Intelligent translation method and system for relay between manpower and AI translator

Abstract

The invention provides a manual and AI translator relay intelligent translation method and a manual and AI translator relay intelligent translation system. The method comprises the following steps: analyzing according to a to-be-translated data set to obtain a to-be-translated data semantic fusion vector and a corresponding to-be-translated priority, evaluating corresponding real-time translation matchable parameters according to translation capability evaluation data and a real-time translation load, and evaluating a to-be-translated difficulty level according to the to-be-translated data set, the method comprises the steps that firstly, a to-be-translated data semantic fusion vector is obtained, then the to-be-translated data semantic fusion vector or a to-be-translated data set is distributed, to-be-translated distributed data corresponding to a manual translator and an AI translator are obtained, then intelligent translation is conducted, initial translation data are obtained, initial translation quality evaluation parameters are evaluated, and finally the real-time translation state corresponding to the AI translator is determined through threshold value comparison. According to the method, the to-be-translated data is fused on the basis of the to-be-translated scene, the to-be-translated tasks are dynamically allocated, the initial translation quality is quantitatively evaluated, and two-stage intelligent iterative optimization is performed, so that manual and AI translators can effectively cooperate to translate in a relay manner, and the translation efficiency and the translation quality are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent translation technology, and more specifically, to an intelligent translation method and system that combines human and AI translators. Background Technology

[0002] Purely human or purely AI translation can lead to insufficient translation efficiency or accuracy. A relay translation approach, combining human and AI translators, can effectively improve both efficiency and accuracy. However, traditional relay translation techniques suffer from several drawbacks. First, they lack scenario adaptation when fusing multimodal data to be translated. Second, there is a mismatch between the translation of professional terminology corresponding to specific fields and scenarios. Third, the human-AI collaboration allocation mechanism is inflexible. Furthermore, existing technologies lack two-way collaboration for real-time correction and long-term iterative optimization, hindering the intelligence and accuracy of human-AI collaboration. Therefore, there is an urgent need for an intelligent relay translation method that can achieve multimodal data fusion based on scenario categories, match professional field and scenario terminology, and provide feedback-closed-loop iterative optimization.

[0003] Effective technical solutions are urgently needed to address the above problems. Summary of the Invention

[0004] The purpose of this application is to provide an intelligent translation method and system that combines human and AI translators, which can achieve effective collaborative translation between human and AI translators by integrating translation data based on the translation scenario, dynamically allocating translation tasks, quantitatively evaluating the initial translation quality, and performing two-level intelligent iterative optimization, thereby improving translation efficiency and quality.

[0005] Firstly, this application provides an intelligent translation method that combines human and AI translators, including the following steps: Obtain the dataset to be translated, perform multimodal feature extraction and confidence calculation based on the dataset, and obtain the semantic fusion vector of the data to be translated and the corresponding translation priority; Acquire and process translation ability assessment data and real-time translation load of human translators and AI translators to obtain real-time translation matching parameters corresponding to human translators and AI translators; The difficulty level of the data to be translated is obtained by analyzing and processing the dataset to be translated. Based on the real-time translation matching parameters, combined with the translation priority and translation difficulty level, the semantic fusion vector or translation dataset of the data to be translated is allocated to obtain the translation allocation data corresponding to human translators and AI translators; Intelligent translation is performed based on the assigned data to be translated to obtain the initial translation data corresponding to the AI ​​translator. The initial translation data is then analyzed and processed to obtain the initial translation quality evaluation parameters corresponding to the AI ​​translator. The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold, and the real-time translation status of the AI ​​translator is determined based on the threshold comparison result.

[0006] Optionally, in the intelligent translation method of relaying human and AI translators described in this application, the step of obtaining the dataset to be translated, performing multimodal feature extraction and confidence calculation based on the dataset to be translated, and obtaining the semantic fusion vector of the data to be translated and the corresponding translation priority includes: Obtain the dataset to be translated, including text data, audio data, and image data; The text data, audio data, and image data to be translated are standardized and feature extracted to obtain semantic vectors of the data to be translated, including semantic feature vectors of the text data to be translated, semantic feature vectors of the audio data to be translated, and fusion feature vectors of the image data to be translated. The semantic feature vector of the text to be translated, the semantic feature vector of the speech to be translated, and the fused feature vector of the image to be translated are input into a preset scene classification and evaluation model for processing to obtain scene category feature data to be translated. The corresponding scene confidence correction parameters are obtained by querying the preset scene confidence correction parameter mapping table based on the scene category feature data to be translated; Obtain the preset initial confidence levels corresponding to the text data, audio data, and image data to be translated; The preset initial confidence level is corrected according to the scenario confidence correction parameters to obtain the confidence level value of the dataset to be translated; The semantic feature vectors of the text to be translated, the semantic feature vectors of the speech to be translated, and the fusion feature vectors of the image to be translated are weighted and fused with the corresponding confidence values ​​to obtain the semantic fusion vector of the data to be translated. The translation priority is determined based on the category feature data of the scene to be translated, including high priority or low priority.

[0007] Optionally, in the intelligent translation method of human and AI translators relaying each other as described in this application, the step of obtaining and processing the translation ability evaluation data and real-time translation load of human and AI translators to obtain real-time translation matching parameters corresponding to human and AI translators includes: Acquire translation ability assessment data and real-time translation workload of human translators and AI translators. The translation ability assessment data includes the total translation volume, average translation latency, and average translation score over a historical preset time period. The total translation volume, average translation delay, and average translation score are input into a preset translation ability assessment model for processing to obtain the translation ability assessment index corresponding to human translators and AI translators; Divide the translation capability assessment index by the real-time translation load to obtain the real-time translation matching parameters for human translators and AI translators.

[0008] Optionally, in the intelligent translation method of relaying human and AI translators described in this application, the step of analyzing and processing the dataset to be translated to obtain the translation difficulty level of the data to be translated includes: Obtain the lexical complexity and sentence complexity of the dataset to be translated; Based on the category feature data of the scene to be translated, query the preset translation difficulty scene correction coefficient mapping table to obtain the translation difficulty scene correction coefficient; The translation difficulty assessment index is obtained by weighting and summing the complexity of the words to be translated and the complexity of the sentences to be translated, and then multiplying the sum by the translation difficulty scenario correction coefficient. The translation difficulty assessment index is compared with the preset translation difficulty assessment threshold, and the translation difficulty level of the data to be translated is obtained according to the threshold range it falls into, including high difficulty or low difficulty.

[0009] Optionally, in the intelligent translation method of relaying human and AI translators described in this application, the step of allocating the semantic fusion vector or dataset of the data to be translated according to the real-time translation matching parameters combined with the priority and difficulty level of the data to be translated, to obtain the assigned data for human and AI translators, includes: Based on the real-time translation matching parameters and the feature data of the scene to be translated, available human translators and AI translators are determined. If the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator and the dataset to be translated will be assigned to the human translator. If the priority of the data to be translated is high and the difficulty level of the data to be translated is low, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator. If the priority of the data to be translated is low, the semantic fusion vector of the data to be translated will be allocated to the AI ​​translator according to a preset ratio, and the dataset to be translated will be allocated to the human translator.

[0010] Optionally, in the intelligent translation method of relaying human and AI translators described in this application, if the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated is assigned to the AI ​​translator, and the dataset to be translated is assigned to the human translator, including: Based on the semantic fusion vector of the data to be translated, extract the corresponding scene semantic prompt information and the scene-related term information; The scene semantic prompts, related terminology information, and the dataset to be translated are assigned to human translators.

[0011] Optionally, in the intelligent translation method of human and AI translators relaying each other as described in this application, the step of performing intelligent translation based on the data to be translated to obtain initial translation data corresponding to the AI ​​translator, and analyzing and processing the initial translation data to obtain initial translation quality evaluation parameters corresponding to the AI ​​translator, includes: Intelligent translation is performed based on the assigned data to be translated, to obtain the initial translation data corresponding to the AI ​​translator; The translation quality is verified by combining the initial translation data with a preset knowledge graph of the domain and scenario to be translated, and the terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are obtained. The terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are weighted and summed to obtain the initial translation quality evaluation parameters for the AI ​​translator.

[0012] Optionally, in the intelligent translation method of human and AI translators relaying each other as described in this application, the step of comparing the initial translation quality evaluation parameters with a preset initial translation quality evaluation threshold, and determining the real-time translation status of the AI ​​translator based on the threshold comparison result, includes: The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold. If the initial translation quality assessment parameter is greater than or equal to the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be qualified. If the initial translation quality assessment parameter is less than the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be unqualified.

[0013] Optionally, the intelligent translation method of relaying human and AI translators described in this application also includes: If the real-time translation status is unqualified, a human translator will make corrections based on the initial translation data to obtain translation correction trace data, including content correction data, format correction data, and human marking data. The initial translation data is replaced with the content correction data, format correction data, and manually marked data to obtain translation correction data; Based on the translation correction trace data, terminology correction data, scene tone adaptation feature data, and scene terminology association feature data are extracted; The preset domain and scene knowledge graph to be translated is updated based on the terminology correction data, scene tone adaptation feature data, and scene term association feature data.

[0014] Secondly, this application provides an intelligent translation system that combines human and AI translators, comprising: a memory and a processor, wherein the memory includes a program for an intelligent translation method that combines human and AI translators, and when the program for the intelligent translation method that combines human and AI translators is executed by the processor, it performs the following steps: Obtain the dataset to be translated, perform multimodal feature extraction and confidence calculation based on the dataset, and obtain the semantic fusion vector of the data to be translated and the corresponding translation priority; Acquire and process translation ability assessment data and real-time translation load of human translators and AI translators to obtain real-time translation matching parameters corresponding to human translators and AI translators; The difficulty level of the data to be translated is obtained by analyzing and processing the dataset to be translated. Based on the real-time translation matching parameters, combined with the translation priority and translation difficulty level, the semantic fusion vector or translation dataset of the data to be translated is allocated to obtain the translation allocation data corresponding to human translators and AI translators; Intelligent translation is performed based on the assigned data to be translated to obtain the initial translation data corresponding to the AI ​​translator. The initial translation data is then analyzed and processed to obtain the initial translation quality evaluation parameters corresponding to the AI ​​translator. The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold, and the real-time translation status of the AI ​​translator is determined based on the threshold comparison result.

[0015] As can be seen from the above, the intelligent translation method and system provided in this application, which integrates translation data based on the translation scenario, dynamically allocates translation tasks, quantitatively evaluates the initial translation quality, and performs two-level intelligent iterative optimization, thereby achieving effective collaborative translation between human and AI translators and improving translation efficiency and quality.

[0016] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1A flowchart illustrating an intelligent translation method involving a human and AI translator working in relay, provided as an embodiment of this application; Figure 2 A flowchart illustrating the process of obtaining the semantic fusion vector of the data to be translated and the corresponding translation priority in an intelligent translation method that combines human and AI translators, as provided in this application embodiment; Figure 3 A flowchart illustrating the process of obtaining real-time translation matchable parameters for human and AI translators in an intelligent translation method that combines human and AI translators, as provided in this application embodiment; Figure 4 This is a flowchart illustrating the process of obtaining the translation difficulty level of the data to be translated in an intelligent translation method involving a relay between human and AI translators, as provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating an intelligent translation method involving a human and AI translator relay, as described in some embodiments of this application. This intelligent translation method involving a human and AI translator relay is used in terminal devices, such as computers and mobile phones. The intelligent translation method involving a human and AI translator relay includes the following steps: S111. Obtain the dataset to be translated, perform multimodal feature extraction and confidence calculation based on the dataset to be translated, and obtain the semantic fusion vector of the data to be translated and the corresponding translation priority; S112. Obtain translation ability assessment data and real-time translation load of human translators and AI translators, process them, and obtain real-time translation matching parameters corresponding to human translators and AI translators; S113. Analyze and process the dataset to be translated to obtain the difficulty level of the data to be translated. S12. Based on the real-time translation matching parameters and the translation priority and translation difficulty level, the semantic fusion vector or translation dataset of the data to be translated is allocated to obtain the translation allocation data corresponding to human translators and AI translators. S13. Perform intelligent translation based on the data to be translated to obtain the initial translation data corresponding to the AI ​​translator. Analyze and process the initial translation data to obtain the initial translation quality evaluation parameters corresponding to the AI ​​translator. S14. Compare the initial translation quality evaluation parameters with the preset initial translation quality evaluation threshold, and determine the real-time translation status of the AI ​​translator based on the threshold comparison result.

[0022] To further explain, in order to improve translation efficiency and accuracy, the multimodal data to be translated is first standardized and its features are extracted, and then dynamically fused based on the evaluated categories of the scenarios to be translated. Then, the priority, difficulty, and translation capabilities of human translators and AI translators are evaluated separately, and then the task is dynamically assigned. After that, intelligent translation is performed, and the translation quality is evaluated. Based on the translation quality, it is determined whether human translators need to make corrections. At the same time, real-time corrections and long-term iterative optimizations are performed based on the correction results.

[0023] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating the process of obtaining the semantic fusion vector of the data to be translated and the corresponding translation priority in an intelligent translation method involving a human and AI translator relay, as described in some embodiments of this application. According to embodiments of the present invention, obtaining the dataset to be translated, performing multimodal feature extraction and confidence calculation based on the dataset to obtain the semantic fusion vector of the data to be translated and the corresponding translation priority, includes: S21. Obtain the dataset to be translated, including text data, audio data, and image data; S211. Standardize and extract features from the text data to be translated, audio data to be translated, and image data to be translated to obtain semantic vectors of the data to be translated, including semantic feature vectors of the text to be translated, semantic feature vectors of the speech to be translated, and fusion feature vectors of the image to be translated. S212. Input the semantic feature vector of the text to be translated, the semantic feature vector of the speech to be translated, and the fusion feature vector of the image to be translated into a preset scene classification and evaluation model for processing to obtain scene category feature data to be translated. S2121. Query the preset scene confidence correction parameter mapping table according to the scene category feature data to be translated, and obtain the corresponding scene confidence correction parameters; S22. Obtain the preset initial confidence levels corresponding to the text data to be translated, the audio data to be translated, and the image data to be translated; S23. Correct the preset initial confidence level according to the scenario confidence correction parameter to obtain the confidence level value of the dataset to be translated; S24. The semantic feature vector of the text to be translated, the semantic feature vector of the speech to be translated, and the fusion feature vector of the image to be translated are weighted and fused with the corresponding confidence values ​​to obtain the semantic fusion vector of the data to be translated. S2122. Determine the translation priority based on the category feature data of the scene to be translated, including high priority or low priority.

[0024] Further explanation is needed regarding the standardization of text, audio, and image data to be translated, which improves the usability of the data. For example, in this paper, the data to be translated is split into sentences to generate a sequence of text segments; the audio data is segmented into a sequence of speech frames with a preset frame length and frame shift; and the text content is extracted from the image data. Then, feature extraction is performed based on the standardized data: semantic feature vectors are extracted from the text data, semantic feature vectors are extracted from the audio data, and visual features are extracted from the image data. Semantic features are then extracted from the extracted text, and finally, the data is concatenated to form the data to be translated. Image fusion feature vectors are used to evaluate the scene category feature data corresponding to the data to be translated, based on the extracted feature vectors. Scene categories include cross-border e-commerce, disaster relief, business meetings, and medical diagnosis. Scene category feature data is represented by a unique identifier. To improve the accuracy of the fused data, a preset scene confidence correction parameter mapping table is queried using the scene category feature data to obtain the corresponding scene confidence correction parameters. The preset initial confidence scores for each modality of the data to be translated are then corrected by multiplying the two parameters to determine the confidence score value corresponding to the dataset to be translated. Examples of preset initial confidence scores are shown in Table 1. Table 1 Sample of Preset Initial Confidence Level

[0025] Dynamic fusion processing is performed based on the extracted feature vectors and normalized confidence values ​​to avoid low reliability modes reducing fusion credibility. At the same time, the corresponding translation priority is determined according to different translation scenario categories through preset rules, such as medical diagnosis being higher than business meetings. The preset scenario confidence correction parameter mapping table and preset rules are pre-constructed by experts in the field and can be dynamically adjusted. The preset translation scenario classification evaluation model is obtained by training with a large number of historical samples of semantic feature vectors of the text to be translated, semantic feature vectors of the speech to be translated, and fused feature vectors of the image to be translated, as well as the corresponding translation scenario category feature data.

[0026] Please refer to Figure 3 , Figure 3This is a flowchart illustrating the process of obtaining real-time translation matching parameters for human and AI translators in an intelligent translation method involving human and AI translators in some embodiments of this application. According to embodiments of the present invention, the step of obtaining and processing the translation ability evaluation data and real-time translation load of human and AI translators to obtain real-time translation matching parameters for human and AI translators includes: S31. Obtain translation ability assessment data and real-time translation load of human translators and AI translators. The translation ability assessment data includes the total translation volume, average translation latency, and average translation score over a historical preset time period. S32. Input the total translation volume, average translation delay, and average translation score into a preset translation ability evaluation model for processing to obtain the translation ability evaluation index corresponding to human translators and AI translators; S33. Divide the translation capability evaluation index by the real-time translation load to obtain the real-time translation matching parameters corresponding to human translators and AI translators.

[0027] Further explanation is needed. By evaluating the total translation volume, average translation delay, and average translation score within a historical preset time period (such as the past six months), the total translation volume refers to the total vocabulary translated, and the average translation delay refers to the average difference between the completion time of each translation and the required translation time. If the difference is negative, it is recorded as 0. A preset translation ability assessment model, pre-trained with a large number of historical samples of the total translation volume, average translation delay, and average translation score, as well as the corresponding translation ability assessment index, processes the real-time total translation volume, average translation delay, and average translation score to determine the corresponding translation ability assessment index. Then, combined with the real-time translation load of human translators and AI translators, the real-time translation load in this embodiment is represented by integers from 0 to 100 to determine its real-time translation matching parameter. The larger the value, the more idle it is, and the more real-time translation tasks can be assigned. If the real-time translation load is 0, that is, there are no translation tasks, the real-time translation matching parameter is directly recorded as the maximum value.

[0028] Please refer to Figure 4 , Figure 4 This is a flowchart illustrating the process of obtaining the translation difficulty level of the data to be translated in an intelligent translation method involving a human and AI translator relay, as described in some embodiments of this application. According to embodiments of the present invention, the step of analyzing and processing the dataset to be translated to obtain the translation difficulty level includes: S411. Obtain the lexical complexity and sentence complexity of the dataset to be translated; S412. Based on the category feature data of the scene to be translated, query the preset translation difficulty scene correction coefficient mapping table to obtain the translation difficulty scene correction coefficient; S42. The complexity of the words to be translated and the complexity of the sentences to be translated are weighted and summed, and then multiplied by the translation difficulty scenario correction coefficient to obtain the translation difficulty assessment index. S43. Compare the translation difficulty assessment index with the preset translation difficulty assessment threshold, and obtain the translation difficulty level of the data to be translated, including high difficulty or low difficulty, according to the threshold range it falls into.

[0029] Further explanation is needed: to assess the translation difficulty of the data to be translated, firstly, the complexity of the vocabulary to be translated is determined by weighted summation after data normalization, considering the proportion of specialized terms (i.e., the ratio of specialized vocabulary to the total vocabulary), the proportion of rare characters, and the proportion of unique words. Rare characters and unique words are pre-constructed into corresponding lexicons by those skilled in the art. Then, the complexity of the sentences to be translated is determined by weighted summation after data normalization, considering the average sentence length (i.e., the ratio of total vocabulary to the number of sentences), the proportion of sentences with complex structures (e.g., the ratio of sentences containing inversions to the total number of sentences), and the semantic association depth (i.e., the number of sentences with pre- and post-sentence dependencies). Simultaneously, based on... The evaluation of the translation scenario category feature data is queried from a preset translation difficulty scenario correction coefficient mapping table to obtain the translation difficulty scenario correction coefficient. The preset translation difficulty scenario correction coefficient mapping table is pre-constructed by those skilled in the art and can be dynamically adjusted. The construction logic is that the correction coefficient is larger for those with high requirements for colloquialism and many professional terms and high update frequency, and smaller for those requiring conciseness and clarity. Finally, the normalized vocabulary complexity and sentence complexity of the translation scenario are weighted and summed and multiplied by the determined translation difficulty scenario correction coefficient to obtain the translation difficulty assessment index. By comparing the thresholds, the translation data is determined to be of high or low difficulty, thereby realizing a two-dimensional difficulty assessment based on vocabulary and sentence.

[0030] According to an embodiment of the present invention, the step of allocating the semantic fusion vector or dataset of the data to be translated based on the real-time translation matching parameters combined with the priority and difficulty level of the data to be translated, to obtain the data to be translated for human translators and AI translators, includes: Based on the real-time translation matching parameters and the feature data of the scene to be translated, available human translators and AI translators are determined. If the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator and the dataset to be translated will be assigned to the human translator. If the priority of the data to be translated is high and the difficulty level of the data to be translated is low, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator. If the priority of the data to be translated is low, the semantic fusion vector of the data to be translated will be allocated to the AI ​​translator according to a preset ratio, and the dataset to be translated will be allocated to the human translator.

[0031] Further explanation is needed regarding the process of accurately assigning translation tasks to suitable human translators, AI translators, or both. First, based on real-time translation matching parameters and the category feature data of the translation scenario, human translators and available AI translators in the corresponding domains are determined. Then, if the data to be translated is assessed as high-priority and high-difficulty, the semantic fusion vector of the data is assigned to an AI translator for initial translation and output of scenario semantics and terminology suggestions. Simultaneously, a human translator quickly and accurately completes the translation based on the dataset, the corresponding initial translation results, scenario semantics, and terminology suggestions. If the task is high-priority and low-difficulty, it is assigned only to an AI translator, and subsequent evaluation of the AI ​​translator's translation is sufficient. If the task is low-priority, to reduce the translation load and balance translation efficiency, the tasks are assigned according to a preset ratio, such as a 7:3 ratio of AI translators to human translators, thus achieving dynamic matching and ensuring accuracy while improving translation efficiency.

[0032] According to an embodiment of the present invention, if the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated is assigned to an AI translator, and the dataset to be translated is assigned to a human translator, including: Based on the semantic fusion vector of the data to be translated, extract the corresponding scene semantic prompt information and the scene-related term information; The scene semantic prompts, related terminology information, and the dataset to be translated are assigned to human translators.

[0033] It needs further explanation that high-priority and high-difficulty data to be translated requires rapid and accurate completion. AI translators and human translators need to collaborate to complete the task. AI translators improve efficiency, while human translators ensure accuracy. AI translators perform an initial translation based on the semantic fusion vector of the data to be translated, while outputting corresponding scene semantic prompts and related terminology information. Scene semantic prompts, such as cross-border e-commerce, help retain colloquial expressions in translation. Related terminology information focuses on related terms in similar translation scenarios, such as "buy blindly" in cross-border e-commerce. When allocating the dataset to be translated to human translators, the scene semantic prompts and related terminology information output by the AI ​​translators are allocated together, thereby improving translation accuracy and efficiency.

[0034] According to an embodiment of the present invention, the step of performing intelligent translation based on the assigned data to be translated to obtain initial translation data corresponding to the AI ​​translator, and analyzing and processing the initial translation data to obtain initial translation quality evaluation parameters corresponding to the AI ​​translator, includes: Intelligent translation is performed based on the assigned data to be translated, to obtain the initial translation data corresponding to the AI ​​translator; The translation quality is verified by combining the initial translation data with a preset knowledge graph of the domain and scenario to be translated, and the terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are obtained. The terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are weighted and summed to obtain the initial translation quality evaluation parameters for the AI ​​translator.

[0035] Further explanation is needed: to accurately evaluate the initial translation results of the AI ​​translator, a two-level knowledge graph containing the translation domain and the translation scenario is pre-constructed. The translation domain covers common core fields, such as medicine and law. Each translation domain is further divided into different translation scenarios, such as the emergency room scenario in the medical translation domain. The initial translation data is verified for translation quality through the pre-set translation domain and scenario knowledge graph. The terminology domain accuracy and terminology scenario accuracy are evaluated through a dual mechanism of pre-set dictionary matching and pre-set BERT terminology recognition model. The terminology consistency rate is determined by the translation text within and across the translated text. The terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are weighted and summed to determine the initial translation quality evaluation parameters corresponding to the AI ​​translator, which are used to evaluate the translation quality of the AI ​​translator.

[0036] According to an embodiment of the present invention, the step of comparing the initial translation quality evaluation parameters with a preset initial translation quality evaluation threshold, and determining the real-time translation status of the AI ​​translator based on the threshold comparison result, includes: The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold. If the initial translation quality assessment parameter is greater than or equal to the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be qualified. If the initial translation quality assessment parameter is less than the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be unqualified.

[0037] It needs further explanation that the threshold comparison is used to determine whether the real-time translation status of the AI ​​translator is up to standard, and then further processing is carried out.

[0038] According to an embodiment of the present invention, it further includes: If the real-time translation status is unqualified, a human translator will make corrections based on the initial translation data to obtain translation correction trace data, including content correction data, format correction data, and human marking data. The initial translation data is replaced with the content correction data, format correction data, and manually marked data to obtain translation correction data; Based on the translation correction trace data, terminology correction data, scene tone adaptation feature data, and scene terminology association feature data are extracted; The preset domain and scene knowledge graph to be translated is updated based on the terminology correction data, scene tone adaptation feature data, and scene term association feature data.

[0039] Further explanation is needed: when the AI ​​translator's real-time translation performance is deemed unsatisfactory, the initial translation data is sent to a human translator for timely correction. Translation correction trace data is collected, including content correction data, format correction data, and human-labeled data. For example, in cross-border e-commerce, the translation of "buy it" is changed from "buy it" to "grab it now" to maintain a conversational style. Format correction data, for example, in disaster relief scenarios, involves breaking long sentences into shorter ones. Human-labeled data, such as highlighting product features in this scenario, replaces the initial translation data with the collected content correction data, format correction data, and human-labeled data to obtain translation correction data, achieving real-time optimization. Simultaneously, to improve the AI ​​translator's adaptability to similar translations in the future, terminology correction data, scene tone adaptation feature data (scene tone adaptation such as conversational expressions), and scene terminology association feature data are extracted from the human translator's translation correction trace data (e.g., translating "getting on a bus" as "grab it" in a cross-border e-commerce scenario, but "join" in a business meeting). The project dynamically updates the pre-built knowledge graph of the target domain and scenario to be translated, and achieves iterative optimization, thus realizing two-level intelligent iterative optimization.

[0040] It is worth mentioning that, according to embodiments of the present invention, it further includes: If term matching fails by using the preset domain and scene knowledge graph, the frequency of occurrence of the corresponding term will be counted. If the frequency of the term is less than a preset frequency threshold, it is determined to be a common word; If the frequency of the term is greater than or equal to a preset frequency threshold, it is determined to be an unrecorded term. Based on the adjacent words corresponding to the uncollected terms, temporary translation data is obtained by analyzing and processing them through a preset BERT terminology recognition model and a preset knowledge graph of the domain and scenario to be translated. The temporary translation data is transmitted to a human translator for translation confirmation. If confirmed, the preset knowledge graph of the domain and scenario to be translated will be optimized based on the uncollected terms and corresponding temporary translation data. If not confirmed, the preset knowledge graph of the domain and scenario to be translated will be optimized based on the uncollected terms and corresponding human translation data.

[0041] It needs further explanation that new words not included in the translation scenario often appear, which can easily lead to omissions or errors in translation. In this embodiment, when term matching fails, it is first determined whether the word is not included by checking whether the frequency of the word in the text exceeds a threshold. Then, temporary translation data is generated through adjacent word analysis and semantic transfer. Finally, it is manually confirmed and verified, thereby realizing the temporary and rapid translation of emerging terms and iterative optimization of the knowledge graph. The preset BERT terminology recognition model generates semantic vectors of unincluded terms and known terms in the same scenario based on unincluded terms and their adjacent words, and performs association similarity analysis. If the similarity is greater than a preset threshold, it is determined to be relevant, thereby determining the temporary translation data.

[0042] This invention also discloses an intelligent translation system that combines human and AI translators, including a memory and a processor. The memory includes a program for an intelligent translation method that combines human and AI translators. When the processor executes the program for the intelligent translation method that combines human and AI translators, it performs the following steps: Obtain the dataset to be translated, perform multimodal feature extraction and confidence calculation based on the dataset, and obtain the semantic fusion vector of the data to be translated and the corresponding translation priority; Acquire and process translation ability assessment data and real-time translation load of human translators and AI translators to obtain real-time translation matching parameters corresponding to human translators and AI translators; The difficulty level of the data to be translated is obtained by analyzing and processing the dataset to be translated. Based on the real-time translation matching parameters, combined with the translation priority and translation difficulty level, the semantic fusion vector or translation dataset of the data to be translated is allocated to obtain the translation allocation data corresponding to human translators and AI translators; Intelligent translation is performed based on the assigned data to be translated to obtain the initial translation data corresponding to the AI ​​translator. The initial translation data is then analyzed and processed to obtain the initial translation quality evaluation parameters corresponding to the AI ​​translator. The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold, and the real-time translation status of the AI ​​translator is determined based on the threshold comparison result.

[0043] To further explain, in order to improve translation efficiency and accuracy, the multimodal data to be translated is first standardized and its features are extracted, and then dynamically fused based on the evaluated categories of the scenarios to be translated. Then, the priority, difficulty, and translation capabilities of human translators and AI translators are evaluated separately, and then the task is dynamically assigned. After that, intelligent translation is performed, and the translation quality is evaluated. Based on the translation quality, it is determined whether human translators need to make corrections. At the same time, real-time corrections and long-term iterative optimizations are performed based on the correction results.

[0044] According to an embodiment of the present invention, the step of obtaining the dataset to be translated, performing multimodal feature extraction and confidence calculation based on the dataset to be translated, and obtaining the semantic fusion vector of the data to be translated and the corresponding translation priority includes: Obtain the dataset to be translated, including text data, audio data, and image data; The text data, audio data, and image data to be translated are standardized and feature extracted to obtain semantic vectors of the data to be translated, including semantic feature vectors of the text data to be translated, semantic feature vectors of the audio data to be translated, and fusion feature vectors of the image data to be translated. The semantic feature vector of the text to be translated, the semantic feature vector of the speech to be translated, and the fused feature vector of the image to be translated are input into a preset scene classification and evaluation model for processing to obtain scene category feature data to be translated. The corresponding scene confidence correction parameters are obtained by querying the preset scene confidence correction parameter mapping table based on the scene category feature data to be translated; Obtain the preset initial confidence levels corresponding to the text data, audio data, and image data to be translated; The preset initial confidence level is corrected according to the scenario confidence correction parameters to obtain the confidence level value of the dataset to be translated; The semantic feature vectors of the text to be translated, the semantic feature vectors of the speech to be translated, and the fusion feature vectors of the image to be translated are weighted and fused with the corresponding confidence values ​​to obtain the semantic fusion vector of the data to be translated. The translation priority is determined based on the category feature data of the scene to be translated, including high priority or low priority.

[0045] Further explanation is needed regarding the standardization of text, audio, and image data to be translated, which improves the usability of the data. For example, in this paper, the data to be translated is split into sentences to generate a sequence of text segments; the audio data is segmented into a sequence of speech frames with a preset frame length and frame shift; and the text content is extracted from the image data. Then, feature extraction is performed based on the standardized data: semantic feature vectors are extracted from the text data, semantic feature vectors are extracted from the audio data, and visual features are extracted from the image data. Semantic features are then extracted from the extracted text, and finally, the data is concatenated to form the data to be translated. Image fusion feature vectors are used to evaluate the scene category feature data corresponding to the data to be translated, based on the extracted feature vectors. Scene categories include cross-border e-commerce, disaster relief, business meetings, and medical diagnosis. Scene category feature data is represented by a unique identifier. To improve the accuracy of the fused data, a preset scene confidence correction parameter mapping table is queried using the scene category feature data to obtain the corresponding scene confidence correction parameters. The preset initial confidence scores for each modality of the data to be translated are then corrected by multiplying the two parameters to determine the confidence score value corresponding to the dataset to be translated. Examples of preset initial confidence scores are shown in Table 1. Table 1 Sample of Preset Initial Confidence Level

[0046] Dynamic fusion processing is performed based on the extracted feature vectors and normalized confidence values ​​to avoid low reliability modes reducing fusion credibility. At the same time, the corresponding translation priority is determined according to different translation scenario categories through preset rules, such as medical diagnosis being higher than business meetings. The preset scenario confidence correction parameter mapping table and preset rules are pre-constructed by experts in the field and can be dynamically adjusted. The preset translation scenario classification evaluation model is obtained by training with a large number of historical samples of semantic feature vectors of the text to be translated, semantic feature vectors of the speech to be translated, and fused feature vectors of the image to be translated, as well as the corresponding translation scenario category feature data.

[0047] According to an embodiment of the present invention, the step of acquiring and processing the translation ability assessment data and real-time translation load of human translators and AI translators to obtain real-time translation matchable parameters corresponding to human translators and AI translators includes: Acquire translation ability assessment data and real-time translation workload of human translators and AI translators. The translation ability assessment data includes the total translation volume, average translation latency, and average translation score over a historical preset time period. The total translation volume, average translation delay, and average translation score are input into a preset translation ability assessment model for processing to obtain the translation ability assessment index corresponding to human translators and AI translators; Divide the translation capability assessment index by the real-time translation load to obtain the real-time translation matching parameters for human translators and AI translators.

[0048] Further explanation is needed. By evaluating the total translation volume, average translation delay, and average translation score within a historical preset time period (such as the past six months), the total translation volume refers to the total vocabulary translated, and the average translation delay refers to the average difference between the completion time of each translation and the required translation time. If the difference is negative, it is recorded as 0. A preset translation ability assessment model, pre-trained with a large number of historical samples of the total translation volume, average translation delay, and average translation score, as well as the corresponding translation ability assessment index, processes the real-time total translation volume, average translation delay, and average translation score to determine the corresponding translation ability assessment index. Then, combined with the real-time translation load of human translators and AI translators, the real-time translation load in this embodiment is represented by integers from 0 to 100 to determine its real-time translation matching parameter. The larger the value, the more idle it is, and the more real-time translation tasks can be assigned. If the real-time translation load is 0, that is, there are no translation tasks, the real-time translation matching parameter is directly recorded as the maximum value.

[0049] According to an embodiment of the present invention, the step of analyzing and processing the dataset to be translated to obtain the translation difficulty level of the data to be translated includes: Obtain the lexical complexity and sentence complexity of the dataset to be translated; Based on the category feature data of the scene to be translated, query the preset translation difficulty scene correction coefficient mapping table to obtain the translation difficulty scene correction coefficient; The translation difficulty assessment index is obtained by weighting and summing the complexity of the words to be translated and the complexity of the sentences to be translated, and then multiplying the sum by the translation difficulty scenario correction coefficient. The translation difficulty assessment index is compared with the preset translation difficulty assessment threshold, and the translation difficulty level of the data to be translated is obtained according to the threshold range it falls into, including high difficulty or low difficulty.

[0050] Further explanation is needed: to assess the translation difficulty of the data to be translated, firstly, the complexity of the vocabulary to be translated is determined by weighted summation after data normalization, considering the proportion of specialized terms (i.e., the ratio of specialized vocabulary to the total vocabulary), the proportion of rare characters, and the proportion of unique words. Rare characters and unique words are pre-constructed into corresponding lexicons by those skilled in the art. Then, the complexity of the sentences to be translated is determined by weighted summation after data normalization, considering the average sentence length (i.e., the ratio of total vocabulary to the number of sentences), the proportion of sentences with complex structures (e.g., the ratio of sentences containing inversions to the total number of sentences), and the semantic association depth (i.e., the number of sentences with pre- and post-sentence dependencies). Simultaneously, based on... The evaluation of the translation scenario category feature data is queried from a preset translation difficulty scenario correction coefficient mapping table to obtain the translation difficulty scenario correction coefficient. The preset translation difficulty scenario correction coefficient mapping table is pre-constructed by those skilled in the art and can be dynamically adjusted. The construction logic is that the correction coefficient is larger for those with high requirements for colloquialism and many professional terms and high update frequency, and smaller for those requiring conciseness and clarity. Finally, the normalized vocabulary complexity and sentence complexity of the translation scenario are weighted and summed and multiplied by the determined translation difficulty scenario correction coefficient to obtain the translation difficulty assessment index. By comparing the thresholds, the translation data is determined to be of high or low difficulty, thereby realizing a two-dimensional difficulty assessment based on vocabulary and sentence.

[0051] According to an embodiment of the present invention, the step of allocating the semantic fusion vector or dataset of the data to be translated based on the real-time translation matching parameters combined with the priority and difficulty level of the data to be translated, to obtain the data to be translated for human translators and AI translators, includes: Based on the real-time translation matching parameters and the feature data of the scene to be translated, available human translators and AI translators are determined. If the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator and the dataset to be translated will be assigned to the human translator. If the priority of the data to be translated is high and the difficulty level of the data to be translated is low, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator. If the priority of the data to be translated is low, the semantic fusion vector of the data to be translated will be allocated to the AI ​​translator according to a preset ratio, and the dataset to be translated will be allocated to the human translator.

[0052] Further explanation is needed regarding the process of accurately assigning translation tasks to suitable human translators, AI translators, or both. First, based on real-time translation matching parameters and the category feature data of the translation scenario, human translators and available AI translators in the corresponding domains are determined. Then, if the data to be translated is assessed as high-priority and high-difficulty, the semantic fusion vector of the data is assigned to an AI translator for initial translation and output of scenario semantics and terminology suggestions. Simultaneously, a human translator quickly and accurately completes the translation based on the dataset, the corresponding initial translation results, scenario semantics, and terminology suggestions. If the task is high-priority and low-difficulty, it is assigned only to an AI translator, and subsequent evaluation of the AI ​​translator's translation is sufficient. If the task is low-priority, to reduce the translation load and balance translation efficiency, the tasks are assigned according to a preset ratio, such as a 7:3 ratio of AI translators to human translators, thus achieving dynamic matching and ensuring accuracy while improving translation efficiency.

[0053] According to an embodiment of the present invention, if the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated is assigned to an AI translator, and the dataset to be translated is assigned to a human translator, including: Based on the semantic fusion vector of the data to be translated, extract the corresponding scene semantic prompt information and the scene-related term information; The scene semantic prompts, related terminology information, and the dataset to be translated are assigned to human translators.

[0054] It needs further explanation that high-priority and high-difficulty data to be translated requires rapid and accurate completion. AI translators and human translators need to collaborate to complete the task. AI translators improve efficiency, while human translators ensure accuracy. AI translators perform an initial translation based on the semantic fusion vector of the data to be translated, while outputting corresponding scene semantic prompts and related terminology information. Scene semantic prompts, such as cross-border e-commerce, help retain colloquial expressions in translation. Related terminology information focuses on related terms in similar translation scenarios, such as "buy blindly" in cross-border e-commerce. When allocating the dataset to be translated to human translators, the scene semantic prompts and related terminology information output by the AI ​​translators are allocated together, thereby improving translation accuracy and efficiency.

[0055] According to an embodiment of the present invention, the step of performing intelligent translation based on the assigned data to be translated to obtain initial translation data corresponding to the AI ​​translator, and analyzing and processing the initial translation data to obtain initial translation quality evaluation parameters corresponding to the AI ​​translator, includes: Intelligent translation is performed based on the assigned data to be translated, to obtain the initial translation data corresponding to the AI ​​translator; The translation quality is verified by combining the initial translation data with a preset knowledge graph of the domain and scenario to be translated, and the terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are obtained. The terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are weighted and summed to obtain the initial translation quality evaluation parameters for the AI ​​translator.

[0056] Further explanation is needed: to accurately evaluate the initial translation results of the AI ​​translator, a two-level knowledge graph containing the translation domain and the translation scenario is pre-constructed. The translation domain covers common core fields, such as medicine and law. Each translation domain is further divided into different translation scenarios, such as the emergency room scenario in the medical translation domain. The initial translation data is verified for translation quality through the pre-set translation domain and scenario knowledge graph. The terminology domain accuracy and terminology scenario accuracy are evaluated through a dual mechanism of pre-set dictionary matching and pre-set BERT terminology recognition model. The terminology consistency rate is determined by the translation text within and across the translated text. The terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are weighted and summed to determine the initial translation quality evaluation parameters corresponding to the AI ​​translator, which are used to evaluate the translation quality of the AI ​​translator.

[0057] According to an embodiment of the present invention, the step of comparing the initial translation quality evaluation parameters with a preset initial translation quality evaluation threshold, and determining the real-time translation status of the AI ​​translator based on the threshold comparison result, includes: The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold. If the initial translation quality assessment parameter is greater than or equal to the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be qualified. If the initial translation quality assessment parameter is less than the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be unqualified.

[0058] It needs further explanation that the threshold comparison is used to determine whether the real-time translation status of the AI ​​translator is up to standard, and then further processing is carried out.

[0059] According to an embodiment of the present invention, it further includes: If the real-time translation status is unqualified, a human translator will make corrections based on the initial translation data to obtain translation correction trace data, including content correction data, format correction data, and human marking data. The initial translation data is replaced with the content correction data, format correction data, and manually marked data to obtain translation correction data; Based on the translation correction trace data, terminology correction data, scene tone adaptation feature data, and scene terminology association feature data are extracted; The preset domain and scene knowledge graph to be translated is updated based on the terminology correction data, scene tone adaptation feature data, and scene term association feature data.

[0060] Further explanation is needed: when the AI ​​translator's real-time translation performance is deemed unsatisfactory, the initial translation data is sent to a human translator for timely correction. Translation correction trace data is collected, including content correction data, format correction data, and human-labeled data. For example, in cross-border e-commerce, the translation of "buy it" is changed from "buy it" to "grab it now" to maintain a conversational style. Format correction data, for example, in disaster relief scenarios, involves breaking long sentences into shorter ones. Human-labeled data, such as highlighting product features in this scenario, replaces the initial translation data with the collected content correction data, format correction data, and human-labeled data to obtain translation correction data, achieving real-time optimization. Simultaneously, to improve the AI ​​translator's adaptability to similar translations in the future, terminology correction data, scene tone adaptation feature data (scene tone adaptation such as conversational expressions), and scene terminology association feature data are extracted from the human translator's translation correction trace data (e.g., translating "getting on a bus" as "grab it" in a cross-border e-commerce scenario, but "join" in a business meeting). The project dynamically updates the pre-built knowledge graph of the target domain and scenario to be translated, and achieves iterative optimization, thus realizing two-level intelligent iterative optimization.

[0061] It is worth mentioning that, according to embodiments of the present invention, it further includes: If term matching fails by using the preset domain and scene knowledge graph, the frequency of occurrence of the corresponding term will be counted. If the frequency of the term is less than a preset frequency threshold, it is determined to be a common word; If the frequency of the term is greater than or equal to a preset frequency threshold, it is determined to be an unrecorded term. Based on the adjacent words corresponding to the uncollected terms, temporary translation data is obtained by analyzing and processing them through a preset BERT terminology recognition model and a preset knowledge graph of the domain and scenario to be translated. The temporary translation data is transmitted to a human translator for translation confirmation. If confirmed, the preset knowledge graph of the domain and scenario to be translated will be optimized based on the uncollected terms and corresponding temporary translation data. If not confirmed, the preset knowledge graph of the domain and scenario to be translated will be optimized based on the uncollected terms and corresponding human translation data.

[0062] It needs further explanation that new words not included in the translation scenario often appear, which can easily lead to omissions or errors in translation. In this embodiment, when term matching fails, it is first determined whether the word is not included by checking whether the frequency of the word in the text exceeds a threshold. Then, temporary translation data is generated through adjacent word analysis and semantic transfer. Finally, it is manually confirmed and verified, thereby realizing the temporary and rapid translation of emerging terms and iterative optimization of the knowledge graph. The preset BERT terminology recognition model generates semantic vectors of unincluded terms and known terms in the same scenario based on unincluded terms and their adjacent words, and performs association similarity analysis. If the similarity is greater than a preset threshold, it is determined to be relevant, thereby determining the temporary translation data.

[0063] This invention discloses an intelligent translation method and system that combines human and AI translators. By integrating translation data based on the translation scenario, dynamically allocating translation tasks, quantitatively evaluating the initial translation quality, and performing two-level intelligent iterative optimization, it enables effective collaborative translation between human and AI translators, thereby improving translation efficiency and quality.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0065] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0066] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0067] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A method for intelligent translation involving a relay between human and AI translators, characterized in that, Includes the following steps: Obtain the dataset to be translated, perform multimodal feature extraction and confidence calculation based on the dataset, and obtain the semantic fusion vector of the data to be translated and the corresponding translation priority; Acquire and process translation ability assessment data and real-time translation load of human translators and AI translators to obtain real-time translation matching parameters corresponding to human translators and AI translators; The difficulty level of the data to be translated is obtained by analyzing and processing the dataset to be translated. Based on the real-time translation matching parameters, combined with the translation priority and translation difficulty level, the semantic fusion vector or translation dataset of the data to be translated is allocated to obtain the translation allocation data corresponding to human translators and AI translators; Intelligent translation is performed based on the assigned data to be translated to obtain the initial translation data corresponding to the AI ​​translator. The initial translation data is then analyzed and processed to obtain the initial translation quality evaluation parameters corresponding to the AI ​​translator. The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold, and the real-time translation status of the AI ​​translator is determined based on the threshold comparison result.

2. The intelligent translation method of relaying human and AI translators according to claim 1, characterized in that, The process of obtaining the dataset to be translated, performing multimodal feature extraction and confidence calculation based on the dataset, and obtaining the semantic fusion vector of the data to be translated and the corresponding translation priority includes: Obtain the dataset to be translated, including text data, audio data, and image data; The text data, audio data, and image data to be translated are standardized and feature extracted to obtain semantic vectors of the data to be translated, including semantic feature vectors of the text data to be translated, semantic feature vectors of the audio data to be translated, and fusion feature vectors of the image data to be translated. The semantic feature vector of the text to be translated, the semantic feature vector of the speech to be translated, and the fused feature vector of the image to be translated are input into a preset scene classification and evaluation model for processing to obtain scene category feature data to be translated. The corresponding scene confidence correction parameters are obtained by querying the preset scene confidence correction parameter mapping table based on the scene category feature data to be translated; Obtain the preset initial confidence levels corresponding to the text data, audio data, and image data to be translated; The preset initial confidence level is corrected according to the scenario confidence correction parameters to obtain the confidence level value of the dataset to be translated; The semantic feature vectors of the text to be translated, the semantic feature vectors of the speech to be translated, and the fusion feature vectors of the image to be translated are weighted and fused with the corresponding confidence values ​​to obtain the semantic fusion vector of the data to be translated. The translation priority is determined based on the category feature data of the scene to be translated, including high priority or low priority.

3. The intelligent translation method of relaying human and AI translators according to claim 2, characterized in that, The process involves acquiring and processing translation capability assessment data and real-time translation load from both human and AI translators to obtain real-time translation matchable parameters for both types of translators, including: Acquire translation ability assessment data and real-time translation workload of human translators and AI translators. The translation ability assessment data includes the total translation volume, average translation latency, and average translation score over a historical preset time period. The total translation volume, average translation delay, and average translation score are input into a preset translation ability assessment model for processing to obtain the translation ability assessment index corresponding to human translators and AI translators; Divide the translation capability assessment index by the real-time translation load to obtain the real-time translation matching parameters for human translators and AI translators.

4. The intelligent translation method of relaying human and AI translators according to claim 3, characterized in that, The step of analyzing and processing the dataset to be translated to obtain the translation difficulty level of the dataset includes: Obtain the lexical complexity and sentence complexity of the dataset to be translated; Based on the category feature data of the scene to be translated, query the preset translation difficulty scene correction coefficient mapping table to obtain the translation difficulty scene correction coefficient; The translation difficulty assessment index is obtained by weighting and summing the complexity of the words to be translated and the complexity of the sentences to be translated, and then multiplying the sum by the translation difficulty scenario correction coefficient. The translation difficulty assessment index is compared with the preset translation difficulty assessment threshold, and the translation difficulty level of the data to be translated is obtained according to the threshold range it falls into, including high difficulty or low difficulty.

5. The intelligent translation method of relaying human and AI translators according to claim 4, characterized in that, The process involves allocating the semantic fusion vector or dataset of the data to be translated based on the real-time translation matching parameters, combined with the translation priority and translation difficulty level, to obtain the translation allocation data corresponding to human translators and AI translators, including: Based on the real-time translation matching parameters and the feature data of the scene to be translated, available human translators and AI translators are determined. If the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator and the dataset to be translated will be assigned to the human translator. If the priority of the data to be translated is high and the difficulty level of the data to be translated is low, then the semantic fusion vector of the data to be translated will be assigned to the AI ​​translator. If the priority of the data to be translated is low, the semantic fusion vector of the data to be translated will be allocated to the AI ​​translator according to a preset ratio, and the dataset to be translated will be allocated to the human translator.

6. The intelligent translation method of relaying human and AI translators according to claim 5, characterized in that, If the priority of the data to be translated is high and the difficulty level of the data to be translated is high, then the semantic fusion vector of the data to be translated is assigned to an AI translator, and the dataset to be translated is assigned to a human translator, including: Based on the semantic fusion vector of the data to be translated, extract the corresponding scene semantic prompt information and the scene-related term information; The scene semantic prompts, related terminology information, and the dataset to be translated are assigned to human translators.

7. The intelligent translation method of relaying human and AI translators according to claim 6, characterized in that, The process involves intelligent translation based on the assigned data to be translated, obtaining initial translation data corresponding to the AI ​​translator, and analyzing and processing the initial translation data to obtain initial translation quality evaluation parameters corresponding to the AI ​​translator, including: Intelligent translation is performed based on the assigned data to be translated to obtain the initial translation data corresponding to the AI ​​translator; The translation quality is verified by combining the initial translation data with a preset knowledge graph of the domain and scenario to be translated, and the terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are obtained. The terminology domain accuracy, terminology scenario accuracy, and terminology consistency rate are weighted and summed to obtain the initial translation quality evaluation parameters for the AI ​​translator.

8. The intelligent translation method of relaying human and AI translators according to claim 7, characterized in that, The step of comparing the initial translation quality evaluation parameters with a preset initial translation quality evaluation threshold, and determining the real-time translation status of the AI ​​translator based on the threshold comparison result, includes: The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold. If the initial translation quality assessment parameter is greater than or equal to the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be qualified. If the initial translation quality assessment parameter is less than the preset initial translation quality assessment threshold, then the real-time translation status of the AI ​​translator is determined to be unqualified.

9. The intelligent translation method of relaying human and AI translators according to claim 8, characterized in that, Also includes: If the real-time translation status is unqualified, a human translator will make corrections based on the initial translation data to obtain translation correction trace data, including content correction data, format correction data, and human marking data. The initial translation data is replaced with the content correction data, format correction data, and manually marked data to obtain translation correction data; Based on the translation correction trace data, terminology correction data, scene tone adaptation feature data, and scene terminology association feature data are extracted; The preset domain and scene knowledge graph to be translated is updated based on the terminology correction data, scene tone adaptation feature data, and scene term association feature data.

10. An intelligent translation system that combines human and AI translators, characterized in that, The system includes a memory and a processor. The memory contains a program for an intelligent translation method that combines human and AI translators. When the program for this intelligent translation method is executed by the processor, it performs the following steps: Obtain the dataset to be translated, perform multimodal feature extraction and confidence calculation based on the dataset, and obtain the semantic fusion vector of the data to be translated and the corresponding translation priority; Acquire and process translation ability assessment data and real-time translation load of human translators and AI translators to obtain real-time translation matching parameters corresponding to human translators and AI translators; The difficulty level of the data to be translated is obtained by analyzing and processing the dataset to be translated. Based on the real-time translation matching parameters, combined with the translation priority and translation difficulty level, the semantic fusion vector or translation dataset of the data to be translated is allocated to obtain the translation allocation data corresponding to human translators and AI translators; Intelligent translation is performed based on the assigned data to be translated to obtain the initial translation data corresponding to the AI ​​translator. The initial translation data is then analyzed and processed to obtain the initial translation quality evaluation parameters corresponding to the AI ​​translator. The initial translation quality evaluation parameters are compared with the preset initial translation quality evaluation threshold, and the real-time translation status of the AI ​​translator is determined based on the threshold comparison result.

Images