Geographic name translation method combining multi-language syllable segmentation and adaptive learning

Abstract

The invention relates to the technical field of geographical name translation, and discloses a multi-language syllable segmentation and adaptive learning combined geographical name translation method, which comprises the following steps of: performing word segmentation and general name translation annotation on a source language geographical name address based on a general dictionary, converting an unannotated proper name part into an IPA phoneme sequence, and translating the IPA phoneme sequence into an IPA phoneme sequence; performing syllable division and translation matching on the IPA phoneme sequence of each word segment, matching a plurality of possible candidate transliteration results for each word segment, and then combining each candidate transliteration result of the special name word segment with a general name translation result to construct a place name address target language translation result candidate set; and performing adaptive feature learning and aggregation optimization on the candidate set by using a deep learning algorithm so as to comprehensively consider semantic information and context coherence of a multi-transliteration version and generate a final place name address target language translation. According to the application, the problems of inaccurate transliteration, rigidity, ambiguity and the like in the traditional place name and address translation method can be solved, and the translation efficiency and quality are improved.

Description

Technical Field

[0001] This application relates to the technical field of place name translation, and more specifically, to a place name translation method that combines multi - language syllable segmentation and adaptive learning. Background Art

[0002] In today's increasingly globalized world, accurately and efficiently translating place names and addresses between different languages has become crucial. Whether it is international trade, logistics transportation, cross - border tourism, or emergency rescue, geographical information system applications, the clarity and accuracy of place names and addresses are directly related to the smooth progress of activities and the effective transmission of information. However, due to language differences, cultural backgrounds, naming habits, and the diversity of writing systems, the translation of place names and addresses is far from a simple word substitution. Place names and addresses usually contain a large number of proper names (specific names), and these names often do not have directly corresponding target - language words and need to be converted through transliteration. Existing place name and address translation solutions mainly rely on pre - established dictionaries or rule - based transliteration systems, or attempt to use general statistical machine translation. Although these methods can handle some place name and address translation problems to a certain extent, they generally have limitations.

[0003] Specifically, due to the huge differences in the phonological systems of different languages, including inconsistent phonemic inventories, different syllable structure rules, and restrictions on legitimate phoneme combinations (phonotactics), a single source syllable may have multiple target transliteration units to approximately represent it, making it difficult for direct, simple one - to - one or fixed - rule mapping to capture all transliteration possibilities and select the optimal solution, often resulting in inaccurate, rigid, or even ambiguous transliteration results. General machine translation models are also prone to producing incorrect or non - conforming transliteration results when dealing with place names and addresses containing a large number of specific names, making it difficult for existing methods to meet the complex and ever - changing place name and address translation requirements with significant cross - language differences.

[0004] Therefore, an optimized place name translation method that combines multi - language syllable segmentation and adaptive learning is needed to solve the above - mentioned technical problems. Summary of the Invention

[0005] To solve the above - mentioned technical problems, this application is proposed.

[0006] According to one aspect of this application, a place name translation method that combines multi - language syllable segmentation and adaptive learning is provided, which includes: Obtain the source - language place name and address to be translated and the target translation language input by the user; Based on a general dictionary and the target translation language, split-match and translation annotate the source language place name and address to be translated to obtain a sequence of source language place name and address word segments containing common name translation annotations; Perform syllable-segmentation-based transliteration matching on each unannotated source language place name and address word segment in the sequence of source language place name and address word segments containing common name translation annotations to obtain a candidate set of target language transliteration results for each unannotated source language place name and address word segment; Assemble the candidate set of target language transliteration results for each unannotated source language place name and address word segment with the sequence of source language place name and address word segments containing common name translation annotations to obtain a candidate set of target language translations of the place name and address; Perform adaptive learning aggregation optimization on the candidate set of target language translations of the place name and address to obtain the target language translation of the place name and address.

[0007] Beneficial effects: Compared with the prior art, the place name translation method combining multi-language syllable segmentation and adaptive learning provided by this application performs word segmentation based on a general dictionary and common name translation annotation on the source language place name and address, and after converting the unannotated proper name part into an IPA phoneme sequence, performs syllable division and translation matching on each word segment IPA phoneme sequence based on a multi-source place name historical transliteration knowledge base, matches multiple possible candidate transliteration results for each word segment, and then combines each candidate transliteration result of the proper name segment with the common name translation result to construct a candidate set of target language translations of the place name and address with multiple transliteration versions, and uses a deep learning algorithm to perform adaptive feature learning and aggregation optimization on this candidate set to comprehensively consider the semantic information and context coherence of multiple transliteration versions and generate the final target language translation of the place name and address. This application can solve problems such as inaccurate, rigid, and ambiguous transliteration existing in traditional place name and address translation methods, and improve translation efficiency and quality. Description of the Drawings

[0008] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The 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 drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 It is a flowchart of a place name translation method combining multi-language syllable segmentation and adaptive learning according to an embodiment of the present application.

[0010] Figure 2Schematic diagram of data flow for the method of translating place names by combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application.

[0011] Figure 3 Flowchart of sub - step S2 of the method of translating place names by combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application.

[0012] Figure 4 Flowchart of sub - step S3 of the method of translating place names by combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application.

[0013] Figure 5 Flowchart of sub - step S5 of the method of translating place names by combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application.

[0014] Figure 6 Flowchart of sub - step S52 of the method of translating place names by combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application.

[0015] Figure 7 Flowchart of sub - step S523 of the method of translating place names by combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application. Detailed implementation mode

[0016] As shown in the present application and the claims, unless the context clearly indicates an exceptional situation, words such as "a", "an", "one" and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "including" and "comprising" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0017] 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.

[0018] Flowcharts are used in the present application 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, they can be executed in reverse order or processed 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.

[0019] 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.

[0020] It should be noted in advance that all relevant processing of the data in the present application is carried out on the premise of complying with the corresponding data protection regulations and policies of the place where it is located and obtaining authorization from the corresponding authority managers.

[0021] Figure 1 It is a flowchart of a place name translation method combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of a place name translation method combining multi - language syllable segmentation and adaptive learning according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the place name translation method combining multi - language syllable segmentation and adaptive learning includes the steps: S1, obtaining the source - language place name address to be translated and the target translation language input by the user; S2, based on a general dictionary and the target translation language, splitting, matching and translation - annotating the source - language place name address to be translated to obtain a sequence of source - language place name address word segments containing common - name translation annotations; S3, performing syllable - segmentation - based transliteration matching on each unannotated source - language place name address word segment in the sequence of source - language place name address word segments containing common - name translation annotations to obtain a candidate set of target - language transliteration results for each unannotated source - language place name address word segment; S4, assembling the candidate set of target - language transliteration results for each unannotated source - language place name address word segment with the sequence of source - language place name address word segments containing common - name translation annotations to obtain a candidate set of target - language translations of the place name address; S5, performing adaptive learning aggregation optimization on the candidate set of target - language translations of the place name address to obtain the target - language translation of the place name address.

[0022] In the above-mentioned method for translating place names by combining multi-language syllable segmentation with adaptive learning, in step S1, the source language place name address to be translated and the target translation language input by the user are obtained. Specifically, in order to establish the initial parameters of the translation task and ensure that the system adapts to the multi-language environment, based on user interaction and the design of the multi-language input interface, this application dynamically configures the language rule library and the transliteration knowledge base for the subsequent processing flow by receiving the source language place name address input by the user and the specified target translation language. In the specific implementation process, the original place name and the target language identifier (such as "English") input by the user are captured through the front-end interface or the API interface, and the legality of the input format is verified (such as filtering non-text characters). In this way, the translation direction can be accurately locked, providing language type constraints for subsequent word segmentation, transliteration, and semantic optimization, and avoiding rule conflicts caused by unclear target languages.

[0023] In the specific implementation process, the user may directly type in a complete address string, such as "Jianguomenwai Avenue, Chaoyang District, Beijing", or may input a specific place name proper noun, such as "Fragrant Hills". At the same time, the user also needs to clearly specify the target translation language, such as English, Japanese, French, etc. This target language identifier not only determines the selection of subsequent transliteration rules but also affects the behavior logic of the multi-language dictionary matching module. To improve the convenience of user use, the system usually provides ways such as drop-down menus and language code selectors to guide the user to correctly input the target language type and avoid the failure of the entire translation process due to language recognition errors.

[0024] After receiving the original input, the system immediately starts the format legality verification mechanism. Since place name addresses often contain non-standard characters such as spaces, numbers, and special symbols, it is necessary to clean and standardize the input content. For example, removing consecutive multiple spaces, filtering illegal control characters, and unifying the case format, etc. For Chinese input, it is also necessary to determine whether there is a mixture of pinyin or other Latin alphabet spelling forms; for non-Latin writing systems such as Arabic and Russian, character encoding conversion is required to ensure that the subsequent processing module can correctly parse. In addition, the system also needs to identify and separate the common name part (such as "district", "road", "avenue") and the proper name part (such as "Fragrant Hills") in the address for subsequent processing using different translation strategies.

[0025] Given the complexity in a multilingual environment, the system must possess flexible language detection capabilities. When the user does not explicitly specify the target language, the system can attempt to automatically infer the target language type based on context clues. However, for the sake of accuracy, it is still recommended that the user actively select the target language. The language detection module is trained on a large-scale multilingual text and can quickly identify the language system to which the input place name belongs, and call the corresponding phoneme mapping rule library in combination with the target language setting. For example, when a Chinese place name is translated into English, the system will first call the mapping table from Chinese pinyin to English pronunciation habits; when a Chinese place name is translated into Japanese, the transliteration rule set based on katakana will be enabled.

[0026] To enhance the user experience and improve the system's intelligence level, the system also introduces an input suggestion and error correction mechanism. During the user's input process, the system can dynamically prompt possible place name options based on the partially entered characters, reducing the interference caused by typing errors. In addition, after submitting the complete address, the system will perform a syntactic structure analysis on the input content to identify whether there are spelling mistakes, duplicates, or illogical combinations. For example, if the user enters "Beijing District, Beijing City", the system can prompt that "Beijing District" may have a spelling deviation and suggest correcting it to a standard place name expression such as "Dongcheng District" or "Xicheng District".

[0027] Throughout the input processing flow, the system always follows the design concept of "dynamic configuration". That is, according to the source language place name address provided by the user and the target language identifier, the corresponding processing rule library and transliteration knowledge base are loaded in real time. This on-demand loading method not only improves the system's response speed but also enhances its adaptability when facing different language combinations. For example, when translating from Chinese to English, the system will enable the syllable segmentation model based on IPA (International Phonetic Alphabet) and the English pronunciation rule database; when translating from Chinese to German, it will switch to the German phonetic system rule library to ensure that the generated transliteration result conforms to the pronunciation habits of the target language.

[0028] In the above method for translating place names by combining multilingual syllable segmentation and adaptive learning, in step S2, based on a general dictionary and the target translation language, the source language place name address to be translated is split, matched, and translation-annotated to obtain a sequence of source language place name address word segments containing common name translation annotations. Among them, Figure 3 is a flowchart of sub-step S2 of the method for translating place names by combining multilingual syllable segmentation and adaptive learning according to an embodiment of the present application. As Figure 3As shown, step S2 includes steps: S21, performing word segmentation processing on the to-be-translated source language place name and address based on general dictionary matching to obtain a sequence of to-be-translated source language place name and address word segments; S22, based on the general dictionary and the target translation language, performing translation annotation on each to-be-translated source language place name and address word segment in the sequence of to-be-translated source language place name and address word segments to obtain the sequence of source language place name and address word segments containing common name translation annotation.

[0029] Specifically, in step S21, word segmentation processing is performed on the to-be-translated source language place name and address based on general dictionary matching to obtain a sequence of to-be-translated source language place name and address word segments. It should be understood that since place names and addresses usually consist of multiple words or phrases, and the address contains various components such as street names, administrative division names, building names, and directions, and different address components require different translation strategies and processing methods. Therefore, in order to separate the common name and proper name parts in the to-be-translated source language place name and address and give priority to processing the general part that can be freely translated, based on the rule matching mechanism of the multilingual geographical term dictionary, the to-be-translated source language place name and address are subjected to word boundary recognition and segment segmentation through a predefined general dictionary (covering common common names and their multilingual translation comparison tables). In the specific implementation process, the bidirectional maximum matching algorithm (BMM) is used to scan the to-be-translated source language place name and address from left to right and from right to left to identify each common name segment in the to-be-translated source language place name and address, and the place name and address are segmented into common name segments and un-matched proper name segments according to the matching results, so as to initially distinguish the general part (such as "street", "district") that can be directly translated in the place name and address and the proper names (such as specific place names, building names) that need further transliteration processing, laying a foundation for subsequent processing.

[0030] Specifically, the step S22, based on the universal dictionary and the target translation language, performs translation annotation on each source language place name and address word segment to be translated in the sequence of source language place name and address word segments to be translated to obtain the sequence of source language place name and address word segments containing common name translation annotations. It should be understood that since the common name translation in the place name and address needs to follow the standardized translation rules of the target language (such as the Chinese "路" corresponds to the English "Road" instead of the transliterated "Lu"), there is no need to perform complex transliteration processing. Therefore, the present application is based on the cross-language common name mapping table stored in the universal dictionary, and through dictionary query, semantic annotation is performed on the common name segments after word segmentation. In the specific implementation process, by traversing the sequence of source language place name and address word segments to be translated, for each identified common name segment, the target language translation pre-stored in the universal dictionary is queried, and the translation result is used as the translation annotation of the common name segment. For example, if a segment matches a common name entry in a general dictionary (such as "山" is marked as a common name in Chinese), it is directly replaced with the standard translation name in the target language (such as "Mountain"), and its position mark is retained; while the unmatched segment is marked as a proper name to be transliterated. In this way, the common name part of the place name address can be quickly and accurately given a standardized translation in the target language, providing a basis for subsequent transliteration and semantic optimization steps.

[0031] In the place name translation method combining multilingual syllable segmentation with adaptive learning, the step S3 performs transliteration matching based on syllable segmentation on each unlabeled source language place name and address word segment to be translated in the sequence of source language place name and address word segments containing common name translation annotations to obtain a candidate set of target language transliteration results for each unlabeled source language place name and address word segment to be translated. Specifically, since the transliteration of proper names needs to solve the problem of cross-language phonological mismatch, in order to capture the correspondence between multilingual phonemes and generate transliteration results, this application is based on the phonological neutrality of the International Phonetic Alphabet (IPA), constructs a phoneme-transliteration unit probability mapping table through a multi-source place name historical transliteration knowledge base, and uses the historical transliteration data in the transliteration knowledge base to query the possible transliteration results of each unlabeled source language place name and address word segment to be translated. Among them, Figure 4 Flow chart of sub-step S3 of the place name translation method combining multilingual syllable segmentation and adaptive learning according to an embodiment of the present application. Figure 4As shown, the step S3 includes the steps of: S31, performing IPA phoneme mapping on the unlabeled source language place name and address word fragment to be translated to obtain an IPA phoneme sequence of the place name and address word fragment to be translated; S32, constructing a multi-source place name historical transliteration knowledge base, and based on the multi-source place name historical transliteration knowledge base, performing syllable segmentation on the IPA phoneme sequence of the place name and address word fragment to be translated and matching it with the target language transliteration unit to obtain a plurality of target language transliteration result initial candidate units; S33, calculating the phoneme sequence edit distance between each of the multiple target language transliteration result initial candidate units and the IPA phoneme sequence of the place name and address word fragment to be translated, and selecting the target language transliteration result initial candidate units whose phoneme sequence edit distance is less than a preset threshold to form a target language transliteration result candidate set of the unlabeled source language place name and address word fragment to be translated.

[0032] Specifically, the step S31 performs IPA phoneme mapping on the unlabeled source language place name and address word segment to be translated to obtain an IPA phoneme sequence of the place name and address word segment to be translated. Specifically, due to the significant differences in phonemes between writing systems of different languages (such as Chinese characters, Japanese kana, Cyrillic letters, etc.), direct transliteration based on the orthography of the source language may result in distortion of phoneme representation. For example, the actual pronunciation of the Japanese character "京都" is , and the "J" in the Spanish place name "Jerez" is pronounced / x / (similar to the German "ch"), and if it is directly mapped according to English spelling habits, it will be incorrectly converted to . Therefore, in order to eliminate orthographic interference and achieve cross-language phoneme alignment, this application is based on the phonological neutrality of the International Phonetic Alphabet (IPA). By calling multilingual IPA mapping tables (such as Chinese pinyin to IPA, Japanese kana to IPA, Spanish orthography to IPA, etc.), combined with sound change rules (such as Chinese erhua, French liaison), phoneme-level standardization conversion is performed to convert source language place name and address word fragments into a unified IPA phoneme sequence independent of the written form. In this way, phoneme misjudgment caused by differences in writing systems can be avoided (such as avoiding converting Russian " " " in the sentence is pronounced as the Latin letter "X"), providing high-fidelity phonological input for subsequent syllable segmentation.

[0033] Specifically, in step S32, a multi-source historical transliteration knowledge base of geographical names is constructed, and based on the multi-source historical transliteration knowledge base of geographical names, syllable segmentation and matching with target language transliteration units are performed on the IPA phoneme sequence of the to-be-translated geographical name address word segment to obtain multiple initial candidate units of target language transliteration results. It should be understood that considering the possible polysemy and cultural sensitivity of transliteration results, for example, when an English geographical name is transliterated into Chinese, one syllable can be transliterated into multiple approximate Chinese characters, and different combinations of Chinese characters may imply different cultural meanings or cause ambiguities. Therefore, in order to generate accurate and culturally appropriate transliteration results, this application constructs a multi-source historical transliteration knowledge base of geographical names to use historical transliteration experience to perform syllable segmentation on the IPA phoneme sequence of the to-be-translated geographical name address word segment and perform query matching at the syllable level to find target language transliteration units that are similar in pronunciation to the source language and come from real historical data. In the specific implementation process, first, multiple historical transliteration cases similar to the IPA phoneme sequence of the current to-be-translated geographical name address word segment are retrieved from the knowledge base, and then, according to the target language syllable templates of each historical transliteration case, the IPA phoneme sequence of the to-be-translated geographical name address word segment is segmented into corresponding syllable units, and the syllable units are matched with the target language transliteration units in the historical transliteration cases, thereby generating multiple initial candidate units of target language transliteration results.

[0034] Specifically, in step S33, the phoneme sequence edit distance between each initial candidate unit of the target language transliteration result in the multiple initial candidate units of the target language transliteration result and the IPA phoneme sequence of the to-be-translated geographical name address word segment is calculated, and the initial candidate units of the target language transliteration result with a phoneme sequence edit distance less than a preset threshold are selected to form the candidate set of the target language transliteration result of the unannotated to-be-translated source language geographical name address word segment. Specifically, since the transliteration process is essentially a phoneme approximation mapping rather than a completely equivalent conversion, and the degree of phonological deviation of different candidate units directly affects the acceptability of the translated name. Therefore, in order to quantify the approximation degree of each initial candidate unit of the transliteration result to the source language pronunciation and exclude significantly distorted schemes to ensure the quality of the candidate set, this application is based on the phoneme sequence edit distance algorithm, and calculates the minimum editing operation cost from the IPA phoneme sequence of the source language geographical name address word segment to the phoneme sequence of the initial candidate unit of the target language transliteration result through dynamic programming, including operations such as insertion, deletion, and replacement. The smaller the edit distance, the higher the similarity between the initial candidate unit of the target language transliteration result and the source language pronunciation. By setting a reasonable edit distance threshold (such as 0.3×sequence length), only the initial candidate units of the transliteration result with an edit distance less than this threshold are retained, thereby forming the final candidate set of the target language transliteration result. In this way, while retaining reasonable transliteration variants, it can effectively suppress excessive phonological distortion and ensure that the candidate set has both phonological rationality and historical normativity.

[0035] In the above-mentioned method for translating geographical names by combining multi-language syllable segmentation and adaptive learning, in step S4, the candidate set of target-language transliteration results of each unlabeled source-language geographical name address word segment to be translated is assembled with the sequence of source-language geographical name address word segments containing generic name translation annotations to obtain a candidate set of target-language translations of geographical name addresses. It should be understood that in the sequence of source-language geographical name address word segments to be translated, the generic name part has been standardized and translated through standard translations, while multiple transliteration result candidates have been generated for the specific name part. To further evaluate the impact of different transliteration combinations on the overall translation effect of geographical name addresses, in this application, the standard translation of the generic name part in the source-language geographical name address to be translated is assembled with the candidate set of transliteration results of the specific name part to generate all possible complete target-language translation strings of geographical name addresses, and they are normalized according to the grammar rules and expression habits of the target language to construct a candidate set of target-language translations of geographical name addresses containing multiple transliteration versions. In this way, a candidate space containing multi-modal transliteration possibilities can be constructed, providing rich materials for the final translation of geographical name addresses.

[0036] In the above-mentioned method for translating geographical names by combining multi-language syllable segmentation and adaptive learning, in step S5, the candidate set of target-language translations of geographical name addresses is adaptively learned, aggregated, and optimized to obtain the target-language translation of geographical name addresses. It should be understood that there may be problems such as expression differences and insufficient cultural adaptability among different transliteration versions in the candidate set of target-language translations of geographical name addresses, and the traditional scoring mechanism only relies on phoneme similarity and cannot capture the semantic rationality and context adaptability of the translation in the target language. Therefore, to further improve the accuracy and cultural adaptability of the translation of geographical name addresses, this application introduces an adaptive learning aggregation and optimization mechanism based on deep learning algorithms. By performing context semantic learning and aggregation optimization on the candidate set of target-language translations of geographical name addresses, the semantic information and context coherence of multiple transliteration versions are comprehensively considered to generate the final target-language translation of geographical name addresses. Among them, Figure 5 is a flowchart of sub-step S5 of the method for translating geographical names by combining multi-language syllable segmentation and adaptive learning according to an embodiment of the present application. As Figure 5 shown, step S5 includes the steps of: S51, extracting the context semantic features of each target-language translation of geographical name addresses in the candidate set of target-language translations of geographical name addresses to obtain a candidate set of semantic encoding vectors of target-language translations of geographical name addresses; S52, performing aggregation analysis on the candidate set of semantic encoding vectors of target-language translations of geographical name addresses to obtain a semantic aggregation analysis encoding vector of target-language translations of geographical name addresses; S53, performing feature decoding on the semantic aggregation analysis encoding vector of target-language translations of geographical name addresses to obtain the target-language translation of geographical name addresses.

[0037] Specifically, in a specific example of the present application, the step S51 includes: performing context semantic encoding based on the Transformer architecture on each place name and address target language translation result in the candidate set of the place name and address target language translation results to obtain a candidate set of semantic encoding vectors of the place name and address target language translation results. It should be understood that in order to evaluate the naturalness and standardization of each place name and address target language translation result in the target language semantic space, the present application is based on the deep context perception capability of the Transformer architecture, and performs context semantic encoding on each place name and address target language translation result, so as to utilize the self-attention mechanism of the Transformer architecture to model the global semantic dependency in the place name and address target language translation results, capture the place name structure characteristics and cultural implicit meanings of the place name and address target language translation results, and convert them into a quantifiable semantic space distribution, and obtain a candidate set of semantic encoding vectors of the place name and address target language translation results, thereby providing basic feature support for the semantic aggregation optimization of subsequent multi-transliteration versions.

[0038] Specifically, the step S52 performs aggregation analysis on the candidate set of semantic coding vectors of the target language translation results of the place name and address to obtain the semantic aggregation analysis coding vectors of the target language translation results of the place name and address. Specifically, the present application takes into account that simply performing mean aggregation on the candidate set of semantic coding vectors of the target language translation results of the place name and address or selecting a certain optimal translation result semantic coding feature as the final representation may not be sufficient to capture the complementary information or potential better combinations contained in multiple candidates. Therefore, in order to further capture the semantic complementarity in multiple transliteration result candidates and dig out potential better combinations, the present application proposes a semantic aggregation optimization method, which first performs deep aggregation analysis on the candidate set of semantic coding vectors of the target language translation results of the place name and address, extracts the main semantic direction of the set, and further embeds the individual semantic offsets of the semantic coding vectors of the target language translation results of each place name and address relative to the main semantic direction in the main semantic direction of the set, so as to simulate the interaction and potential information fusion between multiple candidate translation results, and generate semantic aggregation analysis coding vectors of the target language translation results of the place name and address. In this way, we can achieve deep integration and optimization of multiple candidate translation results at the semantic level, avoid information loss in the direct voting mechanism, and help improve the accuracy and robustness of the target language translation of place names and addresses. Figure 6 FIG. 5 is a flowchart of sub-step S52 of the method for translating place names by combining multilingual syllable segmentation with adaptive learning according to an embodiment of the present application. Figure 6As shown in the figure, step S52 includes the steps of: S521, performing ground state feature aggregation encoding on the candidate set of the semantic encoding vectors of the target language translation result of the place name address to obtain a ground state feature aggregation encoding vector of the target language translation result of the place name address; S522, extracting the semantic compensation excited state features of each semantic encoding vector of the target language translation result of the place name address in the candidate set of the semantic encoding vectors of the target language translation result of the place name address relative to the ground state feature aggregation encoding vector of the target language translation result of the place name address to obtain a set of semantic compensation excited state feature encoding vectors of the target language translation result of the place name address; S523, based on the set of semantic compensation excited state feature encoding vectors of the target language translation result of the place name address, performing gain modulation on the ground state feature aggregation encoding vector of the target language translation result of the place name address to obtain the semantic aggregation analysis encoding vector of the target language translation result of the place name address.

[0039] More specifically, step S521 includes: First, inputting the candidate set of the semantic encoding vectors of the target language translation result of the place name address into a K-Means clustering network to obtain K semantic clustering center encoding vectors of the target language translation result of the place name address, which is expressed by the formula: ; ; Among them, represents the candidate set of the semantic encoding vectors of the target language translation result of the place name address, , , and respectively represent the first, the second, the th, and the th semantic encoding vectors of the target language translation result of the place name address in represents the set of real numbers, represents the dimension of the semantic encoding vector of the target language translation result of the place name address, represents the number of the semantic encoding vectors of the target language translation result of the place name address, represents input into the K-Means clustering network, represents the set of semantic clustering center encoding vectors of the target language translation result of the place name address, , , and respectively represent the first, the second, and the th semantic clustering center encoding vectors of the target language translation result of the place name address in represents the number of the semantic clustering center encoding vectors of the target language translation result of the place name address.

[0040] That is, the K-Means clustering network is used to perform unsupervised learning on the candidate set of semantic encoding vectors of the target language translation results of geographical names and addresses, realizing data condensation and structured detection, identifying the centroids of regions with higher data density from the high-dimensional semantic feature space, and obtaining K semantic clustering center encoding vectors of the target language translation results of geographical names and addresses that can represent the main semantic directions, providing structured pattern primitives for subsequent ground state feature aggregation based on the self-attention mechanism, reducing the complexity of the feature space, and laying a foundation for integrating the semantic information of multiple transliteration versions, performing adaptive feature learning, and aggregation optimization.

[0041] Then, perform set ground state feature aggregation based on the self-attention mechanism on the K semantic clustering center encoding vectors of the target language translation results of geographical names and addresses to obtain the ground state feature aggregation encoding vector of the target language translation results of geographical names and addresses, which is expressed by the formula: ; ; ; ; ; Wherein, 、 and respectively represent the query weight matrix, the key weight matrix, and the value weight matrix, 、 and respectively represent the query matrix, the key matrix, and the value matrix, represents the self-attention layer, represents the normalization exponential function, represents the transpose of the vector, represents the layer normalization function, represents the ground state feature aggregation encoding vector of the target language translation results of geographical names and addresses.

[0042] That is, the self-attention mechanism is used to mine the internal semantic associations and global structure information between the K semantic clustering center encoding vectors of the target language translation results of geographical names and addresses, breaking through the limitation that K-Means clustering only provides local center representations, enabling each semantic clustering center encoding vector of the target language translation results of geographical names and addresses to fuse the context semantic information of other centers, generating the ground state feature aggregation encoding vector of the target language translation results of geographical names and addresses, thereby effectively capturing the deep connections at the semantic level of different transliteration candidate versions, and providing a core feature representation with both local representativeness and global coherence for subsequent adaptive learning and aggregation optimization.

[0043] More specifically, the step S522 is expressed by the formula: ; ; Among them, represents feature concatenation, represents the sigmoid activation function, represents the compensation weight matrix, represents the compensation bias term, represents dot product, represents the set of semantic compensation excited state feature encoding vectors of the translation result of the place name address in the target language, , , and respectively represent the 1st, the 2nd, the th, and the th semantic compensation excited state feature encoding vectors of the translation result of the place name address in the target language.

[0044] That is, by calculating the deviation components of the semantic encoding vectors of the translation results of the place name address in the target language relative to the ground state, the individual-specific semantic information that is not explained by the ground state features in each candidate translation result is separated, and it is explicitly represented as the semantic compensation excited state feature encoding vectors of the translation result of the place name address in the target language, enabling the network to quantify and process the unique semantic differences of different transliteration versions, while retaining the global semantic consensus and highlighting the semantic characteristics of individual candidates, providing a richer feature representation for subsequent adaptive learning.

[0045] Figure 7 is a flowchart of sub-step S523 of the place name translation method combining multilingual syllable segmentation and adaptive learning according to an embodiment of the present application. As Figure 7 shown, the step S523 includes steps: S5231, calculating the excited state significance modulation weight factors of each semantic compensation excited state feature encoding vector in the set of semantic compensation excited state feature encoding vectors of the translation result of the place name address in the target language to obtain a set of semantic compensation significance modulation weight factors of the translation result of the place name address in the target language; S5232, based on the set of semantic compensation significance modulation weight factors of the translation result of the place name address in the target language, performing weighted modulation on the set of semantic compensation excited state feature encoding vectors of the translation result of the place name address in the target language to obtain a set of semantic compensation excited state feature significant encoding vectors of the translation result of the place name address in the target language; S5233, inputting the set of semantic compensation excited state feature significant encoding vectors of the translation result of the place name address in the target language and the ground state feature aggregation encoding vector of the translation result of the place name address in the target language into a feature gain superposition network to obtain the semantic aggregation analysis encoding vector of the translation result of the place name address in the target language.

[0046] Specifically, when calculating the semantic compensation excited state representation of each semantic coding vector of the target language translation result of the geographical name and address with respect to the ground state feature aggregation coding vector of the target language translation result of the geographical name and address, by it essentially defines the derived metric of the metric space of each semantic coding vector of the target language translation result of the geographical name and address with respect to the ground state space of the ground state feature aggregation coding vector of the target language translation result of the geographical name and address, that is, using the feature spaces of the relatively high-dimensional semantic coding vectors of each target language translation result of the geographical name and address as the metric, and constructing a low-dimensional metric within the ground state space of the ground state feature aggregation coding vector of the target language translation result of the geographical name and address through a derived mapping. In this way, it is necessary to consider the interface measurement constraint under the derived metric, that is, the metric of the excited state is restricted by the geometric boundary of the space transition under the induced mapping, rather than a simple high-dimensional to low-dimensional space transformation. Based on this, in a preferred example of the present application, the step S5231 includes: First, based on the key difference boundary limitation between each semantic coding vector of the target language translation result of the geographical name and address in the candidate set of the semantic coding vectors of the target language translation result of the geographical name and address and the ground state feature aggregation coding vector of the target language translation result of the geographical name and address, perform feature correction on each semantic compensation excited state feature coding vector of the target language translation result of the geographical name and address to obtain a set of optimized semantic compensation excited state feature coding vectors of the target language translation result of the geographical name and address.

[0047] Specifically, first calculate with respect to the reduced order excitation factor: ; ; wherein, represents the L1 norm, represents the L2 norm, and represent different reduced order excitation factors respectively.

[0048] Then introduce a reduced-dimensional tangent space structure, so as to decompose the induced mapping in the form of a tangent space, and project it onto the orthogonal basis in the base space: ; ; wherein, and represent different reduced-dimensional tangent space vectors respectively.

[0049] Next, the dimensionality-reduced tangent space structure is used as the metric representation under the local submanifold to perform boundary-space geometric reconstruction, so as to convert the metric complexity of the excited state from being determined by the high-dimensional set representation of the semantic encoding vector of the target language translation result of the place name address to being constrained by the interface geometry of the tangent space metric: ; Among them, is a regulation factor used to compensate for the influence brought by the excessive spatial curvature parameter , represents the exponential function with the natural constant as the base, represents the corrected semantic encoding vector of the target language translation result of the corresponding place name address.

[0050] Then, using to correct as , among which, represents the th optimized semantic compensation excited state feature encoding vector of the set of optimized semantic compensation excited state feature encoding vectors of the target language translation result of the place name address. In this way, in the case of deriving the metric from the submanifold metric to the base space, by making the excitation state measure depend on the boundary rather than the space transformation, the fixed points in the tangential metric flow can be renormalized to achieve an effective measure of the boundary, avoiding the over-expansion of the representation of each semantic compensation excited state feature encoding vector of the target language translation result of the place name address.

[0051] Then, each optimized semantic compensation excited state feature encoding vector in the set of optimized semantic compensation excited state feature encoding vectors of the target language translation result of the place name address is input into the feature transformation compression module based on the ReLU function and then subjected to global normalization processing to obtain the set of semantic compensation significance modulation weight factors of the target language translation result of the place name address, which is expressed by the formula: ; ; Among them, represents the exponential function operation with as the base, represents the ReLU activation function, represents the attention score conversion vector, represents the transpose of the feature, represents the learnable weight parameter matrix, represents the set of semantic compensation significance modulation weight factors of the target language translation result of the place name address, , , and respectively represent the 1st, 2nd, th, and th semantic compensation significance modulation weight factors for the target language translation results of place name and address.

[0052] That is, through the feature transformation compression module based on the ReLU function, non-linear mapping and dimension screening are performed on the optimized semantic compensation excited state feature encoding vectors of the target language translation results of place name and address, highlighting the effective semantic difference features and suppressing redundant information. Through global normalization processing, the feature weights have cross-dimensional comparability, so as to assign reasonable importance measures to the individual semantic deviations of different transliteration candidate versions, construct a set of semantic compensation significance modulation weight factors for the target language translation results of place name and address that can be used for subsequent weighted modulation, provide a scientific basis for importance evaluation for the subsequent weighted modulation process, help focus on key semantic differences and weaken noise interference in the aggregation optimization, and improve the semantic accuracy and context adaptability of the target language translation of place name and address.

[0053] In a specific example of the present application, the step S5232 is expressed by the formula: ; ; where represents the set of significantly encoded vectors of the semantic compensation excited state features of the target language translation results of place name and address, , , and respectively represent the 1st, 2nd, th, and th significantly encoded vectors of the semantic compensation excited state features of the target language translation results of place name and address.

[0054] That is, through weighted modulation, each semantic compensation excited state feature encoding vector of the target language translation results of place name and address can be adjusted according to its corresponding semantic compensation significance modulation weight factor for the target language translation results of place name and address. The important semantic features are enhanced, while the relatively unimportant or possibly noisy features are weakened. In this way, the generated set of significantly encoded vectors of the semantic compensation excited state features of the target language translation results of place name and address can more accurately reflect the key semantic differences between different translation candidate results, thereby improving the quality and accuracy of place name and address translation, and making the generated translation better meet the needs of actual applications.

[0055] In a specific example of the present application, the step S5233 is expressed by the formula: ; where represents a multi-layer perceptron, represents the semantic aggregation analysis coding vector of the translated result of the place name and address in the target language.

[0056] That is, through the processing of the feature gain superposition network, the excited state features reflecting individual differences are effectively organically combined with the ground state features reflecting global commonalities. This fusion not only enriches the dimension of semantic representation but also enhances the ability to capture complex semantic relationships, which helps to better weigh various semantic factors in the subsequent translation decision-making process, thereby generating a more accurate, context-conforming, and practically demanded semantic aggregation analysis coding vector of the translated result of the place name and address in the target language, and improving the performance and practicality of the entire place name translation method.

[0057] Specifically, in step S53, the semantic aggregation analysis coding vector of the translated result of the place name and address in the target language is feature-decoded to obtain the translated text of the place name and address in the target language. Specifically, in order to reconstruct the abstract semantic representation into a natural language symbol sequence conforming to the morphological rules of the target language, this application is based on a Transformer decoder enhanced by adversarial training, and guides the generation process by introducing linguistic constraints (such as the orthographic rules of the target language and the capitalization convention of place names). In the specific implementation process, a two-stage decoding strategy is adopted: first, the standard Transformer decoder is used to decode the semantic aggregation analysis coding vector of the translated result of the place name and address in the target language into a sequence of coarse-grained word segments in the target language, and then a conditional random field (CRF) layer is used to impose structural constraints to correct format errors (such as correcting to "Lhasa City"). At the same time, an adversarial training mechanism is introduced, and the generated result is input into a discriminator (such as a pre-trained geographical entity recognition model) to evaluate its distribution consistency with the real place name, and the decoder parameters are optimized through gradient backpropagation to prompt the decoder to generate a more natural and accurate place name and address translation. For example, when converting the Spanish place name "Ciudad de México" into English, the discriminator can optimize the decoder to generate "Mexico City" instead of the literal translation "Ciudad de Mexico" by learning the English place name distribution. In this way, the translation method combining adversarial training and linguistic constraints can effectively improve the quality of place name and address translation, making it more in line with the expression habits and cultural backgrounds of the target language.

[0058] In summary, a method for translating geographical names that combines multi - language syllable segmentation and adaptive learning based on the embodiments of the present application is elucidated. It performs word segmentation based on a general dictionary and annotation of common name translations for the source - language geographical name address. After converting the unannotated proper name part into an IPA phoneme sequence, it performs syllable segmentation and translation matching for each word - segment IPA phoneme sequence based on a multi - source historical transliteration knowledge base of geographical names, matching multiple possible candidate transliteration results for each word segment. Furthermore, it combines the candidate transliteration results of each proper - name word segment with the common - name translation results to construct a candidate set of target - language translations of geographical name addresses with multiple transliteration versions, and uses a deep - learning algorithm to perform adaptive feature learning and aggregation optimization on this candidate set to comprehensively consider the semantic information and context coherence of multiple transliteration versions and generate the final target - language translation of the geographical name address. This method can solve the problems of inaccurate transliteration, stiffness, ambiguity, etc. existing in traditional geographical name address translation methods and improve translation efficiency and quality.

[0059] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, advantages, effects, etc. mentioned in the present invention are only examples and not limitations. It cannot be considered that these advantages, advantages, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above - mentioned embodiments are only for the purposes of illustration and easy understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above - mentioned specific details for implementation.

[0060] In the above - mentioned embodiments, the descriptions of each embodiment have their own focuses. For parts not detailed or recorded in a certain embodiment, reference can 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 merely illustrative. For example, the unit division is only a logical function division, and there can 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 can 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.

[0061] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, 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 embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

[0062] In addition, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. The multiple elements stated in the system claims can also be implemented by one element through software or hardware.

[0063] 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 place name translation method combining multilingual syllable segmentation and adaptive learning, characterized in that Including: Obtain the source language place name address to be translated and the target translation language input by the user; Based on a general dictionary and the target translation language, perform splitting matching and translation annotation on the source language place name address to be translated to obtain a sequence of source language place name address word segments containing common name translation annotations; Perform syllable-segmentation-based transliteration matching on each unannotated source language place name address word segment in the sequence of source language place name address word segments containing common name translation annotations to obtain a candidate set of target language transliteration results for each unannotated source language place name address word segment; Assemble the candidate set of target language transliteration results for each unannotated source language place name address word segment with the sequence of source language place name address word segments containing common name translation annotations to obtain a candidate set of target language translations of the place name address; Perform adaptive learning aggregation optimization on the candidate set of target language translations of the place name address to obtain the target language translation of the place name address.

2. The method for translating place names by combining multilingual syllable segmentation and adaptive learning according to claim 1, characterized in that Based on a general dictionary and the target translation language, performing splitting matching and translation annotation on the source language place name address to be translated to obtain a sequence of source language place name address word segments containing common name translation annotations includes: Perform word segmentation processing on the source language place name address to be translated based on general dictionary matching to obtain a sequence of source language place name address word segments; Based on the general dictionary and the target translation language, perform translation annotation on each source language place name address word segment in the sequence of source language place name address word segments to obtain the sequence of source language place name address word segments containing common name translation annotations.

3. The method for translating place names by combining multilingual syllable segmentation and adaptive learning according to claim 1, characterized in that, Performing syllable-segmentation-based transliteration matching on each unannotated source language place name address word segment in the sequence of source language place name address word segments containing common name translation annotations to obtain a candidate set of target language transliteration results for each unannotated source language place name address word segment includes: Perform IPA phoneme mapping on the unannotated source language place name address word segment to obtain an IPA phoneme sequence of the source language place name address word segment to be translated; Construct a multi-source historical transliteration knowledge base of place names, and based on the multi-source historical transliteration knowledge base of place names, perform syllable segmentation and target language transliteration unit matching on the IPA phoneme sequence of the source language place name address word segment to be translated to obtain multiple initial candidate units of target language transliteration results; Calculate the phoneme sequence edit distance between each initial candidate unit of the target language transliteration result in the multiple initial candidate units of the target language transliteration result and the IPA phoneme sequence of the source language place name address word segment to be translated, and select the initial candidate units of the target language transliteration result with the phoneme sequence edit distance less than a preset threshold to form the candidate set of target language transliteration results for the unannotated source language place name address word segment.

4. The method for translating geographical names by combining multilingual syllable segmentation and adaptive learning according to claim 3, wherein Performing adaptive learning aggregation optimization on the candidate set of target language translations of the place name address to obtain the target language translation of the place name address includes: Extract the context semantic features of each target language translation result of the place name and address in the candidate set of the target language translation results of the place name and address to obtain a candidate set of semantic encoding vectors of the target language translation results of the place name and address; Perform aggregation analysis on the candidate set of semantic encoding vectors of the target language translation results of the place name and address to obtain a semantic aggregation analysis encoding vector of the target language translation results of the place name and address; Perform feature decoding on the semantic aggregation analysis encoding vector of the target language translation results of the place name and address to obtain the target language translation of the place name and address.

5. The method for translating geographical names by combining multilingual syllable segmentation and adaptive learning according to claim 4, wherein Extract the context semantic features of each target language translation result of the place name and address in the candidate set of the target language translation results of the place name and address to obtain a candidate set of semantic encoding vectors of the target language translation results of the place name and address, including: Perform context semantic encoding based on the Transformer architecture on each target language translation result of the place name and address in the candidate set of the target language translation results of the place name and address to obtain a candidate set of semantic encoding vectors of the target language translation results of the place name and address.

6. The method for translating geographical names by combining multilingual syllable segmentation and adaptive learning according to claim 5, characterized in that Perform aggregation analysis on the candidate set of semantic encoding vectors of the target language translation results of the place name and address to obtain a semantic aggregation analysis encoding vector of the target language translation results of the place name and address, including: Perform ground state feature aggregation encoding on the candidate set of semantic encoding vectors of the target language translation results of the place name and address to obtain a ground state feature aggregation encoding vector of the target language translation results of the place name and address; Extract the semantic compensation excited state features of each semantic encoding vector of the target language translation result of the place name and address in the candidate set of the semantic encoding vectors of the target language translation results of the place name and address relative to the ground state feature aggregation encoding vector of the target language translation results of the place name and address to obtain a set of semantic compensation excited state feature encoding vectors of the target language translation results of the place name and address; Based on the set of semantic compensation excited state feature encoding vectors of the target language translation results of the place name and address, perform gain modulation on the ground state feature aggregation encoding vector of the target language translation results of the place name and address to obtain the semantic aggregation analysis encoding vector of the target language translation results of the place name and address.

7. The method for translating place names by combining multilingual syllable segmentation and adaptive learning according to claim 6, wherein Perform ground state feature aggregation encoding on the candidate set of semantic encoding vectors of the target language translation results of the place name and address to obtain a ground state feature aggregation encoding vector of the target language translation results of the place name and address, including: Input the candidate set of semantic encoding vectors of the target language translation results of the place name and address into a K-Means clustering network to obtain K semantic clustering center encoding vectors of the target language translation results of the place name and address; Perform set ground state feature aggregation on the K semantic clustering center encoding vectors of the target language translation results of the place name and address based on the self-attention mechanism to obtain the ground state feature aggregation encoding vector of the target language translation results of the place name and address.

8. The method for translating geographical names by combining multilingual syllable segmentation and adaptive learning according to claim 7, characterized in that, Based on the set of semantic compensation excited state feature encoding vectors of the target language translation results of the place name and address, perform gain modulation on the ground state feature aggregation encoding vector of the target language translation results of the place name and address to obtain the semantic aggregation analysis encoding vector of the target language translation results of the place name and address, including: Calculate the excited state significance modulation weight factors of each of the semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses to obtain a set of semantic compensation significance modulation weight factors of the target language translation results of the place name addresses; Based on the set of semantic compensation significance modulation weight factors of the target language translation results of the place name addresses, perform weighted modulation on the set of semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses to obtain a set of semantic compensation excited state feature significant encoding vectors of the target language translation results of the place name addresses; Input the set of semantic compensation excited state feature significant encoding vectors of the target language translation results of the place name addresses and the ground state feature aggregation encoding vectors of the target language translation results of the place name addresses into a feature gain superposition network to obtain the semantic aggregation analysis encoding vectors of the target language translation results of the place name addresses.

9. The method for translating geographical names by combining multilingual syllable segmentation and adaptive learning according to claim 8, wherein Calculating the excited state significance modulation weight factors of each of the semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses to obtain a set of semantic compensation significance modulation weight factors of the target language translation results of the place name addresses includes: Based on the key difference boundary limitation between each of the semantic encoding vectors of the target language translation results of the place name addresses in the candidate set of the semantic encoding vectors of the target language translation results of the place name addresses and the ground state feature aggregation encoding vectors of the target language translation results of the place name addresses, perform feature correction on each of the semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses to obtain a set of optimized semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses; Input each of the optimized semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses in the set of optimized semantic compensation excited state feature encoding vectors of the target language translation results of the place name addresses into a feature transformation compression module based on the ReLU function and then perform global normalization processing to obtain the set of semantic compensation significance modulation weight factors of the target language translation results of the place name addresses.

Images