Intelligent adaptation method and system for drama script based on cultural interpretation and storage medium thereof

Abstract

This invention discloses a method, system, and storable medium for intelligent adaptation of theatrical scripts based on cultural translation. It relates to the field of cultural translation technology and achieves full-process standardization and intelligence in the cross-cultural adaptation of theatrical scripts through script data collection and standardized text preprocessing, three-layer cultural semantic translation based on a fine-tuned Transformer model of the script using a cross-cultural semantic library, directional reconstruction of narrative structure combining the target country's theatrical narrative paradigm and audience preference evolution characteristics, audience-specific adaptation verification and iterative optimization, generation of standardized adaptation guidance instructions throughout the entire process, and a systematic design of the entire chain based on audience behavior feedback after online dissemination, including paragraph association weighting and reverse optimization of core parameters. While fully preserving the core cultural connotations and artistic value of the original script, it significantly improves adaptation efficiency and cross-cultural acceptance, constructing a self-optimizing and continuously iterative intelligent adaptation closed-loop system for theatrical scripts.

Description

Technical Field

[0001] This invention relates to the field of cultural translation technology, and in particular to a method, system, and storable medium for intelligent adaptation of theatrical scripts based on cultural translation. Background Technology

[0002] Currently, cross-cultural adaptation of theatrical scripts still relies primarily on manual methods, supplemented by general machine translation tools. This approach suffers from several unresolved technical challenges: First, manual adaptation is extremely inefficient, lacks scalability, has long cycles, high costs, and its effectiveness heavily depends on the adapter's bilingual and bicultural expertise. Quality stability is uncontrollable, and issues such as cultural erosion and semantic ambiguity are common. Second, general translation technologies cannot adapt to the stylistic characteristics of theatrical scripts, only performing literal semantic conversion. They cannot handle culturally specific elements or contextualized emotional expressions, lack intelligent narrative structure reconstruction capabilities, and cannot adapt to the target country's theatrical paradigm and audience aesthetic preferences, resulting in low cross-cultural acceptance. Third, existing technologies lack a complete iterative loop, failing to continuously optimize based on audience feedback from real-world dissemination and failing to develop standardized, reusable adaptation solutions, thus hindering the large-scale cross-cultural dissemination of theatrical works. In short, the industry suffers from low efficiency in manual adaptation, insufficient accuracy in cultural translation, poor target audience suitability, and the lack of a long-term iterative optimization loop. To address the aforementioned technical deficiencies, a high-speed motor control system and method are proposed. Summary of the Invention

[0003] The purpose of this invention is to provide a three-layer cultural semantic translation through script data collection and standardized text preprocessing, a script Transformer model fine-tuned based on a cross-cultural semantic library, a narrative structure-oriented reconstruction combining the target country's dramatic narrative paradigm and audience preference evolution characteristics, audience-group adaptation verification and iterative optimization, generation of standardized adaptation guidance instructions throughout the entire process, and a systematic design of paragraph association weighting and core parameter reverse optimization based on audience behavior feedback after online dissemination. This achieves full-process standardization and intelligence in the cross-cultural adaptation of theatrical scripts, while fully preserving the core cultural connotation and artistic value of the original script, significantly improving adaptation efficiency and cross-cultural dissemination acceptance, and constructing a self-optimizing and continuously iterative intelligent adaptation closed-loop system for theatrical scripts.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent adaptation of theatrical scripts based on cultural translation includes the following steps: The process involves collecting the original script text, corresponding cultural background data, and theatrical aesthetic data from the target country. The collected original script text is then converted and standardized to generate a preprocessed data package. The script's Transformer model is loaded and fine-tuned based on cross-cultural semantic database data. The script text from the preprocessed data package is input into the fine-tuned model, sequentially performing literal semantic translation, cultural semantic translation, and context-adaptive translation to generate a cross-cultural translated script text. Based on the target country's theatrical narrative paradigm and audience preference evolution characteristics, the translated script text undergoes scene optimization, dialogue optimization, core conflict and character arc optimization, and narrative rhythm optimization to generate a reconstructed script text adapted to the target audience. Next, audience adaptation verification is carried out for the reconstructed script text by group, and the script is iteratively optimized based on the verification results, generating corresponding audience adaptation verification reports; the data of the entire script adaptation process is integrated with the audience adaptation verification reports to generate standardized drama script adaptation guidance instructions; all audience behavior feedback data after the adapted script is launched and disseminated online are collected, data preprocessing and paragraph association weighting are completed, and the core parameters of cultural semantic translation and narrative structure reconstruction are optimized in reverse based on the weighting results, forming a closed loop of full-process iterative adaptation of scripts.

[0005] Furthermore, the original text is first obtained through high-definition scanning and text extraction. Simultaneously, cultural background data such as the era, social culture, etiquette and customs, and language habits corresponding to the original text are collected, as well as theatrical aesthetic data such as the age, cultural background, preference for drama genres, preference for dialogue style, and preference for narrative rhythm of the target audience. Then, a full-process preprocessing operation is performed on the original text.

[0006] Furthermore, the text preprocessing operation achieves the required rate of redundant information removal and noise reduction accuracy to meet the constraints. During the preprocessing process, the text undergoes a JSON-formatted structure conversion, and mapping fields are set that include script ID, scene number, scene description, characters, lines, entity annotations, cultural background notes, and paragraph unique IDs. The paragraph unique IDs enable accurate matching between subsequent audience behavior data and text paragraphs.

[0007] Furthermore, the script Transformer model is fine-tuned by loading cross-cultural semantic database data. The fine-tuning process uses the cross-entropy loss function to calculate the model loss until the loss value falls into the preset range. Then, the preprocessed script text is input into the model to complete three layers of semantic translation in sequence: literal semantic translation, cultural semantic translation, and context-adaptive translation.

[0008] Furthermore, during the translation process, a semantic hierarchical encoder is introduced into the script Transformer model to decompose each cultural proprietary item into three levels: surface semantics, cultural semantics, and emotional semantics. Semantic adaptive coefficients are set for the corresponding levels, and attention weight allocation during the translation process is dynamically adjusted in combination with temporal feedback weight coefficients. At the same time, matching thresholds for the three levels of semantics are set. When the semantic matching degree of any level is lower than the threshold, manual review or supplementary cultural annotations are automatically triggered.

[0009] Furthermore, the core quantitative parameters of the target country's dramatic narrative structure are extracted and their average values ​​are calculated as a reconstruction reference standard. Hidden Markov models are used to model the evolution of audience preferences. Clustering algorithms are used to cluster the preferences of audiences of different age groups. Finally, based on the reference standard and clustering results, the translated script text is optimized for group-specific adaptation in terms of scenes, lines, core conflicts, character arcs, and narrative rhythm.

[0010] Furthermore, audience behavior data after the adapted script is released is collected and preprocessed to construct a "paragraph ID-behavior data" mapping table; then, a semantic association graph of script paragraphs is constructed, and graph convolutional networks are used to aggregate the features of adjacent paragraphs, calculate the comprehensive weight value of each paragraph, and divide the optimization priority; finally, based on the priority results, the semantic mapping weight and narrative structure parameters of high optimization priority paragraphs are dynamically adjusted, and the adaptation guidance instructions are updated synchronously to form a closed loop of full-process iteration.

[0011] A drama script intelligent adaptation system based on cultural translation includes a terminal device, a script data acquisition module, a text preprocessing module, a cultural semantic translation module, a narrative structure reconstruction module, an audience adaptation verification module, a guidance instruction generation module, an offline audience optimization module, a data storage and retrieval module, and a storage medium. The terminal equipment includes a local adaptation terminal (TA), a cloud computing terminal (TB), and a data acquisition terminal (TC), which are connected via a dedicated network link. The script data acquisition module is embedded in the data acquisition terminal TC and is connected to the text preprocessing module via signal; the text preprocessing module is deployed on the local adaptation terminal TA and is connected to the script data acquisition module, the cultural semantic translation module, and the data storage and retrieval module via signal, respectively. The cultural semantic translation module, narrative structure reconstruction module, and offline audience optimization module are all deployed on the cloud computing terminal TB. The cultural semantic translation module is connected to the text preprocessing module, narrative structure reconstruction module, data storage and retrieval module, and offline audience optimization module. The narrative structure reconstruction module is connected to the cultural semantic translation module, audience adaptation verification module, guidance instruction generation module, and offline audience optimization module. The offline audience optimization module is connected to the data storage and retrieval module, cultural semantic translation module, and narrative structure reconstruction module. The audience adaptation verification module is deployed on the local adaptation terminal TA and is connected to the narrative structure reconstruction module, the guidance instruction generation module, and the data storage and retrieval module by signal; the guidance instruction generation module is connected to the audience adaptation verification module and the offline audience optimization module by signal. The data storage and retrieval module is connected to each functional module and the storage medium via signals. The storage medium includes local storage medium and cloud storage medium, which are used for dual backup and secure storage of the system's full data.

[0012] Furthermore, both local and cloud-based storage media adopt an edge-cloud collaborative distributed storage architecture, with dedicated storage partitions divided according to data types. The system has a pre-set initial database, including a cross-cultural semantic database, cultural narrative paradigm data, target country audience preference data, and audience behavior feedback rule database, providing data support for the entire system operation.

[0013] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the intelligent adaptation method for a theatrical script based on cultural translation as described in any one of claims 1 to 7.

[0014] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This invention transforms the traditional script adaptation process, which relies on human experience, into a quantifiable and reproducible standardized intelligent process through a systematic design that integrates script data collection, text preprocessing, cultural semantic translation, narrative structure reconstruction, audience adaptation verification, and feedback optimization. This significantly reduces the labor and time costs of cross-cultural script adaptation while avoiding subjective biases inherent in human adaptation and ensuring the consistency and stability of script adaptation quality across multiple batches. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings. Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the modules of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example: like Figure 1 As shown, a method for intelligent adaptation of theatrical scripts based on cultural translation is presented. S1: Script Data Acquisition and Text Preprocessing S1-1: Script data collection, the specific steps are as follows: Sa1: The preset collection scene is a professional drama archive, the collection object is the original script, the original script text is marked as T0, and the text encoding is UTF-8 to avoid Chinese character garbled characters; Sa2: Start the script data acquisition module, scan the original paper script through the high-definition scanning module of the data acquisition terminal TC, convert it into electronic text, and extract the text content, annotation content, and cultural background description from the electronic text through the text acquisition tool. Mark the acquired original text as T1, which includes text volume, plot information, scene information, and character information. Sa3: Collect cultural background data from the original script, including historical context, social culture, etiquette, customs, and language habits, and mark it as T2. This data will be used to supplement the context during subsequent cultural semantic translation. Collect and store theatrical aesthetic data of the target country's culture, and mark it as T3. This data will be used for subsequent audience adaptation verification. It is important to note that theatrical aesthetic data of the target country's culture is applicable to both theatrical translation and reverse translation, such as our own aatrical aesthetic data. Theatrical aesthetic data includes audience age, cultural background, preference for theatrical genres, preference for dialogue style, preference for narrative rhythm, etc.

[0018] Sa4: Store the collected raw text T1, cultural background data T2, and theatrical aesthetic data T3, and send the raw text T1, cultural background data T2, and theatrical aesthetic data T3 to the text preprocessing module.

[0019] S1-2: Text preprocessing, the specific operations are as follows: Sb1: Text denoising process, removing redundant information from T1, such as duplicate comments, incorrectly formatted characters, and blank lines. The denoised text is marked as T1-1. Quantitative denoising standards include a redundant information removal rate ≥98% and a denoising accuracy rate ≥99.5%. The denoising rate calculation formula is as follows: in To quantify the effect of removing redundant information in text denoising rate; is the number of redundant information characters before denoising; is the number of remaining redundant information characters after denoising.

[0020] Among them is the text denoising accuracy rate, quantifying the retention effect of effective content; is the number of effective characters after denoising; is the number of effective characters before denoising.

[0021] Among them, the number of redundant information characters scans the text through Python code to count the total number of blank characters, repeated comments, and character strings with incorrect formats; the number of effective characters is the core text content of the script, such as the number of characters in lines, narration, and scene descriptions; Sb2: Text word segmentation processing. Call the jieba word segmentation tool to segment T1-1, remove stop words, such as words without actual semantics like "de, di, de, zhi, hu, zhe, ye", etc., and mark the segmented text as T1-2. According to the word segmentation accuracy rate calculation formula as follows: Among them is the text word segmentation accuracy rate; is the number of correctly segmented words; is the total number of words in the text. Among them, the number of correctly segmented words is verified through manual sampling. Randomly extract the word segmentation results and manually judge the number of correctly segmented words to ensure that it initially meets the preset standards; the removal of stop words uses a custom stop word list, and the removal rate calculation formula is as follows: Among them is the stop word removal rate; is the number of successfully removed stop words; is the total number of stop words in the text; Sb3: Part-of-speech tagging and named entity recognition. Call the jieba word segmentation tool combined with a lightweight version of the script Transformer model to perform part-of-speech tagging on the words in T1-2. The tagging categories include 10 categories such as nouns, verbs, adjectives, adverbs, and culturally specific nouns to identify core entities such as character names, place names, and culturally specific items in the script. The calculation formula is as follows: Among them is the named entity recognition accuracy rate; is the number of correctly recognized entities; is the total number of entities recognized by the model.

[0022] Among them To improve the recall rate of named entity recognition; The number of correctly identified entities; This represents the total number of entities that actually exist in the text.

[0023] Sb4: Structured transformation, converting T1-2 into JSON format, writing JSON field mapping code, i.e. field mapping logic, so that it corresponds one-to-one with the core information of the original script. Fields include script ID, scene number, scene description, characters, lines, entity annotation, cultural background notes, paragraph unique ID, etc. A new paragraph unique ID is added for accurate matching of subsequent audience behavior data and text paragraphs. The structured text is marked as T1-3. Sb5: Text standardization processing, which unifies the text content in T1-3 to Simplified Chinese, standardizes punctuation, removes variant characters and typos, and marks the standardized text as T1-4. The calculation formula is as follows: in Standardize the accuracy of the text; The number of characters in the standardized correct text; This represents the total number of characters in the text before standardization.

[0024] Sb6: Generate preprocessed data packets from T1-4, T2, and T3, and then synchronously store and send them to the cultural semantic translation module through the data storage and retrieval module.

[0025] S2: Cultural Semantic Translation S2-1: Loading and fine-tuning the Transformer model in the script, the specific steps are as follows: Sa1: Call the script Transformer model, load the preset cross-cultural semantic library data to fine-tune the model. The fine-tuning parameters are the same as the initialization parameters, including the hidden layer dimension, number of attention heads, number of encoder layers, number of decoder layers, learning rate, batch size, dropout coefficient, and number of iterations. During the fine-tuning process, preprocessed data packets from any two countries are used as training samples. The loss model is calculated until the loss model value is within the preset loss range, thus realizing the model training and learning of mutual translation construction. Sa2: The specific calculation formula for the loss model is as follows: in The cross-entropy loss model value quantifies the deviation between the model's prediction and the true label. For the first The true label of each training sample takes a value of 0 or 1, where 1 indicates semantic matching and 0 indicates semantic mismatch; For the model to predict the first The probability of semantic matching for each sample; This represents the total number of training samples; This is the summation operator; It is the natural logarithm function.

[0026] S2-2: The specific steps for cultural semantic translation are as follows: Sb1: The cultural semantic translation module receives the preprocessed T1-4, extracts the core text content, such as lines, narration, scene descriptions and entity E, marks it as T3-1, inputs T3-1 into the fine-tuned script Transformer model, and starts semantic translation. Sb2: Semantic translation is divided into three levels: literal semantic translation, cultural semantic translation, and context-adaptive translation. Each level is supplemented with quantitative judgment criteria, as follows: Literal translation: Translate the Chinese text in T3-1 sentence by sentence into English, ensuring accurate literal meaning. The translation accuracy is calculated using the following formula: in For the accuracy of literal semantic translation; The number of sentences that accurately translate the literal meaning; This represents the total number of sentences translated.

[0027] Cultural semantic translation: For culturally specific items and ambiguous words in T3-1, cultural adaptation translation is performed by combining the cross-cultural semantic database D1 and cultural background data T2 to avoid semantic misunderstandings. The calculation formula is as follows: in For the accuracy of cultural semantic translation; The number of entities / vocabularies accurately translated into cultural semantics; This represents the total number of entities / vocabularies that require cultural translation.

[0028] ③ Context-Adapted Translation: Based on the original script's plot context and characters' emotions, adjust the tone and sentence structure of the translated dialogue to match the target country's theatrical style, ensuring conciseness, impact, and appropriateness to the characters' identities. The calculation formula is as follows: in For contextual fit; The number of sentences that fit the context; This represents the total number of sentences translated.

[0029] Sb3: During the translation process, an attention mechanism is used to calculate the semantic similarity of the text. It supports dynamic adjustment of attention allocation based on feedback weights. The formula for calculating attention weights is as follows: in For the first The input word and the first The attention weights for each output word, with values ​​ranging from [0,1]; This is the weighting coefficient for audience feedback, with an initial value of 1, which can be dynamically adjusted based on the weighting results of the feedback. For the first The input word and the first The semantic similarity score of each output word; For the first The input word and the first The semantic similarity score of each output word; For indexing all output words Summation; : using natural constant An exponential function with base 1. Where attention weights are... The value range is [0,1]. The higher the weight, the stronger the semantic relationship between the two words, and the more important it is to match during translation. Sb4: Generate the translated script text, marked as T3-2. The translated text retains the core plot, character relationships and cultural connotations of the original script, while adapting to the language habits of the counterpart. Each paragraph retains a unique ID for subsequent feedback data matching. Sb5: The translated T3-2 is synchronously stored and sent to the narrative structure reconstruction module through the data storage and retrieval module.

[0030] To improve the translation accuracy of cultural proprietary items, a semantic hierarchical encoder is introduced into the script Transformer model, decomposing each cultural proprietary item into three semantic layers: surface semantics (literal translation), cultural semantics (rituals and customs), and affective semantics (tone / emotion). The labeling hierarchy adaptive coefficient is... The preset values ​​correspond to the attention enhancement weights of the three semantic layers.

[0031] Introducing the hierarchical coefficient and the dynamic factor of feedback weight, the formula is as follows: Here For the first The input word and the first Attention weights for each output word; For the first The input word and the first The semantic similarity score of each output word; For the first The input word and the first The semantic similarity score of each output word; These are semantic level adaptive coefficients, which can be preset to correspond to surface semantics, cultural semantics, and emotional semantics; This is the weighting coefficient for time-series feedback. The initial value can be set to 1, and it can be dynamically updated according to the results of the offline feedback optimization module. For indexing all output words Summation; For the natural constant An exponential function with a base of 0.5 is used. Hierarchical semantic matching thresholds and constraints are then set. When the surface semantic threshold, cultural semantic threshold, and emotional semantic threshold fall below the threshold, manual review or supplementary cultural annotations are automatically triggered. This enhances the accuracy of cultural translation and the adaptability of emotional semantics.

[0032] S3: Narrative Structure Restructuring S3-1: Analysis of cultural narrative paradigms, the specific steps are as follows: Sa1: Call the cultural narrative paradigm data D2 in the narrative structure reconstruction module to extract the core quantitative parameters of the target country's drama narrative structure, and label them as P = {P1, P2, P3, P4, P5}, where: P1 is the number of scenes in a single drama, P2 is the duration of a single scene, P3 is the length of a single line of dialogue, P4 is the number of core conflicts, and P5 is the number of character arc transformation nodes. Sa2: Calculates the average parameter value of the target country's dramatic narrative paradigm as a reference standard for narrative structure reconstruction. The calculation formula is as follows: in The average vector of the core parameters of the target country's dramatic narrative paradigm; This represents the average number of scenes in a single play. This represents the average duration of a single scene. This represents the average length of a single line of dialogue. This represents the average number of core conflicts. This represents the average number of nodes in the character's arc transition.

[0033] S3-2: Running the time-series preference evolution clustering model to analyze audience preferences in the target country. Sb1: Utilize the target country's audience preference data D3, extract the core quantitative indicators of audience preference, and label them as X = {X1, X2, X3, X4, X5}. Simultaneously, divide D3 into T time segments according to time windows, and construct a dynamic preference vector sequence Xt.

[0034] Sb2: Standardize the preference data for each time slice using the same standardization formula as before, eliminating the influence of dimensions.

[0035] Sb3: A Hidden Markov Model (HMM) is used to model the evolution of audience preferences. The number of states is k=3, corresponding to the three audience groups: young, middle-aged, and elderly. The observed values ​​are standardized preference vectors. The model parameters λ = (π, A, B) are estimated using the Baum-Welch model, where π is the initial state distribution, A is the state transition probability matrix, and B is the observation probability matrix.

[0036] Sb4: The dynamic update formula for cluster centers is as follows: in For the first The audience group in the first Cluster centers for each time window; For the first The audience group in the first Cluster centers updated within a time window; The learning rate for the clustering model is fixed at 0.1. For the first Audience preference sample vectors within a time window; For the first Within the time window belonging to the first A sample set of classes; For the first Within the time window belonging to the first The number of samples in each class; This is the time decay factor; For the first Within the first time window Summing all samples of the class.

[0037] Formula for calculating time decay factor: in This is the time decay factor; The time window number; The decay half-life is fixed at 5 time windows; For the natural constant An exponential function with base 0.

[0038] Sb5: Iterate until the rate of change of cluster centers is less than a threshold or the maximum number of iterations is reached. This results in three dynamically evolving sequences of audience center locations. The average preference parameters for each group are extracted for real-time adaptation of the narrative structure reconstruction.

[0039] S3-3: Practical exercises on reconstructing narrative structure, with specific steps as follows: Sc1: Scene optimization, integrating the core scenes of the original script into target scenes. The integration logic is "merging related scenes and deleting redundant scenes". The optimized target scenes are marked, and the duration of each scene is controlled. The scene duration calculation formula is as follows: in Total duration of a single scene, in minutes; This refers to the total number of words in the dialogue for this scene. The average speaking speed of theatrical dialogue; This refers to the duration of the action performance in this scene; Sc2: Dialogue optimization. The translated dialogue is broken down into shorter or longer lines based on a preset number of English words, thus constraining the length of individual lines. The calculation formula is as follows: in Optimize the adaptation of the dialogue; The number of lines that meet the length standard; This represents the total number of lines in the script.

[0040] By optimizing dialogue to suit audience preferences, dialogue optimized for younger audiences is more concise, such as an average of 18 English words / sentences; dialogue optimized for older audiences is slightly longer, such as an average of 25 English words / sentences; and dialogue optimized for middle-aged audiences is of moderate length, such as an average of 22 English words / sentences, thus achieving a biased optimization of dialogue. Sc3: Core conflict and character arc optimization: The core conflict of the original script is retained, and secondary conflicts are added to conform to the narrative characteristics of the target drama "focusing on the core conflict"; In terms of character arc optimization, arc transformation node markers are set for each core character, with each node corresponding to clear emotional changes and behavioral changes. At the same time, in combination with audience preferences, young audiences focus on the rapid changes in character arcs, while older audiences focus on the gradual changes in character arcs. Sc4: Narrative pacing optimization. Based on audience preferences, the narrative pacing of the script is adjusted. For younger audiences, the pacing is tighter, such as a small conflict every 5 minutes; for middle-aged audiences, the pacing is slower, such as a small conflict every 8 minutes; and for older audiences, the pacing is more relaxed, such as a small conflict every 10 minutes. The formula for calculating narrative pacing suitability is as follows: in Narrative pacing fit The number of paragraphs should be adjusted to match the audience's preferred pace. This represents the total number of paragraphs in the script.

[0041] Sc5: Generates the reconstructed script text, marked as T4, which includes scenes, core conflicts, and characters. The reconstructed script retains the core cultural connotations and plot, while also adapting to the target country's dramatic narrative paradigm and the preferences of different audience groups. Each paragraph retains a unique ID and timestamp, and supports dynamic adjustment of paragraph parameters based on feedback weights. Sc6: The reconstructed script text T4 is synchronously stored and sent to the audience adaptation verification module through the data storage and retrieval module.

[0042] S4: Audience Adaptation Validation and Iterative Optimization S4-1: Audience fit verification, the specific steps are as follows: Sa1: Select the target audience in the predetermined country and divide them into 3 clusters. Have them watch the reconstructed script text T4 and fill in the audience acceptance rating form. The rating indicators include plot suitability, cultural understanding, emotional resonance, dialogue fluency, and narrative rhythm suitability. Each indicator has a rating range of 0-10 points. Then, perform weight normalization to calculate the total score. For example, the total score = plot suitability × 0.25 + cultural understanding × 0.25 + emotional resonance × 0.2 + dialogue fluency × 0.15 + narrative rhythm suitability × 0.15. When the total score is ≥ 7, it is considered a suitable candidate. Sa2: Calculates the average rating for each audience group. The formula for calculating the average rating is as follows: in Average acceptance rating for the corresponding target audience; The number of participants in this audience group; The first in this group Total acceptance score of the target audience; To all within this group The sum of ratings from the top audience members.

[0043] Sa3: Next, calculate the audience suitability rate. The formula for calculating the suitability rate is as follows: To adjust the acceptance rate to suit the audience; To suit the appropriate number of audience members; This represents the total number of participants.

[0044] S4-2: Iterative optimization, the specific steps are as follows: Sb1: For issues that did not meet the standards, optimization suggestions are generated based on their preference parameters: ① Dialogue optimization: Adjust the average length of dialogue for the audiences that did not meet the standards, add cultural background annotations, and improve cultural understanding; ② Narrative pacing optimization: Adjust the pacing of plot segments for the audiences, slow down the narrative pace, and improve pacing adaptability; ③ Cultural element adaptation: Simplify some complex cultural elements in the original script, add brief cultural explanations, and improve cultural acceptance. Sb2: Mark the optimized script as T4', select the corresponding audience again for verification, calculate the average score, and verify the average score of the optimized audience and the overall audience suitability rate until the constraint conditions are met. Sb3: Analyze the verification results, generate an audience adaptation verification report, and mark it as R. The report includes detailed scores for each audience group, optimization suggestions, and comparison data before and after optimization. Send the report to the guidance instruction generation module. Sb4: Store and publish the optimized T4' and verification report R to overseas drama reading platforms to collect audience behavior feedback data.

[0045] S5: Instruction Generation and Core Data Storage The guidance instruction generation module receives the audience adaptation verification report R, integrates the narrative structure reconstruction process, optimization suggestions, and verification results, and generates standardized script adaptation guidance instructions, marked as Z. The instructions include text preprocessing details, key points of cultural semantic translation, narrative structure reconstruction parameters, and audience adaptation optimization schemes, which are used for subsequent script adaptation decisions in multiple batches. The guidance instruction Z is stored and displayed.

[0046] S6: Cross-audience communication feedback weight optimization and model iteration closed loop S6-1: Audience behavior feedback data collection and preprocessing, the specific steps are as follows: Sa1: The offline audience optimization module collects full audience behavior data within a preset period after the adapted script T4' is released on overseas platforms via API interface, including single paragraph reading time, reading completion rate, paragraph annotation / sharing data, comment time distribution, and paragraph ID associated with comment content, and marks the original feedback data as F0; Sa2: Deduplication and noise reduction are performed on F0 to remove invalid and abnormal data. The preprocessed valid feedback data is marked as F1. The efficiency calculation formula is as follows: in Efficient data feedback to the audience; To ensure the number of valid feedback data entries; This represents the total number of original feedback data entries.

[0047] Sa3: Accurately match the effective feedback data F1 with the unique ID of the script paragraph to form a "paragraph ID-behavioral data" mapping table, which is used for subsequent weight assignment.

[0048] S6-2: Paragraph Association Weighting Based on Graph Neural Networks Sb1: Construct a semantic association graph G = (V, E) for paragraphs, where node V represents a paragraph in the script, i.e., each paragraph has a unique ID corresponding to a node, and edge E connects paragraph pairs with semantic similarity higher than a threshold. Semantic similarity is obtained by calculating the cosine similarity of paragraph vectors using a pre-trained BERT model.

[0049] Sb2: Call the audience behavior feedback rule base D5 to extract the initial behavior feature vector for each paragraph. The data within the initial behavioral feature vector can be set. A graph convolutional network is used with two convolutional layers to aggregate the features of adjacent segments, as shown in the following formula: in Paragraph node In the Feature vectors of layer graph convolution; Paragraph node In the Feature vectors updated by layer graph convolution; It is the ReLU activation function; Paragraph node The set of adjacent segments; Paragraph node The degree; Adjacent paragraph nodes The degree; Adjacent paragraph nodes In the Feature vectors of layer graph convolution; : No. Learnable weight matrix for layer graph convolution; For nodes All adjacent nodes Sum.

[0050] Sb3: After aggregating the features of adjacent paragraphs, the final feature representation of each paragraph is obtained. Then, it is mapped to the corresponding level weight probability distribution through a fully connected layer. The formula for calculating the paragraph comprehensive weight is as follows: in This represents the overall weighting of the script segments. Weighting levels; Based on the final features of the paragraph The model predicts that this paragraph belongs to the first... The probability of level weight; This is for summing over the weight levels. The probability that a paragraph predicted by the model belongs to the r-th weight.

[0051] Sb4: Based on the weight assignment results, classify the paragraphs: ① High optimization priority paragraphs, such as... ② Optimize priority paragraphs, such as ③ Low optimization priority paragraphs, such as Generate a weight optimization report. This information is sent to the relevant modules for adaptation and reverse optimization.

[0052] S6-3: Backward adaptation decision-making and model iteration, the specific operations are as follows: Sc1: Cultural semantic translation module optimization report based on weights Dynamically adjust the semantic mapping weights of high-priority paragraphs. ,for example paragraphs, The numerical value improves the semantic translation accuracy of the paragraph, supplements cultural annotations, and eliminates semantic ambiguity; Sc2: Narrative Structure Reconstruction Module Optimization Report Based on Weights Adjust the narrative rhythm and dialogue length of high-priority paragraphs. The paragraphs were re-matched with audience preference parameters, the narrative structure was optimized, and the risk of audience churn was reduced; Sc3: The guidance instruction generation module integrates the weight optimization results, updates the script adaptation guidance instruction Z, supplements the key points of weight optimization, and forms a closed loop for the entire adaptation process; Sc4: Mark the optimized secondary adapted script as T4'' and store iterative optimization synchronously.

[0053] A Drama Script Intelligent Adaptation System Based on Cultural Translation and Its Storage Medium like Figure 2 As shown: I. System composition and connection relationship This invention Theatrical script based on cultural translation The system can be adapted, including a script data acquisition module, a text preprocessing module, a cultural semantic translation module, a narrative structure reconstruction module, an audience adaptation verification module, a guidance instruction generation module, an offline audience optimization module, a data storage and retrieval module, storage media, and terminal devices. The signal connections between each module and device are as follows. All modules adopt a modular design, can operate independently and work together to ensure a smooth and scalable process. The terminal equipment includes a local adaptation terminal (TA), a cloud computing terminal (TB), and a data acquisition terminal (TC). The three are connected via a dedicated network link to achieve real-time data transmission, collaborative model operation, and synchronous output of results. The script data acquisition module is embedded in the data acquisition terminal TC and is connected to the text preprocessing module. It is used to collect the original script text data, cultural background data, and cultural audience data, and transmit them to the text preprocessing module for pre-processing. The text preprocessing module is deployed on the local adaptation terminal TA and is connected to the script data acquisition module, cultural semantic translation module, and data storage and retrieval module. It is used to denoise, segment, and structurally transform the acquired script data to provide standardized data for subsequent cultural translation and narrative reconstruction. The cultural semantic translation module is deployed on the cloud computing terminal TB and is connected to the text preprocessing module, narrative structure reconstruction module, data storage and retrieval module, and offline audience optimization module. The core is equipped with the Transformer model to achieve accurate cultural semantic translation of the original script text. It sets up a dynamic adjustment interface for semantic mapping weights and supports optimizing attention weight allocation based on feedback weights to solve cross-cultural semantic ambiguity problems. The narrative structure reconstruction module is deployed on the cloud computing terminal TB and is connected to the cultural semantic translation module, the audience adaptation verification module, the guidance instruction generation module, and the offline audience optimization module. The core is equipped with a K-means clustering model, which is used to reconstruct the narrative structure of the translated script by combining cultural narrative paradigms and audience preferences. It sets up a narrative parameter feedback adaptation interface and supports adjusting the reconstruction parameters according to the feedback weight. The audience adaptation verification module is deployed on the local adaptation terminal TA and is connected to the narrative structure reconstruction module, the guidance instruction generation module, and the data storage and retrieval module. It is used to verify the audience acceptance of the reconstructed script and generate adaptation optimization suggestions. The offline audience optimization module is deployed on the cloud computing terminal TB and is connected to the cultural semantic translation module, the narrative structure reconstruction module, and the data storage and retrieval module. The core is equipped with the XGBoost multi-class weighting model, which is used to collect audience behavior data of the adapted script across platforms, complete multi-dimensional feedback weighting, and drive semantic mapping and narrative reconstruction parameter optimization in reverse to form an adaptation closed loop. The instruction generation module is connected to the audience adaptation verification module and the offline audience optimization module to receive verification results and weight optimization suggestions, and to generate standardized script adaptation instructions for staff to refer to and adjust. The data storage and retrieval module is connected to each core module and the storage medium signal to realize the storage, retrieval and synchronization of script data, processing results, models, feedback weight data and guidance instructions; The storage media includes local storage medium SD1 (1TB NVMe solid-state drive) and cloud storage medium SD2 (200GB Alibaba Cloud OSS object storage), which are connected to the data storage and retrieval module to store all system data, multi-model library files, and instruction files, achieving dual backup and secure storage, and ensuring that the data is traceable and retrievable.

[0054] H1: System Initialization and Deployment of Storage Media H1-1: Terminal device and software environment initialization, the specific steps are as follows: Hb1: Preset terminal device parameters, marking local adaptation terminals as TA, cloud computing terminals as TB, and data acquisition terminals as TC; Hb2: The data acquisition terminal TC has a built-in high-definition scanning module and text acquisition tool. The parameter settings are based on the following: the data acquisition terminal needs to acquire scanned copies of the script, electronic text, and cultural background information, and store the acquired raw data; the three are connected through a dedicated network to ensure that the data transmission can achieve rapid transmission of script text and model parameters. Hb4: Initialization of a multi-model, non-uniform model library, constructing three independent and non-uniform models: a text semantic model, an audience clustering model, and a feedback weight calculation model. H1-2: Storage media initialization and deployment, the specific steps are as follows: Hc1: Marks the local storage medium as SD1 and the cloud storage medium as SD2. Both are connected to the data storage and retrieval module to work together to achieve dual data backup; the storage medium is edge-cloud coordinated distributed storage. Hc2: SD1 is divided into 5 storage partitions: original script data partition, preprocessed data partition, adapted script data partition, feedback weight data partition, and guidance instructions and log partition. These partitions are used to store the original script text, preprocessed standardized data, adapted script files, audience feedback weight data, guidance instructions and system operation logs, respectively. The newly added feedback weight data partition can realize the classified storage of audience behavior data, which facilitates the quick retrieval of optimization data. Hc3: SD2 creates 6 storage partitions, corresponding to original script storage, cultural semantic library storage, multi-model library storage, backup data storage, feedback weight data storage, and user data storage, respectively. It also sets read and write permissions only for the system terminal and has a preset mechanism to synchronize data with SD1 once every periodic time. Partitioning storage by data type can improve data query and retrieval efficiency and avoid data chaos; read and write permissions only for the system terminal can ensure the security of core data, such as models and cultural semantic libraries, and prevent data leakage; the preset data synchronization once every periodic time can balance data security and system operation efficiency, avoid excessively frequent synchronization to consume network bandwidth, and ensure data consistency between local and cloud to prevent data loss. Hc4: Write data read / write adaptation code to ensure compatibility between local SSDs and cloud storage. This is based on the fact that different terminals may have different read / write protocols for their storage media; the adaptation code unifies the data read / write format, preventing data read failures and format errors. A data compression storage mechanism is also implemented to reduce data storage space and improve data transmission and storage efficiency. Furthermore, the ZIP format offers strong compatibility, allowing for normal decompression and reading on different terminals and storage media. Finally, a storage media anomaly detection program is included to monitor storage capacity and read / write speed in real time, promptly detecting storage media failures, preventing data loss, and ensuring normal system operation.

[0055] H1-3: Initialization of the cultural semantic database and related data, the specific operations are as follows: Hd1: Mark cross-cultural semantic database data as D1, cultural narrative paradigm data as D2, target country audience preference data as D3, original script data as D4, and audience behavior feedback rule base data as D5; preset corresponding effective data volume, rules, and behavior-weight mapping standards.

[0056] Hd2: D1 covers the Chinese target country's drama culture-specific items, semantically ambiguous words, cultural context descriptions, etc. Each data entry contains information such as "original cultural semantics, cultural semantics, semantic ambiguity resolution, cultural context supplement, and basic mapping weights", which are used for training the script Transformer model and cultural semantic translation. All data sources have obtained legal usage authorization and have been manually verified in advance to ensure that the data is free from copyright infringement and semantic bias. Hd3:D2 covers paradigmatic data on the narrative structure, scene setting, dialogue style, and character arc development of the target country's dramas. Each data point includes quantitative parameters such as "narrative structure characteristics, number of scenes, dialogue length, and character arc nodes" for narrative structure reconstruction. The data comes from the target country's classic drama script database and drama creation research literature, and is manually annotated and verified by drama creation experts. Hd4: D3 covers theatrical aesthetic preferences of audiences of different age groups in the target country, including audience age, cultural background, preference for theatrical genre, preference for dialogue style, preference for narrative rhythm, etc. Each data point includes a quantitative score, such as 0-10, for K-means clustering and audience fit verification. The data comes from theatrical audience survey platform in the target country and theater performance feedback data to ensure the representativeness and authenticity of the data. Hd5: D5 covers audience behavior-weighting mapping rules for cross-platform dissemination of adapted scripts, clearly defining weighting standards for weight assignment in the XGBoost model. For example, the rules are as follows: ① Level 1 weight, weight W1=1: Single paragraph reading time < 3s, judged as content to be skimmed without interest; ② Level 2 weight, weight W2=3: Single paragraph without annotations, shares, or comments, judged as low-attention content; ③ Level 3 weight, weight W3=5: Script reading completion rate < 50%, judged as content with high churn risk; ④ Level 4 weight, weight W4=8: Single paragraph dwell time > 10s, annotation volume exceeding 20%, judged as high-attention content; ⑤ Level 5 weight, weight W5=10: Within 6-24 hours after script release, the number of comments on a single paragraph accounts for more than 60% of the total number of comments on the entire script, judged as core high-feedback content. Hd6: Import D1, D2, D3, and D5 into the corresponding modules of the cloud computing terminal TB. After model initialization and training, store them in the corresponding storage bucket of SD2. Import the original script data of D4 into the data acquisition terminal TC. After preliminary processing, store it in the original script data partition of SD1 and the original script storage bucket of SD2 for the above-mentioned text preprocessing.

[0057] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor, comprising the steps of a method and system for intelligent adaptation of theatrical scripts based on cultural translation.

[0058] This invention focuses on intelligent script adaptation, specifically addressing cultural translation, narrative structure reconstruction, and audience adaptation optimization at the script text level. It includes a subsystem for optimizing offline audience feedback weights during cross-platform script dissemination. Through a complete process encompassing text preprocessing, cultural semantic translation, narrative reconstruction, audience adaptation verification, and feedback weight optimization, it achieves intelligent iterative adaptation of cross-cultural scripts, solving the technical pain points of traditional manual adaptation, such as low efficiency, inaccurate cultural translation, poor audience adaptability, and lack of a long-term iterative closed loop. The formulas described above are all dimensionless numerical calculations, derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas can be set by those skilled in the art based on actual conditions.

[0059] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent adaptation of theatrical scripts based on cultural translation, characterized in that, Includes the following steps: The original script text, the corresponding cultural background data, and the target country's theatrical aesthetic data are collected. The collected original script text is converted and standardized to generate a preprocessed data package. The script Transformer model is loaded and fine-tuned based on the cross-cultural semantic library data. The script text in the preprocessed data package is input into the fine-tuned model to complete the literal semantic translation, cultural semantic translation, and context-adaptive translation in sequence, generating the cross-cultural translated script text. Based on the evolution of the narrative paradigm and audience preferences of the target country's theatrical culture, the translated script text is optimized in sequence in terms of scene optimization, dialogue optimization, core conflict and character arc optimization, and narrative rhythm optimization, so as to generate a reconstructed script text that is suitable for the target audience. Next, audience adaptation verification is carried out for the reconstructed script text by group, and the script is iteratively optimized based on the verification results, generating corresponding audience adaptation verification reports; the data of the entire script adaptation process is integrated with the audience adaptation verification reports to generate standardized drama script adaptation guidance instructions; all audience behavior feedback data after the adapted script is launched and disseminated online are collected, data preprocessing and paragraph association weighting are completed, and the core parameters of cultural semantic translation and narrative structure reconstruction are optimized in reverse based on the weighting results, forming a closed loop of full-process iterative adaptation of scripts.

2. The intelligent adaptation method for drama scripts based on cultural translation according to claim 1, characterized in that, The original text is first obtained through high-definition scanning and text extraction. Simultaneously, cultural background data such as the era, social culture, etiquette and customs, and language habits corresponding to the original text are collected, as well as theatrical aesthetic data such as the age, cultural background, preference for drama genres, preference for dialogue style, and preference for narrative rhythm of the target audience. Then, a full-process preprocessing operation is performed on the original text.

3. The intelligent adaptation method for drama scripts based on cultural translation according to claim 2, characterized in that, The text preprocessing operation achieves the required rate of redundant information removal and noise reduction accuracy to meet the constraints. During the preprocessing process, the text is converted to JSON format and structured. Mapping fields are set, including script ID, scene number, scene description, characters, lines, entity annotation, cultural background notes, and paragraph unique ID. The paragraph unique ID enables accurate matching between subsequent audience behavior data and text paragraphs.

4. The intelligent adaptation method for drama scripts based on cultural translation according to claim 1, characterized in that, The script Transformer model is fine-tuned by loading cross-cultural semantic database data. The fine-tuning process uses the cross-entropy loss function to calculate the model loss until the loss value falls into the preset range. Then, the pre-processed script text is input into the model to complete three layers of semantic translation in sequence: literal semantic translation, cultural semantic translation, and context-adaptive translation.

5. The intelligent adaptation method for drama scripts based on cultural translation according to claim 4, characterized in that, During the translation process, a semantic hierarchical encoder is introduced into the script Transformer model to decompose each cultural proprietary item into three levels: surface semantics, cultural semantics, and emotional semantics. Semantic adaptive coefficients are set for the corresponding levels, and attention weight allocation in the translation process is dynamically adjusted in combination with temporal feedback weight coefficients. At the same time, matching thresholds for the three levels of semantics are set. When the semantic matching degree of any level is lower than the threshold, manual review or supplementary cultural annotations are automatically triggered.

6. The intelligent adaptation method for drama scripts based on cultural translation according to claim 1, characterized in that, The core quantitative parameters of the target country's dramatic narrative structure are extracted and the average value is calculated as a reconstruction reference standard. Hidden Markov Model is used to model the evolution of audience preferences. Clustering algorithm is used to cluster the preferences of audiences of different age groups. Finally, based on the reference standard and clustering results, the translated script text is optimized for group-specific adaptation in terms of scenes, lines, core conflicts, character arcs and narrative rhythm.

7. The intelligent adaptation method for drama scripts based on cultural translation according to claim 1, characterized in that, Collect and preprocess audience behavior data after the adapted script is released online, and construct a "paragraph ID-behavior data" mapping table; Then, a semantic association graph of script paragraphs is constructed, and graph convolutional networks are used to aggregate the features of adjacent paragraphs. The comprehensive weight value of each paragraph is calculated and optimization priority is divided. Finally, based on the priority results, the semantic mapping weight and narrative structure parameters of paragraphs with high optimization priority are dynamically adjusted, and the adaptation guidance instructions are updated synchronously to form a closed loop of full-process iteration.

8. A theatrical script intelligent adaptation system based on cultural translation, characterized in that, It includes terminal equipment, script data acquisition module, text preprocessing module, cultural semantic translation module, narrative structure reconstruction module, audience adaptation verification module, instruction generation module, offline audience optimization module, data storage and retrieval module, and storage medium; The terminal equipment includes a local adaptation terminal (TA), a cloud computing terminal (TB), and a data acquisition terminal (TC), which are connected via a dedicated network link. The script data acquisition module is embedded in the data acquisition terminal TC and is connected to the text preprocessing module via signal; the text preprocessing module is deployed on the local adaptation terminal TA and is connected to the script data acquisition module, the cultural semantic translation module, and the data storage and retrieval module via signal, respectively. The cultural semantic translation module, narrative structure reconstruction module, and offline audience optimization module are all deployed on the cloud computing terminal TB. The cultural semantic translation module is connected to the text preprocessing module, narrative structure reconstruction module, data storage and retrieval module, and offline audience optimization module. The narrative structure reconstruction module is connected to the cultural semantic translation module, audience adaptation verification module, guidance instruction generation module, and offline audience optimization module. The offline audience optimization module is connected to the data storage and retrieval module, cultural semantic translation module, and narrative structure reconstruction module. The audience adaptation verification module is deployed on the local adaptation terminal TA and is connected to the narrative structure reconstruction module, the guidance instruction generation module, and the data storage and retrieval module by signal; the guidance instruction generation module is connected to the audience adaptation verification module and the offline audience optimization module by signal. The data storage and retrieval module is connected to each functional module and the storage medium via signals. The storage medium includes local storage medium and cloud storage medium, which are used for dual backup and secure storage of the system's full data.

9. The intelligent adaptation system for drama scripts based on cultural translation according to claim 8, characterized in that, Both local and cloud-based storage media adopt an edge-cloud collaborative distributed storage architecture, with dedicated storage partitions divided according to data types. The system has a pre-set initial database, including a cross-cultural semantic database, cultural narrative paradigm data, target country audience preference data, and audience behavior feedback rule database, providing data support for the entire system operation.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the intelligent adaptation method for theatrical scripts based on cultural translation as described in any one of claims 1 to 7.

Images