Intelligent processing system for text translation and semantic optimization

Abstract

The invention provides an intelligent processing system for text translation and semantic optimization, and relates to the field of electric digital data processing, the intelligent processing system comprises an information interaction module, a text translation module, a semantic understanding module and a semantic optimization module, the information interaction module is used for inputting a source text and outputting a translation result, the text translation module is used for performing language conversion on a source text to obtain a target text, the semantic understanding module is used for constructing a semantic structure of the target text, and the semantic optimization module is used for performing retouching optimization on the target text based on the semantic structure; according to the system, multiple structured scoring mechanisms are introduced, so that translation and optimization cooperative control is realized, and the accuracy, naturalness and controllability of translated texts are improved.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, and specifically to an intelligent processing system for text translation and semantic optimization. Background Technology

[0002] With the increasing demand for multilingual communication, AI-based text translation technology is widely used in cross-language communication, international business, and content generation. Most existing translation systems employ end-to-end neural network models, capable of basic language conversion, but they still have many limitations. First, existing translation models typically focus only on sentence-level accuracy, lacking a deep understanding of the contextual semantic structure. This leads to semantic shifts or unnatural expressions when handling long texts, polysemous words, or technical terms. Second, translation results often lack controllability and subsequent optimization mechanisms, making it difficult for users to fine-tune language style or structure according to application scenarios. While some systems support manual polishing or post-processing, they lack automated, structured optimization modules, failing to achieve collaborative work between semantic consistency verification and language expression optimization. Therefore, there is an urgent need for an intelligent translation system with semantic understanding capabilities, controllable optimization mechanisms, and structured scoring methods to improve translation accuracy, naturalness of expression, and semantic fidelity, meeting the practical needs of multilingual intelligent processing.

[0003] The following three algorithms are commonly used for translation in existing technologies;

[0004] Transformer neural translation algorithms based on attention mechanisms: These models model the input sequence through multi-head attention, capturing the dependencies between words to achieve high-quality language translation. Their advantages include flexible structure and high parallel computation efficiency, but their disadvantages include context processing typically limited to a single sentence, weak support for cross-sentence dependencies and discourse consistency, and a lack of explicit modeling of semantic structure.

[0005] Post-editing algorithms based on grammar rules: These methods typically modify neural translation results by introducing language rules and syntactic templates, such as using subject-verb-object structure recognizers and phrase template libraries to optimize sentence structure. Their advantage is that the output results are relatively standardized and suitable for formal scenarios; however, they rely on pre-set rules, have poor scalability, offer limited improvement in semantic accuracy, and are prone to rigid application of rules.

[0006] Candidate selection optimization algorithms based on language model scoring: These algorithms generate multiple translation candidate sentences and use language models to score and rank them, selecting the version with the best semantics or the most natural language. The advantage is that they can adapt to diverse language styles, but their semantic judgment is mostly a black-box process, lacking a clear semantic alignment mechanism, and their ability to detect and correct errors is insufficient, failing to guarantee semantic fidelity.

[0007] In summary, existing algorithms either prioritize model performance while neglecting semantic structure control, or rely on rule-based systems, resulting in insufficient flexibility and versatility. Therefore, there is an urgent need for an intelligent processing system that simultaneously possesses deep semantic understanding, automatic optimization control capabilities, and an interpretable scoring mechanism to achieve higher-quality translation optimization results.

[0008] Many translation systems have been developed. Extensive research and reference have revealed existing systems such as the one disclosed in publication number CN119849514B. These systems generally involve acquiring source language text, segmenting the source language text into sentences to obtain a sequence of source language sentences, segmenting each sentence in the source language sentence sequence into words to obtain a sequence of words for each source language sentence, and classifying the sentiment of each source language sentence in the source language sentence sequence based on a pre-trained first sentiment classification model. However, this system is overly mechanical in the translation process, and the translation effect can only simply restore the meaning, failing to achieve an elegant effect. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings by proposing an intelligent processing system for text translation and semantic optimization.

[0010] The present invention adopts the following technical solution:

[0011] An intelligent processing system for text translation and semantic optimization includes an information interaction module, a text translation module, a semantic understanding module, and a semantic optimization module;

[0012] The information interaction module is used to input source text and output translation results; the text translation module is used to convert the source text into a language to obtain the target text; the semantic understanding module is used to construct the semantic structure of the target text; and the semantic optimization module is used to refine and optimize the target text based on the semantic structure.

[0013] The system's modules are interconnected, forming a closed processing chain from input to output. The information interaction module serves as the input and display end, enabling user interaction with the system; the text translation module provides cross-language semantic conversion capabilities and is the central hub of the processing flow; the semantic understanding module undertakes the task of semantic structure parsing, providing support for downstream optimization modules; and the semantic optimization module refines language and reconstructs expressions based on semantic structure, ultimately returning the results to the information interaction module for display.

[0014] This modularization scheme effectively decouples system functions, allowing each module to focus on a single responsibility, thus improving the system's maintainability and scalability. At the same time, modules interact with each other through data forms such as vector encoding results, structured semantic graphs, and optimization candidate sets, ensuring smooth information flow, high processing efficiency, and facilitating subsequent distributed deployment or microservice expansion.

[0015] The information interaction module includes a text input unit, a language recognition unit, and an output display unit. The text input unit is used to receive source text input by the user. The language recognition unit is used to automatically identify the language type of the source text. The output display unit is used to display the final optimized target text in an interface format.

[0016] By introducing a multilingual language model into the language identification unit, the system is able to automatically adapt to multilingual scenarios. In addition to displaying the final translation results, the output display unit also supports extended functions such as synchronous highlighting of source and target texts, replacement history comparison, and semantic risk annotation, which improves the user's visual operation experience and translation review efficiency.

[0017] The text translation module includes a model selection unit, an encoding conversion unit, and a translation output unit. The model selection unit selects the corresponding translation model according to the identified language type. The encoding conversion unit is used to convert the source text into an encoding vector. The translation output unit is used to convert the encoding vector into text content in the target language.

[0018] The encoding conversion unit introduces a context fusion mechanism, which can correct the encoding result of the current sentence based on the semantics of the preceding and following sentences, effectively reducing the semantic jump problem in paragraph-level translation. The model selection unit supports the coexistence and dynamic scheduling of multiple models, automatically loads the optimal model structure after identifying the language, and retains the model switching interface to facilitate the later expansion of the system's low-resource language support capabilities.

[0019] The semantic understanding module includes an entity recognition unit, a dependency analysis unit, and a graph construction unit. The entity recognition unit is used to identify core entities in the text, the dependency analysis unit is used to analyze and process the dependency relationships between entities, and the graph construction unit is used to model the target text into a semantic graph.

[0020] The semantic optimization module includes a synonym replacement unit, a sentence reconstruction unit, and a risk control unit. The synonym replacement unit selects and optimizes text to replace synonymous content based on a semantic graph. The sentence reconstruction unit is used to reorganize the word order to optimize the structure. The risk control unit is used to ensure that the optimization process does not change the original meaning of the target text.

[0021] To enhance the precision and controllability of semantic processing, the semantic understanding module introduces a semantic type annotation mechanism and a context entity alignment strategy during graph construction. Specifically, the system uses a semantic relationship classifier configured in the graph construction unit to classify the edge relationships between different entity nodes into types such as "action-object" and "subject-behavior," thereby improving the expressive level of the semantic graph. Simultaneously, to ensure consistency between the translation context and the source text context, the system also introduces an entity alignment processor to match key entities in the target text with entities in the source text one-to-one, determining semantic consistency and constructing cross-language semantic bridge relationships. This mechanism provides a structured semantic basis for subsequent optimization modules, avoiding semantic drift or loss during the optimization process. During the semantic optimization process, the synonym replacement unit introduces a context fit scoring mechanism and a replacement candidate filtering strategy, which can automatically exclude replacement words with poor semantic adaptability or inconsistent stylistics; the sentence reconstruction unit supports multiple syntactic template matching and selects the appropriate template to generate sentences based on semantic structure; the risk control unit calculates semantic offset based on semantic vector comparison and automatically rolls back to the previous optimal state when the threshold is exceeded, ensuring that the final optimized translation is both fluent and semantically faithful.

[0022] Furthermore, the encoding conversion unit includes a text segmentation processor, a vector encoding processor, and a context fusion processor. The text segmentation processor is used to segment the text to adapt to the model input requirements. The vector encoding processor converts the segmented text into semantic vectors by calling a forward encoding model. The context fusion processor adjusts the semantic vectors by fusing context by calling a forward encoding model.

[0023] Furthermore, the context fusion processor calculates the fused semantic vector V according to the following formula. f ;

[0024]

[0025] Where V c V is the initial semantic vector. pi Let α be the semantic vector of the i-th context paragraph. α β is the control coefficient of the main clause, n is the influence coefficient of the i-th context paragraph, and n is the number of context paragraphs.

[0026] The influence coefficient is calculated according to the following formula:

[0027]

[0028] Where, d i τ represents the vector distance between the current sentence vector and the i-th context paragraph. τ This is the temperature coefficient.

[0029] Furthermore, the synonym replacement unit includes a semantic reference library, a context fit, and a replacement scorer. The semantic reference library is used to store a multilingual synonym reference table, the context fit is used to evaluate the context fit after replacement, and the replacement scorer scores candidate words based on language fluency and semantic retention rate.

[0030] The context adapter calculates the fluency score F according to the following formula:

[0031]

[0032] Where n is the total number of words in the sentence, w i Let represent the i-th word, and Big() represent the collocation frequency of the two words.

[0033] Furthermore, the replacement scorer calculates the replacement score R of the candidate synonym according to the following formula:

[0034]

[0035] Among them, u i S is the i-th candidate synonym. i For u i Semantic similarity with the original word, E(u) i T) is to make u i The semantic structure offset F generated after replacing the current text T. i The fluency score of the overall sentence after replacement is given, where λ1 is the fidelity coefficient and λ2 is the naturalness coefficient.

[0036] The beneficial effects achieved by this invention are:

[0037] This system integrates multiple modules, including language identification, semantic modeling, and optimization control, to construct a structured processing flow from source text input to optimized translation output. By introducing semantic graph construction and vector comparison mechanisms, the system can identify and retain key semantic elements during translation, effectively reducing the risk of semantic deviation. Through methods such as synonym substitution scoring, context fusion, and grammatical structure adjustment, it achieves automatic optimization of language expression, making the translation results more consistent with contextual logic and language habits. Simultaneously, the system has a difference alarm mechanism that triggers corrections when the semantic drift of the translation exceeds a controllable range, enhancing the controllability and security of translation quality. Compared to existing technologies, this invention significantly improves the accuracy, naturalness, and application flexibility in multilingual translation tasks, possessing good engineering feasibility and promotional value.

[0038] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the overall structural framework of the present invention;

[0040] Figure 2 This is a schematic diagram of the information interaction module of the present invention;

[0041] Figure 3 This is a schematic diagram illustrating the structure of the text translation module of the present invention;

[0042] Figure 4 This is a schematic diagram illustrating the semantic understanding module of the present invention;

[0043] Figure 5 This is a schematic diagram illustrating the semantic optimization module of the present invention;

[0044] Figure 6 This is a comparison chart of the translation accuracy of the present invention with that of three comparative systems;

[0045] Figure 7 This is a comparison chart of the semantic fidelity distribution of the present invention and the comparison system under different sentence counts. Detailed Implementation

[0046] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated beforehand. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0047] Example 1.

[0048] This embodiment provides an intelligent processing system for text translation and semantic optimization, combining... Figure 1 It includes an information interaction module, a text translation module, a semantic understanding module, and a semantic optimization module;

[0049] The information interaction module is used to input source text and output translation results; the text translation module is used to convert the source text into a language to obtain the target text; the semantic understanding module is used to construct the semantic structure of the target text; and the semantic optimization module is used to refine and optimize the target text based on the semantic structure.

[0050] The information interaction module includes a text input unit, a language recognition unit, and an output display unit. The text input unit is used to receive source text input by the user. The language recognition unit is used to automatically identify the language type of the source text. The output display unit is used to display the final optimized target text in an interface format.

[0051] The text translation module includes a model selection unit, an encoding conversion unit, and a translation output unit. The model selection unit selects the corresponding translation model according to the identified language type. The encoding conversion unit is used to convert the source text into an encoding vector. The translation output unit is used to convert the encoding vector into text content in the target language.

[0052] The semantic understanding module includes an entity recognition unit, a dependency analysis unit, and a graph construction unit. The entity recognition unit is used to identify core entities in the text, the dependency analysis unit is used to analyze and process the dependency relationships between entities, and the graph construction unit is used to model the target text into a semantic graph.

[0053] The semantic optimization module includes a synonym replacement unit, a sentence reconstruction unit, and a risk control unit. The synonym replacement unit selects and optimizes text to replace synonymous content based on a semantic graph. The sentence reconstruction unit is used to reorganize the word order to optimize the structure. The risk control unit is used to ensure that the optimization process does not change the original meaning of the target text.

[0054] The encoding conversion unit includes a text segmentation processor, a vector encoding processor, and a context fusion processor. The text segmentation processor is used to segment the text to adapt to the model input requirements. The vector encoding processor converts the segmented text into semantic vectors by calling a forward encoding model. The context fusion processor adjusts the semantic vectors by fusing context by calling a forward encoding model.

[0055] The context fusion processor calculates the fused semantic vector V according to the following formula. f ;

[0056]

[0057] Where V c V is the initial semantic vector. pi Let α be the semantic vector of the i-th context paragraph. α β is the control coefficient of the main clause, n is the influence coefficient of the i-th context paragraph, and n is the number of context paragraphs.

[0058] The influence coefficient is calculated according to the following formula:

[0059]

[0060] Where, di τ represents the vector distance between the current sentence vector and the i-th context paragraph. τ This is the temperature coefficient.

[0061] The synonym replacement unit includes a semantic reference library, a context fit, and a replacement scorer. The semantic reference library is used to store a multilingual synonym reference table. The context fit is used to evaluate the context fit after replacement. The replacement scorer scores candidate words based on language fluency and semantic retention rate.

[0062] The context adapter calculates the fluency score F according to the following formula:

[0063]

[0064] Where n is the total number of words in the sentence, w i Let represent the i-th word, and Big() represent the collocation frequency of the two words.

[0065] The replacement scorer calculates the replacement score R of the candidate synonyms according to the following formula:

[0066]

[0067] Among them, u i S is the i-th candidate synonym. i For u i Semantic similarity with the original word, E(u) i T) is to make u i The semantic structure offset F generated after replacing the current text T. i The fluency score of the overall sentence after replacement is given, where λ1 is the fidelity coefficient and λ2 is the naturalness coefficient.

[0068] Example 2.

[0069] This embodiment includes all the content of Embodiment 1, and provides an intelligent processing system for text translation and semantic optimization, including an information interaction module, a text translation module, a semantic understanding module and a semantic optimization module;

[0070] The information interaction module is used to input source text and output translation results; the text translation module is used to convert the source text into a language to obtain the target text; the semantic understanding module is used to construct the semantic structure of the target text; and the semantic optimization module is used to refine and optimize the target text based on the semantic structure.

[0071] Combination Figure 2The information interaction module includes a text input unit, a language recognition unit, and an output display unit. The text input unit is used to receive source text input by the user. The language recognition unit is used to automatically identify the language type of the source text. The output display unit is used to display the final optimized target text in an interface format.

[0072] Combination Figure 3 The text translation module includes a model selection unit, an encoding conversion unit, and a translation output unit. The model selection unit selects the corresponding translation model according to the identified language type. The encoding conversion unit is used to convert the source text into an encoding vector. The translation output unit is used to convert the encoding vector into text content in the target language.

[0073] Combination Figure 4 The semantic understanding module includes an entity recognition unit, a dependency analysis unit, and a graph construction unit. The entity recognition unit is used to identify core entities in the text, the dependency analysis unit is used to analyze and process the dependency relationships between entities, and the graph construction unit is used to model the target text into a semantic graph.

[0074] Combination Figure 5 The semantic optimization module includes a synonym replacement unit, a sentence reconstruction unit, and a risk control unit. The synonym replacement unit selects and optimizes text to replace synonymous content based on the semantic graph. The sentence reconstruction unit is used to reorganize the word order to optimize the structure. The risk control unit is used to ensure that the optimization process does not change the original meaning of the target text.

[0075] The text input unit includes an input interface processor, a format normalization processor, and an input state buffer. The input interface processor is used to receive raw text data input by the user, the format normalization processor is used to correct the format of the input text, and the input state buffer is used to record the text input process and historical versions.

[0076] The language identification unit includes a language model recognizer, a confidence evaluator, and a recognition label generator. The language model recognizer determines the language type of the input text based on a multilingual model. The confidence evaluator adds a confidence score to the recognition result. The recognition label generator is used to encode language information into labels.

[0077] The output display unit includes a result presentation processor, a highlighting processor, and a file export processor. The result presentation processor is used to display the final translation content, the highlighting processor is used to simultaneously highlight the corresponding content of the source text and the target text, and the file export processor is used to export the target text as a file.

[0078] The model selection unit includes a model library register, a language mapping processor, and a model output processor. The model library register is used to store forward encoding models and reverse decoding models for various languages. The language mapping processor maps the identified language and the target language to the corresponding model. The model output processor is used to output the mapped model to the encoding conversion unit and the translation output unit.

[0079] The encoding conversion unit includes a text segmentation processor, a vector encoding processor, and a context fusion processor. The text segmentation processor is used to segment the text to adapt to the model input requirements. The vector encoding processor converts the segmented text into semantic vectors by calling a forward encoding model. The context fusion processor adjusts the semantic vectors by fusing context by calling a forward encoding model.

[0080] The context fusion processor calculates the fused semantic vector V according to the following formula. f ;

[0081]

[0082] Where V c V is the initial semantic vector. pi Let α be the semantic vector of the i-th context paragraph. α β is the control coefficient of the main clause, n is the influence coefficient of the i-th context paragraph, and n is the number of context paragraphs.

[0083] The influence coefficient is calculated according to the following formula:

[0084]

[0085] Where, d i τ represents the vector distance between the current sentence vector and the i-th context paragraph. τ Temperature coefficient;

[0086] The translation output unit includes a decoding execution processor, a word order restoration processor, and a structure output buffer. The decoding execution processor converts semantic vectors into target text by calling the inverse decoding model. The word order restoration processor adjusts the word order of the target text by calling the inverse decoding model. The structure output buffer is used to cache the target text for easy subsequent use.

[0087] The entity recognition unit includes an entity extraction processor, a phrase block marker, and an ambiguity comparison processor. The entity extraction processor is used to extract entity words from the target text. The phrase block marker is used to group key entity words into phrase groups. The ambiguity comparison processor is used to identify polysemous words and mark risks.

[0088] The dependency analysis unit includes a dependency building processor, a syntactic structure parser, and a structural consistency checker. The dependency building processor is used to generate a syntactic dependency structure graph. The syntactic structure parser is used to identify structural relationships such as subject-verb-object, attributive, adverbial, and complement. The structural consistency checker is used to maintain logical consistency of the syntactic structure among multiple sentences.

[0089] The graph construction unit includes a graph structure generator, an entity alignment processor, and a semantic relation classifier. The graph structure generator is used to generate a semantic graph composed of nodes and edges. The entity alignment processor is used to semantically align the current target text entity with the original text entity. The semantic relation classifier is used to classify the edge relations by type.

[0090] The synonym replacement unit includes a semantic reference library, a context fit, and a replacement scorer. The semantic reference library is used to store a multilingual synonym reference table. The context fit is used to evaluate the context fit after replacement. The replacement scorer scores candidate words based on language fluency and semantic retention rate.

[0091] The context adapter calculates the fluency score F according to the following formula:

[0092]

[0093] Where n is the total number of words in the sentence, w i Let represent the i-th word, and Big() represent the collocation frequency of the two words;

[0094] The replacement scorer calculates the replacement score R of the candidate synonyms according to the following formula:

[0095]

[0096] Among them, u i S is the i-th candidate synonym. i For u i Semantic similarity with the original word, E(u) i T) is to make u i The semantic structure offset F generated after replacing the current text T. i To score the fluency of the overall sentence after replacement, λ1 is the fidelity coefficient and λ2 is the naturalness coefficient;

[0097] The sentence reconstruction unit includes a grammar template matcher, a sentence order adjustment processor, and a fluency detector. The grammar template matcher is used to provide a variety of sentence template libraries. The sentence order adjustment processor is used to adjust the main and subordinate clauses, adverbial positions, and inversion structures. The fluency detector is used to evaluate the fluency of the reconstructed text.

[0098] The risk control unit includes a vector call processor, a vector comparison processor, and a difference alarm processor. The vector call processor is used to convert the target text into a semantic vector by calling a forward encoding model. The vector comparison processor is used to compare and analyze the semantic vectors of the target text and the source text. The difference alarm processor is used to issue an alarm message and re-optimize the semantics when the comparison difference is too large.

[0099] The 'i' and 'j' mentioned above are ordinal numbers used to represent sequence numbers and have no actual meaning.

[0100] The following is a portion of the code for this system:

[0101]

[0102]

[0103]

[0104] The translation test of this system against 100 samples was conducted using three contrast systems, and the results were obtained. Figure 6 The comparison chart shown here compares the semantic guarantee degree distribution of system 1 with that of system 1 at different sentence counts. Figure 7 As shown.

[0105] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. An intelligent processing system for text translation and semantic optimization, characterized in that, It includes an information interaction module, a text translation module, a semantic understanding module, and a semantic optimization module; The information interaction module is used to input source text and output translation results; the text translation module is used to convert the source text into a language to obtain the target text; the semantic understanding module is used to construct the semantic structure of the target text; and the semantic optimization module is used to refine and optimize the target text based on the semantic structure. The information interaction module includes a text input unit, a language recognition unit, and an output display unit. The text input unit is used to receive source text input by the user. The language recognition unit is used to automatically identify the language type of the source text. The output display unit is used to display the final optimized target text in an interface format. The text translation module includes a model selection unit, an encoding conversion unit, and a translation output unit. The model selection unit selects the corresponding translation model according to the identified language type. The encoding conversion unit is used to convert the source text into an encoding vector. The translation output unit is used to convert the encoding vector into text content in the target language. The semantic understanding module includes an entity recognition unit, a dependency analysis unit, and a graph construction unit. The entity recognition unit is used to identify core entities in the text, the dependency analysis unit is used to analyze and process the dependency relationships between entities, and the graph construction unit is used to model the target text into a semantic graph. The semantic optimization module includes a synonym replacement unit, a sentence reconstruction unit, and a risk control unit. The synonym replacement unit selects and optimizes text to replace synonymous content based on the semantic graph. The sentence reconstruction unit is used to reorganize the word order to optimize the structure. The risk control unit is used to ensure that the optimization process does not change the original meaning of the target text.

2. The intelligent processing system for text translation and semantic optimization as described in claim 1, characterized in that, The encoding conversion unit includes a text segmentation processor, a vector encoding processor, and a context fusion processor. The text segmentation processor is used to segment the text to adapt to the model input requirements. The vector encoding processor converts the segmented text into semantic vectors by calling a forward encoding model. The context fusion processor adjusts the semantic vectors by fusing context by calling a forward encoding model.

3. The intelligent processing system for text translation and semantic optimization as described in claim 2, characterized in that, The context fusion processor calculates the fused semantic vector V according to the following formula. f ; Where V c V is the initial semantic vector. pi Let α be the semantic vector of the i-th context paragraph. α β is the control coefficient of the main clause, β is the influence coefficient of the i-th context paragraph, and n is the number of context paragraphs; The influence coefficient is calculated according to the following formula: Where, d i τ represents the vector distance between the current sentence vector and the i-th context paragraph. τ This is the temperature coefficient.

4. The intelligent processing system for text translation and semantic optimization as described in claim 3, characterized in that, The synonym replacement unit includes a semantic reference library, a context fit, and a replacement scorer. The semantic reference library is used to store a multilingual synonym reference table. The context fit is used to evaluate the context fit after replacement. The replacement scorer scores candidate words based on language fluency and semantic retention rate. The context adapter calculates the fluency score F according to the following formula: Where n is the total number of words in the sentence, w i Let represent the i-th word, and Big() represent the collocation frequency of the two words.

5. The intelligent processing system for text translation and semantic optimization as described in claim 4, characterized in that, The replacement scorer calculates the replacement score R of the candidate synonyms according to the following formula: Among them, u i S is the i-th candidate synonym. i For u i Semantic similarity with the original word, E(u) i T) is to make u i The semantic structure offset F generated after replacing the current text T. i The fluency score of the overall sentence after replacement is given, where λ1 is the fidelity coefficient and λ2 is the naturalness coefficient.

Images