Translation title and honorific selection method, device, equipment, medium and system based on dynamic role relationship graph

Abstract

The application provides a translation title and polite language selection method and device based on a dynamic role relationship graph, an electronic device, a computer readable storage medium and a system. The method comprises: obtaining script text data and extracting time-synchronized video data; scanning the script text data to extract the identity identifier of the target role; reading the time label and marking it as a first timestamp; tracking the pixel displacement of the video data to extract the action features; determining the relationship closeness quantitative value under the plot node; generating a relationship state trajectory and constructing a dynamic role relationship graph; and generating a target translation text according to the dynamic role relationship graph.

Description

Technical Field

[0001] This application relates to the field of film and television text translation and multimedia data processing technology. Specifically, it relates to a method, apparatus, electronic device, computer-readable storage medium, and system for selecting translation titles and honorifics based on a dynamic role relationship graph. Background Technology

[0002] In applications such as film and television subtitle translation, TV series localization, and multilingual dialogue generation, the expression of titles and honorifics is not merely a matter of replacing ordinary characters. It is also closely related to the plot position of the dialogue, the narrative relationship between the speakers, and the interaction between characters on screen. When the same character faces different people at different stages of the plot, the level of titles and honorifics often needs to change with the change in the communication relationship, thus placing higher demands on data correlation in film and television text translation processing.

[0003] Existing film and television subtitle translation solutions typically focus on dialogue characters as the primary processing object. They read the sentences to be translated, identify personal pronouns or polite particles within the sentences, and generate the target language text based on a static dictionary or fixed translation template. This approach generally first segments the input dialogue, then determines candidate translations based on a language rule table, and finally uniformly replaces predicates and honorifics according to preset vocabulary replacement rules.

[0004] However, the aforementioned processing methods primarily rely on the character content of the sentences to be translated and static replacement templates, making it difficult to utilize the continuously changing character relationship information during plot progression. When the relationship status of the same character changes in different plot stages, existing methods still tend to use fixed titles or fixed levels of honorifics, resulting in a mismatch between the generated target language text and the character's interaction status in the current plot stage. Therefore, how to link character identity, plot time, and on-screen interaction status during dialogue translation processing, and generate titles and honorifics more suitable for the current plot node, has become a technical problem that needs to be solved in the field of film and television text translation processing. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, computer-readable storage medium, and system for selecting translation titles and honorifics based on a dynamic role relationship graph, in order to at least alleviate the aforementioned technical problems.

[0006] A method for selecting titles and honorifics in translation based on a dynamic role relationship graph, comprising: Acquire script text data and extract video data that is time-synchronized with the script text data; The script text data is scanned for character sequences to extract at least one identifier character for a target character; Read the timestamps attached to the script text data and mark the timestamps as the first timestamp; Pixel displacement tracking is performed on the video data to extract the motion features of the target character; Based on the identity identifier characters and the action features, determine the quantitative values ​​of the relationship between different target characters at plot nodes in the pre-configured plot timeline; Arrange multiple relational closeness quantification values ​​according to the time sequence of the first timestamp to generate a relational state trajectory; Based on the relationship state trajectories corresponding to multiple target roles, a dynamic role relationship graph is constructed. Obtain the dialogue text to be translated, and generate the target translation text based on the dynamic character relationship graph.

[0007] Optionally, the step of obtaining the dialogue text to be translated, and generating the target translation text based on the dynamic character relationship graph, includes: Extract the second timestamp carried by the text of the dialogue to be translated; The second timestamp is compared with the plot nodes in the pre-configured plot timeline to locate the target plot node that matches the second timestamp. In the dynamic character relationship graph, extract the current state identifier of the target character under the target plot node; Based on the current state identifier, extract the corresponding translation replacement rule from the rule base; The original title characters in the text to be translated are replaced using the translation replacement rules to update the title expression; The original honorific characters in the text to be translated are replaced using the translation replacement rules to update the honorific expressions, thereby obtaining the target translated text.

[0008] Optionally, obtaining the script text data includes: Read the global script character stream from a multimedia video file; Identify dialogue boundary markers in the global script character stream; Based on the dialogue boundary symbol, a character segment containing the pronunciation subject information is extracted from the global script character stream to serve as the script text data; The extraction of video data synchronized with the script text data in time includes: Extract the start and end times of the video playback associated with the pronunciation subject information; Based on the start and end times of the video playback, video data synchronized with the script text data is extracted from the main video track of the multimedia video file.

[0009] Optionally, the step of performing pixel displacement tracking on the video data to extract the motion features of the target character includes: The video data is processed frame by frame to obtain a time-series image sequence; Contour recognition is performed in the time-series image sequence to locate the center point of the facial contour representing the target character; Joint recognition is performed in the time-series image sequence to locate the center point of the torso representing the target character; Calculate the first displacement vector of the center point of the facial contour in two adjacent temporal image sequences; Calculate the second displacement vector of the torso center point in two adjacent temporal image sequences; Spatial displacement data is generated based on the first displacement vector and the second displacement vector; The semantic action tags that match the spatial displacement data are queried in the pre-configured action tag mapping library to generate the action features of the target character based on the semantic action tags. The pre-configured action tag mapping library stores the mapping relationship between spatial displacement data and semantic action tags.

[0010] Optionally, determining the quantitative value of the relationship between different target characters at plot nodes in the pre-configured plot timeline based on the identity identifier character and the action feature includes: The first timestamp is compared with the pre-configured story timeline to determine the story node hit by the first timestamp; The frequency of interaction of the action features between the two target characters at the stated plot node is statistically analyzed. The motion features are decomposed to extract the limb contact identifiers contained in the motion features; Based on the interaction frequency, the first intimacy evaluation character is retrieved from the pre-configured activity mapping table, wherein the pre-configured activity mapping table records the mapping relationship between the interaction frequency and the first intimacy evaluation character; Based on the physical contact identifier, relationship correction characters are extracted from a pre-configured correction table, wherein the pre-configured correction table records the mapping relationship between physical contact identifiers and relationship correction characters; Based on the first intimacy evaluation character and the relationship correction character, a combined string is generated, and based on the combined string, the relationship intimacy quantification value is generated.

[0011] Optionally, the step of extracting relationship correction characters from a pre-configured correction table based on the limb contact identifier includes: The physical contact markers are parsed to identify the type of behavioral action; When the behavior action type includes a physical collision behavior flag, the first adjustment character representing the hostile state is retrieved from the pre-configured correction table and used as the relationship correction character; When the behavior action type does not contain a physical collision behavior flag but contains a side-by-side sticking behavior flag, a second adjustment character representing the intimacy state is retrieved from the pre-configured correction table as the relationship correction character.

[0012] Optionally, constructing a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles includes: Establish a blank topological base map with a planar coordinate system; Map the multiple identity identifier characters to multiple entity nodes in the blank topology base map; The relationship state trajectory is parsed into directed edges with direction and label attributes; The directed edges are used to connect corresponding entity nodes that have interactive relationships, and the time sequence corresponding to the relationship state trajectory is attached to the directed edges as a time attribute to complete the construction of the dynamic role relationship graph.

[0013] Optionally, extracting the current state identifier of the target character under the target plot node from the dynamic character relationship graph includes: Read the status tag data of the target character corresponding to the target plot node in the dynamic character relationship graph; The status label data is parsed to extract the hierarchy indicator characters that represent hierarchical subordination and the sentiment indicator characters that represent sentiment inclination. Based on the hierarchical and subordinate indicator characters and the emotional tendency indicator characters, feature-encoded characters are generated, and the current status identifier of the target character is generated based on the feature-encoded characters.

[0014] Optionally, the step of extracting the corresponding translation replacement rule from the rule base based on the current state identifier includes: The current status identifier is input into a pre-configured rule instruction library for matching and querying to obtain alternative control instructions. The pre-configured rule instruction library stores the mapping relationship between status identifiers and control instructions. Extract the control command that matches the current state identifier from the candidate control commands; The control instructions are parsed to obtain the honorific prefix supplementation instruction for adding honorific prefixes and the peer-level address reduction instruction for reducing the honorific level. The translation replacement rules are generated based on the honorific prefix supplementation instruction and the peer-level address dimensionality reduction instruction.

[0015] Optionally, the step of replacing the original appellation characters in the text to be translated using the translation replacement rules to update the appellation expression includes: Lexical scanning is performed on the text of the dialogue to be translated to mark the first position of the original appellation character belonging to the category of personal pronouns; The honorific prefix supplementation instruction in the translation replacement rule is triggered, and a prefix character representing the honorific is inserted before the first position to update the original title character, thereby forming the title expression; The process of using the translation replacement rules to overwrite and replace the original honorific characters in the text to be translated, in order to update the honorific expressions, includes: Lexical scanning is performed on the text of the dialogue to be translated to mark the second position of the original honorific characters belonging to the category of polite particles; The translation replacement rule triggers the peer-level address reduction instruction, which uses the omission of auxiliary words to cover the original honorific character at the second position, thereby updating the original honorific character and forming an honorific expression.

[0016] A translation address and honorific selection device based on a dynamic role relationship graph, comprising: The script video acquisition module is used to acquire script text data and extract video data that is time-synchronized with the script text data; The identity extraction module is used to scan the script text data for character sequences to extract at least one identity character of the target character; The timestamp annotation module is used to read the timestamps attached to the script text data and annotate the timestamps as the first timestamp; The motion feature extraction module is used to perform pixel displacement tracking on the video data in order to extract the motion features of the target character; The relationship closeness determination module is used to determine the quantitative value of the relationship closeness between different target characters at plot nodes in the pre-configured plot timeline based on the identity identification characters and the action features. The relationship trajectory generation module is used to arrange multiple relation closeness quantification values ​​according to the time order of the first timestamp to generate a relationship state trajectory; The relationship graph construction module is used to construct a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles; The translation text generation module is used to obtain the dialogue text to be translated, and generate the target translation text based on the dynamic character relationship graph.

[0017] An electronic device includes a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the translation title and honorific selection method based on a dynamic role relationship graph as described above.

[0018] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the translation title and honorific selection method based on a dynamic role relationship graph as described above.

[0019] A translation address and honorific selection system based on dynamic character relationship graph, including a script video processing end, a character relationship processing end, a relationship graph construction end, and a translation text processing end; The script video processing terminal is used to acquire script text data and extract video data that is time-synchronized with the script text data; The script video processing terminal is also used to perform character sequence scanning on the script text data to extract at least one target character's identity identifier character, and to read the time tag attached to the script text data to mark the time tag as a first timestamp; The character relationship processing terminal is used to perform pixel displacement tracking on the video data in order to extract the motion features of the target character; The character relationship processing terminal is also used to determine the quantitative value of the relationship between different target characters at plot nodes in the pre-configured plot timeline based on the identity identification characters and the action features; The relationship graph construction end is used to arrange multiple relationship closeness quantification values ​​according to the time order of the first timestamp, so as to generate a relationship state trajectory; The relationship graph construction terminal is also used to construct a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles; The translation text processing terminal is used to obtain the dialogue text to be translated, and to generate the target translation text based on the dynamic character relationship graph.

[0020] The technical advantages of the technical solution provided in this application are: This application presents a translation method for addressing and honorifics based on a dynamic character relationship graph. Addressing the shortcomings of existing film and television subtitle translation solutions, which primarily rely on the character content of the translated text and static replacement templates, making it difficult to adapt to changes in character relationships as the plot progresses, this method first acquires the script text data and then extracts video data synchronized with the script text data in time. This establishes a temporal correspondence between the text source and the video source within the same processing chain. Compared to traditional methods that only process the characters of the translated text, this application introduces video data synchronized with the script text data before translation processing. This provides a relevant data source for subsequent identification of target characters and extraction of action changes, thus offering a processing foundation where text and video data jointly participate in the dynamic selection of addressing and honorific expressions.

[0021] Furthermore, this application performs character sequence scanning on the script text data to extract at least one target character's identity identifier character, and reads the timestamps attached to the script text data to mark the timestamps as a first timestamp, thus establishing a correspondence between the target character's character identity and its time position in the plot. Traditional processing solutions typically only focus on word substitution in the dialogue to be translated, making it difficult to distinguish the communication partners and relationship changes of the same character in different plot stages during translation. This application uses the identity identifier character to mark the target character and uses the first timestamp to mark the plot time position of the target character, so that subsequent relationship state analysis no longer stops at the single sentence text level, but can organize the relationship changes between characters along the plot time sequence.

[0022] Furthermore, this application performs pixel displacement tracking on the video data to extract the action features of the target characters, and determines the quantified values ​​of the relationship between different target characters at plot nodes in a pre-configured plot timeline based on the identity identifier characters and the action features. Thus, the interaction changes between target characters in the scene can be transformed into action features that can be used for data processing, and further participate in the determination of the quantified values ​​of relationship closeness together with the identity identifier characters. Compared to static dictionaries or fixed templates that only select titles and honorifics based on the language itself, this application incorporates the on-screen interaction state of target characters at plot nodes into the relationship judgment process, so that the relationship state between different target characters can be expressed by the quantified values ​​of relationship closeness, thereby reducing the probability of fixed titles or fixed honorific levels not matching the current plot node.

[0023] Furthermore, this application arranges multiple relational quantification values ​​according to the time sequence of the first timestamp to generate relational state trajectories, and constructs a dynamic character relationship graph based on the relational state trajectories corresponding to multiple target characters. The relational state trajectories reflect the changing order of the relational quantification values ​​between target characters as the plot progresses, while the dynamic character relationship graph organizes the relational state trajectories corresponding to multiple target characters into a data structure that can be read by subsequent translation processing. Compared to the problem that static character lists or fixed character relationship tables in traditional processing schemes cannot express the order of relational changes, this application preserves the time-varying attributes of character relationships through the relational state trajectories and the dynamic character relationship graph, enabling the text to be translated to obtain the character relationship basis corresponding to the current plot stage during subsequent translation processing.

[0024] Furthermore, this application obtains the dialogue text to be translated and generates the target translation text based on the dynamic character relationship graph, ensuring that the appellations and honorifics in the target translation text originate from the current relationship state represented by the dynamic character relationship graph. Traditional processing methods typically perform character replacement in the absence of plot node relationship state information, easily leading to the same character using the same appellation or honorific level in different plot stages, whether the relationship is distant, close, or antagonistic. This application provides the dialogue text to be translated with the dynamic character relationship graph, updating the relationship state information as the plot progresses, enabling the generated target translation text to better match the changes in the relationship between the target characters at the current plot node, thereby improving the contextual adaptability of appellations and honorifics in film and television text translation. Attached Figure Description

[0025] Figure 1 This application provides a scenario for a translation method for selecting titles and honorifics based on a dynamic role relationship graph. Figure 2 This application provides an embodiment of a method for selecting titles and honorifics in translation based on a dynamic role relationship graph. Figure 3 This application provides an embodiment of a translation address and honorific selection device based on a dynamic role relationship graph; Figure 4 This is an electronic device according to an embodiment of the present application.

[0026] Figure 5 This application provides an embodiment of a computer-readable storage medium.

[0027] Figure 6 This application provides an embodiment of a translation title and honorific selection system based on a dynamic role relationship graph. Detailed Implementation

[0028] like Figure 1 The image shown illustrates a scenario of a translation title and honorific selection method based on a dynamic role relationship graph, according to an embodiment of this application. Figure 2 The image shows an embodiment of this application of a method for selecting translation titles and honorifics based on a dynamic role relationship graph, comprising the following steps: Acquire script text data and extract video data that is time-synchronized with the script text data; The script text data is scanned for character sequences to extract at least one identifier character for a target character; Read the timestamps attached to the script text data and mark the timestamps as the first timestamp; Pixel displacement tracking is performed on the video data to extract the motion features of the target character; Based on the identity identifier characters and the action features, determine the quantitative values ​​of the relationship between different target characters at plot nodes in the pre-configured plot timeline; Arrange multiple relational closeness quantification values ​​according to the time sequence of the first timestamp to generate a relational state trajectory; Based on the relationship state trajectories corresponding to multiple target roles, a dynamic role relationship graph is constructed. Obtain the dialogue text to be translated, and generate the target translation text based on the dynamic character relationship graph.

[0029] Optionally, the step of obtaining the dialogue text to be translated, and generating the target translation text based on the dynamic character relationship graph, includes: Extract the second timestamp carried by the text of the dialogue to be translated; The second timestamp is compared with the plot nodes in the pre-configured plot timeline to locate the target plot node that matches the second timestamp. In the dynamic character relationship graph, extract the current state identifier of the target character under the target plot node; Based on the current state identifier, extract the corresponding translation replacement rule from the rule base; The original title characters in the text to be translated are replaced using the translation replacement rules to update the title expression; The original honorific characters in the text to be translated are replaced using the translation replacement rules to update the honorific expressions, thereby obtaining the target translated text.

[0030] Preferably, the specific implementation process for extracting the second timestamp carried by the text to be translated is as follows: After the construction of the dynamic character relationship graph has been completed, the text to be translated that has entered the translation processing flow is first read, and the text to be translated is structurally split at the character level to form a text structure splitting result. The text structure splitting result is used to distinguish the text characters, pronunciation prompt characters, and time position description characters in the text to be translated. The technical essence of the text to be translated is that it is the text character data waiting to be generated in the target language expression during the translation processing stage. The text to be translated is not an independent sentence detached from the plot position, but carries the time position description characters that can point to the plot playback position. Therefore, the text to be translated provides both the original title characters and original honorific characters that need to be replaced later, and the time positioning basis required for subsequent querying of the dynamic character relationship graph. In specific implementation, the text to be translated is first read line by line to form a line-level reading result, and the time position character segment where the time position description character is located is identified in the line-level reading result. Then, the time position description character in the time position character segment is formatted, and the start playback time, end playback time, or single-point playback time existing in different sources is converted into a second timestamp that can be compared with the pre-configured plot timeline. The technical essence of the second timestamp is the time coordinate of the text to be translated in the plot time dimension. The second timestamp is used to represent the position where the dialogue to be translated occurs, rather than representing the time when the text to be translated is read or translated. If the text to be translated carries the start playback time and the end playback time, the start playback time is read as the second timestamp, or the center position of the time interval formed by the start playback time and the end playback time is read as the second timestamp; if the text to be translated only carries the single-point playback time, the single-point playback time is directly converted into the second timestamp. After the second timestamp is formed, a time-location association is established with the text to be translated. The time-location association is used to continuously bind the second timestamp to the text to be translated, so that the second timestamp can directly participate in the positioning of the target plot node in subsequent time sequence comparison, and so that the subsequent replacement of the original title characters and the replacement of the original honorific characters in the text to be translated fall into the plot position corresponding to the second timestamp and are executed, thereby avoiding static replacement based only on the text characters of the dialogue and ignoring the plot time change.

[0031] Preferably, in the specific technical implementation of comparing the second timestamp with the plot nodes in the pre-configured plot timeline to locate the target plot node matching the second timestamp, the pre-configured plot timeline, which has been kept consistent with the time caliber of the dynamic character relationship graph, is first read, and multiple plot nodes and their corresponding time ranges are extracted from the pre-configured plot timeline. The technical essence of the pre-configured plot timeline is a time index structure configured before film and television text translation processing using time tags in the script text data, the start and end times of video playback, and plot segment boundaries. The pre-configured plot timeline does not directly replace the dynamic character relationship graph, but rather provides a time entry point for the second timestamp to look up the relationship status in the dynamic character relationship graph. The technical essence of the plot node is a time segment in the pre-configured plot timeline used to carry changes in character relationship status. The plot node can correspond to a dialogue scene, a continuous interactive scene, or a playback interval where character relationships change; each plot node has its own time range and corresponds to the status tag data in the dynamic character relationship graph. During the specific comparison, the second timestamp is first converted to a time unit identical to the node time range, and then a range scan is performed according to the chronological order of each plot node in the pre-configured plot timeline. When the second timestamp falls within the node time range of a plot node, that plot node is marked as the target plot node. When the second timestamp is located at the boundary between two adjacent plot nodes, the pronunciation prompt characters in the text to be translated and the node time ranges of the two adjacent plot nodes are read, and the target plot node is selected based on the time distance between the second timestamp and the node time ranges of the two adjacent plot nodes. The technical essence of the target plot node is the plot relationship reading position corresponding to the text to be translated on the pre-configured plot timeline. The target plot node is used to limit the time range for subsequently reading the current state identifier of the target character from the dynamic character relationship graph. Through the above time sequence comparison, the second timestamp is no longer just an auxiliary time information of the text to be translated, but is transformed into a retrieval basis connecting the text to be translated and the dynamic character relationship graph, so that when generating the target translation text later, the character relationship state that matches the plot position of the text to be translated can be used.

[0032] Preferably, the specific implementation process for extracting the current state identifier of the target character under the target plot node in the dynamic character relationship graph is as follows: After the target plot node is located, the time attribute corresponding to the target plot node in the dynamic character relationship graph is read first, and the entity node under the target plot node, the directed edge with direction attribute and label attribute, and the time attribute attached to the directed edge are filtered according to the time attribute. The technical essence of the dynamic character relationship graph is a character relationship data structure constructed by the relationship state trajectories corresponding to multiple target characters. The dynamic character relationship graph uses the entity node to represent the target character corresponding to the identity identifier character, the directed edge with the direction attribute and label attribute to represent the interaction relationship between different target characters, and the time attribute to represent the order in which the interaction relationship occurs in the pre-configured plot timeline. In the specific extraction process, the pronunciation prompt characters in the text to be translated are first matched with the identity identifier characters in the dynamic character relationship graph to determine the entity node corresponding to the target character. Then, the directed edges connected to the entity node corresponding to the target character are read under the target plot node, and the relationship closeness quantification value, hierarchy / subordination indicator characters, and emotional tendency indicator characters are extracted from the label attributes of the directed edges. The technical essence of the current state identifier is an encoded representation of the target character's relationship state under the target plot node. The current state identifier is not simply a character name or a title entry, but a state index character formed by the relationship closeness quantification value, hierarchy / subordination indicator characters, and emotional tendency indicator characters of the target character under the target plot node. Specifically, when forming the current state identifier, the state tag data corresponding to the target character under the target plot node is first read. Then, the state tag data is parsed to obtain the hierarchy indicator character and the emotional tendency indicator character. Subsequently, the hierarchy indicator character, the emotional tendency indicator character, and the relationship closeness quantification value are sequentially encapsulated to form the current state identifier that can be retrieved by the rule base. After the current state identifier is formed, it is written into the translation processing record corresponding to the dialogue text to be translated. The translation processing record continues to carry the correspondence between the dialogue text to be translated, the second timestamp, the target plot node, and the current state identifier, so that when the translation replacement rule is subsequently extracted from the dialogue text to be translated, the character relationship state under the target plot node can be used, avoiding the misuse of relationship states from other plot nodes for the current dialogue translation.

[0033] Preferably, the specific implementation process of extracting the corresponding translation replacement rule from the rule base based on the current state identifier is as follows: After the current state identifier has been formed, the rule base corresponding to the film and television text translation processing task is first read, and the state index field, title replacement field, honorific replacement field, and applicable language field are read from the rule base. The technical essence of the rule base is to convert the current state identifier into structured rule data for text replacement actions. The rule base is not an ordinary dictionary, nor is it a character table that only stores static translated words, but rather records the mapping relationship between the current state identifier and the translation replacement rule; wherein, the state index field is used to store the state identifier that can be compared with the current state identifier, the title replacement field is used to store the replacement method of the original title character, the honorific replacement field is used to store the overwrite replacement method of the original honorific character, and the applicable language field is used to limit the target language environment corresponding to the translation replacement rule. When constructing the rule base, rule entries can be established before the film and television text translation processing task starts, based on a combination of the target language's appellation type, honorific level, hierarchical indicator characters, emotional tendency indicator characters, and relational closeness quantification values. Each rule entry includes a status index field and corresponding appellation replacement fields, honorific replacement fields, and applicable language fields, enabling the rule base to retrieve rule entries matching the target plot node based on the current status identifier. Specifically, during extraction, the current status identifier is first read, and then matched character by character with each of the status index fields in the rule base. When the current status identifier meets the consistency comparison condition with a certain status index field, the appellation replacement field, honorific replacement field, and applicable language field corresponding to that status index field are read, and these fields are encapsulated into the translation replacement rule. The technical essence of the translation replacement rule is the text operation basis for performing replacement processing on the original title characters and the original honorific characters in the text to be translated. The translation replacement rule includes the update caliber of title expression, the update caliber of honorific expression, and the target language adaptation caliber. The update caliber of title expression comes from the title replacement field and is used to update title expressions in the future. The update caliber of honorific expression comes from the honorific replacement field and is used to update honorific expressions in the future. The target language adaptation caliber comes from the applicable language field and is used to limit the target language expression environment of the title expression and the honorific expression.By using the mapping between the current state identifier and the rule base, the translation replacement rule can inherit the relationship state of the target character under the target plot node, so that the update of title expression and honorific expression no longer depends on a fixed template, but is directly related to the state label data in the dynamic character relationship graph.

[0034] Preferably, in the specific technical implementation of replacing the original appellation characters in the text to be translated using the translation replacement rule to update the appellation expression, the text of the dialogue to be translated is first read, and the text of the dialogue to be translated is lexically scanned to mark candidate appellation characters belonging to the categories of personal pronouns, kinship terms, job titles, or role names; then, the candidate appellation characters are correlated and compared with the pronunciation subject prompt characters in the text to be translated, the identity identifier characters corresponding to the target role, and the appellation expression update criteria in the translation replacement rule to determine the original appellation character from the candidate appellation characters. The technical essence of the original appellation character is the appellation character in the text to be translated used to point to the communication object or the person being addressed. The original appellation character directly corresponds to the appellation expression in the target translation text. Therefore, the determination result of the original appellation character will affect the expression of appellation level, appellation prefix, and appellation object in the subsequent target language. During the specific replacement process, the update guidelines for appellation expressions in the translation replacement rules are first read, and the target appellation character, honorific prefix character, or appellation omission indicator character corresponding to the original appellation character is extracted from the update guidelines. Then, the character position of the original appellation character in the text to be translated is read, and this position is recorded as the first position. Appellation replacement processing is then performed at the first position. If the update guidelines indicate that the honorific prefix character is inserted before the original appellation character, the honorific prefix character is written before the first position, and the original appellation character is retained to participate in forming the appellation expression. If the update guidelines indicate that the original appellation character is replaced with the target appellation character, the target appellation character is used to cover the original appellation character at the first position to form the updated appellation expression. If the update guidelines indicate that the original appellation character is omitted, the original appellation character at the first position is deleted, and the adjacent characters after deletion are normalized with whitespace to form the updated appellation expression. Once the title expression is formed, it enters the subsequent honorific expression update process together with the text characters of the dialogue in the text to be translated, so that the title expression in the target translation text can maintain a correspondence with the current status identifier under the target plot node, thereby reducing the context mismatch caused by the same title expression being used for the same target character in different plot nodes.

[0035] Preferably, the specific implementation process of using the translation replacement rule to overwrite and replace the original honorific characters in the text to be translated to update the honorific expressions and thus obtain the target translated text is as follows: After the address expression is updated, the text to be translated, which already contains the address expression, is read again, and the text characters of the text to be translated are lexically scanned to mark candidate honorific characters belonging to the categories of polite particles, honorific suffixes, respectful verbs, or humble expressions; then, the candidate honorific characters are matched with the honorific expression update criteria in the translation replacement rule to determine the original honorific characters from the candidate honorific characters. The technical essence of the original honorific characters is the character content in the text to be translated that reflects the level of politeness or tone between the speaker and the communication object. The original honorific characters and the address expression together affect the relationship expression effect of the target translated text. Therefore, the original honorific characters need to be overwritten and replaced under the relationship state defined by the current state identifier. During the specific overwrite and replacement process, the character position of the original honorific character in the text to be translated is first read, and this character position is recorded as the second position. Then, the honorific expression update guidelines in the translation replacement rules are read, and the target honorific character, the peer-level address reduction indicator character, or the omitted auxiliary word corresponding to the original honorific character is determined based on the honorific expression update guidelines. If the honorific expression update guidelines indicate maintaining a high level of politeness, the original honorific character at the second position is overwritten with the target honorific character to form the updated honorific expression. If the honorific expression update guidelines indicate lowering the honorific level, the peer-level address reduction instruction is triggered, and the omitted auxiliary word is used to overwrite the original honorific character at the second position to form the updated honorific expression. If the honorific expression update guidelines indicate deleting a polite auxiliary word, the original honorific character at the second position is deleted, and adjacent characters in the deleted text are joined together to form the updated honorific expression. The technical essence of the honorific expression is that the politeness level is adjusted according to the current state identifier of the target character under the target plot node. The honorific expression and the previously formed title expression are jointly processed in the text reconstruction process. The text reconstruction process writes the updated title expression, the updated honorific expression, and the unreplaced text characters of the dialogue to be translated in the original character order to form the target translated text.The technical essence of the target translation text is that, under the constraints of the dynamic character relationship graph, the target language dialogue text is updated with new titles and honorifics. The target translation text retains the plot time position of the dialogue text to be translated, and continuously transmits the processing results between the second timestamp, the target plot node, the current state identifier, and the translation replacement rules to the final text generation process, so that the target translation text can correspond to the character relationship state in the current plot node.

[0036] Optionally, obtaining the script text data includes: Read the global script character stream from a multimedia video file; Identify dialogue boundary markers in the global script character stream; Based on the dialogue boundary symbol, a character segment containing the pronunciation subject information is extracted from the global script character stream to serve as the script text data; The extraction of video data synchronized with the script text data in time includes: Extract the start and end times of the video playback associated with the pronunciation subject information; Based on the start and end times of the video playback, video data synchronized with the script text data is extracted from the main video track of the multimedia video file.

[0037] Preferably, the specific implementation process of reading the global script character stream from a multimedia video file is as follows: Before obtaining the script text data, the multimedia video file is first read from the film and television material cache area. The film and television material cache area is used to store the multimedia video file to be parsed and to provide a file addressing location for subsequent reading of the multimedia video file. Subsequently, the multimedia video file is parsed using an encapsulation layer to extract the media encapsulation index record that records the media track type, track start and end positions, track time base, and character encoding method. The technical essence of the multimedia video file is that it is a media data carrier that simultaneously carries the global script character stream, the main video track, and time positioning information. The multimedia video file provides a unified data source for subsequently obtaining the script text data from the same data source and extracting video data that is time-synchronized with the script text data, reducing the situation where the time base is inconsistent when the script text data and the video data come from different files. In specific implementation, firstly, based on the track type marker in the media encapsulation index record, the character track containing dialogue characters, pronunciation prompt characters, time tags, or subtitle text is located from the multimedia video file. The track type marker is used to distinguish between the character track and the main video track in the multimedia video file, ensuring that subsequent character reading and processing falls within the character track. Then, the character encoding method corresponding to the character track is read, and the byte data in the character track is decoded according to the character encoding method to form the global script character stream arranged in playback order. After character decoding, the byte data is converted into dialogue characters, pronunciation prompt characters, and time tags that can be scanned. The technical essence of the global script character stream is a continuous sequence of script characters extracted from the multimedia video file. The global script character stream includes dialogue characters used to express dialogue content, pronunciation prompt characters used to identify the pronunciation subject, and time tags used to limit the playback position of the dialogue. Since it is necessary to extract character segments containing the speech entity information from the global script character stream and further generate the script text data, a character order check is performed on the global script character stream after its formation. This character order check is used to rearrange out-of-order character segments according to the track time base in the media encapsulation index record, ensuring that the dialogue characters, speech entity prompts, and time tags in the global script character stream are consistent with the playback order of the multimedia video file. After the above processing, the global script character stream can serve as the basic character source for subsequent identification of dialogue boundary symbols, extraction of character segments, and determination of the start and end times of the screen playback. This basic character source continues to participate in the subsequent formation of the script text data.

[0038] Preferably, in the specific technical implementation of identifying dialogue boundary symbols in the global script character stream, after the global script character stream is formed, the character arrangement positions in the global script character stream are first read, and the global script character stream is scanned character by character to identify candidate boundary characters that can represent the start or end position of a dialogue. The technical essence of the dialogue boundary symbol is a structural separator character or combination of structural separator characters located in the global script character stream. The dialogue boundary symbol is used to divide continuously arranged dialogue characters into multiple dialogue character segments that can independently express the phonological subject and dialogue content. The dialogue boundary symbol is not part of the semantic content of the dialogue, but rather a data separation basis used to determine the boundary of the character segment truncation. In the specific identification process, the system first reads the combination of newline characters, time stamp separators, separators after the main prompt characters, and whitespace characters between adjacent lines in the global script character stream. This combination of newline characters, time stamp separators, separators after the main prompt characters, and whitespace characters is used as the candidate boundary character. Then, the system reads the character arrangement before and after the candidate boundary character, determining whether there is a previous line of dialogue before the candidate boundary character and whether there is a new main prompt character, time stamp, or line start character after the candidate boundary character. When the character arrangement before and after the candidate boundary character satisfies the above-mentioned character arrangement relationship, the candidate boundary character is marked as the dialogue boundary symbol. To reduce the possibility of misidentifying ordinary punctuation marks in the dialogue text as the dialogue boundary symbol, the system further reads the time stamps and main prompt characters before and after the candidate boundary character. Only when the candidate boundary character forms a dialogue separation relationship with the time stamp and the main prompt character is the candidate boundary character written as the dialogue boundary symbol into the boundary index record. The boundary index record is used to record the character arrangement position of the dialogue boundary symbol in the global script character stream. The boundary index record provides the start and end positioning basis when subsequently extracting character segments from the global script character stream based on the dialogue boundary symbol. Through the above processing, the dialogue boundary symbol can divide the continuous global script character stream into multiple readable dialogue character segments. These dialogue character segments continue to serve as the basis for the character range of subsequent character segment extraction, so that the formation of subsequent script text data no longer depends on the overall reading of the entire character stream, but can instead determine character segments with pronunciation subjects and time tags around each dialogue boundary symbol.

[0039] Preferably, the specific implementation process of extracting character fragments containing pronunciation subject information from the global script character stream based on the dialogue boundary symbols to serve as the script text data is as follows: After the boundary index record is formed, the character arrangement positions of two adjacent dialogue boundary symbols in the global script character stream are read according to the boundary index record, and the character interval between two adjacent dialogue boundary symbols is taken as the character interval to be extracted. In this process, the dialogue boundary symbols are used to define the start and end boundaries of the character fragments, so that the character fragments extracted from the global script character stream correspond to a relatively complete dialogue content, rather than being arbitrarily extracted characters from the global script character stream. Specifically, during extraction, the corresponding dialogue characters, pronunciation subject prompt characters, and time tags are read first according to the character interval to be extracted, and then the pronunciation subject prompt characters are parsed in a field-based manner to form the pronunciation subject information. The technical essence of the pronunciation subject information is a character source field used to characterize which target character utters the dialogue. This pronunciation subject information can be formed from character name characters, character number characters, or subject prompt characters corresponding to the identity identifier characters. This pronunciation subject information subsequently participates in extracting the start and end times of the screen playback and in establishing the time synchronization relationship between the script text data and the video data. The technical essence of the character fragment is local character data extracted from the global script character stream according to the dialogue boundary symbols, carrying dialogue characters, pronunciation subject information, and time tags. This character fragment serves as the text basis for subsequent character sequence scanning, identity identifier character extraction, and first timestamp annotation. If a certain character interval to be extracted lacks the pronunciation subject prompt character, the pronunciation subject information from the adjacent preceding character fragment is read, and combined with the time tag in the current character interval to be extracted to determine whether it belongs to the same continuous pronunciation segment. When the time tags have a continuous relationship, the pronunciation subject information from the adjacent preceding character fragment is continued into the current character fragment, so that the current character fragment still contains the pronunciation subject information that can be used for subsequent processing. After the above extraction is completed, the character fragment containing the pronunciation subject information, dialogue characters, and time tags is used as the script text data. The technical essence of the script text data is text data extracted from the multimedia video file that can directly participate in character recognition and time synchronization processing. In subsequent processing, the script text data is used to extract the identity marker characters, read the time tags, and generate the first timestamp. Simultaneously, it provides the pronunciation subject information and time tags for extracting video data synchronized with the script text data. Thus, a continuous source and usage relationship is formed between the character fragment, the pronunciation subject information, the time tags, and the script text data.

[0040] Preferably, in the specific technical implementation of extracting the playback start and end times associated with the pronunciation subject information, after the script text data has been formed from the character segments, the time tags located in the same character segment as the pronunciation subject information in the script text data are first read, and the start and end time characters in the time tags are formatted to form the playback start and end times with the same time base as the video main track of the multimedia video file. The technical essence of associating the pronunciation subject information is to establish a binding relationship between the pronunciation subject information and the playback time interval of the corresponding lines, so that the pronunciation subject information can not only represent which target character delivers the lines, but also point to the corresponding screen interval of the target character in the video main track. Specifically, the process involves first reading the pronunciation subject information from the character segment and writing it into the screen time association record; then reading the time tag from the same character segment and extracting the start and end time characters from the time tag; when the time tag only contains the start time character, the start time character of the adjacent next character segment is read as the end time reference, or the subtitle display duration corresponding to the character segment in the multimedia video file is read as the end time reference, thereby forming the screen playback start and end time. The technical essence of the screen playback start and end time is to locate the time range of the screen interval corresponding to the script text data in the main video track. The screen playback start and end time includes the screen playback start time and the screen playback end time. The screen playback start time is used to determine the starting point of the cut in the main video track, and the screen playback end time is used to determine the ending point of the cut in the main video track. To ensure the playback start and end times correspond to the main video track, the track time reference in the media encapsulation index record is further read, and the start time character and end time character are converted into track time coordinates recognizable by the main video track. These track time coordinates are used to convert the characterized time in the timestamps into frame-level positioning data in the main video track. The converted playback start and end times, along with the speech subject information, are written into the playback time association record. This record is subsequently used to extract video data synchronized with the script text data from the main video track of the multimedia video file. Through this processing, a readable technical correspondence is established between the speech subject information, the timestamps, and the playback start and end times. This allows for the location of video frame intervals based on the speech subject and speech time position when extracting video data, and the playback time association record continues to serve as the basis for subsequently reading the playback start and end times.

[0041] Preferably, the specific implementation process of extracting video data synchronized with the script text data from the video main track of the multimedia video file based on the playback start and end times is as follows: After the playback time association record is formed, the playback start and end times in the playback time association record are read first, and the video main track index record in the multimedia video file is read. The video main track index record originates from the media encapsulation index record. The video main track index record is used to record the track position, frame time reference, and keyframe position of the video main track in the multimedia video file. The track position in the video main track index record is used to locate the video main track. The frame time reference in the video main track index record is used to map the playback start and end times to frame positions. The keyframe positions in the video main track index record are used to determine the decodeable starting point when reading frame data from the video main track. In this process, the playback start and end times are used to limit the extraction range of the video main track. The playback start and end times are not simply time text for display, but time positioning parameters used to map the script text data to the video playback data. In the specific extraction process, the playback start time is first mapped to the start frame position in the main video track, and the playback end time is mapped to the end frame position in the main video track. Then, based on the start and end frame positions, continuous frame data is read from the main video track and arranged sequentially according to the frame time reference of the main video track to form an initial video segment. If the playback start time falls between adjacent keyframes, reading begins from the nearest keyframe before the playback start time, and after reading, preceding frame data that does not belong to the target playback interval is trimmed according to the playback start time. The target playback interval is defined by both the playback start and end times. If the playback end time falls between adjacent frames, following frame data that does not belong to the target playback interval is trimmed according to the playback end time. After the above frame-level extraction and trimming processing, the initial video segment is converted into video data corresponding to the playback start and end times. The video data is used for pixel displacement tracking in subsequent processing to extract the action features of the target character. Therefore, the video data needs to retain the temporal order and visual continuity corresponding to the script text data. By extracting the video data from the main video track of the same multimedia video file, the voice subject information in the script text data, the time tags in the script text data, and the frame images in the video data maintain a correspondence under the same time reference, providing a source of visual data for subsequently determining the quantitative value of the relationship based on the identity identifier characters and the action features.

[0042] Preferably, the process of forming the video data synchronized with the script text data is as follows: After the video data is extracted from the main video track, the frame time markers corresponding to the video data are read, and the frame time markers are synchronized and verified with the time tags in the script text data to determine whether the time range of the video data covers the time range of the dialogue corresponding to the script text data. The technical essence of this time synchronization with the script text data is that the frame time range in the video data and the dialogue time range in the script text data have a correspondence under the same time base, enabling the video data to reflect the screen state when the dialogue described in the script text data occurs, rather than the screen state of other playback intervals in the multimedia video file. In specific implementation, firstly, the time stamps in the script text data are read, and the text start time and text end time are extracted from the time stamps. Then, the first frame time stamp and the last frame time stamp in the video data are read. The first frame time stamp is compared with the text start time, and the last frame time stamp is compared with the text end time. When the first frame time stamp is earlier than or equal to the text start time, and the last frame time stamp is later than or equal to the text end time, the video data is marked as video data synchronized with the script text data. If the first frame time stamp of the video data is later than the text start time, adjacent frame data is extracted backward based on the text start time. If the last frame time stamp of the video data is earlier than the text end time, adjacent frame data is extracted backward based on the text end time. The extracted adjacent frame data is concatenated with the original video data according to the frame time stamp order, and synchronization verification is performed again to form video data synchronized with the script text data. After the video data synchronized with the script text data is formed, a text-video synchronization association record is established with the script text data. This record records the correspondence between the script text data, the speaker information, the start and end times of the video playback, and the video data, providing a traceable time source for subsequent pixel displacement tracking of the video data. Therefore, the motion features extracted from the video data can be mapped to the target character and dialogue time position in the script text data, allowing the identity markers and motion features to jointly participate in determining the quantification of relationship closeness within the same plot timeframe.

[0043] Optionally, the step of performing pixel displacement tracking on the video data to extract the motion features of the target character includes: The video data is processed frame by frame to obtain a time-series image sequence; Contour recognition is performed in the time-series image sequence to locate the center point of the facial contour representing the target character; Joint recognition is performed in the time-series image sequence to locate the center point of the torso representing the target character; Calculate the first displacement vector of the center point of the facial contour in two adjacent temporal image sequences; Calculate the second displacement vector of the torso center point in two adjacent temporal image sequences; Spatial displacement data is generated based on the first displacement vector and the second displacement vector; The semantic action tags that match the spatial displacement data are queried in the pre-configured action tag mapping library to generate the action features of the target character based on the semantic action tags. The pre-configured action tag mapping library stores the mapping relationship between spatial displacement data and semantic action tags.

[0044] Preferably, the specific implementation process of extracting images frame by frame from the video data to obtain a time-series image sequence is as follows: After the video data has established a time synchronization relationship with the script text data, the frame timestamps, encoded frame order, and picture frame data in the video data are read first. The picture frame data is then checked for order based on the frame timestamps to ensure that the arrangement of the picture frame data corresponds to the timestamps in the script text data. The encoded frame order represents the compressed arrangement order of the picture frame data in the video data, and the frame timestamps represent the playback time order of the picture frame data in the video data. Therefore, if there is a discrepancy between the encoded frame order and the frame timestamps, the picture frame data is rearranged according to the frame timestamps so that the subsequently formed time-series image sequence corresponds to the order in which the lines occur in the script text data. The technical essence of the video data is that it is continuous frame data extracted from the main video track of the multimedia video file according to the start and end times of the playback. The video data carries the scene state when the script text data corresponds to the dialogue. Therefore, the video data is the visual source for subsequently extracting the action features of the target character. In subsequent processing, the visual source continues to participate in the localization of the facial contour center point and the torso center point through the temporal image sequence. Specifically, when performing frame-by-frame image extraction, each frame data is first read according to the frame time marker of the video data, and then the frame data is decoded to convert each frame data into a single frame image that can be used for pixel displacement tracking. Subsequently, each single frame image is bound to the corresponding frame time marker to form an image time binding record. The image time binding record is used to record the one-to-one correspondence between the single frame image and the frame time marker, and is used to maintain the temporal sequence relationship between adjacent frames when calculating the first displacement vector and the second displacement vector. The technical essence of the temporal image sequence is that it is an image sequence formed by multiple single-frame images arranged in the order of the frame time markers. The temporal image sequence preserves both the continuity of the video data and the temporal synchronization relationship between the video data and the script text data. In subsequent processing, the temporal image sequence is used to locate the center point of the facial contour and the center point of the torso, and further provides the pixel coordinate source for the first displacement vector and the second displacement vector. By first converting the video data into the temporal image sequence, and then performing subsequent recognition processing on the temporal image sequence, the changes in the target character's image are no longer limited to the entire video, but are broken down into pixel-level image data with a temporal order, thus providing a stable image foundation for subsequently converting the character's action changes into the spatial displacement data.

[0045] Preferably, in the specific technical implementation of contour recognition in the time-series image sequence to locate the center point of the facial contour representing the target character, each single-frame image in the time-series image sequence is first read from the image time binding record, and grayscale component extraction, boundary gradient calculation, and connected component filtering are performed on each single-frame image to form candidate facial contour regions. The grayscale component extraction is used to convert the color pixels in the single-frame image into grayscale pixels that are easy to read for boundary changes. The boundary gradient calculation is used to extract pixel boundaries with large brightness changes from the grayscale pixels. The connected component filtering is used to filter out candidate facial contour regions from the pixel boundaries that have continuous shapes and areas matching the outer edge of the face. The technical essence of contour recognition is to find the boundary pixel coordinates corresponding to the outer edge of the target character's face in the single-frame image, and extract the pixel center coordinates that can be used for continuous tracking from the boundary pixel coordinates. Contour recognition is not an abstract description of the image semantics, but rather a frame-by-frame processing of pixel grayscale changes, boundary gradient changes, and connected component shapes. In the specific positioning process, the target character to be tracked in the time-series image sequence is first determined based on the pronunciation subject information and identity identifier characters in the script text data. Then, candidate facial contour regions corresponding to the target character are selected in each single-frame image, and the boundary pixel coordinates of the candidate facial contour regions are read. The boundary pixel coordinates are then centered to form the facial contour center point. The technical essence of the facial contour center point is the pixel center coordinates of the target character's facial contour region in a single-frame image. The facial contour center point is used to characterize the changing position of the target character's head in the time-series image sequence. To enable the facial contour center point to continuously participate in calculations between adjacent frames, the facial contour center point corresponding to each frame is bound to the frame time marker, forming a facial contour center point time-series record. This facial contour center point time-series record directly provides the pixel center coordinates of the facial contour center point between adjacent frames when subsequently calculating the first displacement vector, allowing the first displacement vector to reflect the direction and magnitude of the target character's head movement in the image. Therefore, the contour recognition transforms the boundary pixel coordinates in the temporal image sequence into quantifiable facial contour center points, thereby providing a basis for extracting the action features of the target character from changes in screen movement.

[0046] Preferably, the specific implementation process of joint recognition in the time-series image sequence to locate the center point of the torso representing the target character is as follows: After the time-series record of the facial contour center point has been formed, each single-frame image in the time-series image sequence is read from the image time binding record, and pixel connectivity analysis is performed on the upper limb connection region, shoulder connection region, and torso outer edge region of the target character in each single-frame image to form candidate torso joint regions. The upper limb connection region is used to exclude image interference regions that are discontinuous with the torso region of the target character, the shoulder connection region is used to provide a connection reference between the head position and the torso position of the target character, and the torso outer edge region is used to define the pixel distribution range of the target character's body in the single-frame image; the upper limb connection region, the shoulder connection region, and the torso outer edge region together form the candidate torso joint regions after pixel connectivity analysis. The technical essence of the joint recognition is to extract candidate torso joint regions that represent the torso position based on the pixel connection relationships of the target character's body structure in a single-frame image. This joint recognition does not subjectively judge the character's movements, but rather determines the center position of the target character's torso in the image through the pixel position relationships of the candidate torso joint regions. Specifically, during localization, the center point of the facial contour is first used as a reference for the target character's head position in the current single-frame image. Then, the upper limb connection region, the shoulder connection region, and the outer edge region of the torso are read along the pixel region below the center point of the facial contour. Subsequently, the center point of the shoulder connection region and the outer edge region of the torso is calculated, and the calculated area range is verified in conjunction with the upper limb connection region to form the torso center point. The technical essence of the torso center point is the pixel center coordinates of the target character's torso region in a single-frame image. The torso center point is used to characterize the changing position of the target character's main body in the time-series image sequence. The reason for simultaneously locating the facial contour center point and the torso center point is that the head offset and body offset of the target character have different action meanings in the film / video footage. For example, if the facial contour center point shifts while the torso center point shifts less, it can reflect local action changes such as nodding, turning the head, or moving closer to the dialogue partner. If the facial contour center point and the torso center point shift in the same direction simultaneously, it can reflect the overall movement of the target character, moving closer to or away from another target character. To ensure that the torso center point can be used continuously in subsequent calculations, the torso center point corresponding to each frame is also bound to the frame time stamp, forming a torso center point time sequence record.The timing record of the torso center point directly provides the pixel center coordinates of the torso center point between two adjacent frames when calculating the second displacement vector, so that the second displacement vector can reflect the direction and magnitude of movement of the target character's body in the picture.

[0047] Preferably, in the specific implementation of calculating the first displacement vector of the facial contour center point in the temporal image sequence of two adjacent frames, the temporal record of the facial contour center point is first read, and the sequential relationship between the two adjacent frames is determined according to the frame time stamp; then, the previous frame facial pixel coordinates of the facial contour center point are read from the previous frame, and the subsequent frame facial pixel coordinates of the facial contour center point are read from the subsequent frame. Both the previous frame facial pixel coordinates and the subsequent frame facial pixel coordinates are derived from the temporal record of the facial contour center point. The previous frame facial pixel coordinates are used to characterize the head position of the target character in the previous frame, and the subsequent frame facial pixel coordinates are used to characterize the head position of the target character in the subsequent frame. The technical essence of the first displacement vector is two-dimensional pixel displacement data used to characterize the positional change of the center point of the target character's facial contour between two adjacent frames. The first displacement vector includes a horizontal facial displacement component, a vertical facial displacement component, and time interval information corresponding to the frame time marker. The horizontal facial displacement component represents the positional change of the center point of the facial contour in the horizontal direction of the image, the vertical facial displacement component represents the positional change of the center point of the facial contour in the vertical direction of the image, and the time interval information limits the occurrence of this positional change between two adjacent frames. Specifically, in calculation, the horizontal facial displacement component is formed by the horizontal difference between the facial pixel coordinates of the subsequent frame and the facial pixel coordinates of the previous frame, and the vertical facial displacement component is formed by the vertical difference between the facial pixel coordinates of the subsequent frame and the facial pixel coordinates of the previous frame. The horizontal facial displacement component, the vertical facial displacement component, and the time interval information are then encapsulated together as the first displacement vector. The first displacement vector does not directly represent a quantitative value of the relationship between the target characters, but rather serves as the source of head displacement for generating the subsequent spatial displacement data. In subsequent processing, the first displacement vector and the second displacement vector jointly participate in the formation of the spatial displacement data, enabling the spatial displacement data to simultaneously express the changes in the target character's head position and body position. By calculating the displacement of the facial contour center point in two adjacent frames, the head contour changes in the temporal image sequence can be transformed into the first displacement vector with direction and amplitude, thereby providing head motion components for subsequent querying of semantic action tags in the pre-configured action tag mapping library.

[0048] Preferably, the specific implementation process for calculating the second displacement vector of the torso center point in the temporal image sequence of two adjacent frames is as follows: After forming the temporal record of the torso center point, the torso center points corresponding to the two adjacent frames are read according to the frame time stamp, and the torso pixel coordinates of the torso center point in the previous frame are extracted from the previous frame, and the torso pixel coordinates of the torso center point in the following frame are extracted from the following frame. The torso pixel coordinates in the previous frame and the torso pixel coordinates in the following frame are both derived from the temporal record of the torso center point. The torso pixel coordinates in the previous frame are used to represent the position of the target character's main body in the previous frame, and the torso pixel coordinates in the following frame are used to represent the position of the target character's main body in the following frame. The technical essence of the second displacement vector is two-dimensional pixel displacement data used to characterize the positional change of the target character's torso center point between two adjacent frames. The second displacement vector includes a horizontal displacement component, a vertical displacement component, and time interval information corresponding to the frame time markers. The horizontal displacement component represents the positional change of the torso center point in the horizontal direction of the image, the vertical displacement component represents the positional change of the torso center point in the vertical direction, and the time interval information ensures that the second displacement vector and the first displacement vector are within the same adjacent frame interval. Specifically, the horizontal displacement component is formed by the horizontal difference between the torso pixel coordinates of the subsequent frame and the torso pixel coordinates of the preceding frame, and the vertical displacement component is formed by the vertical difference between the torso pixel coordinates of the subsequent frame and the torso pixel coordinates of the preceding frame. The horizontal displacement component, the vertical displacement component, and the time interval information are then encapsulated together as the second displacement vector. The second displacement vector supplements the body movement information that the first displacement vector cannot cover, enabling subsequent spatial displacement data to distinguish between local head movement and overall body movement. If the first displacement vector shows a significant directional change in the center point of the facial contour, while the second displacement vector shows a relatively small change in the center point of the torso, then the subsequent spatial displacement data can reflect local posture changes. If the first and second displacement vectors are close in direction and magnitude within the same adjacent frame interval, then the subsequent spatial displacement data can reflect the overall movement trend. By calculating the second displacement vector, the positional change of the target character's body in the frame can be converted into displacement data that can be processed together with the first displacement vector, thereby providing a basis for the body's movement in the subsequent generation of the motion features.

[0049] Preferably, the specific implementation process of generating spatial displacement data based on the first displacement vector and the second displacement vector, and querying semantic action tags that match the spatial displacement data in the pre-configured action tag mapping library to generate the action features of the target character based on the semantic action tags is as follows: After forming the first displacement vector and the second displacement vector in the same adjacent frame interval, the horizontal displacement component of the face, the vertical displacement component of the face, and the time interval information in the first displacement vector are read first, and the horizontal displacement component of the torso, the vertical displacement component of the torso, and the time interval information in the second displacement vector are read; then, the first displacement vector and the second displacement vector are checked in the same frame interval so that the first displacement vector and the second displacement vector involved in the processing come from the same two adjacent frames. The technical essence of the spatial displacement data is that it combines the head displacement change at the center point of the facial contour and the body displacement change at the center point of the torso to form character motion displacement description data. The spatial displacement data includes a head lateral displacement component, a head longitudinal displacement component, a torso lateral displacement component, a torso longitudinal displacement component, and time interval information. The head lateral displacement component originates from the facial lateral displacement component in the first displacement vector, the head longitudinal displacement component originates from the facial longitudinal displacement component in the first displacement vector, and the torso lateral displacement component and the torso longitudinal displacement component originate from the second displacement vector. The time interval information is used to characterize the frame interval corresponding to the above displacement components. After generating the spatial displacement data, a pre-configured action tag mapping library is read. This library is configured before the film and television text translation processing task starts and stores the mapping relationship between spatial displacement data and semantic action tags. The mapping relationship is determined by the spatial displacement range, the relative displacement relationship between the head and torso, and the time interval range. The spatial displacement range constrains the value range of each displacement component in the spatial displacement data. The relative displacement relationship between the head and torso characterizes the directional difference between changes in head position and changes in body position. The time interval range limits the length of adjacent frame intervals corresponding to the spatial displacement data, enabling the spatial displacement data to be mapped to semantic action tags such as nodding, turning, approaching, moving away, moving side-by-side, or colliding. Specifically, during a query, each displacement component in the spatial displacement data is matched item by item with the spatial displacement range in the pre-configured action tag mapping library. The matching semantic action tag is determined by combining the directional relationships between the head's lateral displacement component and the torso's lateral displacement component, and the head's longitudinal displacement component and the torso's longitudinal displacement component.The technical essence of the semantic action tagging is that it is a tag character derived from the spatial displacement data and categorized into readable actions. The semantic action tag is not directly used as translated text, but rather to form the action features of the target character. Subsequently, the semantic action tag is encapsulated with the corresponding frame time marker, the target character's identity identifier, and the spatial displacement data to generate the target character's action features. The technical essence of these action features is that they are character action data capable of participating in the subsequent determination of the quantifiable values ​​of relationship closeness. These action features convert pixel displacement changes in the video data into action expressions that can be processed together with the identity identifier, enabling the subsequent determination of the quantifiable values ​​of relationship closeness between different target characters based on changes in on-screen interaction.

[0050] Optionally, determining the quantitative value of the relationship between different target characters at plot nodes in the pre-configured plot timeline based on the identity identifier character and the action feature includes: The first timestamp is compared with the pre-configured story timeline to determine the story node hit by the first timestamp; The frequency of interaction of the action features between the two target characters at the stated plot node is statistically analyzed. The motion features are decomposed to extract the limb contact identifiers contained in the motion features; Based on the interaction frequency, the first intimacy evaluation character is retrieved from the pre-configured activity mapping table, wherein the pre-configured activity mapping table records the mapping relationship between the interaction frequency and the first intimacy evaluation character; Based on the physical contact identifier, relationship correction characters are extracted from a pre-configured correction table, wherein the pre-configured correction table records the mapping relationship between physical contact identifiers and relationship correction characters; Based on the first intimacy evaluation character and the relationship correction character, a combined string is generated, and based on the combined string, the relationship intimacy quantification value is generated.

[0051] Preferably, the specific implementation process of comparing the first timestamp with the pre-configured plot timeline is as follows: After the first timestamp has been formed by the time tags in the script text data, the first timestamp is read first, and the pre-configured plot timeline is read simultaneously. The technical essence of the pre-configured plot timeline is a time index structure used to organize the plot playback progress, plot nodes, and changes in character relationship status. The pre-configured plot timeline is not a simple list of times, but rather is configured after the multimedia video file has completed its encapsulation layer parsing, based on the time tags in the script text data, the start and end times of the video playback, and the time intervals between adjacent lines. In specific configuration, firstly, multiple candidate time boundaries are extracted according to the chronological order of multiple time tags in the script text data. Then, the candidate time boundaries are normalized to the same reference based on the start and end times of the video playback corresponding to the video data. This normalization is used to convert the time tags from the script text data and the start and end times of the video playback to the same time caliber, so that the candidate time boundaries and the frame time markers in the video data are on the same time caliber. Subsequently, the playback interval between adjacent candidate time boundaries is divided into plot nodes that can carry changes in the state of character relationships. The node start time, node end time, and node sequence marker of each plot node are written into the pre-configured plot timeline, so that the pre-configured plot timeline can form the node time range corresponding to each plot node through the node start time and node end time. The technical essence of the span comparison is to determine the inclusion relationship between the single time position of the first timestamp and the node time range corresponding to each plot node in the pre-configured plot timeline, rather than performing semantic matching between the first timestamp and the plot node. During the specific comparison, the first timestamp is first converted into a time unit with the same time range as the node. Then, the start time and end time of each node are read sequentially according to the arrangement order of the plot nodes in the pre-configured plot timeline, and it is determined whether the first timestamp falls between the start time and end time of the node. Through the above span comparison, the first timestamp is converted into a time positioning basis that can hit the plot node. The time positioning basis is then used to determine the plot node hit by the first timestamp, so that subsequent statistics on the interaction frequency of the action features, extraction of the body contact identifier, and generation of the relationship closeness quantification value all fall within the same plot time range.

[0052] Preferably, in the specific technical implementation of determining the plot node hit by the first timestamp, after comparing the span of the first timestamp with the pre-configured plot timeline, plot nodes that satisfy the inclusion relationship judgment are read, and these plot nodes are marked as plot nodes hit by the first timestamp. The technical essence of the plot node is a time segment in the pre-configured plot timeline used to carry a continuous plot scene, corresponding script text data, and the interactive state of the target character. The plot node is neither a simple line number nor a simple scene capture number, but rather an index unit used to unify the first timestamp, the script text data, the video data, and subsequent relationship state analysis into the same time range. Specifically, in determining this, the starting span between the first timestamp and the node start time of each plot node is read first, and then the ending span between the first timestamp and the node end time of each plot node is read. When the starting span is not less than zero and the ending span is not greater than zero, the corresponding plot node is determined as the plot node hit by the first timestamp. If the first timestamp falls at the boundary time position of two adjacent plot nodes, the system continues to read the pronunciation information in the script text data, the frame time markers in the video data, and the node time range of the two adjacent plot nodes. Based on the time distance between the first timestamp and the node time range of the two adjacent plot nodes, the plot node with the smaller distance from the first timestamp is selected as the plot node hit by the first timestamp. After the plot node hit by the first timestamp is determined, a plot node association record is established between the plot node hit by the first timestamp and the identity identifier character and the action feature. The plot node association record is used to limit the statistical range when subsequently calculating the interaction frequency of the action features between two target characters, and to provide the time attribution basis under the same plot node when subsequently extracting the body contact identifier, retrieving the first intimacy evaluation character, and generating the relationship closeness quantification value. Thus, the plot node hit by the first timestamp becomes an intermediate index connecting the time annotation results and the character relationship quantification processing, ensuring that the subsequent relationship closeness analysis does not deviate from the plot time position where the dialogue occurs.

[0053] Preferably, the specific implementation process of statistically analyzing the interaction frequency of the action features between two target characters under the plot node is as follows: After determining the plot node hit by the first timestamp, firstly, read the identity character and action feature corresponding to the plot node in the plot node association record, and locate the target character participating in the current dialogue or screen interaction in the script text data based on the identity character. The technical essence of the two target characters is that they are two target characters with dialogue association, the same screen frame range, or action interaction under the same plot node. The two target characters are not a randomly selected combination of characters, but are jointly defined by the identity character, the voice subject information, and the screen position changes in the video data. Specifically, the target character uttering the lines is first determined by the voice subject information in the script text data, and then other identity characters in the video data that are in the same screen frame range as the target character are read; when the target character corresponding to another identity character has a relative position change, approach trend, distance trend, or contact trend with the target character uttering the lines within the plot node, the target character corresponding to the other identity character is determined as the target character that forms an interactive relationship with the target character uttering the lines, thereby forming the two target characters. The technical essence of the interaction frequency is the frequency description data formed between the number of action-related events between the two target characters within the time range corresponding to the plot node and the node's time range. The interaction frequency is used to characterize the intensity of interaction between the two target characters at that plot node. Specifically, the frame time markers, semantic action tags, and spatial displacement data carried in the action features are first read. Then, action features belonging to the plot node are filtered according to the node's start and end times. Subsequently, the action features belonging to the plot node are paired according to the two target characters, and the number of occurrences of action-related events such as approaching, moving away, moving side-by-side, or colliding with the two target characters within the same frame time marker range is counted. The occurrence counts of these action-related events are then time-normalized with the plot node's time range to form the interaction frequency. The interaction frequency is subsequently retrieved from the pre-configured activity mapping table to generate the first intimacy evaluation character, allowing changes in the frequency of character interactions to participate in the formation of the quantified value of the relationship closeness.

[0054] Preferably, the specific implementation process of feature decomposition of the action features is as follows: After completing the interaction frequency statistics, the action features belonging to the plot node are read again, and the action features are split into fields to separate the identity identifier characters, frame time markers, spatial displacement data, and semantic action tags carried in the action features. The technical essence of the feature decomposition is to split the mixed encapsulated character source field, time source field, displacement source field, and action tag source field in the action features into multiple fields that can participate in subsequent judgments, rather than performing generalized classification of the action features. In specific implementation, the identity identifier characters in the action features are read first to determine the target character corresponding to the action features; then the frame time markers in the action features are read to determine which frame interval in the plot node the action features belong to; subsequently, the spatial displacement data in the action features is read to separate the head horizontal displacement component, head vertical displacement component, torso horizontal displacement component, torso vertical displacement component, and time interval information; finally, the semantic action tags in the action features are read to extract the tag characters corresponding to the character action category. The feature decomposition is performed before generating the relational affinity quantification value because the interaction frequency only reflects the density of interactions between the two target characters at the plot node, but cannot distinguish whether the interaction is approaching, moving away, moving side by side, or colliding. By performing feature decomposition on the action features, a field source can be provided for the subsequent extraction of the limb contact identifier, so that the relational affinity quantification value is not only formed based on changes in the number of interactions, but can also be corrected based on changes in the nature of the action. After completing the feature decomposition, the identity identifier character, the frame time marker, the spatial displacement data, and the semantic action tag are written into the action feature decomposition record. The action feature decomposition record is used to carry the decomposition results of the action features and provides an entry point for reading the semantic action tag and the spatial displacement data when extracting the limb contact identifier, so that the limb contact identifier can be formed sequentially from the action features, rather than appearing abruptly in subsequent steps.

[0055] Preferably, the specific implementation process for extracting the limb contact identifiers included in the action features is as follows: After forming the action feature decomposition record, the semantic action tags and spatial displacement data in the action feature decomposition record are first read, and it is determined whether the action features contain action categories related to limb contact based on the semantic action tags. The technical essence of the limb contact identifier is a character used to characterize whether the two target characters are in a state of physical proximity such as contact, close-up contact, collision, or side-by-side contact at the plot node. The limb contact identifier does not directly describe the evaluation result of the closeness or distance relationship, but is used as the basis for the action nature of extracting the relationship correction characters in the pre-configured correction table. During the extraction process, the semantic action tags are first read to determine if there are tags such as collision contact, side-by-side movement, approaching, or moving away. If the semantic action tags include collision contact, the head lateral displacement component, head longitudinal displacement component, torso lateral displacement component, and torso longitudinal displacement component in the spatial displacement data are read to determine whether the spatial displacement data corresponding to the two target characters show a relative displacement relationship under the same frame time marker, and the judgment result corresponding to the relative displacement relationship is written into the limb contact identifier. If the semantic action tags include side-by-side movement, the torso lateral displacement component and torso longitudinal displacement component corresponding to the two target characters are read to determine whether the main body displacement direction of the two target characters remains in the same direction, and the judgment result corresponding to the same direction change is written into the limb contact identifier. If the semantic action tags include moving away, the contact absence state corresponding to the moving away action is written into the limb contact identifier. After the limb contact identifier is formed, a limb contact association record is established with the two target characters, the plot node, and the action feature decomposition record. The limb contact association record is used to provide a reading basis when extracting the relationship correction character based on the limb contact identifier in the subsequent process, so that the relationship correction character can originate from the change in action nature in the action features. By extracting the physical contact markers, the quantitative value of the relationship closeness can be generated by combining both the interaction frequency and the contact nature between the two target characters, thereby reducing the problem of unclear relationship direction when judging the relationship solely based on the number of interactions.

[0056] Preferably, the specific implementation process of retrieving the first intimacy evaluation character from the pre-configured activity mapping table based on the interaction frequency is as follows: After the interaction frequency is formed, the pre-configured activity mapping table configured before the start of the film and television text translation processing task is read first. The technical essence of the pre-configured activity mapping table is to convert the interaction frequency into structured mapping data of the first intimacy evaluation character. The pre-configured activity mapping table is not an ordinary character relationship dictionary, but records the mapping relationship between the interaction frequency and the first intimacy evaluation character; wherein, the interaction frequency is used to describe the density of occurrence of action-related events of the two target characters under the plot node, and the first intimacy evaluation character is used to describe the preliminary intimacy level derived from the density of action-related events. Specifically, during pre-configuration, multiple interaction frequency intervals are configured first according to the node time range, action feature sampling granularity, and semantic action tag type of the plot node in the film and television text translation processing task; then, a corresponding first intimacy evaluation character is configured for each interaction frequency interval, so that different preliminary intimacy levels can be obtained when the interaction frequency falls into different interaction frequency intervals. The technical essence of the mapping relationship between the interaction frequency and the first intimacy evaluation character is to convert continuous frequency description data into evaluation characters that can participate in character combination processing. This mapping relationship provides a first component field for the subsequent generation of the combined string. Specifically, during retrieval, the interaction frequency is first read, and then compared item by item with the interaction frequency intervals in the pre-configured activity mapping table. When the interaction frequency falls into a certain interaction frequency interval, the first intimacy evaluation character corresponding to that interval is read. After the first intimacy evaluation character is formed, a first intimacy association record is established with the plot node, the two target characters, and the interaction frequency. This first intimacy association record is used to provide the first intimacy evaluation character when the combined string is subsequently generated, enabling the combined string to retain the initial intimacy information formed by the interaction frequency. Thus, the interaction frequency is converted into the first intimacy evaluation character through the pre-configured activity mapping table, providing a basic field for the subsequent generation of the relationship closeness quantification value together with the relationship correction character.

[0057] Preferably, the specific implementation process of extracting relationship correction characters from the pre-configured correction table based on the physical contact identifier is as follows: After the physical contact association record is formed, the physical contact identifier in the physical contact association record is first read, and the pre-configured correction table configured before the start of the film and television text translation processing task is read. The technical essence of the pre-configured correction table is to convert the physical contact identifier into structured mapping data for relationship correction characters. The pre-configured correction table records the mapping relationship between physical contact identifiers and relationship correction characters; wherein, the physical contact identifier is used to describe the contact nature between the two target characters, and the relationship correction character is used to supplement the relationship direction of the first intimacy evaluation character obtained from the interaction frequency. Specifically, during pre-configuration, multiple physical contact identifiers are first configured according to the action categories such as collision contact, side-by-side movement, approach, and distance in the semantic action tags, and then corresponding relationship correction characters are configured for each physical contact identifier, so that the same interaction frequency can form different relationship correction results under different contact natures. For example, when the interaction frequency is high but the body contact identifier represents collision contact, the relationship correction character can represent a hostile state; when the interaction frequency is high and the body contact identifier represents side-by-side contact, the relationship correction character can represent an intimate state; when the interaction frequency is low and the body contact identifier represents distance, the relationship correction character can represent a distant state. Specifically, during extraction, the body contact identifier is matched item by item with the body contact identifiers in the pre-configured correction table; when the body contact identifier meets the consistency matching condition with a certain body contact identifier in the pre-configured correction table, the relationship correction character corresponding to that body contact identifier is read. After the relationship correction character is formed, a relationship correction association record is established with the body contact identifier, the plot node, and the two target characters. The relationship correction association record is used to provide the relationship correction character when generating the combined string subsequently, so that the combined string can simultaneously contain the first intimacy evaluation character formed by the interaction frequency and the relationship correction character formed by the body contact identifier, thereby providing a more complete basis for expressing the character relationship for the quantified value of the relationship.

[0058] Preferably, in the specific technical implementation of generating a combined string based on the first intimacy evaluation character and the relationship correction character, the first intimacy evaluation character in the first intimacy association record is read first, and the relationship correction character in the relationship correction association record is read second; then, the plot node, the two target characters, and the first timestamp are read, and the above fields are used as the context source field of the combined string. The context source field is used to limit the character source and plot time source corresponding to the combined string. In this process, the first intimacy evaluation character is used to represent the initial intimacy level derived from the interaction frequency, and the relationship correction character is used to represent the relationship direction correction result derived from the physical contact identifier. The two come from different data processing paths, so they need to be combined before generating the relationship intimacy quantification value. The technical essence of the combined string is that it is a relationship encoding character formed by sequentially encapsulating the first intimacy evaluation character, the relationship correction character, the plot node, and the two target characters. The combined string is not the final relationship intimacy quantification value, but an intermediate carrier character used to generate the relationship intimacy quantification value. In the specific generation process, the identity identifiers corresponding to the two target characters are first sorted according to the interaction direction in the plot node to form a character order field; then, the node order marker of the plot node is written into the plot node field; subsequently, the first intimacy evaluation character and the relationship correction character are written in sequence, so that the combined string simultaneously retains the character source, plot time range, interaction activity level, and action correction direction. To avoid the field order in the combined string affecting subsequent reading, the field arrangement is fixed when generating the combined string, so that when the relationship closeness quantification value is generated based on the combined string, the corresponding information can be read from the same field position. After the combined string is formed, a combined string association record is established with the plot node hit by the first timestamp. The combined string association record is used to carry the combined string and provides a reading entry point when the relationship closeness quantification value is generated subsequently, so that the relationship closeness quantification value can inherit the processing results of the first intimacy evaluation character and the relationship correction character. By generating the combined string, the results of the interaction frequency path and the physical contact path can be merged into the same relationship encoding structure, providing a characterized intermediate result for subsequent quantification processing.

[0059] Preferably, the specific implementation process for generating the relationship closeness quantification value based on the combined string is as follows: After the combined string association record is formed, the combined string is first read, and the character order field, plot node field, first intimacy evaluation character, and relationship correction character are parsed according to the field arrangement of the combined string. The character order field is used to represent the interaction direction between the two target characters, the plot node field is used to represent the plot node corresponding to the combined string, the first intimacy evaluation character is used to represent the initial intimacy level obtained from the interaction frequency, and the relationship correction character is used to represent the relationship direction correction result obtained from the physical contact identifier. The technical essence of the relationship closeness quantification value is to represent the numerical result of the closeness, distance, or antagonism of the relationship between the two target characters under the plot node. The relationship closeness quantification value is not directly obtained from a single action tag, nor is it directly obtained from a single interaction number, but is formed by the joint participation of the first intimacy evaluation character corresponding to the interaction frequency and the relationship correction character corresponding to the physical contact identifier. In the specific generation process, the initial intimacy level is first read based on the first intimacy evaluation character, transforming the interaction density formed by the interaction frequency into a calculable initial quantification basis. Then, the corresponding correction direction is read based on the relationship correction character, adjusting the contact nature formed by the limb contact identifier to directionally adjust the initial quantification basis. Next, combining the character order field and the plot node field, the adjusted quantification result is written into the relationship quantification record corresponding to the two target characters and the plot node, thus forming the relationship intimacy quantification value. The relationship quantification record is used to store the correspondence between the relationship intimacy quantification value and the two target characters, the plot node, and the combined string, and provides a data source when arranging multiple relationship intimacy quantification values ​​in the subsequent time order according to the first timestamp. After the relationship intimacy quantification value is formed, it continues to participate in the generation of the relationship state trajectory, enabling the relationship intimacy quantification values ​​under multiple plot nodes to be organized in chronological order into trajectory data reflecting changes in character relationships. Therefore, the relationship closeness quantification value integrates the processing results of the identity identification character, the action feature, the interaction frequency, the physical contact identifier, the first intimacy evaluation character, the relationship correction character, and the combined string into the same quantification result, providing sortable relationship data with appendable time attributes for the subsequent construction of the dynamic role relationship graph.

[0060] Optionally, the step of extracting relationship correction characters from a pre-configured correction table based on the limb contact identifier includes: The physical contact markers are parsed to identify the type of behavioral action; When the behavior action type includes a physical collision behavior flag, the first adjustment character representing the hostile state is retrieved from the pre-configured correction table and used as the relationship correction character; When the behavior action type does not contain a physical collision behavior flag but contains a side-by-side sticking behavior flag, a second adjustment character representing the intimacy state is retrieved from the pre-configured correction table as the relationship correction character.

[0061] Preferably, the specific implementation process for parsing the limb contact identifier to identify the type of behavior action is as follows: After the limb contact association record has been formed, the limb contact identifier in the limb contact association record is read first, and the action feature decomposition record corresponding to the limb contact identifier is read simultaneously. The technical essence of the limb contact identifier is an action nature identifier character formed by the semantic action tag and spatial displacement data in the action features. The limb contact identifier is used to describe whether two target characters are close, colliding, side by side, or moving away from each other at the same plot node. When parsing the limb contact identifier, the character relationship is not determined directly based on a single tag character. Instead, the action tag field corresponding to the semantic action tag in the limb contact identifier is read first, and then the displacement relationship field corresponding to the spatial displacement data in the limb contact identifier is read. The action tag field and the displacement relationship field are then checked for consistency. The action tag field is used to characterize the source of the target character's action category in the video data, and the displacement relationship field is used to characterize the body displacement direction, head displacement direction, and relative approach trend of the two target characters at the same frame time marker. In specific implementation, the action label field first identifies label characters such as collision contact, side-by-side movement, approaching, or moving away. Then, the displacement relationship field determines whether the torso lateral displacement component, torso longitudinal displacement component, head lateral displacement component, and head longitudinal displacement component of the two target characters exhibit opposite, same, or opposite changes. Subsequently, the label characters corresponding to the action label field are cross-validated with the displacement change results corresponding to the displacement relationship field. If both point to collision contact, a behavior action type containing a physical collision behavior mark is generated; if both point to side-by-side contact, a behavior action type containing a side-by-side contact behavior mark is generated; and if both point to moving away, a behavior action type containing a moving away behavior mark is generated. The technical essence of the behavior action type is the action category result parsed from the limb contact mark, used to enter the pre-configured correction table for relationship direction correction. The behavior action type is used in subsequent processing to select the first adjustment character or the second adjustment character. Once the behavior action type is formed, the limb contact identifier, the behavior action type, the physical collision behavior mark, the side-by-side contact behavior mark, the distance behavior mark, and the corresponding plot node are written into the behavior action type parsing record, so that the relationship correction character can come from the limb contact identifier and the behavior action type under the same plot node, rather than from the character relationship data that has not been corrected by the limb contact identifier and the behavior action type.

[0062] Preferably, when the action type includes a physical collision behavior flag, in the specific technical implementation of retrieving the first adjustment character representing the hostile state from the pre-configured correction table as the relationship correction character, the action type in the action type parsing record is first read, and it is determined whether the action type includes the physical collision behavior flag. The physical collision behavior flag is essentially an action category flag used to represent the action category of two target characters moving towards each other, overlapping in contact, or colliding after their bodies rapidly approach each other at the same plot node. The physical collision behavior flag is not equivalent to a normal approaching action, but is jointly confirmed by the collision contact tag character in the semantic action tag and the relative displacement relationship in the spatial displacement data. To enable the physical collision behavior flag to be read by the subsequent relationship correction character extraction process, the physical collision behavior flag is written into the action type parsing record when the action type is formed. The action type parsing record stores the limb contact identifier, the action type, the physical collision behavior flag, and the corresponding plot node. The technical essence of the pre-configured correction table is structured correction data that records the mapping relationship between body contact markers, behavior action types, and relationship correction characters. This pre-configured correction table is configured before the film and television text translation processing task begins. Specifically, during configuration, firstly, based on the action categories that can be parsed from the semantic action tags and the spatial displacement data, behavior action type fields are configured, including the physical collision behavior marker, the side-by-side contact behavior marker, and the distance movement behavior marker. Then, a corresponding adjustment character field is configured for each behavior action type field, allowing different behavior action types to correspond to different relationship correction directions. Specifically, during retrieval, the physical collision behavior marker is first used as the search field to perform a matching query in the behavior action type fields of the pre-configured correction table. When the behavior action type field matches the physical collision behavior marker, the adjustment character field corresponding to that behavior action type field is read, and the character in the adjustment character field is used as the first adjustment character. The technical essence of the first adjustment character is a relationship correction character source field used to characterize a hostile state. The first adjustment character is used to correct the initial intimacy level formed by the first intimacy evaluation character, ensuring that high-frequency interaction but collision-related character relationships are not simply interpreted as close relationships. The first adjustment character is then written into the relationship correction association record, and the first adjustment character is used as the relationship correction character, so that the relationship correction character can participate in the subsequent generation of combined strings and the generation of relationship closeness quantification values.

[0063] Preferably, when the behavior action type does not contain a physical collision behavior flag but contains a side-by-side contact behavior flag, the specific implementation process of retrieving the second adjustment character representing the intimate state from the pre-configured correction table as the relationship correction character is as follows: After the behavior action type has been parsed, the behavior action type parsing record is read first, and the physical collision behavior flag in the behavior action type parsing record is excluded; when the physical collision behavior flag is not written in the behavior action type, the side-by-side contact behavior flag in the behavior action type parsing record is read again. The technical essence of the side-by-side contact behavior flag is an action category flag used to represent the movement of two target characters in the same plot node, where their bodies move in the same direction at close range or their adjacent positions are stably close. The side-by-side contact behavior flag comes from the side-by-side movement tag character in the semantic action tag and the same-direction displacement relationship in the spatial displacement data. The side-by-side contact behavior flag and the physical collision behavior flag have different technical meanings. The physical collision behavior flag emphasizes collision contact after opposite displacement, while the side-by-side contact behavior flag emphasizes the close relationship under same-direction displacement or adjacent contact states. Therefore, before retrieving the second adjustment character, the physical collision behavior flag needs to be excluded to avoid mismapping collision contact as an intimate state. Specifically, during retrieval, the side-by-side contact behavior flag is first used as the search field to perform a matching query in the behavior action type field of the pre-configured correction table. When the behavior action type field matches the side-by-side contact behavior flag, the adjustment character field corresponding to the behavior action type field is read, and the character in the adjustment character field is used as the second adjustment character. The technical essence of the second adjustment character is a relationship correction character source field used to characterize the intimate state. The second adjustment character is used to correct the direction of the initial intimacy level formed by the first intimacy evaluation character, so that the high-frequency side-by-side contact action state under the same plot node can be reflected as the direction of closeness in the subsequent relationship intimacy quantification value. The second adjustment character is then written into the relationship correction association record, and the second adjustment character is used as the relationship correction character, so that the relationship correction character can participate in the generation of the combined string together with the first intimacy evaluation character. By first excluding the physical collision behavior flag and then confirming the side-by-side bonding behavior flag, the source path of the relationship correction character can have a definite judgment order, reducing mapping conflicts between different behavior action types.

[0064] Preferably, the subsequent processing after the relationship correction character is formed is as follows: After the first adjustment character or the second adjustment character has been written into the relationship correction association record, the relationship correction character in the relationship correction association record is read, and the physical contact identifier, the behavior action type, the plot node, and the two target characters stored in the relationship correction association record are read simultaneously. The technical essence of the relationship correction character is that it is a relationship direction correction character obtained by mapping the behavior action type through the pre-configured correction table. The relationship correction character is used to correct the initial intimacy level obtained solely from the interaction frequency. The relationship correction character does not generate the relationship closeness quantification value independently, but rather participates in the quantification process as a component field of the subsequent combined string. In specific processing, firstly, the first intimacy association record formed when retrieving the first intimacy evaluation character based on the aforementioned interaction frequency is read. Then, the relationship correction character is matched with the first intimacy evaluation character in the first intimacy association record, so that the first intimacy evaluation character represents the interaction density, and the relationship correction character represents the nature of the interaction action. Next, the identity identifier characters corresponding to the two target characters and the node sequence markers corresponding to the plot nodes are read. The identity identifier characters, the node sequence markers, the first intimacy evaluation character, and the relationship correction character are written into the combined string association record according to the field arrangement criteria to generate a subsequently readable combined string. The field arrangement criteria are configured before generating the combined string. The field arrangement criteria are used to constrain the writing order of the identity identifier characters, the node sequence markers, the first intimacy evaluation character, and the relationship correction character in the combined string, so that corresponding information can be read from fixed field positions when parsing the combined string later. The associated record of the combined string is used to carry the combined string and to record whether the relationship correction character comes from the first adjustment character or the second adjustment character. When the relationship correction character comes from the first adjustment character, the combined string uses the relationship direction correction result of the hostile state when generating the relationship intimacy quantification value in the future. When the relationship correction character comes from the second adjustment character, the combined string uses the relationship direction correction result of the intimate state when generating the relationship intimacy quantification value in the future. Thus, the relationship correction character transmits the physical contact identifier, the behavior action type, the pre-configured correction table, and the processing results of the first adjustment character and the second adjustment character to the combined string, and further to the relationship intimacy quantification value, so that the dynamic role relationship graph can subsequently obtain role relationship data with action nature correction.

[0065] Optionally, constructing a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles includes: Establish a blank topological base map with a planar coordinate system; Map the multiple identity identifier characters to multiple entity nodes in the blank topology base map; The relationship state trajectory is parsed into directed edges with direction and label attributes; The directed edges are used to connect corresponding entity nodes that have interactive relationships, and the time sequence corresponding to the relationship state trajectory is attached to the directed edges as a time attribute to complete the construction of the dynamic role relationship graph.

[0066] Preferably, the specific implementation process of establishing a blank topological base map with a planar coordinate system is as follows: After the relationship state trajectories corresponding to multiple target characters have been formed, the identity identifier characters, plot nodes, relationship affinity quantification values, and the time sequence corresponding to the relationship affinity quantification values ​​carried in the relationship state trajectories are read first, and a graph construction carrying record is established. The graph construction carrying record is used to include the entity node sources, directed edge sources, and time attribute sources required for the subsequent construction of dynamic character relationship graphs into the same construction scope. The entity node sources correspond to the identity identifier characters, the directed edge sources correspond to the relationship state trajectories, and the time attribute sources correspond to the time sequence corresponding to the relationship affinity quantification values, thereby avoiding deviation from the relationship state trajectories when mapping the entity nodes, generating the directed edges, and attaching the time attributes in the subsequent process. Subsequently, the topology layout configuration record is read. This record is configured before the video text translation processing task starts. The topology layout configuration record includes the origin of the planar coordinate system, coordinate units, node spacing, edge connection detour, and time attribute writing entry. The origin defines the starting position of the planar coordinate system; the coordinate units define the position measurement of the entity nodes within the planar coordinate system; the node spacing defines the arrangement distance between adjacent entity nodes; the edge connection detour defines the edge connection position when multiple directed edges connect to the same entity node; and the time attribute writing entry defines the field position where the time sequence is subsequently written to the directed edges. The technical essence of the planar coordinate system is a coordinate index structure used to arrange the entity nodes and directed edges in a two-dimensional logical space. This planar coordinate system is not the actual spatial coordinates in the video frame, nor is it the pixel coordinates of the target character in the frame; rather, it provides the topology layout basis for the node positions and edge connection positions of the dynamic character relationship graph. In the specific establishment process, the starting position of the planar coordinate system is first configured according to the origin of the coordinates in the topology layout configuration record. Then, the arable range of nodes in the planar coordinate system is determined according to the number of identity identifier characters involved in the relationship state trajectory and the number of interaction relationships between the relationship state trajectories. Subsequently, multiple node coordinates to be mapped are generated within the arable range of nodes, and the node coordinates to be mapped are written into the graph construction carrying record. The technical essence of the blank topology base map is that it is a graph construction underlying data structure that only contains the planar coordinate system and the coordinates of the nodes to be mapped, and has not yet written the entity nodes and the directed edges. The blank topology base map provides the coordinate carrying basis for subsequently mapping the identity identifier characters to the entity nodes, and provides the connection carrying basis for subsequently parsing the relationship state trajectory into the directed edges.By first establishing the blank topological base map, the construction of the dynamic role relationship graph no longer directly stacks role relationship data, but first forms a topological foundation that can be arranged, connected, and have time attributes attached, so that the subsequent entity nodes, directed edges, and time attributes can all be organized in the same planar coordinate system.

[0067] Preferably, in the specific technical implementation of mapping multiple identity identifier characters to multiple entity nodes in the blank topological structure base map, all identity identifier characters participating in the relationship state trajectory are first read from the graph construction carrier record, and the identity identifier characters are deduplicated and sequentially arranged to form an identity identifier character mapping record. The technical essence of the multiple identity identifier characters is that they are character source characters used to identify different target roles, obtained by scanning the character sequence of the script text data. These multiple identity identifier characters are not ordinary dialogue characters, but rather the basic indexes for forming entity nodes in the subsequent dynamic role relationship graph. The identity identifier character mapping record is used to record the correspondence between each identity identifier character and the coordinates of a node to be mapped in the blank topological structure base map, and provides endpoint query basis when generating the directed edges subsequently. In the specific mapping process, the identity identifier characters are first read according to the order of the plot nodes in the relationship state trajectory. Then, the number of interactions of the identity identifier characters in different relationship state trajectories is combined to determine the node arrangement order of the identity identifier characters in the planar coordinate system. Subsequently, the coordinates of the node to be mapped corresponding to the node arrangement order are selected from the blank topology base map, and the coordinates of the node to be mapped are bound to the identity identifier characters to form the entity node. The technical essence of the entity node is that it is a node data unit used to represent the target character in the dynamic character relationship graph. The entity node includes at least an identity identifier character field, a node coordinate field, and a node order field. The identity identifier character field is used to maintain consistency with the character source of the target character, the node coordinate field is used to determine the position of the entity node in the planar coordinate system, and the node order field is used to determine the reading order when connecting the directed edges later. After the entity node is formed, it continues to be written into the graph construction carrying record and maintains a corresponding relationship with the identity identifier character mapping record, so that when parsing the relationship state trajectory later, the corresponding entity node can be directly found according to the identity identifier characters in the relationship state trajectory. Thus, the identity identifier character is converted from the character source character in the script text data into an entity node in the blank topological structure base map, enabling the dynamic character relationship graph to carry multiple target characters with the entity nodes, and providing a node basis for subsequently expressing the relationship state changes between different target characters.

[0068] Preferably, the specific implementation process of parsing the relationship state trajectory into directed edges with directional and label attributes is as follows: After the entity node has been written into the graph construction record, the relationship state trajectory is read again, and the relationship state trajectory is parsed to extract the two target characters, plot nodes, relationship closeness quantification values, combined strings, and time order corresponding to the relationship state trajectory. The technical essence of the relationship state trajectory is the relationship change trajectory data formed by arranging multiple relationship closeness quantification values ​​according to the time order of the first timestamp. The relationship state trajectory is used to describe the change order of the relationship closeness quantification values ​​between two target characters as the plot progresses. Therefore, the relationship state trajectory provides both the endpoint source of the directed edges and the source of the label attributes and time attributes of the directed edges. Specifically, during parsing, the identity identification characters corresponding to the two target characters are read from the relationship state trajectory, and the entity nodes corresponding to the two identity identification characters are found through the identity identification character mapping record; then, the character order field or interaction direction field in the relationship state trajectory is read, and the directional attribute is generated based on the character order field or the interaction direction field. The technical essence of the directional attribute is a data field used to characterize the connection direction of the directed edge from one entity node to another. The directional attribute can be determined based on the order in which the pronunciation subject information points to the communication object, or it can be determined based on the interaction direction of the two target characters in the relationship state trajectory. The directional attribute allows the directed edge to not only represent a relationship between two entity nodes, but also the source of the directed edge's direction within the story node. Subsequently, the relationship closeness quantification value, the first intimacy evaluation character, the relationship correction character, and the combined string are read from the relationship state trajectory, and these fields are encapsulated into the tag attribute. The technical essence of the tag attribute is a relationship state description field attached to the directed edge. The tag attribute is used to represent the relationship closeness quantification value, the relationship correction character, and the combined string carried by the directed edge, enabling the generation of target translated text based on the dynamic character relationship graph to read the relationship state corresponding to the current story node from the tag attribute. Based on the directional attribute and the tag attribute, the relationship connection data between the two entity nodes is encapsulated into a directed edge to form a directed edge with both directional and tag attributes. The technical essence of the directed edge is that it is an edge data unit that connects two entity nodes and carries the relationship state in the dynamic role relationship graph. The directed edge is then used to connect corresponding entity nodes that have an interactive relationship and to receive additional processing of the time attribute.

[0069] Preferably, in the specific technical implementation of connecting corresponding entity nodes with interactive relationships using the directed edges, the already formed directed edges are first read, and the direction attribute, label attribute, and identity identifier characters corresponding to the two target characters are parsed from the directed edges. Then, the entity nodes corresponding to the two identity identifier characters are found through the identity identifier character mapping record, and the two found entity nodes are marked as corresponding entity nodes. The technical essence of the corresponding entity nodes is a combination of entity nodes mapped from two identity identifier characters in the same relationship state trajectory. The corresponding entity nodes are not arbitrary two entity nodes, but a combination of nodes that have a common origin with the two target characters in the relationship state trajectory. The technical essence of the interactive relationship is that the two target characters have a relationship connection condition represented by the action features, the interaction frequency, the physical contact identifier, or the relationship closeness quantification value under the same plot node; when there are relationship closeness quantification values ​​corresponding to the two target characters in the relationship state trajectory, or when the combined string corresponding to the relationship state trajectory contains the identity identifier characters of the two target characters, it is determined that there is an interactive relationship between the two entity nodes. During the connection process, the starting and ending entity nodes of the directed edge are first determined based on the direction attribute. Then, the starting connection end of the directed edge is bound to the starting entity node, and the ending connection end is bound to the ending entity node. Subsequently, the tag attribute is appended to the edge attribute field of the directed edge, enabling the directed edge to carry the relationship closeness quantification value, the first intimacy evaluation character, the relationship correction character, and the combined string while connecting the corresponding entity nodes. If multiple relationship state trajectories exist for the same group of corresponding entity nodes under different story nodes, different directed edges are generated for each story node, or multiple edge attribute fields corresponding to different story nodes are recorded under the same directed edge, ensuring that the relationship states in different story nodes are not merged into indistinguishable static relationships. After the connection is completed, the corresponding entity node, the directed edge, and the tag attribute are written into the edge connection carrying record. This edge connection carrying record is used to provide a readable edge connection object when the time attribute is subsequently appended. Through the above connection processing, the directed edges connect the entity nodes with the relationship states of the relationship state trajectories, enabling the dynamic role relationship graph to express the interaction relationships between different target roles and the relationship states corresponding to the interaction relationships.

[0070] Preferably, the specific implementation process of attaching the time sequence corresponding to the relationship state trajectory as a time attribute to the directed edge to complete the construction of the dynamic character relationship graph is as follows: After the edge connection carrying record has been formed, first read the time sequence corresponding to the relationship state trajectory, and simultaneously read the directed edge, the corresponding entity node, and the tag attribute in the edge connection carrying record. The technical essence of the time sequence is the sequential arrangement of multiple relationship closeness quantification values ​​under the time sequence of the first timestamp. The time sequence is used to characterize the order in which the relationship state changes of the same group of target characters occur under different plot nodes. The technical essence of the time attribute is the time field attached to the directed edge. The time attribute is used to record the plot node, node start time, node end time, and node sequence mark corresponding to the relationship state carried by the directed edge, so that when the dynamic character relationship graph is queried later, the target plot node can be located according to the second timestamp, and the current state identifier can be read from the corresponding directed edge. Specifically, during the attachment process, the node sequence markers corresponding to the plot nodes are first read from the relationship state trajectory, followed by the node start and end times. Then, the node sequence markers, start and end times are written into the time attribute field of the directed edge, and a field association is established between the time attribute field and the tag attribute field, so that the tag attribute corresponding to a given time attribute can be read synchronously later. If the same directed edge carries the relationship states under multiple plot nodes, multiple time attribute fields are written sequentially according to the time order, and each time attribute field is associated with the tag attribute field under the corresponding plot node, thereby preserving the changing order of the relationship state trajectory as the plot progresses. After the time attribute attachment is completed, all entity nodes, all directed edges, all tag attributes, and all time attributes in the graph construction record are read, and node-edge indexing encapsulation processing is performed on the above data based on the planar coordinate system to form the dynamic character relationship graph. The node-edge index encapsulation process is used to write the node coordinate field of the entity node, the edge attribute field of the directed edge, the label attribute field, and the time attribute field into the same graph index structure. This allows subsequent retrieval of the dynamic character relationship graph according to the target plot node to read the corresponding entity node, corresponding directed edge, corresponding label attribute, and corresponding time attribute from the graph index structure. The technical essence of the dynamic character relationship graph is a character relationship data structure composed of entity nodes, directed edges, direction attributes, label attributes, and time attributes. The dynamic character relationship graph can distinguish the character relationship status within different time ranges according to plot nodes and provides a searchable basis for the relationship status in generating target translated text based on the dynamic character relationship graph.

[0071] Optionally, extracting the current state identifier of the target character under the target plot node from the dynamic character relationship graph includes: Read the status tag data of the target character corresponding to the target plot node in the dynamic character relationship graph; The status label data is parsed to extract the hierarchy indicator characters that represent hierarchical subordination and the sentiment indicator characters that represent sentiment inclination. Based on the hierarchical and subordinate indicator characters and the emotional tendency indicator characters, feature-encoded characters are generated, and the current status identifier of the target character is generated based on the feature-encoded characters.

[0072] Preferably, the specific implementation process for reading the status tag data of the target character corresponding to the target plot node in the dynamic character relationship graph is as follows: After the second timestamp has been compared with the plot nodes in the pre-configured plot timeline and the target plot node has been located, the graph index structure in the dynamic character relationship graph is read first, and the time attribute field corresponding to the target plot node is read from the graph index structure. The graph index structure originates from the node edge index encapsulation processing performed on the entity node, the directed edge, the tag attribute, and the time attribute in the previous step. The graph index structure is used to put the target plot node, the entity node, the directed edge, the tag attribute, and the time attribute into the same searchable data structure, and to provide a graph retrieval entry point for subsequent reading of the status tag data. Specifically, during reading, the time attribute corresponding to the target plot node is first searched in the time attribute field according to the node sequence mark, node start time, and node end time of the target plot node; then, the target entity node corresponding to the identity identifier character is searched in the entity nodes in the dynamic character relationship graph according to the identity identifier character. The technical essence of the target entity node is that it is the node-based representation of the target character in the dynamic character relationship graph. The target entity node does not directly represent a title or honorific; instead, it is used to locate directed edges related to the target character in the dynamic character relationship graph. Subsequently, directed edges connected to the target entity node and matching the time attribute of the target plot node are read and marked as target directed edges. The label attribute field of the target directed edge is then used to read the relationship closeness quantification value, first intimacy evaluation character, relationship correction character, combined string, direction attribute, and the relationship status field corresponding to the target plot node. The technical essence of the status label data is a combination of data fields extracted from the label attribute field and time attribute field of the target directed edge, used to describe the relationship status of the target character under the target plot node. The status label data is not a single character name, nor is it static character relationship text; rather, it includes a quantified relationship status including character origin, relationship direction, relationship closeness, relationship correction direction, and plot time position. After the status tag data is read, the target plot node, the target entity node, the target directed edge, and the status tag data are written into the status tag data reading record. This allows the graph reading results of the same target plot node and the same target character to be used when parsing the fields of the status tag data in the future. It also ensures that the hierarchical subordination indicator characters, emotional tendency indicator characters, feature encoding characters, and current status identifiers formed later can all be traced back to the status tag data.

[0073] Preferably, in the specific technical implementation of field parsing of the status tag data, after the status tag data reading record is formed, the status tag data in the status tag data reading record is read again, and the status tag data is subjected to field boundary identification and field type labeling according to the pre-formed tag attribute field arrangement in the dynamic character relationship graph. The reason why the status tag data needs to be parsed is that the status tag data simultaneously carries the relationship closeness quantification value, the first intimacy evaluation character, the relationship correction character, the combined string, and the direction attribute. If the status tag data is directly input into the subsequent rule base retrieval process, the rule base retrieval process cannot distinguish the hierarchical subordinate indication, emotional tendency indication, and relationship closeness degree of the target character under the target plot node. In the specific parsing process, the target entity node field in the status label data is first read, and the endpoint position of the target role in the directed edge of the target is determined based on the target entity node field. Then, the direction attribute field in the status label data is read, and the target entity node is determined to be on the side of the starting entity node or the ending entity node in the directed edge of the target based on the direction attribute field. Next, the label attribute field in the status label data is read, and the relationship affinity quantification value, the first intimacy evaluation character, the relationship correction character, and the combined string are separated from the label attribute field. The technical essence of this field parsing is to break down the graph endpoint information, relationship strength information, relationship correction information, and time attribution information mixed and encapsulated in the status label data into field-based data that can be read sequentially, rather than performing semantic summarization of the status label data. After completing the field parsing, the target entity node field, the direction attribute field, the relationship affinity quantification value, the first intimacy evaluation character, the relationship correction character, and the combined string are written into the status label data parsing record. The status tag data parsing record is used to carry the field parsing results of the status tag data, and provides the field source when extracting the hierarchy indicator character and the sentiment indicator character in the subsequent process, so that the hierarchy indicator character and the sentiment indicator character can be formed sequentially from the status tag data, and avoids the hierarchy indicator character or the sentiment indicator character from appearing alone outside the graph relationship data in the dynamic role relationship graph.

[0074] Preferably, the specific implementation process for extracting the hierarchy indicator character representing the hierarchical subordination indication is as follows: After the status label data parsing record has been formed, the target entity node field, direction attribute field, combined string, and relational closeness quantification value in the status label data parsing record are read first, and the node position of the target character in the target directed edge is determined according to the target entity node field. The technical essence of the hierarchy indicator character is a status character used to represent the relational direction of the target character relative to another target character under the target plot node. The hierarchy indicator character is not directly generated based on the title text itself, but is jointly determined based on the direction attribute, target entity node field, and relational closeness quantification value in the dynamic character relationship graph. In the specific extraction process, the direction attribute field is first read to determine the pointing relationship between the starting and ending entity nodes of the target directed edge. If the target entity node is located on the side of the starting entity node, the order of the identity identifier characters corresponding to the two target roles in the combined string is used to determine whether the target role is on the side of the speaker, the communication object, or the source of the relationship under the target plot node. If the target entity node is located on the side of the ending entity node, the role order field in the combined string is read again, and the target role is determined to be on the side of the addressed object or the relationship recipient based on the role order field. Subsequently, the relationship closeness metric value is read, and the node position judgment results are merged according to the relationship strength level of the relationship closeness metric value to form the hierarchy and subordination indicator characters. The state merging process converts the graph connection results (starting entity node side, ending entity node side, relationship source side, and relationship inheriting object side) into hierarchical subordinate indicator characters that can be read by translation and replacement rules. These hierarchical subordinate indicator characters express the hierarchical, peer, or unclear relationship direction of the target character under the target plot node. After the hierarchical subordinate indicator characters are formed, they are written into a hierarchical subordinate indicator record along with the target plot node, the target entity node, the target directed edge, and the state label data parsing record. This record provides the hierarchical subordinate indicator characters when generating feature-encoded characters, ensuring that the generated current state identifier retains the relationship direction information of the target character under the target plot node. It also allows the current state identifier to distinguish between different hierarchical subordinate indicator-related title and honorific expression update methods during subsequent rule base retrieval.

[0075] Preferably, the specific implementation process for extracting the emotion tendency indicator character representing the emotion tendency indication is as follows: After the subordinate indication record is formed, the first intimacy evaluation character, relationship correction character, relationship closeness quantification value, and combined string in the status label data parsing record are read, and the first intimacy evaluation character, the relationship correction character, the relationship closeness quantification value, and the combined string are associated with the target plot node. The technical essence of the emotion tendency indicator character is a status character used to represent the relationship tendency direction between the target character and another target character under the target plot node. The emotion tendency indicator character is not a word result obtained by judging the emotion of the dialogue content, but a data-based relationship tendency result formed by the relationship closeness quantification value, the first intimacy evaluation character, and the relationship correction character. Specifically, during extraction, the first intimacy evaluation character is read first to obtain the preliminary intimacy level formed by the interaction frequency conversion; then the relationship correction character is read to obtain the relationship direction correction result formed by the physical contact identifier conversion; subsequently, the relationship closeness quantification value is read to determine the direction of intimacy, distance, or opposition corresponding to the preliminary intimacy level and the relationship direction correction result after quantification. If the relationship correction character originates from a first adjustment character representing a hostile state, then when extracting the sentiment tendency indicator character, the relationship tendency corresponding to the relationship closeness / distance quantification value is merged into the opposing direction; if the relationship correction character originates from a second adjustment character representing a close state, then when extracting the sentiment tendency indicator character, the relationship tendency corresponding to the relationship closeness / distance quantification value is merged into the close direction; if the relationship correction character corresponds to a distance action or a contact absence state, then when extracting the sentiment tendency indicator character, the relationship tendency corresponding to the relationship closeness / distance quantification value is merged into the alienation direction. After the sentiment tendency indicator character is formed, the sentiment tendency indicator character, the hierarchy / subordination indicator character, the target plot node, the target entity node, and the status label data parsing record are written into the sentiment tendency indicator record. The emotional tendency indication record is used to provide the emotional tendency indication character when generating the feature-encoded character in the future, so that the feature-encoded character can simultaneously reflect the relationship direction and emotional tendency indication of the target character under the target plot node, avoiding the subsequent translation replacement rules from ignoring the relationship tendency differences reflected by collision contact, side-by-side contact or distance actions only based on the degree of closeness of the relationship, and enabling the current state identifier to distinguish the updated etiquette expression corresponding to the close direction, the distant direction or the opposing direction in the subsequent rule base retrieval process.

[0076] Preferably, the specific implementation process of generating feature-encoded characters based on the hierarchical subordination indicator characters and the emotional tendency indicator characters, and generating the current state identifier of the target character based on the feature-encoded characters, is as follows: After the hierarchical subordination indicator record and the emotional tendency indicator record have been formed, the hierarchical subordination indicator characters in the hierarchical subordination indicator record are read first, and the emotional tendency indicator characters in the emotional tendency indicator record are read simultaneously; then, the node sequence mark of the target plot node, the identity identifier character corresponding to the target entity node, and the quantified value of the relationship closeness are read, and the node sequence mark of the target plot node, the identity identifier character corresponding to the target entity node, and the quantified value of the relationship closeness are used as the encoding source field of the feature-encoded characters. The technical essence of the feature-encoded characters is that they are state index characters formed by sequentially encapsulating the hierarchical subordination indicator characters, the emotional tendency indicator characters, the target plot node, and the identity identifier character corresponding to the target character according to a pre-configured field arrangement. The feature-encoded characters are not the final translated text, nor are they title entries, but rather intermediate state characters used to generate the current state identifier. Specifically, when generating the feature-encoded characters, firstly, the identity identifier character corresponding to the target character is written into the character field, so that the feature-encoded characters retain the character's origin; then, the node sequence mark of the target plot node is written into the plot node field, so that the feature-encoded characters retain the time position of the target plot node; subsequently, the hierarchy indicator character and the emotional tendency indicator character are written sequentially, so that the feature-encoded characters simultaneously carry the hierarchy relationship and emotional tendency indication; finally, the relationship strength field corresponding to the relationship closeness quantification value is written, so that the feature-encoded characters can provide a basis for relationship strength for subsequent rule base retrieval. The technical essence of the current state identifier is the state index result generated based on the state label data of the target character under the target plot node. The current state identifier is used to extract translation replacement rules in the subsequent rule base, and to enable subsequent updates of title expressions and honorific expressions to use the relationship state corresponding to the target plot node. Specifically, when generating the current state identifier, the feature-encoded character is read, and a state field verification is performed on the feature-encoded character. The state field verification is used to check whether the role field, plot node field, hierarchy indicator character, emotional tendency indicator character, and relationship strength field in the feature-encoded character all originate from the state tag data parsing record. When the role field, plot node field, hierarchy indicator character, emotional tendency indicator character, and relationship strength field in the feature-encoded character all correspond to the state tag data parsing record, the feature-encoded character is encapsulated as the current state identifier.Once the current status identifier is formed, it is written into the translation processing record of the dialogue text to be translated. This allows the subsequent extraction of the translation replacement rule from the rule base based on the current status identifier to reuse the status tag data of the target character under the target plot node, the hierarchy indicator character, and the emotional tendency indicator character. This ensures that the titles and honorifics in the target translated text correspond to the current plot relationship status.

[0077] Optionally, the step of extracting the corresponding translation replacement rule from the rule base based on the current state identifier includes: The current status identifier is input into a pre-configured rule instruction library for matching and querying to obtain alternative control instructions. The pre-configured rule instruction library stores the mapping relationship between status identifiers and control instructions. Extract the control command that matches the current state identifier from the candidate control commands; The control instructions are parsed to obtain the honorific prefix supplementation instruction for adding honorific prefixes and the peer-level address reduction instruction for reducing the honorific level. The translation replacement rules are generated based on the honorific prefix supplementation instruction and the peer-level address dimensionality reduction instruction.

[0078] Preferably, the specific implementation process of inputting the current status identifier into a pre-configured rule instruction library for matching and querying to obtain alternative control instructions is as follows: After the current status identifier has been encapsulated by the feature-encoded characters, the role field, plot node field, hierarchy / subordination indicator character field, emotional tendency indicator character field, and relationship strength field in the current status identifier are read first. Then, the role field, plot node field, hierarchy / subordination indicator character field, emotional tendency indicator character field, and relationship strength field are processed according to the field arrangement of the current status identifier to generate a status query key. The technical essence of the pre-configured rule instruction library is to convert the current status identifier into structured instruction data that can be translated and replaced with executable control content. The pre-configured rule instruction library is not a regular dictionary, nor a fixed translation template, but rather instruction retrieval data that stores the mapping relationship between status identifiers and control instructions according to the field arrangement of the status identifier. The pre-configured rule instruction library is configured before the film and television text translation processing task starts. During configuration, the definition of status identifier fields is first determined. These fields include character fields, plot node fields, hierarchy / subordination indicator character fields, emotional tendency indicator character fields, and relationship strength fields. Then, corresponding control instructions are configured for different hierarchy / subordination indicator character fields, different emotional tendency indicator character fields, and different relationship strength fields. The correspondence between the status identifier field definitions and the control instructions is written into the pre-configured rule instruction library, enabling the current status identifier to be queried along the same status identifier field definition. During matching queries, the status query key is first compared item by item with the status identifier field definitions in the pre-configured rule instruction library. Candidate control instruction fields that satisfy the correspondence with the character field, plot node field, hierarchy / subordination indicator character field, emotional tendency indicator character field, and relationship strength field are read and encapsulated as alternative control instructions. The technical essence of the alternative control instructions is that they are candidate translation control data matched by the current status identifier in the pre-configured rule instruction library. These alternative control instructions are not the final control instructions used to generate the translation replacement rules, but rather intermediate instruction sources for subsequent field validation and instruction filtering based on the current status identifier. Through the above matching query, the dynamic relationship state in the current status identifier is converted into the alternative control instructions, enabling subsequent extraction of the control instructions to continue processing along the field source of the current status identifier, rather than directly generating address and honorific expressions based on static address entries.

[0079] Preferably, in the specific technical implementation where the pre-configured rule instruction library stores the mapping relationship between status identifiers and control instructions, the pre-configured rule instruction library first reads the status identifier field scope that can be expressed by the current status identifier during configuration, and then splits the status identifier field scope into a character field, a plot node field, a hierarchy / subordination indicator character field, an emotional tendency indicator character field, and a relationship strength field. The technical essence of the mapping relationship between status identifiers and control instructions is to convert the dynamic relationship state in the current status identifier into a data connection relationship corresponding to the translation processing action. The mapping relationship between status identifiers and control instructions is used to explain which control instruction should be invoked under different hierarchy / subordination indicator character fields, different emotional tendency indicator character fields, and different relationship strength fields, rather than simply recording a fixed translation term corresponding to a certain character name. In specific configuration, firstly, the subordinate indicator character field is configured with a direction for processing honorific prefixes, processing peer titles, or retaining honorifics; then, the emotional tendency indicator character field is configured with the corresponding adjustment of title expression for closeness, distance, or opposition; subsequently, the relationship strength field is configured with a strong / weak segmented reading approach, so that the same subordinate indicator character field can correspond to different control commands under different relationship strength fields. Each mapping relationship between a status identifier and a control command includes the status identifier field caliber, field matching order, field matching conditions, and a control command field. The field matching order is used to ensure that the current status identifier first limits the current plot scope according to the plot node field, then limits the target character source according to the character field, then limits the relationship state according to the subordinate and emotional tendency indicator character fields, and finally limits the replacement range according to the relationship strength field. The control command field is used to store the command content that can be subsequently parsed into the honorific prefix supplementation command and the peer title reduction command. After the pre-configured rule instruction library is configured, it also writes the mapping relationship between the status identifier and the control instruction into the rule instruction configuration record. This record is used to provide the status identifier field definition, the field matching order, and the field matching conditions during subsequent matching queries, enabling the candidate control instructions to be sequentially matched from the current status identifier. By configuring the mapping relationship between the status identifier and the control instruction as a field-based data connection relationship, the pre-configured rule instruction library can transform the dynamic relationship status provided by the dynamic role relationship graph into readable control instruction fields, providing a structured instruction source for the subsequent generation of the translation replacement rules.

[0080] Preferably, the specific implementation process for extracting control instructions matching the current state identifier from the candidate control instructions is as follows: After the candidate control instructions have been obtained through matching and querying the pre-configured rule instruction library, a candidate control instruction filtering record is first established, and the current state identifier, the state query key, the candidate control instructions, and the rule instruction configuration record are written into the candidate control instruction filtering record. In this process, the candidate control instructions are used to provide multiple candidate translation control directions. The technical essence of the control instructions is that they are selected from the candidate control instructions, have the same field source as the current state identifier, and can be subsequently parsed into specific replacement actions. Specifically, during the extraction process, firstly, the character field and the plot node field in the current status identifier are read, and candidate control instruction fields that do not correspond to the character field or the plot node field are filtered out from the candidate control instructions. Then, the hierarchy / subordination indicator character field and the emotion tendency indicator character field in the current status identifier are read, and these fields are compared item by item with the field matching conditions in the remaining candidate control instruction fields. Next, the relationship strength field in the current status identifier is read, and according to the strong / weak segmentation reading criteria in the rule instruction configuration record, candidate control instruction fields corresponding to the relationship strength field are selected from the remaining candidate control instruction fields. After the above field filtering, candidate control instruction fields that satisfy the character field, the plot node field, the hierarchy / subordination indicator character field, the emotion tendency indicator character field, and the relationship strength field are extracted as control instructions, and these control instructions are written into the control instruction extraction record. The control instruction extraction record is used to record the source field, matching conditions, and matching results of the control instruction, enabling the subsequent parsing of the control instruction to read the correspondence between the control instruction and the current state identifier. By extracting the control instruction from the candidate control instructions, multiple candidate translation control directions from the candidate control instructions can be avoided from entering subsequent processing simultaneously. This ensures that the subsequent honorific prefix supplementation instruction and the peer-level address dimensionality reduction instruction both originate from the same current state identifier, thereby maintaining the correspondence between the translation replacement rule and the current relationship state of the target character under the target plot node.

[0081] Preferably, in the specific technical implementation of parsing the control instructions to obtain the honorific prefix supplementation instruction for adding an honorific prefix and the peer-level address reduction instruction for lowering the honorific level, after the control instruction extraction record is formed, the control instructions in the control instruction extraction record are first read, and the control instructions are parsed according to the pre-configured control instruction field arrangement in the pre-configured rule instruction library. The reason why the control instructions need to be parsed is that the control instructions simultaneously contain the direction of address expression adjustment, the direction of honorific expression adjustment, the replacement position reading caliber, and the replacement character source field. If the instruction field is not parsed, it will be impossible to distinguish whether to perform honorific prefix supplementation or peer-level address reduction when processing the original address characters and original honorific characters in the text to be translated. In the specific analysis, the title processing field is first read from the control instructions, and it is determined whether the title processing field contains a processing direction for adding an honorific prefix. If the title processing field contains a processing direction for adding an honorific prefix, the prefix character source field, insertion position field, and applicable role field in the title processing field are read further, and the prefix character source field, insertion position field, and applicable role field are encapsulated into the honorific prefix supplementation instruction. The technical essence of adding the honorific prefix is ​​to insert a prefix character representing the honorific before the original title character in the text to be translated, so that the title expression can reflect the hierarchical subordinate indicator character field in the current status identifier. Subsequently, the honorific processing field is read from the control instructions, and it is determined whether the honorific processing field contains a processing direction for reducing the honorific level. If the honorific processing field contains a processing direction for reducing the honorific level, the omission particle source field, coverage position field, and applicable honorific field in the honorific processing field are read further, and the omission particle source field, coverage position field, and applicable honorific field are encapsulated into the peer-level address reduction instruction. The technical essence of the technique used to reduce the level of honorifics is to perform a replacement on the original honorific characters belonging to the category of polite particles, based on the hierarchy indicator character field and the sentiment indicator character field in the current state identifier, so that the honorific expression corresponds to the peer-level address relationship in the current plot relationship state. After the honorific prefix supplementation instruction and the peer-level address reduction instruction are formed, they are written into the control instruction parsing record. The control instruction parsing record is used to provide readable title processing fields, honorific processing fields, the honorific prefix supplementation instruction, and the peer-level address reduction instruction when the translation replacement rule is subsequently generated, so that the translation replacement rule can continue to be formed along the parsing result of the control instruction.

[0082] Preferably, the specific implementation process for generating the translation replacement rule based on the honorific prefix supplementation instruction and the peer-level address dimensionality reduction instruction is as follows: After the control instruction parsing record has been formed, the honorific prefix supplementation instruction and the peer-level address dimensionality reduction instruction in the control instruction parsing record are read first, and the current status identifier, the target plot node, and the identity identifier character corresponding to the target character are read simultaneously. The technical essence of the honorific prefix supplementation instruction is an instruction for updating the title expression by controlling the prefix insertion processing of the original title characters in the text to be translated. The honorific prefix supplementation instruction provides a prefix character source field, an insertion position field, and an applicable character field in subsequent processing. The technical essence of the peer-level address dimensionality reduction instruction is an instruction for updating the honorific expression by controlling the overwrite replacement processing of the original honorific characters in the text to be translated. The peer-level address dimensionality reduction instruction provides an omitted auxiliary word source field, an overwrite position field, and an applicable honorific field in subsequent processing. When generating the translation replacement rule, the current state identifier is first written into the rule source field, enabling the translation replacement rule to correspond to the current relationship state of the target character under the target plot node. Then, the honorific prefix supplementation instruction is written into the title replacement field, allowing the translation replacement rule to read the prefix character source field and the insertion position field when processing the original title character. Subsequently, the peer-level address dimensionality reduction instruction is written into the honorific replacement field, enabling the translation replacement rule to read the omitted particle source field and the overwrite position field when processing the original honorific character. Finally, the identity identifier characters corresponding to the target plot node and the target character, along with the control instruction extraction record, are written into the rule application field to form the translation replacement rule. The technical essence of the translation replacement rule is text replacement control data formed by parsing the control instructions corresponding to the current state identifier. The translation replacement rule is not a directly generated target translation text, but rather a rule-based processing basis used for subsequent updates to title expressions and honorific expressions. After the translation replacement rules are formed, they are written into the translation processing record of the dialogue text to be translated, and provide a readable source of instructions when the original title characters and honorific characters in the dialogue text to be translated are replaced or overwritten using the translation replacement rules. Through the above processing, a sequential relationship is formed between the current state identifier, the pre-configured rule instruction library, the alternative control instructions, the control instructions, the honorific prefix supplementation instructions, the peer-level address dimensionality reduction instructions, and the translation replacement rules, so that the title expressions and honorific expressions in the target translation text can be generated according to the current relationship state determined by the dynamic role relationship graph.

[0083] Optionally, the step of replacing the original appellation characters in the text to be translated using the translation replacement rules to update the appellation expression includes: Lexical scanning is performed on the text of the dialogue to be translated to mark the first position of the original appellation character belonging to the category of personal pronouns; The honorific prefix supplementation instruction in the translation replacement rule is triggered, and a prefix character representing the honorific is inserted before the first position to update the original title character, thereby forming the title expression; The process of using the translation replacement rules to overwrite and replace the original honorific characters in the text to be translated, in order to update the honorific expressions, includes: Lexical scanning is performed on the text of the dialogue to be translated to mark the second position of the original honorific characters belonging to the category of polite particles; The translation replacement rule triggers the peer-level address reduction instruction, which uses the omission of auxiliary words to cover the original honorific character at the second position, thereby updating the original honorific character and forming an honorific expression.

[0084] Preferably, the specific implementation process of performing lexical scanning on the text to be translated to mark the first position of the original title character belonging to the category of personal pronouns is as follows: After the translation replacement rule has been written into the translation processing record of the text to be translated, the text to be translated, the current status identifier, the target plot node, the identity identifier character corresponding to the target character, and the translation replacement rule are first read from the translation processing record, and the honorific prefix supplementation instruction is read from the title replacement field of the translation replacement rule. The technical essence of the lexical scanning is to sequentially identify and mark the position of character segments in the text to be translated according to character boundaries, word boundaries, punctuation boundaries, and language lexical rules, so that the subsequent honorific prefix supplementation instruction can locate the original title character that needs to be inserted, rather than performing a general replacement of the text to be translated. In specific implementation, the text to be translated is first converted into a character sequence that retains the original character order, and a character order mark is written for each character in the character sequence. Then, according to the word segmentation boundary of the language to which the text to be translated belongs, the character sequence is divided into segments to form a segment sequence. The segment sequence is used to carry multiple segments obtained from the text to be translated. Each segment retains a start character order mark and an end character order mark, so that when marking the first position, it can return to the original character position in the text to be translated. In order to identify the original title characters belonging to the category of personal pronouns, a personal pronoun category identification table is pre-configured. Before the film and television text translation processing task starts, the personal pronoun category identification table is configured with fields based on personal pronouns, role titles, kinship titles, job titles, and commonly used titles in subtitles in the target language. The personal pronoun category identification table includes a title character field, an applicable language field, a role reference field, and a position mark field. During lexical scanning, the word segments in the dialogue segment sequence are first compared item by item with the title character field in the personal pronoun category identification table. Then, combined with the role field in the current status identifier and the identity identifier character corresponding to the target role, it is determined whether the word segment points to the target role or the target role's communication partner. When a word segment simultaneously matches the title character field and the role pointing field, the word segment is marked as the original title character belonging to the personal pronoun category, and the starting character sequence of the word segment in the dialogue character sequence is marked as the first position. The technical essence of the first position is the addressable character position of the original title character in the dialogue text to be translated. The first position is not the original title character itself, but the positioning basis for the subsequent prefix insertion processing.After the first position is formed, the original title character, the first position, the identity identifier character corresponding to the target role, and the honorific prefix supplementation instruction are written into the title character positioning record. The title character positioning record is used to provide the insertion position source when the honorific prefix supplementation instruction is triggered in the future, so that the update of the title expression can continue to be executed along the current state identifier and the translation replacement rule.

[0085] Preferably, in the specific technical implementation of triggering the honorific prefix supplementation instruction in the translation replacement rule to insert a prefix character representing an honorific title before the first position to update the original title character and thus form the title expression, after the title character positioning record is formed, the original title character and the first position in the title character positioning record are read first, and the prefix character source field, insertion position field, and applicable role field in the honorific prefix supplementation instruction are read simultaneously. The technical essence of inserting a prefix character representing an honorific title before the first position is to convert the hierarchical relationship determined by the current state identifier into the prefix character in the title expression without changing the role pointing relationship of the original title character, so that the target translated text can reflect the role relationship state under the target plot node at the title level. The prefix character source field is used to provide the prefix character to be inserted, the insertion position field is used to limit the prefix character to be written before the first position, and the applicable role field is used to limit the honorific prefix supplementation instruction to be applied to the identity identifier character corresponding to the target role or the identity identifier character corresponding to the communication object of the target role. During the insertion process, the applicable role field is first compared with the identity identifier character corresponding to the target role in the title character positioning record. When a correspondence exists between the applicable role field and the identity identifier character corresponding to the target role, the prefix character in the prefix character source field is read, and the prefix character is confirmed to be written before the first position based on the insertion position field. Subsequently, the preceding character fragment before the first position, the original title character corresponding to the first position, and the following character fragment after the first position in the dialogue character sequence are read. The preceding character fragment, the prefix character, the original title character, and the following character fragment are sequentially concatenated according to the original character order to form a title character insertion record. Both the preceding character fragment and the following character fragment originate from the dialogue character sequence. The preceding character fragment is used to retain the character content that has not been replaced before the first position, and the following character fragment is used to retain the character content that has not been replaced after the first position. The preceding character fragment and the following character fragment, together with the prefix character and the original title character, maintain the original character order of the dialogue text to be translated in the title character insertion record. The title character insertion record is used to carry the title update result after inserting the prefix character, and records the correspondence between the prefix character source field, the first position, and the original title character, so as to avoid the inability to trace the source of the prefix character when generating the title expression later. Then, the original title character after inserting the prefix character is read from the title character insertion record, and the original title character after inserting the prefix character is marked as the title expression.The technical essence of the title expression is that the title update result is formed by performing prefix insertion processing on the original title character by the honorific prefix supplementation instruction in the translation replacement rule. The title expression is subsequently written into the translation processing record of the text to be translated and maintains a corresponding relationship with the current status identifier, so that the subsequent update of honorific expressions can continue to be processed under the same target plot node and the same target character's tone.

[0086] Preferably, the specific implementation process of performing lexical scanning on the text to be translated to mark the second position of the original honorific characters belonging to the category of politeness particles is as follows: After the title expression has been written into the translation processing record of the text to be translated, the text to be translated, the title expression, the current status identifier, the target plot node, and the translation replacement rule are read, and the peer-level title reduction instruction is read from the honorific replacement field of the translation replacement rule. The lexical scanning here follows the character order marking of the aforementioned text character sequence and text segment sequence, but the scanning target is changed from the original title characters belonging to the category of personal pronouns to the original honorific characters belonging to the category of politeness particles, so that the original title characters and the original honorific characters can be located and updated separately in the same text to be translated. The technical essence of the original honorific characters is that they are character fragments in the text to be translated used to express the degree of politeness, honorific level, or tone of respect. These original honorific characters are not equivalent to the original title characters. They are used in subsequent overwrite and replacement processing to form an honorific expression corresponding to the current state identifier. To identify the original honorific characters belonging to the category of polite particles, a polite particle category identification table is pre-configured. This table is field-based based on common polite particles, sentence-ending honorifics, honorific conjunctions, and tone-level characters in the target language before the film and television text translation processing task begins. The polite particle category identification table includes honorific character fields, applicable language fields, honorific level fields, and overwrite permission fields. During the specific scanning process, each segment in the dialogue segment sequence is first read, and each segment is compared item by item with the honorific character field in the politeness particle category identification table. When a segment matches the honorific character field, the hierarchy indicator character field, sentiment indicator character field, and relationship strength field in the current status identifier are read, and it is determined whether the peer-level address dimensionality reduction instruction allows the segment to be overwritten and replaced. If the honorific level field corresponding to the segment corresponds to the applicable honorific field in the peer-level address dimensionality reduction instruction, and the overwriting permission field corresponding to the segment allows the overwriting and replacement process, then the segment is marked as the original honorific character belonging to the politeness particle category, and the starting character sequence of the segment in the dialogue character sequence is marked as the second position. The technical essence of the second position is the addressable character position of the original honorific character in the dialogue text to be translated. The second position is used to limit the character writing range of the omitted auxiliary word covering the original honorific character. After the second position is formed, the original honorific character, the second position, the peer-level address dimensionality reduction instruction, and the title expression are written into the honorific character positioning record, so that the subsequent update of the honorific expression can be in the same translation processing record as the already formed title expression.

[0087] Preferably, in the specific technical implementation of triggering the peer-level address reduction instruction in the translation replacement rule, using omitted auxiliary words to cover the original honorific character at the second position to update the original honorific character, thereby forming an honorific expression, after the honorific character positioning record is formed, the original honorific character and the second position in the honorific character positioning record are first read, and the omitted auxiliary word source field, the coverage position field, and the applicable honorific field in the peer-level address reduction instruction are read simultaneously. The technical essence of using omitted auxiliary words to cover the original honorific character at the second position is to replace the original honorific character in the text to be translated that originally expressed a higher level of honorifics based on the peer-level address relationship or close relationship represented by the current state identifier, so that the honorific expression corresponds to the relationship state under the target plot node. The omission particle source field is used to provide an omission particle to replace the original honorific character. The omission particle can be configured as an empty placeholder character, a weakened tone character, or a character with no output. The coverage position field is used to limit the omission particle to cover the original honorific character at the second position. The applicable honorific field is used to limit the range of original honorific characters that the peer-level address dimensionality reduction instruction can process. Specifically, during the coverage replacement, the original honorific character in the honorific character positioning record is first compared with the applicable honorific field. When there is a correspondence between the original honorific character and the applicable honorific field, the omission particle in the omission particle source field is read, and the coverage start and end points corresponding to the second position are determined according to the coverage position field. Then, the honorific preceding character fragment, the original honorific character at the second position, and the honorific following character fragment after the second position in the dialogue character sequence are read. The honorific preceding character fragment, the omission particle, and the honorific following character fragment are sequentially concatenated according to the original character order to form an honorific character coverage record. Both the honorific prefix character fragment and the honorific suffix character fragment originate from the dialogue character sequence. The honorific prefix character fragment is used to retain the character content that has not been covered before the second position, and the honorific suffix character fragment is used to retain the character content that has not been covered after the second position. The honorific prefix character fragment and the honorific suffix character fragment, together with the omitted auxiliary word, maintain the original character order of the dialogue text to be translated in the honorific character coverage record. The honorific character coverage record is used to carry the honorific update result after the omitted auxiliary word covers the original honorific character, and records the correspondence between the omitted auxiliary word source field, the second position, and the original honorific character, so that the update source of the honorific expression can be read when the target translation text is subsequently generated.Subsequently, the honorific update result after overwriting the original honorific characters is read from the honorific character overwriting record, and the honorific update result after overwriting the original honorific characters is marked as the honorific expression. The technical essence of the honorific expression is that the honorific update result is formed by performing overwriting and replacement processing on the original honorific characters by the peer-level address dimensionality reduction instruction in the translation replacement rule. The honorific expression and the title expression are jointly written into the translation processing record of the dialogue text to be translated, so that the dialogue text to be translated can simultaneously complete the title expression update and the honorific expression update.

[0088] Preferably, after both the title expression and the honorific expression have been formed, the translation processing record of the dialogue text to be translated is read again, and the title character insertion record, the honorific character overwrite record, the title expression, the honorific expression, the current state identifier, and the target plot node are read from the translation processing record. The title expression and the honorific expression correspond to two different types of text update results in this process. The title expression comes from the prefix insertion processing of the original title character by the honorific prefix supplementation instruction, and the honorific expression comes from the overwrite replacement processing of the original honorific character by the peer-level address dimensionality reduction instruction. Both are constrained by the current state identifier, but their text positions, character sources, and replacement methods are different. When generating the target translation text, the character order markers in the text to be translated are first rearranged and verified based on the first position in the title character insertion record and the second position in the honorific character overlay record. This ensures that the character length change after the prefix character insertion does not affect the character position reading when the omitted auxiliary word covers the original honorific character. Subsequently, according to the original character order of the text to be translated, the original title characters that have formed the title expression and the original honorific characters that have formed the honorific expression are written into the same updated dialogue character sequence. The updated dialogue character sequence is used to carry the dialogue character content after the title expression and the honorific expression are completed simultaneously, and retains the correspondence between the current status identifier, the target plot node, and the identity identifier character corresponding to the target character. To enable the source of the title expression and the honorific expression to be traced when reading the target translation text later, the honorific prefix supplementation instruction corresponding to the title expression, the peer-level address reduction instruction corresponding to the honorific expression, and the translation replacement rule are also written into the target translation text generation record. The target translated text generation record is used to carry the correspondence between the updated dialogue character sequence, the translation replacement rule, and the current state identifier, and provides a readable text generation basis when outputting the target translated text. Subsequently, the updated dialogue character sequence is encapsulated in character order to form the target translated text. The technical essence of the target translated text is that, based on the dialogue text to be translated, the current state identifier formed by the dynamic character relationship graph is used, and the original title characters and the original honorific characters are updated respectively through the translation replacement rule to obtain the translated text result. Through the above processing, the first position of the original title character, the second position of the original honorific character, the honorific prefix supplementation instruction, the peer-level address dimensionality reduction instruction, the title expression, and the honorific expression all form a sequential relationship in the same translation processing record, enabling the target translated text to reflect the character relationship state under the target plot node.

[0089] like Figure 3 As shown, this embodiment of the present application provides a translation address and honorific selection device based on a dynamic role relationship graph, which includes: The script video acquisition module is used to acquire script text data and extract video data that is time-synchronized with the script text data; The identity extraction module is used to scan the script text data for character sequences to extract at least one identity character of the target character; The timestamp annotation module is used to read the timestamps attached to the script text data and annotate the timestamps as the first timestamp; The motion feature extraction module is used to perform pixel displacement tracking on the video data in order to extract the motion features of the target character; The relationship closeness determination module is used to determine the quantitative value of the relationship closeness between different target characters at plot nodes in the pre-configured plot timeline based on the identity identification characters and the action features. The relationship trajectory generation module is used to arrange multiple relation closeness quantification values ​​according to the time order of the first timestamp to generate a relationship state trajectory; The relationship graph construction module is used to construct a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles; The translation text generation module is used to obtain the dialogue text to be translated, and generate the target translation text based on the dynamic character relationship graph.

[0090] like Figure 4 As shown, an electronic device according to an embodiment of this application includes a processor and a memory. The memory stores a computer program. When the processor executes the computer program, it implements the translation title and honorific selection method based on dynamic role relationship graph as described in any one of the claims of this application.

[0091] like Figure 5 As shown, this is a computer-readable storage medium according to an embodiment of the present application. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the translation title and honorific selection method based on dynamic role relationship graph as described in any one of the present applications.

[0092] like Figure 6 As shown, this is an embodiment of the present application of a translation title and honorific selection system based on a dynamic role relationship graph, which includes a script video processing terminal, a role relationship processing terminal, a relationship graph construction terminal, and a translation text processing terminal; The script video processing terminal is used to acquire script text data and extract video data that is time-synchronized with the script text data; The script video processing terminal is also used to perform character sequence scanning on the script text data to extract at least one target character's identity identifier character, and to read the time tag attached to the script text data to mark the time tag as a first timestamp; The character relationship processing terminal is used to perform pixel displacement tracking on the video data in order to extract the motion features of the target character; The character relationship processing terminal is also used to determine the quantitative value of the relationship between different target characters at plot nodes in the pre-configured plot timeline based on the identity identification characters and the action features; The relationship graph construction end is used to arrange multiple relationship closeness quantification values ​​according to the time order of the first timestamp, so as to generate a relationship state trajectory; The relationship graph construction terminal is also used to construct a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles; The translation text processing terminal is used to obtain the dialogue text to be translated, and to generate the target translation text based on the dynamic character relationship graph.

[0093] The above Figures 3-6 For an exemplary description, please refer to the above. Figure 1 This will not be elaborated upon here.

Claims

1. A method for selecting a title and a polite expression based on a dynamic role relationship graph, characterized in that, include: Acquire script text data and extract video data that is time-synchronized with the script text data; The script text data is scanned for character sequences to extract at least one identifier character for a target character; Read the timestamps attached to the script text data and mark the timestamps as the first timestamp; Pixel displacement tracking is performed on the video data to extract the motion features of the target character; Based on the identity identifier characters and the action features, determine the quantitative values ​​of the relationship between different target characters at plot nodes in the pre-configured plot timeline; Arrange multiple relational closeness quantification values ​​according to the time sequence of the first timestamp to generate a relational state trajectory; Based on the relationship state trajectories corresponding to multiple target roles, a dynamic role relationship graph is constructed. Obtain the dialogue text to be translated, and generate the target translation text based on the dynamic character relationship graph.

2. The method of claim 1, wherein, The step of obtaining the dialogue text to be translated, and generating the target translation text based on the dynamic character relationship graph, includes: Extract the second timestamp carried by the text of the dialogue to be translated; The second timestamp is compared with the plot nodes in the pre-configured plot timeline to locate the target plot node that matches the second timestamp. In the dynamic character relationship graph, extract the current state identifier of the target character under the target plot node; Based on the current state identifier, extract the corresponding translation replacement rule from the rule base; The original title characters in the text to be translated are replaced using the translation replacement rules to update the title expression; The original honorific characters in the text to be translated are replaced using the translation replacement rules to update the honorific expressions, thereby obtaining the target translated text.

3. The method of claim 1, wherein, The acquisition of script text data includes: Read the global script character stream from a multimedia video file; Identify dialogue boundary markers in the global script character stream; Based on the dialogue boundary symbol, a character segment containing the pronunciation subject information is extracted from the global script character stream to serve as the script text data; The extraction of video data synchronized with the script text data in time includes: Extract the start and end times of the video playback associated with the pronunciation subject information; Based on the start and end times of the video playback, video data synchronized with the script text data is extracted from the main video track of the multimedia video file.

4. The method of claim 1, wherein, The step of performing pixel displacement tracking on the video data to extract the motion features of the target character includes: The video data is processed frame by frame to obtain a time-series image sequence; Contour recognition is performed in the time-series image sequence to locate the center point of the facial contour representing the target character; Joint recognition is performed in the time-series image sequence to locate the center point of the torso representing the target character; Calculate the first displacement vector of the center point of the facial contour in two adjacent temporal image sequences; Calculate the second displacement vector of the torso center point in two adjacent temporal image sequences; Spatial displacement data is generated based on the first displacement vector and the second displacement vector; The semantic action tags that match the spatial displacement data are queried in the pre-configured action tag mapping library to generate the action features of the target character based on the semantic action tags. The pre-configured action tag mapping library stores the mapping relationship between spatial displacement data and semantic action tags.

5. The method of claim 1, wherein, The step of determining the quantitative values ​​of the relationship between different target characters at plot nodes in the pre-configured plot timeline based on the identity identifier characters and the action features includes: The first timestamp is compared with the pre-configured story timeline to determine the story node hit by the first timestamp; The frequency of interaction of the action features between the two target characters at the stated plot node is statistically analyzed. The motion features are decomposed to extract the limb contact identifiers contained in the motion features; Based on the interaction frequency, the first intimacy evaluation character is retrieved from the pre-configured activity mapping table, wherein the pre-configured activity mapping table records the mapping relationship between the interaction frequency and the first intimacy evaluation character; Based on the physical contact identifier, relationship correction characters are extracted from a pre-configured correction table, wherein the pre-configured correction table records the mapping relationship between physical contact identifiers and relationship correction characters; Based on the first intimacy evaluation character and the relationship correction character, a combined string is generated, and based on the combined string, the relationship intimacy quantification value is generated.

6. The method of claim 5, wherein, The step of extracting relationship correction characters from a pre-configured correction table based on the limb contact identifier includes: The physical contact markers are parsed to identify the type of behavioral action; When the behavior action type includes a physical collision behavior flag, the first adjustment character representing the hostile state is retrieved from the pre-configured correction table and used as the relationship correction character; When the behavior action type does not contain a physical collision behavior flag but contains a side-by-side sticking behavior flag, a second adjustment character representing the intimacy state is retrieved from the pre-configured correction table as the relationship correction character.

7. A translation address and honorific selection device based on a dynamic role relationship graph, characterized in that, include: The script video acquisition module is used to acquire script text data and extract video data that is time-synchronized with the script text data; The identity extraction module is used to scan the script text data for character sequences to extract at least one identity character of the target character; The timestamp annotation module is used to read the timestamps attached to the script text data and annotate the timestamps as the first timestamp; The motion feature extraction module is used to perform pixel displacement tracking on the video data in order to extract the motion features of the target character; The relationship closeness determination module is used to determine the quantitative value of the relationship closeness between different target characters at plot nodes in the pre-configured plot timeline based on the identity identification characters and the action features. The relationship trajectory generation module is used to arrange multiple relation closeness quantification values ​​according to the time order of the first timestamp to generate a relationship state trajectory; The relationship graph construction module is used to construct a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles; The translation text generation module is used to obtain the dialogue text to be translated, and generate the target translation text based on the dynamic character relationship graph.

8. An electronic device, comprising: It includes a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the translation title and honorific selection method based on dynamic role relationship graph as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the translation title and honorific selection method based on a dynamic role relationship graph as described in any one of claims 1 to 6.

10. A translation address and honorific selection system based on a dynamic role relationship graph, characterized in that, This includes the script video processing end, the character relationship processing end, the relationship graph construction end, and the translated text processing end; The script video processing terminal is used to acquire script text data and extract video data that is time-synchronized with the script text data; The script video processing terminal is also used to perform character sequence scanning on the script text data to extract at least one target character's identity identifier character, and to read the time tag attached to the script text data to mark the time tag as a first timestamp; The character relationship processing terminal is used to perform pixel displacement tracking on the video data in order to extract the motion features of the target character; The character relationship processing terminal is also used to determine the quantitative value of the relationship between different target characters at plot nodes in the pre-configured plot timeline based on the identity identification characters and the action features; The relationship graph construction end is used to arrange multiple relationship closeness quantification values ​​according to the time order of the first timestamp, so as to generate a relationship state trajectory; The relationship graph construction terminal is also used to construct a dynamic role relationship graph based on the relationship state trajectories corresponding to multiple target roles; The translation text processing terminal is used to obtain the dialogue text to be translated, and to generate the target translation text based on the dynamic character relationship graph.

Images