Multi-language translation method and system

Abstract

The invention discloses a multi-language translation method and system, and relates to the field of electrical digital data process.The method comprises the steps that cultural feature information and semantic structure information are obtained; splitting the source language text into a plurality of semantic units based on the semantic structure information; retrieving cultural connotation information of which the association degree with the semantic unit is greater than a preset threshold value; determining that the semantic unit is a culture sensitive semantic unit; extracting a target language expression mode based on the cultural connotation information; performing semantic combination on the target language expression mode and standard translation results of other semantic units except the culture sensitive semantic unit in the source language text to obtain a translation result; judging whether the semantic coherence of the translation result meets a preset condition or not; if the preset condition is met, outputting a translation result; and if the preset condition is not met, re-executing the step of semantic combination. The accuracy of the cultural connotation and the emotional color contained in the transmission source language is improved, and then the cultural adaptability of the translation result is improved.

Description

Technical Field

[0001] The present application belongs to the field of electronic digital data processing, and in particular, relates to a multi-language translation method and system. Background Art

[0002] With the continuous development of economic globalization, cross-language communication has become increasingly frequent, and multilingual translation technology plays an increasingly important role in people's production and life. Traditional manual translation methods are not only time-consuming and labor-intensive, but also inefficient when processing large-scale texts. In addition, they are prone to misunderstandings and inaccurate expressions, and are no longer able to meet today's rapidly developing international needs.

[0003] In related technologies, an encoder-decoder model can be established to encode the source language text into a vector representation, and then decode it to generate the target language text, which has achieved significant improvements in translation speed and accuracy. This technology uses an attention mechanism to dynamically focus on important information in the source text, and uses a deep learning algorithm to continuously optimize the parameters of the translation model, making the translation results more accurate and smooth.

[0004] However, when translating texts containing idioms, proverbs and other texts with profound cultural connotations, the problem of literal translation often occurs, which results in the loss of the deeper meaning of the original text. It is difficult to accurately convey the cultural connotation and emotional color contained in the source language, resulting in a lack of cultural adaptability in the translation results. Summary of the invention

[0005] The present application provides a multilingual translation method and system for improving the accuracy of conveying the cultural connotations and emotional colors contained in the source language, thereby improving the cultural adaptability of the translation results.

[0006] In a first aspect, the present application provides a multilingual translation method, which obtains a source language text to be translated and a target language type; Acquire cultural feature information corresponding to the source language text from a preset cultural knowledge base; acquire semantic structure information of the source language text from a preset source language semantic knowledge base; Split the source language text into several semantic units based on semantic structure information; For each semantic unit, searching the preset cultural knowledge base for cultural connotation information whose correlation with the semantic unit is greater than a preset threshold; In the case where it is determined that the semantic unit contains cultural characteristic identifiers, the semantic unit is determined to be a culturally sensitive semantic unit; Based on the cultural connotation information, the target language expression is extracted from the preset cultural knowledge base, and the cultural connotation matching degree of the cultural sensitive semantic unit is greater than the preset matching degree; Calling the target language semantic knowledge base, and semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic units, to obtain the translation result; Determine whether the semantic coherence of the translation result meets the preset conditions according to the preset translation evaluation rules; If the preset conditions are met, the translation result is output; If the preset condition is not met, the step of semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic units to obtain the translation result is re-executed.

[0007] By adopting the above technical solutions, through the identification and special processing of culturally sensitive content, the translation results can accurately convey the cultural connotations of the source language. The use of semantic unit splitting and semantic coherence-based evaluation mechanism improves the overall coherence of the translation results, and improves the translation quality and expression accuracy in cross-cultural communication. When encountering a situation that does not meet the preset conditions of semantic coherence, the system will find a better solution by recombining, which enhances the reliability of the translation results and the fluency of expression. Improves the accuracy of conveying the cultural connotations and emotional colors contained in the source language, thereby improving the cultural adaptability of the translation results.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, when determining that a semantic unit includes a cultural feature identifier, the method specifically includes: Obtaining a preset cultural feature identification dictionary, the preset cultural feature identification dictionary includes multiple cultural scenes and feature words corresponding to each cultural scene; Perform word segmentation on the semantic units to obtain word sequences; Matching each word in the word sequence with a characteristic word in the cultural characteristic identification dictionary; When the number of matched feature words is greater than a preset number, the cultural scene corresponding to the semantic unit is determined according to the matched feature words; Extract cultural feature rules from the cultural knowledge base based on cultural scenarios; When it is determined according to the cultural feature rule that the semantic structure of the semantic unit meets the constituent conditions of the cultural feature identifier, it is confirmed that the semantic unit contains the cultural feature identifier.

[0009] By adopting the above technical solution, a corresponding relationship is established between cultural scenes and feature words, semantic units are segmented and matched with feature words, and cultural scenes are determined based on the matching results and corresponding cultural feature rules are extracted, and then it is judged whether the semantic unit contains cultural feature identifiers. The accuracy and reliability of cultural feature identification are improved through double inspections of the number threshold control of feature words and verification of cultural feature rules. The dictionary-based feature word management method facilitates the accumulation and updating of knowledge, enables the system to cope with different types of cultural scenes, and improves the coverage and adaptability of cultural feature identification. The verification of semantic structure through cultural feature rules reduces the misjudgment rate and improves the pertinence and accuracy of subsequent translation processing.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic units to obtain the translation result specifically includes: Constructing a target language syntax tree template, the syntax tree template including a plurality of semantic slots and syntax relations between the semantic slots; Acquire the syntactic structure information of the target language expressions from the target language semantic knowledge base; Fill the target language expressions into the corresponding semantic slots according to the syntactic structure information; Obtaining a semantic association calculation model, where the semantic association calculation model is used to calculate the semantic association strength between different semantic units; Calculate the semantic relevance between the standard translation results of other semantic units and the content in the filled semantic slot based on the semantic relevance calculation model; Sorting the standard translation results of other semantic units according to the semantic relevance to obtain a sorting result; According to the order of the sorting results from high to low, the standard translation results of other semantic units are filled into the remaining semantic slots in sequence; Determine the connectives between the semantic slots according to the syntax tree template; The connectives are filled between the semantic slots and converted into a linear text sequence to obtain the translation result.

[0011] By adopting the above technical solution and a structured translation organization method based on syntax tree, the syntactic relationship between the semantic components is clearly defined, reducing the problem of word order confusion that may be caused by simple splicing. Through the quantitative calculation of semantic relevance and the sorting mechanism based on association strength, the rationality and coherence of each semantic unit in the combination process are improved. The application of syntax tree template makes the translation result have a good syntactic structure, improves the standardization and readability of expression, and improves the expression quality of the translation result through structured combination while maintaining the semantic integrity of the original text.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after re-performing the step of semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic units to obtain the translation result, the method further includes: Obtaining a historical translation sample library, the historical translation sample library comprising a plurality of sample pairs of source language texts and corresponding target language translation texts; Calculating the similarity between the source language text and the source language text in the historical translation sample library based on a preset similarity calculation rule; Filtering sample pairs whose similarity is greater than a preset similarity threshold from the historical translation sample library; Extracting semantic combination patterns of target language translation texts in sample pairs; Applying semantic combination patterns to translation results to generate candidate translation results; Compare the semantic coherence scores of the alternative translations with the translation result; The translation result with high semantic coherence score is selected as the optimized translation result.

[0013] By adopting the above technical solutions, the existing high-quality translation experience is used to optimize the translation results and improve the translation efficiency. The sample screening mechanism based on the similarity threshold ensures the relevance and applicability of the reference cases and reduces the negative impact that may be caused by improper references. By extracting and applying semantic combination patterns, successful translation experience is transformed into reusable knowledge, which improves the learning ability and optimization effect of the translation system, and enhances the practicality and reliability of the system while improving the translation quality.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, calculating the similarity between the source language text and the source language text in the historical translation sample library based on a preset similarity calculation rule specifically includes: Extract features from the source language text and generate text feature vectors; Extract features from the source language text in the historical translation sample library to generate sample feature vectors; Calculate the Euclidean distance between the text feature vector and the sample feature vector in each dimension; The similarity is obtained by weighted summing of the Euclidean distances in each dimension.

[0015] By adopting the above technical solution, the source language text and the source language text in the historical translation sample library are feature extracted and feature vectors are generated. The Euclidean distance of the two feature vectors in each dimension is calculated and weighted summed to obtain the similarity. The similarity between the texts can be accurately quantified from a mathematical perspective, and the feature differences of the texts in multiple dimensions such as word frequency, syntactic structure, and semantic association can be captured. Compared with simple text matching, it has higher accuracy. By setting weight coefficients for features of different dimensions, the solution can highlight the influence of important features and weaken the interference of secondary features, so that the similarity calculation results are more in line with actual language usage scenarios.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after selecting a translation result with a high semantic coherence score as the optimized translation result, the method further includes: Evaluate the optimized translation results according to preset automatic evaluation rules to obtain automatic evaluation results; Sending automatic evaluation results to the detection terminal; Receive the review and evaluation feedback from the detection terminal, and modify the optimized translation result according to the review and evaluation feedback to obtain the final translation result.

[0017] By adopting the above technical solution, the optimized translation results are automatically evaluated and sent to the detection terminal for manual review, achieving dual control of machine evaluation and manual evaluation. The automatic evaluation system can quickly detect common errors and unreasonable expressions in the translation results, improving the evaluation efficiency. The manual review stage can discover deep semantic problems and cultural differences that cannot be identified by automatic evaluation, ensure the quality of translation, and enable the translation system to make targeted modifications and improvements to the translation results based on the evaluation feedback, ensuring that the final output translation results are not only accurate and standardized in grammar and wording, but also can accurately convey the cultural connotation and deep meaning of the source language, thereby improving the quality of translation.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the optimized translation result is modified according to the review evaluation feedback to obtain the final translation result, which specifically includes: Obtain modification suggestions in the review and evaluation feedback to obtain a set of modification suggestions; Sort the set of modification suggestions by importance to obtain a priority list; Based on the priority sorting list, the optimized translation result is replaced with the target language expression template to obtain the final translation result.

[0019] By adopting the above technical solution, the modification suggestions in the review and evaluation feedback are ranked by importance, and the target language expression template is used to replace them based on the priority, so that the system can process various modification suggestions in the best order. The application of the target language expression template ensures that the modified expression conforms to the language habits and expression norms of the target language, improves the efficiency of translation modification, and makes the final translation result more natural and fluent while maintaining the meaning of the original text.

[0020] In a second aspect, an embodiment of the present application provides a multilingual translation system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the system to execute the method described in the first aspect and any possible implementation method of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, comprising instructions, which, when executed on a system, causes the system to execute the method described in the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer program product. When the computer program product runs on a system, the system executes the method described in any possible implementation manner in the first aspect.

[0023] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. The present application provides a multilingual translation method, which enables the translation results to accurately convey the cultural connotations of the source language through the identification and special processing of culturally sensitive content. The semantic unit splitting method and the evaluation mechanism based on semantic coherence are adopted to improve the overall coherence of the translation results, and improve the translation quality and expression accuracy in cross-cultural communication. When encountering a situation that does not meet the preset conditions of semantic coherence, the system will find a better solution by recombining, which enhances the reliability and fluency of the translation results. The accuracy of conveying the cultural connotations and emotional colors contained in the source language is improved, thereby improving the cultural adaptability of the translation results.

[0024] 2. This application provides a multilingual translation method that uses existing high-quality translation experience to optimize translation results and improve translation efficiency. The sample screening mechanism based on similarity thresholds ensures the relevance and applicability of reference cases and reduces the negative impact that may be caused by improper references. By extracting and applying semantic combination patterns, successful translation experience is converted into reusable knowledge, which improves the learning ability and optimization effect of the translation system, and enhances the practicality and reliability of the system while improving translation quality.

[0025] 3. The present application provides a multilingual translation method, which automatically evaluates the optimized translation results and sends them to the detection terminal for manual review, thus achieving dual control of machine evaluation and manual evaluation. The automatic evaluation system can quickly detect common errors and unreasonable expressions in the translation results, thereby improving the evaluation efficiency. The manual review stage can discover deep semantic problems and cultural differences that cannot be identified by automatic evaluation, ensure the quality of translation, and enable the translation system to make targeted modifications and improvements to the translation results based on the evaluation feedback, thereby ensuring that the final output translation results are not only accurate and standardized in grammar and wording, but also can accurately convey the cultural connotation and deep meaning of the source language, thereby improving the quality of translation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of a multi-language translation method in an embodiment of the present application.

[0027] Figure 2 It is a flowchart of an optimization method based on historical translation samples in an embodiment of the present application.

[0028] Figure 3 It is a schematic diagram of the physical device structure of a multi-language translation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to be used as limitations to the present application. As used in the specification and appended claims of the present application, the singular expressions "one", "a kind of", "said", "above", "the" and "this" are intended to also include plural expressions, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations comprising one or more listed items.

[0030] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, unless otherwise specified, "plurality" means two or more.

[0031] The following uses an embodiment and combines Figure 1 , a multilingual translation method in an embodiment of the present application is described: See also Figure 1 , which is a flow chart of a multi-language translation method in an embodiment of the present application.

[0032] S101, obtaining a source language text to be translated and a target language type; The system first needs to obtain the source language text to be translated that is input or uploaded by the user. The source language text can be provided to the system by the user through keyboard input, voice recognition conversion, image recognition conversion, document upload, etc. At the same time, the system also needs to obtain the target language type selected or set by the user. The target language type can be selected by the user in the list of alternative languages ​​provided by the system through a drop-down menu, radio button, etc., or it can be the language name or code directly entered by the user.

[0033] The system can implement this step by constructing an input interface, which includes a text input box, a voice input button, an image upload button, a document upload button, etc. The user can provide the source language text to the system through these interfaces. At the same time, the input interface also includes a language selection drop-down menu or radio button, and the user can select the target language type through these controls. After obtaining the source language text and the target language type, the system passes them to the subsequent processing module.

[0034] S102, obtaining cultural feature information corresponding to the source language text from a preset cultural knowledge base; obtaining semantic structure information of the source language text from a preset source language semantic knowledge base; The system needs to obtain information related to the source language text from two preset knowledge bases. First, the system needs to obtain the cultural feature information corresponding to the source language text from the preset cultural knowledge base. The cultural feature information may include the cultural background, customs, traditions, values, etc. involved in the source language text. Secondly, the system needs to obtain the semantic structure information of the source language text from the preset source language semantic knowledge base. The semantic structure information may include the part of speech, syntactic structure, semantic role, etc. of the source language text.

[0035] The system can implement this step by building a knowledge base query interface. The knowledge base query interface receives the source language text as input, then searches in the preset cultural knowledge base and source language semantic knowledge base, and returns the cultural feature information and semantic structure information found. The knowledge base can use a variety of forms such as relational databases and graph databases to store and organize data, and the query interface can be implemented using a variety of query languages ​​such as SQL and SPARQL.

[0036] S103, splitting the source language text into a plurality of semantic units based on the semantic structure information; The system needs to split the source language text into several independent semantic units based on the semantic structure information obtained from the source language semantic knowledge base. Semantic units can be language fragments of different lengths and complexities, such as words, phrases, and clauses. Each semantic unit can express a relatively complete meaning. The purpose of splitting semantic units is to perform more fine-grained processing and translation in subsequent steps.

[0037] The system can achieve this step by building a semantic analysis model. The semantic analysis model receives the source language text and its corresponding semantic structure information as input, and then divides the text into several semantic units based on information such as part of speech and syntactic roles in the semantic structure. Common semantic analysis models include rule-based methods and statistical-based methods. The former divides semantic units by manually defining rules, while the latter automatically learns the division rules of semantic units from a large-scale corpus through machine learning algorithms.

[0038] S104, searching for cultural connotation information with a correlation degree greater than a preset threshold value for each semantic unit in a preset cultural knowledge base; In this step, the system needs to search the preset cultural knowledge base for each semantic unit to find cultural connotation information with a high correlation with the semantic unit. Cultural connotation information can include cultural background, customs, traditions, values, etc. related to the semantic unit. By comparing the correlation between the semantic unit and the cultural connotation information, the system screens out the cultural connotation information with a correlation greater than a preset threshold as the cultural feature of the semantic unit.

[0039] The system can implement this step by building a semantic matching model. The semantic matching model receives semantic units and candidate cultural connotation information as input, and obtains the correlation between the two by calculating their similarity or distance in the semantic space. Common semantic matching methods include methods based on vector space models (such as cosine similarity) and methods based on deep learning (such as twin networks). The system can set a correlation threshold to retain cultural connotation information with a correlation greater than the threshold.

[0040] In this step, if the preset cultural knowledge base is not fully covered or the semantic matching model is not accurate enough, the retrieved cultural connotation information may be inconsistent with the actual cultural characteristics of the semantic unit. To solve this problem, the system can use transfer learning to expand the cultural knowledge base using knowledge from other fields (such as common sense and encyclopedias), and fine-tune the semantic matching model with a small amount of annotated data to improve the accuracy and comprehensiveness of cultural connotation information retrieval.

[0041] S105, when it is determined that the semantic unit contains a cultural feature identifier, determining the semantic unit as a culturally sensitive semantic unit; As for how the system determines that a semantic unit includes a cultural feature identifier, the steps are as follows: obtaining a preset cultural feature identifier dictionary, the preset cultural feature identifier dictionary includes multiple cultural scenes and feature words corresponding to each cultural scene; Perform word segmentation on the semantic units to obtain word sequences; Matching each word in the word sequence with a characteristic word in the cultural characteristic identification dictionary; When the number of matched feature words is greater than a preset number, the cultural scene corresponding to the semantic unit is determined according to the matched feature words; Extract cultural feature rules from the cultural knowledge base based on cultural scenarios; When it is determined according to the cultural feature rule that the semantic structure of the semantic unit meets the constituent conditions of the cultural feature identifier, it is confirmed that the semantic unit contains the cultural feature identifier. Then the system determines that the semantic unit is a culturally sensitive semantic unit.

[0042] In this step, the system needs to determine whether the semantic unit is a culturally sensitive semantic unit based on whether it contains cultural feature identifiers. Cultural feature identifiers can be keywords, idioms, and phrases that reflect specific cultural connotations. If a semantic unit contains a cultural feature identifier corresponding to the cultural connotation information that matches it, it can be considered that the semantic unit is sensitive to this cultural connotation and needs to be specially processed during translation.

[0043] As for how to determine whether a semantic unit contains a cultural feature identifier, the system can adopt the following steps: First, the system needs to obtain a preset cultural feature identifier dictionary, which includes various cultural scenes and their corresponding feature words. Then, the system performs word segmentation on the semantic unit to obtain a word sequence. Next, the system matches each word in the word sequence with the feature words in the cultural feature identifier dictionary. If the number of matched feature words exceeds a certain threshold, the cultural scene corresponding to the semantic unit can be determined based on these feature words. Finally, the system extracts the feature rules of the cultural scene from the cultural knowledge base, and determines whether the semantic structure of the semantic unit meets the composition conditions of the cultural feature identifier based on the feature rules. If so, the semantic unit is confirmed to be a culturally sensitive semantic unit.

[0044] S106, extracting a target language expression whose cultural connotation matching degree of the cultural sensitive semantic unit is greater than a preset matching degree from a preset cultural knowledge base based on the cultural connotation information; In this step, the system needs to find the target language expression that matches the cultural connotation of each culturally sensitive semantic unit from the preset cultural knowledge base. The cultural connotation matching degree indicates the degree of similarity between the target language expression and the source language culturally sensitive semantic unit in terms of semantics, emotion, rhetoric, etc. The higher the matching degree, the more accurately the expression can convey the cultural connotation contained in the source language semantic unit. The system sets a matching degree threshold to screen all target language expressions in the knowledge base and extract expressions with a matching degree greater than the threshold.

[0045] The system can achieve this step by building a cross-language cultural connotation matching model. The model takes the source language culturally sensitive semantic units and the target language candidate expressions as input, performs cross-language semantic encoding on them through a deep learning network, and calculates their similarity in the semantic space to obtain the cultural connotation matching degree. The training data of the model can come from a manually translated corpus or can be automatically mined from a large-scale bilingual corpus in an unsupervised manner.

[0046] S107, calling the target language semantic knowledge base, and semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic units, to obtain a translation result; The system calls the target language semantic knowledge base, and semantically combines the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic units to obtain the translation result, which specifically includes: constructing a target language syntax tree template, the syntax tree template includes multiple semantic slots and syntactic relations between the semantic slots; obtaining the syntactic structure information of the target language expression from the target language semantic knowledge base; filling the target language expression into the corresponding semantic slot according to the syntactic structure information; obtaining a semantic association calculation model, the semantic association calculation model is used to calculate the semantic association strength between different semantic units; calculating the semantic association between the standard translation results of other semantic units and the content in the filled semantic slots based on the semantic association calculation model; sorting the standard translation results of other semantic units according to the semantic association to obtain the sorting result; filling the standard translation results of other semantic units into the remaining semantic slots in order from high to low in the sorting result; determining the connecting words between each semantic slot according to the syntax tree template; filling the connecting words between each semantic slot and converting them into a linear text sequence to obtain the translation result.

[0047] In this step, the system needs to semantically combine the target language expression of the culturally sensitive semantic unit with the standard translation results of other semantic units in the source language text to obtain a complete target language translation result. The standard translation result can be obtained through a conventional machine translation system. In the semantic combination process, the system needs to call the target language semantic knowledge base, use the grammatical and semantic rules therein, reorder, insert, replace, and add appropriate conjunctions to the translation results of different semantic units, and finally form a semantically coherent and well-structured target language text.

[0048] The system can implement this step by building a semantic combination model based on a syntax tree. The model first builds a target language syntax tree template, which contains several semantic slots and syntactic dependencies between the slots. Then, the model fills the target language expression of the culturally sensitive semantic unit into the corresponding semantic slot. Next, based on the results of the semantic relevance calculation, the model fills the standard translation results of other semantic units into the remaining semantic slots in descending order. Finally, according to the structure of the syntax tree, the model adds conjunctions between the slots and linearizes the slot content into the target language text. The semantic relevance can be calculated by a pre-trained language model, or end-to-end relevance learning can be performed directly on the syntax tree through methods such as graph neural networks.

[0049] In this step, if the rules of the target language semantic knowledge base are not fully covered, or the semantic combination model cannot coordinate the translation results of different semantic units well, the translation results may be incoherent and chaotic in structure. To solve these problems, the system can introduce a large-scale target language corpus to expand the rules of the semantic knowledge base in an unsupervised or semi-supervised manner; in addition, the system can also add some flexibility and fault tolerance to the syntactic tree template, allowing local adjustments to the syntactic structure to meet the needs of translation results of different semantic units; at the same time, the system can combine reinforcement learning and other technologies to guide the semantic combination model to generate more fluent and natural translation results by setting a suitable reward function.

[0050] S108, judging whether the semantic coherence of the translation result meets the preset conditions according to the preset translation evaluation rules; This step needs to evaluate the quality of the translation result outputted in step S107 to determine whether it meets the system preset conditions to ensure the readability and comprehensibility of the translation. The translation evaluation rules may include indicators of multiple aspects such as grammatical correctness, semantic integrity, logical consistency, and appropriateness of language style, and each indicator is provided with corresponding judgment criteria and thresholds. Only when the translation result meets the standards in all indicators can it be judged to meet the preset conditions. The preset conditions refer to a series of measurement criteria and thresholds used by the system when judging whether the translation result meets the requirements of semantic coherence. These preset conditions are set by the system according to specific task requirements and domain knowledge before applying the translation method, as the basis for evaluating the quality of the translation. Specifically, the preset conditions may include the following aspects: Grammatical correctness condition: The translation results must conform to the grammatical rules of the target language, such as word order, sentence pattern, tense, and number agreement, to avoid grammatical errors or unreasonable expressions. The system can set specific grammatical error types and quantity thresholds. Translation results that exceed the threshold will be judged as not meeting the grammatical correctness condition.

[0051] Semantic completeness condition: The translation result needs to fully convey the semantic information of the source language text, and there must be no omissions, mistranslations, or the addition of irrelevant information. The system can set quantitative indicators and thresholds for semantic similarity or coverage to determine whether the translation result meets the semantic completeness condition.

[0052] Logical consistency condition: The translation results need to be consistent in terms of logic and facts, avoiding self-contradictions or expressions that are inconsistent with objective facts. The system can set some common sense rules and reasoning rules to check the logical consistency of the translation results and set corresponding threshold conditions.

[0053] Appropriateness of language: The style and tone of the translation result must match the source language text and conform to the expression habits of a specific register and communication scenario. The system can predefine the language features of different registers and scenarios and set similarity thresholds to determine whether the translation result meets the appropriateness of language.

[0054] Translation fluency condition: The translation result needs to conform to the expression habits of the target language, read smoothly and naturally, without stiffness, verbosity, repetition, etc. The system can establish a language model, calculate the naturalness score of the translation result, and set a fluency score threshold to determine whether it meets the translation fluency condition.

[0055] The system can achieve this step by building a multi-indicator translation quality assessment model. The model takes the translation results as input and judges and scores them from different perspectives. In terms of grammatical correctness, the model can use rule-based or statistical grammar checkers to identify grammatical errors in the translation results; in terms of semantic completeness, the model can use pre-trained semantic representation models to determine whether the translation results fully convey the semantic information of the source language text; in terms of logical consistency, the model can use common sense reasoning, reference resolution and other techniques to check whether the translation results are inconsistent at the logical and factual levels; in terms of language appropriateness, the model can combine domain classification and sentiment analysis to determine whether the style and tone of the translation results match the source language text. Finally, the model weights and sums the scores of each indicator and compares them with the preset threshold to obtain the final evaluation result.

[0056] S109: Output the translation result.

[0057] If the preset conditions are met, the translation result is output; if the preset conditions are not met, step S107 is executed again.

[0058] This step mainly realizes outputting the translation result that has been evaluated in step S108 to the user. If the evaluation result meets the preset semantic coherence condition, the system will directly present the translation result to the user; if the evaluation result does not meet the standard, the system needs to return to step S107 to adjust and optimize the result of the semantic combination, and then perform quality evaluation again until a translation result that meets the requirements is generated.

[0059] The output form of the translation results can be customized according to the user's needs and usage scenarios. The system can support multiple output methods such as text, voice, and image, which is convenient for users to obtain translation results in different devices and environments. While outputting the translation results, the system can also provide some auxiliary information, such as the original content, keyword explanation, cultural background introduction, etc., to help users better understand and use the translation results. In addition, the system can provide users with interactive functions, allowing users to evaluate and provide feedback on the translation results, and continuously optimize the system's translation strategy.

[0060] In the above embodiment, by identifying and specially processing culturally sensitive content, the translation result can accurately convey the cultural connotation in the source language. The semantic unit splitting method and the evaluation mechanism based on semantic coherence are adopted to improve the overall coherence of the translation result, and improve the translation quality and expression accuracy in cross-cultural communication. When encountering a situation that does not meet the preset conditions of semantic coherence, the system will find a better solution by recombining, which enhances the reliability and fluency of the translation result. Improve the accuracy of conveying the cultural connotation and emotional color contained in the source language, thereby improving the cultural adaptability of the translation result.

[0061] After completing the basic translation process, in order to further improve the translation quality, this application also provides an optimization method based on historical translation samples. By similarity matching and extraction of semantic combination patterns, more references and optimization space are provided for the current translation task. Figure 2 , an optimization method based on historical translation samples in an embodiment of the present application is described: See also Figure 2 , which is a flow chart of an optimization method based on historical translation samples in an embodiment of the present application.

[0062] S201, obtaining a historical translation sample library; The system obtains a historical translation sample library, which includes a plurality of sample pairs of source language texts and corresponding target language translation texts.

[0063] In this step, the system needs to obtain a historical translation sample library, which contains a large number of source language texts and corresponding target language translation texts. These text pairs constitute a collection of translation samples. The historical translation sample library can come from multiple channels, such as manual translation corpora, parallel corpora generated by machine translation systems, translation memory libraries, etc. The scale and quality of the sample library have an important impact on the subsequent optimization of translation results.

[0064] The system can obtain and build a historical translation sample library in a variety of ways. One common way is to extract samples from existing bilingual or multilingual parallel corpora, which can be public data sets or corpora accumulated within the system. Another way is to use crawler technology to crawl a large number of bilingual web pages from the Internet, and form a sample library after alignment and filtering. In addition, the system can also access the manual translation interface, allowing users to upload and manage their own translation samples, and continuously enrich and improve the sample library.

[0065] S202, calculating the similarity between the source language text and the source language text in the historical translation sample library based on a preset similarity calculation rule; The system calculates the similarity between the source language text and the source language text in the historical translation sample library based on the preset similarity calculation rules, specifically including: extracting features from the source language text to generate a text feature vector; extracting features from the source language text in the historical translation sample library to generate a sample feature vector; calculating the Euclidean distance between the text feature vector and the sample feature vector in each dimension; and performing weighted summation of the Euclidean distances in each dimension to obtain the similarity.

[0066] In this step, the system needs to calculate the similarity between the current source language text to be translated and all source language texts in the historical translation sample library as the basis for subsequent screening of similar samples. The similarity calculation rules can be based on a variety of text representation methods and distance metrics, such as vector space model, edit distance, etc. Different calculation rules are suitable for different text characteristics and task requirements.

[0067] Specifically, the system can use the following steps to calculate the similarity between the source language text and the source language text in the historical translation sample library: first, perform feature extraction on the current source language text and convert it into a text feature vector of fixed dimension; then, perform the same feature extraction on each source language text in the historical translation sample library to obtain a series of sample feature vectors; next, the system calculates the Euclidean distance between the text feature vector and each sample feature vector in each dimension to obtain a distance vector; finally, the system performs weighted summation on each component in the distance vector to obtain the similarity between the source language text and the sample text.

[0068] S203, selecting sample pairs whose similarity is greater than a preset similarity threshold from the historical translation sample library; In this step, the system needs to screen out translation sample pairs with high similarity based on the similarity between the source language text and the source language text in the historical translation sample library, providing a reference for the subsequent semantic combination pattern extraction. The setting of the similarity threshold needs to balance the quantity and quality of samples, ensuring that enough similar samples are screened out while avoiding the introduction of too many irrelevant noise samples.

[0069] The system can use sorting and truncation to screen similar samples. Specifically, the system first sorts all samples in descending order according to the similarity score, and then selects the first N samples with scores greater than or equal to the threshold as similar sample subsets based on the preset similarity threshold. The size of the threshold can be set according to the size and distribution characteristics of the sample library, and can also be automatically optimized through methods such as cross-validation. In addition, the system can further filter the screened similar samples, such as removing duplicate samples and low-quality samples, to improve the purity of similar samples.

[0070] S204, extracting semantic combination patterns of target language translation texts in the sample pairs; In this step, the system needs to extract the semantic combination patterns contained in the target language translation text from the selected similar translation sample pairs as a reference for optimizing the current translation results. The semantic combination pattern reflects the idiomatic structure and expression of the target language when expressing specific semantics, which can help the system generate translation results that are more in line with the habits of the target language.

[0071] The system can extract semantic combination patterns through the following methods: first, perform syntactic analysis on the target language translation text to obtain a syntactic dependency tree; then, identify important components in the syntactic tree, such as subject, predicate, object, etc., and extract the semantic relations between these components; next, the system abstracts the semantic relations into semantic combination templates, which contain information such as the type of semantic components, semantic roles, and combination order; finally, the system summarizes and generalizes the semantic combination templates extracted from multiple sample texts to obtain some common semantic combination patterns.

[0072] S205, applying the semantic combination pattern to the translation result to generate an alternative translation result; In this step, the system needs to apply the semantic combination patterns extracted from similar samples to the current translation results to generate one or more alternative translation results. The alternative translation results are semantically equivalent to the original translation results, but the expression method may be more in line with the idiomatic usage of the target language. By generating a variety of alternative translation results, the system can provide users with more choices and provide greater optimization space for subsequent translation optimization.

[0073] Specifically, the system can apply the semantic combination pattern to the translation results by taking the following steps: first, perform semantic analysis on the original translation results to extract key semantic components, such as terms, named entities, semantic roles, etc.; then, retrieve combination templates matching these semantic components from the semantic combination pattern library, and fill the semantic components into the corresponding positions of the template according to the semantic relations and combination order in the template to form a new semantic combination structure; finally, the system generates one or more alternative translation results based on the new semantic combination structure, which use different expressions while keeping the original semantics unchanged.

[0074] S206, comparing the semantic coherence scores of the candidate translation results and the translation result; In this step, the system needs to evaluate the semantic coherence of the alternative translation results and the original translation results, and calculate their respective coherence scores as the basis for optimizing the translation results. Semantic coherence reflects the consistency and continuity of the translation results in terms of logic, semantics, pragmatics, etc., and is one of the important indicators for measuring translation quality. The higher the coherence score, the better the readability and comprehensibility of the translation results.

[0075] The system can use a variety of methods to calculate the semantic coherence score. Common methods include language model-based methods and discourse analysis-based methods. The language model-based method establishes an n-gram language model of the target language and calculates the probability or perplexity of the translation result under the language model. The greater the probability or the smaller the perplexity, the more the translation result conforms to the usage habits of the target language and the better the coherence. The discourse analysis-based method calculates the integrity and rationality of the discourse relationship by identifying the discourse relationship in the translation result, such as reference, connection, substitution, etc. The closer and more reasonable the discourse relationship is, the better the coherence of the translation result is.

[0076] In the process of semantic coherence evaluation, the training corpus of the language model may be insufficient or mismatched with the target domain, resulting in inaccurate coherence scores; the discourse analysis method may incorrectly identify the discourse structure and semantic roles of the translation results, affecting the reliability of coherence calculation; the coherence score is inconsistent with the manual evaluation results, and it is difficult to serve as a direct basis for translation optimization. In order to solve these problems, the system can use domain adaptation technology to dynamically adjust the parameters of the language model according to the characteristics of the target domain to improve the domain adaptability of the language model; at the same time, the system can introduce a discourse analysis model based on deep learning to improve the accuracy of discourse relationship recognition and semantic role labeling through end-to-end joint learning; in addition, the system can also combine coherence scores with other translation quality indicators through multi-objective learning, reinforcement learning and other methods to learn a comprehensive translation optimization model, so that the optimized translation results are improved in multiple dimensions.

[0077] S207, selecting the translation result with a high semantic coherence score as the optimized translation result.

[0078] In this step, the system needs to select the one with the highest score as the final optimization result based on the semantic coherence scores of the alternative translation results and the original translation results. The optimized translation results ensure semantic correctness while making the language expression more fluent and natural, and more in line with the expression habits of the target language.

[0079] Specifically, the system can sort the original translation results and all alternative translation results according to the semantic coherence score to obtain an ordered candidate set; then, the system selects the result with the highest score from the candidate set as the preliminary result; next, the system conducts further checks and verifications on the preliminary results, such as grammar checks, terminology consistency checks, etc., to ensure that there are no obvious errors or unreasonableness; finally, the system outputs the results that have passed the inspection as the optimized translation results, and updates the translation memory to provide a reference for subsequent translation tasks.

[0080] In the above embodiment, the existing high-quality translation experience is used to optimize the translation results, thereby improving the translation efficiency. The sample screening mechanism based on the similarity threshold ensures the relevance and applicability of the reference cases, and reduces the negative impact that may be caused by improper references. By extracting and applying semantic combination patterns, successful translation experience is converted into reusable knowledge, which improves the learning ability and optimization effect of the translation system, and enhances the practicality and reliability of the system while improving the translation quality.

[0081] Furthermore, after selecting the translation result with a high semantic coherence score as the optimized translation result, the system can also evaluate the optimized translation result according to preset automatic evaluation rules to obtain an automatic evaluation result; send the automatic evaluation result to the detection terminal; receive the review evaluation feedback from the detection terminal, and modify the optimized translation result according to the review evaluation feedback to obtain the final translation result, which specifically includes: obtaining the modification suggestions in the review evaluation feedback to obtain a modification suggestion set; sorting the modification suggestion set by importance to obtain a priority sorting list; based on the priority sorting list, replacing the optimized translation result with the target language expression template to obtain the final translation result.

[0082] The system can automatically evaluate the optimized translation results according to preset automatic evaluation rules. The automatic evaluation rules may include grammar checking, terminology consistency checking, semantic integrity checking, etc. The system generates comprehensive automatic evaluation results by applying these rules; then, the system sends the automatic evaluation results to the detection terminal, and the manual evaluators review the automatic evaluation results. The manual evaluators confirm, correct, and supplement the automatic evaluation results based on their professional knowledge and experience, form review evaluation feedback, and send the feedback back to the system; finally, the system receives the review evaluation feedback returned by the detection terminal, and modifies the optimized translation results according to the modification suggestions in the feedback to obtain the final translation results. During the modification process, the system first extracts all modification suggestions from the review and evaluation feedback to form a set of modification suggestions; then, the system evaluates the importance of each suggestion in the modification suggestion set according to the preset importance evaluation rules to obtain the importance score of each suggestion; next, the system sorts the modification suggestion set according to the importance score to obtain a priority list arranged from high to low importance; finally, the system traverses the priority list and uses the expression template of the target language to apply the modification suggestions in the list to the optimized translation result in turn, and finally obtains the modified final translation result.

[0083] In the above embodiment, the technical solution of automatically evaluating the optimized translation results and sending them to the detection terminal for manual review realizes the dual control of machine evaluation and manual evaluation. The automatic evaluation system can quickly detect common errors and unreasonable expressions in the translation results, improving the evaluation efficiency. The manual review link can discover deep semantic problems and cultural differences that cannot be identified by automatic evaluation, ensure the quality of translation, and enable the translation system to make targeted modifications and improvements to the translation results based on the evaluation feedback, ensuring that the final output translation results are not only accurate and standardized in grammar and wording, but also can accurately convey the cultural connotation and deep meaning of the source language, thereby improving the quality of translation.

[0084] The following describes the system in the embodiment of the present invention from the perspective of hardware processing. Figure 3 , which is a schematic diagram of the physical device structure of a multilingual translation system provided in an embodiment of the present application.

[0085] It should be noted that Figure 3 The structure of the system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0086] like Figure 3 As shown, the system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage part 308 to the random access memory (RAM) 303, such as executing the method in the above embodiment. In RAM 303, various programs and data required for system operation are also stored. CPU 301, ROM 302 and RAM 303 are connected to each other through a bus 304. Input / output (I / O) interface 305 is also connected to bus 304.

[0087] The following components are connected to the I / O interface 305: an input section 306 including a camera, an infrared sensor, etc.; an output section 307 including a liquid crystal display (LCD) and a speaker, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed so that a computer program read therefrom is installed into the storage section 308 as needed.

[0088] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 309, and / or installed from a removable medium 311. When the computer program is executed by the central processing unit (CPU) 301, various functions defined in the present invention are performed.

[0089] It should be noted that the computer-readable medium shown in the embodiment of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable computer program. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.

[0090] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Among them, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0091] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the system described in the above embodiment; or may exist independently without being assembled into the system. The above storage medium carries one or more computer programs, and when the above one or more computer programs are executed by a processor of a system, the system implements the method provided in the above embodiment.

[0092] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0093] As used in the above embodiments, the term "when..." may be interpreted as "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. Similarly, the phrases "upon determining..." or "if (the stated condition or event) is detected" may be interpreted as "if determining..." or "in response to determining..." or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)", depending on the context.

[0094] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk), etc.

[0095] Those skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by computer programs to instruct related hardware, and the programs can be stored in computer-readable storage media. When the programs are executed, they can include the processes of the above-mentioned method embodiments. The aforementioned storage media include: ROM or random access memory RAM, magnetic disk or optical disk and other media that can store program codes.

Claims

1. A multilingual translation method, characterized in that: include: Get the source language text and target language type to be translated; Acquiring cultural feature information corresponding to the source language text from a preset cultural knowledge base; Acquiring semantic structure information of the source language text from a preset source language semantic knowledge base; Splitting the source language text into a plurality of semantic units based on the semantic structure information; For each of the semantic units, cultural connotation information having a correlation degree with the semantic unit greater than a preset threshold is retrieved from the preset cultural knowledge base; In the case where it is determined that the semantic unit contains a cultural feature identifier, determining that the semantic unit is a culturally sensitive semantic unit; Extracting a target language expression having a cultural connotation matching degree greater than a preset matching degree of the cultural connotation of the culturally sensitive semantic unit from the preset cultural knowledge base based on the cultural connotation information; Calling a target language semantic knowledge base, and semantically combining the target language expression with standard translation results of other semantic units in the source language text except the culturally sensitive semantic unit, to obtain a translation result; Determining whether the semantic coherence of the translation result meets a preset condition according to a preset translation evaluation rule; If the preset condition is met, outputting the translation result; If the preset condition is not met, the step of semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic unit to obtain the translation result is performed again.

2. The method according to claim 1, characterized in that In the case where it is determined that the semantic unit contains a cultural feature identifier, the method specifically includes: Obtaining a preset cultural feature identification dictionary, wherein the preset cultural feature identification dictionary includes a plurality of cultural scenes and a feature word corresponding to each of the cultural scenes; Performing word segmentation processing on the semantic unit to obtain a word sequence; Matching each word in the word sequence with a characteristic word in the cultural characteristic identification dictionary; When the number of matched feature words is greater than a preset number, determining the cultural scene corresponding to the semantic unit according to the matched feature words; Extracting cultural feature rules from the cultural knowledge base based on the cultural scenario; When it is determined according to the cultural feature rule that the semantic structure of the semantic unit satisfies the constituent conditions of the cultural feature identifier, it is confirmed that the semantic unit includes the cultural feature identifier.

3. The method according to claim 1, characterized in that The semantic combination of the target language expression and the standard translation results of other semantic units in the source language text except the culturally sensitive semantic unit to obtain the translation result specifically includes: Constructing a target language syntax tree template, wherein the syntax tree template includes a plurality of semantic slots and syntax relations between the semantic slots; Acquiring syntactic structure information of the target language expression from the target language semantic knowledge base; Filling the target language expression into the corresponding semantic slot according to the syntactic structure information; Acquire a semantic relevance calculation model, wherein the semantic relevance calculation model is used to calculate the semantic relevance strength between different semantic units; Calculating the semantic relevance between the standard translation results of the other semantic units and the content in the filled semantic slots based on the semantic relevance calculation model; Sorting the standard translation results of the other semantic units according to the semantic association to obtain a sorting result; According to the order of the sorting results from high to low, the standard translation results of the other semantic units are sequentially filled into the remaining semantic slots; Determining the connectives between the semantic slots according to the syntax tree template; The connecting words are filled between the semantic slots and converted into a linear text sequence to obtain a translation result.

4. The method according to claim 1, characterized in that After the step of re-performing the step of semantically combining the target language expression with the standard translation results of other semantic units in the source language text except the culturally sensitive semantic unit to obtain the translation result, the method further includes: Acquire a historical translation sample library, wherein the historical translation sample library includes a plurality of sample pairs of the source language text and the corresponding target language translation text; Calculating the similarity between the source language text and the source language text in the historical translation sample library based on a preset similarity calculation rule; Filtering out sample pairs whose similarity is greater than a preset similarity threshold from the historical translation sample library; Extracting semantic combination patterns of target language translation texts in the sample pairs; Applying the semantic combination pattern to the translation result to generate an alternative translation result; comparing the semantic coherence scores of the candidate translation results and the translation result; The translation result with a high semantic coherence score is selected as the optimized translation result.

5. The method according to claim 4, characterized in that The calculating the similarity between the source language text and the source language text in the historical translation sample library based on the preset similarity calculation rule specifically includes: Extracting features from the source language text to generate a text feature vector; Extracting features from the source language text in the historical translation sample library to generate a sample feature vector; Calculating the Euclidean distance between the text feature vector and the sample feature vector in each dimension; The Euclidean distances on each dimension are weighted and summed to obtain the similarity.

6. The method according to claim 4, characterized in that After selecting the translation result with a high semantic coherence score as the optimized translation result, the method further includes: Evaluate the optimized translation result according to preset automatic evaluation rules to obtain an automatic evaluation result; Sending the automatic evaluation result to a detection terminal; The review and evaluation feedback from the detection terminal is received, and the optimized translation result is modified according to the review and evaluation feedback to obtain a final translation result.

7. The method according to claim 6, characterized in that The step of modifying the optimized translation result according to the review and evaluation feedback to obtain the final translation result specifically includes: Obtain modification suggestions in the review and evaluation feedback to obtain a set of modification suggestions; Sorting the set of modification suggestions according to importance to obtain a priority sorting list; Based on the priority sorting list, the optimized translation result is replaced with a target language expression template to obtain a final translation result.

8. A multilingual translation system, characterized in that: The system comprises: One or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the system to execute the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on a system, the system is caused to execute the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is run on a system, the system is caused to execute the method according to any one of claims 1 to 7.

Images