A traditional Chinese medicine literature translation system and method based on multi-agent cooperation and RAG technology

Abstract

The application relates to a traditional Chinese medicine literature translation system and method based on multi-agent collaboration and RAG technology, aiming to solve the technical problems of mistranslation caused by polysemy of terms, distortion of classical citation translation, and inconsistency of translation context logic in traditional Chinese medicine literature translation. The system is divided into labor cooperation by multi-agent, including general semantic agent, term specialization agent, ancient Chinese translation agent and classics interpretation agent, which respectively process the tasks of general semantics, professional terms, ancient Chinese conversion and classics citation translation; and combines the RAG module to dynamically search the traditional Chinese medicine term library and classical literature knowledge, to assist the intelligent agent to accurately understand the term context and the classics background. In the translation process, the system ensures the accuracy of term translation, the reliability of classical citation restoration and the logical coherence of translation through conflict detection and context consistency optimization. In addition, the system supports continuous learning and model optimization based on user feedback, dynamically improving the translation quality. Through multi-agent collaboration and knowledge enhancement technology, the application significantly improves the professionalism, accuracy and consistency of traditional Chinese medicine literature translation, effectively promotes the international communication and exchange of traditional Chinese medicine culture.

Description

Technical Field

[0001] This invention relates to the field of multi-agent RAG (Retrieval-Augmented Generation) TCM literature translation technology, specifically a TCM literature translation system and method based on multi-agent collaboration and RAG technology. Background Technology

[0002] Currently, the translation of TCM terminology mainly relies on human translation and simple machine translation. While human translation can guarantee high translation quality, it is inefficient and cannot meet the translation needs of a large number of documents. On the other hand, existing machine translation technology often results in inaccurate and ambiguous translations when dealing with TCM terminology due to the special and complex nature of the terminology, and thus cannot meet professional requirements.

[0003] According to the utility model patent with Chinese patent application number 202510065937.5, a method and system for translating traditional Chinese medicine (TCM) terminology based on natural language processing (NLP) technology is proposed. This method and system can solve at least the following problems or defects in the existing technology: First, it cannot update and collect new terms in TCM literature in real time, resulting in a lag in translation content; second, the semantic understanding of TCM terminology is not deep enough, making it impossible to accurately judge semantic changes, thus affecting the accuracy of translation; third, it lacks an effective ambiguity analysis and processing mechanism, and cannot provide accurate translation results when encountering polysemous terms; fourth, the existing translation system has a low degree of modularity and cannot flexibly cope with the translation needs of different types of TCM terminology. The TCM terminology translation method and system based on NLP technology uses a single NLP model for translation, and the terminology database is a static pre-built structure, which cannot respond to semantic ambiguities in the translation process in real time, and lacks a cross-segment collaboration mechanism, resulting in poor consistency of translations. Therefore, it is necessary to propose a TCM literature translation system and method based on multi-agent collaboration and RAG technology to solve the above-mentioned problems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a TCM literature translation system and method based on multi-agent collaboration and RAG technology, which has the advantages of being able to respond to semantic ambiguities in the translation process in real time and having a cross-segment collaboration mechanism.

[0005] To achieve the above objectives, this invention provides a TCM literature translation system based on multi-agent collaboration and RAG technology, comprising:

[0006] The text input and preprocessing module is used to preprocess the classical Chinese source text containing TCM professional terms in the TCM literature to be translated. It encapsulates each preprocessed text block, the source document meta-information, and the list of TCM professional terms extracted from the text block into an array object.

[0007] The translation task decomposition and allocation module is used to treat an array object as a whole translation task and decompose the whole translation task into multiple sub-translation tasks according to a preset multi-stage translation process.

[0008] A multi-agent translation group consists of multiple translation agents with specific expertise, which are used to receive and process corresponding sub-translation tasks and generate corresponding initial translation fragments.

[0009] The RAG module connects to at least one TCM terminology and classic text knowledge base that stores TCM knowledge. It is used to receive keyword or phrase queries containing contextual information from the translation agent, retrieve relevant terminology definitions, classic text citations, concept explanations, usage examples or graph association information from the TCM terminology and classic text knowledge base based on vector similarity calculation, and return the retrieval results to the translation agent in the form of semantically enhanced fragments.

[0010] The results coordination and integration module receives the initial translation fragments generated by the translation agent, and performs collaborative optimization, correction and splicing integration on each initial translation fragment based on the correlation and contextual consistency between sub-translation tasks to generate the final target language translation text.

[0011] Among them, the multi-agent translation group and RAG module dynamically update the model and knowledge based on user feedback and / or expert proofreading of the corrected translation through the following methods:

[0012] For translation agents in a multi-agent translation group, the corrected translation and the corresponding initial translation are constructed into high-quality supervised fine-tuning data pairs. A parameter-efficient fine-tuning method is used to incrementally train the basic large model of a specific agent. If the term translation is incorrect, the model parameters of the term-specialized agent are fine-tuned using the corrected term example sentences to enhance its mapping ability for the term. If the error is due to a misunderstanding of classical Chinese, the classical Chinese-modern Chinese translation agent is trained in a targeted manner using correct classical Chinese-modern Chinese comparison samples.

[0013] For the RAG module, based on the identified retrieval failure cases, the retrieval model is fine-tuned using query-classical passage pairs containing positive and negative samples through comparative learning to improve its ability to capture key contextual information. For the ranking model, the relevance scores annotated by experts are used to incrementally train the re-ranking model of retrieval results to optimize its ability to prioritize important classical passages.

[0014] Furthermore, the translation agent includes at least one of the following: a general semantic agent, a terminology-specialized agent, a classical Chinese translation agent, and a classic text interpretation agent.

[0015] A general semantic agent is used to handle basic translation tasks involving conventional semantics;

[0016] A terminology-specialized intelligent agent is used to process the translation and parsing of TCM professional terms.

[0017] The Classical Chinese to Chinese translation intelligent agent is used to convert Classical Chinese grammar into more easily understood modern Chinese;

[0018] The Classical Text Interpretation Intelligent Agent is used to process classical text citations and associate them with contextual information.

[0019] Furthermore, the TCM terminology and classical text knowledge base connected to the RAG module is a structured TCM terminology database, including:

[0020] Standardized translations of terminology, synonym comparisons, concept definitions, and semantic network connections;

[0021] A database of original texts and multiple translations of classic Chinese medicine texts;

[0022] Database of contemporary TCM works, textbooks, and clinical guidelines;

[0023] Traditional Chinese Medicine knowledge graph, linking entities, concepts, attributes, and relationships;

[0024] A rule base for TCM translation and methods for handling common ambiguities.

[0025] Furthermore, the RAG module is used to respond to a specific query request from any translation agent, retrieving auxiliary information semantically related to the query request from a knowledge base of TCM terminology and classical texts. This auxiliary information includes:

[0026] Term definitions or standard translations;

[0027] Interpretation or usage examples of classic original texts;

[0028] Knowledge graph association information;

[0029] Ambiguity handling suggestions in the translation rule base;

[0030] The auxiliary information is fed back to the translation agent that initiated the query, so that it can be integrated and utilized when generating or optimizing the initial translation fragment;

[0031] During the dynamic update process of the multi-agent translation group and the RAG module, translation errors or points for optimization are identified based on the differences between the output of the translation correction and result collaboration and integration module.

[0032] Furthermore, the implementation process of the results collaboration and integration module includes:

[0033] Receive initial translation fragments generated by different translation agents, including classical Chinese translation results, preliminary bilingual translation results, and translation results of technical terms;

[0034] The translation conflict detection unit is used to detect inconsistencies or conflicts in terminology, concept interpretation, logical coherence, or stylistic features among different translated text fragments.

[0035] The contextual consistency optimization unit is used to adjust the coherence and semantic consistency of the corresponding translation based on the overall context of the source language text.

[0036] The text fluency optimization unit is used to splice and combine the optimized and adjusted translation fragments to output the final translated text.

[0037] This invention also provides a method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology, which includes:

[0038] Step S1, Text Input and Preprocessing: Receive the classical Chinese source text containing TCM terminology from the TCM literature to be translated, and preprocess the source text.

[0039] Step S2, Translation Task Decomposition and Allocation: The preprocessed text block is taken as the whole translation task and decomposed into multiple sub-translation tasks, which are then allocated to the corresponding translation agents in the multi-agent translation group;

[0040] Step S3, Translation Agent and RAG Assistance: Each translation agent executes its assigned sub-translation task. When needed, any translation agent initiates a specific query request containing contextual information to the RAG module. The RAG module retrieves relevant auxiliary information from the TCM terminology and classic knowledge base and feeds it back to the translation agent. The translation agent integrates the auxiliary information to generate or optimize the initial translation fragment.

[0041] Step S4, Results Collaboration and Integration: Collect the initial translation fragments generated by each translation agent, and based on the correlation and contextual consistency between sub-translation tasks, collaboratively optimize, correct, and splice and integrate each initial translation fragment to generate the final target language translation text;

[0042] Step S5, Dynamic Update of Model and Knowledge: Based on user feedback and / or the revised translation after expert proofreading, the model and knowledge are dynamically updated through the following methods:

[0043] For translation agents in a multi-agent translation group, the corrected translation and the corresponding initial translation are constructed into high-quality supervised fine-tuning data pairs. A parameter-efficient fine-tuning method (such as LoRA) is used to incrementally train the basic large model of a specific agent. If the term translation is incorrect, the model parameters of the term-specialized agent are fine-tuned using the corrected term example sentences to enhance its mapping ability for the term. If the error is due to a misunderstanding of classical Chinese, the classical Chinese-modern Chinese translation agent is trained in a targeted manner using correct classical Chinese-modern Chinese comparison samples.

[0044] For the RAG module, based on the identified retrieval failure cases, the retrieval model is fine-tuned using query-classical passage pairs containing positive and negative samples through comparative learning to improve its ability to capture key contextual information. For the ranking model, the relevance scores annotated by experts are used to incrementally train the re-ranking model of retrieval results to optimize its ability to prioritize important classical passages.

[0045] Furthermore, the translation agent includes at least one of the following: a general semantic agent, a terminology-specialized agent, a classical Chinese translation agent, and a classic text interpretation agent.

[0046] A general semantic agent is used to handle basic translation tasks involving conventional semantics;

[0047] A terminology-specialized intelligent agent is used to process the translation and parsing of TCM professional terms.

[0048] The Classical Chinese to Chinese translation intelligent agent is used to convert Classical Chinese grammar into more easily understood modern Chinese;

[0049] The Classical Text Interpretation Intelligent Agent is used to process classical text citations and associate them with contextual information.

[0050] Furthermore, the TCM terminology and classical text knowledge base connected to the RAG module is a structured TCM terminology database, including:

[0051] Standardized translations of terminology, synonym comparisons, concept definitions, and semantic network connections;

[0052] A database of original texts and multiple translations of classic Chinese medicine texts;

[0053] Database of contemporary TCM works, textbooks, and clinical guidelines;

[0054] Traditional Chinese Medicine knowledge graph, linking entities, concepts, attributes, and relationships;

[0055] A rule base for TCM translation and methods for handling common ambiguities.

[0056] Furthermore, the RAG module is used to respond to a specific query request from any translation agent, retrieving auxiliary information semantically related to the query request from a knowledge base of TCM terminology and classical texts. This auxiliary information includes:

[0057] Term definitions or standard translations;

[0058] Interpretation or usage examples of classic original texts;

[0059] Knowledge graph association information;

[0060] Ambiguity handling suggestions in the translation rule base;

[0061] The auxiliary information is fed back to the translation agent that initiated the query, so that it can be integrated and utilized when generating or optimizing the initial translation fragment;

[0062] During the dynamic update process of the multi-agent translation group and the RAG module, translation errors or points for optimization are identified based on the differences between the output of the translation correction and result collaboration and integration module.

[0063] Furthermore, step S4 specifically includes:

[0064] Receive initial translation fragments generated by different translation agents, including classical Chinese translation results, preliminary bilingual translation results, and translation results of technical terms;

[0065] The translation conflict detection unit is used to detect inconsistencies or conflicts in terminology, concept interpretation, logical coherence, or stylistic features among different translated text fragments.

[0066] The contextual consistency optimization unit is used to adjust the coherence and semantic consistency of the corresponding translation based on the overall context of the source language text.

[0067] The text fluency optimization unit is used to splice and combine the optimized and adjusted translation fragments to output the final translated text.

[0068] Compared with existing technologies, this invention provides a TCM literature translation system and method based on multi-agent collaboration and RAG technology, which has the following beneficial effects:

[0069] 1. This TCM literature translation system and method, based on multi-agent collaboration and RAG technology, addresses the polysemy of TCM terminology by having a specialized terminology agent call the RAG module to retrieve standardized translations and synonyms from the knowledge base in real time, and dynamically select the translation method based on the context. Compared with traditional single NLP models, it reduces the mistranslation rate of terminology (based on testing with 5,000 terms in "Basic Terminology of TCM"). When the general semantic agent encounters ambiguous words, it triggers RAG to retrieve related information from the knowledge graph, ensuring accurate translation.

[0070] 2. This TCM literature translation system and method based on multi-agent collaboration and RAG technology, for classic citations, the intelligent agent for classic interpretation requests the RAG module: retrieves the original text of "Shang Han Lun" and multiple authoritative translations (Wiseman / Nigel version); returns semantically enhanced fragments of compatibility principles and indications; breaks through the limitations of the comparative static terminology database, improves the restoration accuracy of classic citations, and ensures academic rigor.

[0071] 3. This TCM literature translation system and method based on multi-agent collaboration and RAG technology identifies contradictions between adjacent translations through a translation conflict detection unit (e.g., "Shanghan" is translated as "Cold Damage" in the previous text, but "Febrile Disease" is mistakenly translated as "Febrile Disease" in the later text); the context consistency optimization unit unifies terminology based on the full-text semantic network (e.g., the main line of "Six Channels Differentiation") and reconstructs logical chains (e.g., the transmission relationship of "Taiyang Disease → Yangming Disease"); it solves the defect of lack of cross-segment collaboration in the comparison ratio and ensures context consistency. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the structure of the TCM literature translation system of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1 The TCM literature translation system based on multi-agent collaboration and RAG technology provided in this embodiment of the invention includes a text input and preprocessing module 101, a translation task decomposition and allocation module 102, a multi-agent translation group 103, a RAG module 104, a result collaboration and integration module 106, and a model management and update module 107. The text input and preprocessing module 101 is used to preprocess the classical Chinese source text of the TCM literature to be translated, which contains TCM professional terms. Each preprocessed text block, the source document meta-information, and the list of TCM professional terms extracted from the text block are encapsulated together into a JSON format array object.

[0076] The preprocessing includes the following operations, but the numbers below do not indicate the order of operations:

[0077] 1. Word segmentation: Using natural paragraphs as the basic unit, the input lengthy source language text is divided into multiple independent text blocks. The filtered text blocks are compared with a pre-set TCM professional terminology database to extract all TCM professional terms contained in the text blocks and output them in sequence form. This ensures the suitability of the subsequent translation task processing unit.

[0078] 2. Clean the text, removing non-informational content such as page numbers, headers, footers, and advertising text, and standardize punctuation.

[0079] 3. Perform structural analysis on the sequence, such as on the subject and sentence structure.

[0080] The translation task decomposition and allocation module 102 is used to treat the array object of the text input and preprocessing module 101 as a whole translation task, and decompose the whole translation task into multiple sub-translation tasks according to the preset multi-stage translation process.

[0081] The multi-agent translation group 103 includes several translation agents with specific specializations, such as: a general semantic agent for basic semantic translation and handling routine descriptions; a terminology-specialized agent for translating traditional Chinese medicine terminology, resolving terms like "Taiyang disease" and "stroke"; a classical Chinese translation agent for processing classical Chinese grammar into modern Chinese, translating specialized classical texts into more easily understood Mandarin; and a classic text interpretation agent for processing classic text citations and relating them to the context of the *Shanghan Lun* (Treatise on Cold Damage). Of course, other specialized translation agents can be set up according to actual needs and requirements, which will not be described in detail here.

[0082] After receiving a sub-translation task, the translation agent generates the corresponding initial translation fragment.

[0083] RAG module 104 is connected to at least one TCM terminology and classic text knowledge base 105 storing TCM knowledge. It is understood that: TCM terminology may have multiple meanings or uncertain standard translations; it is necessary to obtain the original texts and authoritative explanations of specific concepts or terms in TCM classics; it is necessary to understand the contextual background knowledge of specific sentences in ancient texts; it is necessary to obtain background knowledge related to a certain disease or prescription to assist in accurate translation; and it is necessary to understand the true meaning of classical Chinese sentences or rhetoric with the help of background knowledge. Therefore, in this embodiment, the TCM terminology and classic text knowledge base 105 is set as a structured TCM terminology database, including standardized translations of terms, synonym comparisons, concept definitions, semantic network associations; a database of original texts of TCM classics and their multiple versions of translations; a database of contemporary TCM works, textbooks, and clinical guidelines; a TCM knowledge graph, linking entities, concepts, attributes, and relationships; and a TCM translation rule base and common ambiguity handling methods.

[0084] RAG module 104 is used to receive keyword or phrase queries containing contextual information from the translation agent, and retrieve relevant term definitions, classical text citations, concept explanations, usage examples or graph association information from the TCM terminology and classic knowledge base 105 based on vector similarity calculation, and return the retrieval results to the translation agent that made the query request in the form of semantically enhanced fragments.

[0085] The RAG module 104 is used to respond to specific query requests from any translation agent. It retrieves auxiliary information semantically related to the query request from the TCM terminology and classic text knowledge base 105. The auxiliary information includes: directly applying the retrieved term definitions or standard translations to the translation; understanding and translating difficult sentences by combining the interpretations or usage examples of the original classic texts; using knowledge graph association information to assist in understanding the logical relationships and implicit information in the text; and applying the ambiguity handling suggestions retrieved according to the translation rule base. The RAG module 104 feeds back the retrieved auxiliary information to the translation agent that initiated the query, so that it can integrate and utilize the auxiliary information when generating or optimizing the initial translation fragment, and output the optimized initial translation fragment.

[0086] The result coordination and integration module 106 receives the initial translation fragments generated by the translation agent. Based on the correlation and contextual consistency between sub-translation tasks, it performs collaborative optimization, correction, and splicing integration on each initial translation fragment to generate the final target language translation text. The implementation process of the result coordination and integration module 106 includes:

[0087] First, the system receives the results of classical Chinese translations into modern Chinese generated by the classical Chinese translation agent, preliminary bilingual translations generated by the general semantic agent, and translations of specialized terms generated by the terminology-specialized agent. Using the accurate terminology translations provided by the specialized agent as the authoritative benchmark, the system leverages the semantic understanding capabilities of a large-scale language model. Then, the translation conflict detection unit ensures the accuracy of terminology usage by detecting inconsistencies or conflicts in terminology translation, conceptual explanation, logical coherence, or stylistic features between adjacent or related translation fragments output by different translation agents. Next, the contextual consistency optimization unit adjusts and optimizes the coherence and semantic consistency of the corresponding translations based on the overall context, discourse structure, and identified conflicts of the source language text. Finally, the text fluency optimization unit combines the optimized translation fragments according to the original text's structural order, performs final optimization on the overall translation for language fluency, target language habits, and readability, and outputs a final translated text that is accurate in specialized terminology, coherent in contextual logic, and conforms to the target language's expression habits.

[0088] The model management and update module 107 is used to receive user feedback and / or corrected translations after expert review; based on the differences between the corrected translations and the output of the result collaboration and integration module 106, it uses the language capabilities of the large model to identify translation errors or points that can be optimized; using the identified information, it performs incremental learning or fine-tuning updates on the relevant translation agents in the multi-agent translation group 103 and / or the retrieval or ranking model of the RAG module 104.

[0089] For the relevant translation agents in the multi-agent translation group 103, this module constructs high-quality supervised fine-tuning data pairs between the corrected translation and the corresponding initial translation. Then, it employs parameter-efficient fine-tuning methods (such as LoRA) to incrementally train the base model of the specific agent. This incremental training includes: if there is a terminology translation error, the model parameters of the terminology-specialized agent are fine-tuned using corrected terminology examples to enhance its mapping ability for that term; if there is a misunderstanding of classical Chinese, the classical Chinese-to-modern Chinese translation agent is trained specifically using correct classical Chinese-modern Chinese comparison samples.

[0090] (ii) For RAG module 104, this module performs one or both of the following two updates based on the failed retrieval cases in the identified information: for the retrieval model, it uses query-classical passage pairs containing positive and negative samples to fine-tune it through comparative learning to improve its ability to capture key contextual information; for the ranking model, it uses the relevance scores annotated by experts to incrementally train the re-ranking model of the retrieval results to optimize its ability to prioritize important classical passages.

[0091] The method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology provided in this invention includes:

[0092] Step S1, Text Input and Preprocessing: Receive the classical Chinese source text containing TCM terminology from the TCM literature to be translated, and preprocess the source text. Please refer to the method mentioned above for the preprocessing method.

[0093] Step S2, Translation Task Decomposition and Allocation: Based on the content of the preprocessed source language text, the overall translation task is decomposed into multiple sub-translation tasks, and the sub-translation tasks are allocated to translation agents with corresponding expertise in the multi-agent translation group.

[0094] Step S3, Agent Translation and RAG Assistance: Each translation agent executes its assigned sub-translation task. During execution:

[0095] When needed, any translation agent can send a specific query request containing contextual information to the RAG module;

[0096] Based on the query request, the RAG module retrieves semantically relevant auxiliary information from a knowledge base of TCM terminology and classic texts that specifically stores knowledge in the field of TCM.

[0097] The RAG module feeds back the retrieved auxiliary information to the translation agent that initiated the query;

[0098] When generating or optimizing its corresponding initial translation fragments, the translation agent incorporates auxiliary information from feedback.

[0099] Step S4: Results Collaboration and Integration: Collect initial translation fragments generated by each translation agent in the multi-agent translation group. Based on the correlation and contextual consistency between sub-translation tasks, collaboratively optimize, correct, and integrate each initial translation fragment to generate the final target language translation text.

[0100] Below is a specific example:

[0101] The implementation process of this invention will be explained in detail through a typical case of translating traditional Chinese medicine literature, taking the classic passage of "Treatise on Febrile Diseases" as an example: The source language text is: "In Taiyang disease, headache, fever, sweating, aversion to wind, and slow pulse are called stroke."

[0102] Step S1, Text Input and Preprocessing: Text input and preprocessing module 101 receives the original text:

[0103] Text cleaning: Remove irrelevant characters and standardize punctuation;

[0104] Word segmentation processing, output sequence: ["Taiyangbing", ",", "headache", "fever", ",", "sweating", "aversion to wind", ",", "slow pulse", "those", ",", "named", "stroke" and "."];

[0105] Structural analysis: Identify key components, such as "Taiyang disease" as the subject and "those with slow pulse" as a judgment sentence.

[0106] Step S2, Translation Task Decomposition and Allocation:

[0107] Examples of task decomposition within the translation task decomposition and allocation module 102, which includes agents with different functionalities, include:

[0108] A general semantic agent is used for basic semantic translation, handling routine descriptions such as "headache and fever".

[0109] A terminology-specialized intelligent agent is used for translating TCM terminology, resolving terms such as "Taiyang disease" and "stroke".

[0110] The Classical Text Interpretation Intelligent Agent is used for classical text citation processing and relating it to the context of the *Shanghan Lun* (Treatise on Cold Damage).

[0111] The ancient Chinese text translation agent is used for grammar vernacularization processing of ancient Chinese texts, to translate professional ancient Chinese literature texts into a more understandable Mandarin form.

[0112] Step S3, agent translation and RAG module 104 assistance:

[0113] For example, the query request sent by the terminology specialization agent to the RAG module 104 is the standard translation method and pathogenesis explanation of "Taiyang Disease". After the RAG module 104 is triggered, it returns: the WHO standard translated name "Taiyang Disease", and retrieves from the traditional Chinese medicine terminology and classic knowledge base 105. The retrieved auxiliary information includes: first, the original text interpretation of *Treatise on Cold Damage*: "Taiyang disease is characterized by floating pulse, stiff pain in the head and nape, and aversion to cold"; second, the knowledge graph: associating the pathogenesis of "Taiyang Disease" with the exterior wind-cold syndrome. The terminology specialization agent fuses the queried auxiliary information to obtain a semantic enhanced fragment: "Taiyang Disease (exterior wind-cold syndrome)", which is output as the optimized initial translation fragment.

[0114] For another example: the classic interpretation agent sends a query instruction to the RAG module 104 to process "Wind Stroke": after the RAG module 104 is triggered, it returns the translated name of "Zhongfeng" in the traditional Chinese medicine context as "Wind Stroke" (not the Western medical "Stroke"), accompanied by the differential interpretation from *Essentials of the Golden Chamber*.

[0115] Step S4, result collaboration and integration:

[0116] The translation conflict detection unit detects a contradiction: the general agent initially translates "aversion to wind" as "aversion to wind", but the terminology agent returns a pathogenesis prompt that "exterior deficiency" needs to be emphasized;

[0117] The context consistency optimization unit performs optimization: based on the full-text semantics (Taiyang disease exterior syndrome), adjusts "aversion to wind" to "aversion to wind due to exterior deficiency";

[0118] Reconstruct the logical chain: integrate the clauses into a coherent expression;

[0119] Final output: "Taiyang Disease manifests with headache, fever, sweating, aversion to wind due to exterior deficiency, and a slow pulse; this is identified as the Wind Stroke pattern";

[0120] Fluency optimization: Adjusted word order: "When Taiyang Disease presents with headache, fever, sweating, aversion to wind, and a slow pulse, it is diagnosed as WindStroke pattern."

[0121] In summary, this TCM literature translation system and method based on multi-agent collaboration and RAG technology addresses the polysemy of TCM terminology by having a specialized terminology agent invoke the RAG module to retrieve standardized translations and synonyms from the knowledge base in real time, dynamically selecting the translation method based on context. Compared with traditional single NLP models, it reduces the mistranslation rate of terminology (based on testing with 5000 terms from "Basic Terminology of Traditional Chinese Medicine"). When the general semantic agent encounters ambiguous words, it triggers RAG to retrieve related information from the knowledge graph, ensuring accurate translation.

[0122] Furthermore, for classic citations, the intelligent agent for interpreting classical texts requests the RAG module to: retrieve the original text of the *Shanghan Lun* and multiple authoritative translations (Wiseman / Nigel versions); return semantically enhanced fragments on compatibility principles and indications; overcome the limitations of the comparative static terminology database, improve the accuracy of classical citation restoration, and ensure academic rigor.

[0123] Furthermore, the translation conflict detection unit identifies contradictions between adjacent translations (e.g., "Shanghan" was translated as "Cold Damage" in the previous text, but mistakenly as "Febrile Disease" in the later text); the contextual consistency optimization unit unifies terminology based on the full-text semantic network (e.g., the main line of "Six Classics Differentiation") and reconstructs logical chains (e.g., the transmission relationship of "Taiyang disease → Yangming disease"); and solves the deficiency of cross-segment collaboration in the comparison ratio, ensuring contextual consistency.

[0124] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A TCM literature translation system based on multi-agent collaboration and RAG technology, characterized in that, include: The text input and preprocessing module is used to preprocess the classical Chinese source text containing TCM professional terms in the TCM literature to be translated. It encapsulates each preprocessed text block, the source document meta-information, and the list of TCM professional terms extracted from the text block into an array object. The translation task decomposition and allocation module is used to treat an array object as a whole translation task and decompose the whole translation task into multiple sub-translation tasks according to a preset multi-stage translation process. A multi-agent translation group consists of multiple translation agents with specific expertise, which are used to receive and process corresponding sub-translation tasks and generate corresponding initial translation fragments. The RAG module connects to at least one TCM terminology and classic text knowledge base that stores TCM knowledge. It is used to receive keyword or phrase queries containing contextual information from the translation agent, retrieve relevant terminology definitions, classic text citations, concept explanations, usage examples or graph association information from the TCM terminology and classic text knowledge base based on vector similarity calculation, and return the retrieval results to the translation agent in the form of semantically enhanced fragments. The results coordination and integration module receives the initial translation fragments generated by the translation agent, and performs collaborative optimization, correction and splicing integration on each initial translation fragment based on the correlation and contextual consistency between sub-translation tasks to generate the final target language translation text. Among them, the multi-agent translation group and RAG module dynamically update the model and knowledge based on user feedback and / or expert proofreading of the corrected translation through the following methods: For translation agents in a multi-agent translation group, the corrected translation and the corresponding initial translation are constructed into high-quality supervised fine-tuning data pairs. A parameter-efficient fine-tuning method is used to incrementally train the basic large model of a specific agent. If the term translation is incorrect, the model parameters of the term-specialized agent are fine-tuned using the corrected term example sentences to enhance its mapping ability for the term. If the error is due to a misunderstanding of classical Chinese, the classical Chinese-modern Chinese translation agent is trained in a targeted manner using correct classical Chinese-modern Chinese comparison samples. For the RAG module, based on the identified retrieval failure cases, the retrieval model is fine-tuned using query-classical passage pairs containing positive and negative samples through comparative learning to improve its ability to capture key contextual information. For the ranking model, the relevance scores annotated by experts are used to incrementally train the re-ranking model of retrieval results to optimize its ability to prioritize important classical passages.

2. The TCM literature translation system based on multi-agent collaboration and RAG technology according to claim 1, characterized in that, Translation agents include at least one of the following: general semantic agents, terminology-specialized agents, classical Chinese translation agents, and classical text interpretation agents. A general semantic agent is used to handle basic translation tasks involving conventional semantics; A terminology-specialized intelligent agent is used to process the translation and parsing of TCM professional terms. The Classical Chinese to Chinese translation intelligent agent is used to convert Classical Chinese grammar into more easily understood modern Chinese; The Classical Text Interpretation Intelligent Agent is used to process classical text citations and associate them with contextual information.

3. The TCM literature translation system based on multi-agent collaboration and RAG technology according to claim 1, characterized in that, The RAG module connects to a structured TCM terminology and classic text knowledge base, which includes: Standardized translations of terminology, synonym comparisons, concept definitions, and semantic network connections; A database of original texts and multiple translations of classic Chinese medicine texts; Database of contemporary TCM works, textbooks, and clinical guidelines; Traditional Chinese Medicine knowledge graph, linking entities, concepts, attributes, and relationships; A rule base for TCM translation and methods for handling common ambiguities.

4. The TCM literature translation system based on multi-agent collaboration and RAG technology according to any one of claims 1-3, characterized in that, The RAG module is used to respond to a specific query request from any translation agent, retrieving auxiliary information semantically related to the query request from a knowledge base of TCM terminology and classical texts. This auxiliary information includes: Term definitions or standard translations; Interpretation or usage examples of classic original texts; Knowledge graph association information; Ambiguity handling suggestions in the translation rule base; The auxiliary information is fed back to the translation agent that initiated the query, so that it can be integrated and utilized when generating or optimizing the initial translation fragment; During the dynamic update process of the multi-agent translation group and the RAG module, translation errors or points for optimization are identified based on the differences between the output of the translation correction and result collaboration and integration module.

5. The TCM literature translation system based on multi-agent collaboration and RAG technology according to claim 4, characterized in that, The implementation process of the results collaboration and integration module includes: Receive initial translation fragments generated by different translation agents, including classical Chinese translation results, preliminary bilingual translation results, and translation results of technical terms; The translation conflict detection unit is used to detect inconsistencies or conflicts in terminology, concept interpretation, logical coherence, or stylistic features among different translated text fragments. The contextual consistency optimization unit is used to adjust the coherence and semantic consistency of the corresponding translation based on the overall context of the source language text. The text fluency optimization unit is used to splice and combine the optimized and adjusted translation fragments to output the final translated text.

6. A method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology, characterized in that, include: Step S1: Receive the classical Chinese source text containing TCM terminology from the TCM literature to be translated, and preprocess the source text. Step S2: The preprocessed text block is treated as a whole translation task and decomposed into multiple sub-translation tasks, which are then assigned to the corresponding translation agents in the multi-agent translation group. In step S3, each translation agent performs its assigned sub-translation task. When needed, any translation agent initiates a specific query request containing contextual information to the RAG module. The RAG module retrieves relevant auxiliary information from the TCM terminology and classic knowledge base and feeds it back to the translation agent. The translation agent integrates the auxiliary information to generate or optimize the initial translation fragment. Step S4: Collect the initial translation fragments generated by each translation agent. Based on the correlation and contextual consistency between sub-translation tasks, perform collaborative optimization, correction and splicing integration on each initial translation fragment to generate the final target language translation text. Step S5: Based on user feedback and / or the revised translation after expert proofreading, dynamically update the model and knowledge in the following ways: For translation agents in a multi-agent translation group, the corrected translation and the corresponding initial translation are constructed into high-quality supervised fine-tuning data pairs. Efficient parameter fine-tuning methods (such as LoRA) are used to incrementally train the basic large model of a specific agent. If the term translation is incorrect, the model parameters of the term-specialized agent are fine-tuned using the corrected term example sentences to enhance its mapping ability for the term. If the error is due to misunderstanding of classical Chinese, the classical Chinese-modern Chinese translation agent is trained in a targeted manner using correct classical Chinese-modern Chinese comparison samples. For the RAG module, based on the identified retrieval failure cases, the retrieval model is fine-tuned using query-classical passage pairs containing positive and negative samples through comparative learning to improve its ability to capture key contextual information. For the ranking model, the relevance scores annotated by experts are used to incrementally train the re-ranking model of retrieval results to optimize its ability to prioritize important classical passages.

7. The method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology according to claim 6, characterized in that, Translation agents include at least one of the following: general semantic agents, terminology-specialized agents, classical Chinese translation agents, and classical text interpretation agents. A general semantic agent is used to handle basic translation tasks involving conventional semantics; A terminology-specialized intelligent agent is used to process the translation and parsing of TCM professional terms. The Classical Chinese to Chinese translation intelligent agent is used to convert Classical Chinese grammar into more easily understood modern Chinese; The Classical Text Interpretation Intelligent Agent is used to process classical text citations and associate them with contextual information.

8. The method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology according to claim 6, characterized in that, The RAG module connects to a structured TCM terminology and classic text knowledge base, which includes: Standardized translations of terminology, synonym comparisons, concept definitions, and semantic network connections; A database of original texts and multiple translations of classic Chinese medicine texts; Database of contemporary TCM works, textbooks, and clinical guidelines; Traditional Chinese Medicine knowledge graph, linking entities, concepts, attributes, and relationships; A rule base for TCM translation and methods for handling common ambiguities.

9. The method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology according to any one of claims 6-8, characterized in that, The RAG module is used to respond to a specific query request from any translation agent, retrieving auxiliary information semantically related to the query request from a knowledge base of TCM terminology and classical texts. This auxiliary information includes: Term definitions or standard translations; Interpretation or usage examples of classic original texts; Knowledge graph association information; Ambiguity handling suggestions in the translation rule base; The auxiliary information is fed back to the translation agent that initiated the query, so that it can be integrated and utilized when generating or optimizing the initial translation fragment; During the dynamic update process of the multi-agent translation group and the RAG module, translation errors or points for optimization are identified based on the differences between the output of the translation correction and result collaboration and integration module.

10. The method for translating traditional Chinese medicine literature based on multi-agent collaboration and RAG technology according to claim 9, characterized in that, Step S4 specifically includes: Receive initial translation fragments generated by different translation agents, including classical Chinese translation results, preliminary bilingual translation results, and translation results of technical terms; The translation conflict detection unit is used to detect inconsistencies or conflicts in terminology, concept interpretation, logical coherence, or stylistic features among different translated text fragments. The contextual consistency optimization unit is used to adjust the coherence and semantic consistency of the corresponding translation based on the overall context of the source language text. The text fluency optimization unit is used to splice and combine the optimized and adjusted translation fragments to output the final translated text.

Images