Translation support system and translation support method

Abstract

To assist in the translation of a source document in a first language into a translated document in a second language. [Solution] The system comprises a translation acquisition unit that acquires bilingual information including an original text in a first language and a translation of the original text in a second language, and a correction processing unit that provides input including the bilingual information and correction instructions for the translation to a generative model and acquires a response from the generative model, and the correction processing unit repeats inputting the correction instructions and acquiring the response until a repetition termination condition is met.

Description

[Technical Field]

[0001] The present invention relates to a translation support system and a translation support method for supporting the translation of an original document in a first language into a translated document in a second language. [Background technology]

[0002] Traditionally, document translation work required professional translators to read the original text and manually create the translation while appropriately interpreting the context and meaning of terms. This type of work was time-consuming and labor-intensive, relying heavily on the translator's language skills and specialized knowledge. It was also difficult to maintain consistent translation quality, and efficiency was required for large-scale translation work.

[0003] In recent years, advances in artificial intelligence technology have led to the practical application of machine translation models, and machine translation systems that automatically convert source text into translations are becoming widely used. This has made it possible to process the initial stages of translation quickly and efficiently, reducing the burden of translation work. In particular, the technique of "post-editing," in which translations generated by machine translation are manually corrected, is becoming more common, and there is a demand for both high-quality and efficient translation.

[0004] The translation device described in Patent Document 1 discloses a configuration in which a first text written in a first language is converted into a translation in a second language using a neural machine translation model. When multiple different translations for the same source language are output by machine translation, this device generates a unified third text by replacing the other translations with one representative translation. Furthermore, the device is configured to display the representative translation so that it can be distinguished from the other translations, and measures are taken to help post-translation editors easily recognize and correct inconsistencies in the translations.

[0005] Patent Document 2 discloses a translation support device capable of generating natural-looking translations in which designated words are translated with designated translations. The translation support device described in Patent Document 2 acquires bilingual information consisting of original and translated sentences obtained by machine translation, and further associates and references designated words registered in a terminology database with the corresponding designated translations. The translation support device generates instructions for translating designated words included in the original sentence into sentences with designated translations, and performs sentence generation processing using a generative model to obtain corrected translations. This realizes a means for achieving both consistency in terminology and natural translations. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2024-128094 [Patent Document 2] Japanese Patent Publication No. 2025-005833 Summary of the Invention [Problem to be solved by the invention]

[0007] Prior art documents disclose techniques for correcting inconsistencies in terminology in translations using machine translation models or generative models. However, post-editing requires corrections not only to correct inconsistencies in terminology in translations, but also to avoid ambiguity, correct grammatical issues, mistranslations, omissions, and additions. In other words, there is a problem in that these general corrections cannot be made efficiently.

[0008] An object of the present invention is to provide a technique for supporting the task of translating an original document in a first language into a translated document in a second language. [Means for solving the problem]

[0009] [1] A translation support system that supports the task of translating an original document in a first language into a translated document in a second language, comprising: a translation acquisition unit that acquires bilingual information including an original document in the first language and a translation of the original document into the second language; and a correction processing unit that provides input including the bilingual information and correction instructions for the translation to a generative model and acquires a response from the generative model, wherein the correction processing unit repeats inputting the correction instructions and acquiring the response until a repetition termination condition is met. [2] The translation support system described in [1], wherein the condition for ending the repetition includes at least one of the following: the response from the generative model does not require modification; and a predetermined number of responses have been obtained. [3] The translation support system described in [1] or [2], wherein the correction processing unit inputs and obtains responses to the generation model for multiple sentences or multiple segments included in the parallel translation information, and excludes translations that are determined not to require correction based on the previous response from being targets for input of correction instructions from the next time onwards. [4] The translation support system described in any of [1] to [3], wherein the correction processing unit generates multiple subsets each including multiple sentences or multiple segments included in the parallel translation information, provides each of the subsets to a generative model, obtains a response for each of the subsets from the generative model, and, if it is determined based on the previous response that none of the sentences or segments included in the subset require correction, excludes the subset from targets for input of correction instructions from the next time onwards. [5] The translation support system described in any of [1] to [4], wherein the correction processing unit provides the generative model with a processing request to generate a response including the original text, the translation, and a corrected translation obtained by correcting the translation, obtains the response from the generative model, and sets the corrected translation included in the response as the translation for the next processing request. [6] A translation support system according to any one of [1] to [5], wherein the correction instructions include instructions to reflect in the correction of the translation at least one of the quantity, elements, order, and relative positional relationship of tag information contained in the original text without substantially changing it. [7] The translation support system according to any one of [1] to [6], wherein the correction processing unit generates an explanation in a specified language for the correction portion of the translation.

[0010] According to the invention of [1], by repeatedly correcting the translation using a generative model, the accuracy of the translation can be improved and post-editing can be made more efficient. The invention according to [2] makes it possible to make the generative model execute the process repeatedly an appropriate number of times, thereby optimizing the accuracy of the translation and the processing time. The invention of [3] can reduce the amount of sentences processed by the generative model and shorten the translation time. [4] The invention of the present invention processes the target sentence or segment in subset units, thereby reducing the number of requests and responses to the generative model and enabling parallel processing of subsets, thereby shortening processing time. The invention of [5] makes it possible to appropriately structure the response of the generative model and realize effective iteration of corrections. The invention of [6] makes it possible to have a generative model infer whether tag information in the original text is properly reflected in the translation, and correct it as necessary. The invention described in [7] makes it possible to obtain an explanation of the corrections. [Effects of the Invention]

[0011] According to the present invention, it is possible to provide a technique for supporting the task of translating an original document in a first language into a translated document in a second language. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a system configuration diagram according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram illustrating a hardware configuration according to the present embodiment. [Figure 3] FIG. 2 is a processing sequence diagram according to the embodiment. [Figure 4] FIG. 2 is a data structure diagram according to the embodiment. [Figure 5] 3 is a processing flowchart according to the present embodiment. [Figure 6] 10 is a flowchart illustrating a different process according to the present embodiment. [Figure 7] 10 is a graph showing the relationship between the number of corrections and accuracy according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] A translation support system and a translation support method according to an embodiment of the present invention will be described below with reference to the accompanying drawings. Note that the embodiment shown below is an example of the present invention, and the present invention is not limited to the following embodiment, and various configurations can be adopted.

[0014] In this embodiment, the configuration, operation, etc. of a translation support system and a translation support device are described, but a translation support method, a computer program, and a program recording medium on which the program is recorded, each having a similar configuration, also achieve the same effects. For example, by using a program recording medium, the program can be installed on a computer. The series of processes according to this embodiment described below are provided as a computer-executable program, and can be provided via a non-transitory computer-readable recording medium such as a CD-ROM or a flexible disk, or even via a communication line.

[0015] The translation support system is composed of a computer device. The computer device has a processor such as a CPU (Central Processing Unit) and a memory. The computer device can function as a translation support device by executing a translation support program stored in the memory using an arithmetic unit. The translation support method is realized by processing the computer device including the translation support device.

[0016] In this embodiment, translation refers to converting a sentence in a first language into a sentence in a second language. The first language and the second language include various languages ​​used as natural languages. For example, the first language is Japanese and the second language is English, but these are not limited thereto.

[0017] In this embodiment, translation support refers to support for post-editing, a part of translation work. Post-editing refers to the process of making necessary corrections to a translation obtained by machine translation of an original text. This correction includes clarifying ambiguous expressions, correcting grammatical errors, correcting mistranslations or missing translations, deleting unnecessary words, adding phrases, and standardizing notation.

[0018] In this embodiment, the original document refers to a document file containing sentences in a first language. The document file may be any file format that contains at least text. The original text refers to the sentences that make up the original document. A translated document refers to a document file containing a sentence in a first language that is included in an original document and translated into a second language. A translated sentence refers to a sentence that constitutes a translated document. In this specification, the terms translated document and translated sentence are defined to include both pre-edited and post-edited translated documents and sentences, regardless of whether they have been revised by post-editing.

[0019] In this case, the "pre-correction translation" refers to, for example, a sentence that has been translated from the original text by machine translation, and the "corrected translation" refers to a sentence that has been corrected one or more times by post-editing to the pre-correction translation. When distinguishing the number of corrections in the corrected translation, it is written as an "n-th-corrected translation." This is a natural number equal to or greater than 1 and indicates the number of corrections. In this embodiment, the n-th-corrected translation refers to a sentence that has been corrected from the n-1-th-corrected translation. When n=1, the n-1-th-corrected translation refers to the pre-correction translation. Similarly, pre-correction translated documents, corrected translated documents, and n-th-corrected translated documents are defined.

[0020] In this embodiment, the bilingual information is information including an original text and its corresponding translation. The bilingual information manages the original text (source) and the translation (target) as a pair in one file. In this embodiment, the bilingual information holds tag information consisting of font information and document structure information of the original text and the translation. Specifically, the bilingual information can be in the form of an XLIFF (XML Localization Interchange File Format) file or the like, but the format is not limited to this.

[0021] Fig. 1 shows a system configuration diagram of a translation support system 1 according to this embodiment. As shown in Fig. 1, the translation support system 1 includes a translation support device 2, a user terminal 3, and a generation device 4. The translation support device 2, the user terminal 3, and the generation device 4 are all connected via a communication network NW.

[0022] The translation support device 2 is a computer device installed by the operator of the translation support system 1. The translation support device 2 includes, as functional components, a document acquisition unit 21 that accepts input of an original document or a document pair including an original document and a translated document, a bilingual acquisition unit 22 that acquires bilingual information including the original text and its translation, a correction processing unit 23 that corrects the translation, and an output unit 24 that outputs the correction result including the corrected translation.

[0023] The user terminal 3 is a terminal device operated by a user. The user can perform operations such as specifying an original document, document pair, and bilingual information via the user terminal 3, and obtain the correction results. The user terminal 3 has, as functional components, an input unit that accepts operation instructions and input of various data, and an output unit that displays on the screen and outputs audio. Although only one user terminal 3 is shown in FIG. 1, multiple user terminals 3 may be present.

[0024] The generation device 4 is a computer device installed by the provider of the translation support system 1 or an external provider. The generation device 4 receives a processing request from the translation support device 2 and generates a response to the request. In this embodiment, the processing request includes bilingual information and instructions to correct the translation. The generation device 4 returns the generated response to the translation support device 2. Although only one generation device 4 is shown in FIG. 1, multiple generation devices 4 may be installed.

[0025] The generation device 4 can generate a response by inputting data included in the processing request into a generative model. A known model trained by machine learning can be used as the generative model. In this embodiment, large-scale language models (LLMs) are used as the generative model. LLMs are a type of natural language processing model trained by machine learning using large-scale natural language data. LLMs receive, for example, text or various files containing questions or instructions as input, and output text as a response to the input. Note that LLMs may have a multimodal configuration and can handle data input and output in various formats, such as images, videos, and audio. In this embodiment, the generative model is not limited to LLMs, and other machine learning models or algorithms capable of generating a response to an input may be used.

[0026] In this embodiment, an example is described in which the generative model is stored in the generation device 4, but a configuration in which the translation support device 2 is equipped with the generative model may also be adopted. Furthermore, the generation device 4 may be installed inside the translation support device 2, or installed outside the translation support device 2 and connected to that device via a wired or wireless connection.

[0027] In this embodiment, the translation support system 1 is described as including a translation support device 2 and a user terminal 3, but is not limited to this. For example, the translation support device 2 may be configured as a terminal device operated by a user, and may be configured to accept input of source documents, document pairs, and bilingual information. In this case, the translation support system 1 may be configured not to include the user terminal 3.

[0028] FIG. 2(a) shows a hardware configuration diagram of the translation support device 2. The translation support device 2 includes a processor 201, a memory 202, and a communication interface 203 as its hardware configuration. In this embodiment, the translation support device 2 can be a computer device such as a server or a personal computer. Note that the translation support device 2 may be configured with multiple computer devices, and is not limited to the configuration shown in FIG. 2(a) as long as the above-described functional components (21-23) can be realized as a whole. The generation device 4 has the same hardware configuration as the translation support device 2.

[0029] The processor 201 is configured with one or more processors such as a CPU, and controls the overall processing of the translation support device 2 by executing a translation support program, an OS (Operating System), and other applications. The memory 202 is configured with one or more transient and / or non-transient storage devices such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), a flash memory, or a RAM (Random Access Memory), and stores the translation support program and various data. The communication interface 203 controls communication with the communication network NW, and realizes data communication with the user terminal 3, the generation device 4, and the storage unit 5. The processor 201 executes the translation support program, causing the computer to function as the translation support device 2 and execute a translation support method.

[0030] 2(b) shows a hardware configuration diagram of the user terminal 3. The user terminal 3 includes, as its hardware configuration, a processor 301, a memory 302, a communication device 303, an input interface 304, and an output device 305. In this embodiment, the user terminal 3 may be a smartphone, a personal computer, a tablet terminal, or the like.

[0031] The processor 301 is composed of one or more processors such as a CPU, and controls the overall processing of the user terminal 3 by executing an OS and other applications. The memory 302 is an HDD, SSD, flash memory, RAM, etc., and stores a browser application and various data. The communication device 303 controls communication with the communication network NW and realizes data communication with the translation support device 2, etc. The input interface 304 accepts operation instructions from an operator and is composed of at least one or more devices such as a touch panel, mouse, keyboard, etc. The output device 305 is composed of devices such as a display that displays the processing results by the processor 301 and a speaker that outputs audio.

[0032] FIG. 3 shows a processing sequence in the translation support system 1.

[0033] In step S101, the user terminal 3 transmits an original document or a document pair including an original document and a translation document to the translation support device 2. The translation support device 2 acquires bilingual information based on the acquired document. In step S101, the user terminal 3 may be configured to transmit bilingual information including the original text and the translation to the translation support device 2.

[0034] In step S102, the translation support device 2 transmits to the generation device 4 a processing request including the bilingual information and instructions to correct the translation. In step S103, the generation device 4 transmits the response generated in response to the acquired processing request to the translation support device 2. This response may include at least a corrected translation obtained by correcting the translation.

[0035] In this embodiment, the translation support device 2 repeatedly executes the processes of steps S102 and S103 until a repetition termination condition, which will be described in detail later, is satisfied. If the response in step S103 includes a corrected translation, the corrected translation is set as the translation for the next processing request.

[0036] When the repetition end condition is satisfied, in step S104, the translation support device 2 generates a correction result based on the response from the generation device 4 and transmits it to the user terminal 3. The correction result is based on the corrected translation or the corrected translation document.

[0037] Figure 4 shows examples of the configuration of various information handled by the translation support system 1. Figure 4(a) shows an example of the configuration of bilingual information. Figures 4(b) and 4(c) show examples of the configuration of a response (response information) from a generative model specified by a processing request.

[0038] As shown in FIG. 4(a), bilingual information includes segments, original texts, and translations. Bilingual information can be extracted from original documents and translations of the original documents, such as machine translation. A segment is information for identifying pairs of original texts and translations as a predetermined unit. A segment is set as a unit of one or more sentences, and each unit is assigned a sequential number or the like. For example, segment "02" includes the Japanese original text "It's nice weather, isn't it?" and the English translation "It's a nice day." Segments are set, for example, to divide a sentence into single sentences based on a sentence-ending symbol. However, this is not limited to this, and segments may be divided to include multiple sentences based on a predetermined sentence structure or other criteria. Furthermore, some documents may contain sentences written without using a sentence-ending symbol or sentences with ambiguous syntax. Therefore, even when segments are set to divide into single sentences, they may contain multiple sentences or fragments less than a sentence.

[0039] In this embodiment, the original text and the translation text include tag information that is made up of font information, document structure information, etc. in addition to text information. For example, the segment "01" is the original Japanese text " <g id=" / "1 / ”"> Hello.< / g> The text information in this original sentence is "Hello." and the tag information is " <g id=" / "1 / ”"> 」、「< / g> " Tag information is used to add font information and the like to text information. Font information includes, but is not limited to, typeface, character size, bold, italic, underline, strikethrough, color, and the like. Note that translations can also include font information and tag information.

[0040] As shown in Figure 4(b), the response information includes a segment, an original text, a translation, a corrected translation, and an explanation. The response information is information generated by a generative model by providing the model with input (processing request) including bilingual information and instructions for correcting the translation of the bilingual information. In addition to the segment, original text, and translation of the bilingual information, the response information also includes a newly generated corrected translation and an explanation.

[0041] The corrected translation is a sentence obtained by correcting the translation. If there are no corrections from the translation, the corrected translation may be the same as the translation, or may be left blank. The explanation is an explanatory text about the corrected translation in the specified language (Japanese in the illustrated example). The explanation is generated as an explanatory text indicating what corrections were made between the translation and the corrected translation. The explanation may be generated as an explanatory text for each point of correction, such as mistranslation, omission, addition, ambiguity, grammar problem, etc. If there are no corrections from the translation, an explanation such as "no corrections" is generated to indicate that no corrections are necessary.

[0042] In the example of Figure 4(b), segment "01" is the translated sentence "Hello." <g id=" / "1 / ”">< / g> " is the revised translation " <g id=" / "1 / ”"> Hello.< / g> " This correction is the original text " <g id=" / "1 / ”"> Hello.< / g>", the position of the tag information in the translation has been corrected based on the position of the tag information in the original text. The response information generates an explanation for this correction: "The tag position has been corrected. The translation is unchanged." Note that corrections are not limited to tag information, and it is of course possible for text information to be corrected as well.

[0043] In this embodiment, a processing request to the generative model and a response thereto are repeatedly executed. Therefore, response information is generated multiple times. The corrected translation of the response information is set as the "translation" for the next processing request. Figure 4(c) shows an example of response information resulting from a second correction in response to the response information resulting from the first correction in Figure 4(b). In Figure 4(c), the translation of segment "01" is set to a translation after one correction. Furthermore, the translation after two corrections has not been corrected from the translation after one correction, and the explanation "no correction" is generated. In this embodiment, translations determined to not require correction can be excluded from the input targets for correction instructions from the next time onwards. This reduces the processing in the generative model and shortens processing time.

[0044] 5 is a process flowchart of the translation support device 2 according to an embodiment. In FIG. 5, step S201 indicates a document acquisition step, step S202 indicates a translation acquisition step, steps S203 to S207 indicate correction processing steps, and step S208 indicates an output step.

[0045] In step S201, the document acquisition unit 21 accepts input of an original document in a first language or a document pair including an original document in a first language and a translated document obtained by translating the original document into a second language. In this embodiment, the document acquisition unit 21 acquires the original document and generates an XLIFF file including the original document (source). The document acquisition unit 21 can generate a translated document (target) by applying machine translation to the XLIFF file. Note that in step S201, the document acquisition unit 21 may be configured to acquire the original document and its translated document, and may also be configured to directly acquire a document pair, as appropriate.

[0046] In step S202, the bilingual acquisition unit 22 acquires bilingual information including an original text in a first language and a translation of the original text in a second language, and stores the acquired information in the memory 202. The bilingual acquisition unit 22 can acquire the bilingual information by extracting the original text and the translation based on the information acquired in step S201. The bilingual acquisition unit 22 may acquire the bilingual information by accepting input of the bilingual information via the user terminal 3. In this case, the document acquisition step (S201) can be omitted.

[0047] In step S203, the correction processing unit 23 transmits a processing request including bilingual information and correction instructions for the translation to the generation device 4. The correction instructions are configured as prompts that refer to the bilingual information and generate response information including at least a corrected translation. In this embodiment, the correction instructions are instructions to generate a corrected translation for the translation, focusing on mistranslations, omissions, additions, ambiguities, grammatical issues, etc. In addition, instructions to generate a corrected translation without substantially changing at least one of the number, elements, order, and relative positional relationship of tag information between the original text and the corrected translation may be further adopted. Note that "without substantially changing" means that at least one of the number, elements, order, and relative positional relationship of tag information is maintained between the original text and the corrected translation. However, even if the order of tag information, etc., is changed due to grammatical differences between languages, this embodiment accepts this as "without substantially changing" as long as the change does not impair the semantic and structural consistency of the translation content.

[0048] In this embodiment, the processing request can include prompts related to the output format, constraints, and correction instructions. The output format includes, for example, an instruction to output each data item of the response information shown in FIG. 4 in JSON format, and an instruction to specify the first language and the second language. The constraints set predetermined rules to improve the accuracy of the response.

[0049] In step S203, the bilingual information included in the processing request may be in units of one or more segments. If the number of segments included in the processing request exceeds a predetermined number, the processing request may be sent to the generation device 4 in multiple batches.

[0050] In step S204, the modification processing unit 23 obtains a response to the processing request from the generation device 4.

[0051] In step S205, the correction processing unit 23 refers to at least one of the corrected translation and the explanation included in the response, and determines whether the translation corresponding thereto does not require correction. In this embodiment, for example, if the translation included in the response information (corrected translation) matches the immediately preceding uncorrected translation, the correction processing unit 23 can determine that the translation does not require correction. Note that the correction processing unit 23 may also determine that the translation does not require correction if, for example, the explanation corresponding to the translation included in the response information (corrected translation) is "no correction."

[0052] If it is determined that the translation does not require correction (YES in step S205), the correction processing unit 23 excludes the translation from the targets for input of correction instructions from the next time onwards (step S206). The excluded translation is stored in the memory 202 so as to be distinguishable from translations that are not excluded. If it is not determined that the translation does not require correction (NO in step S205), the correction processing unit 23 sets the corresponding corrected translation as the translation (target) of the next bilingual information and stores it in the memory 202.

[0053] In this embodiment, in steps S205 and S206, the correction processing unit 23 may set exception conditions for determining whether correction is unnecessary. Specifically, the exception conditions include not excluding a translation that is determined not to require correction in the first response from targets for input of correction instructions in the next response. The exception conditions are set for the purpose of having the generative model infer at least once whether there are any errors in the content of the corrections included in the first response.

[0054] In step S207, the correction processing unit 23 determines whether the repetition end condition is satisfied. If it is determined that the repetition end condition is not satisfied (NO in step S207), the correction processing unit 23 returns to step S203 and repeatedly executes input of correction instructions and acquisition of responses thereto in steps S203 to S206. These processes are repeatedly executed until the repetition end condition is satisfied.

[0055] In this embodiment, the repetition termination condition includes at least one of the following: the response from the generative model indicating that no correction is required; and the predetermined number of responses being obtained. Here, the "no correction is required" condition refers to a state in which no correction is required for all target translations, for example, a state in which all translations included in a translated document no longer require correction. The predetermined number of responses being obtained can be set arbitrarily by the administrator, for example, to 30 or less. In this embodiment, the repetition termination condition is set as a condition in which a predetermined number of responses are obtained, and may also be a condition in which the repetition is terminated before the predetermined number of responses are obtained, provided that the response from the generative model indicates that no correction is required.

[0056] If it is determined that the repetition termination condition is met (NO in step S207), the output unit 24 outputs the correction result based on the response information (step S208). The correction result is generated based on the corrected translation and the explanation. In one aspect, the correction result is output in a format that allows a comparison of the difference between the pre-correction translation and the final corrected translation. Here, an explanation corresponding to the final corrected translation may be written side by side. In one aspect, the correction result is output in a list format of the pre-correction translation and a series of corrected translations. Also, an explanation corresponding to the series of corrected translations may be written side by side. Also, the correction result may include a translation document file that reflects all of the final corrected translations. By outputting the correction result to the user terminal 3, the user can obtain a translation document with improved translation accuracy, thereby making post-editing more efficient.

[0057] 6 shows a processing flowchart of a translation support device 2 according to a different embodiment. In this embodiment, the translation support device 2 generates subsets each including a plurality of sentences included in the bilingual information, and issues correction instructions to the generative model and acquires a response for each subset. Note that explanations of processes common to the steps in FIG. 5 will be omitted as appropriate.

[0058] In step S303, the correction processing unit 23 generates a plurality of subsets, each including a plurality of segments included in the bilingual information. In this embodiment, a subset is a unit that groups together a plurality of segments, and is set for the purposes of appropriately adjusting the number of processing requests to the generative model and processing the processing requests in parallel. Note that a subset may be a unit that groups together a plurality of sentences instead of a segment.

[0059] As a specific example, assume that the bilingual information has 100 segments (or 100 sentences). For a generative model to process and respond to 100 segments in a single processing request poses problems such as the processing time required and limitations on the number of characters. In this embodiment, the 100 segments are divided into 10 subsets, and each of these subsets can be processed with 10 processing requests. For example, the first subset includes bilingual information for segments "01" to "10," and the second subset includes bilingual information for segments "11" to "20." The number of segments set in a subset is not limited to 10 and can be set appropriately taking into account processing time and processing efficiency.

[0060] In step S304, the correction processing unit 23 transmits a processing request for each subset to the generating device 4. Here, multiple processing requests related to multiple subsets may be transmitted to multiple generating devices 4 or multiple generative models and processed in parallel.

[0061] In step S305, the correction processing unit 23 acquires a response for each subset. In step S306, the correction processing unit 23 determines a subset that does not require correction. A subset that does not require correction includes a case where all translations in the subset are determined to not require correction.

[0062] If it is determined that the subset does not require correction (YES in step S306), in step S307 the correction processing unit 23 excludes the subset from the targets of subsequent processing requests. If it is determined that the subset does not require correction (NO in step S306), the correction processing unit 23 sets the corrected translation included in the subset as the translation (target) of the bilingual information for the next processing request and stores it in memory 202. At this time, translations in the subset that are determined not to require correction may be excluded from the targets of subsequent processing requests. Furthermore, by excluding translations from the subset, translations equivalent to the capacity may be appropriated from other subsets. In other words, the subset may be optimized to have a set number of segments.

[0063] In step S308, the correction processing unit 23 determines whether the repetition end condition is satisfied, and repeatedly executes input of a correction instruction and acquisition of a response thereto until the repetition end condition is satisfied. In step S309, the output unit 24 generates a correction result based on the acquired response and transmits it to the user terminal 3.

[0064] Next, we will explain the results of an analysis of the relationship between the number of iterations of correction using the generative model and the accuracy of the translation, with reference to Tables 1, 2 and Figure 7.

[0065] Table 1 shows the accuracy of translated documents A to F before machine translation (MT) and after n rounds of revisions by the generative model. The accuracy of the translated document is expressed as 100% when there are no further post-editing corrections made by humans and the document is complete. "na" in Table 1 indicates that the response from the generative model no longer requires any revisions.

[0066] Table 1 shows the accuracy when a document translated using MT before correction is corrected up to 10 times using a generative model. For example, Rev:1 represents a document translated after one correction, Rev:2 represents a document translated after two corrections, and Final represents a document translated after 10 corrections. The number of targets indicates the number of segments to be translated. In this example, translated documents A to F each have a different file format and document content. Specifically, translated document B is a presentation file, and translated document F is a document file containing table data. The other translated documents are document files containing only text. The document content includes information on management, economics, news, etc.

[0067] [Table 1]

[0068] Table 1 shows that the first revision of translated documents A to E results in a correction of more than 70% of the final accuracy. Furthermore, the accuracy of translated documents A to E generally improves up to the sixth revision, after which revisions are no longer necessary or the degree of accuracy improvement becomes smaller. Based on these results, in this embodiment, the predetermined number of response acquisitions can be set to 10 or less, and more preferably 6 or less. Note that translated document F is a document containing accounting terms, and the accuracy in Table 1 was calculated based on the translation list published by the Financial Services Agency. Although translated document F did not match the translations in the translation list, the translations were at a standard level. When calculated assuming these were correct, translated document F's accuracy improved by an overall average of 2 to 3%.

[0069] FIG. 7 is a graph showing the relationship between the number of revisions for each translated document in Table 1 and the accuracy of the translation. In FIG. 7, the vertical axis represents accuracy, and the horizontal axis represents the number of revisions (repetitions). As can be seen from FIG. 7, the accuracy for each document improves as the number of revision repetitions increases. In this way, the invention according to this embodiment can provide a technology that improves the efficiency of post-editing by repeatedly executing revisions using a generative model.

[0070] Table 2 shows the accuracy of translated documents A to F before revision using machine translation (MT) and after n revisions using a generative model. While Table 1 shows the results for segments, Table 2 shows the results for sentences contained in the segments. The number of subjects in Table 2 indicates the number of sentences to be translated. Note that Table 2 only evaluates the accuracy of the final revised translated documents.

[0071] Table 2 shows that the accuracy of translated documents A to F has improved compared to the pre-edited and post-edited translated documents in Table 1. One reason for this result is that when a segment requiring revision contains multiple sentences, not all of those sentences necessarily require revision. Therefore, the results in Table 2 can be said to more accurately represent the reduction rate of work in actual post-editing.

[0072] [Table 2] [Explanation of symbols]

[0073] 1 Translation support system 2. Translation support device 21 Document Acquisition Department 22 Bilingual Acquisition Department 23 Correction processing section 24 Output section 3. User terminal 4 Generator

Claims

1. A translation support system that supports a task of translating an original document in a first language into a translated document in a second language, comprising: a bilingual acquisition unit that acquires bilingual information including an original text in a first language and a translation of the original text in a second language; a correction processing unit that provides a processing request including the bilingual information and a correction instruction for the translation to a generative model and obtains a response from the generative model, the correction processing unit sets the corrected translation included in at least the first response as the translation of the next processing request, and repeats inputting the correction instruction and obtaining the response thereto in the second and subsequent responses until a repetition termination condition is satisfied; A translation support system, wherein the iteration termination condition includes a condition indicating that the response from the generative model does not require modification.

2. 2. The translation support system according to claim 1, wherein the condition for ending the repetition includes reaching a predetermined number of responses obtained.

3. 3. The translation support system according to claim 1, wherein the correction processing unit inputs to the generative model and obtains a response for each of a plurality of sentences or segments included in the parallel translation information, and excludes translations that are determined to not require correction based on the previous response from being targets for input of correction instructions from the next time onwards.

4. 3. The translation support system according to claim 1, wherein the correction processing unit generates a plurality of subsets each including a plurality of sentences or a plurality of segments included in the parallel translation information, provides each of the subsets to a generative model, obtains a response for each of the subsets from the generative model, and, if it is determined based on the previous responses that none of the sentences or segments included in the subset require correction, excludes the subset from targets for input of correction instructions from the next time onwards.

5. 3. The translation support system according to claim 1, wherein the correction processing unit provides the generative model with a processing request for generating a response including the original text, the translation, and a corrected translation obtained by correcting the translation, obtains the response from the generative model, and sets the corrected translation included in the response as the translation for the next processing request.

6. 3. The translation support system according to claim 1, wherein the correction instructions include instructions to reflect in the correction of the translation without substantially changing at least one or more of the number, elements, order, and relative positional relationship of tag information contained in the original text.

7. 3. The translation support system according to claim 1, wherein the correction processing unit generates an explanation in a specified language for the correction portion of the translation.

8. A translation support method for supporting a task of translating an original document in a first language into a translated document in a second language, comprising: a bilingual information acquisition step of acquiring bilingual information including an original text in a first language and a translation of the original text in a second language; a correction processing step of providing a processing request including the bilingual information and a correction instruction for the translation to a generative model and obtaining a response from the generative model, the correction processing step sets the corrected translation included in at least the first response as the translation of the next processing request, and repeats inputting the correction instruction and obtaining the response thereto in the second and subsequent responses until a repetition termination condition is satisfied; A translation support method, wherein the iteration termination condition includes a condition indicating that a response from the generative model does not require modification.

Images