A multi-lingual cultural content generation method and system based on natural language understanding

Abstract

This application provides a method and system for generating multilingual cultural content based on natural language understanding. The method first performs deep semantic analysis on the source language text content, extracting semantic structural elements, cultural feature elements, and contextual information. It then performs semantic alignment processing between these elements and a pre-defined multilingual cross-cultural knowledge graph. For semantic elements with cultural differences, it performs rule-constrained fuzzy reasoning to generate cultural adaptation control parameters to guide the content generation process. In the multilingual generation model, during content generation, the selection of vocabulary and expression methods are dynamically adjusted based on the cultural adaptation control parameters, outputting target language text content that conforms to the target language's cultural context. This solution enables cross-cultural dynamic adaptation generation based on fuzzy reasoning, thereby improving the cultural accuracy and contextual appropriateness of multilingual content output.

Description

Technical Field

[0001] This application relates to the field of semantic processing technology, and more specifically, to a method and system for generating multilingual cultural content based on natural language understanding. Background Technology

[0002] In the field of multilingual cultural content generation, traditional machine translation and early cross-cultural processing methods mainly rely on grammatical transformation and lexical mapping, with the core being the establishment of formal correspondences between languages. While these methods can convey basic semantics, their processing dimension typically remains at the surface structure of sentences, lacking a systematic analysis of the text's deeper cultural connotations, social context, and emotional tendencies. Existing technologies largely focus on language structure transformation, failing to integrate cultural features as a computable and inferable independent dimension into the generation process. This results in significant deficiencies in the cultural appropriateness and contextual suitability of the generated content.

[0003] Currently, the core technical challenge in multilingual content generation lies in the difficulty of achieving accurate and dynamic cross-cultural adaptation using existing methods. Traditional methods typically rely on static rule bases or limited cultural filtering vocabularies, resulting in rigid and discrete processing logic that struggles to handle the complex, implicit, and interconnected cultural elements in the source language text. For example, when processing sentences with specific cultural metaphors or allusions, existing technologies either fail to recognize them and resort to literal translation, leading to loss of meaning or misunderstanding, or they can only perform rigid substitutions based on a few preset rules, ignoring the continuity and contextual dependence of cultural transformation. This rigid approach results in poor cultural expressiveness in the generated content and may even cause cross-cultural communication barriers. If cross-cultural dynamic adaptation generation based on fuzzy reasoning could be achieved, the vocabulary and expressions of the text could be dynamically adjusted to conform to the target language's cultural context, thereby improving the cultural accuracy and contextual appropriateness of the generated content. Therefore, how to achieve cross-cultural dynamic adaptation generation based on fuzzy reasoning to improve the cultural accuracy and contextual appropriateness of multilingual content output has become a challenge facing the industry. Summary of the Invention

[0004] This application provides a method and system for generating multilingual cultural content based on natural language understanding, which can realize cross-cultural dynamic adaptation generation based on fuzzy reasoning, thereby improving the cultural accuracy and contextual appropriateness of multilingual content output.

[0005] In a first aspect, this application provides a method for generating multilingual cultural content based on natural language understanding, comprising the following steps: Receive source language text content and target language identifier; Deep semantic analysis is performed on the source language text content to extract semantic structural elements, cultural feature elements and contextual information of the source language text content; The semantic structural elements, cultural feature elements, and contextual information are semantically aligned with a preset multilingual cross-cultural knowledge graph. Fuzzy reasoning under rule constraints is performed on semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process. The source language text content, target language identifier, and cultural adaptation control parameters are input into a preset multilingual generation model. During the content generation process, the vocabulary selection and expression are dynamically adjusted according to the cultural adaptation control parameters to output target language text content that conforms to the cultural context of the target language.

[0006] Preferably, performing deep semantic analysis on the source language text content to extract semantic structural elements, cultural feature elements, and contextual information specifically includes: The source language text content is segmented and tagged with parts of speech to construct a lexical sequence with syntactic dependencies; Based on a pre-trained semantic understanding model, semantic roles are labeled on the word sequence to identify predicates, arguments and their semantic relationships in the source language text content, thereby obtaining semantic structural elements; Cultural elements are identified in the vocabulary sequence, and cultural feature elements including cultural customs, historical allusions, social taboos, and rhetorical habits are extracted. These cultural feature elements are used to characterize the implicit cultural norms and constraints in the text and the expressive biases of different languages. The source language text content is subjected to discourse-level contextual dependency analysis to extract logical connections, referential relationships and topic coherence information between sentences, forming contextual association information.

[0007] Preferably, the semantic alignment processing of the semantic structural elements, cultural feature elements, and contextual information with the preset multilingual cross-cultural knowledge graph specifically includes: The entities and concepts in the semantic structure elements are mapped to corresponding nodes in a multilingual cross-cultural knowledge graph through cross-linguistic word vectors, and the mapping confidence is calculated. The cultural feature elements are mapped to the cultural entity nodes corresponding to the source language side in the multilingual cross-cultural knowledge graph, and then the cultural entity nodes are compared with the cultural norm nodes corresponding to the target language side to identify cultural differences and their degree of difference. The semantic alignment process is weighted according to the contextual association information, and ambiguous or conflicting areas that occur during the semantic alignment process are marked and confidence-attenuated.

[0008] Preferably, constructing the multilingual cross-cultural knowledge graph specifically includes: Extract cultural entities, cultural relationships, and cultural constraint rules from a multilingual cultural corpus; The extracted cultural entities are subjected to cross-linguistic alignment and semantic normalization to construct a cultural concept network with source-target language mapping relationships; The cultural relationships and cultural constraint rules are associated with the corresponding cultural entity nodes in the cultural concept network, and the language affiliation and applicable scope of cultural constraints of each cultural entity are marked. The results are then represented and stored in the form of an attribute graph, forming a multilingual cross-cultural knowledge graph with semantic hierarchy and network association features.

[0009] Preferably, the semantic alignment process further includes: Based on the mapping confidence, the semantic difference coefficient between the source language and the target language in terms of semantic features is calculated; Based on the degree of difference, the cultural difference coefficient between the source language and the target language in terms of cultural characteristics is calculated; Based on the semantic difference coefficient and the cultural difference coefficient, a comprehensive cultural difference degree is determined by fusion. The comprehensive cultural difference degree is compared with a preset difference threshold to filter out semantic elements with significant cultural differences, and each filtered semantic element is labeled with a cultural difference type and processing priority, wherein the processing priority is determined by the value of the comprehensive cultural difference degree.

[0010] Preferably, the fuzzy reasoning under rule constraints performed on semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process specifically includes: Based on the cultural rule base in the multilingual cross-cultural knowledge graph, we identify cultural transformation rules related to cultural differences and extract the applicable conditions and execution weights of the rules. Furthermore, based on the fuzzy reasoning mechanism, according to the cultural difference type and processing priority of the semantic elements, the contextual acceptability and cultural sensitivity of the semantic elements in the target language are comprehensively evaluated to generate cultural adaptation membership degree; Based on the processing priority, cultural adaptation membership degree, and preset conversion strategy, cultural adaptation control parameters, including vocabulary replacement tendency, sentence structure adjustment intensity, and cultural annotation insertion strategy, are dynamically generated.

[0011] Preferably, during the content generation process, the selection of vocabulary and expression methods are dynamically adjusted according to the cultural adaptation control parameters to output target language text content that conforms to the target language cultural context. Specifically, this includes: Input the source language text content and the cultural adaptation control parameters into the multilingual generation model; During the decoding phase, the probability distribution of words is adjusted according to the cultural adaptation control parameters, and words that conform to the cultural habits of the target language are selected first. Based on the sentence structure adjustment intensity in the cultural adaptation control parameters, the complexity and word order of the generated sentences are dynamically adjusted. Based on the cultural annotation insertion strategy and processing priority in the cultural adaptation control parameters, cultural background descriptions or adaptation prompts are added to the corresponding positions of cultural difference points with higher priority in the generated text, forming natural, fluent and culturally adapted target language text content.

[0012] Secondly, this application provides a multilingual cultural content generation system based on natural language understanding, including: The receiving module is used to receive source language text content and target language identifier; The processing module is used to perform deep semantic analysis on the source language text content, and extract the semantic structural elements, cultural feature elements and contextual information of the source language text content; The processing module is also used to perform semantic alignment processing on the semantic structural elements, cultural feature elements and contextual association information with the preset multilingual cross-cultural knowledge graph, and to perform fuzzy reasoning under rule constraints for semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process. The execution module is used to input the source language text content, the target language identifier, and the cultural adaptation control parameters into a preset multilingual generation model. During the content generation process, the module dynamically adjusts the vocabulary selection and expression methods according to the cultural adaptation control parameters, and outputs target language text content that conforms to the cultural context of the target language.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described method for generating multilingual cultural content based on natural language understanding.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for generating multilingual cultural content based on natural language understanding.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The proposed solution effectively addresses the problem of insufficient cultural adaptation in existing multilingual content generation. First, it performs deep semantic analysis on the source language text, extracting semantic structural elements, cultural feature elements, and contextual information. This extraction enables a systematic representation of the core semantic relationships, implicit cultural information, and textual context, providing structured and computable foundational data for cross-cultural adaptation. Second, it performs semantic alignment processing with a pre-defined multilingual cross-cultural knowledge graph, identifying semantic elements with significant cultural differences. This alignment process accurately maps semantic nodes and cultural entities between the source and target languages, quantifying cultural differences and their degree, thereby achieving continuous and measurable matching in both semantic and cultural dimensions. Then, fuzzy reasoning under rule constraints is applied to semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process. Through fuzzy reasoning, semantic difference coefficients, cultural difference coefficients, processing priorities, and cultural adaptation membership can be comprehensively considered to dynamically adjust the vocabulary selection, sentence structure, and cultural expression of the target language text, achieving continuous and adjustable control of cross-cultural generation. Finally, during the content generation process, the vocabulary selection and expression are dynamically adjusted according to the cultural adaptation control parameters to output target language text content that conforms to the target language cultural context. This ensures that semantic integrity, cultural appropriateness, and text coherence are simultaneously satisfied, achieving natural, accurate, and comprehensible text generation in a cross-cultural context. In summary, the proposed solution can achieve cross-cultural dynamic adaptation generation based on fuzzy reasoning, thereby improving the cultural accuracy and contextual suitability of multilingual content output. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of a multilingual cultural content generation method based on natural language understanding, according to some embodiments of this application. Figure 2 This is an exemplary flowchart of a method for generating multilingual cultural content based on natural language understanding, according to some embodiments of this application; Figure 3 This is a schematic diagram illustrating the process of constructing a multilingual cross-cultural knowledge graph according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a multilingual cultural content generation system based on natural language understanding, according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device that implements a method for generating multilingual cultural content based on natural language understanding, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] refer to Figure 1 This figure is a schematic diagram of an application scenario for a multilingual cultural content generation method based on natural language understanding, according to some embodiments of this application. The figure includes an input device, a server, a communication network, and a terminal. The input device is connected to the server via the network, and the terminal is connected to the server via the communication network. The server obtains the source language text content and target language identifier transmitted by the input device, performs deep semantic analysis on the source language text content, and extracts semantic structural elements, cultural feature elements, and contextual information. The server performs semantic alignment processing with a preset multilingual cross-cultural knowledge graph, performs fuzzy reasoning under rule constraints on semantic elements with cultural differences, and generates cultural adaptation control parameters. The server then inputs the source language text content, target language identifier, and cultural adaptation control parameters into a preset multilingual generation model, dynamically adjusts the vocabulary selection and expression methods according to the cultural adaptation control parameters, and outputs target language text content that conforms to the target language cultural context. The server can feed back the generated target language text content to the terminal for users to view and use. The input devices may include a text content input interface, a language identifier selection module, and a source language file import tool; the terminals may include a multilingual content management backend, a content generation client, and a culture adaptation and review terminal; the server may be a local cultural content processing cluster or a cloud-based multilingual cultural content intelligent generation service platform.

[0019] refer to Figure 2 The figure is an exemplary flowchart of a multilingual cultural content generation method based on natural language understanding, according to some embodiments of this application. This method mainly includes the following steps: In step 101, the source language text content and the target language identifier are received.

[0020] In some embodiments, receiving source language text content and target language identifier can be achieved in the following way: the text to be converted submitted by the user can be received through an interactive interface or application programming interface, and the target language identifier selected by the user can be obtained simultaneously. The target language identifier is parsed and mapped to a standardized language code containing specific regional cultural information that is relied upon by the internal processing flow. Then, the identifier is associated with the corresponding cultural norm system in the knowledge graph. After the parsing and association are completed, the source language text content and the identified target language identifier are passed together to the subsequent semantic parsing module. The target language identifier is specifically a standardized code that indicates the cultural variant of a country or region, which is used to accurately locate the target cultural context that the content generation needs to adapt to.

[0021] In step 102, deep semantic analysis is performed on the source language text content to extract semantic structural elements, cultural feature elements, and contextual information of the source language text content.

[0022] In some embodiments, performing deep semantic analysis on the source language text content to extract semantic structural elements, cultural feature elements, and contextual information can be achieved through the following steps: The source language text content is segmented and tagged with parts of speech to construct a lexical sequence with syntactic dependencies; Based on a pre-trained semantic understanding model, semantic roles are labeled on the word sequence to identify predicates, arguments and their semantic relationships in the source language text content, thereby obtaining semantic structural elements; Cultural elements are identified in the vocabulary sequence, and cultural feature elements including cultural customs, historical allusions, social taboos, and rhetorical habits are extracted. These cultural feature elements are used to characterize the implicit cultural norms and constraints in the text and the expressive biases of different languages. The source language text content is subjected to discourse-level contextual dependency analysis to extract logical connections, referential relationships and topic coherence information between sentences, forming contextual association information.

[0023] It should be noted that the semantic structure elements in this application are structured semantic information used to characterize the semantic role relationship between predicates and arguments and the logical structure of core events in the source language text content; the cultural feature elements are structured cultural information used to characterize the implicit cultural normative constraints, cultural background information and language expression bias features in the source language text content; the context association information is discourse-level contextual constraint information used to characterize the semantic continuity relationship, referential correspondence relationship and topic coherence structure between sentences in the source language text content, and is used to constrain semantic alignment and overall consistency and expression coherence in the content generation process.

[0024] In specific implementation, firstly, the source language text content is segmented and tagged with parts of speech. Constructing a lexical sequence with syntactic dependencies can be achieved as follows: The received source language text content is segmented and tagged with parts of speech. This segmentation and tagging process involves using specialized natural language processing tools corresponding to the source language to cut the continuous text string into independent lexical units, and tagging each lexical unit with its part-of-speech category, such as noun, verb, or adjective. Based on this, a syntactic analysis model is used to perform syntactic parsing on the tagged lexical sequence. This syntactic parsing refers to the process of analyzing the lexical sequence and... Its part-of-speech tags calculate all possible dependency relationships between words and select the optimal dependency structure based on the scoring algorithm, thereby constructing a lexical sequence with syntactic dependencies that can reflect grammatical relationships such as modification relationships, subject-predicate relationships, and verb-object relationships between words. This lexical sequence is usually represented in the computer as a dependency syntax tree or dependency graph data structure, where each node corresponds to a lexical unit and each directed edge corresponds to a specific grammatical dependency relationship type. The lexical sequence with syntactic dependencies is a graph structure data containing lexical nodes and their dependency relationship edges used to represent the surface grammatical structure of sentences.

[0025] Secondly, semantic role labeling is performed on the lexical sequence based on a pre-trained semantic understanding model to identify predicates, arguments, and their semantic relationships in the source language text content, thereby obtaining semantic structural elements. This can be achieved in the following way: A lexical sequence with syntactic dependencies can be input as input data into a pre-trained semantic understanding model. This semantic understanding model is typically a large-scale deep neural network based on the Transformer architecture. Its pre-training process first involves learning the general grammar, word meaning, and contextual representation capabilities of a language through self-supervised learning tasks on a massive unlabeled general corpus, such as masked language modeling and next-sentence prediction. Afterward, the semantic understanding model is further labeled with manually assigned semantic roles. Supervised fine-tuning training is performed on a large-scale corpus of words to learn to accurately identify predicates in sentences, label each predicate with its related core argument components (such as agent, patient, instrument, location), and assign the correct semantic role label to each argument. Through the reasoning of this semantic understanding model, it can automatically identify all predicates in the source language text content, that is, the core words expressing actions or states, identify the related argument components around each predicate, and accurately determine the semantic relationship between each argument and the predicate. Finally, the labeling results of the semantic understanding model, which consists of predicate-argument-role triples, are structured and integrated to obtain semantic structural elements used to abstractly represent the core event logic and participant relationships in the text.

[0026] Then, cultural element identification is performed on the vocabulary sequence to extract cultural feature elements, including cultural customs, historical allusions, social taboos, and rhetorical habits. The cultural feature elements are used to characterize the implicit cultural normative constraints and the expression bias characteristics of different languages ​​in the text. This can be achieved in the following way: cultural element identification is performed on the vocabulary sequence with syntactic dependencies. The cultural element identification refers to calling a pre-built cultural knowledge base and rule base, and using pattern matching and entity linking technology to identify and label words or phrases in the text that involve specific cultural fields. The extracted cultural feature elements specifically include language units and their contexts that directly or indirectly point to cultural customs, historical allusions, social taboos, and rhetorical habits. The cultural feature elements are a set of key information used to characterize the implicit cultural normative constraints in the text and reveal the potential expression bias characteristics between the source language and the target language.

[0027] Finally, a discourse-level contextual dependency analysis is performed on the source language text content to extract logical connections, referential relationships, and topic coherence information between sentences. The formation of contextual information can be achieved in the following way: Discourse-level contextual dependency analysis is performed on the source language text content. Discourse-level contextual dependency analysis refers to going beyond the scope of a single sentence and using discourse analysis models to automatically identify and infer the logical connections between multiple sentences in the text, the specific antecedents of pronoun references, and the continuation and transformation of core topics. The logical connections, referential relationships, and topic coherence information extracted in this way together form contextual information. The contextual information is a background constraint used to ensure cross-sentence semantic consistency, eliminate ambiguity, and maintain the overall coherence of the generated content.

[0028] In step 103, the semantic structural elements, cultural feature elements, and contextual information are semantically aligned with a preset multilingual cross-cultural knowledge graph. Fuzzy reasoning under rule constraints is performed on semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process.

[0029] In some embodiments, semantic alignment of the semantic structural elements, cultural feature elements, and contextual information with a preset multilingual cross-cultural knowledge graph can be achieved through the following steps: The entities and concepts in the semantic structure elements are mapped to corresponding nodes in a multilingual cross-cultural knowledge graph through cross-linguistic word vectors, and the mapping confidence is calculated. The cultural feature elements are mapped to the cultural entity nodes corresponding to the source language side in the multilingual cross-cultural knowledge graph, and then the cultural entity nodes are compared with the cultural norm nodes corresponding to the target language side to identify cultural differences and their degree of difference. The semantic alignment process is weighted according to the contextual association information, and ambiguous or conflicting areas that occur during the semantic alignment process are marked and confidence-attenuated.

[0030] It should be noted that the mapping confidence in this application is a quantitative evaluation index used to characterize the reliability of semantic matching when entities and concepts in semantic structure elements are mapped to corresponding nodes in a multilingual cross-cultural knowledge graph; the cultural difference point is a key corresponding position that characterizes the significant semantic deviation or cultural inconsistency between the source language cultural entity node and the target language cultural norm node; the difference degree is a standardized numerical index that quantitatively describes the degree of deviation of the cultural difference point in dimensions such as emotional attributes, usage scenarios, and cultural constraints; the confidence attenuation processing is an adjustment mechanism to reduce the influence of ambiguous mapping or conflict mapping results on the weight during semantic alignment.

[0031] In specific implementation, firstly, the entities and concepts in the semantic structure elements are mapped to corresponding nodes in the multilingual cross-cultural knowledge graph using cross-lingual word vectors, and the mapping confidence is calculated. This can be achieved as follows: The entities and core concepts identified in the semantic structure elements can be traversed. For each item to be mapped, a pre-trained cross-lingual word vector model is used to convert its lexical units into vector representations in a high-dimensional semantic space. Simultaneously, in the target language seed graph of the multilingual cross-cultural knowledge graph, each graph node has already generated its corresponding vector representation using the same or compatible model and stored it in the node attributes. The mapping confidence is calculated by determining the cosine similarity between the source language vector and all candidate target node vectors. For each source language entity or concept, one or more candidate nodes with the highest similarity are found. The result of this similarity calculation is the mapping confidence score, which is a value between 0 and 1 used to quantify the reliability of the mapping relationship. The system will retain the best matching node and its mapping confidence score to form an initial list of cross-language mapping pairs. It should be noted that the cross-language word vector model is a trained neural network model. Its principle is to map words from different languages ​​to the same shared semantic vector space, so that the vector representations of words with the same or similar meanings in different languages ​​are also close to each other in this space. This is achieved by training on a large-scale parallel corpus or by using cross-language alignment constraints in a monolingual corpus.

[0032] Secondly, mapping the cultural feature elements to the corresponding cultural entity nodes on the source language side of the multilingual cross-cultural knowledge graph, and then comparing these cultural entity nodes with the corresponding cultural normative nodes on the target language side to identify cultural differences and their degree of difference, can be achieved in the following way: Specific cultural elements extracted from the cultural feature elements can be located to specific cultural entity nodes on the source language side of the graph based on predefined membership or instantiation relationships in the knowledge graph. Then, using explicitly labeled cross-cultural correspondence edges or cultural equivalence edges in the graph, one or more cultural normative nodes associated with the source language cultural entity node on the target language side can be found, and these two nodes can be compared procedurally. The attribute values ​​attached to each point are defined and assigned values ​​when constructing the knowledge graph. These attributes include sentiment polarity values, applicable scenario tags, taboo level index, and symbolic meaning descriptions. The deviation between each pair of comparable attributes is further calculated. The specific method includes calculating the absolute value of the numerical difference for numerical attributes such as sentiment polarity, and querying the discrete distance value of symbolic tags according to a preset semantic distance table. The semantic distance table defines the quantitative relationship of semantic gap between different tag concepts. After all attribute deviations are weighted and averaged with preset weights and then normalized, the degree of difference for this cultural element is formed. This degree of difference is also a standardized scalar value used to measure the severity of cultural inconsistency.

[0033] Then, the semantic alignment process is weighted according to the contextual information, and ambiguous or conflicting regions that appear in the semantic alignment process are marked and confidence decayed. This can be achieved in the following way: The preliminary results of the above two steps can be globally optimized according to the contextual information. Specifically, the system uses the referential relationship chain obtained by discourse analysis. If multiple entities belong to the same referential chain, the system will apply a consistency constraint, which will force them to have close semantic association in the mapping nodes in the target graph. The system will also iteratively adjust the weight of their mapping selection to make the results more consistent. At the same time, using topic coherence information, multiple concepts belonging to the same core topic will be assigned higher global decision weights in their mapping process to maintain semantic consistency at the topic level. For ambiguous or conflicting regions that appear in the mapping or difference identification process, a typical situation is that a source language concept is mapped to multiple target nodes and the mapping confidence is very close, or a cultural entity has multiple possible corresponding norms in the target culture and the difference calculation results are contradictory. The system marks such regions as unresolved ambiguity regions. For the initial results in these regions, the system applies a confidence decay process. Specifically, the original confidence or difference is multiplied by a pre-set decay coefficient less than 1 to significantly reduce its influence in subsequent inference and avoid overall decision-making errors due to local uncertainties.

[0034] Preferably, in some embodiments, reference is made to Figure 3 As shown in the figure, this is a schematic diagram of the process of constructing a multilingual cross-cultural knowledge graph in some embodiments of this application. The construction of a multilingual cross-cultural knowledge graph in this embodiment can be achieved by the following steps: In step 1031, cultural entities, cultural relationships, and cultural constraint rules are extracted from a multilingual cultural corpus; In step 1032, the extracted cultural entities are subjected to cross-language alignment and semantic normalization to construct a cultural concept network with source-target language mapping relationship; In step 1033, the cultural relationships and cultural constraint rules are associated with the corresponding cultural entity nodes in the cultural concept network, the language affiliation and applicable scope of cultural constraints of each cultural entity are marked, and the structured representation and storage are performed in the form of attribute graph to form a multilingual cross-cultural knowledge graph with semantic hierarchy and network association features.

[0035] It should be noted that the cultural entities in this application are structured units used to represent specific cultural concepts or objects and their attribute information; the cultural relationships are structured edges representing logical connections such as affiliation, origin, or symbolism between cultural entities; the cultural constraint rules are logical descriptions used to represent the behavioral norms or applicable restrictions of cultural entities in a specific context; and the multilingual cross-cultural knowledge graph is a structured knowledge network used to integrate cultural entities, relationships, and constraint rules of different languages.

[0036] In practice, the extraction of cultural entities, cultural relationships, and cultural constraint rules from a multilingual cultural corpus can be achieved as follows: Information is extracted from a pre-collected and cleaned multilingual cultural corpus. Named entity recognition technology is used to identify and extract cultural entities representing specific cultural connotations from the corpus, such as traditional festivals, historical allusions, specific customary names, and objects with cultural symbolic significance. Furthermore, through a relationship extraction model or a matching method based on predefined relationship patterns, various cultural relationships between these cultural entities are identified and extracted. These cultural relationships specifically include belonging relationships, origin relationships, symbolic relationships, and relationships commonly used in specific scenarios. Simultaneously, by combining rule template matching and statistical pattern mining methods, cultural constraint rules describing cultural norms, social taboos, or behavioral suggestions are extracted from the corpus. These rules exist in the form of logical statements. Finally, the extracted set of cultural entities, cultural relationships, and cultural constraint rules are used as the original knowledge material.

[0037] Secondly, the extracted cultural entities undergo cross-linguistic alignment and semantic normalization to construct a cultural concept network with source-target language mapping relationships. This can be achieved in the following way: using a large-scale bilingual dictionary, cross-lingual word vector model, or an existing authoritative bilingual knowledge base as alignment references, entities expressing the same or highly similar cultural concepts in different languages ​​are associated and clustered by calculating semantic similarity and setting a threshold, forming cross-lingual cultural concept clusters. Then, semantic normalization is performed on each concept cluster, that is, a globally unique normalized concept identifier is assigned to the cluster, and its core semantic description is automatically generated by experts or by integrating multi-source definitions. In this process, the system establishes and records the mapping relationship from each source language entity to its corresponding normalized concept, as well as the equivalence mapping relationship between normalized concepts in different languages, thereby constructing a basic cultural concept network with normalized concepts as nodes and cross-lingual equivalence mapping relationships as edges.

[0038] Then, the cultural relationships and cultural constraint rules are associated with the corresponding cultural entity nodes in the cultural concept network. The language affiliation and applicable scope of each cultural entity are labeled, and the structured representation and storage are performed in the form of an attribute graph. The formation of a multilingual cross-cultural knowledge graph with semantic hierarchy and network association features can be achieved in the following way: The extracted cultural relationships and cultural constraint rules can be associated with the corresponding normalized concept nodes in the constructed cultural concept network. For cultural relationships, they are treated as directed edges with specific type labels, connecting two related concept nodes. For cultural constraint rules, they are transformed into structured logical expressions or links pointing to detailed descriptive texts, and attached as attributes to the related concept nodes. At the same time, each concept node is labeled with its language affiliation list, recording the language and culture from which the concept originates or is applicable; and the applicable scope of the relevant cultural constraint rules is labeled, indicating the specific cultural context conditions under which each rule takes effect. Finally, all the above nodes, edges and their attributes are organized in an attribute graph data model and stored in a dedicated graph database, thereby forming a multilingual cross-cultural knowledge graph with rich semantic associations that can be directly used for semantic computation and reasoning.

[0039] Preferably, in some embodiments, the semantic alignment process further includes: Based on the mapping confidence, the semantic difference coefficient between the source language and the target language in terms of semantic features is calculated; Based on the degree of difference, the cultural difference coefficient between the source language and the target language in terms of cultural characteristics is calculated; Based on the semantic difference coefficient and the cultural difference coefficient, a comprehensive cultural difference degree is determined by fusion. The comprehensive cultural difference degree is compared with a preset difference threshold to filter out semantic elements with significant cultural differences, and each filtered semantic element is labeled with a cultural difference type and processing priority, wherein the processing priority is determined by the value of the comprehensive cultural difference degree.

[0040] It should be noted that the semantic difference coefficient in this application is a quantitative indicator used to measure the degree of deviation between the semantic structural elements in the source language text and the corresponding concepts in the target language in terms of semantic feature matching; the cultural difference coefficient is a quantitative indicator used to measure the degree of deviation between the cultural feature elements in the source language text and the corresponding cultural norms in the target language in terms of cultural adaptation; the comprehensive cultural difference degree is a quantitative indicator used to characterize the degree of deviation of the source language text from the target language at the semantic and cultural levels; the cultural difference type is a classification label used to identify the nature of the deviation between the cultural features in the source language text and the cultural norms of the target language; and the processing priority is an importance indicator indicating the order of cultural adaptation processing of each semantic element in content generation.

[0041] In practice, firstly, based on the mapping confidence, the semantic difference coefficient between the source language and the target language in terms of semantic features can be calculated in the following way: For each semantic element to be evaluated, the mapping confidence obtained in the semantic alignment step and calculated when mapping all its subordinate entities and concepts to the target language knowledge graph nodes are collected. Then, the arithmetic mean of these mapping confidences is calculated to obtain a preliminary aggregate value. Then, this aggregate value is converted into a positive difference index through a preset linear transformation function. The design principle of this transformation function is: the higher the average value of the mapping confidence, the lower the difference value obtained after transformation; the lower the average value of the mapping confidence, the higher the difference value obtained after transformation. Finally, the normalized value of the positive difference index obtained after transformation is used as the semantic difference coefficient of the semantic element.

[0042] Secondly, based on the degree of difference, the cultural difference coefficient between the source language and the target language in terms of cultural features can be calculated in the following way: For the same semantic element, collect the quantified cultural difference degrees obtained after cross-cultural comparison of all its associated cultural feature elements, and calculate a weighted average of these cultural difference degrees. The weights can be preset according to the type of cultural feature element or its own importance. For example, the weight of the difference degree of social taboo can be higher than the weight of the difference degree of general custom. The calculated weighted average directly reflects the overall deviation level of the semantic element at the cultural level. Then, the value obtained by normalizing this weighted average is used as the cultural difference coefficient of the semantic element.

[0043] Then, based on the semantic difference coefficient and the cultural difference coefficient, the comprehensive cultural difference degree can be determined by the following method: For each semantic element, the calculated semantic difference coefficient and cultural difference coefficient are input into a fusion function. This fusion function usually adopts a weighted summation model, that is, a weight factor is assigned to the semantic difference coefficient and the cultural difference coefficient respectively, and the weighted sum of the two is used as the fusion result. The specific value of the weight factor can be preset according to the different emphases on semantic accuracy and cultural adaptability in the actual task. The result of this weighted summation is then used as the comprehensive cultural difference degree of the semantic element.

[0044] Finally, the comprehensive cultural difference degree is compared with a preset difference threshold to filter out semantic elements with significant cultural differences. Each filtered semantic element is then labeled with a cultural difference type and processing priority. The processing priority is determined by the value of the comprehensive cultural difference degree, which can be achieved as follows: the comprehensive cultural difference degree of all semantic elements is compared with a preset global difference threshold. This threshold can be set based on historical experience data or domain expert opinions. Any semantic element whose comprehensive cultural difference degree exceeds this threshold is filtered as a pending element with significant cultural differences. For each filtered semantic element, firstly, based on the main category to which its cultural characteristic elements belong, a specific cultural difference type is labeled, such as "allusion difference" or "taboo conflict." Simultaneously, it is sorted in descending order according to the specific value of its comprehensive cultural difference degree, with larger values ​​ranking higher. Processing priorities are then assigned according to this ranking; for example, the element ranked first is assigned the highest priority. Finally, this list containing the pending semantic elements and their corresponding difference types and priorities serves as a key input to guide subsequent fuzzy reasoning and content generation.

[0045] In some embodiments, performing fuzzy reasoning under rule constraints on semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process can be achieved through the following steps: Based on the cultural rule base in the multilingual cross-cultural knowledge graph, we identify cultural transformation rules related to cultural differences and extract the applicable conditions and execution weights of the rules. Furthermore, based on the fuzzy reasoning mechanism, according to the cultural difference type and processing priority of the semantic elements, the contextual acceptability and cultural sensitivity of the semantic elements in the target language are comprehensively evaluated to generate cultural adaptation membership degree; Based on the processing priority, cultural adaptation membership degree, and preset conversion strategy, cultural adaptation control parameters, including vocabulary replacement tendency, sentence structure adjustment intensity, and cultural annotation insertion strategy, are dynamically generated.

[0046] It should be noted that the applicable conditions and execution weights of the rules in this application are parameters used to indicate the scope of application and degree of influence of cultural transformation rules in specific semantic contexts; the contextual acceptability is an indicator used to measure the degree to which semantic elements conform to language habits and cultural norms in the target language; the cultural sensitivity is an indicator used to measure the degree to which semantic elements touch upon cultural taboos or differences in the target language; the cultural adaptation membership degree is an indicator used to measure the degree to which semantic elements conform to cultural norms in the target context; the transformation strategy is a set of rules used to guide the adjustment of cultural difference semantic elements in the target language; and the cultural adaptation control parameter is a quantifiable adjustment indicator used to regulate the cultural expression mode during the target language text generation process.

[0047] In specific implementation, firstly, based on the cultural rule base in the multilingual cross-cultural knowledge graph, cultural transformation rules related to cultural differences are identified, and the applicable conditions and execution weights of the rules are extracted. This can be achieved in the following way: traverse each semantic element and its cultural difference point marked in the difference screening step. For each difference point, query and match in the cultural rule base attached to the multilingual cross-cultural knowledge graph. The cultural rule base stores predefined "if-then" form cultural transformation rules, such as "if the source cultural concept is A and the target culture is B, then it is recommended to adopt processing strategy C". Retrieve all cultural transformation rules whose preconditions match the difference point to form an initial candidate rule set. Then, parse and extract its complete applicable condition description and a preset rule execution weight from each matching rule. This execution weight is usually set in advance by domain experts when building the rule base based on the universality and importance of the rule. Further, associate the matched rule and its extracted applicable conditions and execution weight with the semantic element to form the initial rule matching result.

[0048] Secondly, based on the fuzzy reasoning mechanism, and according to the cultural difference type and processing priority of the semantic element, the contextual acceptability and cultural sensitivity of the semantic element in the target language are comprehensively evaluated, and the cultural adaptation membership degree is generated. This can be achieved in the following way: The obtained rule matching result, the cultural difference type of the semantic element, and its processing priority are input into a preset fuzzy reasoning engine. The reasoning engine first fuzzifies the clear input values, i.e., the variables such as "cultural difference type", "processing priority", and the degree of satisfaction of the rule "applicability conditions", and calculates the degree to which it belongs to the fuzzy language values ​​such as "high", "medium", and "low" according to the predefined membership degree function. The core of this engine is a rule base containing a series of fuzzy inference rules. These rules define the logical mapping relationship between the aforementioned fuzzy inputs and fuzzy outputs such as "contextual acceptability" and "cultural sensitivity". By applying synthetic inference methods in fuzzy logic, such as Madani inference, composite operations are performed on all activated fuzzy rules to obtain a fuzzy set of output variables. Finally, defuzzification methods such as centroid method or maximum membership method are used to transform the output fuzzy set into a clear quantized value between 0 and 1, which is the cultural adaptation membership degree of the current semantic element for a certain preset processing method. This value represents the applicability and acceptability of the processing method in the target cultural context.

[0049] Then, based on the processing priority, cultural adaptation membership degree, and preset conversion strategy, the dynamic generation of cultural adaptation control parameters, including word substitution tendency, sentence structure adjustment intensity, and cultural annotation insertion strategy, can be achieved in the following way: Taking into account the processing priority of the current semantic element, the cultural adaptation membership degree calculated for various processing methods, and a preset overall strategy configuration file that defines the basic conversion tendency under different scenarios, decisions are made based on this information: for elements with high priority and low cultural adaptation membership degree, a more proactive and intervention-oriented conversion strategy is preferred; for elements with low priority and high cultural adaptation membership degree, a more conservative strategy may be adopted. Based on this decision, the system dynamically generates a set of specific and operable cultural adaptation control parameters. These parameters are represented in the form of key-value pairs or structured objects, clearly indicating the operations that the generation model needs to perform. For example, the "lexical substitution tendency" is set to point to the equivalent words in the target culture, the "sentence structure adjustment intensity" is set to a specific value to control the extent of word order changes, and the "cultural annotation insertion strategy" is set to "add a brief explanation at the end of the sentence". Finally, all the control parameters generated for different semantic elements are summarized and structured to form a complete set of cultural adaptation control parameters covering the entire text, which serves as an instruction set to directly control the final content generation process.

[0050] Furthermore, it should be noted that this application addresses the problem in existing multilingual content generation technologies that lack fine-grained handling of cultural differences, easily leading to cultural incompatibility or contextual deviations in target language texts. It proposes a method for generating cultural adaptation control parameters based on rule-constrained fuzzy reasoning. This method extracts cultural transformation rules related to cultural differences from a multilingual cross-cultural knowledge graph and, combined with the rules' applicability conditions and execution weights, comprehensively evaluates semantic elements with cultural differences. This quantifies contextual acceptability and cultural sensitivity, generating cultural adaptation membership degrees. Based on processing priorities, cultural adaptation membership degrees, and preset transformation strategies, it dynamically generates cultural adaptation control parameters such as word substitution tendencies, sentence structure adjustment strengths, and cultural annotation insertion strategies. This allows for cultural optimization and precise adjustment of semantic expression during content generation. This solution significantly improves the cultural adaptability and contextual rationality of generated text in the target language, solving technical problems such as insufficient handling of cultural differences, singular adaptation strategies, and significant semantic deviations in existing technologies. It possesses strong practicality and innovation.

[0051] In step 104, the source language text content, the target language identifier, and the cultural adaptation control parameters are input into a preset multilingual generation model. During the content generation process, the vocabulary selection and expression are dynamically adjusted according to the cultural adaptation control parameters to output target language text content that conforms to the cultural context of the target language.

[0052] In this embodiment, the pre-training process of the multilingual generation model includes the following steps: First, collect multilingual text corpora covering the source and target languages, and standardize the corpora, including word segmentation, part-of-speech tagging, and syntactic structure parsing; second, construct a unified multilingual word vector space, and map words from different languages ​​to a shared vector space through a cross-language alignment algorithm to achieve semantically consistent representation; then, train a deep neural network model on multilingual text sequences, using self-supervised learning strategies, such as language model prediction tasks, mask prediction tasks, and sequence alignment tasks, to enable the model to capture semantic structure, contextual dependencies, and cross-language expression rules; simultaneously, combine multilingual cross-cultural knowledge graphs for auxiliary training, injecting cultural entities, cultural relationships, and cultural constraints from the text into the model to enhance the model's sensitivity and adaptability to cultural features; finally, iteratively optimize the model parameters to achieve high-precision representation capabilities of multilingual texts in terms of semantic understanding, cross-language mapping, and cultural adaptation, thereby completing the model training.

[0053] In some embodiments, the following steps can be used to dynamically adjust the vocabulary selection and expression methods according to the cultural adaptation control parameters during the content generation process, and output target language text content that conforms to the target language cultural context: Input the source language text content and the cultural adaptation control parameters into the multilingual generation model; During the decoding phase, the probability distribution of words is adjusted according to the cultural adaptation control parameters, and words that conform to the cultural habits of the target language are selected first. Based on the sentence structure adjustment intensity in the cultural adaptation control parameters, the complexity and word order of the generated sentences are dynamically adjusted. Based on the cultural annotation insertion strategy and processing priority in the cultural adaptation control parameters, cultural background descriptions or adaptation prompts are added to the corresponding positions of cultural difference points with higher priority in the generated text, forming natural, fluent and culturally adapted target language text content.

[0054] In specific implementation, firstly, the source language text content and cultural adaptation control parameters are input into a multilingual generation model. The source language text content processed in the aforementioned steps, along with the set of cultural adaptation control parameters, are fed as input into a pre-trained multilingual generation model. The cultural adaptation control parameters are integrated into the model's input encoder in a structured form, such as as a set of additional conditional labels or a control vector. This allows the generation model to understand the semantics of the source text while simultaneously perceiving the specific transformation and adaptation instructions required for each cultural difference point in the text. Thus, the source text and external cultural constraints are co-encoded into an internal contextual representation that integrates semantics and culture, serving as an intermediate representation to be decoded, with accompanying cultural constraints. Secondly, during the decoding stage, the word probability distribution is adjusted according to the cultural adaptation control parameters. During the model's autoregressive decoding process, whenever the model needs to predict the next generated word, the system retrieves the cultural adaptation control parameters associated with the current generation position in real time. These parameters may exist in the form of a target culture preferred word list and its bias weights. These parameters apply a bias adjustment to the probability distribution of words in the model's original output that covers the entire vocabulary. For example, they can increase the probability weight of recommended words in the list or decrease the probability weight of inappropriate words. In this way, the model is dynamically guided in word selection at each step, thus prioritizing the generation of words that conform to the cultural habits, expression preferences, or specific taboo avoidance requirements of the target language, achieving cultural adaptation at the vocabulary level. Then, the complexity and word order of the generated sentences are dynamically adjusted according to the sentence structure adjustment intensity. During the decoding and generation of complete sentences, the instructions on the sentence structure adjustment intensity in the cultural adaptation control parameters are applied. The intensity parameter may be a quantified level value. By adjusting the attention mechanism preference or hidden state guidance of the generation model during decoding, the generated sentence structure is affected. For example, a higher adjustment intensity may prompt the model to generate sentences that are more in line with the typical word order of the target language, or actively split long sentences in the source language into short sentences that conform to the reading habits of the target language, or increase or decrease the frequency of certain modifying clauses, thereby achieving cultural adaptation and style adjustment at the syntactic structure level.Finally, based on the cultural annotation insertion strategy and processing priority, cultural background information is added to the generated text. After generating the main content of the text, the system inserts preset or dynamically generated cultural background information, equivalent explanations, or adaptation tips at appropriate positions in the text (such as at the end of sentences, paragraphs, or in the form of parenthetical explanations) according to the cultural annotation insertion strategy set for high-priority cultural differences in the cultural adaptation control parameters. This insertion operation is a post-processing step performed after the model decoding is completed. It aims to ensure the smooth transmission of core information while making necessary cultural differences transparent or providing supplementary explanations to help target language users understand. Through the above comprehensive dynamic adjustment from vocabulary to sentence structure to supplementary explanations, the system finally outputs a target language text that is semantically accurate, naturally expressed, and culturally appropriate.

[0055] On the other hand, in some embodiments, this application provides a multilingual cultural content generation system based on natural language understanding, with reference to... Figure 4 The figure is a schematic diagram of the structure of a multilingual cultural content generation system based on natural language understanding, according to some embodiments of this application. The multilingual cultural content generation system 400 based on natural language understanding includes: a receiving module 401, a processing module 402, and an execution module 403, which are described below: The receiving module 401 in this application is mainly used to receive source language text content and target language identifier; Processing module 402, in this application, is used to perform deep semantic analysis on the source language text content, and extract the semantic structure elements, cultural feature elements and contextual information of the source language text content; In this application, the processing module 402 is also used to perform semantic alignment processing on the semantic structural elements, cultural feature elements and contextual association information with the preset multilingual cross-cultural knowledge graph, and to perform fuzzy reasoning under rule constraints for semantic elements with cultural differences, and generate cultural adaptation control parameters to guide the content generation process. The execution module 403 in this application is mainly used to input the source language text content, the target language identifier and the cultural adaptation control parameters into a preset multilingual generation model. During the content generation process, the word selection and expression mode are dynamically adjusted according to the cultural adaptation control parameters, and the target language text content that conforms to the cultural context of the target language is output.

[0056] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described method for generating multilingual cultural content based on natural language understanding.

[0057] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a method for generating multilingual cultural content based on natural language understanding, according to some embodiments of this application. The method for generating multilingual cultural content based on natural language understanding in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0058] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0059] The communication bus 502 can be used to transmit information between the aforementioned components.

[0060] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0061] The memory 503 stores program code that executes the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The multilingual cultural content generation method based on natural language understanding in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0062] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0063] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0064] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0065] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for generating multilingual cultural content based on natural language understanding.

[0066] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0067] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for generating multilingual cultural content based on natural language understanding, characterized in that, include: Receive source language text content and target language identifier; Deep semantic analysis is performed on the source language text content to extract semantic structural elements, cultural feature elements and contextual information of the source language text content; The semantic structural elements, cultural feature elements, and contextual information are semantically aligned with a preset multilingual cross-cultural knowledge graph. Fuzzy reasoning under rule constraints is performed on semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process. The source language text content, target language identifier, and cultural adaptation control parameters are input into a preset multilingual generation model. During the content generation process, the vocabulary selection and expression are dynamically adjusted according to the cultural adaptation control parameters to output target language text content that conforms to the cultural context of the target language.

2. The method as described in claim 1, characterized in that, Deep semantic analysis of the source language text content, extracting semantic structural elements, cultural feature elements, and contextual information, specifically includes: The source language text content is segmented and tagged with parts of speech to construct a lexical sequence with syntactic dependencies; Based on a pre-trained semantic understanding model, semantic roles are labeled on the word sequence to identify predicates, arguments and their semantic relationships in the source language text content, thereby obtaining semantic structural elements; Cultural elements are identified in the vocabulary sequence, and cultural feature elements including cultural customs, historical allusions, social taboos, and rhetorical habits are extracted. These cultural feature elements are used to characterize the implicit cultural norms and constraints in the text and the expressive biases of different languages. The source language text content is subjected to discourse-level contextual dependency analysis to extract logical connections, referential relationships and topic coherence information between sentences, forming contextual association information.

3. The method as described in claim 1, characterized in that, The semantic alignment process between the semantic structural elements, cultural feature elements, and contextual information and the preset multilingual cross-cultural knowledge graph specifically includes: The entities and concepts in the semantic structure elements are mapped to corresponding nodes in a multilingual cross-cultural knowledge graph through cross-linguistic word vectors, and the mapping confidence is calculated. The cultural feature elements are mapped to the cultural entity nodes corresponding to the source language side in the multilingual cross-cultural knowledge graph, and then the cultural entity nodes are compared with the cultural norm nodes corresponding to the target language side to identify cultural differences and their degree of difference. The semantic alignment process is weighted according to the contextual information, and ambiguous or conflicting areas that occur during the semantic alignment process are marked and confidence-attenuated.

4. The method as described in claim 3, characterized in that, The construction of the multilingual cross-cultural knowledge graph specifically includes: Extract cultural entities, cultural relationships, and cultural constraint rules from a multilingual cultural corpus; The extracted cultural entities are subjected to cross-linguistic alignment and semantic normalization to construct a cultural concept network with source-target language mapping relationships; The cultural relationships and cultural constraint rules are associated with the corresponding cultural entity nodes in the cultural concept network, and the language affiliation and applicable scope of cultural constraints of each cultural entity are marked. The results are then represented and stored in the form of an attribute graph, forming a multilingual cross-cultural knowledge graph with semantic hierarchy and network association features.

5. The method as described in claim 1, characterized in that, The semantic alignment process also includes: Based on the mapping confidence, the semantic difference coefficient between the source language and the target language in terms of semantic features is calculated; Based on the degree of difference, the cultural difference coefficient between the source language and the target language in terms of cultural characteristics is calculated; Based on the semantic difference coefficient and the cultural difference coefficient, a comprehensive cultural difference degree is determined by fusion. The comprehensive cultural difference degree is compared with a preset difference threshold to filter out semantic elements with significant cultural differences, and each filtered semantic element is labeled with a cultural difference type and processing priority, wherein the processing priority is determined by the value of the comprehensive cultural difference degree.

6. The method as described in claim 1, characterized in that, For semantic elements with cultural differences, fuzzy reasoning under rule constraints is performed to generate cultural adaptation control parameters to guide the content generation process. These parameters specifically include: Based on the cultural rule base in the multilingual cross-cultural knowledge graph, we identify cultural transformation rules related to cultural differences and extract the applicable conditions and execution weights of the rules. Furthermore, based on the fuzzy reasoning mechanism, according to the cultural difference type and processing priority of the semantic elements, the contextual acceptability and cultural sensitivity of the semantic elements in the target language are comprehensively evaluated to generate cultural adaptation membership degree; Based on the processing priority, cultural adaptation membership degree, and preset conversion strategy, cultural adaptation control parameters, including vocabulary replacement tendency, sentence structure adjustment intensity, and cultural annotation insertion strategy, are dynamically generated.

7. The method as described in claim 1, characterized in that, During the content generation process, the selection of vocabulary and expression methods are dynamically adjusted according to the aforementioned cultural adaptation control parameters, and the output of target language text content that conforms to the target language cultural context specifically includes: Input the source language text content and the cultural adaptation control parameters into the multilingual generation model; During the decoding phase, the probability distribution of words is adjusted according to the cultural adaptation control parameters, and words that conform to the cultural habits of the target language are selected first. Based on the sentence structure adjustment intensity in the cultural adaptation control parameters, the complexity and word order of the generated sentences are dynamically adjusted. Based on the cultural annotation insertion strategy and processing priority in the cultural adaptation control parameters, cultural background descriptions or adaptation prompts are added to the corresponding positions of cultural difference points with higher priority in the generated text, forming natural, fluent and culturally adapted target language text content.

8. A multilingual cultural content generation system based on natural language understanding, characterized in that, include: The receiving module is used to receive source language text content and target language identifier; The processing module is used to perform deep semantic analysis on the source language text content, and extract the semantic structural elements, cultural feature elements and contextual information of the source language text content; The processing module is also used to perform semantic alignment processing on the semantic structural elements, cultural feature elements and contextual association information with the preset multilingual cross-cultural knowledge graph, and to perform fuzzy reasoning under rule constraints for semantic elements with cultural differences to generate cultural adaptation control parameters to guide the content generation process. The execution module is used to input the source language text content, the target language identifier, and the cultural adaptation control parameters into a preset multilingual generation model. During the content generation process, the module dynamically adjusts the vocabulary selection and expression methods according to the cultural adaptation control parameters, and outputs target language text content that conforms to the cultural context of the target language.

9. A computer device comprising a memory and a processor, the memory storing code, characterized in that, The processor is configured to acquire the code and execute the multilingual cultural content generation method based on natural language understanding as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the multilingual cultural content generation method based on natural language understanding as described in any one of claims 1 to 7.

Images