Translation support device and program

Abstract

To provide a translation support device and a program that can obtain natural translated sentences using designated words as designated translation words.SOLUTION: A translation support device for acquiring a sentence generation processing result using generation model information in a machine-learned state to generate a sentence based on an instruction in a language acquires bilingual information including an original sentence and a translated sentence obtained by machine translation, refers to a terminology database that holds designated words in a language of the original sentence and corresponding designated translation words in a language of the translation sentence in association with each other to generate an instruction that the translated sentence of the bilingual information should be a sentence in which the designated word included in the original sentence of the bilingual information is translated with the designated translation word held in the terminology database in association with the designated word, and acquires a sentence that is the sentence generation processing result obtained using the generation model information as modified translated sentence information.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a translation support device and a program. [Background technology]

[0002] In recent years, technology that integrates terminology databases, translation memories, and machine translation software has become widely used to assist translation work. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] China Patent Application Publication No. 108984540 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional machine translation software translates source text using a machine learning model, and does not allow for the setting of a specific translation for a pre-specified word (designated word). Therefore, when there is a designated word for which a designated translation is specified for translation, the translator must edit the machine-translated text himself.

[0005] Patent Document 1 discloses a technology for automatically searching and matching special terms in the translation results during the translation process. However, the technology disclosed in Patent Document 1 has a problem in that if the translation corresponding to the designated word is not the designated translation, the translation is replaced with the designated translation as is, resulting in an incorrect translation if the replacement is not appropriate.

[0006] The present invention has been made in consideration of the above-mentioned situation, and one of its objectives is to provide a translation support device and program which, by referring to bilingual information obtained by machine translation or the like, can obtain a natural translation in which the designated word is used as the designated translation, while utilizing the translation result when the translation result contains a translation that corresponds to the designated word but is not the designated translation. [Means for solving the problem]

[0007] One aspect of the present invention that solves the problems of the conventional examples is a translation support device that includes: generation information acquisition means for acquiring a sentence generation process result using generation model information that has been machine-trained to generate sentences based on an instruction in a language; acquisition means for acquiring bilingual information including an original sentence and a translation obtained by machine translation; instruction generation means for referencing a terminology database that associates and holds a designated word in the original language with a corresponding designated translation word in the translation language, and generating an instruction that the translation of the bilingual information should be a sentence in which a designated word included in the original sentence of the bilingual information is translated with a designated translation word held in the terminology database in association with the designated word; and modified translation acquisition means for providing the generated instruction to the generation information acquisition means and acquiring the sentence that is the result of the sentence generation process acquired by the generation information acquisition means as modified translation information, and the modified translation information acquired by the modified translation acquisition means is provided to a predetermined translation work support process. Effect of the Invention

[0008] According to the present invention, it is possible to obtain a natural translation in which a designated word is used as a designated translation while utilizing the results of an already performed translation. [Brief description of the drawings]

[0009] [Figure 1] 1 is a configuration block diagram illustrating an example of a translation support device according to an embodiment of the present invention. [Diagram 2] 1 is a functional block diagram illustrating an example of a translation support device according to an embodiment of the present invention. [Diagram 3] 1 is an explanatory diagram illustrating an example of bilingual information used by a translation support device according to an embodiment of the present invention; [Figure 4] FIG. 4 is a flowchart illustrating an example of the operation of the translation support device according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] An embodiment of the present invention will be described with reference to the drawings. A translation support device 1 according to the embodiment of the present invention is realized by a general computer system, as shown in Fig. 1, and includes, for example, a control unit 11, a storage unit 12, an operation unit 13, a display unit 14, and a communication unit 15.

[0011] The control unit 11 is a program control device such as a CPU, and operates according to a program stored in the storage unit 12. In one example of the present embodiment, the control unit 11 acquires bilingual information (for example, so-called XLIFF (XML Localization Interchange File Format) data) including an original text and a translation obtained by machine translation, and stores it in the storage unit 12. The control unit 11 also refers to a term database that stores designated words in the original language and corresponding designated translation words in the translation language in association with each other, and generates an instruction to translate the translation of the bilingual information into a sentence in which a designated word included in the original text of the bilingual information is translated with a designated translation word stored in the term database in association with the designated word.

[0012] This instruction is written in any language, and is used as input to generate a sentence, and is used in a process of acquiring a sentence generation process result using generative model information that has been machine-learned to output the generated sentence. Specifically, this process can be realized by inputting the instruction to ChatGPT (registered trademark) of Open AI, Inc., and acquiring the sentence generated by ChatGPT. In this example, the control unit 11 that sends an instruction to ChatGPT and performs a process of acquiring the generated sentence corresponds to the generation information acquisition means of the present invention.

[0013] The control unit 11 outputs the acquired sentence resulting from the sentence generation process as corrected translation information, for example, to provide it for the translation work support process. The operation of the control unit 11 will be described later, including modified examples.

[0014] The storage unit 12 is a memory device, a disk device, or the like, and holds a program executed by the control unit 11. This program may be provided by being stored in a computer-readable and non-transitory recording medium, and may be copied to the storage unit 12. The storage unit 12 also operates as a work memory for the control unit 11.

[0015] In this embodiment, the storage unit 12 holds a term database that holds designated terms in the original language and corresponding designated translated terms in the translation language in association with each other.

[0016] The operation unit 13 includes a mouse, a keyboard, etc. The operation unit 13 accepts instructions from a user of the translation support device 1, and outputs information indicating the contents of the accepted instructions to the control unit 11. The display unit 14 is a display device or the like, and displays information according to instructions input from the control unit 11.

[0017] The communication unit 15 is a network interface or the like, and transmits and receives information to and from an external server R (for example, a server providing the above-mentioned ChatGPT service) in accordance with instructions input from the control unit 11.

[0018] Next, a description will be given of an example of the operation of the control unit 11. In one example of the present embodiment, the control unit 11 executes a program stored in the storage unit 12, and is configured to include a creation information acquisition unit 20, a parallel translation information acquisition unit 30, a match determination unit 31, an instruction generation unit 32, a corrected translation acquisition unit 33, and an output processing unit 34, as exemplified in FIG.

[0019] The generation information acquisition unit 20 executes a process (processing by a generation AI) of generating a sentence using an instruction written in a predetermined language as an input, and acquiring a sentence generation process result using generative model information that has been machine-learned to output the generated sentence. As an example, as already mentioned, the generation information acquisition unit 20 sends an instruction generated by the instruction generation unit 32, which will be described later, to the ChatGPT server of Open AI, Inc., and acquires the sentence generated by the ChatGPT server. This process can be realized by a well-known method of calling the API (Application Program Interface) of ChatGPT.

[0020] However, the generation information acquisition unit 20 of this embodiment is not limited to this example, and the control unit 11 itself may take instructions written in a specified language as input, generate a sentence, and execute a sentence generation process using generative model information that has been machine-trained to output the generated sentence.

[0021] The bilingual information acquisition unit 30 acquires bilingual information including machine-translated original text and translation. In one example of the present embodiment, the control unit 11 accepts original text information written in a first language (e.g., Japanese), divides the original text information into predetermined segment units, generates a translation in a second language for each segment unit obtained by the division, and records the corresponding segment unit as an original text in association with the generated translation in the storage unit 12 (this recording may be performed in a widely known format such as XLIFF), thereby generating bilingual information.

[0022] In this example, the machine translation is a machine-learned system that takes an original text written in a first language as input and outputs a translation written in a second language (a language different from the first language) that corresponds to the original text, using a machine learning model different from that which generates a sentence equivalent to a translation in response to an instruction (a so-called prompt). This type of machine translation is widely known as NMT (Neural Machine Translation), etc.

[0023] In one example of this embodiment, the bilingual information, as shown in FIG. 3, associates, for each segment unit included in the original text information, identification information N (which may be a unique sequential number for each sentence), the original text (S), and a translation (T) which is the result of a machine translation of the original text.

[0024] In this example, the bilingual information acquisition unit 30 acquires the bilingual information generated by the control unit 11 and stored in the storage unit 12 as the processing target.

[0025] In another example of this embodiment, the parallel translation information acquiring unit 30 acquires, via a network, parallel translation information (which may also be in XLIFF format in this example) obtained by machine translation processing in another computer.

[0026] The match determination unit 31 performs the following process by referring to the bilingual information acquired by the bilingual information acquisition unit 30. For each original sentence in segment units included in the bilingual information, the match determination unit 31 determines whether or not the original sentence contains a designated word stored in the terminology database. When the match determination unit 31 determines that the original sentence contains a designated word stored in the terminology database, it checks whether or not a translation (called a corresponding translation) included in the bilingual information corresponding to the original sentence contains a designated translation word that corresponds to the designated word that it has determined to contain.

[0027] If the match determination unit 31 determines that the corresponding translation does not include a designated translation corresponding to the designated word found in the original text, it outputs identification information associated with the corresponding translation, the found designated word, and the corresponding designated translation to the instruction generation unit 32 to generate an instruction.

[0028] That is, the match determination unit 31 performs the following for each of the corresponding original sentences and translations (corresponding translations) included in the bilingual information: (1) The source text contains a designated term stored in the terminology database, and (2) The corresponding translation does not contain a specified translation that corresponds to the specified term. If the above conditions (1) and (2) are not satisfied, the match determination unit 31 repeats the above determination process for the original text of the next segment unit.

[0029] The instruction generating unit 32 receives the identification information, the found designated word, and the corresponding designated translation from the match determining unit 31, and generates an instruction to output to the generated information acquiring unit 20. This instruction generating unit 32 generates an instruction to change the translation recorded in the bilingual information in association with the input identification information to a sentence in which the input reference word (a designated word included in the original text of the bilingual information associated with the identification information) is translated with a designated translation held in the terminology database in association with the designated word.

[0030] As a specific example, the instruction generating unit 32: "Replace [corresponding translation] with a sentence in which [specified word] is translated by [specified translation]" For example, if the corresponding machine-translated Japanese-English translation is: Translation support devices display machine-translated text and its corresponding original text associated with each other in response to user instructions. (Original text: In response to a user's instruction, the translation support device displays the machine-translated text and the corresponding original text in association with each other.) In this case, the term database contains Designated word: Translation support device Translation assistance apparatus When recorded as “Replace the following English text and appropriately add an indefinite article or a definite article the word “translation assistance apparatus” is translated as “translation assistance apparatus”: “Translation support devices display machine-translated text and its corresponding original text associated with each other in response to user instructions.” Generate instructions as follows.

[0031] The modified translation acquisition unit 33 outputs the instruction generated by the instruction generation unit 32 to the generation information acquisition unit 20. Then, the modified translation acquisition unit 33 associates the sentence obtained as a result of the sentence generation process output by the generation information acquisition unit 20 with the identification information output by the match determination unit 31, and acquires the following modified translation information related to the translation (corresponding translation) included in the bilingual information: A translation assistance apparatus displays machine-translated text and its corresponding original text associated with each other in response to user instructions.

[0032] Note that, generally, articles are not included in the term database that uses the specified term as the specified translation term. This is because, conventionally, regarding how to handle singular and plural expressions in the entire translated text, whether to use the definite article or the indefinite article in the singular case, etc., the translator appropriately judged on the translation editor of the translation support tool and adjusted the translation term and the surrounding translated text. In contrast, for the corrected text using the generation model (the corrected text by the generative AI), the generative AI makes judgments such as whether to attach an article to the specified translation term, which of the indefinite article or the definite article is appropriate if an article is to be attached, and which of the singular or plural is appropriate. As a result, it is possible to generate a corrected text that cannot be achieved by the conventional simple translation term replacement.

[0033] As another specific example, in the case of English-Japanese translation, when the corresponding machine translation of the text is: He thought he could use it very well as a cradle, some day when he had children of his own. (Original text: He thought he could use it very well as a cradle, some day when he had children of his own.) When it is such a thing, in the term database, Specified term: use it very well Specified translation term: うまく使える when it is recorded, the instruction generation unit 32 will Replace the following Japanese text and show only the corrected text wherein the word “use it very well” is translated as “うまく使える”.: 「彼は、いつの日か自分の子供ができたときに、それをゆりかごとして非常によく使うことができると思った。」 generate an instruction to the effect.

[0034] The modified translation acquisition unit 33 then outputs the instruction generated by the instruction generation unit 32 to the generation information acquisition unit 20. The modified translation acquisition unit 33 associates the sentence obtained as a result of the sentence generation process output by the generation information acquisition unit 20 with the identification information output by the match determination unit 31 and acquires the following modified translation information related to the translation (corresponding translation) included in the bilingual information: "He thought that one day when he had children of his own, he could put it to good use as a cradle."

[0035] Previously, if a verb or adverb was in the terminology database with the designated word as the designated translation, it was not possible to mechanically replace the translation. For this reason, in the past, translators would make appropriate judgments on the bilingual editor of a translation support tool when adjusting translations that included verbs or adverbs, and would then adjust the translation and the surrounding translations. On the other hand, in modified sentences using generative AI, even if the designated translation is a verb or adverb, a generative model is used to determine which parts should be replaced, and even when replacing a verb or adverb, modified results are generated based on the results of machine learning for the surrounding words. This makes it possible to generate modified sentences that include translations that consist of verbs or adverbs, which could not be achieved by conventional mechanical replacement.

[0036] The output processing unit 34 receives the corrected translation information and the corresponding identification information, and provides the corrected translation information for a predetermined translation work support process. Specifically, the output processing unit 34 replaces the translation associated with the identification information input from the corrected translation acquisition unit 33 in the bilingual information acquired by the bilingual information acquisition unit 30 with the corrected translation information, thereby updating the bilingual information.

[0037] When the translator displays this updated bilingual information, instead of the translation obtained by machine translation, revised translation information obtained by rewriting the machine-translated translation through processing using the generative model information is displayed on the display unit 14. At this time, the translation support device 1 may also display the corresponding original text.

[0038] [Operation] This embodiment has the above configuration and operates as follows. In the explanation below, a case where Japanese text (original text information in Japanese) as a first language is translated into translation text information in English as a second language is taken as an example. However, this is only an example, and the language is not limited to Japanese, English, etc.

[0039] In the following example, the process of taking instructions written in a specific language as input, generating sentences, and obtaining the results of the sentence generation process using generative model information that has been machine-trained to output the generated sentences is assumed to be performed by an external server, and in this case, the translation support device 1 of this embodiment sends instructions to the above server via an API to make a call, and receives and obtains the results of the sentence generation process from the server.

[0040] Furthermore, the terminology database stores information in advance that corresponds a word expressed in a first language that should be replaced with a specific translation in a second language as a designated word, and the translation to be replaced is also stored as a designated translation. In other words, as already exemplified, this terminology database stores: {Specified word: translation assistance apparatus, specified translation: translation assistance apparatus}, {Specified term: Generative AI, Specified translation: Generative AI} … Entries such as:

[0041] A user of the translation support device 1 inputs original text information to the translation support device 1. The translation support device 1 divides the original text information into a predetermined segment unit (for example, a paragraph unit or a sentence unit; in the following, a paragraph unit is taken as an example, and a segment unit is replaced with a word "paragraph"), and generates a translation in the second language, English, for each paragraph obtained by the division. This machine translation process may be performed by a general NMT. ​​The translation support device 1 then associates unique identification information (here, paragraph number; N) issued for each paragraph, a part of the original text information included in the paragraph (original text; S(N)), and a correspondingly generated translation (T(N)) with each other, and records them as bilingual information in the XLIFF format (FIG. 3). Hereinafter, the original text corresponding to the identification information N is represented as S(N), and the translation corresponding to the identification information N is represented as T(N).

[0042] Next, the translation support device 1 reads out the obtained bilingual information and starts the process illustrated in Fig. 4. As illustrated in Fig. 4, the translation support device 1 repeats the following process while sequentially selecting identification information included in the bilingual information as a processing target (S11).

[0043] That is, the translation support device 1 refers to the original text S(N) associated with the identification information N that is the subject of processing, and determines whether or not the designated word is present and satisfies the condition (called the designated word match condition) that the original text S(N) contains a designated word W stored in the terminology database and the corresponding translated text T(N) does not contain a designated translation word TW that corresponds to the designated word W (match determination; S12).

[0044] If it is determined that there is a designated word that satisfies the designated word matching condition (S12: Yes), the translation support device 1 generates an instruction to an external server to perform sentence generation processing using the generation model information, using the identification information (N) of the paragraph being processed, the designated word W found in step S12, and the corresponding designated translation word TW (S13).

[0045] As in the example already given, in paragraph (n) with bilingual information, Original text S(n): In response to an instruction from a user, the translation support device displays a machine-translated sentence and the corresponding original text in association with each other. In response to TranslationT(n):Translation support devices display machine-translated text and its corresponding original text associated with each other in response to user instructions. In this case, the original text S(n) contains the designated word W "translation assistance apparatus" contained in the example database, but the translation T(n) does not contain the corresponding designated translation TW "A translation assistance apparatus." Therefore, the translation support device 1 determines in step S12 that the designated word agreement condition is satisfied, and in step S13, using the translation T(n) associated with the identification information n of the paragraph, issues the instruction: “Replace the following English text:”Translation support devices display machine-translated text and its corresponding original text associated with each other in response to user instructions.” However, this instruction is just one example, and other expressions may be used as long as they are intended to instruct the translation text T(N) to be rewritten along with the meaning that "in the translation text T(N), the designated word W should be translated into the designated translation word TW."

[0046] The translation support device 1 sends the instruction generated in step S13 to an external server that performs sentence generation processing using the generation model information to cause the server to perform the sentence generation processing, and acquires the result of the sentence generation processing as corrected translation information (S14).

[0047] Then, the translation support device 1 replaces the translation TW(N) included in the translation information associated with the identification information N of the paragraph being processed with the corrected translation information acquired in step S14, thereby updating the translation information (S15).

[0048] If it is determined in step S12 that the designated word satisfying the designated word matching condition is not present in the original text or the corresponding translation (S12: No), the translation support device 1 skips and does not execute the processes in steps S13 to S15.

[0049] The translation support device 1 repeatedly executes the processes of steps S12 to S15 for each paragraph included in the translation information, and when the processes for all paragraphs are completed, outputs and displays the updated translation information on the display unit 14.

[0050] This allows a user who is a translator to use the results of machine translation while also being able to refer to natural translations in which the designated word is used as the designated translation.

[0051] [When multiple designated words are found] In addition, in the above processing, an example has been described in which processing is performed using a pair of designated words W and corresponding designated translation words TW. However, there is not necessarily a single pair of designated words W and designated translation words TW that satisfy the designated word agreement condition in step S12.

[0052] For example, if there is an example of a term in the above database, Original text: "The computer-assisted translation device compares the machine-translated English sentences with the English sentences generated by the generative AI." Against Supported translations:Translation support devices compare machine-translated English text with English text generated by AI. If it is, {Specified term: translation assistance apparatus, Specified translation term: A translation assistance apparatus}, {Specified term: Generative AI, Specified translation: Generative AI} It is determined that the two pairs above satisfy the specified word matching condition.

[0053] In this case, the translation support device 1: (a) Initially, for a designated word that satisfies a designated word matching condition, an instruction to modify the corresponding translation is generated while the designated word is being translated with a designated translation, and a sentence generation processing result is obtained using the generative model information. Thereafter, for each designated word that satisfies the designated word matching condition, an instruction to further modify the previously obtained sentence generation processing result is generated while the designated word is being translated with a designated translation, and a sentence generation processing result is obtained using the generative model information. This process may be repeated.

[0054] In addition, the translation support device 1 (b) It is also possible to generate an instruction to correct the corresponding translation, assuming that each of a plurality of designated words should be translated with a corresponding designated translation, and obtain a sentence generation processing result using the generative model information.

[0055] In the former case (a), in the above example, the translation support device 1 first “Replace the following English text:”Translation support devices compare machine-translated English text with English text generated by AI.” therefore the word “translation assistance device” is translated as “A translation assistance apparatus” The following instructions are generated and the sentence generation process results are based on the generative model information: A translation assistance apparatus compares machine-translated English text with English text generated by AI. Then, for the designated word "Generation AI" that satisfies another designated word matching condition, “Replace the following English text:”A translation assistance apparatus compares machine-translated English text with English text generated by AI.” then the word “Generative AI” is translated as “Generative AI” The following instructions are generated and the sentence generation process results are generated using the generative model information: A translation assistance apparatus compares machine-translated English text with English text generated by a generative AI. get.

[0056] In the case of the above process (b), the translation support device 1 “Replace the following English text:”Translation support devices compare machine-translated English text with English text generated by AI.” therefore the word “translation assistance apparatus” is translated as “A translation assistance apparatus”, and the word “Generative AI” is translated as “Generative AI” The following instructions are generated and the sentence generation process results are generated using the generative model information: A translation assistance apparatus compares machine-translated English text with English text generated by a generative AI. get.

[0057] However, in the former (a), in the second or subsequent processing, the designated translation replaced in the preceding processing may be replaced by another translation, and in the latter (b) where the instruction is redundant, a load may be placed on the processing using the generative model information. Therefore, the translation support device 1 may use a combination of the above processing, for example, by performing processing (b) when the number of pairs of the designated word W and the designated translation word TW that satisfy the designated word matching condition is less than a predetermined threshold, and performing processing (a) otherwise.

[0058] [Effects of the embodiment] In the case of translation from Japanese to English, the conventional technology had a problem in that it could not handle how to add articles to the translation of nouns or noun phrases, or how to handle changes in endings of singular and plural, and could only simply replace the specified term with the specified translation. In addition, in the case of translations containing verbs or adverbs, the translation replacement of the conventional technology resulted in an incorrect translation if adjustments to the wording after the translation were required. In the present embodiment, it is possible to appropriately handle the addition of appropriate articles in the case of nouns, changes in endings to singular and plural, and adjustments to the wording in the case of verbs and adverbs.

[0059] Machine translation using neural networks has brought about a huge improvement in accuracy, but general-purpose machine translation engines produce average translations, which is still an issue that cannot be adapted to field-specific terminology. In addition, translations are determined on a sentence-by-sentence basis, so there is a problem of translation inconsistency when viewed as a whole document. To solve this problem, computer-assisted translation tools are used to highlight the relevant terms in the original text and display the translations in the terminology database for segments where the original terms in the terminology database match in the bilingual translation of the original text and the machine translation, and the translator then corrects the translation by making adjustments to articles and singular / plural, or to the translation in the case of verbs and adverbs.

[0060] In the computer-aided translation tool of this embodiment, for segments where original terms from the terminology database match in the bilingual text between the original text and the machine-translated text, the translations from the terminology database are appropriately displayed, eliminating the need for translators to adjust the translations. As a result, the translation correction work, which is the core of post-editing of machine-translated text by translators, becomes more efficient, dramatically improving translation productivity. [Explanation of symbols]

[0061] 1 translation support device, 11 control unit, 12 memory unit, 13 operation unit, 14 display unit, 15 communication unit, 20 generated information acquisition unit, 30 bilingual information acquisition unit, 31 match determination unit, 32 instruction generation unit, 33 corrected translation acquisition unit, 34 output processing unit.

Claims

1. A generation information acquisition means for acquiring a sentence generation process result using generation model information that has been machine-trained to generate a sentence based on an instruction in a language; An acquisition means for acquiring bilingual information including an original text and a translation obtained by machine translation; an instruction generating means for referring to a term database which holds a designated word in an original language and a corresponding designated translation word in a translation language in association with each other, and generating an instruction for translating the translation of the bilingual information into a sentence in which a designated word included in the original text of the bilingual information is translated with a designated translation word held in the term database in association with the designated word; a corrected translation acquisition means for providing the generated instruction to the generation information acquisition means and acquiring, as corrected translation information, a sentence that is a result of the sentence generation process acquired by the generation information acquisition means; the revised translation information acquired by the revised translation acquisition means is used for a predetermined translation work support process.

2. 2. The translation support device according to claim 1, a translation means for receiving source information in a first language, dividing the source information into predetermined segment units, generating a translation in a second language for each segment unit obtained by dividing the source information, and generating bilingual information recorded in association with the generated translation, with the corresponding segment unit as the source text; The translation means generates a translation in a second language from a segment unit in the first language using a machine learning model different from the generative model information.

3. 2. The translation support device according to claim 1, The instruction generating means, for each of the corresponding original sentences and translation sentences included in the parallel translation information, The source text contains a designated term stored in the terminology database, and The translation does not contain the specified translation that corresponds to the specified term. in this case, the translation support device generates an instruction to the effect that the translation of the bilingual information should be a sentence in which the designated word is translated with a designated translation word stored in a term database in association with the designated word.

4. Computer, A generation information acquisition means for acquiring a sentence generation process result using generation model information that has been machine-trained to generate a sentence based on an instruction in a language; An acquisition means for acquiring bilingual information including an original text and a translation obtained by machine translation; an instruction generating means for referring to a term database which holds a designated word in an original language and a corresponding designated translation word in a translation language in association with each other, and generating an instruction for translating the translation of the bilingual information into a sentence in which a designated word included in the original text of the bilingual information is translated with a designated translation word held in the term database in association with the designated word; a corrected translation acquisition means for providing the generated instruction to the generation information acquisition means and acquiring, as corrected translation information, a sentence that is a result of the sentence generation process acquired by the generation information acquisition means; and controls a computer so as to provide the revised translation information acquired by the revised translation acquisition means for a predetermined translation work support process.

Images