An automatic transliteration method for geographical names based on a specific rule base

Abstract

This application relates to the technical field of geographical name transliteration. Specifically, it discloses an automatic geographical name transliteration method based on a specific rule library. First, it splits the geographical name to be transliterated input by the user to obtain multiple geographical name segments, and loads the corresponding rule library based on the type of each geographical name segment to parse each geographical name segment to obtain multiple initial candidate translations. Furthermore, it introduces natural language processing technology based on deep learning to perform semantic parsing and context semantic association encoding on each initial candidate translation to obtain the global context semantic association features of the candidate translations, and with the help of a large language model, performs geographical name translation based on the context semantic information of the candidate translations to generate the geographical name transliteration result. By combining a specific rule library and semantic parsing technology, this method can comprehensively consider the language characteristics, cultural background, and context of geographical names, thereby improving the accuracy of geographical name transliteration.

Description

Technical Field

[0001] This application relates to the technical field of place name transliteration, and more specifically, to an automatic place name transliteration method based on a specific rule library. Background Art

[0002] With the acceleration of the globalization process, cross-border exchanges have become increasingly frequent. As an important part of geographical information, place names play an indispensable role in fields such as international exchanges, tourism, and logistics. Place name transliteration, as an important link in cross-cultural communication and geographical information processing, requires accurately and appropriately converting place names from one language to another while maintaining the meaning, cultural characteristics, and geographical orientation of the place names.

[0003] However, traditional place name transliteration methods often rely on manual translation or automatic translation based on simple rules. On the one hand, the efficiency of manual transliteration is low. Facing a large number of place name translation requirements, manual translation often fails to meet the timeliness requirements, seriously affecting the project progress. On the other hand, although the rule-based transliteration method improves the transliteration efficiency to a certain extent, the formulation of rules is often based on limited language samples. In reality, place names have a wide range of sources, complex and diverse language structures, and many place names have unique historical, cultural, and regional backgrounds, making it difficult to fully cover them with rules, resulting in insufficient accuracy and cultural adaptability of the rule-based transliteration results.

[0004] Therefore, an optimized automatic place name transliteration method based on a specific rule library is needed to solve the above technical problems. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide an automatic place name transliteration method based on a specific rule library. First, the place name to be transliterated input by the user is split to obtain multiple place name segments, and the corresponding rule library is loaded based on the type of each place name segment to parse each place name segment to obtain multiple initial candidate translations. Then, natural language processing technology based on deep learning is introduced to perform semantic parsing and context semantic association encoding on each initial candidate translation to obtain the global context semantic association features of the candidate translations, and with the help of a large language model, place name translation is performed based on the context semantic information of the candidate translations to generate the place name transliteration result. By combining a specific rule library and semantic parsing technology, this method can comprehensively consider the language characteristics, cultural background, and context of place names, thereby improving the accuracy of place name transliteration.

[0006] According to one aspect of this application, an automatic place name transliteration method based on a specific rule library is provided, which includes:

[0007] Obtain the place name to be transliterated input by the user;

[0008] Split the to-be-translated place name to obtain the sequence distribution of place name segments;

[0009] Based on the types of the individual place name segments in the sequence distribution of the place name segments, parse the individual place name segments based on a specific rule base to obtain the sequence distribution of initial candidate translations;

[0010] Perform semantic embedding encoding on the individual initial candidate translations in the sequence distribution of the initial candidate translations to obtain the sequence distribution of initial candidate translation semantic embedding encoding vectors;

[0011] Perform context association encoding based on semantic change perception on the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain candidate translation context semantic association encoding vectors;

[0012] Generate the place name translation result based on the candidate translation context semantic association encoding vectors.

[0013] Specifically, based on the types of the individual place name segments in the sequence distribution of the place name segments, parsing the individual place name segments based on a specific rule base to obtain the sequence distribution of initial candidate translations includes:

[0014] Identify the types of the individual place name segments in the sequence distribution of the place name segments based on a place name classifier to obtain the sequence distribution of place name segment types;

[0015] Based on the sequence distribution of the place name segment types, load the rule base for the corresponding place name types;

[0016] Based on the rule base for the corresponding place name types, parse the individual place name segments in the sequence distribution of the place name segments to obtain the sequence distribution of the initial candidate translations.

[0017] Specifically, performing semantic embedding encoding on the individual initial candidate translations in the sequence distribution of the initial candidate translations to obtain the sequence distribution of initial candidate translation semantic embedding encoding vectors includes:

[0018] Use a semantic embedding encoder based on the Bert model to perform semantic embedding encoding on the individual initial candidate translations in the sequence distribution of the initial candidate translations to obtain the sequence distribution of the initial candidate translation semantic embedding encoding vectors.

[0019] Specifically, performing context association encoding based on semantic change perception on the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain candidate translation context semantic association encoding vectors includes:

[0020] Performing context - neighborhood - aware semantic transfer significance measurement on each initial candidate translation semantic embedding encoding vector in the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the sequence distribution of candidate translation semantic transfer significant factors;

[0021] Based on the sequence distribution of the candidate translation semantic transfer significant factors, performing feature modulation transfer aggregation encoding on the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the candidate translation context semantic association encoding vectors.

[0022] Specifically, performing context - neighborhood - aware semantic transfer significance measurement on each initial candidate translation semantic embedding encoding vector in the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the sequence distribution of candidate translation semantic transfer significant factors includes:

[0023] Calculating the semantic jump degree of each initial candidate translation semantic embedding encoding vector in the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the sequence distribution of candidate translation semantic jump degrees;

[0024] Calculating the semantic transfer space span of each initial candidate translation semantic embedding encoding vector in the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the sequence distribution of candidate translation semantic transfer space spans;

[0025] Based on the semantic jump degree and semantic transfer space span of each initial candidate translation semantic embedding encoding vector in the sequence distribution of the initial candidate translation semantic embedding encoding vectors, calculating the semantic transfer significant factor of each initial candidate translation semantic embedding encoding vector to obtain the sequence distribution of the candidate translation semantic transfer significant factors.

[0026] Specifically, based on the sequence distribution of the candidate translation semantic transfer significant factors, performing feature modulation transfer aggregation encoding on the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the candidate translation context semantic association encoding vectors includes:

[0027] Inputting the sequence distribution of the candidate translation semantic transfer significant factors into a gated transfer unit containing a softmax normalization function to obtain the sequence distribution of candidate translation semantic transfer significant weights;

[0028] Based on the sequence distribution of the candidate translation semantic transfer significant weights, performing weighted aggregation on the sequence distribution of the initial candidate translation semantic embedding encoding vectors to obtain the candidate translation context semantic association encoding vectors.

[0029] Specifically, based on the candidate translation context semantic association encoding vectors, generating a result of place - name translation and writing, including:

[0030] After adding a prompt word to the tail of the context semantic association encoding vector of the candidate translation, input it into a translation polishing tool based on a large language model to obtain the transliteration result of the place name.

[0031] Specifically, the prompt word is: "Polish the place name translation based on the context semantic information of the candidate translation".

[0032] The present application has at least the following technical effects:

[0033] Compared with the prior art, the automatic place name transliteration method based on a specific rule library provided by the present application first splits the place name to be transliterated input by the user to obtain multiple place name segments, and loads the corresponding rule library based on the type of each place name segment to parse each place name segment to obtain multiple initial candidate translations. Furthermore, a natural language processing technology based on deep learning is introduced to perform semantic parsing and context semantic association encoding on each initial candidate translation to obtain the global context semantic association features of the candidate translation, and with the help of a large language model, perform place name translation based on the context semantic information of the candidate translation to generate the transliteration result of the place name. By combining a specific rule library and semantic parsing technology, this method can comprehensively consider the language characteristics, cultural background and context of the place name, thereby improving the accuracy of place name transliteration. Brief Description of the Drawings

[0034] By describing the embodiments of the present application in more detail with reference to the accompanying drawings, the above and other objects, features and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0035] Figure 1 It is a flowchart of an automatic place name transliteration method based on a specific rule library according to an embodiment of the present application.

[0036] Figure 2 It is a schematic diagram of data flow of an automatic place name transliteration method based on a specific rule library according to an embodiment of the present application.

[0037] Figure 3 It is a flowchart of sub-step S3 of an automatic place name transliteration method based on a specific rule library according to an embodiment of the present application.

[0038] Figure 4 It is a flowchart of sub-step S5 of an automatic place name transliteration method based on a specific rule library according to an embodiment of the present application.

[0039] Figure 5It is a flowchart of sub-step S51 of the method for automatically transliterating geographical names based on a specific rule library according to an embodiment of the present application.

[0040] Figure 6 It is a flowchart of sub-step S52 of the method for automatically transliterating geographical names based on a specific rule library according to an embodiment of the present application. Detailed implementation manners

[0041] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0042] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0043] In the present application, flowcharts are used to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in order. On the contrary, as needed, various steps can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0044] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0045] It should be noted that all actions of obtaining data in the present application are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where it is located and obtaining the authorization given by the owner of the corresponding device.

[0046] In view of the technical problems described in the above background art, the present application proposes an optimized automatic place name transliteration method based on a specific rule base. First, the place name to be transliterated input by the user is split to obtain multiple place name segments, and the corresponding rule base is loaded based on the type of each place name segment to parse each place name segment to obtain multiple initial candidate translations. Then, a natural language processing technology based on deep learning is introduced to perform semantic parsing and context semantic association encoding on each initial candidate translation to obtain the global context semantic association features of the candidate translations, and with the help of a large language model, place name translation is performed based on the context semantic information of the candidate translations to generate the place name transliteration result. By combining a specific rule base and semantic parsing technology, this method can comprehensively consider the language characteristics, cultural background and context of place names, thereby improving the accuracy of place name transliteration.

[0047] Figure 1 It is a flowchart of the automatic place name transliteration method based on a specific rule base according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of the automatic place name transliteration method based on a specific rule base according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the automatic place name transliteration method based on a specific rule base includes the steps: S1, obtaining the place name to be transliterated input by the user; S2, splitting the place name to be transliterated to obtain the sequence distribution of place name segments; S3, based on the type of each place name segment in the sequence distribution of place name segments, performing parsing based on a specific rule base on each place name segment to obtain the sequence distribution of initial candidate translations; S4, performing semantic embedding encoding on each initial candidate translation in the sequence distribution of initial candidate translations to obtain the sequence distribution of initial candidate translation semantic embedding encoding vectors; S5, performing context association encoding based on semantic change perception on the sequence distribution of initial candidate translation semantic embedding encoding vectors to obtain candidate translation context semantic association encoding vectors; S6, generating the place name transliteration result based on the candidate translation context semantic association encoding vectors.

[0048] In the above method for automatic transliteration of geographical names based on a specific rule library, in step S1, the geographical name to be transliterated input by the user is obtained. It should be understood that the purpose of automatic transliteration of geographical names is to meet the user's need for translating specific geographical names. In practical applications, a dedicated input box is usually set up in the software interface or application program, and the user can directly manually input the geographical name to be transliterated in this input box. At the same time, to improve the convenience and accuracy of input, the system may provide some auxiliary functions. For example, the automatic association function. When the user inputs some characters, the system provides possible complete geographical name options for the user based on historical data or a common geographical name library, facilitating the user to quickly select. It may also set up an error correction prompt function. When the content input by the user has obvious spelling mistakes or does not conform to the conventional geographical name format, the system promptly gives a prompt to guide the user to correct it.

[0049] Specifically, to adapt to the application requirements in different scenarios, the system supports multiple ways to receive the geographical name information to be transliterated provided by the user. On the one hand, for users accessing the system through a web page or a mobile application, the interface design is intuitive and friendly, providing a simple and clear text input box that allows the user to directly type in the geographical name to be translated. At the same time, to improve the user's input efficiency and reduce possible spelling mistakes, the system integrates an intelligent prompt function that can quickly associate and display a list of possible target geographical names for selection based on the input partial characters. This not only speeds up the information entry speed but also increases the accuracy of the initial input. In addition, the intelligent prompt function also combines historical search records and geographical location information, making the recommended results more personalized and relevant. On the other hand, considering that some users may not know the correct writing of specific geographical names or hope to search for a specific location within a larger geographical range, the system also provides a map interaction function. The user can mark the target location on the map, and the system will automatically retrieve and provide the main geographical names in the nearby area for the user to confirm. This method is especially suitable for use by people with special geographical location requirements such as travel enthusiasts and logistics practitioners. Even users who are not very familiar with the local language can accurately specify the geographical name to be translated. The map interaction interface also provides rich geographical information query functions, such as surrounding facilities and transportation routes, further enhancing the user experience.

[0050] To meet the requirements of batch processing, such as updating the address database of multinational enterprises or large-scale place name conversion tasks in academic research, the system has opened a file upload interface, supporting the import of data files in various common formats such as Excel and CSV. Users can submit a document containing a large number of place name records at one time, and the system will parse and store these place names one by one, preparing for the next translation process. To ensure data security and privacy protection, an encrypted transmission protocol is adopted during the upload process, and relevant laws and regulations regarding the handling of personal information are strictly adhered to. In addition, to facilitate users to manage and track the status of upload tasks, the system also provides detailed log records and progress reports.

[0051] The application of speech recognition technology has greatly expanded the service scope of the system, especially for user groups who are not good at typing or unable to use traditional input devices. The system is compatible with speech recognition in multiple languages and dialects, and has strong noise resistance and speech rate adaptability, ensuring that even in a noisy environment or when the speech rate varies, the user's intention can be accurately captured. The speech assistant can not only handle single place name query requests, but also understand relatively complex natural language instructions, such as "find all cities along the shortest path from Beijing to Paris", and convert these instructions into a specific list of place names. Users can select from the list of place names to determine the place names to be transliterated.

[0052] In the above method for automatically transliterating place names based on a specific rule library, in step S2, the place name to be transliterated is split to obtain the sequence distribution of place name segments. It should be understood that considering that place names often consist of multiple parts with different semantic and grammatical functions, and these parts may follow different translation rules. For example, a complete place name may include different parts such as administrative division names, natural geographical entity names, landmark building or facility names, etc., and different translation strategies may be required during translation. Therefore, in this application, the place name to be transliterated is split to decompose it into multiple place name segments, forming the sequence distribution of place name segments. In this way, each place name segment corresponds to a component in the original place name, so that the corresponding rule library can be selected for each place name segment for parsing, thereby improving the accuracy and flexibility of translation.

[0053] Specifically, considering that a geographical name may consist of multiple parts, such as a city name, a street name, a building name, or a number, etc., the system adopts a multi-level analysis method. For compound geographical names, the system first identifies each component within the overall framework, for example, by detecting specific conjunctions or punctuation marks to distinguish different elements. This initial classification helps break down complex geographical names into smaller and more manageable units. On this basis, for each individual component, the system applies a more detailed language model for further analysis. For example, for geographical names containing direction indicators (east, west, south, north) or descriptive words (big, small, new, old), the system will identify these modifiers according to predefined patterns and separate the corresponding parts as independent geographical name segments. In addition, the system can also identify and separate geographical name segments with time attributes, such as "old", "new", and adjectives indicating size, such as "big", "small".

[0054] In the above method for automatically transliterating geographical names based on a specific rule library, in step S3, based on the types of each geographical name segment in the sequence distribution of the geographical name segments, each geographical name segment is parsed based on the specific rule library to obtain a sequence distribution of initial candidate translations. Among them, Figure 3 FIG. is a flowchart of sub-step S3 of the method for automatically transliterating geographical names based on a specific rule library according to an embodiment of the present application. As Figure 3 shown, step S3 includes steps: S31, based on a geographical name classifier, identifying the types of each geographical name segment in the sequence distribution of the geographical name segments to obtain a sequence distribution of geographical name segment types; S32, based on the sequence distribution of the geographical name segment types, loading the rule library corresponding to the geographical name types; S33, based on the corresponding rule library of the geographical name types, parsing each geographical name segment in the sequence distribution of the geographical name segments to obtain the sequence distribution of the initial candidate translations.

[0055] Specifically, in step S31, based on a geographical name classifier, the types of each geographical name segment in the sequence distribution of the geographical name segments are identified to obtain a sequence distribution of geographical name segment types. It should be understood that since different types of geographical name segments have different translation habits and rules. For example, the translation of administrative division names often follows official translation standards, while the translation of natural geographical entity names may rely more on cultural habits and geographical features. Therefore, the present application further uses a geographical name classifier to identify the types of geographical name segments to ensure that an appropriate rule library can be selected for geographical name translation in the follow-up. In an embodiment of the present application, the geographical name classifier adopts a neural network model based on deep learning. By training a large amount of geographical name data, the decision boundary between different types of geographical name data is established to map each geographical name segment to a preset category label respectively, so as to achieve accurate identification of the types of geographical name segments.

[0056] Specifically, in step S32, based on the sequence distribution of the place name segment types, the rule base corresponding to the respective place name types is loaded. It should be understood that each rule base contains translation rules formulated for specific types of place name segments. For example, for administrative division names, the rule base may contain official standard translated names to ensure the accuracy of translation; for natural geographical entity names, the rule base may contain translated recommended names based on geographical features and cultural habits to enhance the cultural adaptability and geographical orientation of the translation. By loading the rule base that matches the place name segment type, the present application can adopt appropriate translation rules and strategies for different types of place name segments, thereby improving the pertinence and accuracy of translation. Specifically, the present application pre-establishes a rule base management system, classifies and stores different types of place name rule bases, and in the rule base management system, an index table is established to associate the place name type with the corresponding rule base file path or database table name. After receiving the sequence distribution of the place name segment types, by traversing this sequence, for each place name segment type, the storage location of the corresponding rule base is found according to the index table. For example, if a certain place name segment is identified as "administrative division type", the system finds the storage path of the administrative division type rule base according to the index table, and then loads this rule base into the memory through file reading operations or database query operations for subsequent parsing of this place name segment.

[0057] Specifically, in step S33, based on the rule base of the corresponding place name type, each place name segment in the sequence distribution of the place name segments is parsed to obtain the sequence distribution of the initial candidate translated names. Specifically, for each place name segment, a one-by-one matching query is performed in the rule base of the corresponding place name type to find possible translation options for each place name segment, that is, the initial candidate translated names, so as to obtain the sequence distribution of the initial candidate translated names, which is used as a preliminary translation result to lay the foundation for the final place name translation.

[0058] In the above method for automatically transliterating geographical names based on a specific rule base, in step S4, semantic embedding encoding is performed on each initial candidate transliteration in the sequence distribution of the initial candidate transliterations to obtain a sequence distribution of initial candidate transliteration semantic embedding encoding vectors. It should be understood that considering that translating each geographical name segment individually may lead to incoherence in the overall geographical name transliteration result and insufficient context adaptability, for this, the present application further introduces natural language processing technology based on deep learning to perform context semantic mining and optimization processing on the sequence distribution of the initial candidate transliterations, so as to generate a more accurate and semantically fluent geographical name transliteration result. In a specific example of the present application, step S4 includes: using a semantic embedding encoder based on the Bert model to perform semantic embedding encoding on each initial candidate transliteration in the sequence distribution of the initial candidate transliterations to obtain the sequence distribution of the initial candidate transliteration semantic embedding encoding vectors. Specifically, the present application uses the Bert model, which has excellent performance in natural language processing tasks, as the semantic embedding encoder to perform semantic embedding encoding on each initial candidate transliteration, so as to utilize the strong semantic understanding ability and representation ability pre-trained by the Bert model in a large-scale corpus to capture the deep semantic information contained in each initial candidate transliteration, and represent each initial candidate transliteration as an embedding vector representation in a high-dimensional semantic space, thereby generating a sequence distribution of initial candidate transliteration semantic embedding encoding vectors.

[0059] In the above method for automatically transliterating geographical names based on a specific rule base, in step S5, context association encoding based on semantic change perception is performed on the sequence distribution of the initial candidate transliteration semantic embedding encoding vectors to obtain candidate transliteration context semantic association encoding vectors. It should be understood that during the process of translating geographical names, there are semantic correlations and coherences between different geographical name segments. Therefore, in order to effectively capture the semantic transitions and connections between each initial candidate transliteration, the present application adopts a context association encoding method based on semantic change perception, which captures the semantic transitions and coherences between adjacent names by performing semantic association analysis between adjacent initial candidate transliteration semantic embedding encoding vectors, thereby dynamically adjusting and optimizing the semantic expressions of each initial candidate transliteration in the overall geographical name transliteration result to enhance the semantic coherence and context adaptability of the overall geographical name transliteration. Among them, Figure 4 is a flowchart of sub-step S5 of the method for automatically transliterating geographical names based on a specific rule base according to an embodiment of the present application. As Figure 4As shown, step S5 includes steps: S51, performing context neighborhood-aware semantic transfer significance measurement on each initial candidate translation semantic embedding coding vector in the sequence distribution of the initial candidate translation semantic embedding coding vectors to obtain a sequence distribution of candidate translation semantic transfer significant factors; S52, based on the sequence distribution of the candidate translation semantic transfer significant factors, performing feature modulation transfer aggregation coding on the sequence distribution of the initial candidate translation semantic embedding coding vectors to obtain the candidate translation context semantic association coding vectors.

[0060] Figure 5 It is a flowchart of sub-step S51 of the automatic place name translation method based on a specific rule library according to an embodiment of the present application. As Figure 5 shown, step S51 includes steps: S511, calculating the semantic jump degree of each initial candidate translation semantic embedding coding vector in the sequence distribution of the initial candidate translation semantic embedding coding vectors to obtain a sequence distribution of candidate translation semantic jump degrees; S512, calculating the semantic transfer space span of each initial candidate translation semantic embedding coding vector in the sequence distribution of the initial candidate translation semantic embedding coding vectors to obtain a sequence distribution of candidate translation semantic transfer space spans; S513, based on the semantic jump degree and semantic transfer space span of each initial candidate translation semantic embedding coding vector in the sequence distribution of the initial candidate translation semantic embedding coding vectors, calculating the semantic transfer significant factor of each initial candidate translation semantic embedding coding vector to obtain the sequence distribution of the candidate translation semantic transfer significant factors.

[0061] More specifically, step S511 is represented by the formula:

[0062] ;

[0063] ;

[0064] ;

[0065] where represents the sequence distribution of the initial candidate translation semantic embedding coding vectors, , , and respectively represent the 1st, 2nd, th, and th initial candidate translation semantic embedding coding vectors in the sequence distribution of the initial candidate translation semantic embedding coding vectors, is the number of feature vectors in the sequence distribution of the initial candidate translation semantic embedding coding vectors, represents the The eigenvalue at the -th position in the semantic embedding coding vector of the -th initial candidate translation name, is the feature scale value of the semantic embedding coding vector of the -th initial candidate translation name, is the semantic intensity factor of the semantic embedding coding vector of the -th initial candidate translation name, denotes the semantic intensity factor of the semantic embedding coding vector of the -th initial candidate translation name, denotes the semantic jump degree of the semantic embedding coding vector of the

[0066] -th initial candidate translation name. Specifically, in order to keenly capture the semantic correlation fluctuations among the initial candidate translation names, the present application first calculates the semantic jump degree of the semantic embedding coding vectors of the initial candidate translation names to measure the severity of the semantic change relative to the adjacent candidate translation names. It should be understood that during the context transfer coding process of the semantic information of the initial candidate translation names, the semantic embedding coding vectors of the initial candidate translation names perform semantic interactions in sequence according to their order in the original sequence distribution. If the semantic jump degree between a certain initial candidate translation name and the subsequent one is large, that is, the semantic change is relatively severe, it means that there may be a large break or incoherence between them semantically. At this time, a higher semantic attention needs to be given to this initial candidate translation name for semantic smoothing processing to ensure the semantic coherence and fluency of the overall place name translation.

[0067] More specifically, the step S512 is expressed by the formula:

[0068] ;

[0069] where denotes the semantic transfer space span of the semantic embedding coding vector of the -th initial candidate translation name, denotes and the number of feature vectors separated between them.

[0070] Specifically, considering that the position of each candidate translation name in the overall sequence distribution also has a certain impact on its semantic importance. Therefore, the present application further calculates the semantic transfer space span of the semantic embedding coding vectors of the initial candidate translation names to reveal the position of each candidate translation name in the overall sequence distribution, so as to further evaluate its semantic contribution degree in the overall place name translation result.

[0071] More specifically, the step S513 is expressed by the formula:

[0072] ;

[0073] Among them, and are preset weight parameters used to balance the influence of semantic jump degree and semantic transmission space span, represents the semantic transmission significant factor of the semantic embedding coding vector of the th initial candidate translation name.

[0074] Specifically, in order to comprehensively consider the two important factors of semantic jump degree and semantic transmission space span and more accurately measure the importance of each initial candidate translation name semantic embedding coding vector in the context semantic transmission process, the present application further calculates its semantic transmission significant factor based on the semantic jump degree and semantic transmission space span of each initial candidate translation name semantic embedding coding vector, so as to focus more on important candidate translation names in the subsequent semantic transmission process, reduce the attention to secondary information, and thus perform more refined semantic adjustment and optimization on the semantic feature expression of each initial candidate translation name.

[0075] Specifically, in step S52, based on the sequence distribution of the candidate translation name semantic transmission significant factor, feature modulation transfer aggregation coding is performed on the sequence distribution of the initial candidate translation name semantic embedding coding vector to obtain the candidate translation name context semantic association coding vector. Among them, Figure 6 is a flowchart of sub-step S52 of the automatic geographical name transliteration method based on a specific rule library according to an embodiment of the present application. As Figure 6 shown, step S52 includes steps: S521, inputting the sequence distribution of the candidate translation name semantic transmission significant factor into a gated transfer unit including a softmax normalization function to obtain the sequence distribution of the candidate translation name semantic transmission significant weight; S522, based on the sequence distribution of the candidate translation name semantic transmission significant weight, performing weighted aggregation on the sequence distribution of the initial candidate translation name semantic embedding coding vector to obtain the candidate translation name context semantic association coding vector.

[0076] More specifically, step S521 is represented by the formula:

[0077] ;

[0078] Among them, is the normalization exponential function, is the gating mask function, is the gating threshold, is the candidate translation name semantic transmission significant weight of the .

[0079] Specifically, the present application further introduces a gating mechanism to perform gating screening on the semantic transmission significant factors of each initial candidate translation name semantic embedding coding vector, generating a sequence distribution of the candidate translation name semantic transmission significant weights.

[0080] More specifically, the step S522 is represented by the formula:

[0081] ;

[0082] where represents the candidate translation name context semantic association coding vector.

[0083] Specifically, based on the candidate translation name semantic transmission significant weights, a weighted aggregation is performed on the sequence distribution of the original initial candidate translation name semantic embedding coding vectors, thereby generating a candidate translation name context semantic association coding vector. In this way, the semantic roles and importance of each initial candidate translation name in the overall place name translation result can be more accurately characterized, making the final place name translation result not only accurately convey the meaning of the original place name, but also be more coherent and fluent semantically, thereby improving the accuracy and readability of place name translation.

[0084] In the above place name automatic translation method based on a specific rule library, the step S6 generates a place name translation result based on the candidate translation name context semantic association coding vector. In a specific example of the present application, the step S6 includes: after adding a prompt word to the tail of the candidate translation name context semantic association coding vector, inputting it into a translation polishing tool based on a large language model to obtain the place name translation result. Among them, the prompt word is: "Polish the place name translation based on the context semantic information of the candidate translation name". It should be understood that a large language model refers to an artificial intelligence model trained on a large scale of text data, with extensive knowledge and complex language understanding capabilities. Such models can capture the subtle differences in language and make reasonable inferences based on context information. In the application scenario of the present application, the translation polishing tool uses a pre-trained large language model as the basic framework and is fine-tuned for the place name translation task, enabling the model to better understand and handle the unique challenges of geographical names, such as the correct conversion of proper nouns and the retention of local characteristics.

[0085] Specifically, through training on a large scale of corpora, the large language model has accumulated rich language knowledge and translation experience. Therefore, after receiving the candidate translation name context semantic association coding vector containing the prompt word, the large language model can, according to the indication of the prompt word, deeply understand and analyze the candidate translation name context semantic association coding vector, comprehensively consider the semantic information and context association of the candidate translation name, and use its powerful language generation ability and translation experience to generate an accurate and fluent place name translation result.

[0086] In a preferred example of the present application, since each initial candidate translation semantic embedding coding vector in the sequence distribution of the initial candidate translation semantic embedding coding vectors respectively represents the semantic embedding coding features of the initial candidate translations of the corresponding place name segments, when performing context semantic coding of feature sequence message passing based on the feature jump degree, the difference in the population attributes of the embedding coding semantic features in the local semantic space will cause differences in the fairness of the feature jump degree level, thereby affecting the inclusiveness of the context semantic association distribution transfer of the candidate translation context semantic association coding vector in the global semantic space, and reducing the semantic quality of the place name translation result obtained by inputting it into the translation polishing tool based on the large language name.

[0087] Therefore, in this preferred example, after adding a prompt word to the tail of the candidate translation context semantic association coding vector and inputting it into the translation polishing tool based on the large language name, the candidate translation context semantic association coding vector is first optimized. The specific optimization process is as follows:

[0088] Determine the number of hyperdistribution eigenvalue numbers in the candidate translation context semantic association coding vector that are greater than the threshold difference from the feature mean, and calculate the reciprocal of the logarithm to the base 2 of the number of hyperdistribution eigenvalue numbers and the exponential value to the base of the natural constant of the reciprocal of the number of hyperdistribution eigenvalue numbers respectively to obtain the first candidate translation context semantic association non-linear trajectory representation value and the second candidate translation context semantic association non-linear trajectory representation value :

[0089] ;

[0090] ;

[0091] ;

[0092] wherein, represents the candidate translation context semantic association coding vector, represents the th eigenvalue of the candidate translation context semantic association coding vector, represents the mean of all eigenvalues of the candidate translation context semantic association coding vector, represents the calculation of number, is the number of hyperdistribution eigenvalue numbers in the candidate translation context semantic association coding vector, and is the threshold hyperparameter;

[0093] Calculate the hyperbolic sine function value of the sum of the squares of all eigenvalues of the candidate translation context semantic association coding vector:

[0094] ;

[0095] Wherein, represents the length of the context semantic association coding vector of the candidate translation, represents the hyperbolic sine function, represents the value of the hyperbolic sine function;

[0096] And calculate the exponential value of the hyperbolic sine function value with the natural constant as the base, and then divide it by the square of the length of the context semantic association coding vector of the candidate translation to obtain the context semantic association potential manifold value of the candidate translation , wherein, represents the natural constant, represents the context semantic association potential manifold value of the candidate translation;

[0097] Calculate the context semantic association covariance integration value of the context semantic association potential manifold value of the candidate translation with respect to the first context semantic association non-linear trajectory representation value and the second context semantic association non-linear trajectory representation value of the candidate translation:

[0098] ;

[0099] Wherein, represents the context semantic association covariance integration value of the candidate translation;

[0100] Calculate the power function eigenvector of the context semantic association coding vector of the candidate translation with the reciprocal of the context semantic association covariance integration value of the candidate translation, and calculate the autocorrelation matrix of the power function eigenvector to obtain the context semantic association adaptive representation matrix, that is:

[0101] ;

[0102] Wherein, represents calculating the power function eigenvector of the context semantic association coding vector of the candidate translation with the reciprocal of the context semantic association covariance integration value of the candidate translation, represents matrix multiplication, represents the transpose of the vector, represents the context semantic association adaptive representation matrix;

[0103] Multiply the context semantic association coding vector of the candidate translation with the context semantic association adaptive representation matrix to obtain the optimized context semantic association coding vector of the candidate translation, that is:

[0104] ;

[0105] Here, the candidate translation context semantic association coding vector is a row vector, representing the optimized candidate translation context semantic association coding vector.

[0106] Specifically, for the dynamic mode clusters of the candidate translation context semantic association coding vector, under the constraint of the generation of the trajectory in the random field domain, the cross-modal localization energy consumption fluctuation phenomenon mainly stems from exceeding the global action threshold of the adjacent scope. By implementing the dynamic manifold latent variable analysis method based on covariance tensor decomposition, the heterogeneous association graph hidden in its feature space is analyzed, and by constructing an elastic manifold latent variable framework with dimensional elasticity, a dynamic coupling representation protocol between the dynamic interaction time series projection parameters is re-established. Thus, through this reverse modeling scheme of the energy state transition path, the discrete phase transition modeling under the global association of parameters is effectively realized, significantly enhancing the context semantic association distribution transmission inclusiveness of the candidate translation context semantic association coding vector in the global semantic space. In this way, the accuracy of the place name translation results obtained by its input of the translation polishers based on large language names is improved.

[0107] In order to maintain the original cultural characteristics and regional orientation of place names, the translation polisher has a built-in evaluation system for measuring whether the candidate translations meet the established criteria. This evaluation system takes into account various factors such as phonetic beauty, consistency of literal meaning, local idiomatic expressions, etc., to ensure that the output results are both scientific and reasonable and can resonate. In addition, the system introduces a user feedback mechanism, allowing the performance of the model to be continuously improved through correction suggestions in actual applications, so as to achieve continuous optimization. This dynamic adjustment process not only improves the performance of the model but also accumulates valuable experience for future place name translation tasks.

[0108] During this process, special attention also needs to be paid to protecting user privacy and data security. All transmitted data uses an encryption protocol to ensure that information will not be intercepted or tampered with during transmission. At the same time, relevant laws and regulations regarding the processing of personal information need to be strictly complied with to ensure that the rights and interests of users are not violated. In addition, to improve the transparency and trust of the system, the system also provides detailed log records and progress reports, allowing users to understand the status and progress of the task at any time.

[0109] In summary, the method for automatically transliterating place names based on a specific rule library according to the embodiments of the present application is elucidated. First, the place name to be transliterated input by the user is split to obtain multiple place name segments, and the corresponding rule library is loaded based on the type of each place name segment to parse each place name segment to obtain multiple initial candidate translations. Furthermore, a natural language processing technology based on deep learning is introduced to perform semantic parsing and context semantic association encoding on each initial candidate translation to obtain the global context semantic association features of the candidate translations, and with the help of a large language model, place name translation is performed based on the context semantic information of the candidate translations to generate the place name transliteration result. By combining a specific rule library and semantic parsing technology, this method can comprehensively consider the language characteristics, cultural background, and context of place names, thereby improving the accuracy of place name transliteration.

[0110] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and easy understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above specific details for implementation.

[0111] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0112] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be encompassed by the present invention. Any associated drawing marks in the claims should not be regarded as limiting the claimed rights.

[0113] In addition, it is obvious that the term "including" does not exclude other units or steps, and the singular form does not exclude the plural form. A plurality of units stated in the system claims can also be implemented by one unit through software or hardware.

[0114] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for automatically translating place names based on a specific rule base, characterized in that: include: Obtaining the place name to be translated input by the user; Splitting the place name to be translated to obtain a sequence distribution of place name fragments; Based on the type of each place name segment in the sequence distribution of the place name segment, the each place name segment is parsed based on a specific rule base to obtain a sequence distribution of initial candidate translated names; Performing semantic embedding coding on each initial candidate translated name in the sequence distribution of the initial candidate translated names to obtain a sequence distribution of semantic embedding coding vectors of the initial candidate translated names; Performing context-association coding based on semantic change perception on the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain a candidate translated name context semantic association coding vector; Based on the candidate translated name context semantic association coding vector, a place name translation result is generated.

2. The method for automatically translating place names based on a specific rule base according to claim 1 is characterized in that: Based on the type of each place name segment in the sequence distribution of the place name segment, the each place name segment is parsed based on a specific rule base to obtain the sequence distribution of the initial candidate translated name, including: Identifying the type of each place name segment in the sequence distribution of the place name segments based on the place name classifier to obtain a sequence distribution of the place name segment types; Based on the sequence distribution of the place name segment type, loading a rule library of the corresponding place name type; Based on the rule base of the corresponding place name type, each place name segment in the sequence distribution of the place name segments is parsed to obtain the sequence distribution of the initial candidate translated names.

3. The method for automatically translating place names based on a specific rule base according to claim 2 is characterized in that: Performing semantic embedding coding on each initial candidate translated name in the sequence distribution of the initial candidate translated names to obtain a sequence distribution of semantic embedding coding vectors of the initial candidate translated names, including: A semantic embedding encoder based on the Bert model is used to perform semantic embedding encoding on each initial candidate translated name in the sequence distribution of the initial candidate translated names to obtain a sequence distribution of the semantic embedding encoding vectors of the initial candidate translated names.

4. The method for automatically translating place names based on a specific rule base according to claim 3 is characterized in that: The sequence distribution of the initial candidate translated name semantic embedding coding vector is subjected to contextual association coding based on semantic change perception to obtain a candidate translated name contextual semantic association coding vector, including: Performing a context-neighborhood-aware semantic transfer significance measurement on each initial candidate translated name semantic embedding coding vector in the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain a sequence distribution of candidate translated name semantic transfer significance factors; Based on the sequence distribution of the semantic transfer significance factors of the candidate translated names, feature modulation transfer aggregation coding is performed on the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain the candidate translated name context semantic association coding vector.

5. The method for automatically translating place names based on a specific rule base according to claim 4 is characterized in that: The method of performing a context-neighborhood-aware semantic transfer significance measurement on each initial candidate translated name semantic embedding coding vector in the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain a sequence distribution of candidate translated name semantic transfer significance factors includes: Calculating the semantic jump degree of each initial candidate translated name semantic embedding coding vector in the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain the sequence distribution of the candidate translated name semantic jump degree; Calculating the semantic transfer space span of each initial candidate translated name semantic embedding coding vector in the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain the sequence distribution of the candidate translated name semantic transfer space span; Based on the semantic jump degree and semantic transfer space span of each initial candidate translated name semantic embedding coding vector in the sequence distribution of the initial candidate translated name semantic embedding coding vector, the semantic transfer significance factor of each initial candidate translated name semantic embedding coding vector is calculated to obtain the sequence distribution of the candidate translated name semantic transfer significance factor.

6. The method for automatically translating place names based on a specific rule base according to claim 5 is characterized in that: Based on the sequence distribution of the semantic transfer significance factor of the candidate translated name, feature modulation transfer aggregation coding is performed on the sequence distribution of the initial candidate translated name semantic embedding coding vector to obtain the candidate translated name context semantic association coding vector, including: Inputting the sequence distribution of the semantic transfer significance factors of the candidate translated names into a gated transfer unit including a softmax normalization function to obtain the sequence distribution of the semantic transfer significance weights of the candidate translated names; Based on the sequence distribution of the candidate translated name semantically significant weights, weighted aggregation is performed on the sequence distribution of the initial candidate translated name semantic embedding coding vectors to obtain the candidate translated name context semantic association coding vector.

7. The method for automatically translating place names based on a specific rule base according to claim 6 is characterized in that: Based on the candidate translated name context semantic association encoding vector, generating a place name translation result, including: After adding a prompt word to the tail of the candidate translated name context semantic association encoding vector, it is input into a translation polisher based on a large language model to obtain the place name translation result.

8. The method for automatically translating place names based on a specific rule base according to claim 7 is characterized in that: The prompt word is: "polish the place name translation based on the contextual semantic information of the candidate translated names."

Images