Translation method and system

Abstract

The embodiment of the invention provides a translation method and system, computer equipment, a computer readable storage medium and a computer program product, and belongs to the field of data processing. The translation method comprises the steps that a first source language text is received, and a first structured context constraint is generated through retrieval based on the first source language text by means of a multi-dimensional knowledge base; generating a first translation first draft based on the first structure context constraint; performing evaluation-revision translation iteration based on the first translation draft by utilizing a scheduling control module to obtain a first translation result; and performing document level evaluation based on the first translation result, and obtaining and outputting a translated text based on the first translation result under the condition that the document level evaluation meets a preset requirement. According to the technical scheme, the translation quality and efficiency can be improved, and the large-scale and engineering delivery requirements can be met.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a translation method, system, computer device, computer-readable storage medium, and computer program product. Background Technology

[0002] In the field of game localization, relevant technical solutions can be mainly divided into three categories: The first category is single-round translation solutions based on Large Language Models (LLM). While simple to integrate, these solutions are prone to terminology drift and stylistic inconsistencies when dealing with long storylines and multi-character dialogues, resulting in high maintenance costs. The second category consists of traditional computer-aided translation tools, such as Trados. Their core flaw lies in their reliance on rigid string fuzzy matching, making it difficult to handle variables, cross-sentence dependencies, and non-linear narratives in game text. Furthermore, they lack the ability to structurally model character styles and sentence templates, failing to guarantee tonal consistency in large-scale dialogues. The third category is simple Retrieval-Enhanced Generation (RAG) translation solutions. While these can improve terminology consistency and contextual coherence to some extent, their knowledge dimensions are limited, and they largely follow a "single-round translation" paradigm. Lacking a full-process system architecture, they lead to a disconnect between translation production and version management, making it difficult to support the large-scale, automated delivery needs of large-scale game projects.

[0003] It should be noted that the above content is not necessarily prior art, nor is it intended to limit the scope of patent protection of this application. Summary of the Invention

[0004] This application provides a translation method, system, computer device, computer-readable storage medium, and computer program product to solve or alleviate one or more of the technical problems mentioned above.

[0005] One aspect of this application provides a translation method, the method comprising: Receive the first source language text, use a multidimensional knowledge base to search based on the first source language text, and generate the first structured context constraint based on the search results; A first draft translation is generated based on the first structural context constraints; The first initial translation draft is used as the translation draft for the current round at the start of the iteration. The translation is iterated using the scheduling control module. If the translated draft meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft for the current round, revising the translation draft for the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round for further iteration. Based on the first translation result, a document-level evaluation is performed. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation result.

[0006] Optionally, the method further includes: Receive the second source language text and the corresponding historical translation text; The second source language text is aligned with the historical translation text using a sliding window semantic alignment method to obtain aligned translation units, and a translation memory is constructed based on the aligned translation units. The translation memory and the multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results; A second initial translation draft is generated based on the second structural context constraints; The second initial translation draft is used as the translation draft for the current round at the start of the iteration. The scheduling control module is used to perform translation iterations. If the translated draft after the iteration meets the preset quality requirements or reaches the preset number of iterations, the second translation result of the second source language text is obtained. Based on the second translation result, a document-level evaluation is performed. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the second translation result.

[0007] Optionally, before using the scheduling control module to perform translation iterations with the first initial translation draft as the translation draft for the current round at the start of the iteration, the method further includes: The complexity of the first source language text is determined using the scheduling control module. If the complexity assessment result is simple text, a document-level evaluation is performed based on the first translation draft. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation draft. If the complexity assessment result indicates that the text is not simple, the first translation draft is used as the translation draft for the current round at the start of the iteration, and the translation iteration and subsequent steps are performed using the scheduling control module.

[0008] Optionally, generating the first translation draft based on the first structural context constraints includes: The first structured context constraint and the first source language text are input into the generative agent, and the generative agent is used to generate the first translation draft. Correspondingly, the step of using the first initial translation draft as the translation draft for the current round at the start of the iteration, and performing translation iterations using the scheduling control module, includes: The translation of the current round is input into the evaluation agent, which evaluates the translation of the current round to obtain the evaluation result. The evaluation results are fed back to the revision agent, which then revises the translation of the current round to obtain the revised translation. The revised translation is evaluated for quality, and the results are fed back to the scheduling and control module. The scheduling and control module then determines whether to proceed to the next iteration or submit the revised translation to the manual review terminal based on the evaluation results.

[0009] Optionally, the step of evaluating the translation of the current round using the evaluation agent to obtain the evaluation result includes: The evaluation agent is used to evaluate the translation of the current round and generate a structured evaluation result, which includes error type, error location, severity, and revision suggestions.

[0010] Optionally, the document-level evaluation based on the first translation result includes: A global consistency check is performed based on the first translation result. The global consistency check includes consistency checks across sentences, paragraphs, chapters, and documents. The consistency check includes terminology consistency, sentence structure consistency, and style consistency checks. If the global consistency check fails, the first translation result is returned to the revision agent for revision. If the global consistency check passes, a document-level evaluation is performed based on the first translation result.

[0011] Optionally, if the document-level evaluation meets preset requirements, obtaining and outputting the translated text based on the first translation result includes: If the document-level evaluation meets the preset requirements, an online verification is performed based on the first translation result. The online verification includes at least one of the following: project format, code tag integrity, and user interface length. If the online verification fails but the conditions for automatic repair are met, the rule script is invoked to correct the first translation result; If the online verification fails and the conditions for automatic repair are not met, the first translation result is returned to the revision agent for revision.

[0012] Optionally, the step of using the translation memory and multidimensional knowledge base to perform retrieval based on the second source language text, and generating a second structured context constraint based on the retrieval results, includes: Determine the increment of the second source language text relative to the historical translated text; When the increment meets the preset conditions, the translation memory is used as a strong constraint. The translation memory and the multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results.

[0013] Optionally, the multidimensional knowledge base includes at least two of the following: a terminology knowledge base, a sentence structure knowledge base, a style knowledge base, and a version knowledge base.

[0014] Another aspect of this application provides a translation system, the system comprising: The retrieval module is used to receive the first source language text, perform retrieval based on the first source language text using a multidimensional knowledge base, and generate the first structured context constraint based on the retrieval results. The generation module is used to generate a first translation draft based on the first structural context constraints; An iteration module is used to take the first initial translation draft as the translation draft for the current round at the start of the iteration, and use the scheduling control module to perform translation iteration. If the translation draft after iteration meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft for the current round, revising the translation draft for the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round for continued iteration. The output module is used to perform a document-level evaluation based on the first translation result, and if the document-level evaluation meets preset requirements, to obtain and output the translated text based on the first translation result.

[0015] Another aspect of this application provides a computer device, including: At least one processor; and A memory that is communicatively connected to the at least one processor; Wherein: the memory stores instructions that can be executed by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described above.

[0016] Another aspect of this application provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method described above.

[0017] Another aspect of this application provides a computer program product including a computer program that, when executed by a processor, implements the method described above.

[0018] The embodiments of this application employing the above-described technical solution may have the following advantages: By leveraging a multidimensional knowledge base to generate structured context constraints, creating initial drafts based on these constraints, and iterating through multiple rounds of evaluation and revision of the initial drafts, the system ultimately performs document-level evaluation of the translated document and outputs the translated text based on the evaluation results. Through structured context constraints, multiple rounds of evaluation, iterative revisions, and document-level assessment, the system avoids terminology drift and style inconsistencies that can occur with long storylines and multi-character dialogues, reducing later maintenance costs. It can also handle variables, cross-sentence dependencies, and non-linear narratives in game texts, possessing structured modeling capabilities for character styles and sentence templates to ensure consistent tone in large-scale dialogues. Furthermore, the system architecture across the entire translation process ensures translation production and version management, effectively supporting the large-scale, automated delivery needs of large-scale game projects. Attached Figure Description

[0019] The accompanying drawings exemplify embodiments and form part of the specification, serving together with the textual description to explain exemplary implementations of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. Throughout the drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0020] Figure 1 The diagram schematically illustrates the operating environment of the translation method according to Embodiment 1 of this application; Figure 2 A flowchart illustrating a translation method according to Embodiment 1 of this application is shown schematically; Figure 3 The illustration schematically shows the additional flow of the translation method according to Embodiment 1 of this application; Figure 4 This illustration schematically shows another additional process in the translation method according to Embodiment 1 of this application; Figure 5 Schematic illustration Figure 2 Flowchart of the sub-steps in step S204; Figure 6 Schematic illustration Figure 2 Flowchart of the sub-steps in step S206; Figure 7 Schematic illustration Figure 2 Flowchart of another sub-step in step S206; Figure 8 Schematic illustration Figure 2Flowchart of the sub-steps in step S304; Figure 9 The schematic diagram illustrates the principle framework of the translation method according to Embodiment 1 of this application; Figure 10 A flowchart illustrating a translation method according to Embodiment 1 of this application is shown schematically. Figure 11 A block diagram of a translation system according to Embodiment 2 of this application is schematically shown; and Figure 12 A schematic diagram of the hardware architecture of a computer device according to Embodiment 3 of this application is shown. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0022] It should be noted that the descriptions involving "first," "second," etc., in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0023] It should be noted that, in any stage of this application involving the collection, storage, use, transmission, and processing of data, each stage strictly adheres to the laws, regulations, industry standards, and regulatory requirements of the data source, usage location, and relevant countries and regions to ensure the legality and compliance of data activities. In the collection stage, the purpose, method, and scope of collection are clearly communicated to the data subject in a prominent manner. Collection is conducted only after obtaining the data subject's legal authorization, ensuring that the collection process follows the "minimum necessary" principle and does not exceed the scope of data collection. In the storage stage, storage periods are limited, and data is promptly deleted or anonymized / encrypted after the storage purpose is achieved. In the usage stage, a strict data security protection mechanism is implemented, using field-level desensitization technology and processing the original data according to preset desensitization rules. For different types of data, multiple desensitization strategies, such as data generalization, data anonymization, and data encryption, are employed to effectively mitigate the risk of sensitive information leakage and ensure that all data used is securely processed and desensitized, comprehensively protecting the rights and interests of data subjects and data security. In the transmission and processing stages, the confidentiality and security of data are ensured during transmission and processing.

[0024] In the description of this application, it should be understood that the numerical labels before the steps do not indicate the order of the steps, but are only used to facilitate the description of this application and to distinguish each step, and therefore should not be construed as a limitation of this application.

[0025] First, a definition of the terminology used in this application is provided: Game localization refers to the entire process of adapting, translating, and transforming a game from its original language and cultural environment to suit the language, culture, laws, and player habits of the target market.

[0026] Sliding Window Semantic Matching (SWSM): Applied in the field of natural language processing, it uses a "sliding window" approach to segment long texts and combines semantic matching algorithms to achieve semantic alignment of content across document versions and between documents.

[0027] An intelligent agent is an agent that can perceive its environment and take actions to achieve specific goals. In the field of artificial intelligence, it is usually based on a large language model and has the ability to autonomously perceive, understand, plan, make decisions, remember, act, and use tools. It can automatically perform complex tasks with less human intervention and exhibits characteristics such as autonomy, interactivity, responsiveness, and adaptability.

[0028] Secondly, to facilitate understanding of the technical solutions provided in the embodiments of this application by those skilled in the art, the relevant technologies are described below: Among related technologies, single-turn translation solutions based on large language modules are prone to terminology drift and style inconsistencies when dealing with long storylines and multi-character dialogues, resulting in high maintenance costs in the later stages. Traditional computer-aided translation tools struggle to handle variables, cross-sentence dependencies, and non-linear narratives in game texts, and lack the ability to structurally model character styles and sentence templates, making it impossible to guarantee the consistency of tone in large-scale dialogues. Simple search-enhanced translation solutions have a single knowledge dimension and are mostly based on the "single-turn translation" paradigm, lacking a full-process system architecture, leading to a disconnect between translation production and version management, making it difficult to support the large-scale and automated delivery needs of large-scale game projects.

[0029] To address this, this application provides a translation technology solution. This solution utilizes a multi-dimensional knowledge base to generate structured context constraints, generates a preliminary translation draft based on these constraints, iterates through the draft multiple times to obtain the final translation document, and finally performs a document-level evaluation. Based on the evaluation results, the translated text is output. This approach avoids terminology drift and style inconsistencies that can occur with long storylines and multi-character dialogues, reducing later maintenance costs. It can also handle variables, cross-sentence dependencies, and non-linear narratives in game texts, possessing structured modeling capabilities for character styles and sentence templates, ensuring consistency in tone across large-scale dialogues. Furthermore, the system architecture across the entire translation process ensures translation production and version management, effectively supporting the large-scale, automated delivery needs of large-scale game projects. Details are provided below.

[0030] Finally, for ease of understanding, an exemplary operating environment is provided below.

[0031] like Figure 1 As shown in the diagram, the runtime environment includes service platform 2, network 4, and client 6, where: Service platform 2 can consist of one or more computing devices. These computing devices may include virtualized computing instances. Virtualized computing instances may include virtual machines, such as emulations of computer systems, operating systems, servers, etc. The computing devices can load virtual machines based on virtual images and / or other data that define specific software used for emulation (e.g., operating systems, dedicated applications, servers). As the demand for different types of processing services changes, different virtual machines can be loaded and / or terminated on one or more computing devices. A hypervisor can be implemented to manage the use of different virtual machines on the same computing device.

[0032] Service platform 2 can be configured to communicate with clients 6, etc., via network 4. Network 4 includes various network devices, such as routers, switches, multiplexers, hubs, modems, bridges, repeaters, firewalls, proxy devices, and / or similar devices. Network 4 may include physical links, such as coaxial cable links, twisted pair cable links, fiber optic links, or combinations thereof, or wireless links, such as cellular links, satellite links, Wi-Fi links, etc.

[0033] Service platform 2 can provide services such as storage, reading, writing, querying, and deleting, such as providing translation services to clients.

[0034] Client 6 can be an electronic device running operating systems such as Windows, Android™, or iOS, including smartphones, tablets, laptops, virtual reality devices, gaming devices, set-top boxes, in-vehicle terminals, and smart TVs. Based on these operating systems, various applications can run, such as translation applications.

[0035] Client 6 can provide / configure user access pages, which can be used to control service platform 2 or upload objects, etc.

[0036] It should be noted that the above-mentioned equipment is exemplary, and the number and type of equipment can be adjusted in different scenarios or according to different needs.

[0037] The technical solutions of this application are described below through multiple embodiments, using a service platform as the implementing entity. It should be understood that these embodiments can be implemented in many different forms and should not be construed as being limited to the embodiments described herein.

[0038] Example 1 Figure 2 A flowchart illustrating a translation method according to Embodiment 1 of this application is shown schematically.

[0039] like Figure 2 As shown, the translation method may include steps S200~S206, wherein: Step S200: Receive the first source language text, perform a retrieval based on the first source language text using a multidimensional knowledge base, and generate the first structured context constraint based on the retrieval results.

[0040] Step S202: Generate the first translation draft based on the first structural context constraints.

[0041] Step S204: The first translation draft is used as the translation draft for the current round at the start of the iteration. The translation is iterated using the scheduling control module. If the translation draft after the iteration meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft for the current round, revising the translation draft for the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round for continued iteration.

[0042] Step S206: Perform a document-level evaluation based on the first translation result. If the document-level evaluation meets the preset requirements, obtain the translated text based on the first translation result and output it.

[0043] The translation method provided in this embodiment generates structured context constraints using a multi-dimensional knowledge base, produces an initial draft based on these constraints, and iterates through multiple rounds of evaluation and revision of the initial draft. Finally, it performs a document-level evaluation of the translated document and outputs the translated text based on the evaluation results. Through structured context constraints, multiple rounds of evaluation, iterative revisions, and document-level evaluation, it avoids terminology drift and style inconsistencies that can occur in long storylines and multi-character dialogues, reducing later maintenance costs. It can also handle variables, cross-sentence dependencies, and non-linear narratives in game texts, and has the ability to structurally model character styles and sentence templates, ensuring consistency in tone during large-scale dialogues. Furthermore, through a comprehensive translation system architecture, it can guarantee translation production and version management, effectively supporting the large-scale, automated delivery needs of large-scale game projects.

[0044] The following combination Figure 2 The steps in steps S200 to S206, as well as other optional steps, are described in detail.

[0045] Step S200 It receives the first source language text, performs a retrieval based on the first source language text using a multidimensional knowledge base, and generates the first structured context constraint based on the retrieval results.

[0046] The service platform receives the original language text uploaded by the client. The client can provide an input interface through which users can upload the original language text.

[0047] Optionally, the multidimensional knowledge base may include at least two of the following: a terminology knowledge base, a sentence structure knowledge base, a style knowledge base, and a version knowledge base. Specifically, the terminology knowledge base ensures consistent translation of proper nouns; the sentence structure knowledge base stores abstracted sentence templates and their corresponding translations to guarantee the standardization of sentence expression; the style knowledge base stores the style characteristics of each role to maintain the role's stylistic features; and the version knowledge base stores translation information for each version to support incremental translation and asset traceability.

[0048] The input for the retrieval in step S200 can be the first source language text and its metadata (such as text type, role ID, and scene). The retrieval results can include recall content, relevance scores, and constraint strength indicators for various dimensions. During the retrieval, one or more methods can be combined, such as BM25, vector retrieval, structural matching, and metadata query.

[0049] Specifically, the first source language text can be broken down into retrieval units (such as sentences, phrases, or terms). Optionally, basic tags (such as keywords like topic and domain) can be added. For the multidimensional knowledge base, information related to the first source language text can be retrieved from the multidimensional knowledge base. For example, if role A appears in the first source language text, style information corresponding to role A can be retrieved from the style knowledge base. After the retrieval is completed, the results of the multidimensional knowledge base retrieval can be integrated to obtain constraints (such as the terminology translation rules, contextual logical association requirements, role style, sentence structure norms, etc.), which are then organized into a first structured context constraint in a structured format. During the retrieval, multiple dimensions of retrieval can be executed in parallel to improve response efficiency.

[0050] Step S202 The first translation draft is generated based on the first structural context constraints.

[0051] Specifically, the first structural context constraints can be transformed into guidance parameters (such as input prompts for the model) that the translation model (such as a large language model) can read. The first source language text is then input into the translation model, the transformed guidance parameters are loaded, and the first translation draft is generated.

[0052] Step S204 The first translation draft is used as the translation draft for the current round at the start of the iteration. The translation is iterated using the scheduling control module. If the translated draft meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft of the current round, revising the translation draft of the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round and the iteration continues.

[0053] Specifically, the first draft is used as the translation draft for the current round at the start of the translation iteration (i.e., the first round of translation draft). The scheduling control module performs translation iterations based on the current round of translation draft. This module controls resource scheduling and iteration strategies during the translation iteration process. For example, it allocates resources to evaluate the current round of translation draft or determines whether to continue iteration based on the quality of the revised translation draft. To facilitate the execution of the scheduling control module, relevant status reports can be sent to it. For instance, information such as the quality of the revised translation draft and the current number of iterations can be fed back to the scheduling control module, allowing it to decide whether to continue or terminate the iteration based on this information. When evaluating the current round of translation, a pre-built evaluation model or rules can be used to assess aspects such as terminology consistency, sentence structure standardization, and role style matching. Evaluation results are then output based on these dimensions. Targeted revisions are then made based on these results, such as replacing non-standard terminology, adjusting sentence structure, or correcting logical inconsistencies. After revision, the revised translation is checked to see if it meets the preset quality requirements and if the preset number of iterations has been reached. If the revised translation meets the preset quality requirements or has reached the preset number of iterations, the iteration can end. If the revised translation does not meet the preset quality requirements and has not reached the preset number of iterations, the revised translation is used as the next round of evaluation and iteration continues until it meets the preset quality requirements or reaches the preset number of iterations. At the end of the iteration, the current translation is taken as the first translation result.

[0054] Step S206 Based on the first translation result, a document-level evaluation is performed. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation result.

[0055] Document-level evaluation can include, but is not limited to, consistency of terminology or style throughout the document, logical coherence, formatting standardization, and overall quality achievement. Consistency of terminology or style throughout the document includes, for example, whether the translation of the same term is consistent throughout the document, and whether the character's style is consistent throughout the document. Logical coherence includes, for example, whether the transitions between paragraphs are smooth, and whether the semantics of the context are consistent. Formatting standardization includes, for example, whether the document structure, punctuation, and professional formatting meet the requirements of the target scenario. Overall quality achievement includes, for example, whether the preset translation accuracy rate is met.

[0056] Specifically, the first translation result can be evaluated through a full-text scan, based on pre-built document-level evaluation requirements and methods. If the evaluation result meets the preset requirements, the first translation result can be directly used as the final translated text output. If the evaluation result does not meet the preset requirements, the first translation result can be returned to the earlier stage for further revision or submitted for manual review.

[0057] In optional embodiments, such as Figure 3 As shown, the translation method provided in this application embodiment may further include: Step S300: Receive the second source language text and the historical translation text corresponding to the second source language text.

[0058] Step S302: Align the second source language text with the historical translation text using the sliding window semantic alignment method to obtain aligned translation units, and construct a translation memory based on the aligned translation units.

[0059] Step S304: Use the translation memory and multidimensional knowledge base to perform retrieval based on the second source language text, and generate a second structured context constraint based on the retrieval results.

[0060] Step S306: Generate a second initial translation draft based on the second structural context constraints.

[0061] Step S308: The second translation draft is used as the translation draft for the current round at the start of the iteration. The translation is iterated using the scheduling control module. If the translated draft after the iteration meets the preset quality requirements or reaches the preset number of iterations, the second translation result of the second source language text is obtained.

[0062] Step S310: Perform a document-level evaluation based on the second translation result. If the document-level evaluation meets the preset requirements, obtain the translated text based on the second translation result and output it.

[0063] It should be noted that the historical translation and the translated text correspond to the same target language.

[0064] In the gaming industry, games are often modified, requiring the release of new translated versions. In such cases, the historical translated text can be the original translated text from before the modification. Of course, the translation method of this application is not limited to gaming scenarios; it can also be applied to film and television subtitle translation, comic or light novel localization, user interface (UI) and help documentation translation for large software or platforms, and multilingual operational activity pages and announcements. Multiple historical translated texts can be used.

[0065] In steps S300-S304, the service platform receives the second source language text and the corresponding historical translation text uploaded by the client. Then, a sliding window can be used to segment the first source language text and the historical translation text into several continuous small text segments. The window size of the sliding window can be set according to the text length and semantic integrity. Next, semantic similarity is calculated for each of the two text segments. Based on the semantic similarity calculation results, the semantically corresponding parts in the text segments of the first source language text and the historical translation text are found, forming aligned translation units. Algorithms such as Transformer can be used to calculate semantic similarity. Finally, all aligned translation units can be stored uniformly to form a reusable translation memory.

[0066] Optionally, after alignment is completed, the alignment completion status can be reported to the scheduling control module, which will then decide whether to proceed to the next stage.

[0067] In step S304, the second source language text can be divided into retrieval units (such as sentences, phrases, or terms). Optionally, basic tags (such as keywords like topic and domain) can be added. For the translation memory, the retrieval units can be matched with the aligned "source text fragments - translated text fragments" in the translation memory to filter out historical translation cases with high semantic similarity. For the multidimensional knowledge base, information related to the first source language text can be retrieved from the multidimensional knowledge base. Finally, the results of the retrieval from the translation memory and the multidimensional knowledge base can be integrated to obtain constraints (such as the terminology translation rules, contextual logical association requirements, role style, sentence structure norms, etc.), which are then organized into the first structured context constraints in a structured format.

[0068] It is understood that this embodiment is applicable to scenarios with historical translated texts. Except for steps S300-S304, the other parts are similar to or the same as the previous embodiment. For the same parts, please refer to the previous embodiment, which will not be repeated here.

[0069] In this embodiment, a sliding window semantic matching method is used to align the second source language text with the historical translated text and construct a translation memory. The translation memory and a multidimensional knowledge base are used to generate a second structured context constraint based on the second source language text. A second draft translation is generated based on the second structured context constraint. The second draft translation is iterated to obtain a second translation result. A document-level evaluation is performed based on the second translation result. If the document-level evaluation meets preset requirements, the translated text is obtained and output based on the second translation result. Further constraints can be generated through the construction of the translation memory, thereby minimizing terminology drift and style inconsistencies that may occur in long narratives and multi-character dialogues, further improving the quality of the translation.

[0070] In an optional embodiment, before step S204, that is, before the translation iteration is performed using the scheduling control module when the first translation draft is used as the current round's translation draft at the start of the iteration, such as... Figure 4 As shown, the translation method in this application embodiment may further include: Step S400: Use the scheduling control module to determine the complexity of the first source language text.

[0071] In step S402, if the complexity judgment result is simple text, a document-level evaluation is performed based on the first translation draft. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation draft.

[0072] Step S404: If the complexity judgment result is that the text is not simple, the first translation draft is used as the translation draft of the current round at the start of the iteration, and the translation iteration and subsequent steps are performed using the scheduling control module.

[0073] The scheduling and control module can determine the complexity of the source language text based on its length, content difficulty, and structural complexity. Content difficulty can be assessed by determining whether it contains technical terms, complex sentence structures, multi-role styles, etc., or by judging the proportion of such complex content. Structural complexity can be assessed by determining whether there are connections between paragraphs or multiple layers of logic. Different judgment results and corresponding criteria can be pre-configured for different levels of complexity. These different results could be categorized as simple text, general text, and difficult text, or simple text and core text, etc. The corresponding judgment criteria could be, for example, complex content less than 10%, 10 ≤ complex content < 50%, and complex content ≥ 50%.

[0074] If the complexity assessment result indicates simple text, the multi-round iteration process in step S204 can be skipped, and the generated first translation draft can be directly evaluated at the document level. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation draft. Optionally, if the document-level evaluation does not meet the preset requirements, step S204 can be entered for multi-round iteration, or the first translation draft can be submitted for manual review. If the complexity assessment result indicates non-simple text, the first translation draft is used as the translation draft for the current round at the start of the iteration, and the scheduling control module performs translation iterations and subsequent steps based on the translation draft for the current round.

[0075] In this embodiment, the complexity of the first source language text is determined by the scheduling control module. If the complexity determination result is simple text, the first translation draft is directly evaluated at the document level and then operated on. If the complexity determination result is not simple text, the translation iteration and subsequent operations are performed. The scheduling control module can be used to realize the function of adaptive routing, saving translation computing resources when the text to be translated is simple text.

[0076] In an optional embodiment, step S202, generating a first translation draft based on the first structured context constraint, may include: inputting the first structured context constraint and the first source language text into the generating agent, and using the generating agent to generate the first translation draft; correspondingly, in step S204, the first translation draft is used as the translation draft for the current round at the start of the iteration, and the translation iteration is performed using the scheduling control module, such as... Figure 5 As shown, it may include: In step S500, the translation of the current round is input into the evaluation agent, and the evaluation agent is used to evaluate the translation of the current round to obtain the evaluation result.

[0077] Step S502: Feedback the evaluation results to the revision agent, and use the revision agent to revise the translation in the current round to obtain the revised translation.

[0078] Step S504: Conduct a quality assessment on the revised translation and feed the assessment results back to the scheduling control module. The scheduling control module then determines whether to proceed with the next iteration or submit the revised translation to the manual review terminal based on the assessment results.

[0079] Specifically, the first structured context constraint can be transformed into a prompt for the generating agent. The first source language text and the generated prompt are input into the generating agent, which generates the first draft translation. During the translation iteration process, the current draft translation is input into the evaluation agent. The evaluation agent automatically detects the current draft translation according to preset evaluation dimensions and outputs the evaluation result. Then, the evaluation result can be fed back to the revision agent, which revises the current draft translation based on the evaluation result, resulting in the revised translation. The revised translation is then subjected to a quality evaluation, and the quality evaluation result is fed back to the scheduling control module. The scheduling control module determines whether to proceed to the next iteration or submit the revised translation to a human review terminal based on the quality assessment results. For example, if the quality assessment results show that the preset quality requirements are met, the scheduling control module can end the iteration. If the quality assessment results show that the preset quality requirements are not met and the preset number of iterations has not been reached, the scheduling control module can decide to proceed to the next iteration. If the quality assessment results show that the preset quality requirements are not met and the preset number of iterations has been reached, the scheduling control module can decide to end the iteration and submit the revised translation to a human review terminal for manual revision. Optionally, if the quality assessment results show that the preset quality requirements are not met and the preset number of iterations has been reached, the revised translation can be marked as "low confidence" to distinguish it from translations that meet the preset quality requirements after iteration.

[0080] The evaluation dimensions for assessing an agent can include multiple dimensions such as terminology consistency, style consistency, sentence structure regularity, semantic fidelity, contextual coherence, and format correctness. The assessment agent can be independent of the generated agent to avoid confirmation bias.

[0081] When revising, the agent makes targeted revisions based on the evaluation results, focusing on local revisions rather than a complete rewrite, thus maintaining the stability of the correct parts.

[0082] To enable the revision agent to perform targeted revisions, in an optional embodiment, step S500 involves using an evaluation agent to evaluate the current round of translation and obtain evaluation results. This may include: using the evaluation agent to evaluate the current round of translation and generate evaluation results in a structured format (such as JSON format). The structured evaluation results include error type, error location, severity, and revision suggestions. That is, the evaluation results obtained by the evaluation agent are in a machine-parseable structured format, facilitating the revision agent to parse the evaluation results and then make targeted revisions to the current round of translation based on the parsed results.

[0083] In some other embodiments, the generating agent, the evaluating agent, and the revising agent can be combined into two or one model, and different roles can be implemented by using different prompts, as long as a closed loop of "generation-evaluation-revision-convergence" can be achieved.

[0084] In this embodiment, a first draft translation is generated by a generating agent, an evaluation agent evaluates the current draft translation, and a revision agent revises the current draft translation based on the evaluation results. The revised draft translation undergoes a quality assessment, and the quality assessment results are fed back to the scheduling control module. The scheduling control module then determines whether to proceed to the next iteration or submit the revised draft translation to a human review terminal based on the quality assessment results. Translation iteration can be performed through the collaboration of multiple agents (generating, evaluating, and revising agents) to obtain high-quality translations, avoiding confirmation biases that may be caused by a single model or agent, and improving translation accuracy. Simultaneously, the quality assessment results are fed back to the scheduling control module, which can make appropriate decisions based on specific circumstances and effectively manage the collaboration between multiple agents.

[0085] In an optional embodiment, in step S206, a document-level evaluation is performed based on the first translation result, such as... Figure 6 As shown, it may include: Step S600: Perform a global consistency check based on the first translation result. The global consistency check includes consistency checks across sentences, paragraphs, chapters, and documents. The consistency check includes terminology consistency, sentence structure consistency, and style consistency checks.

[0086] In step S602, if the global consistency check fails, the first translation result is returned to the revision agent for revision.

[0087] Step S604: If the global consistency check passes, perform a document-level evaluation based on the first translation result.

[0088] When performing global consistency verification, a global dependency graph can be constructed first, and then the first translation result can be verified for global consistency based on the global dependency graph. The global dependency graph can include node types and edge types. Node types can include entities such as roles, events, skills, items, and place names, while edge types can include the occurrence relationships of entities in different text paragraphs or files. Specifically, the entity recognition capabilities of a large language model can be used to pre-scan the entire text, extract entities, and construct a preliminary topology graph, which can then be corrected and supplemented using a manually imported knowledge base.

[0089] Terminology consistency checks can be used to verify whether the translations of the same term are consistent in different positions; sentence structure consistency checks can be used to verify whether texts matching the same sentence structure template use consistent translations; style consistency checks can be used to calculate the style vector of the same character in different dialogues and detect whether there are significant shifts.

[0090] If the global consistency check fails, the first translation result is returned to the revision agent for revision. To facilitate the revision process, consistency suggestions can be generated and fed back to the revision agent. When the result is returned to the revision agent for revision, the revision can be counted in the total iteration count. If the total number of iterations reaches a preset number, iteration will be forcibly stopped to prevent an infinite loop. Additionally, to prevent infinite rollbacks, a maximum number of rollbacks can be set; if the maximum number of rollbacks is exceeded, the result can be submitted to a human review terminal for processing.

[0091] In this embodiment, a global consistency check is performed based on the first translation result. The global consistency check includes consistency checks across sentences, paragraphs, chapters, and documents, as well as consistency checks on terminology, sentence structure, and style. If the global consistency check fails, the first translation result is returned to the revision agent for revision. If the global consistency check passes, document-level evaluation and subsequent operations are performed. This ensures the consistency of the global translation and improves the accuracy of the translation.

[0092] In an optional embodiment, in step S206, if the document-level evaluation meets preset requirements, the translated text is obtained and output based on the first translation result, such as... Figure 7 As shown, it may include: Step S700: If the document-level evaluation meets the preset requirements, an online verification is performed based on the first translation result. The online verification includes at least one of the following: project format, code tag integrity, and user interface length.

[0093] Step S702: If the online verification fails and the conditions for automatic repair are met, the rule script is invoked to correct the first translation result.

[0094] In step S704, if the online verification fails and the conditions for automatic repair are not met, the first translation result is returned to the revision agent for revision.

[0095] To ensure that the initial translation meets engineering requirements before actual deployment, further online verification can be performed on the initial translation, provided that the document-level evaluation meets the preset requirements. Online verification includes at least one of the following: project format verification, code tag integrity verification, and user interface length verification. Specifically, project format verification checks whether it conforms to the target system or platform's format specifications, such as whether the structure of JSON, XML, or YAML is complete and whether key-value pairs match. Code tag integrity verification ensures that HTML tags, variable placeholders, links, code blocks, etc., in the original source language text have not been corrupted or lost during translation. User interface length verification checks whether the display of the initial translation in the user interface exceeds layout limitations.

[0096] If all validations pass, the final translated text can be obtained and output to the client. If the online validation fails, it can be further determined whether the conditions for automatic repair are met. If the conditions for automatic repair are met, a predefined rule script (such as regular expression replacement, template matching, etc.) can be called to perform batch or targeted corrections on the first translation result, such as automatically completing missing HTML tags or adjusting variable order. After the repair is completed, the online validation can be re-executed until it passes or it is transferred to manual revision. If the online validation fails and the conditions for automatic repair are not met, the first translation result is returned to the revision agent for revision. The conditions for automatic repair are, for example, that the problem type is a predetermined type, such as a missing comma at the end of JSON, or an incorrect color code format. The specific conditions for automatic repair can be set according to the actual situation.

[0097] In this embodiment, if the document-level evaluation meets the preset requirements, an online verification is performed based on the first translation result. If the online verification fails but the conditions for automatic repair are met, a rule script is invoked to correct the first translation result. If the online verification fails and the conditions for automatic repair are not met, the first translation result is returned to the agent for revision. Online verification can quickly determine whether the translation result meets the online requirements, and it can also quickly repair some scenarios that can be automatically repaired, thereby improving translation efficiency and reducing later maintenance costs.

[0098] In an optional embodiment, in step S304, a retrieval is performed based on the second source language text using a translation memory and a multidimensional knowledge base, and a second structured context constraint is generated based on the retrieval results, such as... Figure 8 As shown, it may include: Step S800: Determine the increment of the second source language text relative to the historical translation text.

[0099] Step S802: If the increment meets the preset conditions, the translation memory is used as a strong constraint. The translation memory and multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results.

[0100] Specifically, fuzzy matching algorithms (such as edit distance) can be used to compare the first source language text with the historical translated text to determine the degree of matching, thereby determining the increment of the first source language text relative to the historical translated text. Preset conditions can represent conditions within a limited range, such as the percentage of incremental content being below a certain threshold (e.g., 20%). After determining the increment, it can be judged whether the increment meets the preset conditions. If the increment meets the preset conditions, it means that the first source language text has only undergone minor adjustments or limited changes compared to the corresponding original text in the historical translated text. Therefore, when using translation memory and multidimensional knowledge bases for retrieval based on the first source language text, the translation memory can be used as a strong constraint, while the multidimensional knowledge base can be used as a relatively weak constraint, thus ensuring that the translation remains as consistent as possible with the historical translated text in terms of style, terminology, and sentence structure.

[0101] In this embodiment, by determining the increment of the first source language text relative to the historical translation text, and under the condition that the increment meets the preset conditions, the translation memory is used as a strong constraint to generate the first structured context constraint. This allows for translation as close as possible to the historical translation text when the modification range of the first source language text is limited, thereby ensuring that the translation is as consistent as possible with the historical translation text in terms of style, terminology, and sentence structure, thus improving the accuracy of the translation.

[0102] To make this application easier to understand, the following is combined with... Figure 9 and Figure 10 An example application is provided. Figure 9 This is a schematic diagram illustrating the principle framework of the translation method according to an embodiment of this application. Figure 10 for Figure 9 The corresponding flowchart example may generally include the following: 1. Assembly line scheduling system (i.e., scheduling control module).

[0103] This is the "central nervous system" of the system, responsible for resource scheduling of multiple models, coordinating the orderly operation of each module, and controlling abnormal circuit breakers, connecting each module into an organic whole.

[0104] Process orchestration: such as adjusting the order of modules, turning certain functions on or off.

[0105] Resource scheduling: Allocate computing resources (such as GPU / CPU) reasonably and manage the parallel execution of multiple tasks.

[0106] Iterative convergence control: Controlling the number of iterations and optimization strategies in the feedback loop to avoid infinite loops.

[0107] 2. Input layer: Items and text.

[0108] Source language text: Supports multiple formats (script, JSON, Excel, Lua, etc.).

[0109] Project configuration: including target language, text type, and role information.

[0110] 3. Sliding window semantic matching module.

[0111] Initial text segmentation and cleaning: Remove irrelevant characters and standardize the format.

[0112] Window construction: Dividing text into overlapping or non-overlapping "windows" (e.g., each paragraph or each few sentences).

[0113] Semantic vector computation: Each window is converted into a vector using an embedding model such as BERT.

[0114] Local window matching: Search for semantically similar past translations in the translation memory or knowledge base.

[0115] Alignment Result Output TU (Translation Unit): Outputs the translation unit, which is the corresponding segment of the source text and the historical translation.

[0116] Functional positioning: As a unified "pre-translation alignment entry point" of the system, it solves the problem of adding, deleting and rearranging text content across versions and platforms.

[0117] Input interface: Receives source language documents; if performing an asset building task, it also receives target language history documents; if performing a pure translation task, it only receives source documents. Output interface: Outputs high-quality aligned translation units as the basic data for building translation memory and multidimensional knowledge base; Core competency requirements: Robustness: It can tolerate content deletion, rearrangement, and quantity mismatch; Accuracy: Alignment accuracy must reach the system-set threshold (e.g., >95%). Traceability: Alignment confidence information is retained for use by downstream modules.

[0118] Interaction with the pipeline scheduling system: After alignment is completed, the status is reported to the pipeline scheduling system, which then decides whether to proceed to the next stage. 4. Multidimensional knowledge base collaborative retrieval module.

[0119] Terminology knowledge: Ensure consistent translation of proper nouns; Memory-based knowledge: Reusing high-quality historical translations; Sentence structure knowledge: Stores abstracted sentence templates and their corresponding translations to ensure the standardization of fixed expressions; Style-related knowledge: Maintaining the distinctive language of characters; Version-related knowledge: Supports incremental translation and asset traceability.

[0120] Functional positioning: To provide multi-dimensional constraint context for translation generation, solving the problem that a single knowledge source cannot guarantee translation quality.

[0121] Input interface: Receives the text to be translated and its metadata (text type, role ID, scene, etc.); Output interface: Outputs structured search results, including recall content, relevance scores, and constraint strength indicators for each dimension.

[0122] Core competency requirements: Differentiated retrieval: Adaptive retrieval strategies are used for different types of knowledge (e.g., incremental translation); Parallel Recall: Multi-dimensional retrieval is performed in parallel to ensure response efficiency; Quality stratification: distinguishing between strong constraints, weak constraints, and no constraints.

[0123] Data flow with other modules: Upstream: The TU aligned by the sliding window semantic matching module is used for the construction of the translation memory. Downstream: Provides structured context to generate and revise agents.

[0124] 5. Feedback-driven multi-model collaborative optimization module.

[0125] Character separation design: Generative agent: Responsible for generating the first translation draft based on the retrieval results of the multidimensional knowledge base collaborative retrieval module, focusing on the first translation generation "from scratch"; Evaluation agent: Responsible for multi-dimensional quality diagnosis, independent of the generating agent to avoid confirmation bias. Evaluation dimensions include: terminology consistency, style consistency, sentence structure standardization, semantic fidelity, contextual coherence, and UI / format correctness; Revising agent: Responsible for making targeted corrections based on the structured feedback from the evaluation agent, focusing on "local repair" rather than a complete rewrite, in order to maintain the stability of the correct parts.

[0126] Functional positioning: To realize an automated quality closed loop of "generation → evaluation → feedback → revision → convergence" and solve the problem of unstable quality in a single generation.

[0127] Input interface: Receives source text, search results from the multidimensional knowledge base collaborative retrieval module (including terminology constraints, translation memory references, sentence templates, and character style profiles), and quality target configuration; Output interface: Outputs translations that meet the quality threshold, along with a complete iterative process record and quality assessment report.

[0128] Core competency requirements: Structured Feedback: The evaluation results of the agent must be in a machine-parseable structured format (JSON), including fields such as error type, error location, severity, and modification suggestions, so that the agent can perform automated revisions. Targeted revision: The revised agent only modifies the parts that raise issues, while keeping the other correct parts unchanged; Sentence pattern library constraint: When the source language text matches a sentence pattern template in the sentence pattern knowledge base, the generating agent and the revising agent must give priority to the translation corresponding to the template; Convergence guarantee: A maximum iteration round mechanism is introduced (default N=3). If the quality threshold is not reached within the specified number of rounds, the system will forcibly terminate the iteration, output the current best version and mark it as "low confidence", and transfer it to the manual review queue.

[0129] Interaction with the pipeline scheduling system: The pipeline scheduling system configures iteration parameters (maximum number of rounds, quality threshold, etc.), and the feedback-driven multi-model collaborative optimization module reports the status to the pipeline scheduling system after each iteration, so that the pipeline scheduling system can make a decision to pass, continue, or circuit breaker.

[0130] 6. Context consistency control module.

[0131] Core processing flow: Global dependency graph construction: Node types: Entities such as characters, events, skills / items, and place names; Edge type: The relationship of an entity's appearance in different text paragraphs / files; Construction method: The entity recognition capability of the large language model is used to pre-scan the whole text, extract entities and construct a preliminary topology map, and then combined with manually imported knowledge bases (terminology base, role base) for correction and supplementation.

[0132] Consistency conflict detection: Terminology consistency: Check whether the translations of the same term are consistent in different positions; Sentence consistency: Check whether texts matching the same sentence template use consistent translation. Style consistency: Calculate the style vector of the same character in different dialogues and detect whether there is a significant shift.

[0133] Conflict resolution strategies: If a conflict is detected, a consensus proposal is generated and a revision agent is triggered to perform a local rewrite. The rewritten result will be fed back to the evaluation agent for rapid verification. This rewrite action will be included in the total iteration count of the multi-model collaborative optimization module. If the maximum number of rounds is exceeded, the iteration will be forcibly stopped to prevent it from falling into an infinite loop. To prevent infinite rollbacks, the pipeline scheduling system sets a maximum number of rollbacks (default M=2). If this limit is exceeded, the system is marked as "consistency pending" and transferred to manual review.

[0134] Functional positioning: After the multi-model collaborative optimization module achieves closed-loop convergence, it performs consistency checks on the translation across sentences, paragraphs, chapters, and files to ensure the uniformity of the global translation.

[0135] Input interface: Receives the translation output from the multi-model collaborative optimization module, the global dependency graph, and the terminology and style library from the multi-dimensional knowledge base collaborative retrieval module; Output interface: Outputs the translation that has passed the consistency check, or the translation to be reviewed with conflict markers.

[0136] 7. Output layer: Results and validation.

[0137] The document-level quality assessment module uses a hybrid approach of large language model as the assessment model and rule detection to perform document-level quality assessment on the final candidate translations.

[0138] Working mechanism: The evaluation results of the document-level quality assessment module do not directly participate in the generation or partial modification of the translation, but only serve as the decision-making basis for the pipeline scheduling system.

[0139] Routing strategy: The pipeline scheduling system determines the routing direction based on the document-level quality assessment module score. Pass: If the score is higher than the threshold, it is allowed to enter the online verification module for engineering verification; Blocking: If the score is below the threshold (such as the discovery of serious cross-paragraph logical errors or tone inconsistencies), the system determines that the cost of automated repair is too high. Instead of reverting to the multi-model collaborative optimization module, the system directly triggers the circuit breaker, marks the batch of tasks as "requiring manual review", and transfers them to the manual review process.

[0140] Online verification module: Functional positioning: To perform hard checks on the final translated resource package, such as project format, code tag integrity, and UI length, which are not semantic.

[0141] A dual strategy of "rule repair + blocking alarm" is adopted: Automatic repair: For simple formatting errors (such as missing commas at the end of JSON or incorrect color code formatting), the rule script is invoked to automatically correct them; Rollback or blocking: For semantic-related errors such as UI length overflow, rollback can be performed to multi-model collaborative optimization module rewriting; for engineering errors that cannot be automatically corrected (such as missing key tags), output can be blocked directly and an alarm can be triggered, entering the manual processing flow.

[0142] Example 2 Figure 11 The diagram schematically illustrates a translation system according to Embodiment 2 of this application. This system can be divided into one or more program modules. One or more program modules are stored in a storage medium and executed by one or more processors to complete the embodiments of this application. The program module referred to in the embodiments of this application refers to a series of computer program instruction segments capable of performing a specific function. The following description will specifically introduce the functions of each program module in this embodiment. For example... Figure 11 As shown, the system 900 may include: a retrieval module 910, a generation module 920, an iteration module 930, and an output module 940, wherein: The retrieval module 910 is used to receive the first source language text, perform retrieval based on the first source language text using a multidimensional knowledge base, and generate a first structured context constraint based on the retrieval results. Generation module 920 is used to generate a first translation draft based on the first structural context constraints; The iteration module 930 is used to take the first translation draft as the translation draft of the current round at the start of the iteration, and use the scheduling control module to perform translation iteration. If the translation draft after iteration meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft of the current round, revising the translation draft of the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round for further iteration. The output module 940 is used to perform a document-level evaluation based on the first translation result, and if the document-level evaluation meets preset requirements, to obtain and output the translated text based on the first translation result.

[0143] In an optional embodiment, the system 900 is further used for: Receive the second source language text and the corresponding historical translation text; The second source language text is aligned with the historical translation text using a sliding window semantic alignment method to obtain aligned translation units, and a translation memory is constructed based on the aligned translation units. The translation memory and the multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results; A second initial translation draft is generated based on the second structural context constraints; The second initial translation draft is used as the translation draft for the current round at the start of the iteration. The scheduling control module is used to perform translation iterations. If the translated draft after the iteration meets the preset quality requirements or reaches the preset number of iterations, the second translation result of the second source language text is obtained. Based on the second translation result, a document-level evaluation is performed. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the second translation result.

[0144] In an optional embodiment, the system 900 is further used for: The complexity of the first source language text is determined using the scheduling control module. If the complexity assessment result is simple text, a document-level evaluation is performed based on the first translation draft. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation draft. If the complexity assessment result indicates that the text is not simple, the first translation draft is used as the translation draft for the current round at the start of the iteration, and the translation iteration and subsequent steps are performed using the scheduling control module.

[0145] In an optional embodiment, the generation module 920 is further configured to: The first structured context constraint and the first source language text are input into the generative agent, and the generative agent is used to generate the first translation draft. Correspondingly, the iteration module 930 is also used for: The translation of the current round is input into the evaluation agent, which evaluates the translation of the current round to obtain the evaluation result. The evaluation results are fed back to the revision agent, which then revises the translation of the current round to obtain the revised translation. The revised translation is evaluated for quality, and the results are fed back to the scheduling and control module. The scheduling and control module then determines whether to proceed to the next iteration or submit the revised translation to the manual review terminal based on the evaluation results.

[0146] In an optional embodiment, the iteration module 930 is further configured to: The evaluation agent is used to evaluate the translation of the current round and generate a structured evaluation result, which includes error type, error location, severity, and revision suggestions.

[0147] In an optional embodiment, the output module 940 is further configured to: A global consistency check is performed based on the first translation result. The global consistency check includes consistency checks across sentences, paragraphs, chapters, and documents. The consistency check includes terminology consistency, sentence structure consistency, and style consistency checks. If the global consistency check fails, the first translation result is returned to the revision agent for revision. If the global consistency check passes, a document-level evaluation is performed based on the first translation result.

[0148] In an optional embodiment, the output module 940 is further configured to: If the document-level evaluation meets the preset requirements, an online verification is performed based on the first translation result. The online verification includes at least one of the following: project format, code tag integrity, and user interface length. If the online verification fails but the conditions for automatic repair are met, the rule script is invoked to correct the first translation result; If the online verification fails and the conditions for automatic repair are not met, the first translation result is returned to the revision agent for revision.

[0149] In an optional embodiment, the system 900 is further used for: Determine the increment of the second source language text relative to the historical translated text; When the increment meets the preset conditions, the translation memory is used as a strong constraint. The translation memory and the multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results.

[0150] In an optional embodiment, the multidimensional knowledge base includes at least two of the following: a terminology knowledge base, a sentence structure knowledge base, a style knowledge base, and a version knowledge base.

[0151] Example 3 Figure 12 This illustration schematically shows a hardware architecture diagram of a computer device 10000 suitable for implementing a translation method according to Embodiment 3 of this application. In some embodiments, the computer device 10000 may be a terminal device such as a smartphone, wearable device, tablet computer, personal computer, in-vehicle terminal, game console, virtual device, workbench, digital assistant, set-top box, robot, etc. In other embodiments, the computer device 10000 may be a rack server, blade server, tower server, or cabinet server (including independent servers or server clusters composed of multiple servers), etc. Figure 12As shown, the computer device 10000 includes, but is not limited to: a memory 10010, a processor 10020, and a network interface 10030 that can communicate and be linked with each other via a system bus. Wherein: The memory 10010 includes at least one type of computer-readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 10010 may be an internal storage module of a computer device 10000, such as the hard disk or memory of the computer device 10000. In other embodiments, the memory 10010 may also be an external storage device of the computer device 10000, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 10000. Of course, the memory 10010 may also include both the internal storage module and the external storage device of the computer device 10000. In this embodiment, the memory 10010 is typically used to store the operating system and various application software installed on the computer device 10000, such as the program code for translation methods. In addition, the memory 10010 can also be used to temporarily store various types of data that have been output or will be output.

[0152] In some embodiments, processor 10020 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other chip. Processor 10020 is typically used to control the overall operation of computer device 10000, such as performing control and processing related to data interaction or communication with computer device 10000. In this embodiment, processor 10020 is used to run program code stored in memory 10010 or process data.

[0153] Network interface 10030 may include a wireless network interface or a wired network interface, which is typically used to establish a communication link between computer device 10000 and other computer devices. For example, network interface 10030 is used to connect computer device 10000 to an external terminal via a network, establishing a data transmission channel and communication link between computer device 10000 and the external terminal. The network may be an intranet, the Internet, Global System for Mobile Communication (GSM), Wideband Code Division Multiple Access (WCDMA), 4G network, 5G network, Bluetooth, Wi-Fi, or other wireless or wired networks.

[0154] It should be pointed out that, Figure 12 Only computer devices with components 10010-10030 are shown; however, it should be understood that it is not required to implement all of the shown components, and more or fewer components may be implemented instead.

[0155] In this embodiment, the translation method stored in memory 10010 can be further divided into one or more program modules and executed by one or more processors (such as processor 10020) to complete the embodiment of this application.

[0156] Example 4 This application also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the translation method in the embodiments.

[0157] In this embodiment, the computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the computer-readable storage medium can be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device. Of course, the computer-readable storage medium can also include both the internal storage unit and the external storage device of the computer device. In this embodiment, the computer-readable storage medium is typically used to store the operating system and various application software installed on the computer device, such as the program code of the translation method in the embodiment. In addition, the computer-readable storage medium can also be used to temporarily store various types of data that have been output or will be output.

[0158] Example 5 This application also provides a computer program product, including a computer program that, when executed by a processor, implements the methods described in the above embodiments.

[0159] Obviously, those skilled in the art should understand that the modules or steps of the embodiments of this application described above can be implemented using general-purpose computer devices. They can be centralized on a single computer device or distributed across a network of multiple computer devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computer device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of this application are not limited to any particular combination of hardware and software.

[0160] It should be noted that the above are merely preferred embodiments of this application and do not limit the scope of patent protection of this application. Any equivalent structural or procedural changes made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A translation method, characterized in that, The method includes: Receive the first source language text, use a multidimensional knowledge base to search based on the first source language text, and generate the first structured context constraint based on the search results; A first draft translation is generated based on the first structural context constraints; The first initial translation draft is used as the translation draft for the current round at the start of the iteration. The translation is iterated using the scheduling control module. If the translated draft meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft for the current round, revising the translation draft for the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round for further iteration. Based on the first translation result, a document-level evaluation is performed. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation result.

2. The method according to claim 1, characterized in that, The method further includes: Receive the second source language text and the corresponding historical translation text; The second source language text is aligned with the historical translation text using a sliding window semantic alignment method to obtain aligned translation units, and a translation memory is constructed based on the aligned translation units. The translation memory and the multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results; A second initial translation draft is generated based on the second structural context constraints; The second initial translation draft is used as the translation draft for the current round at the start of the iteration. The scheduling control module is used to perform translation iterations. If the translated draft after the iteration meets the preset quality requirements or reaches the preset number of iterations, the second translation result of the second source language text is obtained. Based on the second translation result, a document-level evaluation is performed. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the second translation result.

3. The method according to claim 1, characterized in that, Before using the scheduling control module to perform translation iterations with the first initial translation draft as the translation draft for the current round at the start of the iteration, the method further includes: The complexity of the first source language text is determined using the scheduling control module. If the complexity assessment result is simple text, a document-level evaluation is performed based on the first translation draft. If the document-level evaluation meets the preset requirements, the translated text is obtained and output based on the first translation draft. If the complexity assessment result indicates that the text is not simple, the first translation draft is used as the translation draft for the current round at the start of the iteration, and the translation iteration and subsequent steps are performed using the scheduling control module.

4. The method according to claim 1, characterized in that, The process of generating the first translation draft based on the first structural context constraints includes: The first structured context constraint and the first source language text are input into the generative agent, and the generative agent is used to generate the first translation draft. Correspondingly, the step of using the first initial translation draft as the translation draft for the current round at the start of the iteration, and performing translation iterations using the scheduling control module, includes: The translation of the current round is input into the evaluation agent, which evaluates the translation of the current round to obtain the evaluation result. The evaluation results are fed back to the revision agent, which then revises the translation of the current round to obtain the revised translation. The revised translation is evaluated for quality, and the results are fed back to the scheduling and control module. The scheduling and control module then determines whether to proceed to the next iteration or submit the revised translation to the manual review terminal based on the evaluation results.

5. The method according to claim 4, characterized in that, The evaluation of the translation in the current round using the evaluation agent to obtain the evaluation result includes: The evaluation agent is used to evaluate the translation of the current round and generate a structured evaluation result, which includes error type, error location, severity, and revision suggestions.

6. The method according to claim 4, characterized in that, The document-level evaluation based on the first translation result includes: A global consistency check is performed based on the first translation result. The global consistency check includes consistency checks across sentences, paragraphs, chapters, and documents. The consistency check includes terminology consistency, sentence structure consistency, and style consistency checks. If the global consistency check fails, the first translation result is returned to the revision agent for revision. If the global consistency check passes, a document-level evaluation is performed based on the first translation result.

7. The method according to claim 4, characterized in that, If the document-level evaluation meets preset requirements, the translated text is obtained and output based on the first translation result, including: If the document-level evaluation meets the preset requirements, an online verification is performed based on the first translation result. The online verification includes at least one of the following: project format, code tag integrity, and user interface length. If the online verification fails but the conditions for automatic repair are met, the rule script is invoked to correct the first translation result; If the online verification fails and the conditions for automatic repair are not met, the first translation result is returned to the revision agent for revision.

8. The method according to claim 2, characterized in that, The step of using the translation memory and multidimensional knowledge base to perform retrieval based on the second source language text, and generating a second structured context constraint based on the retrieval results, includes: Determine the increment of the second source language text relative to the historical translated text; When the increment meets the preset conditions, the translation memory is used as a strong constraint. The translation memory and the multidimensional knowledge base are used to perform retrieval based on the second source language text, and a second structured context constraint is generated based on the retrieval results.

9. The method according to any one of claims 1-7, characterized in that, The multidimensional knowledge base includes at least two of the following: a terminology knowledge base, a sentence structure knowledge base, a style knowledge base, and a version knowledge base.

10. A translation system, characterized in that, The system includes: The retrieval module is used to receive the first source language text, perform retrieval based on the first source language text using a multidimensional knowledge base, and generate the first structured context constraint based on the retrieval results. The generation module is used to generate a first translation draft based on the first structural context constraints; An iteration module is used to take the first initial translation draft as the translation draft for the current round at the start of the iteration, and use the scheduling control module to perform translation iteration. If the translation draft after iteration meets the preset quality requirements or reaches the preset number of iterations, the first translation result of the first source language text is obtained. The translation iteration process includes evaluating the translation draft for the current round, revising the translation draft for the current round based on the evaluation results, and if the revised translation draft does not meet the preset quality requirements, the revised translation draft is used as the translation draft to be evaluated in the next round for continued iteration. The output module is used to perform a document-level evaluation based on the first translation result, and if the document-level evaluation meets preset requirements, to obtain and output the translated text based on the first translation result.

11. A computer device, characterized in that, include: At least one processor; and A memory communicatively connected to the at least one processor; wherein: The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in any one of claims 1 to 9.

13. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 9.

Images