Game translation collaborative management and monitoring method, system, equipment and medium

Abstract

The invention discloses a game translation collaborative management and monitoring method, system and device and a medium. The method specifically comprises the steps that a term library version tree is constructed according to preprocessed game resources; based on the term library version tree, game texts are classified and marked according to game contexts, a game context vector space is constructed in combination with game screenshots, a 3D scene model and audio dialogues, and the game context vector space is used for generating a unique context fingerprint for each segment of game text; constructing an ability graph based on historical translation data of translators, performing intelligent task allocation in combination with text categories identified by context fingerprints and context features, and recommending an optimal translator combination for a specific game item; and in combination with a pre-constructed translation risk prediction model, marking a high-risk text and automatically generating an alternative translation method. According to the method, a comprehensive and efficient game translation collaborative management and monitoring system is constructed, and the quality and efficiency of game translation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, system, device, and medium for collaborative management and monitoring of game translation. Background Technology

[0002] With the booming development of the global game market, game localization has become a crucial link for the game industry to expand into international markets and enhance the gaming experience for global users. Currently, game localization involves various file types, covering UI text, story scripts, item descriptions, achievement systems, and more. However, traditional translation management systems have revealed many limitations when addressing the needs of game localization, specifically as follows: 1. File Type Handling Challenges: Traditional translation management systems struggle to handle the diverse file types involved in games uniformly. Different file types have unique formats and structures, such as Unity's .prefab files, which traditional systems cannot directly and effectively translate. This leads to formatting errors or information loss during translation, affecting the game's integrity and playability.

[0003] 2. Terminology Management Challenges: Game terminology is highly industry-specific, and existing systems lack a professional game-specific terminology database for management. Inconsistencies in game terminology across different languages ​​lead to differences in semantic expression and style in translation results, disrupting the game's integrity and consistency, and reducing the player's gaming experience.

[0004] 3. Lack of Multilingual Progress Collaboration Management: The demand for simultaneous release of multiple language versions is increasingly strong, but traditional systems lack effective multilingual progress collaboration management mechanisms. The progress of multiple language versions is opaque, communication between translation teams for different languages ​​is poor, and it is difficult to coordinate work progress, affecting the global simultaneous release of the game and causing missed market opportunities.

[0005] 4. Original format corruption issue: Game text often contains variables and code snippets. Traditional translation tools are prone to corrupting the original format when processing this text, leading to abnormal game functions or display errors, which increases the workload and cost of subsequent repairs.

[0006] 5. Inefficient Incremental Translation Management: Frequent game updates lead to inefficient incremental translation management during version iterations. Traditional systems cannot effectively identify version changes, resulting in high translation duplication rates, wasting significant human and time resources, and reducing the efficiency and quality of translation work.

[0007] 6. Lack of context-assisted translation: The lack of auxiliary translation support for game context (characters, scenes) makes it difficult for translators to accurately understand the specific situation in which the game text is located, resulting in translation results that do not match the actual content of the game, affecting the accuracy of the translation and the immersion of the game.

[0008] 7. Stylistic Consistency Challenge: The game's diverse text types are managed in a fragmented manner, making it difficult to maintain stylistic consistency. The lack of unified stylistic guidance during the translation of different text types results in inconsistent overall language style, reducing the game's cultural quality and appeal.

[0009] 8. Insufficient Quality Inspection Rules: The game lacks specific quality inspection rules for its text, such as character length limits. Traditional quality inspection methods cannot meet the unique requirements of game localization, resulting in translations that do not conform to game specifications in terms of format and content, thus affecting the normal operation of the game and the user experience. Summary of the Invention

[0010] The purpose of this invention is to provide a method, system, device, and medium for collaborative management and monitoring of game translation, thereby constructing a comprehensive and efficient collaborative management and monitoring system for game translation, significantly improving the quality and efficiency of game translation, and solving at least one of the aforementioned problems in the prior art.

[0011] In a first aspect, the present invention provides a method for collaborative management and monitoring of game translation, the method specifically comprising: Game resources are extracted and preprocessed using a dedicated parsing plugin for the game engine. A terminology version tree is then constructed based on the preprocessed game resources, and the terminology version tree is bound and associated with the game version number. Based on the terminology version tree, game text is classified and labeled according to game context, and a game context vector space is constructed by combining game screenshots, 3D scene models and audio dialogues. The game context vector space is used to generate a unique context fingerprint for each piece of game text. A capability map is constructed based on translators' historical translation data. Intelligent task allocation is performed by combining text categories and contextual features identified by contextual fingerprints to recommend the optimal translator combination for specific game projects. In the process of performing context-assisted translation based on the optimal translator combination, the visualization reference provided by the game context vector space is used, combined with a pre-built translation risk prediction model, to mark high-risk texts and automatically generate alternative translations; The translated text is quality checked through the integrated GQA model in the real-time collaborative monitoring interface, and the results of the GQA model are fed back to the terminology version tree.

[0012] Secondly, the present invention provides a game translation collaborative management and monitoring system, the system specifically comprising: The first processing module is used to extract and preprocess game resources through a dedicated parsing plugin for the game engine, build a terminology version tree based on the preprocessed game resources, and bind and associate the terminology version tree with the game version number. The second processing module is used to classify and label game text based on the terminology version tree and the game context, and to construct a game context vector space by combining game screenshots, 3D scene models and audio dialogues. The game context vector space is used to generate a unique context fingerprint for each piece of game text. The third processing module is used to construct a capability map based on the translator's historical translation data, and to intelligently allocate tasks by combining the text categories and contextual features identified by the context fingerprint, and recommend the optimal translator combination for a specific game project. The fourth processing module is used to mark high-risk texts and automatically generate alternative translations during the context-assisted translation process based on the optimal translator combination, by utilizing the visual references provided by the game context vector space and combining them with a pre-built translation risk prediction model. The fifth processing module is used to perform quality checks on the translated text through the integrated GQA model in the real-time collaborative monitoring interface, and to feed back the GQA model's check results to the terminology version tree.

[0013] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, it implements the game translation collaborative management and monitoring method as described in any of the above methods.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the game translation collaborative management and monitoring method as described in any of the above methods.

[0015] Compared with the prior art, the present invention has at least one of the following technical effects: 1. This invention constructs a comprehensive and efficient collaborative management and monitoring system for game translation, which significantly improves the quality and efficiency of game translation.

[0016] 2. This invention extracts and preprocesses game resources through a dedicated parsing plugin for game engines, enabling unified processing of multiple file types, effectively protecting the original format of variables and code snippets in game text, avoiding format corruption issues caused by traditional translation tools, and ensuring the integrity and playability of the game.

[0017] 3. This invention constructs a terminology version tree and binds it to the game version number, establishing a terminology change tracking mechanism between versions. When the game version is updated, it can automatically identify terminology differences, trigger translation memory updates, ensure the consistency of game terminology across languages, and improve the accuracy and professionalism of translation.

[0018] 4. This invention constructs a capability map based on translators' historical translation data and uses contextual fingerprinting for intelligent task allocation, recommending the optimal translator combination for specific game projects. Simultaneously, it enables transparent management of multilingual progress through a real-time collaborative monitoring interface, promoting information communication and collaborative work among translation teams of different languages ​​and improving the efficiency of simultaneous release of multilingual versions.

[0019] 5. This invention utilizes a terminology change tracking mechanism to accurately identify incremental content during game version iterations, avoid translation duplication, improve the efficiency of incremental translation management, and reduce manpower and time costs.

[0020] 6. This invention constructs a game context vector space, generating a unique context fingerprint for each piece of game text and providing a visual reference. Translators can utilize this immersive translation context to accurately understand the meaning of the game text, improving the accuracy and quality of their translations.

[0021] 7. This invention, by classifying and marking game texts and translating them in conjunction with contextual features, can maintain stylistic consistency across various types of game texts, thereby enhancing the cultural quality and appeal of the game.

[0022] 8. This invention integrates a GQA model into the real-time collaborative monitoring interface to perform quality checks on translated text and feeds the results back to the terminology version tree. By establishing a traceability path from quality issues to the root causes of terminology, it is possible to analyze the scope of the impact of terminology changes on translated text, formulate progressive update strategies, continuously improve translation quality, and meet the unique quality check requirements of game text. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a collaborative management and monitoring method for game translation provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a game translation collaborative management and monitoring system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0025] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0026] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0027] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0028] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0029] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0030] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0031] In this application embodiment, the entity executing the process includes a terminal device. This terminal device includes, but is not limited to, devices capable of executing the methods disclosed in this application, such as servers, computers, smartphones, and tablets. Figure 1 A flowchart illustrating a game translation collaborative management and monitoring method according to an embodiment of the present invention is shown below: S101 extracts and preprocesses game resources through a dedicated parsing plugin for the game engine, constructs a terminology version tree based on the preprocessed game resources, and binds and associates the terminology version tree with the game version number.

[0032] In this embodiment, a dedicated parsing plugin is developed for the Unity engine. This plugin has the ability to deeply parse Unity project files, and can identify and process various resource files in the Unity project, including but not limited to .prefab files (prefab files, used to store configuration information of game objects), .scene files (scene files, containing the layout and elements of the game scene), .asset files (resource files, such as textures, audio, etc.), and script files (such as C# scripts).

[0033] During game development, when localization translation is required, a dedicated parsing plugin for the game engine is activated. The plugin first scans the Unity project directory, identifying all resource files containing text information. For .prefab files, the plugin parses UI text, item descriptions, and other text content; for .scene files, it extracts dialogue text and prompts from the scene; and for script files, it analyzes variable names, function names, and other text that may be relevant to translation. Simultaneously, the plugin records the resource file path of each text file, its location in the game (e.g., specific scene, UI interface), and the functional module it belongs to (e.g., story system, item system). In this way, all translatable text resources in the game are comprehensively and accurately extracted.

[0034] The extracted game resources may have inconsistent formats and encodings, requiring preprocessing. First, the text is encoded to ensure all text uses a consistent encoding format, such as UTF-8, to avoid garbled characters on different platforms and tools. Then, the text is cleaned, removing unnecessary spaces, line breaks, and other irrelevant characters to standardize the text format. Next, the text is segmented, breaking long sentences into appropriate lexical units based on the characteristics of game terminology and translation needs, facilitating subsequent terminology extraction and management. For example, "magic shield generator" is broken down into words like "magic," "shield," and "generator." Furthermore, text containing variables and code snippets is specially marked and processed to ensure its original format and functionality are not compromised during subsequent translation.

[0035] After preprocessing the game resources, a terminology version tree is constructed based on the preprocessed text information. First, all extracted and preprocessed terms are categorized into primary categories according to the game's functional modules, such as plot terms, item terms, and UI terms. Under each primary category, further subcategories are created based on term type; for example, plot terms are categorized into character names, dialogue text, and quest descriptions; item terms are categorized into weapon names, equipment attributes, and consumable descriptions. This multi-level categorization method forms a hierarchical terminology database structure. Then, based on the game's version iterations, an independent branch is created for each version, recording any added, modified, or deleted terms under that branch, thus forming the terminology version tree. For example, if the initial game version is 1.0, and during the development of version 1.1, some item descriptions were modified and some new plot dialogues were added; these changes would be recorded under the 1.1 branch of the terminology version tree.

[0036] To ensure consistency and traceability between the terminology library version tree and the game version, the terminology library version tree is bound to the game version number. During game development, each time a version is released, the current game version number is recorded and associated with the corresponding branch of the terminology library version tree in the terminology library management system. For example, when version 1.2 is released, the branch corresponding to version 1.2 is found in the terminology library management system and bound to the game version number "1.2". This allows for quick location of the corresponding terminology library version through the game version number during subsequent translation and management processes. This facilitates accurate terminology information for translators and administrators, and also enables comparison and analysis of game terminology across different versions, ensuring consistency and accuracy in translation.

[0037] In this embodiment, the process of extracting and preprocessing game resources through a dedicated parsing plugin for the game engine, constructing a terminology version tree, and binding the terminology version tree with the game version number is realized, providing a solid foundation for game localization translation.

[0038] S102, based on the terminology version tree, classify and label the game text according to the game context, and construct a game context vector space by combining game screenshots, 3D scene models and audio dialogues. The game context vector space is used to generate a unique context fingerprint for each piece of game text.

[0039] In this embodiment, game text information is obtained from a pre-constructed terminology version tree. The terminology version tree is meticulously categorized according to game version and functional modules, including different branches such as the initial version and subsequent updates. Each branch is further subdivided into multiple functional modules such as plot, items, quests, and UI. By traversing the terminology version tree, each piece of game text can be accurately located, and its corresponding game version, functional module, and other basic information can be obtained. For example, under the plot functional module of a certain updated version, a text about the protagonist's dialogue with a mysterious NPC can be found, along with the corresponding game version number and plot category.

[0040] After acquiring basic text information, the text is further categorized and labeled in greater detail based on the game context. Game context encompasses multiple aspects, including the scene where the text appears (e.g., town, dungeon, wilderness), the characters involved (protagonist, NPCs, monsters, etc.), and the text's function (dialogue, hints, quest descriptions, etc.). For example, the text describing the protagonist's dialogue with the mysterious NPC is labeled "Ancient Ruins Entrance" because the dialogue takes place at the entrance to an ancient ruin; the characters involved are the protagonist and the mysterious NPC, each labeled separately; and since the text guides the player into the ruins, its function is labeled "Guiding Dialogue." This categorization and labeling allows for a more comprehensive description of the game environment and usage scenarios of the text.

[0041] To construct a rich game context vector space, it is necessary to collect game screenshots, 3D scene models, and audio dialogues related to the game text. For each piece of game text, its specific location within the game development environment is identified, and a screenshot containing the interface displaying that text is captured. For example, for the dialogue text at the entrance of the ancient ruins mentioned above, a screenshot of the interface when the protagonist talks to the mysterious NPC is captured, showing the surrounding environment and character appearance. Simultaneously, a 3D scene model of the scene is obtained to understand the environment in which the text is situated from a spatial and layout perspective. Furthermore, audio dialogues related to the text are recorded, including character voices and background sound effects, to enrich the contextual information from an auditory perspective.

[0042] The collected game screenshots, 3D scene models, and audio dialogues are integrated with the categorized and labeled game text to construct a game context vector space. First, feature extraction is performed on the game screenshots, analyzing visual features such as color, shape, and object distribution, and converting them into numerical vectors. For example, the main color values, character and object coordinates, and other information are extracted from the screenshots to form visual feature vectors. For 3D scene models, features such as geometric shape, texture, and lighting are extracted and converted into spatial feature vectors. For example, the shape of buildings, ground texture, and the direction and intensity of lighting are analyzed. For audio dialogues, audio feature analysis is performed, extracting features such as pitch, volume, speech rate, and background sound effects to form auditory feature vectors. Then, these visual, spatial, and auditory feature vectors are fused with the categorized and labeled information of the game text to construct a multi-dimensional game context vector space. In this space, each dimension represents a contextual feature, and different game texts occupy different positions in the space according to their contextual features.

[0043] After the game context vector space is constructed, a unique context fingerprint is generated for each piece of game text. A specific algorithm is used to comprehensively process the feature vectors in the game context vector space, compressing multi-dimensional feature information into a unique identifier, the context fingerprint. This algorithm considers all relevant contextual features, including visual, spatial, and auditory features, as well as text classification labeling information, ensuring that different game texts, even with partially similar contextual features, will generate different context fingerprints. For example, for two dialogue texts in different scenes but involving similar characters, due to differences in the visual features of the scenes and the background sound effects of the audio, the algorithm will generate different context fingerprints. In this way, each piece of game text has its unique context fingerprint, accurately and uniquely identifying its game context.

[0044] In this embodiment, a process is implemented that classifies and tags game text based on the game context using a terminology version tree, and constructs a game context vector space by combining game screenshots, 3D scene models, and audio dialogues to generate a unique context fingerprint for each piece of game text. This provides strong support for subsequent intelligent task allocation and context-assisted translation.

[0045] S103 constructs a capability map based on translators' historical translation data, and intelligently allocates tasks by combining text categories and contextual features identified by contextual fingerprints, recommending the optimal translator combination for specific game projects.

[0046] In this embodiment, historical translation data of translators is comprehensively collected from the company's internal translator management system and past game translation project records. This data covers multiple dimensions, including the type of game projects the translator participated in (e.g., role-playing, strategy, casual games), the language of translation (e.g., English, Japanese, Korean), the type of text translated (e.g., UI text, story scripts, item descriptions), translation quality assessment results (scored by professional reviewers based on indicators such as accuracy, fluency, and stylistic consistency), translation efficiency data (e.g., the number of words translated per unit of time), and the translator's role in the project (e.g., lead translator, proofreader). For example, for translator A, data shows that they participated in the translation of three strategy mobile games, involving both English and Japanese, translating story scripts and item descriptions. In one strategy mobile game translation, they received a quality assessment score of 85 points, averaging approximately 500 words per hour, and frequently served as the lead translator.

[0047] The collected historical translation data of translators is organized and analyzed to construct a translator competency map. Using translators as core nodes, information such as the types of game projects they have participated in, the languages ​​they have translated, the types of texts they have translated, quality assessment results, translation efficiency data, and the roles they have played are used as branch attributes to form a multi-dimensional competency map. In this competency map, each translator has a unique combination of attributes, which can intuitively demonstrate the translator's abilities and experience in different areas. For example, translator A's competency map shows that they have extensive experience in translating strategy mobile games, strong English and Japanese translation skills, are proficient in translating plot scripts and item descriptions, have high quality assessments, moderate translation efficiency, and have served as a lead translator. In this way, the translators' abilities and experience are quantified, providing a basis for subsequent intelligent task allocation.

[0048] In this strategy mobile game translation project, the method described above for constructing a game context vector space based on a terminology version tree has been used to generate a unique context fingerprint for each piece of game text. Text categories and contextual features are extracted from these fingerprints. Text categories include UI text, dialogue, quest descriptions, and skill descriptions; contextual features cover game scenes (such as castles, forests, deserts), involved characters (such as generals, soldiers, mages), and text functions (such as guidance, hints, and plot progression). For example, for a dialogue about a general issuing a combat quest in a castle, its text category is dialogue, its contextual features are a castle game scene, a general involved character, and plot progression text function.

[0049] Based on the specific requirements of this strategy mobile game translation project, the required translators' abilities for different text types and contextual features were clearly defined. For example, for dialogue texts, translators are required to have strong literary expression skills and the ability to grasp the characters' emotions; for skill description texts, translators are required to have accurate technical terminology translation skills and concise and clear expression skills. The text tasks in the project were subdivided according to text type and contextual features, and then matched and analyzed with information in the translator competency map. By comparing the fit between the translators' past translated text types, languages, quality assessment results, etc., and the current task requirements, a matching score was calculated for each translator and different sub-tasks. For example, if translator A has previously translated dialogue texts for strategy mobile games with high quality assessments, then their matching score for similar dialogue tasks in this project will be high.

[0050] Based on matching scores, an intelligent algorithm is used to allocate translators and tasks. The algorithm comprehensively considers multiple factors, such as translator availability, current workload, and collaborative experience among translators, to ensure the rationality and efficiency of task allocation. For each sub-task, the translator with the highest matching score is selected. Simultaneously, considering the overall project and coordination, translator combinations are optimized. For example, if one translator is highly skilled in translating dialogue but has low efficiency, while another translator is highly efficient but has less experience in translating dialogue, they can be combined. The more skilled translator can handle the core dialogue, while the more efficient translator assists with auxiliary tasks, forming an optimal translator combination. In this way, the most suitable translator combination is recommended for this strategy mobile game translation project, ensuring the project is completed with high quality and efficiency.

[0051] In this embodiment, a capability map is constructed based on translators' historical translation data. Intelligent task allocation is performed by combining text categories and contextual features identified by contextual fingerprints. The process of recommending the optimal translator combination for a specific game project is realized, which effectively improves the management efficiency and translation quality of game translation projects.

[0052] S104, in the process of performing context-assisted translation based on the optimal translator combination, utilizes the visual reference provided by the game context vector space, combined with a pre-built translation risk prediction model, to mark high-risk texts and automatically generate alternative translations.

[0053] In this embodiment, a visualization interface for the game's context vector space is loaded. This interface intuitively displays various contextual elements within the game, such as 3D models of different game scenes, character appearances and animations, and screenshots of key plot points. For example, when a translator needs to translate a dialogue text set in a mysterious forest, the visualization interface displays a 3D scene of the mysterious forest, including the types of trees, the brightness of the light, and possible monsters, making the translator feel as if they are in the game scene. Simultaneously, the interface also labels the character information involved in the currently translated text, such as the character's physical characteristics, personality traits, and role in the game, helping the translator better understand the context of the text.

[0054] An integrated, pre-built translation risk prediction model is used. This model is built upon extensive data and experience from past game translation projects and can identify high-risk factors that may lead to translation errors or inconsistencies with the game's context. These high-risk factors include: inappropriate use of technical terms, semantic misunderstandings due to cultural differences, text length exceeding the game's display limits, and expressions inconsistent with the overall game style. For example, the model analyzes whether the translated text contains game-specific technical terms; if the translator uses inaccurate terminology, the model identifies it as high-risk.

[0055] When a translator begins translating, the translation risk prediction model analyzes the input text in real time. If the model detects high-risk factors in the translation, it will clearly alert the translator through prominent markers on the translation tool's interface. For example, for translations that may contain inappropriate terminology, the model will highlight the sentence containing the term in red and display a pop-up message explaining the potential risks and suggested corrections. For translations that may exceed the text length limit, the model will display warning signs at the edges of the text boxes, reminding the translator to adjust the translation length.

[0056] While identifying high-risk text, the translation risk prediction model automatically generates alternative translations based on information provided by the game context vector space and the model's own knowledge base. These alternative translations are optimized solutions proposed by the model after comprehensively considering factors such as game context, terminology accuracy, and cultural adaptability. For example, when a translator translates a description of a magic skill in a game, if the translator's translation is inaccurate or inconsistent with the overall style of the game, the model will generate multiple alternative translations. Each alternative translation will be accompanied by a detailed explanation of its advantages in terms of context, terminology, and cultural aspects. The translator can then choose the most suitable translation based on these explanations and their own judgment.

[0057] After seeing high-risk text markers and automatically generated alternative translations, translators carefully evaluate this information. They use visual references provided by the game's context vector space to determine if the alternative translations truly fit the game's actual situation. Simultaneously, translators consider their own translation style and project requirements to select the most suitable alternative. If translators deem the automatically generated alternative translations unsatisfactory, they can manually modify the translation based on their experience and understanding of the game. For example, when evaluating alternative translations, if a translator finds that a certain translation, while technically accurate, is too stiff and doesn't fit the character's personality, they will choose a more suitable alternative or further optimize the original translation.

[0058] After translators complete the processing of high-risk texts, the system records their choices and processing steps. These records are fed into the knowledge base of the translation risk prediction model to continuously optimize its accuracy and usability. Simultaneously, project managers can use these records to understand the problems and difficulties encountered by translators during the translation process, providing timely support and guidance. For example, if multiple translators choose similar alternative translations when handling a certain type of high-risk text, it indicates a certain regularity in the translation of that type of text. Project managers can summarize these experiences and share them with other translators to improve the overall translation efficiency and quality of the entire team.

[0059] In this embodiment, during the context-assisted translation process based on the optimal translator combination, the visualization references provided by the game context vector space can be fully utilized. Combined with a pre-built translation risk prediction model, high-risk texts can be effectively marked and alternative translations can be automatically generated, thereby improving the accuracy and quality of game translation and ensuring the smooth completion of the game localization project.

[0060] S105 performs quality checks on the translated text through an integrated GQA model in the real-time collaborative monitoring interface, and feeds back the GQA model's check results to the terminology version tree.

[0061] In this embodiment, a dedicated real-time collaborative monitoring interface is developed and deployed before the game translation project begins. This interface supports simultaneous online operation by multiple users, allowing translator teams and project managers from different language backgrounds to work on the same platform. For example, the interface is divided into modules such as a translator workspace, a project progress display area, and a message notification area. The translator workspace allows translators to input and modify translated text; the project progress display area shows the completion status of translation tasks in each language in chart form; and the message notification area promptly pushes project-related notifications and reminders. Furthermore, it ensures that this interface can stably interact with the GQA (General Quality Assessment) model to achieve real-time quality checking functionality.

[0062] The pre-built GQA model is integrated into the real-time collaborative monitoring interface. The GQA model is a quality inspection model specifically trained for the characteristics of game text. It comprehensively considers multiple dimensions of quality inspection rules, including the accuracy of game terminology, the fit of the text with the game context, character length limits, and grammatical correctness. During integration, it is crucial to ensure that the model can accurately receive the translated text input by the translator and clearly display the inspection results on the monitoring interface. For example, a quality inspection result display box is set up next to the translator's workspace. When the translator completes and submits a piece of text, the GQA model immediately analyzes the text and displays the inspection results in the display box using different colored indicators. Green indicates acceptable text quality, yellow indicates minor issues that can be optimized, and red indicates serious quality problems.

[0063] After translators complete a piece of game text, they submit it to the real-time collaborative monitoring interface. Upon submission, the system automatically records the submission time, translator information, and the content of the translated text. For example, after translating a text about a game character's skills, the translator clicks the submit button, and the text enters the GQA model's quality check process. Simultaneously, the system associates this text with corresponding game context information (such as character information, skill usage scenarios, etc.) so that the GQA model can more accurately assess its quality.

[0064] After receiving the translated text submitted by the translator, the GQA model begins a comprehensive quality check. First, the model checks the accuracy of terminology usage by comparing it to a pre-built terminology version tree to ensure consistency with standard terminology in the game. For example, for the specific item name "Magic Crystal," the model checks whether the translator used the correct translation, avoiding inconsistencies between different translators. Second, the model analyzes the text's fit with the game context, using a game context vector space constructed from game screenshots, 3D scene models, and audio dialogue to determine if the translation accurately reflects the actual situation in the game. For instance, if a dialogue takes place in a dark underground castle scene, the model checks whether the translation uses vocabulary and expressions appropriate for that atmosphere. Furthermore, the model also checks whether the text's character length meets the display limits of the game interface and whether the grammar is correct.

[0065] After completing the quality check, the GQA model generates a detailed quality check report. The report includes the types and locations of quality issues found in the text, along with suggested modifications. For example, if the model finds a term translated incorrectly, the report will indicate the term's location in the original text, the incorrect translation, and provide a correct translation suggestion. For issues exceeding character length limits, the report will specify the number of characters exceeding the limit and recommend simplification. Furthermore, the report will be categorized according to the severity of the quality issues, allowing project managers and translators to quickly understand the urgency of the problems.

[0066] The generated quality inspection results report is clearly displayed on the real-time collaborative monitoring interface. In addition to simply displaying the inspection results in a color-coded display box next to the translator's workspace, a detailed results viewing button can be added. When a translator or project manager clicks this button, a detailed report window pops up, displaying the complete quality inspection results. For example, the detailed report window can show all discovered quality issues in a list format, with each issue accompanied by a specific description and suggested modifications. This allows translators to easily view problems in their translated text and make targeted revisions.

[0067] The system automatically feeds back the quality check results of the GQA model to the terminology version tree. For any errors or inconsistencies in terminology usage found during the check, the corresponding terms in the terminology version tree are updated and corrected. For example, if a term is found to differ in translation across different languages, the system determines the most accurate translation based on the quality check results and updates that translation to the corresponding language branch in the terminology version tree. Simultaneously, for text content that needs modification due to quality issues, the system records the before-and-after comparison information and associates it with the terminology version tree for subsequent tracking and analysis of terminology usage. In this way, by continuously feeding back quality check results to the terminology version tree, the terminology database can be continuously improved, enhancing the accuracy and consistency of subsequent translation work.

[0068] Project managers can comprehensively view the quality check status of translation tasks in each language and the translators' work progress on the real-time collaborative monitoring interface. Based on the quality check results of the GQA model, managers can promptly identify common quality issues in the project and organize training and guidance for translators. For example, if multiple translators are found to have inaccurate terminology usage when translating a certain type of game text, managers can arrange specialized terminology training courses to help translators improve their terminology translation skills. Simultaneously, managers can also coordinate the work progress of translation teams in different languages ​​based on the quality check results to ensure that the game can be released globally simultaneously as planned.

[0069] In this embodiment, the translated text is quality checked through an integrated GQA model in the real-time collaborative monitoring interface, and the inspection results of the GQA model are fed back to the terminology version tree, which can effectively improve the quality and efficiency of game translation and ensure the smooth progress of game localization projects.

[0070] In some embodiments, step S101 above, which involves extracting and preprocessing game resources using a dedicated game engine parsing plugin, constructing a terminology version tree based on the preprocessed game resources, and binding the terminology version tree to the game version number, specifically includes: Game resources are extracted and preprocessed using a dedicated parsing plugin for the game engine, and a mapping between text resources and game context resources is established. Based on the association mapping, construct a terminology version tree; Bind the terminology version tree to the game version number to establish a terminology change tracking mechanism between versions; When a game version update is detected, the terminology change tracking mechanism automatically identifies terminology differences and triggers an update of the translation memory of the associated text in the terminology version tree based on these differences.

[0071] In this embodiment, a dedicated parsing plugin that matches the game engine is selected. This plugin needs to have the ability to deeply parse the game engine file format and accurately identify various resources in the game, such as text resources (including UI text, plot scripts, item descriptions, achievement systems, etc.), image resources, audio resources, and 3D model resources.

[0072] After the parsing plugin starts working, it performs a full scan of the game project folder, locating and reading all relevant resource files. For text resources, the plugin extracts the text content and performs preliminary formatting, removing unnecessary spaces, line breaks, and other irrelevant characters, converting it into a uniform text format for easier subsequent processing. Simultaneously, for game context resources, such as game scene screenshots and character images in image resources, character dialogue and background music in audio resources, and character and scene models in 3D model resources, the plugin extracts their key feature information, such as image resolution and color mode, audio duration and sampling rate, and 3D model vertex count and material information.

[0073] After resource extraction, a mapping is established between text resources and game context resources. Taking story script text as an example, when a passage describing a character's dialogue in a specific scene is extracted, the plugin simultaneously obtains a screenshot of that scene and the 3D model information of the relevant character, and associates this information with the text content. Through this mapping, the game context of each piece of text can be clearly understood, providing rich background information for subsequent terminology extraction and translation. For example, when translating a passage about the release of a magic skill, combining a screenshot of the scene at the time of the skill release with the character's motion model allows for a more accurate understanding of the skill's effects and functions, thus enabling the selection of more appropriate terminology for translation.

[0074] A deep analysis was performed on all extracted text resources to identify terms with industry-specific and game-specific meanings. These terms may include character names, skill names, item names, location names, etc., from the game. For example, the unique magic skill "Starfall" and the character-specific weapon "Dragon Breath Longsword" are examples. The identified terms were then categorized according to certain classification rules, such as by the game module to which the terms belong (character system, skill system, item system, etc.) or by the semantic type of the terms (noun, verb, adjective, etc.). After classification, these terms were organized into a tree structure, forming the initial framework of the terminology database version tree. The root node of the tree represents the entire game project, first-level child nodes represent different game modules, second-level child nodes represent specific terminology categories under that module, and third-level child nodes and below are specific terminology entries.

[0075] In constructing the terminology version tree, the previously established association mapping information is fully utilized. For each term entry, its game context is recorded, including relevant scene screenshots, character models, audio dialogues, and other information. This allows translators to intuitively understand the usage scenarios of terms when querying and using them later, avoiding misuse. For example, when querying the skill term "Starfall," one can simultaneously see a scene screenshot and a character animation model when the skill is activated, helping translators better understand the skill's effects and functions, thus choosing a more accurate translation.

[0076] After constructing the terminology version tree, it is linked to the game version number. Whenever the game development team releases a new game version, a unique version number is assigned to it, such as V1.0, V1.1, etc. Simultaneously, the current version of the terminology version tree is associated with this version number and recorded in the system's database.

[0077] To establish a terminology change tracking mechanism between versions, a detailed comparative analysis of the terminology version tree is performed with each version update. Specifically, the terminology version tree of the new version is compared item by item with that of the previous version, recording the addition, deletion, and modification of terms. For example, if a new character, "Phantom Assassin," is added in a new version, a corresponding term entry is added to the terminology version tree, recording the game module (character system), semantic type (noun), and the game context in which the term appears. If the translation of a term is modified, such as changing "Dragon Breath Sword" from "Dragon Breath Sword" to "Dragon Fire Sword," the translation before and after the modification, as well as the reason for the modification, are recorded.

[0078] This inter-version terminology change tracking mechanism provides a clear understanding of how terminology evolves across different game versions, offering historical reference for subsequent translation work. For example, when translating a new version, translators can review the changes in terminology across different versions to understand how terminology usage has evolved and avoid inconsistencies in translation.

[0079] During game operation, a version detection module is set up to monitor changes in the game version in real time. When a new version is detected by the game development team, the version detection module will immediately notify the terminology management system.

[0080] Upon receiving the notification, the terminology management system automatically identifies the terminology differences between the new and previous versions based on the established terminology change tracking mechanism. Specifically, the system compares the terminology version trees of the two versions to identify newly added, deleted, and modified terminology entries. For newly added terminology entries, the system adds them to the translation memory and records their initial translation. For deleted terminology entries, the system removes them from the translation memory but retains their historical records for later querying and analysis. For modified terminology entries, the system updates the translation content in the translation memory and records the reason for and time of the modification.

[0081] After a translation memory update is triggered, the system will push the updated translation memory information to relevant translators and project managers. Translators can refer to historical translations in the translation memory when working on a new version, improving efficiency and accuracy. For example, if the new version contains text similar to the previous version, translators can directly use the translation results from the translation memory or make appropriate adjustments based on the revised terminology, avoiding duplication of effort and inconsistencies. Meanwhile, project managers can monitor the translation memory update status to understand the project's progress and quality, enabling timely coordination and management.

[0082] In this embodiment, a process is implemented to extract and preprocess game resources through a dedicated parsing plugin for the game engine, construct a terminology version tree, and bind it to the game version number. This effectively improves the quality and efficiency of game localization translation and ensures the consistency of terminology and the accuracy of translation across different game versions.

[0083] In some embodiments, step S102 above, which involves classifying and labeling game text based on the terminology version tree according to the game context, and constructing a game context vector space by combining game screenshots, 3D scene models, and audio dialogues, specifically includes: Based on the terminology context information stored in the terminology version tree, the game text is classified and marked in multiple dimensions according to the terminology category attributes and text function positions to determine the game function attributes of the game text. Based on the game's functional attributes, and combining game screenshots, 3D scene models, and audio dialogues, a layered coding architecture is used to construct the game context vector space. The classified game text is mapped into the game context vector space. The high-dimensional context vector is converted into a fixed-length fingerprint code using a hash algorithm, generating a unique context fingerprint for each piece of game text.

[0084] In this embodiment, stored terminology context information is extracted from the terminology version tree. This information covers the specific usage scenarios and associated objects of the term in the game. For example, for the term "resource gathering" in the game, its context information may include the location of gathering (such as a mine or forest), the tools required for gathering (such as a pickaxe or axe), and the type of resource obtained after gathering (such as ore or wood).

[0085] Based on the attributes of terminology categories, terms in game texts are divided into different categories. For example, according to the type of game element, they can be divided into character-related terms (such as character names, character skill names), item-related terms (such as weapon names, prop names), and scene-related terms (such as map names, location names), etc.; according to the importance of function, they can be divided into core gameplay-related terms (such as combat rule terms, resource management terms) and auxiliary function-related terms (such as interface prompt terms, help document terms).

[0086] At the same time, the game text is further marked according to its functional location. The functional location of the text refers to the display position and function of the text in the game interface, such as the main interface text (e.g., game title, menu options), the battle interface text (e.g., battle prompts, damage value display), and the quest interface text (e.g., quest description, quest reward description), etc.

[0087] By combining multi-dimensional classification and labeling of terminology category attributes and text functional location, the game function attributes of each piece of game text can be accurately determined. For example, a piece of text displayed on the main interface that introduces the core gameplay can be identified as "Main Interface - Core Gameplay Introduction"; a piece of text that appears on the battle interface and prompts players about the cooldown time of their character's skills can be identified as "Battle Interface - Skill Hints".

[0088] After determining the game functional attributes of the game text, the game context vector space is constructed. A hierarchical encoding architecture is adopted, consisting of three layers: a basic information layer, a related information layer, and a comprehensive context layer.

[0089] The basic information layer mainly contains basic information directly related to the game text, such as text content, text length, and text language. For each piece of game text, its text content is converted into a unified character encoding format, the text length (in characters) is recorded, and the language of the text is indicated (such as Chinese, English, etc.).

[0090] The related information layer combines game screenshots, 3D scene models, and audio dialogues to extract relevant information related to the game text. For game screenshots, image recognition technology is used to extract key elements such as character appearance, item appearance, and scene layout, and these are converted into feature vectors. For example, for a screenshot showing a character in combat, information such as the character's equipment features and posture is extracted to generate a corresponding feature vector. For 3D scene models, geometric features (such as the number of vertices and faces), material information (such as texture and color), and spatial location information are extracted and converted into feature vectors. For audio dialogues, speech recognition technology is used to convert audio into text, and information such as keywords and sentiment in the dialogue is extracted to generate feature vectors.

[0091] The integrated context layer combines the feature vectors from the basic information layer and the related information layer. Based on the game's functional attributes, the feature vectors at different levels are weighted. For example, for core gameplay text, the weight of feature vectors related to game rules and operation prompts is increased; for auxiliary function text, the weight of feature vectors related to interface display and help information is increased. Through this weighted integration, all feature vectors are combined into a high-dimensional game context vector, thus constructing a game context vector space. In this space, each dimension represents a specific aspect of the game context. The context vectors of different game texts have different positions and directions in this space, intuitively reflecting the contextual differences between them.

[0092] After constructing the game context vector space, the classified game texts are mapped one by one into this space. For each piece of game text, based on its game function attributes, the basic information feature vector of the text is extracted from the basic information layer, and the feature vectors of game screenshots, 3D scene models and audio dialogues related to the text are extracted from the associated information layer. Then, the feature vectors are weighted and integrated in the comprehensive context layer to obtain the game context vector corresponding to the text.

[0093] To facilitate the storage and comparison of contextual information from different game texts, a hash algorithm is used to convert high-dimensional game context vectors into fixed-length fingerprint codes. Hash algorithms can map input data of arbitrary length to fixed-length output values, exhibiting uniqueness and stability. In this embodiment, a hash algorithm suitable for processing high-dimensional vectors is selected, such as the MD5 hash algorithm or the SHA-1 hash algorithm.

[0094] The game context vector is input into a hash algorithm, which generates a fixed-length fingerprint code. This fingerprint code acts like an "identity card" for the game text, uniquely identifying the contextual information of each piece of game text. For example, even if two pieces of game text have similar content, their generated game context vectors will differ because they exist in different game contexts (such as different game scenes or different character dialogue situations), and consequently, the fingerprint codes obtained through the hash algorithm will also be different. In this way, a unique contextual fingerprint is generated for each piece of game text, facilitating subsequent text retrieval, matching, and translation quality assessment during the game localization process.

[0095] In this embodiment, the process of classifying and labeling game text based on a terminology version tree, constructing a game context vector space, and generating context fingerprints for game text is realized. This can effectively improve the accuracy and efficiency of game localization and ensure the consistency of game context and user experience in different language environments.

[0096] In some embodiments, step S103 above, which involves constructing a capability map based on translators' historical translation data, combining the text category and contextual features identified by contextual fingerprints to perform intelligent task allocation, and recommending the optimal translator combination for a specific game project, specifically includes: Based on multi-dimensional analysis of translators' historical translation data, a translator competency map reflecting translators' professional abilities is constructed. Based on the text category and contextual features identified by the contextual fingerprint, a requirement vector for the translation task is generated; The matching calculation between the demand vector and the translator capability vector in the translator capability map determines the fit between the translator and the translation task. Based on the fit results, candidate translators are selected for specific game projects. The translators are then assigned according to their collaboration history, time zone distribution, and workload to generate an optimal translator combination recommendation scheme.

[0097] In this embodiment, historical translation data is collected from all translators who have participated in similar game projects or have experience in game translation. This data covers multiple dimensions, including the type of translated text (such as plot text, interface text, skill description text, etc.), the language pairs (such as Chinese-English, Chinese-French, Chinese-Japanese, etc.), translation quality assessment results (scored by professional reviewers based on indicators such as accuracy, fluency, and stylistic consistency), translation efficiency (the number of translated words or the amount of work completed per unit of time), and customer feedback (such as evaluations of the translator's professionalism and communication skills).

[0098] In-depth analysis was conducted on the collected historical translation data of translators. For the text type dimension, the number of translation experiences and average quality scores of translators on different text types were statistically analyzed. For example, a translator had 20 experiences translating narrative texts with an average quality score of 85; and 15 experiences translating interface texts with an average quality score of 80. For the language pair dimension, the translator's proficiency in translating different language pairs was analyzed, assessed by calculating the number of translation tasks completed and the stability of quality for each language pair. For example, a translator completed 50 tasks in English-Chinese translation with relatively small quality fluctuations, but only completed 10 tasks in French-Chinese translation with some quality fluctuations.

[0099] Based on multi-dimensional analysis, a translator competency map reflecting each translator's professional abilities is constructed. This map is presented in the form of nodes and edges. Nodes represent translators' competency indicators across different dimensions, such as the ability to translate narrative text and English-Chinese translation, while edges represent the relationships between these competency indicators. By quantifying the competency value of each node (e.g., calculating a weighted average based on quality scores and experience counts), a translator competency vector is formed, comprehensively and intuitively showcasing the translator's professional competency characteristics. For example, translator A's competency map shows strong abilities in translating character skill description text and English-Chinese language pairs, with competency values ​​of 90 and 88 respectively, while relatively weaker abilities in translating interface text and French-Chinese language pairs, with competency values ​​of 75 and 70 respectively.

[0100] In the text processing phase of game projects, contextual fingerprints are utilized. Contextual fingerprints can accurately identify the category (such as character dialogue text, task description text, item description text, etc.) and contextual features (such as the game scene in which the text is set, the type of character involved, emotional tendencies, etc.) of each piece of text.

[0101] Based on the information identified by the contextual fingerprint, the specific requirements of the translation task are determined. Different text categories have different requirements for the translator's translation skills and style. For example, character dialogue texts require the translator to accurately grasp the character's personality traits and colloquial expressions, while task description texts emphasize accuracy and logic. Regarding contextual features, such as a game setting in a fantasy magical world involving numerous fictional magical terms and mythological elements, this requires the translator to possess relevant background knowledge and imagination.

[0102] The requirements for a translation task are transformed into a requirement vector. This vector contains multiple dimensions, each corresponding to a requirement indicator, such as text category fit, contextual knowledge requirements, and language style requirements. By assigning weights to each requirement indicator (based on the project's importance and urgency) and mapping the information reflected by the contextual fingerprint to the corresponding indicator, the numerical value of each indicator is calculated, thus generating the translation task's requirement vector. For example, for a task translating a dialogue from a fantasy world, the text category fit indicator has a weight of 0.3, the contextual knowledge requirement indicator has a weight of 0.4, and the language style requirement indicator has a weight of 0.3. Contextual fingerprint analysis shows that the task requires a text category fit value of 0.8 (indicating a high suitability for translating dialogue with strong translator skills), a contextual knowledge requirement value of 0.9 (indicating the need for translators with extensive fantasy / magical knowledge), and a language style requirement value of 0.85 (indicating the need for translators capable of vivid and conversational translation).

[0103] The generated translation task requirement vector is matched with the translator ability vector in each translator's ability graph. The matching calculation uses a similarity algorithm, calculating the numerical differences between the requirement vector and the translator ability vector across various dimensions to obtain a similarity score. A higher similarity score indicates a higher suitability between the translator and the translation task. For example, for the aforementioned fantasy world character dialogue text translation task, the requirement vector (0.8, 0.9, 0.85) and translator A's ability vector (0.7, 0.8, 0.8) have numerical differences across various dimensions of 0.1, 0.1, and 0.05, respectively. Using a similarity algorithm (such as cosine similarity), the calculated similarity score is 0.92, indicating a high suitability for translator A. After matching all translators, the suitability results for each translator are presented in a list, including translator name and suitability score. By sorting the suitability scores, it becomes clear which translators are more suitable for completing the translation task.

[0104] Based on the translator-translation task fit results, translators with higher fit scores are selected as candidate translators. A fit threshold, such as 0.8, is typically set, and translators with fit scores greater than or equal to this threshold are included in the candidate translator list.

[0105] Consider the collaboration history between candidate translators. By reviewing historical data in the project management system, understand whether candidate translators have previously collaborated and the results of those collaborations. If two candidate translators have worked well together in previous projects, communicating and collaborating efficiently, then pairing them together may improve the translation efficiency and quality of this project. For example, if translator B and translator C have collaborated on a previous game localization project, completing several high-quality translation tasks with positive client feedback, then pairing them together again for this project could be considered.

[0106] Consider the time zone distribution of candidate translators. Since the game project may target players in different regions around the world, translation tasks need to be completed within a reasonable timeframe. Therefore, selecting a translator team with a reasonable time zone distribution ensures that translators are available to handle translation tasks promptly at different times. For example, if the project requires simultaneous translation of Chinese, English, and Japanese, one translator located in the China time zone can handle Chinese translation, one in the US time zone can handle English translation, and one in the Japan time zone can handle Japanese translation, thus enabling 24 / 7 uninterrupted translation work.

[0107] Consider the workload of potential translators. Check their other project assignments to ensure they have sufficient time and energy to complete the translation tasks for this game project. Avoid including translators with excessive workloads in the team, as this could affect translation quality and schedule. For example, if translator D is currently involved in three other projects with a heavy workload, they should not be the primary translator for this project.

[0108] Taking all the above factors into account, candidate translators are rationally allocated to generate an optimal translator combination recommendation scheme. This scheme details the responsibilities and task allocation of each translator in the project, as well as the strengths and expected results of the entire translator combination. For example, the recommendation scheme indicates that translator B is responsible for translating the character dialogue text, and translator C is responsible for translating the task description text. They have collaborated before and are in similar time zones, enabling them to work efficiently together and are expected to complete the high-quality translation task within the stipulated time.

[0109] In this embodiment, a capability graph is constructed based on the translator's historical translation data, and intelligent task allocation is performed by combining contextual fingerprints. This process recommends the optimal translator combination for a specific game project, which can effectively improve the translation quality and efficiency of game localization projects and ensure the smooth progress of the project.

[0110] In some embodiments, step S104 above, which involves using the visual references provided by the game context vector space, combined with a pre-built translation risk prediction model, to mark high-risk texts and automatically generate alternative translations, specifically includes: By utilizing the multimodal visualization references provided by the game context vector space, translators can be given an immersive translation context. Based on a pre-built translation risk prediction model, a multi-dimensional risk assessment is performed on the current translated text to identify and mark high-risk text segments. For the identified high-risk texts, a strategy is combined using a rule base, translation memory, and neural networks to generate multiple alternative translation schemes. The alternative translation schemes are then prioritized using a quality assessment algorithm.

[0111] In this embodiment, when a translator begins translation work, the system retrieves relevant multimodal visualization references from the game context vector space based on the game content involved in the current translation text. For example, if the current translation text is a description of a mysterious forest scene in the game, the system will retrieve a high-resolution screenshot of the forest scene from the vector database, displaying details such as trees, flowers, and terrain in the forest; at the same time, it will play the background music of the scene to create a mysterious and tranquil atmosphere; it will also extract audio clips of character dialogue related to the scene, allowing the translator to understand the character's language style and emotional expression in that scene.

[0112] The system presents these multimodal visual references to translators in an intuitive way, such as displaying game screenshots on one side of the translation interface and audio playback controls on the other, along with relevant scene descriptions. This provides translators with an immersive translation context, enabling them to more accurately understand the game content and grasp the context and style of the translation. For example, after seeing a screenshot of the mysterious forest and hearing the corresponding background music, translators can better appreciate the "deep and tranquil forest" described in the text, thus choosing more appropriate vocabulary and expressions during translation.

[0113] Before the project began, a translation risk prediction model was built based on a large amount of historical translation data and expertise in the gaming field. This model considers risk factors in multiple dimensions, including language difficulty (such as vocabulary complexity and grammatical complexity), cultural differences (such as cultural taboos and customs), gaming expertise (such as gaming terminology and gameplay rules), and contextual relevance.

[0114] Once the translator inputs the text to be translated, the system feeds it into a pre-built translation risk prediction model. The model analyzes the text word by word and sentence by sentence, assessing it based on risk factors across various dimensions. For example, for a text containing complex magical terminology, the model identifies the specialized vocabulary and assesses the difficulty and risk of translation based on the term's common translations and frequency of use in the game. If a term has been prone to errors or ambiguity in previous translations, the model marks it as a high-risk term.

[0115] For the entire text segment, the model comprehensively considers risk factors across various dimensions and calculates a risk score. A risk threshold is set; when a text segment's risk score exceeds this threshold, it is identified and marked as a high-risk text segment. For example, a text describing a new gameplay mechanic, which involves multiple game mechanics and terms not seen in previous projects and has a relatively complex language structure, might have a risk score of 0.8 calculated by the model (with a threshold set at 0.6), thus marking the text segment as high-risk. The system prominently marks high-risk text segments on the translation interface (e.g., highlighted in red) to alert translators to pay close attention.

[0116] Once the system identifies high-risk text segments, it initiates a mechanism to generate alternative translations. This mechanism combines three techniques: rule base, translation memory, and neural networks.

[0117] The rule base stores a large number of game translation rules, including word substitution rules, grammar adjustment rules, and cultural adaptation rules. For example, for words with specific cultural connotations, the rule base specifies corresponding replacement words or interpretations. Translation memory records the translation results of similar texts from previous projects. When encountering high-risk text, the system searches for similar text fragments in the translation memory and extracts their corresponding translations as a reference. The neural network, through deep learning algorithms, learns from a large amount of game translation data and can generate natural and accurate translations based on the context.

[0118] The system combines these three techniques to generate multiple alternative translation schemes for the marked high-risk text. For example, for a high-risk description of a game skill, the rule base may provide a translation based on word substitution, translation memory may provide a translation of a similar skill from the past, and the neural network may generate an innovative translation that is more in line with the current context.

[0119] After generating candidate translation schemes, the system prioritizes each scheme using a quality assessment algorithm. The algorithm considers multiple evaluation metrics, including accuracy (the degree of faithfulness to the original meaning), fluency (the naturalness of the language), stylistic consistency (the degree of matching with the overall style of the game), and cultural adaptability (whether it conforms to the habits of the target language culture). The system assigns weights to each evaluation metric and calculates a comprehensive quality score based on the performance of each candidate translation scheme across all metrics.

[0120] Based on the overall quality score, the system ranks the candidate translation options from highest to lowest priority and displays them to the translator on the translation interface. Translators can choose the most suitable translation based on the ranking and their own professional judgment. For example, the top-ranked candidate translation option performs well in terms of accuracy, fluency, and stylistic consistency, and the translator can prioritize using this option.

[0121] In this embodiment, a process is implemented that utilizes the visual references provided by the game context vector space, combined with a pre-built translation risk prediction model, to mark high-risk texts and automatically generate alternative translations. This can effectively improve the quality and efficiency of game translation and reduce translation risks.

[0122] Furthermore, the pre-construction steps of the translation risk prediction model include: Establish a multi-level risk feature identification system to determine various risk features in game localization. Based on various risk characteristics, training samples were collected and labeled from historical error databases, expert-annotated corpora, and cross-cultural comparison data to construct a basic dataset; Based on the basic dataset and combined with domain knowledge of the game industry, a multi-task learning framework is used to train a hierarchical hybrid neural model architecture to generate a translation risk prediction model. The translation risk prediction model includes a feature extraction layer, an attention mechanism layer, and a multi-task output layer.

[0123] In this embodiment, building a multi-layered risk feature identification system is fundamental to accurately predicting translation risks in game localization projects. This system is constructed from three main levels: language, culture, and game expertise.

[0124] At the linguistic level, three sub-levels are considered: vocabulary, grammar, and semantics. At the vocabulary level, potential risks arise from game terminology, obscure words, and words with multiple meanings. For example, inaccurate translations of game item names and skill names can lead to player misunderstandings; differences in the understanding and expression of obscure words across different languages ​​can also pose risks; and accurate interpretation of words with multiple meanings within a specific game context is crucial. At the grammatical level, differences in grammatical structures across languages ​​are considered, such as subject-verb-object order, tense, and voice. For example, English and Chinese have significantly different grammatical structures, and direct translation may result in grammatical errors. At the semantic level, the overall semantics of sentences and paragraphs are analyzed to identify potential ambiguities, logical contradictions, and other issues. For example, unclear semantic expressions in game story text can negatively impact the player's gaming experience.

[0125] At the cultural level, risks such as cultural taboos, differences in customs, and conflicts of values ​​must be considered. Different cultures have varying degrees of acceptance of certain things, and if the game content involves cultural taboos, it may cause resentment among players in the target language region. For example, certain colors and numbers have special meanings in different cultures and must be handled with care in game translation. Regarding differences in customs, such as festivals and etiquette, the relevant descriptions in the game must conform to the customs of the target language region. Regarding conflicts of values, the values ​​conveyed by the game should align with the values ​​of the target language region to avoid player resistance due to value conflicts.

[0126] At the game expertise level, risk characteristics such as game mechanics, gameplay rules, and character design must be considered. The translation of game mechanics and gameplay rules must be accurate; otherwise, players may not be able to correctly understand the game's mechanics. For example, in a strategy game, an unclear translation of the resource management mechanism can affect players' strategic planning. Regarding character design, the translation of descriptions of a character's personality, background, and abilities must conform to the overall game setting, maintaining consistency in the character's image.

[0127] Through the above multi-level risk feature identification system, we comprehensively sort out various risk features that may occur in game localization, providing accurate feature basis for subsequent model training.

[0128] After identifying the characteristics of various risks, training samples were collected from multiple data sources to build the basic dataset.

[0129] First, samples are collected from a historical error database. This database records translation errors that occurred in previous game localization projects, covering multiple aspects such as language, culture, and game expertise. For example, the database might record incorrect translations of a game term in past translations, along with the problems caused by those errors. Error samples related to identified risk characteristics are then collected from the historical error database as part of the training samples.

[0130] Secondly, experts in game localization were invited to annotate the corpus. These experts possess extensive experience and expertise in game translation, enabling them to accurately identify and annotate risk features within the corpus. The corpus provided to the experts included the original game text, preliminary translations, and relevant game background information. Based on a risk feature identification system, the experts annotated risk points in the corpus, such as potentially ambiguous words or expressions that do not conform to cultural customs. The expert-annotated corpus provides high-quality training samples, ensuring that the model learns accurate risk features.

[0131] Finally, cross-cultural comparative data was used for sample collection. This data includes comparisons of game texts from different linguistic and cultural backgrounds, which can reveal translation risks caused by cultural differences. For example, comparing the texts of the same game in English and Chinese versions allows for analysis of expressions that need adjustment due to cultural differences. Collecting representative samples from this cross-cultural comparative data enriches the diversity of the underlying dataset.

[0132] When labeling the collected training samples, each sample is labeled with a corresponding risk type and risk level according to the risk feature identification system. For example, for a sample with a translation error caused by cultural taboo, its risk type is labeled as "cultural taboo risk," and its risk level is labeled as "high," "medium," or "low" according to the severity of the error. In this way, a basic dataset containing rich samples and accurate labels is constructed, providing reliable data support for model training.

[0133] After constructing the basic dataset, we combined domain knowledge from the gaming industry and used a multi-task learning framework to train the hierarchical hybrid neural model architecture.

[0134] Domain knowledge in the gaming industry includes knowledge of game genres, target audiences, and game styles. For example, different types of games (such as role-playing games, strategy games, and casual games) have different focuses and requirements in translation; factors such as the age and cultural background of the target audience also influence translation strategies; and game styles (such as fantasy, science fiction, and realistic styles) determine the language style and vocabulary selection in translation. Integrating this domain knowledge into model training allows the model to better adapt to the characteristics of game localization.

[0135] A multi-task learning framework is employed to simultaneously handle multiple tasks related to translation risk. For example, tasks such as language accuracy prediction, cultural fit prediction, and game expertise compliance prediction are set up. Through multi-task learning, the model can share information across different tasks, improving its generalization ability and prediction accuracy.

[0136] The hierarchical hybrid neural model architecture includes a feature extraction layer, an attention mechanism layer, and a multi-task output layer. The feature extraction layer is responsible for extracting features from the input text data, transforming the text into numerical features that a computer can process. This layer can employ structures such as Convolutional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs) to extract linguistic and semantic features from the text. For example, CNNs can capture local features in text, while RNNs can process the sequential information of the text.

[0137] The attention mechanism layer is used to weight the extracted features, highlighting important features and suppressing irrelevant features. In game localization, different parts of information have varying degrees of impact on translation risk, and the attention mechanism can automatically learn and assign different weights to each feature. For example, the attention mechanism will give higher weight to the game terminology section because the accuracy of terminology translation is crucial for game understanding.

[0138] The multi-task output layer outputs prediction results for multiple tasks simultaneously based on the processing results of the preceding layers. For example, it outputs language accuracy scores, cultural fit scores, and game expertise compliance scores. Through multi-task output, the model can comprehensively assess the risk profile of the translated text.

[0139] During training, the base dataset is divided into training, validation, and test sets. The model is trained using the training set, and its parameters are continuously adjusted to minimize the loss function value on the training set. The model is evaluated using the validation set, and the optimal model parameters are selected. Finally, the trained model is tested using the test set to evaluate its performance in real-world applications. After multiple iterations of training and optimization, a high-performance translation risk prediction model is generated, capable of accurately predicting translation risks in game localization.

[0140] In some embodiments, step S105 above, which involves performing a quality check on the translated text using an integrated GQA model on the real-time collaborative monitoring interface and feeding back the GQA model's check results to the terminology version tree, specifically includes: The GQA model is integrated into the real-time collaborative monitoring interface to perform real-time quality assessment of translated texts and obtain quality assessment results. Based on the quality assessment results, the anomalies found in the translated text are classified and graded according to severity, and the priority of handling the anomalies is determined. Based on the processing priority, abnormal issues are mapped one by one to the corresponding nodes in the terminology library version tree, establishing a traceability path from quality issues to the root cause of the terminology. Based on the tracing pathway, the impact of terminology changes on translated texts is analyzed, and a gradual update strategy is formulated.

[0141] In this embodiment, the GQA model is integrated into a real-time collaborative monitoring interface. This monitoring interface serves as a collaborative work platform for translators, reviewers, and project managers, providing real-time display of translation progress, text content, and interactive communication capabilities.

[0142] When a translator inputs or modifies translated text in the interface, the GQA model integrated in the background immediately initiates a quality assessment process. The GQA model comprehensively considers multiple dimensions of quality indicators, including language accuracy (such as appropriate vocabulary selection and grammatical correctness), fluency (whether the text flows smoothly and naturally), consistency (whether it aligns with the overall style and terminology of the project), and cultural fit (whether it conforms to the cultural habits and values ​​of the target language region). For example, when translating character dialogue in a game, the GQA model checks whether the vocabulary in the dialogue matches the character's identity and personality. If a rugged warrior character uses overly elegant vocabulary, the model will identify it as a language accuracy issue. Simultaneously, the model analyzes the sentence structure for logical flow and ambiguity to ensure text fluency. After a comprehensive evaluation, the GQA model generates a detailed quality assessment report, including scores for each indicator and an overall quality rating (such as Excellent, Good, Satisfactory, Unsatisfactory), as the quality assessment result.

[0143] After receiving the quality assessment report generated by the GQA model, the reviewers will conduct a detailed analysis of the anomalies identified in the report. Based on the degree to which these anomalies affect translation quality and the gaming experience, they are categorized into different severity levels, typically high, medium, and low. For example, if an anomaly leads to misunderstanding of the game's plot and severely impacts the player's progress, it is classified as a high-level issue; if it only involves inaccurate wording of a few words but has a minor impact on overall understanding, it is classified as a low-level issue. Furthermore, anomalies are categorized for further processing. Common categories include language errors (such as spelling and grammar errors), inconsistencies in terminology (the same term translated differently in different places), and cultural conflicts (content not conforming to the culture of the target language region).

[0144] After severity grading and classification, the priority of handling anomalies is determined based on the project's urgency and importance. High-level issues classified as language errors or cultural conflicts are given the highest priority and require immediate attention; low-level issues classified as terminology inconsistencies can be addressed later. Reviewers will mark each anomaly in the real-time collaborative monitoring interface, noting its severity, classification, and handling priority, so that relevant personnel can clearly understand the issue.

[0145] The terminology version tree is a crucial tool for managing game terminology in this project. It's organized according to the game's modules, functions, and hierarchical structure, with each node representing a specific term or set of terms. Based on the priority of handling exceptions, issues are mapped one by one to the corresponding nodes in the terminology version tree. For example, if inconsistent translations of a game item are found in different places, reviewers will find the corresponding node for that item in the terminology version tree and record the inconsistencies under that node. If a language error is caused by a misunderstanding of terminology, the definition and explanation nodes of the relevant term in the terminology library will be traced back to indicate the cause of the problem.

[0146] In this way, a traceability path is established from quality issues to the root cause of terminology problems. Project team members can use this path to quickly pinpoint the root cause of problems and understand whether the translation anomalies are due to unclear terminology definitions, untimely updates, or other reasons. For example, when a translation error is found in a game skill name, the traceability path can be used to view the initial definition and subsequent changes of that skill name in the terminology database, thereby accurately identifying the problem.

[0147] After establishing the tracing pathway, the project team used it to analyze the scope of the impact of terminology changes on the translated text. For example, when a core term in the terminology database was modified, the tracing pathway allowed them to quickly find all translated text paragraphs that used that term.

[0148] After analyzing the scope of impact, a gradual update strategy was developed. For high-level and widely impactful terminology changes, an immediate update strategy was adopted, notifying all relevant translators and reviewers to simultaneously revise the translated texts involving the term, ensuring consistency and accuracy of the game content. For example, if an important character's name in the game was changed, all relevant texts needed to be updated immediately to avoid player confusion.

[0149] For minor terminology changes with limited impact, a phased update strategy will be adopted. Based on project progress and resource availability, the translated texts involving the term will be updated gradually at appropriate times. For example, minor changes to game item names can be implemented gradually in subsequent version updates.

[0150] Meanwhile, during the update process, the time and content of each update are recorded, and the relevant node information of the terminology database version tree is updated to ensure the accuracy of the terminology database's version management and traceability functions. This gradual update strategy ensures timely improvement in translation quality while avoiding resource constraints and workflow chaos caused by large-scale simultaneous updates.

[0151] In this embodiment, the quality of translated text is checked through an integrated GQA model in a real-time collaborative monitoring interface, and the check results are fed back to the terminology version tree, which effectively improves the quality and efficiency of game localization translation.

[0152] Reference Figure 2 An embodiment of the present invention provides a game translation collaborative management and monitoring system 2, wherein system 2 specifically includes: The first processing module 201 is used to extract and preprocess game resources through a dedicated parsing plugin for the game engine, construct a terminology version tree based on the preprocessed game resources, and bind and associate the terminology version tree with the game version number. The second processing module 202 is used to classify and label game text based on the terminology version tree and the game context, and to construct a game context vector space by combining game screenshots, 3D scene models and audio dialogues. The game context vector space is used to generate a unique context fingerprint for each piece of game text. The third processing module 203 is used to construct a capability map based on the translator's historical translation data, and to intelligently allocate tasks by combining the text category and context features identified by the context fingerprint, and recommend the optimal translator combination for a specific game project. The fourth processing module 204 is used to mark high-risk texts and automatically generate alternative translations during the context-assisted translation process based on the optimal translator combination, by utilizing the visual references provided by the game context vector space and combining them with a pre-built translation risk prediction model. The fifth processing module 205 is used to perform quality checks on the translated text through the integrated GQA model in the real-time collaborative monitoring interface, and to feed back the inspection results of the GQA model to the terminology version tree.

[0153] It is understandable that, such as Figure 1 The content shown in the game translation collaborative management and monitoring method embodiments is applicable to the game translation collaborative management and monitoring system embodiments. The specific functions implemented in the game translation collaborative management and monitoring system embodiments are as follows: Figure 1 The game translation collaborative management and monitoring method shown in the embodiment is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the illustrated game translation collaborative management and monitoring method embodiment are also the same.

[0154] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0155] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0156] Reference Figure 3 The present invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored on the memory 302. When the computer program 303 is executed on the processor 301, it implements the game translation collaborative management and monitoring method as described in any of the above methods.

[0157] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0158] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0159] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0160] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the game translation collaborative management and monitoring method as described in any of the above methods.

[0161] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0162] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0163] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0164] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0165] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for collaborative management and monitoring of game translation, characterized in that, The method specifically includes: Game resources are extracted and preprocessed using a dedicated parsing plugin for the game engine. A terminology version tree is then constructed based on the preprocessed game resources, and the terminology version tree is bound and associated with the game version number. Based on the terminology version tree, game text is classified and labeled according to game context, and a game context vector space is constructed by combining game screenshots, 3D scene models and audio dialogues. The game context vector space is used to generate a unique context fingerprint for each piece of game text. A capability map is constructed based on translators' historical translation data. Intelligent task allocation is performed by combining text categories and contextual features identified by contextual fingerprints to recommend the optimal translator combination for specific game projects. In the process of performing context-assisted translation based on the optimal translator combination, the visualization reference provided by the game context vector space is used, combined with a pre-built translation risk prediction model, to mark high-risk texts and automatically generate alternative translations; The translated text is quality checked through the integrated GQA model in the real-time collaborative monitoring interface, and the results of the GQA model are fed back to the terminology version tree.

2. The method according to claim 1, characterized in that, The process involves extracting and preprocessing game resources using a dedicated game engine parsing plugin, constructing a terminology version tree based on the preprocessed game resources, and binding the terminology version tree to the game version number. Specifically, this includes: Game resources are extracted and preprocessed using a dedicated parsing plugin for the game engine, and a mapping between text resources and game context resources is established. Based on the association mapping, construct a terminology version tree; Bind the terminology version tree to the game version number to establish a terminology change tracking mechanism between versions; When a game version update is detected, the terminology change tracking mechanism automatically identifies terminology differences and triggers an update of the translation memory of the associated text in the terminology version tree based on these differences.

3. The method according to claim 1, characterized in that, The method, based on a terminology version tree, categorizes and tags game text according to game context, and constructs a game context vector space by combining game screenshots, 3D scene models, and audio dialogues. Specifically, this includes: Based on the terminology context information stored in the terminology version tree, the game text is classified and marked in multiple dimensions according to the terminology category attributes and text function positions to determine the game function attributes of the game text. Based on the game's functional attributes, and combining game screenshots, 3D scene models, and audio dialogues, a layered coding architecture is used to construct the game context vector space. The classified game text is mapped into the game context vector space. The high-dimensional context vector is converted into a fixed-length fingerprint code using a hash algorithm, generating a unique context fingerprint for each piece of game text.

4. The method according to claim 1, characterized in that, The method involves constructing a capability map based on translators' historical translation data, combining text categories and contextual features identified by contextual fingerprints for intelligent task allocation, and recommending the optimal translator combination for a specific game project. Specifically, this includes: Based on multi-dimensional analysis of translators' historical translation data, a translator competency map reflecting translators' professional abilities is constructed. Based on the text category and contextual features identified by the contextual fingerprint, a requirement vector for the translation task is generated; The matching calculation between the demand vector and the translator capability vector in the translator capability map determines the fit between the translator and the translation task. Based on the fit results, candidate translators are selected for specific game projects. The translators are then assigned according to their collaboration history, time zone distribution, and workload to generate an optimal translator combination recommendation scheme.

5. The method according to claim 1, characterized in that, The method utilizes the visual references provided by the game context vector space, combined with a pre-built translation risk prediction model, to mark high-risk texts and automatically generate alternative translations, specifically including: By utilizing the multimodal visualization references provided by the game context vector space, translators can be given an immersive translation context. Based on a pre-built translation risk prediction model, a multi-dimensional risk assessment is performed on the current translated text to identify and mark high-risk text segments. For the identified high-risk texts, a strategy is combined using a rule base, translation memory, and neural networks to generate multiple alternative translation schemes. The alternative translation schemes are then prioritized using a quality assessment algorithm.

6. The method according to claim 5, characterized in that, The pre-construction steps of the translation risk prediction model include: Establish a multi-level risk feature identification system to determine various risk features in game localization. Based on various risk characteristics, training samples were collected and labeled from historical error databases, expert-annotated corpora, and cross-cultural comparison data to construct a basic dataset; Based on the basic dataset and combined with domain knowledge of the game industry, a multi-task learning framework is used to train a hierarchical hybrid neural model architecture to generate a translation risk prediction model. The translation risk prediction model includes a feature extraction layer, an attention mechanism layer, and a multi-task output layer.

7. The method according to any one of claims 1 to 6, characterized in that, The process of performing quality checks on the translated text through an integrated GQA model in the real-time collaborative monitoring interface, and feeding back the GQA model's check results to the terminology version tree, specifically includes: The GQA model is integrated into the real-time collaborative monitoring interface to perform real-time quality assessment of translated texts and obtain quality assessment results. Based on the quality assessment results, the anomalies found in the translated text are classified and graded according to severity, and the priority of handling the anomalies is determined. Based on the processing priority, abnormal issues are mapped one by one to the corresponding nodes in the terminology library version tree, establishing a traceability path from quality issues to the root cause of the terminology. Based on the tracing pathway, the impact of terminology changes on translated texts is analyzed, and a gradual update strategy is formulated.

8. A game translation collaborative management and monitoring system, characterized in that, The system specifically includes: The first processing module is used to extract and preprocess game resources through a dedicated parsing plugin for the game engine, build a terminology version tree based on the preprocessed game resources, and bind and associate the terminology version tree with the game version number. The second processing module is used to classify and label game text based on the terminology version tree and the game context, and to construct a game context vector space by combining game screenshots, 3D scene models and audio dialogues. The game context vector space is used to generate a unique context fingerprint for each piece of game text. The third processing module is used to construct a capability map based on the translator's historical translation data, and to intelligently allocate tasks by combining the text categories and contextual features identified by the context fingerprint, and recommend the optimal translator combination for a specific game project. The fourth processing module is used to mark high-risk texts and automatically generate alternative translations during the context-assisted translation process based on the optimal translator combination, by utilizing the visual references provided by the game context vector space and combining them with a pre-built translation risk prediction model. The fifth processing module is used to perform quality checks on the translated text through the integrated GQA model in the real-time collaborative monitoring interface, and to feed back the GQA model's check results to the terminology version tree.

9. A computer device, characterized in that, include: The memory and processor, and the computer program stored in the memory, when the computer program is executed on the processor, implement the game translation collaborative management and monitoring method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the game translation collaborative management and monitoring method as described in any one of claims 1 to 7.

Images