JP7333933B2 - TRANSLATION DEVICE, CONTROL PROGRAM FOR TRANSLATION DEVICE, AND TRANSLATION METHOD USING TRANSLATION DEVICE - Google Patents

Description

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

It was in the 1950's in the Christian era that research into basic techniques for machine translation (MT), such as morphological analysis and dependency analysis, began. In the 1980s of the Christian era, as computer performance improved, rule-based machine translation technology became practical to some extent.

In the 1990s AD, the technique of Statistical Base Machine Translation (STM) was developed. This statistically processes the correspondence between words and sentence structures between different languages, and selects words and translation sentences with a high probability of being translated from translation candidates. In statistical machine translation, a translation model determines the word corresponding to the original word and the word order of the words according to probability. Furthermore, the language model determines the correct arrangement of words after translation according to probability. Statistical machine translation requires a large amount of bilingual data (training data) 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 and the correct translation.

In statistical machine translation, in order to determine the correct word order, the sentences in the source language are syntactically analyzed, and the word order of the words in the source language is preliminarily changed to the word order of the target language using the dependency relationship. Techniques for converting are known.

In the 2010s, neural machine translation (NMT) using deep learning by neural networks appeared. It has achieved rapid practical application since around 2015 AD, and is replacing statistical machine translation. Neural machine translation generally translates text directly (end-to-end) using a neural network without pre-processing the original text to be translated (reversing word order, etc.).

In neural machine translation, words are treated as real-valued vectors and treated as distributed representations of hundreds of dimensions. That is, the vector expresses the semantic and syntactic information of a word (word embedding). From the semantic relationship between one original word and its translated word, the semantic relationship between another original word and its translated word can be inferred to some extent, so flexible translation is possible.

In machine translation, the length of input data (original text length) is not constant, and it is necessary to use data input in the past. For this reason, neural machine translation generally uses a recurrent neural network (RNN). When the words (or characters) of the original sentence are 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, the weighting coefficient of each neuron (node) serves as translation knowledge, so a table for translation (a table that associates original words with translated words, such as a phrase table) is not required. Unlike statistical machine translation, which translates by replacing and rearranging the constituent words of the original text, neural machine translation creates a new translation from the input text to match the learned language model. It can be said that it is a thing.

JP 2018-120584 A

Compared to rule-based machine translation and statistical machine translation, neural machine translation can flexibly translate even unlearned sentences (by word embedding, etc.), even if the translated sentence is fluent. has the advantage of being

However, unlike rule-based machine translation and statistical machine translation, neural machine translation is not a technology that translates by replacing the original text, so it is difficult to accurately translate the information contained in the input sentence. There was a problem (possibility of omission of translation). In addition, duplicated translations were sometimes output.

Furthermore, in neural machine translation, it is not possible to convert the word order of the words in the source language in advance, which was used in statistical machine translation, so the learning that has been used so far cannot be used. There is a problem that it is difficult to use data (dictionary data, word order data, etc.).

Furthermore, in neural machine translation, word embedding enables flexible translation, but on the other hand, by selecting words with completely different meanings, translated sentences with completely different meanings may be created (especially words with a low frequency of occurrence, infrequent proper nouns, etc.). In addition, in the case of long sentences and complicated sentences, translations that do not make sense are sometimes created.

Furthermore, in neural machine translation, unlike rule-based machine translation and statistical machine translation, it is generally difficult for humans to understand the process of translation. I had a difficult problem. As a result, neural machine translation has a problem in that mistranslation tends to occur, and the mistranslated sentence is likely to be provided as the final product.

Fluency and flexibility of the generated translations are more important than accuracy of content for daily conversations and documents for conveying unimportant information, even if there are some mistranslations or omissions in translation. is sometimes emphasized. Neural machine translation is extremely 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. Therefore, even if neural machine translation is used, it is necessary to spend a considerable amount of time and effort to check for mistranslations and missing translations. Insufficient checks may lead to critical mistranslations or missing translations, resulting in financial damage.

Another problem with neural machine translation is the difficulty in handling character decorations and styles, such as superscripts, subscripts, italic, bold, and underlined characters. In sentences, for example, if it is a superscript character, it is indicated that it is a superscript character by enclosing a normal character in HTML with tags ^~, or a character code such as Unicode In some cases, superscripts are expressed directly in . If these are mixed and used for machine learning, or if the handling of character ornaments and styles in a document during translation differs from that during learning, there is a possibility that incorrect translation results will be output.

Superscripts and subscripts are used to describe chemical symbols, chemical formulas, formulas, variables, etc., but in neural machine translation, due to the effects of Word Embedding, seemingly similar but completely different chemical symbols, chemical formulas, formulas, variables may be output as a translation result. Since it is difficult to find errors in chemical symbols, chemical formulas, formulas, and variables at first glance, they often remain as mistranslations in translation products.

Furthermore, in neural machine translation, there is a problem that the same word in the original text is translated into different types of words. For example, the word "address" in the original document may be translated as "address", "address", or "address", depending on where it appears, even though it has the same meaning. There are times when it is done. Also, for example, the word "display unit" included in the original document may be translated as "display unit", "display portion", or " It is sometimes translated as "display means". Such fluctuations in notation and the output of words with different meanings are not preferable particularly in the translation of legal documents and patent documents, which require strictness.

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

In order to reduce the cost of neural machine translation, consider using a trained external computer (Internet-connected sites, servers, etc. operated by other companies or others) that is not dedicated to your company. be done. One business provides a trained neural machine translation system on the Internet, and a plurality of users (unspecified number of users) use the system on the Internet. The use by an unspecified number of people means that the site (server) is open to the public and can be accessed from various places.

When using such a system, users of the system send document data to a trained neural machine translation computer operated by another company via the Internet, and the results of machine translation performed by that server are sent via the Internet. receive. Transmission and reception can be performed using an API (Application Programming Interface) provided by a system provider, or can be performed using general communication protocols such as HTTP and HTTPS.

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

In the above-described jointly usable neural machine translation system, system users do not need to prepare a huge amount of bilingual sentences for learning, and do not need a lot of time and maintenance for learning. System users do not need to own computer resources (servers, GPUs, etc.) for system construction. In other words, the user can use the neural machine translation service while minimizing the time required for learning the neural network, the hardware cost, and the maintenance cost and effort, which is convenient.

On the other hand, when translating highly confidential documents (for example, legal documents such as contracts, especially patent documents before disclosure, confidential documents within the company, documents for research and development, etc.), it is necessary to transmit these confidential documents to the outside via the Internet. Avoid sending highly sensitive documents.

Even if the communication path is encrypted by using HTTPS communication or the like, the transmitted document is naturally converted into plaintext at the communication destination. Therefore, when using services provided by other companies, it is necessary to avoid sending highly confidential documents over the Internet. Ensuring security by encrypting the communication path and security risk by passing confidential information to the outside (other companies, etc.) are two different problems, and need to be dealt with independently. There is

In this regard, it is conceivable to conclude a non-disclosure agreement with the recipient of the document (a business that provides neural machine translation services on the Internet). However, in the Internet society, there is a problem that information is easily leaked, and the leak is difficult to be made clear. Therefore, it is impractical and dangerous to conceal information by means of a non-disclosure agreement. In addition, the reliable performance of such non-disclosure agreements cannot be guaranteed. Furthermore, it is difficult to regularly supervise and manage the ongoing performance of non-disclosure agreements.

Today, personal computers and smartphones have become familiar tools, and the Internet and web browsers have become daily necessities that anyone can easily use. On the other hand, due to the lack of understanding of Internet systems, communication protocols, and information transmission routes, there is no end to cases in which highly confidential documents are easily sent to machine translation services on the Internet. Also, the risk of information leakage due to the use of machine translation services on the Internet is often overlooked.

In translation services and dictionary services via the WEB, users input sentences or words to be translated from a WEB browser or the like (input methods include typing, voice input, copy-and-paste, and the like). The entered sentences and words are transmitted to a server owned (or managed) by a provider of translation services and dictionary services, and stored together with the sender's information.

Businesses providing translation services and dictionary services can establish usage policies (terms of use) for the accumulated information, but there are also businesses that do not establish such policies. In addition, depending on the terms of use, "When a user uploads, provides, or transmits content, the user may request the service provider (and third parties working with the service provider) to use the content. , grants a worldwide license to store, reproduce, modify, prepare derivative works from, (publicly) transmit, publish, publicly display and distribute."

A translator at a translation company, or an individual translator who has received an order from a translation company, can easily enter highly confidential sentences (or partial sentences or words) into the web browser of a computer or smartphone, thereby reducing confidentiality. There have been many cases where documents with a high score are leaked to the outside (especially overseas). There is also the problem that such leaks have continued for years without the knowledge of the translator, the translator's manager, the translation company, or the source of the translation.

In particular, leaking highly confidential documents such as contracts, research and development documents, and unpublished patents to the outside (especially overseas) is not only from the perspective of protecting the interests of a company, but also from the perspective of protecting national interests. Must be avoided at all costs.

For example, in the press release dated February 20, 2015 by the Information-technology Promotion Agency, Japan, "Warning: Beware of Unintentional Leakage of Information Entered into Cloud Services," the following items are stated: It is (Quoted from https://www.ipa.go.jp/about/press/20150220.html.)

“IPA (Independent Administrative Agency, Information-technology Promotion Agency, President: Kazumasa Fujie) was informed of the problem that sentences entered into an online translation service were published as they were on the Internet. In order to raise the awareness of users regarding the use of cloud services, we decided to once again issue a warning regarding the mindset when using cloud services. In recent years, various cloud services have been enhanced, and the use of services not only for companies but also for individuals is progressing. For example, by storing data such as photos and materials on the Internet, there are services that can be used anytime, anywhere, and can be easily translated. However, while cloud services have rapidly spread due to their convenience, the problem of unintentional information leaks has been repeatedly pointed out due to users using the services without properly recognizing the content and risks of the services. *1).

In 2013, users of Google Groups, one of the services provided by Google, did not correctly recognize the setting of the scope of information disclosure, so that even non-related parties could view the exchanges. I have a problem. In addition, in the IPA survey (*2), regarding the fact that "information entered into the browser and search history are collected by the company that provides the browser," respondents answered that they were "not at all concerned" or "not very concerned." Since people accounted for 31.3% of the total, it is necessary to improve user awareness and knowledge of cloud services.

On the other hand, in order to prevent information leaks due to insufficient explanation to users or lack of awareness by users, service providers are also required to provide easy-to-understand explanations of the content of services and the handling of information on the service side. 』

In addition, the Information-technology Promotion Agency's "February 2014 Reminder" states the following: (Quoted from https://www.ipa.go.jp/security/txt/2014/02outline.html.)

“Are you leaking information to the outside without knowing it?” ~ Vital points in using cloud services ~ (Omitted)

While cloud services are convenient to use, it is inevitable that some information will be passed on to the service provider.

The three cloud services in Table 1 below are services that we casually use in our daily work. In particular, the "online translation service" allows you to easily translate sentences without installing translation software on your computer, but it can also be said that the original sentence itself is handed over to the operator.

If you handle confidential information in your business, please be aware of this and use it with caution.

Table 1: How to use cloud services that may lead to information leaks

online translation service

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

Risk of use

If the content of the copied and pasted translation source document is sent to the outside, and if the content corresponds to confidential information, it will violate internal regulations without knowing it.

Assumed example of damage

An employee exchanged contract negotiations with an overseas company via encrypted email. The decrypted email was written in English, so I used an online translation service to translate it into Japanese. As a result, the contract information was leaked. I used encrypted email to prevent information leaks, but information was leaked through another channel, online translation. 』

Despite the above warnings, security awareness among those who use external translation systems (clouds, etc.) remains low due to their high convenience.

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

In order to achieve the above object, according to one aspect of the present invention, there is provided a computer used by an unspecified number of people, in which a sentence written in a first language is input via the Internet and written in a second language. in a translation device that communicates with a computer that translates the translated sentences into a new sentence and transmits the translated sentences through the Internet, wherein the sentences written in the first language are partly composed of words that are nouns and their and the translation device replaces word portions in the set of the noun word and the reference sign immediately following it with a string corresponding to the word. a first replacement means; a transmission means for transmitting the text replaced by the first replacement means to the computer via the Internet; and a text written in the translated second language from the computer. through the Internet.

Preferably, the translation device further comprises second replacing means for replacing the character string replaced by the first replacing means in the text received by the receiving means with a corresponding word in the second language. .

Preferably, the character string after replacement by the first replacement means is a word in the second language corresponding to the noun word or a cipher character string.

Preferably, said noun words include superscripts or subscripts.

Preferably, the first replacement means includes reading means for reading a word and its replacement character string in one set from a plurality of sets registered in a database, which associate a word with its replacement character string; The processing of searching for one word read by the means from sentences written in the first language and, if found, replacing the word with the replacement character string associated with it. and a first processing means for repeatedly performing in a sentence written in a first language, wherein after all the replacements in the sentence written in the first language are performed by the first processing means, the The reading means reads out the words in the next set and their replacement character strings, and processes them in the first processing means.

Preferably, the transmitting means transmits to the computer one of a plurality of unit parts obtained by dividing the text written in the first language, and the receiving means transmits the Receive a sentence written in the second language after translation for one unit part, and perform processing by the transmitting means and processing by the receiving means for each of the plurality of unit parts. conduct.

Preferably, the unit portion is a portion obtained by dividing a sentence written in the first language at at least one of a period, a comma, and a line feed code.

Preferably, the unit part is a part obtained by dividing a sentence written in the first language at the position of a period, and in the unit part, at least a part of the phrase or clause included therein is followed by a linefeed code. inserted.

Preferably, the receiving means sequentially receives sentences written in the second language after translation of the unit parts, and joins the received documents by inserting a space therebetween.

Preferably, the translation device further comprises post-processing means for post-processing the text received by the receiving means, wherein the post-processing means deletes spaces in the text, unifies words, and adjusts uppercase/lowercase letters in the text. At least one of converting, spell checking, and correcting plural words.

Preferably, the text written in the first language partly includes a paragraph number, and the transmitting means does not transmit the paragraph number included in the text.

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

Preferably, the first replacing means replaces a specific word in a sentence written in the first language with a corresponding character string, and at least either before or after the character string, a specific word Insert character code.

Preferably, the specific character code is a code representing a space character, or a mark for distinguishing the enclosed portion such as brackets, quotation marks, or the like.

According to another aspect of the present invention, an external computer used by an unspecified number of people inputs a text written in a first language through the Internet and translates it into a text written in a second language. , a control program for a translation device that communicates with an external computer that transmits translated text over the Internet, wherein the translation device includes a computer, and the text written in the first language is part of contains a set consisting of a noun word followed immediately by a reference sign, and the control program of the translation device is configured to interpret a word part in the set consisting of the noun word followed by a reference sign. is replaced with a character string corresponding to the word, a transmission step of transmitting the text replaced in the first replacement step to the external computer via the Internet, from the external computer, and a receiving step of receiving the translated text written in the second language via the Internet.

According to still another aspect of the present invention, an external computer used by an unspecified number of people receives a sentence written in a first language through the Internet and translates it into a sentence written in a second language. and, in the translation method using a translation device that communicates with an external computer that transmits the translated text via the Internet, the translation device includes a computer, and the text written in the first language is: A part thereof includes a set consisting of a noun word followed immediately by a reference sign, and the translation method comprises a set of the noun word followed by a reference sign by the translation device. a first replacement step of replacing a word portion in the set with a character string corresponding to the word; and transmitting the sentence replaced in the first replacement step by the translation device to the external computer via the Internet. and a receiving step of receiving the translated text written in the second language from the external computer by the translation device via the Internet.

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

1 is a block diagram showing the configuration of a translation system according to a first embodiment of the present invention; FIG. 1 is a functional block diagram showing the configuration of a translation system according to a first embodiment of the present invention; FIG. 3 is a diagram showing a specific example of a data structure stored in a database 207; FIG. 4 is a flowchart showing Japanese-English translation processing of a computer program included in the translation system according to the first embodiment of the present invention; 5 is a flowchart continued from FIG. 4; 4 is a flow chart showing word and code registration processing of a computer program included in the translation system according to the first embodiment of the present invention. FIG. 10 is a flow chart showing word and code registration processing of a computer program included in the translation system according to the second embodiment of the present invention; FIG. FIG. 11 is a flow chart showing Japanese-English translation processing of a computer program included in the translation system according to the third embodiment of the present invention; FIG. FIG. 12 is a flowchart showing Japanese-English translation processing of a computer program included in the translation system according to the fourth embodiment of the present invention; FIG.

[First embodiment]

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

Referring to the figure, the translation system comprises a computer 100 operated by a user (system user), a LAN (Local Area Network) 300, the Internet 400, other company's computer resources 200, and an in-house computer resource 500. be done.

A computer 100 operated by a 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, a graphic unit 117, It includes a keyboard 119 and a mouse 121 .

Computer 100 is connected to LAN 300 , and LAN 300 is connected to Internet 400 . A company 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 computers, NAS (Network Attached Storage), etc.), and information communication paths (networks, load balancers, switches, routers, etc.), and their components. CPU, memory, storage device (hard disk, optical or magnetic storage device, SSD or other semiconductor device), ROM, motherboard, input device such as keyboard, mouse, microphone, output device such as display and speaker, and storage device All or part of software such as computer programs and data that are read from and generally run on memory.

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

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

The other company's computer resource 200 is a computer resource that is 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 power and storage of the other company's computer resource 200 . It does not matter whether the form of use is paid or free of charge.

The translation original text data, which is highly confidential text, is written in HTML or plain text and recorded in the computer resource 500 of the company or the computer 100 . When the original text data for translation is an image, it is converted into an HTML document or a text document by OCR (Optical Character Recognition/Reader) software recorded in the company's own computer resource 500 or computer 100, and processed. be.

The communication protocol between the company computer resource 500 and the computer 100 and the other company's computer resource 200 is HTTPS (Hypertext Transfer Protocol Secure) or the like. That is, since HTTP communication is performed on secure encryption (data encryption method using private key and public key) provided by SSL (Secure Sockets Layer) / TLS (Transport Layer Security) protocol, The content of communication between the resource 500 and the computer 100 and the other company's computer resource 200 can be made confidential, and the content of communication can be prevented from being leaked to a third party. Encrypted communication other than HTTPS may be used as long as it is possible to prevent the contents of communication from being leaked to a third party.

Communication contents encrypted by HTTPS or the like are naturally decrypted by the other company's computer resource 200 . For this reason, even if a communication protocol such as HTTPS is used, it is not desirable to send the plain text of a document that should be kept confidential within the user or the organization to which the user belongs to the other company's computer resource 200 . In addition, there are cases where the operator of the other company's computer resource 200 reserves the right to disclose and use the sent data as the terms of use. In such a case, although it is technically possible to transmit the plain text of a document that should be kept confidential within the user or the organization to which the user belongs to the other company's computer resource 200, it is impractical from a business point of view. It is possible.

The user does not know the details of how the communication contents returned to plaintext by the other company's computer resource 200 are actually stored and used by the other company. Further, generally, the user cannot know the configuration and connection of the other company's computer resource 200, and cannot know the specific handling of received data inside the other company.

In reality, leakage of confidential information occurs frequently and it is difficult to prevent it. It can be said that it is not desirable to send the plaintext of the document to be stored to the other company's computer resource 200 .

In other words, even if encrypted communication such as HTTPS is used, highly confidential sentences to be translated or part thereof should not be sent to the external computer resources 200 of other companies.

Therefore, the computer 100 in this embodiment encrypts a part of the highly confidential text to be translated by replacing it with a character string that is difficult to decipher (hereinafter referred to as “first encryption”). ), and then, when the text is transmitted to the external computer resource 200 of another company, it is further encrypted by a protocol such as HTTPS (SSL encrypted communication) (hereinafter referred to as "second encryption"). .

In the other company's computer resource 200, the communication content encrypted by the second encryption is returned to plain text by a protocol such as HTTPS (because it corresponds to the "second encryption", this is hereinafter referred to as " second decryption"). However, the first encryption is not decrypted, and the document still encrypted by the first encryption is translated by the translation program in the other company's computer resource 200 . The translated sentence (text document or HTML document) created by the translation is encrypted with a protocol such as HTTPS (i.e., encrypted by the "second encryption" method) and sent to the computer 100 via the Internet 400. be done. The computer 100 restores the received text to plaintext by a protocol such as HTTPS, which is a "second decryption" technique. In this state, the first encryption has not yet been decrypted.

The computer 100 performs "first decryption", which is a decryption process for the "first encryption", on the received translated data. In addition, post-processing is performed as necessary. This allows the user of the computer 100 to obtain a complete translated sentence.

Even if communication data is leaked to a company other than one's own company and other companies shown in FIG. 1, it is double-encrypted by the first encryption and the second encryption. This prevents information from being leaked to a third party other than the communicating parties.

In addition, the other company shown in FIG. 1 can obtain the decrypted information for the second encryption, but cannot 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 companies (persons on the side of the operator of the computer resource 200) shown in FIG. 1 can obtain only the first encrypted text. In addition, the sentence after translation also remains in the first encryption. For this reason, the other companies shown in FIG. 1 cannot obtain both unencrypted complete pre-translation data and unencrypted complete post-translation data.

The first encryption needs to be encryption that is strong enough to enable machine translation of the encrypted text for the other company's computer resources 200, and strong enough that anyone other than the company cannot understand the content. There is

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

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

By executing the program, as shown in FIG. / functions as an output unit 211 and an input unit 213 .

A control unit 201 is a functional block that performs various controls of the translation apparatus as a whole. The communication unit 103 is a functional block for communicating with internal/external computer resources. The search/replace unit 203 searches for specific elements (words, clauses, etc.) in at least one of the document to be translated and the machine-translated document, and replaces the searched elements with corresponding character strings or codes. It is a function block to replace with. A 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 word dictionaries. As the database 207, database server software may be employed, or simple text files such as CSV files, tables of spreadsheet software (such as EXCEL) may be recorded, searched, and the corresponding data read out. It may be used as a database.

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

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

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

As words, nouns are registered, but verbs, adverbs, adjectives, etc. may also be registered. Nouns also include compound nouns consisting of multiple nouns (such as "electric vehicle", which is "electricity + vehicle").

The number column records the serial number of one record. The word column registers the words of the translation source text. The cipher column registers the cipher corresponding to the word. The parallel translation word column registers the translated word corresponding to the word (and code).

For example, in FIG. 3, translation data used for Japanese and English translation are registered. That is, Japanese words, corresponding English words (translated words), and corresponding ciphers are registered. By using the table of FIG. 3, Japanese words can be converted into their corresponding ciphers, Japanese words can be converted into their corresponding English words, and ciphers can be converted into their corresponding Japanese words. , ciphers can be converted into their corresponding English words, English words can be converted into their corresponding ciphers, and English words can be converted into their corresponding Japanese words.

Further, by adding a column and registering parallel translation words of 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-Chinese, Chinese-Japanese, English-Chinese, and Chinese-English translations (and conversion between words in each language and codes) are possible.

For example, as a record numbered "0", the database associates and records the original word "electric vehicle", the encrypted character string "AAA", and the translated word "electric vehicle".

Such a database may be created in advance by the user, or may be created during or outside of the translation work by repeating the processing shown in the flowchart of FIG. 6 or 7 . Alternatively, a vendor can create a database and provide it to the user.

By using such a database, words can be converted into codes, and words can be converted into parallel words. In addition, it is also possible to convert ciphers into parallel words. Furthermore, it is also possible to convert a bilingual word into the original word (back-translation).

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

The cipher here consists of three letters of the alphabet and functions as a virtual word during machine translation. Here, the cipher is expressed as a hexadecimal number in which a record number (number) written in decimal numbers starting with AAA is assigned alphabetic characters A to Z to numerical values 0 to 25, respectively.

That is, the last 1 digit of the ciphertext is the 0th power of 26, the second digit of the ciphertext is the 1st power of 26, and the last 3rd digit (most significant) of the ciphertext is the square of 26. It is the place of

For example, if the "number" column is 0, the encryption is "AAA" (A indicates hexadecimal 0), and if the "number" column is 1, it is encrypted with "AAB" (A is the hexadecimal number). indicates zero, and B indicates 1 in hexadecimal). Similarly, for example, if the "number" column is 3501, the encryption of "FER" (3501 in hexadecimal) is applied.

That is, the most significant (last third digit) of the cipher is assigned an alphabet corresponding to the integer portion of the number divided by 26 squared (ie, 676). The middle digit (lower second digit) of the cipher is assigned an alphabet corresponding to the integer part of the value obtained by dividing the remainder of the above division by the first power of 26 (ie, 26). The least significant digit (lowest first digit) of the cipher is assigned the alphabet corresponding to the remainder of the division. This is equivalent to converting from decimal to hexadecimal.

Note that the number of digits of the encryption is not limited to three. In addition, three-digit codes that have meaning as words (e.g. CPU, RAM, USB, NOX, etc.) should not be registered as reserved words with their corresponding numbers in order to prevent misunderstandings and mistranslations. It is good as

Also, here, the cipher is obtained by converting the record number into a hexadecimal number, but the cipher may be determined randomly or may be determined according to another rule.

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

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

The cipher may be expressed in 26-decimal notation following the usual hexadecimal notation (using 0-9 and A-F), and may be expressed using 0-9 and A-P. In addition, lowercase letters of the alphabet and symbols (!”#$%&, etc.) may be used. The cipher may be a random character string that does not overlap among a plurality of ciphers. Symbols or combinations of alphabets and symbols are desirable, and pictograms such as emoticons may be used as encryption.

As in the present embodiment, if the cipher is AAA, AAB, AAC, . For example, 26*26*26=17576 words can be registered in the database. If the number of digits is insufficient, the number of digits of the cipher can be increased to register in the database as many records as the power of 26 (combination of words, ciphers, and translated words).

As in the present embodiment, the cipher is expressed in hexadecimal numbers where A to Z are 0 to 25, respectively. In addition, in neural machine translation, since a string of alphabets and symbols is recognized as one word, there is an advantage that machine translation can be performed while leaving the cipher as one word in the original text.

That is, when translating using neural machine translation, strings of alphabets and symbols are output as they are without translations (although there are exceptions).

For example, if the translation source text of "Electric vehicle 100 includes engine 101." is replaced with the table of FIG. 3, the ciphertext of "AAA 100 includes AAB 101." is generated. This sentence is translated into the sentence "AAA 100 includes AAB 101." by neural machine translation. After the neural machine translation is performed, the table in Fig. 3 is used to convert the cipher in it into the translation of the original word to obtain the translated sentence of "electric vehicle 100 includes engine 101." can be done.

It is possible that the cipher is converted into a plural form, for example, the original text of "plural AAAs" is translated into "a plural of AAAs". Also in this case, by replacing the code part of "AAA" of "AAAs" with the English word corresponding to the word (for example, "book"), we can obtain the translation of "books" with "s" at the end of the word. can be done.

However, with this method, words such as "bus", "leaf", and "city" (words that do not form the correct plural form by simply adding an "s") are misspelled. For this reason, prior art spell-checking routines must be run after translation to ensure that these words are spelled correctly. Also, for example, prepare a table that associates ``buss'' with ``buses'', ``leafs'' with ``leaves'', and ``city'' with ``cities'' (a table that associates misspellings with correct spellings). Batch conversion may also be performed.

Also, depending on the learning result of the neural network, the cipher in the input original text may differ from the corresponding cipher in the translated text (a type of mistranslation).

In order to prevent this, when performing neural machine translation of the translation source text, it is advisable to enclose the encrypted part with characters that indicate that it is a special string (insert special characters before and after the encrypted part). . For example, "AAA 100 includes AAB 101." By inserting space (blank) characters before and after the cipher part and sending it to the neural network, the cipher is converted into another cipher. (a type of mistranslation) can be prevented.

In addition to characters that indicate spaces, it is also effective to enclose encrypted parts in the document before translation with square brackets (""). Similarly, quotation characters ('', "", ””, '', etc.), parentheses, double parentheses, double brackets, square brackets, double square brackets, curly braces, tortoiseshell brackets, double It is also valid to enclose in hexagonal brackets, angle brackets, double angle brackets, gume, or corner brackets. As a result, neural machine translation can be performed in a state indicating that the enclosed part is a part (encryption) 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 a cipher is enclosed in brackets and machine translated as "AAA" 100, it will be translated to "AAA" 100. The double quotes are not in the original text and were added during the above processing, so they are deleted after machine translation and become AAA 100. Deletion is possible using regular expressions.

If you use characters other than the alphabet or symbols such as kanji or hiragana for encryption, there is a problem that it will be translated into English and it will not be possible to decrypt it. "shallow" (English translation of "shallow"), and meaningless character strings such as "aai" are translated to "Aai"). Therefore, it is desirable that the cipher is a string of alphabets and symbols.

Also, when the table of FIG. 3 is used, the first letter of one sentence in the translated English sentence may be a lowercase letter (in English, the sentence should start with a capital letter). It is better to capitalize the beginning of sentences in the later spell check for this as well. For example, there is a process of retrieving alphabets at the beginning of a sentence (or characters after a period) using a regular expression and making them all capital letters.

The data in the database of FIG. 3 may be registered by purchasing and obtaining 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 according to the first embodiment of the present invention. FIG. 5 is a flowchart following FIG.

The processing shown in this flowchart is executed by reading computer programs recorded in the storage device 111 onto the RAM 109 and sequentially executing them by the CPU 101 . The program may be stored in a compiled machine language executable format, a format in which the source code is sequentially executed by an interpreter, or a format described in an intermediate language.

Here, it is assumed that word processing software (such as Microsoft's WORD and open source OpenOffice) is used and the program shown in the flowchart of FIG. 4 is executed using its macros (such as VBA: Visual Basic For Applications). do. The database may be a dedicated database or a CSV file. Here, the database is created using spreadsheet software such as Microsoft's EXCEL, and is accessed using word processing software VBA. and The types of word processing software and programming language used are not limited in the implementation of the invention.

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

The document file may be a text file described in plain text format, a word processing software document file, or an HTML file. Alternatively, the text contained in the image may be a file converted to text by OCR. An image file may be input, and a file may be created by converting the character portion into text data using the OCR function in step S101. In addition, the format of the file is not limited to a specific one as long as it can describe sentences. Documents may be received from other computers using a file transfer protocol or e-mail software, or may be input from storage such as 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 a patent specification describing an electric vehicle is to be translated, and in the document file,

"[0023]

An electric vehicle 100 includes an engine 101 and a display unit 102 , and the engine 101 is connected to signal input means 103 and the display unit 102 . If the input of signal input means 103 is high, display 102 displays a warning. ” was written. The translation processing in the present embodiment will be described by taking the translation of this sentence as an example. [0023] is the paragraph number in the document.

In step S103, preprocessing is performed on the document to be translated. This aims to:

(1) Clarify the units of sentences to be transmitted to the other company's computer resource 200 at one time (breaks between sentences).

(2) To reduce mistranslation, shorten the length of one sentence processed at one time in the other company's computer resource 200 .

(3) Clarify the part where the meaning in one sentence is divided.

For example, in this embodiment, the unit of sentences to be transmitted to the other company's computer resource 200 at one time is from the beginning of the sentence to the full stop (.). Assuming that the paragraph number is also a unit of sentences, in the preprocessing, the paragraph number part is searched using a regular expression and a full stop is added after it (in the above example, a full stop is added after [0023]). .

Also, compound sentences (sentences composed of two or more coordinating clauses (a set of words including a combination of a subject and a predicate)) are less likely to be mistranslated if they are broken down into short sentences. Therefore, in the preprocessing, compound sentences are decomposed into two or more sentences (in the above example, "the electric vehicle 100 includes an engine 101 and a display unit 102, and the engine 101 is connected to the signal input means 103 and the display unit 102. is converted into two sentences: "Electric vehicle 100 includes engine 101 and display unit 102. Engine 101 is connected to signal input means 103 and display unit 102.").

Compound sentences consisting of adverbial clauses and main clauses indicating condition, time, cause, reason, etc. are processed as one sentence rather than being converted into two sentences because the adverbial clause and the main clause are related. more accurate translation is possible. On the other hand, there is a trade-off in that a longer sentence is more likely to be mistranslated. In the present embodiment, an adverbial clause and a main clause are treated as one processing unit, but by inserting a linefeed code between them, machine translation is performed as if they are separate clauses (compound sentences are Machine translation is performed after inserting a line feed code between the constituent clauses). This makes it possible to make the translation more accurate. (In the above example, the linefeed code is inserted after "If the input of the signal input means 103 is high, the display unit 102 displays a warning.").

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

By the processing in step S103,

"[0023]

An electric vehicle 100 includes an engine 101 and a display unit 102 , and the engine 101 is connected to signal input means 103 and the display unit 102 . If the input of signal input means 103 is high, display 102 displays a warning. ” sentence is

"[0023].

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

Display unit 102 displays a warning. ” is converted to the sentence. (A newline is inserted after "if high").

In step S105, the database in FIG. 3 is used to convert the words in the sentence to be translated after the processing in step S103 into parallel translation words.

This process reads records in order from the record number 0 in the database in FIG. This is the process of replacing with the bilingual word described in the record. It is also possible to replace the words with the ciphers described in the record here, but since the readability for humans in the later check is reduced, here the words are replaced with parallel words whose meaning is understandable to humans (translators). is desirable.

In the example of FIG. 3, first, "electric vehicle" with record number 0 is set as the search word in the document to be translated, and "electric vehicle" is set as the word to replace it (the word after replacement). By the processing of the CPU, the word "electric vehicle" (target word) is searched from the beginning of the document to be translated, and if it exists, it is replaced with "electric vehicle". Before replacement, a dialog box such as "Do you want to replace ○○ with △△?" may be Further, full-text search and batch replacement (replacement of all target words in a document) may be mechanically performed without obtaining the user's consent. When the processing of record number 0 is completed, it is determined whether or not there is a next record, and the same processing is performed up to the last record.

In addition, it is desirable to insert half-width or full-width spaces (blank marks) before and after the replacement word in order to improve readability and to clarify that it is a word break at the time of machine translation. Also, the word of the replacement word may be enclosed in brackets ("") to make it clear that it is a fixed word. Quotation characters ('', "", ””, '', etc.), parentheses, double parentheses, double brackets, square brackets, double square brackets, curly braces, tortoiseshell brackets, double tortoiseshell brackets, Angle brackets, double angle brackets, square brackets, corner brackets, or other symbols may be used. In this way, the enclosed part may be surrounded by a symbol indicating that it is different from other parts.

If data is not yet registered in the database (Fig. 3) (such as at the beginning of system introduction), the process of step S105 in Fig. 4 is passed (it is not executed and the text to be translated does not change). .

Assuming that the data in Fig. 3 is registered in the database,

"[0023].

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

Display unit 102 displays a warning. ” sentence is

"[0023].

An electric vehicle 100 has an engine 101 and a display 102 . Engine 101 is connected to signal input unit 103 and display 102 . If the input of signal input unit 103 is high,

The display 102 displays warnings. " is converted to the sentence. (Here, half-width spaces (blank marks) are inserted before and after the replaced word to improve readability.)

In step S107, the user visually confirms the document created by the processing up to step S105, and if necessary, edits the sentence, registers new words and codes, and collectively converts the newly registered words in the document. conduct.

This is done with the aim of:

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

(2) Correct the words erroneously converted in the batch conversion in step S105.

(3) Batch conversion of words to be replaced with bilingual words and words to be encrypted, and registration in the database.

That is, the user visually confirms the document created by the processing up to step S105, and converts the remaining long sentences into short sentences. For example, a compound sentence consisting of two or more sentences is divided by a period to make two or more sentences having the same meaning.

In addition, the user corrects the erroneously converted words in the batch conversion in step S105. Due to differences in technical fields, there are cases where even the same word should be translated into different words. In such cases, the user converts the word with batch conversion. In addition, if necessary, update the database with the correct words, and prepare multiple databases that are used according to technical fields.

Since the words were not registered in the database, for the words that were not converted in step S105 (especially the words before the reference numerals), in this step S107 the user newly registers new words, parallel translation words and ciphers, and , batch conversion of newly registered words in the document.

FIG. 6 is a flow chart showing word and code registration processing of a computer program included in the translation system according to the first embodiment of the present invention. This process is executed, for example, in step S107 of FIG. 4 in response to user input.

Referring to FIG. 6, it is assumed that the user is inputting and editing text in a document at step S203. If the user believes that a new word needs to be entered into the database, the user selects that word in the document at step S205. This is done by moving the cursor to the beginning (or end) of the word to be registered while the document is being displayed in word processing software, and pressing the arrow keys to the end (or beginning) of the word to be registered while holding down the shift key. When pressed, the word to be registered is highlighted (or changed in color or underlined so that it can be distinguished from other words). A word may be selected by dragging the word with a mouse.

When a 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 translation of the selected word. By inputting the parallel translation of the selected word in this field, the user confirms the parallel translation of the selected word. In addition, a database different from that shown in FIG. 3 or a trained neural network is used to display candidate translations of the selected word in a dialog box, and by receiving a selection from the user, the translation of the selected word is confirmed. You can do it.

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

In step S209, a full text search is performed for the same word as the word selected in step S205, and in step S211 it is replaced with the translated word determined in step S207. In this replacement, as in step S105 in FIG. 4, a dialog box such as "Do you want to replace ○○ with △△?" A decision may be made as to whether or not to replace the word. Further, full-text search and batch replacement (replacement of all target words in a document) may be mechanically performed without obtaining the user's consent.

In step S213, the bottom line of the database (FIG. 3) where data registration is performed is searched. The line immediately below that line is set as the registration line of the current word, and a cipher is created from the record number in step S215. In step S217, numbers, words, codes, parallel translation words, etc. are newly registered in the row immediately below the bottom row of the database. As a result, the database, which is translation assets, is updated.

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

Note that the event for starting the word registration process in step S205 is the pressing of a shortcut key for word registration, the selection of a display indicating word registration from the displayed menu, or ("word registration , etc.), but it is desirable to use a shortcut key (because words can be registered without using a mouse or voice input). It is desirable that the shortcut key is assigned to, for example, pressing both the control key and a specific key so as not to interfere with character input.

If the processing in step S107 of FIG. 4 has ended, it is checked in step S109 whether or not the document generated by the processing up to step S107 contains a portion that is difficult to machine translate. Specifically, this is as follows.

(1) If the length of one sentence (from the beginning of the sentence to the full stop) is greater than or equal to a predetermined length, the sentence is judged to be difficult to machine-translate, and a warning is output.

(2) If the length from the beginning of a sentence or line feed code to the next line feed code or period is greater than or equal to a predetermined length, output a warning that the text is difficult to machine translate. .

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

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

(5) If there is no correspondence between the subject and the predicate in one sentence, output a warning.

(6) Output a warning if there is more than one subject and one predicate in one sentence.

In step S109, if the check fails, the user continues the processing in step S107. Note that the processing in step S109 may be omitted.

In step S111, the document obtained by the processes up to step S109 is subjected to collective conversion of English words into ciphers.

This process is similar to the process in step S105. Records are read in order from the record numbered 0 in the database in FIG. and replace the found word with the cipher described in the same record in FIG.

This will

"[0023].

An electric vehicle 100 has an engine 101 and a display 102 . Engine 101 is connected to signal input unit 103 and display 102 . If the input of signal input unit 103 is high,

The display 102 displays warnings. ” sentence is

"[0023].

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,

FER 102 displays a warning. ” is converted to

In this way, the words are replaced with the cipher described in the record, so that a document that is incomprehensible to humans is generated (first encryption).

When reading records in order from the record numbered 0 in the database in FIG.

For example, "display unit" and "information display unit" may be registered as bilingual words in the database. When there is the word "information display unit" in the sentence, if only the "display unit" part in it is searched first and encrypted, then "information" remains unencrypted. Therefore, by searching in the order of the length of the translated words (the order in which more words are included), the longer words can be encrypted first. Search terms may be picked up from the database in the order of length of bilingual words (the order containing more words) without sorting the database data.

In step S113, it may be searched whether there is a word (particularly a noun) before the reference sign that cannot be converted into a code. That is, if there is a word before the reference numeral that cannot be converted into a cipher, the user is prompted to register it in the database in step S115 as in step S109. The processing in steps S113 and S115 ensures that the word before the reference sign is encrypted. Note that the processing of steps S113 and S115 may be omitted.

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

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

"[0023].

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,

FER 102 displays a warning. If the sentence is

The first sentence becomes "[0023]."

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

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

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

FER 102 displays a warning. ”. These sentences are processed in order from the first sentence, and are processed in step S121. Although the second and fourth sentences are one sentence to be processed, they contain a line feed code.

In step S121, it is determined whether the 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. In this way, if there is only one sentence consisting of symbols, numbers, and alphabets, there is no need to translate it. Alternatively, the addition may be performed after processing such as converting full-width characters to half-width characters. That is, if YES in step S121, the one sentence is processed as necessary in step S123 and added to the end of the queue. As a result, communication costs and translation costs can be reduced. Also, it is possible to prevent mistranslation from occurring.

If NO in step S121, the sentence is transmitted to the external computer resource 200 of another company by HTTPS communication in step S125. Encryption performed in HTTPS communication is the second encryption.

In step S127, the translation result of the one sentence is received from the external computer resource 200 of another company. This communication is also encrypted by HTTPS, and 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) may be inserted after the period. That is, the received documents may be combined with spaces inserted between them. This is a process for improving the readability of the translated text.

In step S129, it is determined whether or not the process has been completed up to the last sentence (the n-th sentence), and if NO, the process from step S121 (processing of the next sentence) is repeated. This completes the processing from the first sentence to the nth sentence.

If YES in step S129, the queue contains translation results. For example, in the example above,

"[0023].

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,

FER 102 displays a warning. ” sentence 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,

FER 102 displays warning." and it will be included in the queue. In this way, since the translation is also encrypted, even if a person other than the company obtains the translation, they cannot know the contents of the translation.

Here, in step S131, a process of converting the ciphers in the queue into English words is performed.

This process is similar to the process in step S105. Records are read in order from the record numbered 0 in the database in FIG. This is a process of replacing the discovered cipher with the translated word described in the same record in FIG.

This allows 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 sentence 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 a sentence of By performing this conversion in-house, the confidentiality of the document is ensured.

Note that in this conversion, it is desirable to set the full-text search flag so that the cipher does not match part of a word. For example, it prevents "ABS" in the word ABSTRACT from being erroneously converted as a cipher. For example, if Microsoft's VBA is used, such processing can be realized by performing a full-text search with the MatchWholeWord property set to True.

Also, in order to prevent words with meanings such as CPU and RAM from being translated as codes, it is desirable to prevent words with such meanings from being registered in the database as codes (as reserved words).

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

(1) Processing for deleting unnecessary spaces. For example, when two or more spaces are inserted at a position that is not after a period, processing to merge them into one. Also, a process of deleting spaces between numbers and alphabets in a reference code consisting of alphabets and numbers (such as 100A).

(2) A process of deleting a line feed code existing after a comma or a line feed code other than a line feed code existing after a period. For example, in the above example, the processing is to delete the line feed code after "When an input of signal input unit 103 is high,".

(3) the process of converting incorrect plural forms to correct forms; For example, the process of converting boxes to boxes.

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

(5) If a special code is added and translated, such as enclosing the code in brackets, the code is deleted. For example, if the cipher of AAA100 is machine-translated with brackets such as "AAA" 100, a non-original double quotation such as "AAA" 100 is added, so processing to delete this.

(6) Processing to correct articles. An indefinite article is added to the word that appears for the first time, and a definite article is added to the second and subsequent occurrences.

(7) Processing for deleting double-byte full-width punctuation. (This is the process of deleting the punctuation marks in [0023] above.)

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

In step S137, a spell checker in word processing software is used to correct any word or syntax errors, and at the same time correct any plural form errors. The user corrects and edits the errors to complete the final translation.

In this manner, machine translation using external computer resources can be performed while maintaining the confidentiality of both the original translation and the translation.

Also, by registering words that include superscripts, subscripts, italics, bold letters, and underlined letters in the database, they are machine-translated in a coded state. Since the cipher can later be converted back into words containing superscript, subscript, italic, bold, and underlined characters, machine translation will render superscript, subscript, italic, bold, and underline. Mistranslation of words containing the attached characters is prevented. For example, a word may be registered in the database as "SiO ₂ ", its corresponding code as "ABC", and a parallel word as "SiO ₂ ". good. If the word before translation and the translated word are the same, only the word and the code may be registered in the database, and the translated word may not be registered. Words may be registered in the database as full-width characters and bilingual words as half-width characters (or vice versa).

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

There are several translations of "electric vehicle" such as "electric vehicle", "electrical vehicle", "electric-powered vehicle", "electronic vehicle", "battery car", and "battery vehicle". When machine translation is performed using a neural network, it is not possible to know which translated word is appropriate depending on the context, resulting in fluctuations in word expressions after translation. In addition, the appropriate translation often differs depending on the field of the document to be translated. In this embodiment, words are converted into standardized translations in advance before machine translation by a neural network. Therefore, fluctuations in expression are eliminated.

In addition, when machine translation from Japanese to English is performed using full-width characters as encryption, they become half-width characters after machine translation. Therefore, in the conversion in step S131, it is necessary to convert the full text in consideration of this. If the cipher is set to half-width characters from the beginning, such a problem will not occur.

[Second embodiment]

FIG. 7 is a flow chart showing word and code registration processing of a computer program included in the translation system according to the second embodiment of the present invention.

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

That is, it is also possible to perform the processing after step S203 (same as the processing after step S203 in FIG. 6) after opening the document with the word processing software in step S201 of FIG.

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

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

The processing shown in such a flowchart is executed by reading computer programs recorded in the storage device 111 onto the RAM 109 and sequentially executing them by the CPU 101 . The program may be stored in a compiled machine language executable format, a format in which the source code is sequentially executed by an interpreter, or a format described in an intermediate language.

Word processing software (such as WORD from Microsoft Corporation and OpenOffice published as an open source system) is used, and the programs of the flowcharts of FIGS. good. The database may be a dedicated database or a CSV file. Here, the database is created using spreadsheet software such as Microsoft's EXCEL, and is accessed using word processing software VBA. and

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

The document file may be a text file described in plain text format, a word processing software document file, or an HTML file. Alternatively, the text contained in the image may be a file converted to text by OCR. An image file may be input, and a file may be created by converting the character portion into text data using the OCR function in step S201. In addition, the format of the file is not limited to a specific one as long as it can describe sentences. Documents may be received from an external computer using a file transfer protocol or e-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 a patent specification describing an electric vehicle is to be translated, and in the document file,

"The electric vehicle 100 comprises an engine 101 and a display section 102, and the engine 101 is connected to a signal input means 103 and a display section 102. When the input of the signal input means 103 is high, the display section 102 issues a warning. It is assumed that the sentence "display."

In step S203 of FIG. 4, the user checks the content of the opened document and corrects it if necessary. In addition, sentences are added (input) as necessary.

In step S205 of FIG. 4, the user selects a word to be registered in the database by viewing the text. This moves the cursor to the beginning or end of a word, and by pressing SHIFT+arrow key or dragging with the mouse, the target word is highlighted (or displayed with a different character and/or background color than the rest). It is. Words are registered mainly for words existing before reference signs, but words without reference signs may also be registered. Reference numerals are numerals for referring to numerals and characters used together with lead lines to indicate constituent elements, which are described in the drawings. For example, "100" in "electric vehicle 100" is a reference number, and the word before the reference number means "electric vehicle".

Words may be selected by human judgment (key or mouse input) as described above, or, for example, a word existing before a reference sign may be automatically detected and selected. In addition, words may be automatically detected and selected regardless of the presence or absence of reference signs.

If the registration button (or the shortcut key for registration) displayed on the screen is pressed while a word is selected, in step S207, CPU 101 registers the translation of the word selected in step S205. Accept input. This may be one that displays a dialog box on the screen and accepts input from the user using the keyboard or voice, or uses a UI (user interface) that displays multiple word candidates from a dictionary and allows the user to select a parallel translation from among them. You can This determines the correspondence between the words before translation and the words after translation.

In step S209, CPU 101 performs a full-text search of the document to be translated, searches for pre-translation words determined in the processing up to step S207, and converts them into corresponding post-translation words in step S211. A full-text search may be performed without asking the user for confirmation, and each time a word is searched, a dialog box such as "Do you want to replace with (translated word)? (YES/NO)" is displayed. may be displayed on the screen to allow the user to select whether to replace the word. As a result, the words in the entire sentence, including the word selected in step S205, are collectively converted into the translated word determined in step S207. For example, if the word "electric vehicle" is associated with the word "electric vehicle", the word "electric vehicle" in the document is collectively converted to "electric vehicle".

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

The number obtained in step S215 is converted into a 26-decimal number to create a cipher for the target word. In step S217, the code is recorded as one record along with the word before translation and the word after translation in the row below the bottom row of registered information in the database. The information registered in the database will be an asset as a translation dictionary to be used in the future.

After the processing of step S217, the processing from step S203 is repeated. Further, when the registration button is not pressed in step S205, the processing from step S203 is repeated.

When the full text has been processed, the user saves the document. This is original text data in which a part of the document is translated, and may be subjected to subsequent neural machine translation.

Words registered in the database in the processing of FIGS. 6 and 7 are mainly words written before reference numerals. For example, if the original sentence to be translated is "Electric vehicle 100 comprises engine 101 and display unit 102, engine 101 is connected to signal input means 103 and display unit 102. When the input of signal input means 103 is high, , the display unit 102 displays a warning.”, “electric vehicle”, “engine”, “display unit”, and “signal input means” are to be registered.

Words such as "input" and "warning" without reference numerals may also be registered in the database 207 of FIG.

Furthermore, at least some of the words with reference signs may be excluded from registration. That is, at least some of the "electric vehicle", "engine", "display unit", and "signal input means" may not be registered.

After registration in step S213, the words in the original text may be sorted in order of length. Longer words are sorted near number 0 (upper side) and shorter words are sorted higher numbered side (lower side).

By the above processing, "the electric vehicle 100 has an engine 101 and a display section 102, and the engine 101 is connected to the signal input means 103 and the display section 102. When the input of the signal input means 103 is high, the display Unit 102 displays a warning."

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

[Third embodiment]

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

The processing shown in this flowchart is executed by reading computer programs recorded in the storage device 111 onto the RAM 109 and sequentially executing them by the CPU 101 . The program may be stored in a compiled machine language executable format, a format in which the source code is sequentially executed by an interpreter, or a format described 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 the text after the conversion is performed by the processing in FIG. Here, "electric vehicle 100 comprises engine 101 and display 102, engine 101 is connected to signal input unit 103 and display 102. If the input of signal input unit 103 is high, display 102 gives a warning. will be displayed.” is processed.

In step S103, the CPU 101 searches the input document for parts that are difficult to machine translate. For example, if you are translating from Japanese, divide the document into multiple sentences by delimiting them with commas such as "." (that is, whether the length of each sentence is equal to or less than a predetermined threshold value) (as the threshold value, for example, a value of about 80 to 120 characters is selected according to the performance of the other company's computer resource 200) ).

If the length of one sentence is long, the machine translation tends to be inaccurate, so the length of the sentence is checked in advance and long sentences cannot be translated. For a long sentence, for example, if it is a compound sentence, a sentence such as "A is B and C is D." A conversion/editing process is performed to convert the sentences into sentences and shorten each sentence. Such conversion may be performed automatically, or a sentence judged to be long may be presented to the user and the user may be prompted to manually correct it. In the case of automatic conversion, it is desirable to present the text after automatic conversion (and the text before conversion if necessary) to the user so that the user can check whether the conversion has been performed correctly.

As a check for parts that have difficulty in machine translation, there is a subject in one sentence (for example, does it contain a phrase equivalent to "○○ is" or "○○ is"), You can also check whether the predicates are compatible. For more accurate machine translation without mistranslation, for example, a translatable sentence may have only one subject and one predicate.

These checks may be performed by matching sentence patterns using regular expressions. A check may be performed by morphological analysis. Also, for example, using a trained recurrent neural network (RNN), the text to be checked is input, and the output is a text that is not difficult to machine translate, or the output is a signal indicating whether there is difficulty or not. It is also possible to prompt the user to correct in some cases.

In step S105, it is checked whether the text can be machine-translated, and if there is a difficult part, the processing in step S103 is repeated.

If it is determined in step S105 that machine translation is possible, the text is preprocessed in step S107. This is a process that facilitates machine translation so that there are no mistakes in machine translation. For example, a compound sentence has the structure "When A is B, C will do D." This is the process of embedding a code that serves as a mark in the . The code is a line feed code, space, delimiter character, etc., and the code is selected so that the delimiter is easy for machines to understand without affecting the translation. For example, a compound sentence having a structure such as "A is B and C is D." You can embed the code. The process of step S107 reduces the unit of sentences to be processed in machine translation, thereby preventing the selection of wrong subject-predicate combinations and the occurrence of other mistranslations in machine translation.

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

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

"Electric vehicle 100 comprises engine 101 and display 102. Engine 101 is connected to signal input unit 103 and display 102. When the input of signal input unit 103 is high, (newline code)

The display 102 displays warnings. "

(Single sentences are created in step S103, and line feed codes are inserted between sentences to create complex sentences in step S107).

In step S109, the database in the storage device 111 is referred to, and the words (especially nouns) written before the reference characters in the document to be translated are converted into encryption (first encryption). The database compares the original text, the encrypted string, and the translated string.

By collectively converting the words before the reference numerals into ciphers in step S109,

"Electric vehicle 100 comprises engine 101 and display 102. Engine 101 is connected to signal input unit 103 and display 102. When the input of signal input unit 103 is high, (newline code)

The display 102 displays warnings. ” sentence is

"AAA 100 comprises AAB 101 and FER 102. AAB 101 is connected to AAC 103 and FER 102. When the input of AAC 103 is high,

FER 102 displays a warning. "

is replaced by the ciphertext of

In such a ciphertext, since hexadecimal digits such as "AAA" are processed as words, it can be machine-translated by an external third-party computer resource 200 without revealing the true meaning to the outside.

The processing in step S109 is performed as follows.

(1) From the beginning of the document, search for a constituent part of a sentence consisting of a combination of words and reference signs.

(2) searching the database for the words of the searched component;

(3) Get the corresponding cipher from the database.

(4) Replace the words of the component parts retrieved in (2) above with the ciphers obtained in (3) above.

(5) Searching for the constituent part of the sentence that appears next in the sentence and consists of a combination of a word and a reference code.

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

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

(1) Read the first word registered in the database

(2) All read words are searched from the beginning of the document and replaced with the corresponding ciphers registered in the database (whole text replacement).

(3) Read out the next word registered in the database and repeat the process of (2) above. When 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 is sent to the other company's computer resource 200 that has undergone neural network training by HTTPS transmission in order from the first sentence. .

More specifically, the first sentence of the data before translation (for example, the sentence up to the full stop of ".") is sent to the external computer resource 200 by HTTPS in step S119. This transmission is encrypted (second encryption). The external computer resource 200 of another company translates the received data into the target language and transmits it to the computer 100 in response to the request.

In step S121, the computer 100 obtains one-sentence data after translation from the external computer resource 200 of another company as an HTTPS response.

In step S123, it is determined whether translation has been completed up to the last sentence, and if NO, the next sentence is processed in step S119. If translation has been completed up to the last sentence, the process proceeds to step S125.

In step S125, post-processing of the translated document is performed. Here, the translations for each sentence obtained in steps S119 to S123 are combined into one document. In addition, since the translated text obtained by the first encryption in steps S109 to S113 contains a cipher, a process of collectively converting the cipher into corresponding translated words is performed. This is done, for example, as follows.

(1) Read out the first cipher (such as "AAA") registered in the database.

(2) All read out ciphertexts are searched from the beginning of the document and replaced with the corresponding translated words in the database (whole text replacement).

(3) Read out the next cipher registered in the database, and repeat the processing from (2) above. When all the registered ciphers in the database have been processed, the process ends.

In step S127, spell checking, such as checking for plural words, is performed. The user makes a final check of the translation and corrects or edits it if there is a problem.

By performing the processing 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, (linefeed code)

FER 102 displays alert."

From the article 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, (linefeed code)

display 102 displays alert."

sentence is obtained.

By performing the same processes as steps S135 and S137 in FIG. 5 in steps S125 and S127, the same effect as in 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 the translation system according to the fourth embodiment of the present invention.

The processing in this flowchart is the processing in steps S111 and S113 added in addition to the processing in FIG.

After the processing in step S109, in step S111, it is determined whether there is a word before the reference sign that could not be converted because it is not registered in the database. This may be determined by using a regular expression to determine whether or not there is a constituent portion of a sentence composed of a combination of words and reference signs, or by the user visually confirming.

If YES in step S111, the process proceeds to step S113. In step S113, the words that could not be converted in step S111, their codes, and their translations are set and registered in the database. For example, after highlighting the words that could not be converted by black-and-white reversed display, display a dialog box containing the word "Please enter the translation of XX" and a word input field to accept input from the user. may For example, highlight the word "electric vehicle" in a document, display a dialog box containing the word "Please enter the translation of 'electric vehicle'" and a word input field, and prompt the user to enter the word "electric vehicle". It accepts input. A word, its code, and its translation are set and registered in a database. In addition, it is desirable to automatically create the cipher.

The processing in step S113 is repeated until NO is determined in step S111.

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

[others]

By combining or replacing the multiple embodiments described above and the elements included therein (part of configuration, part of processing), new other embodiments can be created.

When performing full-text batch conversion (a process of converting a word into a parallel translation word or a parallel translation word into a code), it is desirable to read the words to be retrieved from the database in order from the longest word. For example, if there are two records in the database, one that converts "detection means" to "detection means" and the other that converts "information detection means" to "information detection means," the long word "information After batch conversion of "detection means" to "information detection means", shorter words "detection means" are batch converted to "detection means". As a result, it is possible to prevent only the "detection means" portion of the "information detection means" in the document from being converted, thereby improving the translation efficiency. Similarly, when converting words into codes, by reading the words from the database in order from the longest words, it is possible to prevent only a part of a set of words having a series of meanings from being converted into codes.

When words are read in ascending order of record number and batch conversion is performed, the records may be sorted according to the number of characters in the registered words.

In the case of a patent specification, there is an item "Description of Reference Symbols" at the end of the document, in which important reference characters and the constituent elements (words) indicated by them are generally explained there. It is also possible to encrypt only the words described in the item of explanation of the code. Prior to machine translation, a warning may be issued if the words listed in the code description item are not encrypted in the document.

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

In documents such as patent specifications that refer to reference signs in the drawings, the noun words that precede the reference signs in the document are of particular significance. By replacing at least part of the words in front of the reference numerals with codes (or parallel translations), the present invention can be highly effective. Of course, at least some of the words other than the words in front of the reference numerals in the document may be replaced with codes (or parallel translations). In addition, at least some of the words other than the words in the document that precede the reference signs are encrypted (or parallel words) without replacing the words that precede the reference signs in the document with the code (or parallel words). It may be replaced.

Of the set of noun words and reference codes that appear in a document (a set consisting of a noun word and its immediately following reference code), only the noun word parts are stored in the database as targets of batch conversion. It is desirable to register and not include the reference code in the target of batch conversion. In this way, the database can be used to process the same words with different reference signs (to make better use of the database when translating other documents in the future). can be done).

The invention is also applicable to documents that do not contain reference signs. In this case, machine translation is performed after replacing at least some of the words in the document with codes (or parallel translations).

The batch conversion process using the database in step S105 of FIG. 4 and step S109 of FIG. 8 may be performed multiple times in succession. For example, by recording a record in the database in which the user converts "detection means" to "detection means" in advance, "information detection means" in the translated document may be converted to "information detection means". . This is the case when part of a word is converted. In this case, the user selects "information detection means" as a word and records "information detection means" as a parallel translation in the database as a new record. When translating the following document using the database, if the word "information detection means" exists in the document, the word is first converted to "information detection means" in the batch conversion using the database. be. If the search term "information detection means" appears after it in the search order of the database, "information detection means" can be successfully converted to "information detection means". However, if the search order in the database is such that the "information detection means" precedes the "detection means", the "information detection means" will be It is not converted to "information detection means".

Therefore, by repeating the batch conversion process using the database twice (or more), batch conversion is performed again. As a result, even a word such as "information detection means" that includes a partial translation in the original text is converted into a correct word such as "information detection means."

In addition, in the embodiment, an example of performing machine translation using a neural network was given, but statistical machine translation, phrase-based, rule-based machine translation, artificial intelligence machine translation, deep learning machine translation, etc. Also, the present invention is not limited to machine translation using neural networks.

The method of transmitting/receiving information to/from the external computer resource 200 of another company may be performed via a WEB page. Information may be sent and received by an automated approach to browser operations. Alternatively, information may be transmitted and received using an API provided by a company that operates external computer resources of other companies.

The post-processing may include deleting unnecessary spaces in sentences, unifying words, converting uppercase letters to lowercase letters in sentences, spell checking, correcting plural words, and the like.

[Effects of Embodiment]

According to the above embodiment, both the translation original text and the translated text are encrypted in two steps, ie, the first encryption and the second encryption. The confidentiality of the document is maintained for a third party by both encryptions. In addition, the confidentiality of the document is maintained by the first encryption with respect to the translation service provider. As a result, it is possible to prevent information from being leaked to the outside while using an external translation service.

Confidentiality of information can be ensured even if computer resources that transmit information are computer resources that are used by an unspecified number of people (or by a plurality of persons or others). It should be noted that "unspecified number of users" means that the site (server) is open to the public and can be accessed from various locations.

In addition, the conventional neural machine translation has a problem that the information included in the input sentence cannot be accurately translated (there is a possibility that translation omissions occur). In addition, duplicated translations were sometimes output. In this embodiment, the original text can be compressed (shorter and simpler) by encryption. Also, the sentence structure can be simplified (shorter and simpler). As a result, even if neural machine translation is adopted, the information included in the input sentence can be translated in just the right amount, with the effect that omissions in translation are less likely to occur. In addition, there is an effect that duplicate translations are less likely to be output.

Furthermore, there is an effect that learning data (bilingual dictionary data, word order data, etc.) that has been used so far can be easily used as data to be registered in the database.

Furthermore, in conventional neural machine translation, word embedding enables flexible translation, but on the other hand, by selecting completely different words (sometimes meaningless character strings), translation sentences with completely different meanings can be created. There were (especially words with a low frequency of occurrence, proper nouns with a low frequency of occurrence, etc.). In addition, in the case of long sentences and complicated sentences, translations that do not make sense are sometimes created. In the present embodiment, words are collectively converted into bilingual words or codes in advance, and then machine-translated (preferably machine-translated with brackets added). This prevents the machine translation from translating the original word into a parallel word with a completely different meaning.

The processing in the above-described embodiments may be performed by software or by using hardware circuits. It is also possible to provide a program for executing the processes in the above-described embodiments, or record the program on a recording medium such as a CD-ROM, flexible disk, hard disk, ROM, RAM, memory card, etc. and provide it to the user. You may decide to A program is executed by a computer such as a CPU. Alternatively, the program may be downloaded to the device via a communication line such as the Internet.

The above-described embodiments should be considered as illustrative in all respects and not restrictive. The scope of the present invention is indicated by the scope of the claims rather than the above description, and is intended to include all modifications within the scope and meaning equivalent to the scope of the claims.

100 computer 101 CPU
103 communication unit 109 RAM
111 storage device 200 other company's computer resource 203 search/replace unit 205 document editing/word registration unit 207 database 209 memory 211 display/output unit 213 input unit 400 Internet 500 own computer resource

Claims (1)

A translation device that communicates with a computer that inputs text written in a first language via the Internet, translates it into text written in a second language, and transmits the translated text via the Internet. ,
the sentence written in the first language includes a set of a word that is a noun and a reference sign that immediately follows;
The translation device is
a first replacement means for replacing a word portion in a set of a noun word and a reference sign immediately following it with a character string corresponding to the word;
a transmitting means for transmitting the text replaced by the first replacing means to the computer via the Internet;
receiving means for receiving, from the computer, a sentence written in the second language after translation through the Internet ;
The first replacement means is
reading means for reading a word and its replacement character string in one set from a plurality of sets registered in a database, which associate a word with its replacement character string;
a process of retrieving one word read by the reading means from sentences written in the first language and, if found, replacing the word with a replacement character string associated with it; , and a first processing means for repeatedly performing within a sentence written in the first language,
After all the replacements in the text written in the first language are performed by the first processing means, the reading means reads the words in the next set and their replacement character strings, and the first processing is performed. process by means,
selection means for allowing a user to select a portion in a sentence written in the first language;
confirmation means for confirming the translation of the selected portion;
A second replacement for performing a full-text search from the text written in the first language for the same portion as the text written in the first language selected by the user, and replacing it with the confirmed parallel translation. means and
A translation device comprising registration means for registering the selected portion and its parallel translation in a database .

Images