convert the parallel translation word into the original word (reverse translation).

  One basic word ("car", "means", "part", "vehicle", "device", "unit", etc.) may be registered in one record for the word and the bilingual word, or a compound word Words ("electric car", "signal input means", "display unit", "electric vehicle", "signal input unit", "display unit", etc.) may be registered in one record.

  The encryption here consists of three alphabets and functions as a virtual word at the time of machine translation. In this case, the encryption starts from AAA and represents a record number (number) described in a decimal number as a 26-base number in which the alphabet of A to Z is assigned to a numerical value of 0 to 25 respectively.

  That is, the lower one digit of the code is 0 to the 26th power, the lower 2nd digit of the code is the 1st power of 26 and the lower 3rd digit (most significant) of the code is the power of 26 It is the place of.

  For example, if the column of "number" is 0, encryption of "AAA" (A indicates a 26-base zero) is applied, and if the column of "number" is 1, "AAB" (A is 26-base) , And B is a 26-digit 1) code. Similarly, for example, if the column of "number" is 3501, the encryption of "FER" (hexadecimal number 3501) is applied.

  That is, the highest digit of the code (lower third digit) is assigned the alphabet corresponding to the integer part of the value divided by 26 (ie, 676). The middle order (lower second digit) of the code is assigned the alphabet corresponding to the integer part of the value obtained by dividing the remainder of the above division by the 26th power (ie 26). The lowermost (lower first digit) of the code is assigned the alphabet corresponding to the remainder of the division. This is equivalent to the conversion from decimal number to 26 number.

  The number of digits of the encryption is not limited to three. Also, among the 3-digit codes, those with meaning as words (for example, CPU, RAM, USB, NOX, etc.) should not be registered as reserved words together with the corresponding numbers, in order to prevent misunderstanding and mistranslation. As well.

  Furthermore, here, the one in which the record number is converted to the 26-adic number is used as the encryption, but the encryption may be determined at random or may be determined by another rule.

  The same word and the translation word may be registered. For example, in chemical formulas and abbreviations (HCl, CPU, etc.), words and parallel words may be registered as the same. In this case, the same word is registered in both Japanese and English. Furthermore, the word (Japanese) may be full-width characters, and the parallel word (English) may be the half-width characters.

  Also, it is possible to register a word in which the part of the original word is described in the translated language, such as the word "information display unit" and the parallel translation word "information display unit".

  The encryption may be expressed as a 26-ary notation following the usual hexadecimal notation (0 to 9 using A to F), and expressed using 0 to 9 and A to P. In addition, lower case letters and symbols (! "# $% & Etc.) may be used. A cipher may be a random character string that does not overlap among multiple ciphers. It is desirable to be a symbol or a combination of an alphabet and a symbol, or a pictogram such as an emoticon may be used as a code.

  As in the present embodiment, let the encryption be AAA, AAB, AAC,..., ZZY, ZZZ, one digit be a hexadecimal number from A to Z, and be represented by a three digit (three letter) alphabet. For example, 26 * 26 * 26 = 17576 words can be registered in the database. If insufficient, by increasing the number of digits of the code, it is possible to register 26 powers of records (a set of a word, a code and a bilingual word) in the database.

  As in the present embodiment, representing a code in a 26-adic notation in which A to Z are each 0 to 25 has an advantage that it is not obvious at first glance what it is a word indicating. Further, in neural machine translation, since a character string of alphabets and symbols is recognized as one word, there is an advantage that machine translation can be performed while leaving the code as a single word in the original text.

  That is, when translating using neural machine translation, the character string of alphabets and symbols is output as it is without being translated (but there are exceptions).

  For example, when the translated text of “the electric vehicle 100 includes the engine 101” is replaced by the table of FIG. 3, the ciphertext of “the AAA 100 includes the AAB 101” is generated. This sentence is translated into a sentence of "AAA 100 includes AAB 101." by neural machine translation. After the neural machine translation is performed, a translated sentence of "electric vehicle 100 includes engine 101." is obtained by converting the code in it into the translation of the original word using the table of FIG. Can.

  In addition, for example, the original text of “a plurality of AAAs” may be translated into “a plural of AAAs”, etc., and the cipher may be converted into plural forms. Also in this case, by replacing the encrypted part of "AAA" of "AAAs" with the English word (for example, "book") corresponding to the word, a translated word of "books" with "s" added at the end is obtained. Can.

  However, in this method, words such as “bus”, “leaf”, and “city” (words that do not form the correct plural form only by adding “s”) become misspelled (spelling error). For this reason, it is necessary to execute post-translational spell-checking routines to handle these words to be spelled correctly after translation. For example, prepare a table that associates “buss” with “buses”, “leafs” with “leaves”, and “city” with “cities” (a table that associates miss spelling with correct spelling), and after machine translation It is also possible to perform batch conversion.

  Also, depending on the learning result of the neural network, the code in the input original text and the code in the corresponding translated text may be different (a kind of mistranslation).

  In order to prevent this, it is better to surround the part of the code with a character indicating that it is a special character string (by inserting a special character before and after the code part) when performing a neural machine translation of the original translation . For example, as shown in "AAA 100 includes AAB 101.", a character of space (blank) is inserted before and after the encryption part and transmitted to the neural network to convert the encryption into another encryption. Can prevent things (a kind of mistranslation).

  In addition to the character indicating a space, it is also effective to enclose the encrypted portion in the document before translation in brackets (""). Similarly, quotation characters ('', "", "", '' etc.), parentheses, double parentheses, double brackets, square brackets, double brackets, braces, brackets, double brackets It is also effective to enclose it in turtle brackets, angle brackets, double angle brackets, narrow corners, or corner brackets. In this way, neural machine translation can be performed on the condition that the enclosed part indicates that it is a part (cipher) having a special meaning different from other parts.

  Characters inserted (added) by these processes are deleted in post-processing after machine translation. For example, if the code is bracketed and machine translated as "AAA" 100 etc., it is translated to "AAA" 100. The double quotation mark is not included in the original text and is added during the above processing, so it is deleted after machine translation to be AAA 100. Deletion is possible by using a regular expression.

  If characters other than alphabets and symbols such as kanji and hiragana are used for encryption, they are translated into English and there is a problem that they can not be decrypted (for example, the character string "assai" assigned as the encryption character string is " "Shallow" (English translation of "shallow"), or a string without meaning such as "oh" is translated into "Aai". Therefore, it is desirable that the encryption be a string of alphabets and symbols.

  In addition, when the table of FIG. 3 is used, the first letter of one sentence of the translated English sentence may be lowercase (if it is English, the beginning of the sentence should start with the capital letter). Also for this, it is better to perform processing to capitalize the beginning of sentences in the later spell check. For example, a process of searching for an alphabet at the beginning of a sentence (or a letter after a period) with a regular expression and using all upper case letters.

  The data in the database of FIG. 3 may be registered by purchasing from a vendor, or may be registered by the user.

  FIG. 4 is a flow chart showing Japanese-English translation processing of a computer program included in the translation system in the first embodiment of the present invention. FIG. 5 is a flowchart following to FIG.

  The processing shown in this flowchart is executed by the computer program stored in the storage device 111 being read onto the RAM 109 and sequentially executed by the CPU 101. The program may be stored in a compiled form of machine language, or the source code may be sequentially executed by an interpreter, or may be written in an intermediate language.

  In this case, the program of the flowchart in FIG. 4 is executed using a word processing software (WORD of Microsoft, OpenOffice published by open source method, etc.) and its macro (VBA: Visual Basic For Applications etc.) Do. The database may be a dedicated database or a CSV file, but here, the database is formed using spreadsheet software such as Microsoft EXCEL, and the database is accessed using the word processing software VBA. I assume. The types of word processing software and program language used are not limited in the practice of the invention.

  Referring to FIG. 4, in step S101, the user opens a document file in which a sentence to be translated is recorded by word processing software (the document file is expanded from the auxiliary storage device to the RAM). Also, open spreadsheet software.

  The document file may be a text file described in a plain text format, may be a document file of word processing software, or may be an HTML file. In addition, the characters included in the image may be files converted to text by OCR. An image file may be input, and in step S101, a file may be created in which the character portion is converted into text data by the OCR function. Besides, the format of the file is not limited to a specific one, as long as sentences can be described. The document may be received from another computer using a file transfer protocol or electronic mail software, or may be input from a storage such as a USB memory. Also, it may be input by the keyboard 119 or a microphone. Documents may be downloaded from the Internet.

  Here, it is assumed that the patent specification described for the electric vehicle is translated, and the document file is

  "[0023]

  The electric vehicle 100 includes an engine 101 and a display unit 102, and the engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input unit 103 is high, the display unit 102 displays a warning. It is assumed that the sentence of "has been described. The translation process in this embodiment will be described by taking the translation of this sentence as an example. [0023] is a paragraph number in the document.

  At step S103, preprocessing is performed on the document to be translated. This aims at the following.

  (1) Define the unit of sentence (delimiter of sentence) to be sent to the other company computer resource 200 at one time.

  (2) In order to reduce mistranslations, the length of one sentence processed at one time in another computer resource 200 is shortened.

  (3) Make clear the part where the meaning in one sentence is separated.

  For example, in the present embodiment, the unit of the sentence transmitted to the other company computer resource 200 at one time is from the beginning of the sentence to the phrase point (.). In the pre-processing, the paragraph number part is searched using a regular expression as a paragraph number is also a unit of a sentence, and a phrase point is added after that (in the above example, the phrase point is added after [0023]) .

  In addition, a heavy sentence (a sentence composed of two or more equal clauses (a set of words including a combination of a subject and a predicate)) is mistranslated less when it is broken into short sentences. Therefore, in the pre-processing, the heavy sentence is divided into two or more sentences (in the above example, “the electric vehicle 100 includes the engine 101 and the display unit 102, and the engine 101 is connected to the signal input unit 103 and the display unit 102 The sentence of “The electric vehicle 100 includes the engine 101 and the display unit 102. The engine 101 is converted to the two sentences of the signal input means 103 and the display unit 102”.

  Compound sentences consisting of adverbial clauses and main clauses that indicate conditions, times, causes, reasons, etc. are related to adverb clauses and main clauses, so they were treated as one sentence rather than being converted into two sentences and processed More accurate translation is possible. On the other hand, there is a trade-off that if one sentence is long, mistranslation is likely to occur. In this embodiment, the adverbial clause and the main clause are taken as one processing unit, but by inserting a line feed code between them, machine translation is performed assuming that both are separate clauses (a compound sentence Perform a machine translation after inserting a line feed code between the clauses to be composed). This makes it possible to make the translation more accurate. (In the above example, a line feed code is inserted after “if it is high,” after the sentence “When the input of the signal input means 103 is high, the display unit 102 displays a warning.”).

  The same process may be performed not only on adverbial clauses but also on adverbial phrases, adjective clauses and adjective phrases.

  By the process at step S103,

  "[0023]

  The electric vehicle 100 includes an engine 101 and a display unit 102, and the engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input unit 103 is high, the display unit 102 displays a warning. The sentence of "is

  "[0023].

  The electric vehicle 100 includes an engine 101 and a display unit 102. The engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input means 103 is high,

The display unit 102 displays a warning. It is converted to the sentence of ". (A line feed code is inserted after "if high").

  In step S105, using the database of FIG. 3, the word in the sentence to be translated after the process of step S103 is subjected to decision conversion into a parallel translation word.

  In this process, the records are read sequentially from the record of the database number 0 in FIG. 3, the full text search of the document to be translated is performed with the words described in the record as the target words, and the found words are the same as those in FIG. It is a process to replace with the bilingual word described in the record. Note that the word may be replaced with the code described in the record here, but since the human readability in the later check is reduced, the word is replaced with the parallel word whose meaning can be understood by the human (translator) here. Is desirable.

  In the example of FIG. 3, first, “electric car” of record number 0 is set as a search word in the document to be translated, and “electric vehicle” is set as a word to replace it (word after substitution). By the processing of the CPU, the word (target word) of “electric car” is retrieved from the top of the document to be translated, and if it is present, it is replaced with “electric vehicle”. Note that a dialog box such as “Do you want to replace 置 き 換 え with ??” is displayed to the user before replacement, and whether to replace the word is determined based on the user's YES / NO input. It may be In addition, full text search and batch replacement (replacement of all target words in a document) may be performed mechanically without obtaining the user's consent. When the process of record number 0 is completed, it is determined whether there is a next record, and the same process is performed until the final record.

  In addition, it is desirable to insert a half-width or full-width space (blank mark) before and after the word of the replacement word in order to make it easy to read out and to make it clear that it is a word break during machine translation. Also, in order to make it clear that the word is a fixed word, the word of the replacement word may be enclosed by square brackets (""). Quotation characters ("," "," ",", etc.), parentheses, double parentheses, double brackets, square brackets, double brackets, braces, turtle brackets, double brackets, It may be enclosed by angle brackets, double angle brackets, gimme, corner brackets, other symbols, etc. Thus, the enclosed portion may be surrounded by a symbol indicating that it is different from other portions.

  If the data is not yet registered in the database (FIG. 3) (the initial stage of system introduction, etc.), the process of step S105 in FIG. 4 is passed (not executed and the text to be translated does not change). .

  Assuming that the data of FIG. 3 is registered in the database,

  "[0023].

  The electric vehicle 100 includes an engine 101 and a display unit 102. The engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input means 103 is high,

The display unit 102 displays a warning. The sentence of "is

  "[0023].

   The electric vehicle 100 includes an engine 101 and a display 102. The engine 101 is connected to the signal input unit 103 and the display 102. If the signal input unit 103 input is high,

 The display 102 displays a warning. Is converted to the statement. (Here, in order to improve readability, a half-width space (space mark) is inserted before and after the replaced word.)

  In step S107, the user visually confirms the document created in the processing up to step S105, and if necessary, edits the text, newly registers new words and ciphers, and performs batch conversion of newly registered words in the document. Do.

  This is done for the following purpose.

  (1) Change the sentences that are still left long, to short sentences.

  (2) The word incorrectly converted in the batch conversion in step S105 is corrected.

  (3) The word to be replaced with the parallel word and the word to be encrypted are collectively converted and registered in the database.

  That is, the user visually confirms the document created in the process up to step S105, and converts the sentences that are still left long into short sentences. For example, a double sentence consisting of two or more sentences is cut at a punctuation mark to make two or more sentences having the same meaning.

  In addition, the user corrects the word erroneously converted in the batch conversion in step S105. Depending on the technical field, etc., there are cases where even the same word should be translated into different words. In such a case, the user converts the word by batch conversion. Also, if necessary, update the database to the correct word, or prepare multiple databases to be used depending on the technical field.

  Since the word was not registered in the database, the user registered a new word, a parallel translation word and a new registration of the cipher in this step S107 for the word not converted in step S105 (in particular, the word before the reference code). Perform batch conversion within the document of newly registered words.

  FIG. 6 is a flow chart showing a process of registering words of a computer program included in the translation system in the first embodiment of the present invention. This process is executed, for example, in response to the user's input in step S107 of FIG.

  Referring to FIG. 6, it is assumed that the user is performing input processing and editing processing of the text in the document in step S203. If the user thinks that registration of the new word in the database is necessary, the user selects the word in the document at step S205. This moves the cursor to the beginning (or end) of the word to be registered when the document is displayed in word processing software, and while pressing the shift key, the direction key to the end (or beginning) of the word to be registered By pressing, the word to be registered is displayed in reverse (or it can be distinguished from other parts by changing the color, adding an underline, etc.). Selection of a word may be performed by dragging the word with a mouse.

  When the shortcut key for word registration is pressed (or a display indicating word registration is selected from the displayed menu), a dialog box is displayed in step S207. The dialog box has a field for entering the parallel word of the selected word. The user inputs the selected word translation into this field to confirm the selected word translation. In addition, by displaying a candidate for bilingual word of the selected word in a dialog box by a database different from that in FIG. 3 or a learned neural network, the bilingual word of the selected word is determined by accepting the selection from the user. You may do it.

  The word selected in step S205 may be a word of which a part is converted in step S105 of FIG. For example, when a parallel translation in which "display unit" is "display unit" is recorded in the database, "information display unit" in the document is converted to "information display unit" by the process of step S105. The user may select "information" and may register the word "information" in the database as a translation, but may select "information display unit" and enter the word "information display unit" in the database as a translation. You may register.

  In step S209, the same word as the word selected in step S205 is full-text searched, and in step S211, it is replaced with the bilingual word determined in step S207. Also in this replacement, a dialog box such as “Do you want to replace ○ with △?” Is displayed to the user before the replacement as in step S105 of FIG. 4, and based on the user's YES / NO input. It may be determined whether to replace the word. In addition, full text search and batch replacement (replacement of all target words in a document) may be performed mechanically without obtaining the user's consent.

  In step S213, the lowermost row in which data registration of the database (FIG. 3) is performed is searched. The next lower line is used as the registration line of the current word, and in step S215, a code is created from the record number. In step S217, a number, a word, a code, a translation word, and the like are newly registered in a line immediately below the bottom line of the database. This updates the database, which is a translation asset.

  After the process in step S217, the process returns to the process from step S203. If the word is not registered in step S205, the process from step S203 is performed.

  As for the event of start of the word registration process in step S205, a shortcut key for word registration is pressed, or a display indicating word registration is selected from the displayed menu, It is preferable to use a shortcut key (because word registration can be performed without using a mouse or voice input), although it may be voice input etc.). It is desirable that the shortcut key be assigned, for example, to pressing of both the control key and the specific key so as not to disturb the character input.

  If the process in step S107 in FIG. 4 is completed, it is checked in step S109 whether or not there is a portion that is difficult to machine-translate in the document generated in the process up to step S107. Specifically, this is as follows.

  (1) If the length of one sentence (from the beginning of a sentence to a phrase point) is equal to or longer than a predetermined length, a warning is output as a portion where the sentence is difficult to machine translate.

  (2) If the length from the beginning of a sentence or a line feed code to the next line feed code or to a punctuation point is equal to or longer than a predetermined length, a warning is output as a portion that is difficult to machine translate the sentence .

  (3) If there is a word that appears a predetermined number of times or more and is not converted (not registered in the database) in the process of step S105 or S107, a warning is output to convert.

  (4) If there is a word that appears before the reference code and is not converted (not registered in the database) in the process of step S105 or S107, a warning is output to convert.

  (5) A warning is output when the correspondence between the subject and the predicate in one sentence is not taken.

  (6) If one subject and one predicate are not included in one sentence, a warning is output.

  In step S109, when the user is stuck in the check, the user continues the process in step S107. The process in step S109 may be omitted.

  In step S111, a process of collectively converting English words into a code is performed on the document obtained in the process up to step S109.

  This process is a process similar to the process at step S105, and records are read sequentially from the record of the database number 0 in FIG. 3 and the full text search of the document to be translated with the bilingual word described in the record as the target word And replace the found word with the code described in the same record of FIG.

  By this,

  "[0023].

   The electric vehicle 100 includes an engine 101 and a display 102. The engine 101 is connected to the signal input unit 103 and the display 102. If the signal input unit 103 input is high,

 The display 102 displays a warning. The sentence of "is

  "[0023].

   The AAA 100 comprises an AAB 101 and a FER 102. The AAB 101 is connected to the AAC 103 and the FER 102. If the AAC 103 input is high,

 The FER 102 displays a warning. Converted to

  In this way, since the word is replaced with the code described in the record, a document having no meaning to humans is generated (first encryption).

  When reading records sequentially from the record of the number 0 of the database in FIG. 3, in the processing in step S111, it is desirable to sort the data in the order of longer parallel words (or in the order including more words).

  For example, "display unit" and "information display unit" may be registered as parallel words in the database. When the word "information display unit" is found in the sentence, only the "display unit" portion is searched first, and if encrypted, "information" remains unencrypted. Therefore, long words can be encrypted first by performing a search in the long order of bilingual words (in order including more words). The search terms may be picked up from the database in the long order of the parallel words (in the order including more words) without sorting the data of the database.

  In step S113, it may be searched whether there is a word (especially noun) before the reference code that can not be converted into a code. That is, if there is a word before the reference code which can not be converted into the code, in step S115, the user is prompted to register in the database as in step S109. By the processes of steps S113 and S115, the word before the reference code can be securely encrypted. The processing in steps S113 and S115 may be omitted.

  In step S117, the document data created by the word processing software by the processing up to step S115 is converted into text data (data such as UTF-8). It may be converted to HTML data or the like, but in order to reduce the amount of data to be communicated and the amount of data to be translated, it is desirable to use text data.

  Next, in step S121 of FIG. 5, one sentence (processing unit to be machine translated at a time) to be processed is specified sequentially from the top of the data obtained in step S117. In this process, the first processing data is from the beginning of the data to the next phrase, and the second processing data is from the next phrase until the end of the sentence (end of the n-th processing data) It is. The last (data end) of the nth processed data is a punctuation mark or the last character of the data.

  "[0023].

   The AAA 100 comprises an AAB 101 and a FER 102. The AAB 101 is connected to the AAC 103 and the FER 102. If the AAC 103 input is high,

 The FER 102 displays a warning. If it is

  The first sentence is "[0023]."

  The second sentence is "(Line feed code) AAA 100 includes AAB 101 and FER 102."

  The third sentence is "AAB 101 is connected to AAC 103 and FER 102."

  The fourth sentence, "If the input of AAC 103 is high, (line feed code)

 The FER 102 displays a warning. It becomes ". These are processed in order from the first sentence and processed in step S121. Although the second and fourth sentences are one sentence to be processed, a line feed code is included therein.

  In step S121, it is determined whether one sentence to be processed does not require translation. In the above example, the first sentence is the paragraph number "[0023]" and contains only symbols and numbers. As described above, if it is one sentence consisting only of a symbol, a number and an alphabet, it is not necessary to translate, so it is added as it is to the variable (queue) to which the translation result is substituted. Alternatively, processing may be performed after conversion of full-width characters to half-width characters, or the like. That is, if YES in step S121, the one sentence is processed as needed in step S123 and added to the end of the queue. This can reduce communication costs and translation costs. Moreover, it can prevent that a mistranslation arises.

  If "NO" in the step S121, the one sentence is transmitted to the external competitor's computer resource 200 by the HTTPS communication in a step S125. The encryption performed in HTTPS communication is the second encryption.

  In step S127, the translation result of the one sentence is received from an external third-party computer resource 200. This communication is also encrypted by HTTPS, and is decrypted by decryption (second decryption) by the computer 100. The translation result is added to the end of the queue.

  When adding the translation result to the end of the queue, a space (or two spaces (two spaces)) may be inserted if it is after the period. That is, received documents may be combined by inserting a space between them. This is a process for improving the readability of the sentence after translation.

  In step S129, it is determined whether the processing has been completed up to the last sentence (nth sentence). If NO, the process from step S121 (processing of the next sentence) is repeated. This completes the processing from the first sentence to the n-th sentence.

  If “YES” in the step S129, the queue includes the translation result. For example, in the above example,

  "[0023].

   The AAA 100 comprises an AAB 101 and a FER 102. The AAB 101 is connected to the AAC 103 and the FER 102. If the AAC 103 input is high,

 The FER 102 displays a warning. The statement is

  "[0023].

   AAA 100 includes AAB 101 and FER 102. AAB 101 is connected to AAC 103 and FER 102. When an input of AAC 103 is high,

 It is converted into a sentence like "FER 102 displays warning." And is included in the queue. Thus, since the translated text is also encrypted, even if someone other than the company obtains the translated text, it can not know the contents.

  Here, in step S131, processing is performed to convert the code in the queue into English words.

  This process is a process similar to the process at step S105, and reads the records sequentially from the record of number 0 of the database in FIG. 3 and performs full-text search of the document to be translated with the code described in the record as the target word. The process is a process of replacing the found cipher with the bilingual word described in the same record of FIG.

  By this, the above

  "[0023].

   AAA 100 includes AAB 101 and FER 102. AAB 101 is connected to AAC 103 and FER 102. When an input of AAC 103 is high,

 FER 102 displays warning. "

  The statement is

  "[0023].

   electric vehicle 100 includes engine 101 and display 102. engine 101 is connected to signal input unit 103 and display 102. when an input of signal input unit 103 is high,

 display 102 displays warning. "

  Is converted to By performing this conversion in-house, the secrecy of the document is secured.

  In this conversion, it is desirable to set a flag for full text search so that the code does not match part of a word. For example, it is intended to prevent erroneous conversion as "ABS" in the word ABSTRACT is a code. Such processing is realized, for example, by performing a full text search with the MatchWholeWord property set to True when using Microsoft's VBA.

  In addition, in order to prevent a word having a meaning such as a CPU or a RAM from being translated as a code, it is desirable that a word having such a meaning not be registered in the database as a code (regarded as a reserved word).

  In step S133, post processing is performed. Specifically, this is the following processing.

  (1) Process of deleting unnecessary space. For example, if two or more spaces are inserted at positions not after the period, processing to make them one. Also, the process of deleting the space between the alphabet and the numerals of reference signs (such as 100A) consisting of alphabets and numerals.

  (2) A process of deleting a line feed code present after a comma or a line feed code other than a line feed code present after a period. For example, in the above example, the line feed code after “When an input of signal input unit 103 is high,” is deleted.

  (3) A process of converting plural form errors into a correct form. For example, it is processing to convert boxes into boxes.

  (4) A process of converting "said" assigned to the translation of "the above" into "the" which is a general word.

  (5) If a special code is added and translated such as putting a code in square brackets, the process of deleting it. For example, if the AAA 100 code is given a bracket such as “AAA” 100 and machine translated, a non-original double quotation mark such as “AAA” 100 is added.

  (6) Correcting the article. The word appearing first will have an indefinite article, and if it appears after a second time, it will have a definite article.

  (7) A process of deleting full-width punctuation marks. (This is a process of deleting the punctuation of the above [0023] and the like.)

  In step S135, the user checks the post-processed document and corrects any errors.

  In step S137, if there is a word or syntax error, the spelling checker in the word processing software corrects it, and if there is a plurality of errors at the same time, correct it. By correcting and editing the errors, the final translated text is completed.

  In this way, machine translation using an external computer resource can be performed while maintaining the secrecy of both the original and translated sentences.

In addition, by registering words including superscripts, subscripts, italics, bold letters, and underlined letters in the database, they are machine-translated in the state of being converted into encryption. Since the ciphers can be converted back to words including superscripts, subscripts, italics, bold, and underlined letters, the machine translation allows superscripts, subscripts, italics, bolds, and underlines. It is prevented that a word including an attached letter is mistranslated. For example, in the database, the word may be registered as "SiO ₂ ", the corresponding code as "ABC", and the parallel word may be registered as "SiO ₂ ", or may be registered as the 3504 record in FIG. Good. If the word before translation and the word of the translation are the same, only the word and the code may be registered in the database, and the parallel word may not be registered. The words may be registered in the database as full-width, double-translational words as half-width or vice versa.

  If the document is low in secrecy, the encryption in steps S111 to S115 and the decryption in step S131 may not be performed, and the document partially converted into parallel translation words may be machine-translated. In this case, the first encryption and the first decryption are not performed, and only the second encryption and the second decryption are performed. This method is effective when there is no need to keep the contents of the translation confidential from the other companies in FIG. In this way, it is possible to prevent the word fluctuation of the document after machine translation. In this method, machine translation may be performed after adding double brackets or the like surrounding the parallel translation word converted by batch conversion. As a result, since machine translation can be performed in a state that it is indicated that the word does not need to be translated, the fluctuation of words in the document after machine translation and mistranslations are further reduced.

  As a translation of "electric vehicle", there are a plurality of translations such as "electric vehicle", "electrical vehicle", "electric-powered vehicle", "electronic vehicle", "battery car", and "battery vehicle". When machine translation is performed by a neural network, it does not know which translation is applied depending on the context, and fluctuations occur in the translated word expression. In addition, the appropriate translation often differs depending on the difference in the field of the document to be translated. In the present embodiment, before machine translation by a neural network, words are converted into translated words in advance. Therefore, the fluctuation of expression is eliminated.

  In addition, when machine translation of Japanese → English is made by using full-size characters for encryption, they become half-size characters after machine translation. For this reason, in the conversion of step S131, it is necessary to take full-text conversion into consideration. This problem does not occur if the encryption is half-width characters from the beginning.

  Second Embodiment

  FIG. 7 is a flow chart showing a process of registering words of a computer program included in the translation system according to the second embodiment of the present invention.

  The process shown in FIG. 6 is not limited to the process in the translation system of FIG. 1, and can be performed independently as a word registration process (database creation process).

  That is, after the text is opened with the word processor software in step S201 of FIG. 7, the processing of step S203 and subsequent steps (the same as the processing of step S203 and subsequent steps of FIG. 6) can be performed.

  In the first and second embodiments, if translation is performed without using encryption (when a document that does not need to be confidential is translated), there is no need to register the encryption in the database. In this case, the process of step S215 is unnecessary.

  Although the encryption is registered in the database, for example, an encryption uniquely determined from the record number may be used, and the encryption may not be recorded in the database but may be calculated from the record number by the computer 100.

  The processing shown by such a flowchart is executed by the computer program stored in the storage device 111 being read onto the RAM 109 and sequentially executed by the CPU 101. The program may be stored in a compiled form of machine language, or the source code may be sequentially executed by an interpreter, or may be written in an intermediate language.

  Even if the program in the flowcharts of FIG. 6 and FIG. 7 is executed using a word processing software (WORD of Microsoft Corporation, OpenOffice published by open source method, etc.) and its macro (VBA: Visual Basic For Applications etc.) Good. The database may be a dedicated database or a CSV file, but here, the database is formed using spreadsheet software such as Microsoft EXCEL, and the database is accessed using the word processing software VBA. I assume.

  Referring to FIG. 7, in step S201, the user opens the document file in which the sentence to be translated is recorded by the word processing software (the document file is expanded from the auxiliary storage device to the RAM). Also, open spreadsheet software.

  The document file may be a text file described in a plain text format, may be a document file of word processing software, or may be an HTML file. In addition, the characters included in the image may be files converted to text by OCR. An image file may be input, and in step S201, a file may be created by converting character parts into text data by the OCR function. Besides, the format of the file is not limited to a specific one, as long as sentences can be described. Documents may be received from an external computer using a file transfer protocol or electronic mail software, or may be input from a storage such as a USB memory. Also, it may be input by the keyboard 119 or a microphone. Documents may be downloaded from the Internet.

  Here, it is assumed that the patent specification described for the electric vehicle is translated, and the document file is

  “The electric vehicle 100 includes the engine 101 and the display unit 102, and the engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input unit 103 is high, the display unit 102 displays a warning. It is assumed that the text of "display" is described.

  In step S203 of FIG. 4, the user checks the content of the opened document and corrects it as necessary. Also add (input) sentences as necessary.

  In step S205 of FIG. 4, the user visually selects sentences to select words to be registered in the database. This moves the cursor to the beginning or end of a word and causes the target word to be highlighted (or displayed with the text and / or background color changed from other parts) by pressing the SHIFT + arrow keys or dragging with the mouse. It is a thing. The word registration is mainly performed on words existing before the reference code, but words without the reference code may also be registered. The reference code is a code for referring to a code (a numeral, a letter, or the like indicating a component used with a lead wire) described in the drawings. For example, "100" of "electric vehicle 100" is a reference code, and the word existing before the reference code means "electric vehicle".

  Selection of a word may be made by human judgment (input of a key or a mouse) as described above, but for example, a word existing before a reference symbol may be automatically detected and selected. Also, not only the presence or absence of the reference code, but a word may be automatically detected and selected.

  If the registration button (or the shortcut key for registration) displayed on the screen is pressed in the state where the word is selected, then in step S207, the CPU 101 selects the parallel word of the word selected in step S205. Accept input. This may be a dialog box displayed on the screen, a keyboard or voice input from the user may be accepted, or a plurality of word candidates may be displayed from the dictionary, and a UI (user interface) that allows the user to select parallel words from among them. You may. This determines the correspondence between the pre-translational word and the post-translational word.

  In step S209, the CPU 101 performs full-text search of the document to be translated, searches for a word before translation determined in the process up to step S207, and converts it into a corresponding translated word in step S211. A full-text search may be performed without confirmation from the user, or a dialog box such as "Do you want to replace with (translational word)? (YES / NO)" every time a word is searched May be displayed on the screen to allow the user to select whether or not to replace a word. Thereby, the word in the full text including the word selected in step S205 is collectively converted to the translated word determined in step S207. For example, if the word "electric vehicle" is associated with "electric vehicle", the word "electric vehicle" in the document is collectively converted to "electric vehicle".

  In step S213, the CPU 101 searches the lowermost line (lowermost line in FIG. 3) in which the information of the database 207 is registered, and acquires the number (serial number) of the blank line which is the next line.

  The number obtained in step S215 is converted into a 26-base number to create a code of the target word. In step S217, the code is recorded as one record together with the pre-translational word and the post-translational word in the lower line in which the information of the database is registered. The information registered in the database is an asset as a translation dictionary to be used in the future.

  After the process of step S217, the process from step S203 is repeatedly performed. When the registration button is not pressed in step S205, the processing from step S203 is repeated.

  When the processing of the whole text is completed, the user saves the document. This is original text data partially translated in the document, and may be subjected to subsequent neural machine translation.

  The words registered in the database in the process of FIG. 6 and FIG. 7 are mainly the words described before the reference symbols. For example, in the original text of the translation object, "the electric vehicle 100 includes the engine 101 and the display unit 102, and the engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input unit 103 is high. If the display unit 102 displays a warning, the “electric car”, the “engine”, the “display unit”, and the “signal input unit” among them are to be registered.

  Note that words such as “input” and “warning” not accompanied by reference numerals may be registered in the database 207 of FIG. 4 together with the encryption and parallel translation words.

  Furthermore, at least a part of the words accompanied by the reference code may be excluded from the registration. That is, at least a part of the “electric car”, the “engine”, the “display unit”, and the “signal input unit” may not be registered.

  After registration in step S213, original words may be sorted in order of length. The longer words are sorted so that the shorter words are located closer to the number 0 (upper side) and the shorter words are located on the higher number side (lower side).

  According to the above process, “the electric vehicle 100 includes the engine 101 and the display unit 102, and the engine 101 is connected to the signal input unit 103 and the display unit 102. When the input of the signal input unit 103 is high, the display is performed. The section 102 displays a warning. "

  The electric vehicle 100 comprises an engine 101 and a display 102, and the engine 101 is connected to the signal input unit 103 and the display 102. If the input of the signal input unit 103 is high, the display 102 displays a warning. It is converted to the sentence of ". The user checks this sentence and determines whether there is an error. If there is an error, edit on the word processor software.

  Third Embodiment

  FIG. 8 is a flow chart showing Japanese-English translation processing of a computer program included in a translation system according to the third embodiment of the present invention.

  The processing shown in this flowchart is executed by the computer program stored in the storage device 111 being read onto the RAM 109 and sequentially executed by the CPU 101. The program may be stored in a compiled form of machine language, or the source code may be sequentially executed by an interpreter, or may be written in an intermediate language.

  Referring to the figure, in step S101, CPU 101 inputs a document (sentence) to be translated according to a user's instruction. It is assumed that the document input here is a sentence after conversion by the processing in FIG. 7. Here, “electric vehicle 100 has engine 101 and display 102, and engine 101 is connected to signal input unit 103 and display 102. If the input of signal input unit 103 is high, display 102 will warn you. An example in which the sentence of "display" is processed will be described.

  In step S103, the CPU 101 searches the input document for a portion that has difficulty in machine translation. For example, when translating from Japanese, the document is divided into multiple sentences by dividing the document by a division point such as "." (Period in the case of translation from English), and each sentence is within a predetermined length. (Ie, whether the length of each sentence is equal to or less than a predetermined threshold) (for example, a value of about 80 to 120 characters is selected as the threshold according to the performance of another company's computer resource 200 ).

  If the length of one sentence is long, machine translation is likely to be inaccurate, so the length of the sentence is checked in advance, and a long sentence is not translatable. For a long sentence, for example, in the case of a complex sentence, a sentence such as “A is B and C is D.” A plurality of sentences such as “A is B. C is D.” A conversion / editing process is performed to convert into sentences and shorten each sentence. Such conversion may be performed automatically, or the user may indicate a sentence determined to be long, and the user may manually request correction. Further, in the case of performing conversion automatically, it is desirable to show the user the sentence after the automatic conversion (and the sentence before conversion if necessary) to make the user check whether the conversion is properly performed.

  As a check of parts that have difficulty in machine translation, do you have the subject in one sentence (for example, does the phrase corresponding to “○○ are” or “○○ is” included?) Or the subject in one sentence You may check whether the predicates correspond or not. Furthermore, if machine translation is performed accurately and without mistranslation, for example, the condition that there is only one subject and one predicate in one sentence may be used as the condition of the translatable sentence.

  These checks may be performed by matching sentence patterns using regular expressions. The check may be performed by morphological analysis. Also, for example, using a learned recurrent neural network (RNN), a sentence to be checked is used as an input, and a sentence having no difficulty in machine translation is output as an output, or a signal having difficulty or noness is output as an output. In some cases, it is also possible to prompt the user to make corrections.

  In step S105, it is checked whether the text is machine-translatable. If there is a difficult part, the process in step S103 is repeated.

  If it is determined in step S105 that machine translation is possible, text preprocessing is performed in step S107. This is a process that facilitates machine translation so that there is no mistake in machine translation. For example, if it is a compound sentence, it has a structure of “When A is B, C does D.” However, in order to prevent a translation error, a compound sentence delimiter (in the case of the above sentence, immediately after “when”) Is a process of embedding a code to be a mark. The code is a line feed code, a space, a character serving as a delimiter, etc. The code is selected to make the delimiter easy to understand without affecting the translation. Also, for example, a double sentence having a structure such as “A is B and C is D.” becomes the above-mentioned mark at the break part of the double sentence (in the case of the above sentence, immediately after “is and”). You may embed code. By the process of step S107, since the unit of sentences to be processed in the machine translation becomes smaller, it is possible to prevent the selection of the wrong subject-predicate combination in the machine translation and the occurrence of other mistranslations.

  By the processing in steps S103 to S107, for example,

  The electric vehicle 100 comprises an engine 101 and a display 102, and the engine 101 is connected to the signal input unit 103 and the display 102. If the input of the signal input unit 103 is high, the display 102 displays a warning. The sentence of "is

  The electric vehicle 100 comprises an engine 101 and a display 102. The engine 101 is connected to a signal input unit 103 and a display 102. When the input of the signal input unit 103 is high (line feed code)

 The display 102 displays a warning. "

  (In step S103, a single culture is performed, and in step S107, a line feed code is inserted at the delimitation of a sentence that forms a compound sentence).

  In step S109, referring to the database in the storage device 111, the words (especially nouns) described before the reference code in the document to be translated are converted into a code (first encryption). The database contrasts the original text, the encrypted string, and the translated string.

  By collectively converting the word before the reference code into the code in step S109, for example,

  The electric vehicle 100 comprises an engine 101 and a display 102. The engine 101 is connected to a signal input unit 103 and a display 102. When the input of the signal input unit 103 is high (line feed code)

 The display 102 displays a warning. The sentence of "is

  "AAA 100 comprises AAB 101 and FER 102. AAB 101 is connected to AAC 103 and FER 102. If the input of AAC 103 is high (line feed code)

The FER 102 displays a warning. "

  Is replaced by the ciphertext of

  In such a ciphertext, since a hexadecimal number such as "AAA" is processed as a word, it is possible to machine-translate with other external computer resources 200 without leaking the true meaning to the outside.

  The process in step S109 is performed as follows.

  (1) From the beginning of the document, search for a component part of a sentence consisting of a combination of a word and a reference code

  (2) The word of the retrieved component part is retrieved from the database.

  (3) Obtain the corresponding cipher from the database.

  (4) The word of the component part retrieved in (2) above is replaced with the code obtained in (3) above.

  (5) Search for a component part of a sentence consisting of a combination of a word and a reference sign which appears next in the sentence.

  (6) Repeat the process from the above (1) until all the component parts are processed. If all components have been processed, exit.

  Alternatively, the process in step S109 may be performed as follows.

  (1) Read out the first word registered in the database

  (2) The read out words are all searched from the top of the document, and replaced with the corresponding encryption registered in the database (full text batch replacement).

  (3) The next word registered in the database is read out and the process of (2) above is repeated. If all the registered words in the database have been processed, the process ends.

  In the processing after step S119, the data before machine translation obtained in the processing up to step S109 will be transmitted to the other company computer resource 200, which has already undergone neural network learning, by HTTPS transmission sequentially from the first sentence. .

  More specifically, the first sentence of the data before translation (for example, the sentence up to the phrase point “.”) Is transmitted to the external competitor computer resource 200 by HTTPS in step S119. This transmission is encrypted (second encryption). The external third-party computer resource 200 translates the received data into a target language in response to the request, and transmits it to the computer 100.

  In step S121, the computer 100 obtains one sentence of translated data from another external computer resource 200 as a response by HTTPS.

  In step S123, it is determined whether the translation is completed up to the last sentence. If NO, the process of the next one sentence is performed in step S119. If the translation is completed to the last sentence, the process proceeds to step S125.

  In step S125, post-translational processing of the document is performed. Here, the translated sentences for each sentence obtained in the process of steps S119 to S123 are combined into one document. In addition, since the translation obtained by the first encryption of steps S109 to S113 includes the encryption, the process of collectively converting the encryption to the corresponding translation is performed. This is performed, for example, as follows.

  (1) Read out the first cipher ("AAA", etc.) registered in the database.

  (2) All the read out ciphers are searched from the top of the document, and replaced with the corresponding translated word in the database (full text batch replacement).

  (3) The next cipher registered in the database is read out and the process from the above (2) is repeated. If processing for all registered encryptions in the database is completed, the process ends.

  In step S127, spell check such as plural word check is performed. The user makes a final check of the translation, and corrects and edits any problems.

  By performing the process in step S127,

  “AAA 100 comprises AAB 101 and FER 102. AAB 101 is connected to AAC 103 and FER 102. When an input of AAC 103 is high, (line feed code)

FER 102 displays alerts. "

  From the sentence of

  “Electric vehicle 100 comprises engine 101 and display 102. engine 101 is connected to signal input unit 103 and display 102. When an input of signal input unit 103 is high, (line feed code)

display 102 displays alerts. "

  Sentence is obtained.

  By performing the same processes as steps S135 and S137 of FIG. 5 in steps S125 and S127, the same effects as those of the first embodiment can be obtained.

  Fourth Embodiment

  FIG. 9 is a flow chart showing Japanese-English translation processing of a computer program included in a translation system according to the fourth embodiment of the present invention.

  The processing in this flowchart is obtained by adding the processing of steps S111 and S113 to the processing of FIG.

  After the process in step S109, it is determined in step S111 whether there is a word before the reference code that could not be converted because there is no registration in the database. This may be determined using a regular expression whether or not there is a text component consisting of a combination of a word and a reference code, or the user may check visually.

  If YES in step S111, the process proceeds to step S113. In step S113, the word that could not be converted in step S111, its code and translated word are registered in the database as a set. For example, a word that could not be converted is highlighted as black and white inverted display, and then a dialog box including the words “Please enter translated ○” and a word input field is displayed, and input is accepted from the user. May be For example, highlight the word "Electric car" in the document, display a dialog box containing the words "Please enter the translation of" Electric car "" and the word entry field, and from the user "electric vehicle" It accepts input. Words and their ciphers and translations are made into sets and registered in the database. In addition, it is desirable to create a cipher automatically.

  The process in step S113 is repeatedly performed until NO in step S111.

  If NO in step S111, the process proceeds to step S119.

  [Others]

  A plurality of the above-described embodiments and elements included in the above (a part of the configuration, a part of the processing) may be combined or replaced to provide another new embodiment.

  When performing full-text batch conversion (processing to convert words into parallel words or parallel words into encryption), it is desirable that the search target words be read from the database in order from the long words. For example, if there are two records in the database, one for converting "detection means" to "detection means" and one for converting "information detection means" to "information detection means" After the processing of collectively converting “detection means” into “information detection means”, the processing of collectively converting shorter words “detection means” into “detection means” is performed. As a result, it is possible to prevent the conversion of only the portion of the "detection means" of the "information detection means" in the document, and to increase the efficiency of translation. Further, also when converting words into codes, reading out from the database in order from long words can prevent only a part of a set of words having a series of meanings from being converted into codes.

  When the words are read out in order from the smallest record number and subjected to batch conversion, the records may be sorted according to the number of registered words.

  In the case of a patent specification, there is an item "explanation of symbols" at the end, in which generally important reference symbols and the components (words) that they indicate are explained there. Only the words described in the item of the explanation of the code may be encrypted. Before machine translation, a warning may be issued if the word described in the item of the code description is not encrypted in the document.

  In the case of a patent specification, a mode for carrying out the invention, a brief description of the drawings, an effect of the invention, a means for solving the problem, a document name, a title of the invention, a technical field, a background art, Prior art documents, patent documents, outlines of the invention, explanation of symbols, document names, claims, document names, abstracts, etc. may be regarded as words and encrypted. This makes it difficult to understand that the text to be translated is a patent specification.

  In a document, such as a patent specification, which is described with reference to the reference numerals of the drawings, words that are nouns preceding the reference numerals in the document have a particularly important meaning. The present invention can be highly effective by replacing at least a part of a word preceding a reference code with a code (or a parallel word). Of course, at least a part of words other than the word before the reference code in the document may be replaced with a code (or a parallel word). Also, do not replace words in the document before reference signs with the code (or equivalent words), but at least partially encrypt words other than the words in the document before the reference signs It is also possible to replace.

  Of the set of words that are nouns and reference signs that appear in the document (a set of words that are nouns and the reference signs that immediately follow them), only the word parts that are nouns are subject to batch conversion in the database It is desirable to register, and not to include reference symbols in the target of batch conversion. In this way, the same words with different reference signs can also be processed using that database (in the future, even when translating other documents, more effectively utilizing that database) Can)

  The invention is also applicable to documents which do not contain reference signs. In this case, at least a part of the words in the document is machine-translated after being replaced with a code (or a parallel word).

  The batch conversion process using the database in step S105 of FIG. 4 and step S109 of FIG. 8 may be performed plural times in succession. For example, the user may convert the "information detection means" in the translated document into "information detection means" by recording in the database a record in which "detection means" → "detection means" is converted in advance. . This is the case where some of the words have been converted. In this case, the user selects "information detection means" as a word, and records "information detection means" as a new record in the database as a parallel translation. When the next document is translated using that database, if the word "information detection means" is present in the document, the word is first converted to "information detection means" by batch conversion using the database. Ru. If the search term "information detection means" subsequently appears in the search order of the database, "information detection means" is converted into "information detection means" without any problem. However, if “information detection means” is earlier than “detection means” in the search order in the database, “information detection means” is It is not converted to "information detection means".

  Therefore, the batch conversion process using the database is repeated twice (or more) to perform the batch conversion again. As a result, even a word such as "information detection means" that includes a partially translated word in the original text is converted into a correct word such as "information detection means".

  In the embodiment, an example of performing machine translation using a neural network has been described, but statistical machine translation, phrase-based, rule-based machine translation, artificial intelligence machine translation, deep learning machine translation, etc. are used. Also, the present invention is not limited to machine translation using a neural network.

  A method of transmitting and receiving information to / from the external third party computer resource 200 may be performed via a WEB page. Information may be transmitted and received by a browser operation automation method. Also, information may be transmitted / received using an API provided by a company operating an external third party computer resource.

  As the above-mentioned post-processing, deletion of unnecessary spaces in sentences, unification of words, conversion of upper case and lower case in sentences, spell check, processing of converting plural words into correct notation and the like may be performed.

  [Effect of the embodiment]

  According to the above embodiment, two-stage encryption of the first encryption and the second encryption is performed on both the translated original text and the post-translation text. For the third party, both encryptions maintain the secrecy of the document. Further, for the provider of the translation service, secrecy of the document is maintained by the first encryption. This makes it possible to avoid the external outflow of information while using an external translation service.

  Even if the computer resource transmitting the information is a computer resource used by an unspecified number (or a plurality of persons, others), the confidentiality of the information can be ensured. Note that "use of unspecified number" means that the site (server) is open to the public and accessible from various places.

  In addition, in the prior art neural machine translation, there is a problem that the information contained in the input sentence can not be translated exactly (there is a possibility that translation omission may occur). In addition, duplicate parts of the translation may be output. In this embodiment, the original text can be compressed (short and simple) by encryption. In addition, the structure of the sentence can be simplified (short and simple). As a result, even if the neural machine translation is adopted, the information contained in the input sentence can be translated without excess or deficiency, and there is an effect that the omission of translation hardly occurs. In addition, there is an effect that it is difficult to output duplicate portions of translation.

  Furthermore, there is an effect that it is easy to use the learning data (bilingual dictionary data, word order data, etc.) used so far as data registered in the database.

  Furthermore, in the prior art neural machine translation, while flexible translation can be performed by Word Embedding, completely different words (sometimes meaningless character strings) may be selected to create translated sentences with completely different meanings. (In particular, words with low frequency of occurrence, proper nouns with low frequency of occurrence, etc.). In addition, in the case of long sentences or complex sentences, translated sentences having meaningless were sometimes created. In this embodiment, a word is converted into a bilingual word or a code in advance and then machine-translated (preferably, machine-translated in a state in which brackets are additionally provided). This prevents machine translation from translating the original words into bilingual words that have completely different meanings.

  The processing in the above embodiment may be performed by software or may be performed using a hardware circuit. In addition, a program for executing the processing in the above-described embodiment can be provided, and the program is provided to the user by being recorded on a recording medium such as a CD-ROM, a flexible disk, a hard disk, a ROM, a RAM, and a memory card. You may decide to do it. The program is executed by a computer such as a CPU. Also, the program may be downloaded to the device via a communication line such as the Internet.

  It should be understood that the above embodiments are illustrative and non-restrictive in every respect. The scope of the present invention is indicated not by the above description but by claims, and is intended to include all modifications within the meaning and scope equivalent to claims.

100 computers 101 CPU
103 Communication unit 109 RAM
111 Storage device 200 Other company computer resource 203 Search / replacement unit 205 Document editing / word registration unit 207 Database 209 Memory 211 Display / output unit 213 Input unit 400 Internet 500 Own company computer resource

Claims (16)

A computer used by an unspecified number of people, which inputs sentences written in a first language through the Internet, translates them into sentences written in a second language, and transmits the translated sentences through the Internet In a translation device that communicates with a computer,
The sentence described in the first language includes, in its part, a set of a word which is a noun and a reference code immediately following it;
The translation device
First replacing means for replacing a word portion in a set consisting of the word which is the noun and a reference code immediately following it with a character string corresponding to the word;
Transmitting means for transmitting the text replaced by the first replacing means to the computer through the Internet;
And a receiving unit configured to receive, from the computer, translated sentences written in the second language through the Internet.   The method according to claim 1, further comprising: second replacing means for replacing the character string replaced by the first replacing means in the text received by the receiving means with the corresponding word of the second language. Translation device as described.   The translation apparatus according to claim 1, wherein the character string after replacement by the first replacement unit is a word of the second language corresponding to the word that is the noun, or a cipher string.   The translation device according to any one of claims 1 to 3, wherein the word which is the noun includes superscripts or subscripts. The first replacement means is
Reading means for reading out a word in one set and a replacement string thereof from a plurality of sets which are registered in the database and which associate words with replacement strings thereof;
A process of searching for one word read by the reading means from a sentence described in the first language, and replacing the word with a replacement character string associated with the word if it is found A first processing unit that repeatedly performs in a sentence described in the first language;
After all substitution in the sentence described in the first language is performed by the first processing means, the reading means reads the word in the next set and the replacement character string, and the first process The translation apparatus according to any one of claims 1 to 4, wherein the processing is performed by means. The transmitting means transmits one unit part of the plurality of unit parts obtained by dividing the sentence described in the first language to the computer.
The receiving means receives a sentence described in the second language after translation for the one unit part,
The translation apparatus according to any one of claims 1 to 5, wherein the processing in the transmission unit and the processing in the reception unit are performed on each of the plurality of unit parts.   The translation device according to claim 6, wherein the unit part is a part obtained by dividing a sentence described in the first language at at least one position of a phrase, a reading point, and a line feed code. The unit part is a part obtained by dividing the sentence described in the first language at the position of the phrase point,
The translation device according to claim 6, wherein a line feed code is inserted in the unit part after at least a part of phrases or clauses included in the unit part.   9. The reception method according to claim 6, wherein the reception means sequentially receives the sentences described in the second language after the translation for the unit part, inserts the spaces between the received documents, and combines them. The translation device according to any of the above. It further comprises post-processing means for post-processing the text received by the receiving means,
The post-processing means performs at least one of deletion of a space in a sentence, unification of words, conversion of upper / lower case in a sentence, spell check, and correct notation of plural words. The translation apparatus according to any one of. The sentences described in the first language include, in part, paragraph numbers,
The translation apparatus according to any one of claims 1 to 10, wherein the transmission means does not transmit a paragraph number included in the sentence.   The translation apparatus according to claim 11, wherein a paragraph number included in the sentence is converted from full-width to half-width, and inserted into a corresponding position of the text received by the receiving means.   The first replacing means replaces a specific word in a sentence described in the first language with a corresponding character string, and a specific character code at least before or after the character string. The translation device according to any one of claims 1 to 12, which is inserted.   The translation apparatus according to claim 13, wherein the specific character code is a code representing a space character or a mark for distinguishing brackets, quotation marks, and other enclosed parts from others. It is an external computer used by an unspecified number of people, and it inputs sentences written in the first language through the Internet, translates it into sentences written in the second language, and sends the translated sentences through the Internet In a control program of a translation apparatus that communicates with an external computer
The translation device includes a computer
The sentence described in the first language includes, in its part, a set of a word which is a noun and a reference code immediately following it;
The control program of the translation device is
A first replacement step of replacing a word portion in a set consisting of the word which is the noun and a reference code immediately after it with a character string corresponding to the word;
Transmitting the text replaced in the first replacement step to the external computer through the Internet;
A control program of a translation apparatus, which causes a computer to execute a receiving step of receiving a sentence described in the second language after translation from the external computer through the Internet. It is an external computer used by an unspecified number of people, and it inputs sentences written in the first language through the Internet, translates it into sentences written in the second language, and sends the translated sentences through the Internet In a translation method using a translation device that communicates with an external computer
The translation device includes a computer
The sentence described in the first language includes, in its part, a set of a word which is a noun and a reference code immediately following it;
The translation method is
A first replacement step of replacing, by the translation device, a word portion in a set consisting of the word which is the noun and a reference code immediately following it, with a character string corresponding to the word;
Transmitting, by the translation device, the text replaced in the first replacement step to the external computer through the Internet;
Receiving at the translation apparatus the sentence described in the second language after translation from the external computer via the Internet.

Images