Multi-language translation method based on large language model agent and related equipment

Abstract

The invention discloses a multi-language translation method based on a large language model agent and related equipment, and relates to the technical field of natural language processing, and the method comprises the steps: obtaining a to-be-translated text and a plurality of language translation targets, so as to determine a plurality of target agents used for translating the to-be-translated text, the method comprises the steps of translating standard files of different language versions in parallel on the basis of a preset large model base and a plurality of target agents to obtain a plurality of initial translation results, and performing calibration and disambiguation operations on the plurality of initial translation results on the basis of a preset single-language calibration agent to obtain a target translation result corresponding to each language target. Based on different language translation targets, a plurality of corresponding target agents are adopted to execute the translation task in parallel, the multi-language translation efficiency is improved, meanwhile, calibration and disambiguation operations are executed in combination with translation results corresponding to standard files of different language versions, the quality of multi-language translation results is improved, and the multi-language translation efficiency is improved. And high-quality and high-efficiency multi-language translation is realized.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular to multilingual translation methods and related devices based on large language model intelligent agents. Background Technology

[0002] As the globalization of digital content such as software and games accelerates, products going global need to translate text content into more than ten languages, including many less common ones. Currently, mainstream solutions still heavily rely on human translation and verification: multilingual professionals are required to cross-check terminology, eliminate ambiguity, unify style, and handle dynamic text (such as variables and placeholders in code).

[0003] This model has significant drawbacks: First, the process is cumbersome and time-consuming, with a single round of translation deployment often taking several months; second, there is a shortage of translators and disambiguators proficient in multiple languages, especially less commonly spoken languages, resulting in high costs; third, manual collaboration makes it difficult to ensure consistency of terminology across multiple languages, and the judgment and handling of textual ambiguity are highly subjective and inefficient. Therefore, there is an urgent need for a multilingual translation calibration and disambiguation solution that is automated, efficient, low-cost, and guarantees translation quality and consistency. Summary of the Invention

[0004] The main purpose of this application is to provide a multilingual translation method and related equipment based on a large language model intelligent agent, aiming to solve the technical problem of how to perform multilingual translation with high quality and efficiency.

[0005] To achieve the above objectives, this application proposes a multilingual translation method based on a large language model intelligent agent, which includes: In response to a multilingual translation instruction, the system acquires the text to be translated and multiple language translation targets, wherein the text to be translated contains at least two different language versions of standard translation text. Based on the multiple language translation targets, multiple target agents are determined for translating the text to be translated; Based on the preset large model base and the multiple target agents, the standard files of different language versions in the text to be translated are translated in parallel to obtain multiple initial translation results corresponding to each language target. The multiple initial translation results are aligned to obtain multiple aligned initial translation results. Based on a preset single-language calibration agent, calibration and disambiguation operations are performed on the multiple initial translation results after alignment to obtain the target translation result corresponding to each language target.

[0006] In one embodiment, the step of performing calibration and disambiguation operations on the multiple aligned initial translation results based on a preset single-language calibration agent to obtain the target translation result corresponding to each language target further includes: Obtain the background knowledge, terminology database, and context of the text to be translated corresponding to the multilingual translation instructions; Based on the background knowledge, the terminology database, the context of the text to be translated, and the preset scoring rules, a preset monolingual calibration agent is used to perform translation quality polysemy scoring and translation quality suitability scoring operations on multiple initial translation results after alignment, so as to obtain the polysemy score and suitability score corresponding to each initial translation result. Based on the polysemy score and the adaptability score, the translation quality corresponding to each initial translation result is determined; Based on the translation quality, the target translation result corresponding to each language target is obtained.

[0007] In one embodiment, the step of obtaining the target translation result corresponding to each language target based on the translation quality further includes: Based on the translation quality, determine whether the initial translation result needs to be retranslated; If all initial translation target results need to be retranslated, then the text to be translated is retranslated to update multiple initial translation results, and the translation quality corresponding to each initial translation result is updated; Based on the updated translation quality, the process returns to the step of determining whether the initial translation result needs to be retranslated based on the translation quality, until there is an initial translation result among all the initial translation target results that does not need to be retranslated, and the target translation result is obtained.

[0008] In one embodiment, after the step of determining whether the initial translation result needs to be retranslated based on the translation quality, the method further includes: If there are initial translation results that need to be retranslated, construct an agent-optimized dataset based on the initial translation results that need to be retranslated; Based on the agent optimization dataset, optimize the target agent, the preset monolingual calibration agent, and the preset scoring rules.

[0009] In one embodiment, the step of obtaining the target translation result until there is an initial translation result among all the initial translation target results that does not need to be retranslated further includes: If there is only one initial translation result among all the initial translation target results that does not need to be retranslated, then the initial translation result that does not need to be retranslated is set as the target translation result; If there are multiple initial translation results that do not require retranslation among all initial translation target results, the initial translation result with the highest adaptability score among these multiple initial translation results that do not require retranslation is set as the target translation result.

[0010] In one embodiment, before the step of performing calibration and disambiguation operations on all the aligned initial translation results based on a preset monolingual calibration agent, the method further includes: Acquire sample data, which corresponds to the first data processing result; The sample data is processed using the current monolingual calibration agent to obtain a second data processing result; Determine whether the first data processing result is consistent with the second data processing result; If the first data processing result is inconsistent with the second data processing result, the parameters of the current monolingual calibration agent are adjusted. Based on the current monolingual calibration agent with adjusted parameters, the process of using the current monolingual calibration agent to process the sample data and obtain the second data processing result is returned until the first data processing result is consistent with the second data processing result, thus obtaining the preset monolingual calibration agent.

[0011] Furthermore, to achieve the above objectives, this application also proposes a multilingual translation device based on a large language model intelligent agent, the multilingual translation device based on a large language model intelligent agent comprising: The acquisition module is used to acquire the text to be translated and multiple language translation targets in response to a multilingual translation instruction; The determining module is used to determine multiple target agents for translating the text to be translated based on the multiple language translation targets; The translation module is used to translate standard files in different language versions of the text to be translated in parallel based on a preset large model base and the multiple target agents, to obtain multiple initial translation results corresponding to each language target, and to align the multiple initial translation results to obtain multiple aligned initial translation results; The calibration and disambiguation module is used to perform calibration and disambiguation operations on multiple initial translation results after alignment based on a preset single-language calibration agent, so as to obtain the target translation result corresponding to each language target.

[0012] In one embodiment, the calibration and disambiguation module further includes: The first acquisition unit is used to acquire the background knowledge, terminology database, and context of the text to be translated corresponding to the multilingual translation instructions. The scoring unit is used to perform translation quality polysemy scoring and translation quality suitability scoring operations on multiple initial translation results after alignment, based on the background knowledge, the terminology database, the context of the text to be translated, and preset scoring rules, using a preset monolingual calibration agent, to obtain the polysemy score and suitability score corresponding to each initial translation result. The first determining unit is used to determine the translation quality corresponding to each initial translation result based on the polysemy score and the adaptability score. The first obtaining unit is used to obtain the target translation result corresponding to each language target based on the translation quality.

[0013] In one embodiment, the calibration and disambiguation module further includes: The first judgment unit is used to determine, based on the translation quality, whether the initial translation result needs to be retranslated; The retranslation unit is used to retranslate the text to be translated if all initial translation target results need to be retranslated, so as to update multiple initial translation results and update the translation quality corresponding to each initial translation result; The second obtaining unit is used to return a step based on the updated translation quality to determine whether the initial translation result needs to be retranslated, until there is an initial translation result among all the initial translation target results that does not need to be retranslated, and then obtain the target translation result.

[0014] In one embodiment, the calibration and disambiguation module further includes: The building unit is used to construct an agent-optimized dataset based on the initial translation results that need to be retranslated if there are initial translation results that need to be retranslated. An optimization unit is used to optimize the target agent, the preset monolingual calibration agent, and the preset scoring rules based on the agent optimization dataset.

[0015] In one embodiment, the calibration and disambiguation module further includes: The first setting unit is used to set the initial translation result that does not need to be retranslated as the target translation result if there is only one initial translation result that does not need to be retranslated among all the initial translation target results; The second setting unit is used to set the initial translation result with the highest adaptability score among the multiple initial translation results that do not need to be retranslated as the target translation result if there are multiple initial translation results that do not need to be retranslated.

[0016] In one embodiment, the multilingual translation device based on a large language model agent further includes an agent training module, which in turn includes: The second acquisition unit is used to acquire sample data, which corresponds to the first data processing result. A data processing unit is used to process the sample data using the current monolingual calibration agent to obtain a second data processing result; The second judgment unit is used to determine whether the first data processing result is consistent with the second data processing result; An iterative training unit is used to adjust the parameters of the current monolingual calibration agent if the first data processing result is inconsistent with the second data processing result. Based on the current monolingual calibration agent with adjusted parameters, the unit returns to the step of using the current monolingual calibration agent to process the sample data and obtain the second data processing result, until the first data processing result is consistent with the second data processing result, thus obtaining a preset monolingual calibration agent.

[0017] Furthermore, to achieve the above objectives, this application also proposes a multilingual translation device based on a large language model intelligent agent. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the multilingual translation method based on a large language model intelligent agent as described above.

[0018] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the multilingual translation method based on a large language model intelligent agent as described above.

[0019] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the multilingual translation method based on a large language model agent as described above.

[0020] One or more technical solutions proposed in this application have at least the following technical effects: This application proposes a multilingual translation method and related equipment based on a large language model intelligent agent, relating to the field of natural language processing technology. In related technologies, mainstream solutions still heavily rely on manual translation and calibration: multilingual professionals are required to cross-check terminology, eliminate ambiguity, unify style, and process dynamic text (such as variables and placeholders in code). This model has significant drawbacks: firstly, the process is cumbersome and time-consuming, with a single round of translation deployment often taking months; secondly, multilingual, especially minority language, translators and calibrators are scarce and costly; thirdly, manual collaboration makes it difficult to guarantee consistency of terminology across languages, and the judgment and handling of textual ambiguity are highly subjective and inefficient. In contrast, this application, firstly, responds to multilingual translation guidance... The process involves obtaining a text to be translated and multiple language translation targets, wherein the text to be translated contains standard translation texts in at least two different language versions. Then, based on the multiple language translation targets, multiple target agents are determined for translating the text to be translated. Further, based on a preset large model base and the multiple target agents, the standard documents in different language versions of the text to be translated are translated in parallel to obtain multiple initial translation results corresponding to each language target. The multiple initial translation results are aligned to obtain multiple aligned initial translation results. Finally, based on a preset single-language calibration agent, calibration and disambiguation operations are performed on all the aligned initial translation results to obtain the target translation result corresponding to each language target. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating an embodiment of the multilingual translation method based on a large language model intelligent agent in this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the multilingual translation method based on a large language model intelligent agent in this application; Figure 3 This is a flowchart illustrating Embodiment 3 of the multilingual translation method based on a large language model intelligent agent in this application; Figure 4 This is a schematic diagram of the module structure of the multilingual translation device based on a large language model intelligent agent according to an embodiment of this application; Figure 5This is a schematic diagram of the device structure of the hardware operating environment involved in the multilingual translation method based on a large language model intelligent agent in the embodiments of this application.

[0024] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0026] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0027] The main solution in this application's embodiments is: In this embodiment, for ease of description, the following description uses a multilingual translation device based on a large language model agent as the execution subject.

[0028] Due to the relevant technologies, the mainstream solutions still heavily rely on human translation and calibration: multilingual professionals are required to cross-check terminology, eliminate ambiguity, unify style, and handle dynamic text (such as variables and placeholders in code). This model has significant drawbacks: First, the process is cumbersome and time-consuming, with a single round of translation deployment often taking several months; second, multilingual, especially less common, translators and calibrators are scarce and costly; third, human collaboration makes it difficult to ensure consistency of terminology across multiple languages, and the judgment and handling of textual ambiguity are highly subjective and inefficient.

[0029] This application provides a solution in which: first, in response to a multilingual translation instruction, a text to be translated and multiple language translation targets are obtained, wherein the text to be translated contains at least two different language versions of standard translation text; then, based on the multiple language translation targets, multiple target agents are determined for translating the text to be translated; further, based on a preset large model base and the multiple target agents, the standard documents of different language versions in the text to be translated are translated in parallel to obtain multiple initial translation results corresponding to each language target; the multiple initial translation results are aligned to obtain multiple aligned initial translation results; finally, based on a preset single-language calibration agent, calibration and disambiguation operations are performed on all the aligned multiple initial translation results to obtain the target translation result corresponding to each language target.

[0030] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, a multilingual translation device based on a large language model intelligent agent, etc. The following description uses a multilingual translation device based on a large language model intelligent agent as an example to illustrate this embodiment and the subsequent embodiments.

[0031] Based on this, embodiments of this application provide a multilingual translation method based on a large language model agent, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the multilingual translation method based on a large language model intelligent agent according to this application.

[0032] In this embodiment, the multilingual translation method based on a large language model agent includes steps S10 to S40: Step S10: In response to a multilingual translation instruction, obtain the text to be translated and multiple language translation targets, wherein the text to be translated contains at least two different language versions of standard translation text; It should be noted that multilingual translation commands refer to instructions issued by users to translate a piece of text into multiple different languages.

[0033] It should be noted that the text to be translated refers to the original text that needs to be translated.

[0034] It should be noted that the language translation target refers to the language into which the text to be translated is specified by the user.

[0035] It should be noted that a standard translation text refers to a high-quality translation that conforms to language norms and conventions, and can be used as a reference standard.

[0036] In this embodiment, after receiving a multilingual translation instruction from the user, the system first extracts the content of the text to be translated from the instruction, and simultaneously obtains multiple language translation targets specified by the user. Then, the extracted text to be translated is analyzed to confirm that it contains standard translation text in at least two different language versions.

[0037] Understandably, in this embodiment, accurately obtaining the user's translation requirements, including the text to be translated and the target language, provides a clear direction for subsequent translation work. The text to be translated contains standard translation texts in multiple languages, providing high-quality references for subsequent translation and helping to improve the accuracy and standardization of the translation results.

[0038] Step S20: Based on the multiple language translation targets, determine multiple target agents for translating the text to be translated; It should be noted that a target intelligent agent refers to an intelligent translation model or system designed to complete a specific translation task. Different target intelligent agents exist for different language translation targets. It should also be noted that in this application, "cognate languages ​​+ minor languages" are combined into one intelligent agent, for example, one agent for "Spanish, Portuguese, and Italian," one agent for "Japanese and Korean," and one agent for "Nordic minor languages."

[0039] In this embodiment, based on the multiple language translation targets obtained in step S10, the system selects multiple target agents from its internal agent library that match these target languages. Each target agent is specifically responsible for translating the text to be translated into the corresponding target language.

[0040] Understandably, in this embodiment, assigning a dedicated agent to each target language fully leverages the strengths of each agent in specific language translation, improving translation accuracy and efficiency. Selecting appropriate target agents based on different language goals avoids the resource waste and reduced translation quality that can result from using a single agent for multilingual translation.

[0041] Step S30: Based on the preset large model base and the multiple target agents, translate the standard files of different language versions in the text to be translated in parallel to obtain multiple initial translation results corresponding to each language target. Align the multiple initial translation results to obtain multiple aligned initial translation results. It should be noted that the pre-set large model base refers to a pre-set basic model with powerful language understanding and generation capabilities, which provides underlying support for translation work.

[0042] It should be noted that parallel translation refers to translating standard documents in multiple different language versions simultaneously, rather than translating them one by one.

[0043] It should be noted that alignment refers to matching the initial translation results of different target languages ​​according to their content, so that they can correspond to each other in terms of structure and semantics.

[0044] Understandably, in this embodiment, based on a pre-defined large model base, multiple target agents determined in step S20 are used to simultaneously translate standard documents in different language versions of the text to be translated in step S10. Each target agent is responsible for translating into the corresponding target language, obtaining multiple initial translation results for each language target. Then, these initial translation results in different target languages ​​are aligned so that they correspond to each other in content and structure, resulting in multiple aligned initial translation results.

[0045] Understandably, in this embodiment, parallel translation is used to process translation tasks for multiple language versions simultaneously, significantly reducing translation time and improving efficiency. The aligned initial translation results facilitate subsequent calibration and disambiguation operations, allowing for cross-referencing between translation results of different language versions, making it easier to identify and correct errors.

[0046] Step S40: Based on a preset single-language calibration agent, calibration and disambiguation operations are performed on the multiple initial translation results after alignment to obtain the target translation result corresponding to each language target.

[0047] It should be noted that the preset single-language calibration agent refers to an agent specifically designed to calibrate and disambiguate the translation results of a single language. There is a corresponding single-language calibration agent for each target language.

[0048] It should be noted that calibration refers to checking and adjusting the translation results to make them more consistent with the grammar, expression habits, and semantic requirements of the target language.

[0049] It should be noted that disambiguation refers to eliminating potential ambiguities in the translation results, making the translation more explicit and accurate.

[0050] In this embodiment, a preset monolingual calibration agent is used to calibrate and disambiguate the multiple initial translation results obtained in step S30 after alignment. Each monolingual calibration agent processes the corresponding target language, correcting and adjusting grammatical errors, unnatural expressions, and semantic ambiguities in the translation results, ultimately obtaining the target translation result for each language target.

[0051] Understandably, in this embodiment, calibration and disambiguation operations make the translation results more consistent with the norms and conventions of the target language, improving the accuracy and readability of the translation. Disambiguation operations can eliminate multiple possible interpretations in the translation results, making them clearer and more explicit, and avoiding misunderstandings caused by ambiguity.

[0052] Specifically, the step of performing calibration and disambiguation operations on the multiple initial translation results after alignment based on a preset single-language calibration agent to obtain the target translation result corresponding to each language target further includes steps S41 to S44: Step S41: Obtain the background knowledge, terminology database, and context of the text to be translated corresponding to the multilingual translation instructions; It should be noted that background knowledge refers to external information such as domain common sense, scenario descriptions, and user intent related to this translation task.

[0053] A terminology database is a standardized bilingual or multilingual glossary that contains specialized vocabulary and their designated translations.

[0054] Context refers to the paragraphs, sentences, or conversational turns that are adjacent to the text to be translated, and is used to identify contextual features such as reference, tone, and domain of the current sentence.

[0055] In this embodiment, metadata fields (such as domain, use-case, and user-id) in the multilingual translation instructions are parsed to extract corresponding background knowledge from the enterprise knowledge graph or task cards. Simultaneously, a terminology database is retrieved using domain tags, and a bilingual terminology list matching the current task is loaded into memory. The text to be translated, along with its preceding and following N sentences (or adjacent paragraphs of the same document), is packaged into a context fragment, and sentence boundary recognition and tokenization are performed to form a structured context object.

[0056] Understandably, in this embodiment, "naked text" is upgraded to "contextual information packages," so that subsequent scoring no longer considers individual sentences in isolation, reducing the probability of domain misuse and incorrect referencing. The terminology database pre-determines the unique translation of key concepts, reducing subsequent scoring noise caused by multiple translations of the same word.

[0057] Step S42: Based on the background knowledge, the terminology database, the context of the text to be translated, and the preset scoring rules, a preset monolingual calibration agent is used to perform translation quality polysemy scoring and translation quality suitability scoring operations on the multiple initial translation results after alignment, so as to obtain the polysemy score and suitability score corresponding to each initial translation result; It should be noted that the polysemy score refers to the quantification of the tendency of a translation to have multiple reasonable interpretations in the same context; the higher the score, the greater the ambiguity.

[0058] It should be noted that the applicability score refers to the quantitative degree of fit between the translation and the background knowledge, terminology database, and context in terms of grammar, pragmatics, and style. The higher the score, the more "fitting" the translation is.

[0059] It should be noted that the preset scoring rules refer to a set of rules or a fine-tuned discriminative model written with interpretable features (term hit rate, reference consistency, syntactic validity, style tag matching degree, etc.) and their weights.

[0060] The specific implementation process is as follows: First, the monolingual calibration agent generates a feature vector for each initial translation. The rule engine maps the feature vector to the range of 0–100 and outputs two scores for each initial translation based on the mapping results.

[0061] Understandably, in this embodiment, subjective judgments of "good / bad" are transformed into interpretable and traceable numerical values, facilitating subsequent horizontal comparisons. Rules and models are decoupled, allowing business users to quickly adapt to new domains by adjusting weights or adding / removing features without retraining a large model.

[0062] Step S43: Based on the polysemy score and the adaptability score, determine the translation quality corresponding to each initial translation result; It should be noted that translation quality refers to the final quality level given after considering both ambiguity and applicability, which can be expressed as a score, grade, or ranking number.

[0063] In this embodiment, multiple initial translations of the same target language are sorted lexicographically by "descending order of applicability score and ascending order of ambiguity score." The first-ranked translation is selected as a candidate. If the difference between the scores of the top two translations is less than a threshold, a fusion strategy is triggered: the terminology of the highly applicability sentence is mixed with the structure of the less ambiguous sentence at the fragment level, and the scoring is run again until the difference is greater than the threshold. The final quality value (ranking number or fused composite score) for each initial translation is output.

[0064] Understandably, in this embodiment, "dual-dimensional + automatic fusion" replaces manual sentence selection, ensuring that the selected translation is both less ambiguous and highly relevant to the context. The difference threshold mechanism avoids over-fusion, preserves the stylistic differences of the original text, and balances diversity and accuracy.

[0065] Step S44: Based on the translation quality, obtain the target translation result corresponding to each language target.

[0066] It should be noted that the target translation result refers to the final translation that has been officially adopted and returned to the user after being scored, ranked, or integrated.

[0067] In this embodiment, based on the sorting results of step S43, the highest quality translation is written to the return buffer; if the user requests the return of Top-k candidates, they are packaged together. The terms matched in the terminology database, background knowledge IDs, and scoring details are attached as metadata to the translation object, forming a traceable "translation certificate." The response message is assembled according to the language target and output to the caller or downstream processes (such as manual review or multilingual typesetting).

[0068] Understandably, in this embodiment, users receive a translation "with quality assurance." If disputes arise later, they can directly access the scoring details and terminology to quickly pinpoint the problem. Metadata is automatically accumulated into new training / fine-tuning samples, forming a closed loop of "translation-scoring-feedback-iteration" for continuous self-optimization.

[0069] This application proposes a multilingual translation method and related equipment based on a large language model intelligent agent, relating to the field of natural language processing technology. In related technologies, mainstream solutions still heavily rely on manual translation and calibration: multilingual professionals are required to cross-check terminology, eliminate ambiguity, unify style, and process dynamic text (such as variables and placeholders in code). This model has significant drawbacks: firstly, the process is cumbersome and time-consuming, with a single round of translation deployment often taking months; secondly, multilingual, especially minority language, translators and calibrators are scarce and costly; thirdly, manual collaboration makes it difficult to guarantee consistency of terminology across languages, and the judgment and handling of textual ambiguity are highly subjective and inefficient. In contrast, this application, firstly, responds to multilingual translation guidance... The process involves obtaining a text to be translated and multiple language translation targets, wherein the text to be translated contains standard translation texts in at least two different language versions. Then, based on the multiple language translation targets, multiple target agents are determined for translating the text to be translated. Further, based on a preset large model base and the multiple target agents, the standard documents in different language versions of the text to be translated are translated in parallel to obtain multiple initial translation results corresponding to each language target. The multiple initial translation results are aligned to obtain multiple aligned initial translation results. Finally, based on a preset single-language calibration agent, calibration and disambiguation operations are performed on all the aligned initial translation results to obtain the target translation result corresponding to each language target.

[0070] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The step of obtaining the target translation result corresponding to each language target based on the translation quality further includes steps A10 to A30: Step A10: Based on the translation quality, determine whether the initial translation result needs to be retranslated; It should be noted that translation quality refers to the quantitative indicators given by the pre-scoring process, which can be scores, grades, or "pass / fail" signs, used to measure whether a single initial translation is acceptable.

[0071] The retranslation trigger condition refers to the threshold rules preset by the business, such as "applicability score < 70 and ambiguity score > 30" or "ranking not in Top-1 and difference from Top-1 < 5 points".

[0072] In this embodiment, the monolingual calibration agent compares the translation quality of each initial translation with the triggering conditions item by item. If any condition is met, the translation is marked as "needs retranslation"; otherwise, it is marked as "can be used directly". The results of all target languages ​​are summarized to form a "retranslation list" and the corresponding quality details are cached for subsequent steps to write back.

[0073] Understandably, in this embodiment, transforming "good or bad" into a binary decision of "whether to re-implement" is logically simple, explainable, and easy to implement in engineering. The threshold can be hot-updated, allowing business users to adjust the quality threshold without restarting the service, thus achieving agile operations.

[0074] Specifically, after the step of determining whether the initial translation result needs to be retranslated based on the translation quality, the method further includes steps A11-A12: Step A11: If there are initial translation results that need to be retranslated, construct an agent optimization dataset based on the initial translation results that need to be retranslated. It should be noted that the initial translation result that needs to be retranslated refers to the translation that was determined to be "not up to standard and must be retranslated" in step A10. The corresponding original text fragments, terminology hits, background knowledge, and quality score details are all completely preserved.

[0075] The agent optimization dataset refers to a hybrid "training + evaluation" data package specifically constructed for iterative improvement, containing: the original text, old translations, quality defect labels, and expected attribute groups. The agent optimization dataset can be directly used for fine-tuning or rule-based optimization.

[0076] In this embodiment, the original text that "needs to be retranslated" is aligned at the sentence or paragraph level, and the smallest trainable unit is cut out. If the old translation is incorrect in terms of terminology, a "terminology violation" label is set, and the standard translation from the terminology database is written into the expected field; if there is ambiguity in reference, a "ambiguity" label is set, and the contextual coreference chain is written into the expected field; if the style is inconsistent, a "style drift" label is set, and the style vector specified by the background knowledge is recorded.

[0077] It should be noted that in this embodiment, for language pairs where the frequency of the same defect is greater than k, the samples are automatically copied and the sampling weight is increased to ensure that high-frequency problems are learned first.

[0078] Ultimately, we obtain the agent optimization dataset.

[0079] Understandably, in this embodiment, "failed cases" are instantly converted into "targeted agent optimization datasets" without waiting for manual collection, achieving zero-latency data closure. Defect labels are fine-grained down to the "terminology / ambiguity / style" dimension, allowing for targeted fine-tuning later, avoiding a one-size-fits-all approach of retraining the entire model and saving computing power.

[0080] Step A12: Based on the agent optimization dataset, optimize the target agent, the preset monolingual calibration agent, and the preset scoring rules.

[0081] It should be noted that optimizing the target agent refers to making lightweight fine-tuning or reconstructing of prompt words on the generated model (target agent) to reduce the number of similar errors in similar scenarios.

[0082] Optimizing the preset monolingual calibration agent refers to updating its internal feature extractor or discrimination layer weights to improve the consistency between the score and human judgment.

[0083] Optimizing the preset scoring rules means recalculating feature weights, thresholds, or replacing the discriminant sub-model to make the scoring function more closely match the latest data distribution.

[0084] Understandably, in this embodiment, failure data immediately feeds back into the model and rules, forming a closed loop of "translation-failure-labeling-optimization", which effectively reduces the online error rate and does not affect the original effects in other languages ​​or domains.

[0085] Step A20: If all initial translation target results need to be retranslated, then the text to be translated is retranslated to update multiple initial translation results and update the translation quality corresponding to each initial translation result; It should be noted that re-translation refers to generating a new translation by adjusting prompt words, sampling parameters, or switching to a backup model, while keeping the original text, terminology database, and background knowledge unchanged.

[0086] In this embodiment, if the length of the "retranslation list" equals the total number of languages, it means that none of the translation results meet the requirements, triggering a global retranslation.

[0087] Understandably, this embodiment avoids a rigid "one-time fix" process, giving the system a chance to "self-correct," significantly improving the final pass rate. By generating new results using "the same original text + different decoding strategies," the closed loop can be completed within the machine without manual intervention, reducing operating costs.

[0088] Step A30: Based on the updated translation quality, return to the step of determining whether the initial translation result needs to be retranslated based on the translation quality, until there is an initial translation result among all initial translation target results that does not need to be retranslated, and obtain the target translation result.

[0089] It should be noted that the iteration exit condition means that the loop ends when at least one translation of the target language is marked as "directly usable".

[0090] In this embodiment, the updated retranslation list is checked: if all translations still need to be retranslated, the translation is repeated and the process returns to A10 for further evaluation. If any translation meets the "can be used directly" requirement, the loop terminates immediately. The highest quality translations for each language (which may come from different iterations) are assembled into the final response and written to the "target translation result" field. The quality curve and decoding parameters for each iteration are recorded as log output for subsequent analysis and model fine-tuning.

[0091] Understandably, by using lightweight iteration, the maximum quality gain can be achieved with the minimum computing power cost. At the same time, iteration conditions are set to ensure that the system will not retry indefinitely, thus controlling latency and avoiding the waste of large model resources in extreme cases.

[0092] Specifically, the step of obtaining the target translation result until there is an initial translation result that does not need to be retranslated among all the initial translation target results further includes steps A31 to A32: Step A31: If there is only one initial translation result that does not need to be retranslated among all the initial translation target results, set the initial translation result that does not need to be retranslated as the target translation result; It should be noted that the initial translation result that does not require retranslation refers to the translation that has been determined to be "of acceptable quality and ready for direct use" in step A10.

[0093] The target translation result refers to the unique translation that is ultimately adopted by the system and returned to the user.

[0094] In this embodiment, all language versions are scanned for "Pass / Requires Retranslation" markers. When the counter shows "Pass" count = 1, the uniquely passed translation is directly copied to the target translation result field, skipping any additional comparisons. Simultaneously, the terminology hit rate and quality score details corresponding to this translation are packaged as metadata for use by downstream modules.

[0095] Understandably, when only one criterion is met, the result is generated immediately, eliminating the overhead of sorting and comparison. No further compromises are needed, resulting in a consistent and predictable user experience.

[0096] Step A32: If there are multiple initial translation results that do not need to be retranslated among all the initial translation target results, set the initial translation result with the highest adaptability score among the multiple initial translation results that do not need to be retranslated as the target translation result.

[0097] It should be noted that the adaptability score refers to the quantitative index output by step S42. The higher the value, the more the translation fits the background knowledge, terminology database and contextual style.

[0098] In this embodiment, all translations marked "passed" are filtered out to form a "candidate pool". Within the candidate pool, translations are sorted in descending order of adaptability score; if there is a tie for the highest score, they are then sorted in ascending order of ambiguity score as a second keyword; if still tied, the latest version from the latest iteration round is selected. The translation ranked first is written to the target translation result field, and the selection criteria (adaptability score, ambiguity score, iteration round) are recorded in the log.

[0099] Understandably, even in scenarios with multiple standards, the system can still ensure "selecting the best from the best," avoiding quality fluctuations caused by random selection. The sorting rules are fully explainable, and the system's decision-making path can be quickly reproduced during subsequent audits or manual reviews.

[0100] Based on the first and second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to that in embodiments one and two above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 Before the step of performing calibration and disambiguation operations on the multiple initial translation results after alignment based on a preset single-language calibration agent, steps B10 to B40 are further included: Step B10: Obtain sample data, which corresponds to the first data processing result; It should be noted that the sample data refers to the set of labeled instances. Each sample contains: the original translation, the expected calibrated translation, and the expected score, which is used to measure whether the output of the calibrating agent meets the standard.

[0101] It should be noted that the first data processing result refers to the "expected result" given by the sample's own human or authoritative system, that is, the correct standard, which will be used as the only correct reference in the future.

[0102] Step B20: Process the sample data using the current monolingual calibration agent to obtain a second data processing result; It should be noted that the current monolingual calibration agent refers to a scoring-calibration model that is in a state of being tested / adjusted, which includes a feature extractor, a discriminator, and scoring rule weights.

[0103] The second data processing result refers to the "calibrated translation + score" that the current agent actually outputs for the same sample.

[0104] Step B30: Determine whether the first data processing result is consistent with the second data processing result; It should be noted that "consistency" means that the second result and the first result are completely matched in both the "calibrated translation" and "expected score" dimensions; if there is a deviation in either one, it is judged as inconsistent.

[0105] Step B40: If the first data processing result is inconsistent with the second data processing result, adjust the parameters of the current monolingual calibration agent. Based on the current monolingual calibration agent with adjusted parameters, return to the step of using the current monolingual calibration agent to process the sample data and obtain the second data processing result, until the first data processing result is consistent with the second data processing result, and obtain the preset monolingual calibration agent.

[0106] Adjusting parameters refers to any reversible changes such as gradient updates, rule weight refactoring, threshold shifting, or prompt word correction; the adjustment range is automatically calculated based on the preset step size or learning rate.

[0107] In this embodiment, if there is a discrepancy, the corresponding parameter tuning strategy is automatically selected. After parameter tuning, the same sample is immediately returned to step B20 to generate a new second result. The comparison and parameter tuning are repeated until "consistency" is achieved. Once all samples pass the consistency test, the model weights and rule snapshot at this moment are fixed, named "Preset Monolingual Calibration Agent", written to the model repository and a version number is added.

[0108] It is understandable that using "completely consistent with the first data processing result" as the convergence condition ensures the accuracy of the output result of the preset single-language calibration agent obtained after iterative training.

[0109] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the multilingual translation method based on large language model intelligent agents in this application. Any simple transformations based on this technical concept are within the protection scope of this application.

[0110] This application also provides a multilingual translation device based on a large language model intelligent agent; please refer to [reference needed]. Figure 4 The multilingual translation device based on a large language model agent includes: The acquisition module 10 is used to acquire the text to be translated and multiple language translation targets in response to a multilingual translation instruction; The determining module 20 is used to determine multiple target agents for translating the text to be translated based on the multiple language translation targets; Translation module 30 is used to translate standard files in different language versions of the text to be translated in parallel based on a preset large model base and the multiple target agents, to obtain multiple initial translation results corresponding to each language target, and to align the multiple initial translation results to obtain multiple aligned initial translation results; The calibration and disambiguation module 40 is used to perform calibration and disambiguation operations on the multiple initial translation results after alignment based on a preset single-language calibration agent, so as to obtain the target translation result corresponding to each language target.

[0111] In one embodiment, the calibration and disambiguation module further includes: The first acquisition unit is used to acquire the background knowledge, terminology database, and context of the text to be translated corresponding to the multilingual translation instructions. The scoring unit is used to perform translation quality polysemy scoring and translation quality suitability scoring operations on multiple initial translation results after alignment, based on the background knowledge, the terminology database, the context of the text to be translated, and preset scoring rules, using a preset monolingual calibration agent, to obtain the polysemy score and suitability score corresponding to each initial translation result. The first determining unit is used to determine the translation quality corresponding to each initial translation result based on the polysemy score and the adaptability score. The first obtaining unit is used to obtain the target translation result corresponding to each language target based on the translation quality.

[0112] In one embodiment, the calibration and disambiguation module further includes: The first judgment unit is used to determine, based on the translation quality, whether the initial translation result needs to be retranslated; The retranslation unit is used to retranslate the text to be translated if all initial translation target results need to be retranslated, so as to update multiple initial translation results and update the translation quality corresponding to each initial translation result; The second obtaining unit is used to return a step based on the updated translation quality to determine whether the initial translation result needs to be retranslated, until there is an initial translation result among all the initial translation target results that does not need to be retranslated, and then obtain the target translation result.

[0113] In one embodiment, the calibration and disambiguation module further includes: The building unit is used to construct an agent-optimized dataset based on the initial translation results that need to be retranslated if there are initial translation results that need to be retranslated. An optimization unit is used to optimize the target agent, the preset monolingual calibration agent, and the preset scoring rules based on the agent optimization dataset.

[0114] In one embodiment, the calibration and disambiguation module further includes: The first setting unit is used to set the initial translation result that does not need to be retranslated as the target translation result if there is only one initial translation result that does not need to be retranslated among all the initial translation target results; The second setting unit is used to set the initial translation result with the highest adaptability score among the multiple initial translation results that do not need to be retranslated as the target translation result if there are multiple initial translation results that do not need to be retranslated.

[0115] In one embodiment, the multilingual translation device based on a large language model agent further includes an agent training module, which in turn includes: The second acquisition unit is used to acquire sample data, which corresponds to the first data processing result. A data processing unit is used to process the sample data using the current monolingual calibration agent to obtain a second data processing result; The second judgment unit is used to determine whether the first data processing result is consistent with the second data processing result; An iterative training unit is used to adjust the parameters of the current monolingual calibration agent if the first data processing result is inconsistent with the second data processing result. Based on the current monolingual calibration agent with adjusted parameters, the unit returns to the step of using the current monolingual calibration agent to process the sample data and obtain the second data processing result, until the first data processing result is consistent with the second data processing result, thus obtaining a preset monolingual calibration agent.

[0116] The multilingual translation device based on a large language model agent provided in this application employs the multilingual translation method based on a large language model agent in the above embodiments, and can solve the technical problems of multilingual translation based on a large language model agent. Compared with related technologies, the beneficial effects of the multilingual translation device based on a large language model agent provided in this application are the same as the beneficial effects of the multilingual translation method based on a large language model agent provided in the above embodiments, and other technical features in the multilingual translation device based on a large language model agent are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0117] This application provides a multilingual translation device based on a large language model intelligent agent. The multilingual translation device based on a large language model intelligent agent includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the multilingual translation method based on a large language model intelligent agent in the above embodiment 1.

[0118] The following is for reference. Figure 5This document illustrates a structural schematic diagram of a multilingual translation device based on a large language model intelligent agent, suitable for implementing embodiments of this application. The multilingual translation device based on a large language model intelligent agent in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The multilingual translation device based on a large language model agent shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0119] like Figure 5 As shown, a multilingual translation device based on a large language model agent may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the multilingual translation device based on the large language model agent. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the multilingual translation device based on a large language model agent to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a multilingual translation device based on a large language model agent with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0120] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0121] The multilingual translation device based on a large language model intelligent agent provided in this application employs the multilingual translation method based on a large language model intelligent agent in the above embodiments, and can solve the technical problems of multilingual translation based on a large language model intelligent agent. Compared with related technologies, the beneficial effects of the multilingual translation device based on a large language model intelligent agent provided in this application are the same as the beneficial effects of the multilingual translation method based on a large language model intelligent agent provided in the above embodiments, and other technical features in this multilingual translation device based on a large language model intelligent agent are the same as the features disclosed in the method of the previous embodiment, and will not be repeated here.

[0122] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0124] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the multilingual translation method based on a large language model agent in the above embodiments.

[0125] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0126] The aforementioned computer-readable storage medium may be included in a multilingual translation device based on a large language model intelligent agent; or it may exist independently and not be assembled into a multilingual translation device based on a large language model intelligent agent.

[0127] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a multilingual translation device based on a large language model intelligent agent, cause the multilingual translation device based on a large language model intelligent agent to: In response to a multilingual translation instruction, the system acquires the text to be translated and multiple language translation targets, wherein the text to be translated contains at least two different language versions of standard translation text. Based on the multiple language translation targets, multiple target agents are determined for translating the text to be translated; Based on the preset large model base and the multiple target agents, the standard files of different language versions in the text to be translated are translated in parallel to obtain multiple initial translation results corresponding to each language target. The multiple initial translation results are aligned to obtain multiple aligned initial translation results. Based on a preset single-language calibration agent, calibration and disambiguation operations are performed on the multiple initial translation results after alignment to obtain the target translation result corresponding to each language target.

[0128] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0129] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0130] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0131] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described multilingual translation method based on a large language model intelligent agent, thereby solving the technical problems of multilingual translation based on a large language model intelligent agent. Compared with related technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the multilingual translation method based on a large language model intelligent agent provided in the above embodiments, and will not be repeated here.

[0132] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the multilingual translation method based on a large language model agent as described above.

[0133] The computer program product provided in this application can solve the technical problem of multilingual translation based on large language model intelligent agents. Compared with related technologies, the beneficial effects of the computer program product provided in this application are the same as those of the multilingual translation method based on large language model intelligent agents provided in the above embodiments, and will not be repeated here.

[0134] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A multilingual translation method based on a large language model intelligent agent, characterized in that, The multilingual translation method based on a large language model agent includes: In response to a multilingual translation instruction, the system acquires the text to be translated and multiple language translation targets, wherein the text to be translated contains at least two different language versions of standard translation text. Based on the multiple language translation targets, multiple target agents are determined for translating the text to be translated; Based on the preset large model base and the multiple target agents, the standard files of different language versions in the text to be translated are translated in parallel to obtain multiple initial translation results corresponding to each language target. The multiple initial translation results are aligned to obtain multiple aligned initial translation results. Based on a preset single-language calibration agent, calibration and disambiguation operations are performed on the multiple initial translation results after alignment to obtain the target translation result corresponding to each language target.

2. The multilingual translation method based on a large language model agent as described in claim 1, characterized in that, The step of performing calibration and disambiguation operations on multiple aligned initial translation results based on a preset single-language calibration agent to obtain the target translation result corresponding to each language target further includes: Obtain the background knowledge, terminology database, and context of the text to be translated corresponding to the multilingual translation instructions; Based on the background knowledge, the terminology database, the context of the text to be translated, and the preset scoring rules, a preset monolingual calibration agent is used to perform translation quality polysemy scoring and translation quality suitability scoring operations on multiple initial translation results after alignment, so as to obtain the polysemy score and suitability score corresponding to each initial translation result. Based on the polysemy score and the adaptability score, the translation quality corresponding to each initial translation result is determined; Based on the translation quality, the target translation result corresponding to each language target is obtained.

3. The multilingual translation method based on a large language model agent as described in claim 2, characterized in that, The step of obtaining the target translation result corresponding to each language target based on the translation quality further includes: Based on the translation quality, determine whether the initial translation result needs to be retranslated; If all initial translation target results need to be retranslated, then the text to be translated is retranslated to update multiple initial translation results, and the translation quality corresponding to each initial translation result is updated; Based on the updated translation quality, the process returns to the step of determining whether the initial translation result needs to be retranslated based on the translation quality, until there is an initial translation result among all the initial translation target results that does not need to be retranslated, and the target translation result is obtained.

4. The multilingual translation method based on a large language model agent as described in claim 3, characterized in that, After the step of determining whether the initial translation result needs to be retranslated based on the translation quality, the method further includes: If there are initial translation results that need to be retranslated, construct an agent-optimized dataset based on the initial translation results that need to be retranslated; Based on the agent optimization dataset, optimize the target agent, the preset monolingual calibration agent, and the preset scoring rules.

5. The multilingual translation method based on a large language model agent as described in claim 3, characterized in that, The step of obtaining the target translation result until there is an initial translation result among all the initial translation target results that does not need to be retranslated also includes: If there is only one initial translation result among all the initial translation target results that does not need to be retranslated, then the initial translation result that does not need to be retranslated is set as the target translation result; If there are multiple initial translation results that do not require retranslation among all initial translation target results, the initial translation result with the highest adaptability score among these multiple initial translation results that do not require retranslation is set as the target translation result.

6. The multilingual translation method based on a large language model agent as described in claim 1, characterized in that, Before the step of performing calibration and disambiguation operations on the multiple initial translation results after alignment based on a preset single-language calibration agent, the method further includes: Acquire sample data, which corresponds to the first data processing result; The sample data is processed using the current monolingual calibration agent to obtain a second data processing result; Determine whether the first data processing result is consistent with the second data processing result; If the first data processing result is inconsistent with the second data processing result, the parameters of the current monolingual calibration agent are adjusted. Based on the current monolingual calibration agent with adjusted parameters, the process of using the current monolingual calibration agent to process the sample data and obtain the second data processing result is returned until the first data processing result is consistent with the second data processing result, thus obtaining the preset monolingual calibration agent.

7. A multilingual translation device based on a large language model intelligent agent, characterized in that, The multilingual translation device based on a large language model agent includes: The acquisition module is used to acquire the text to be translated and multiple language translation targets in response to a multilingual translation instruction; The determining module is used to determine multiple target agents for translating the text to be translated based on the multiple language translation targets; The translation module is used to translate standard files in different language versions of the text to be translated in parallel based on a preset large model base and the multiple target agents, to obtain multiple initial translation results corresponding to each language target, and to align the multiple initial translation results to obtain multiple aligned initial translation results; The calibration and disambiguation module is used to perform calibration and disambiguation operations on multiple initial translation results after alignment based on a preset single-language calibration agent, so as to obtain the target translation result corresponding to each language target.

8. A multilingual translation device based on a large language model intelligent agent, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the multilingual translation method based on a large language model agent as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the multilingual translation method based on a large language model agent as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the multilingual translation method based on a large language model agent as described in any one of claims 1 to 6.

Images