Intelligent geographical name address translation method based on multi-language syllable segmentation

Abstract

The invention relates to the technical field of intelligent translation, and particularly discloses an intelligent geographical name and address translation method based on multi-language syllable segmentation, which comprises the following steps of: performing adaptive translation strategy decision on each vocabulary unit in a source language geographical name and address text by introducing a deep learning algorithm; according to the method, proper nouns which should adopt transliteration and general vocabularies which should adopt dictionary translation are automatically distinguished, and accurate transliteration of the proper nouns is realized by performing syllable-level fine-grained segmentation and target language syllable mapping on the proper nouns which are labeled as transliteration; and meanwhile, the general vocabularies marked as dictionary translation are accurately translated by adopting a dictionary matching algorithm, so that the transliteration result of the proper nouns and the translation result of the general vocabularies are normalized and recombined to obtain a place name address translation result. According to the application, the translation effect of the special place name during mixed output with the general word can be effectively improved, and the translation accuracy and flexibility of the place name address in a multi-language environment are enhanced.

Description

Technical Field

[0001] This application relates to the field of intelligent translation technology, and more specifically, to an intelligent place name and address translation method based on multi-language syllable segmentation. Background Art

[0002] With the continuous acceleration of the internationalization process, cross-language place name and address translation has increasingly become the core requirement for information interaction in multiple fields such as logistics, map services, cross-border e-commerce, and international tourism. Place names and addresses not only contain common general vocabulary but also a large number of proper nouns (such as landmarks, villages, road names, etc.). These proper nouns often span multiple language systems and have obvious differences in pronunciation and reading / writing rules. Simply relying on a global dictionary or traditional rule mapping often makes it difficult to accurately restore the pronunciation or geographical semantics of place names, resulting in ambiguity in information understanding and confusion in actual scenarios. To ensure the accurate transmission of place names and addresses in a multi-language environment, it is necessary to build an intelligent translation technology that can take into account semantics, phonology, and structure to achieve accurate and natural conversion of place names and addresses between different languages.

[0003] In the prior art, place name and address translation mainly relies on rule-based dictionary translation or end-to-end neural machine translation. The dictionary translation method can provide relatively accurate matches for general vocabulary, but its processing ability for proper nouns is limited, especially when faced with emerging landmarks or obscure place names, it is often helpless. Although end-to-end neural machine translation has a certain generalization ability, it is difficult to balance the pronunciation restoration and semantic consistency of place name proper nouns, and there are frequent phenomena of "literal translation" or "mistranslation" of place names, seriously affecting the recognizability and usability in actual applications, and it is difficult to meet the needs of accurate translation of actual cross-language place name and address information.

[0004] Therefore, there is a need to provide an optimized intelligent place name and address translation method based on multi-language syllable segmentation to solve the above technical problems. Summary of the Invention

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

[0006] According to one aspect of this application, there is provided an intelligent place name and address translation method based on multi-language syllable segmentation, which includes: Obtain the source language place name and address text and the target language specified by the user; Make a translation strategy decision for each lexical unit in the source language place name and address text to obtain a sequence of source language place name and address text words with translation strategy annotations, where the translation strategies include transliteration and dictionary translation; Based on the target language, call the general vocabulary translation dictionary of the target language to perform standard translation on the source language place name and address text words marked as dictionary translation in the sequence of source language place name and address text words containing translation strategy annotations, so as to obtain a sequence of general vocabulary translation results of place names and addresses; Based on the target language, perform syllable segmentation mapping on the source language place name and address text words marked as transliteration in the sequence of source language place name and address text words containing translation strategy annotations, so as to obtain a sequence of transliteration results of proper nouns of place names and addresses; Perform structured reconstruction on the sequence of general vocabulary translation results of place names and addresses and the sequence of transliteration results of proper nouns of place names and addresses to obtain the target language place name and address translation results.

[0007] Beneficial effects: Compared with the prior art, the intelligent place name and address translation method based on multi-language syllable segmentation provided by this application makes an adaptive translation strategy decision on each lexical unit in the source language place name and address text by introducing a deep learning algorithm, so as to automatically distinguish proper nouns that should use transliteration and general vocabulary that should use dictionary translation. Then, by performing fine-grained segmentation at the syllable level and target language syllable mapping on the proper nouns marked as transliteration, accurate transliteration of proper nouns can be achieved; at the same time, for general vocabulary marked as dictionary translation, a dictionary matching algorithm is used for accurate translation. This application standardizes and reorganizes the transliteration results of proper nouns and the translation results of general vocabulary to obtain the place name and address translation results, which can effectively improve the translation effect of proper place names when mixed with general words, and enhance the translation accuracy and flexibility of place names and addresses in a multi-language environment. 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. Together with the embodiments of the present application, they are used to explain 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 the intelligent place name and address translation method based on multi-language syllable segmentation according to the embodiment of the present application.

[0010] Figure 2 It is a schematic diagram of data flow of the intelligent place name and address translation method based on multi-language syllable segmentation according to the embodiment of the present application.

[0011] Figure 3 It is a flowchart of sub-step S2 of the intelligent place name and address translation method based on multi-language syllable segmentation according to the embodiment of the present application.

[0012] Figure 4 It is a flowchart of sub-step S4 of the intelligent geographical name and address translation method based on multi-language syllable segmentation according to an embodiment of the present application.

[0013] Figure 5 It is a flowchart of sub-step S42 of the intelligent geographical name and address translation method based on multi-language syllable segmentation according to an embodiment of the present application.

[0014] Figure 6 It is a flowchart of sub-step S422 of the intelligent geographical name and address translation method based on multi-language syllable segmentation according to an embodiment of the present application.

[0015] Figure 7 It is a flowchart of sub-step S4222 of the intelligent geographical name and address translation method based on multi-language syllable segmentation according to an embodiment of the present application. Detailed implementation manners

[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 "include" and "comprise" 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.

[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 sequence. On the contrary, according to the need, 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.

[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 the acquisition and processing of information or data in the present application are carried out on the premise of complying with the corresponding data protection regulations and policies and obtaining the authorization given by the corresponding authority manager.

[0021] Figure 1 It is a flowchart of an intelligent geographical name and address translation method based on multilingual syllable segmentation according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of an intelligent geographical name and address translation method based on multilingual syllable segmentation according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the intelligent geographical name and address translation method based on multilingual syllable segmentation includes the steps of: S1, obtaining a source language geographical name and address text and a target language specified by a user; S2, making a translation strategy decision on each lexical unit in the source language geographical name and address text to obtain a sequence of source language geographical name and address text words with translation strategy annotations, where the translation strategies include transliteration and dictionary translation; S3, based on the target language, calling a target language general vocabulary translation dictionary to perform standard translation on the source language geographical name and address text words marked as dictionary translation in the sequence of source language geographical name and address text words with translation strategy annotations to obtain a sequence of general vocabulary translation results of geographical names and addresses; S4, based on the target language, performing syllable segmentation mapping on the source language geographical name and address text words marked as transliteration in the sequence of source language geographical name and address text words with translation strategy annotations to obtain a sequence of transliteration results of proper nouns of geographical names and addresses; S5, performing structured reconstruction on the sequence of general vocabulary translation results of geographical names and addresses and the sequence of transliteration results of proper nouns of geographical names and addresses to obtain a target language geographical name and address translation result.

[0022] In the above intelligent geographical name and address translation method based on multilingual syllable segmentation, in step S1, a source language geographical name and address text and a target language specified by a user are obtained. It should be understood that in practical applications, the translation requirements of geographical names and addresses have obvious diversified characteristics. In scenarios such as cross-border e-commerce, international logistics, and map services, users need to perform targeted conversions on different source language geographical names and addresses and display them in the target language required by the target user or business scenario. Therefore, in order to ensure the pertinence and adaptability of the translation service, based on a customized input mechanism, the present application first collects the source language geographical name and address text and at the same time allows the user to specify the target language to be translated, providing the basic input for the subsequent translation process. In the specific implementation process, the required original geographical name and address information and target language instructions can be obtained through a user input interface, a one-stop interface, or a batch import method to achieve the standardized start of the translation process and the flexibility of multilingual coverage.

[0023] In the process of specific implementation, the collection of source language place name and address text and target language instructions can be achieved through various methods. For example, in a cross-border e-commerce platform, the system can embed a multi-language address input component, allowing users to fill in the receiving or shipping address in natural language form, and automatically identify the language system used by the address when the form is submitted, and extract it as the source language place name and address text. At the same time, when the user selects the delivery destination area, the system will automatically match the official language or common language of that area and record it as the target language instruction. This intelligent guidance mechanism based on user behavior not only improves the convenience of data collection but also enhances the ability to identify the language attributes of address information.

[0024] In the context of international logistics, address information often comes from customers, suppliers, or transportation nodes in different countries, and there are significant differences in its format and language style. To uniformly manage and efficiently process this heterogeneous data, the system can provide a batch import interface, allowing users to upload files containing multiple place names and addresses (such as Excel spreadsheets or CSV files), and perform language detection on each line of address content during the file parsing stage to automatically label the source language type to which it belongs. At the same time, the user can manually specify the target language into which the data to be imported this time needs to be translated on the upload interface, or automatically recommend target language options according to preset rules (such as based on the country of the recipient). This method is particularly suitable for large-scale address translation tasks, which can significantly improve data processing efficiency and reduce errors caused by manual intervention.

[0025] Map service applications are another typical application environment. In such systems, when users need to convert a specific place name and address into another language version during navigation, searching for a location, or sharing a location, the system can receive the user's oral address information through a speech recognition module, use a language recognition model to judge the language type of the text after converting the speech to text, and use it as the source language place name and address text. Subsequently, the user can switch the language display mode on the map interface, and the system determines the target language accordingly and includes the currently voice-entered address information in the translation queue. This real-time interactive address collection method not only improves the user experience but also provides immediate and accurate language input for subsequent translation processing.

[0026] In addition, in some highly customized business systems, the automated acquisition of source language place name and address text and target language instructions can also be achieved through API interface calls. For example, if the internal information management system of an enterprise needs to interface with external multilingual address translation services, it can automatically send requests to the translation server by configuring standard RESTful API interfaces each time address data is added or modified. The request body should include the original address text, its corresponding source language identifier, and the target language parameters set by the user or the system. This method is applicable to information systems with a high degree of integration and helps to achieve automated cross-platform and cross-system address translation.

[0027] At the technical implementation level, the acquisition of source language place name and address text depends on language recognition technology in natural language processing. Usually, language classification models based on deep learning, such as FastText and BERT, are used to determine the language type of the input text. After being trained with large-scale multilingual corpora, these models can complete the determination of the language attribution of any text within milliseconds, with high accuracy, especially suitable for complex scenarios with mixed languages or multilingual mixed inputs. The specification of the target language depends more on the design of the user interaction logic. Methods such as dropdown menu selection, automatic location matching, and default language setting can all be used as effective input channels.

[0028] It is worth noting that during the actual deployment process, the integrity and standardization of the address text also need to be considered. For example, some non-standard addresses may lack necessary administrative division level information, or have incorrect spellings and chaotic grammatical structures, which will increase the difficulty of language recognition. Therefore, when acquiring the source language place name and address text, an address standardization module can be combined to initially clean and structure the original text, such as removing redundant characters, filling in missing fields, and unifying unit expressions, to improve the stability of the subsequent translation process.

[0029] In the above intelligent geographical name and address translation method based on multilingual syllable segmentation, in step S2, translation strategy decisions are made for each lexical unit in the source language geographical name and address text to obtain a sequence of source language geographical name and address text words with translation strategy annotations, where the translation strategies include transliteration and dictionary translation. That is, considering that geographical names and addresses often contain both proper nouns that need to retain phonetic features (such as landmarks, road names, community names), and common nouns that must strictly reflect semantics and conform to local expression habits (such as "street", "road", "district", etc.), a single translation strategy cannot fully meet the requirements of information fidelity and usability. Therefore, in order to flexibly adopt different processing strategies at each lexical granularity, this application introduces a translation strategy adaptive decision-making mechanism based on a deep neural network model, which automatically identifies and annotates the translation strategy to be adopted for each lexical unit through in-depth semantic analysis and context awareness of the source language geographical name and address text. Among them, Figure 3 is a flowchart of sub-step S2 of the intelligent geographical name and address translation method based on multilingual syllable segmentation according to an embodiment of the present application. As Figure 3 shown, step S2 includes the steps of: S21, performing word segmentation on the source language geographical name and address text to obtain a sequence of source language geographical name and address text words; S22, performing semantic embedding encoding based on the mBERT model on each source language geographical name and address text word in the sequence of source language geographical name and address text words to obtain a sequence of source language geographical name and address word granularity semantic embedding encoding vectors; S23, inputting the sequence of source language geographical name and address word granularity semantic embedding encoding vectors into a translation strategy classifier based on the Transformer architecture to obtain the sequence of source language geographical name and address text words with translation strategy annotations.

[0030] Specifically, in step S21, word segmentation is performed on the source language place name and address text to obtain a sequence of source language place name and address text words. It should be understood that place name and address text usually consists of multiple lexical units, including both general vocabulary that requires dictionary translation and proper nouns or landmarks that require transliteration. To achieve a fine-grained segmentation of the place name and address text in subsequent translation strategy decisions, this application, based on a word segmentation algorithm, performs preprocessing of word segmentation on the input source language place name and address text to split the continuous string into semantically independent lexical units, forming a sequence of source language place name and address text words. In the specific implementation process, according to the structural characteristics of different languages, the system selects an appropriate word segmentation tool. For example, in the Chinese environment, a word segmentation method based on word frequency statistics and dictionary matching is used, while in English or other Latin-based languages, space and punctuation are used for segmentation. In this way, a sequence of source language place name and address text words can be established at the level of the smallest semantic unit of the address, providing a standardized data input format for the subsequent translation strategy annotation process, fundamentally eliminating the problems of fuzzy word boundaries and the mixing of proper words and general words, and greatly improving the fineness and intelligence level of the overall translation processing flow.

[0031] Specifically, in step S22, semantic embedding encoding based on the mBERT model is performed on each source language place name and address text word in the sequence of source language place name and address text words to obtain a sequence of source language place name and address word-level semantic embedding encoding vectors. Specifically, since each lexical unit in the place name and address text not only has surface spelling information but also has context semantic dependency relationships with other word units. If only classified and annotated based on the word list information, it is difficult to accurately identify the specific category attributes of the vocabulary, especially proper nouns such as landmarks and place names, which often can only be effectively distinguished in specific contexts. Therefore, in order to capture the deep semantic information of each source language place name and address text word in the context, this application introduces a multilingual BERT (mBERT) deep pre-trained semantic model to perform semantic embedding encoding on each source language place name and address text word to obtain the semantic embedding representation of each text word. It should be understood that the mBERT model has strong cross-language understanding ability, can achieve effective semantic encoding of multiple language texts, and has accumulated rich context semantic knowledge through the pre-training mechanism. In this application, based on the strong cross-language semantic representation ability of the mBERT model, each source language place name and address text word can be mapped to a high-dimensional semantic space to capture the deep semantic associations and context dependencies between the words, thereby obtaining a sequence of source language place name and address word-level semantic embedding encoding vectors, providing accurate and comprehensive semantic information support for the subsequent adaptive decision-making of translation strategies.

[0032] Specifically, in step S23, the sequence of granular semantic embedding encoding vectors of the source language place name and address words is input into a translation strategy classifier based on the Transformer architecture to obtain the sequence of source language place name and address text words with translation strategy annotations. It should be understood that when making translation strategy classification decisions for each source language place name and address text word, traditional fully connected networks often suffer from overfitting problems due to fixed structures and excessive parameters, and it is difficult to capture long-distance semantic dependencies in the sequence. In response, to achieve fine-grained context-sensitive classification, this application uses a Transformer model to construct a translation strategy classifier, leveraging the self-attention mechanism and positional encoding of the Transformer architecture to handle long-distance dependencies in the sequence, while maintaining the sensitivity of the model to each position in the input sequence and the ability for parallel computing, so as to achieve an in-depth context understanding of the semantic information of each source language place name and address text word. Finally, a binary label (transliteration / dictionary translation) for each source language place name and address text word is output through the Softmax layer, thereby adaptively determining the most suitable translation strategy for each text word and providing strong technical support for the multilingual translation service of place names and addresses.

[0033] In the above intelligent place name and address translation method based on multilingual syllable segmentation, in step S3, based on the target language, the target language general vocabulary translation dictionary is called to perform standard translation on the source language place name and address text words marked as dictionary translation in the sequence of source language place name and address text words with translation strategy annotations to obtain the sequence of general vocabulary translation results of place names and addresses. Specifically, since there are often clear and stable equivalent expressions for general vocabulary across different languages, directly using an authoritative target language dictionary for translation can ensure accurate semantic transmission. Therefore, to ensure the standardization, clarity, and universality for local audiences of common descriptive words in the address (such as directions, ordinals, types, etc.), this application, based on a pre-set multilingual dictionary, retrieves and matches each text word marked as dictionary translation in the source language place name and address text by calling the target language general vocabulary translation dictionary, directly looking up its corresponding target language expression, thereby generating the corresponding general vocabulary translation result. Through this standardization process, it is possible to effectively avoid mis-translation or ambiguity that may occur in existing neural translation models and achieve high-quality matching of general vocabulary.

[0034] In the above intelligent geographical name and address translation method based on multilingual syllable segmentation, in step S4, based on the target language, syllable segmentation mapping is performed on the source language geographical name and address text words marked as phonetic transcription in the sequence of source language geographical name and address text words containing translation strategy annotations, so as to obtain a sequence of phonetic transcription results of geographical name and address specific vocabulary. Specifically, due to the high regionality and phonetic characteristics of specific geographical names, the phonetic transcription method can best retain the pronunciation information in its original language, thereby realizing cross-language user recognition and scenario restoration. At the same time, considering that direct whole-word phonetic transcription may cause distorted pronunciation in the target language (for example, transliterating the Chinese "Xi'an" as "Xian" is easily misread as "xian"). Therefore, in order to ensure the reliability and fluency of the phonetic transcription process, the present application further performs fine-grained segmentation at the syllable level on each source language geographical name and address text word marked as phonetic transcription, disassembles it into the smallest pronunciation unit, and performs one-by-one mapping according to the syllable structure characteristics of the target language, so as to ensure that each syllable can find the closest pronunciation expression in the target language, thereby ensuring the natural fluency and easy understanding of the phonetic transcription result. Among them, Figure 4 is a flowchart of sub-step S4 of the intelligent geographical name and address translation method based on multilingual syllable segmentation according to an embodiment of the present application. As Figure 4 shown, step S4 includes steps: S41, using a G2P model to perform phoneme conversion on the source language geographical name and address text words marked as phonetic transcription, so as to obtain a sequence of phonemes of source language geographical name and address specific vocabulary; S42, performing syllable segmentation on the sequence of phonemes of source language geographical name and address specific vocabulary, so as to obtain a sequence of syllables of source language geographical name and address specific vocabulary; S43, performing syllable mapping based on the syllable library of the target language on each syllable of the sequence of syllables of source language geographical name and address specific vocabulary, so as to obtain a sequence of syllables of target language geographical name and address specific vocabulary; S44, using a P2G model to convert the sequence of syllables of target language geographical name and address specific vocabulary into text, so as to obtain the phonetic transcription result of the geographical name and address specific vocabulary.

[0035] Specifically, in step S41, a G2P model is used to perform phoneme conversion on the source language geographical name and address text words marked as phonetic transcription, so as to obtain a sequence of phonemes of source language geographical name and address specific vocabulary. It should be understood that the present application considers that the written forms (glyphs) of specific nouns in geographical names and addresses (such as street name " " or city name " (Seoul)") vary greatly in different languages. Direct character-based phonetic transcription is easily interfered by spelling rules (for example, the English "Leicester" is pronounced as / / rather than the surface pronunciation of the letter combination), and the phonetic systems of multiple languages are significantly different (for example, the Chinese pinyin "x" corresponds to the International Phonetic Alphabet / / , the Spanish "j" is pronounced as / x / . Therefore, in order to accurately capture the actual pronunciation characteristics of the source language place names, based on the principle of phonetic phoneme conversion, this application trains a multi-task G2P (Grapheme-to-Phoneme) model to convert spelling characters into International Phonetic Alphabet (IPA). In the specific implementation process, load the pre-trained Transformer architecture G2P model for the source language (such as using the Pinyin-to-IPA model based on the decomposition of initials and finals for Chinese, and using the mapping rule from Cyrillic letters to X-SAMPA symbols for Russian), and perform phoneme-level conversion on the source language place name address text words marked as transliteration (such as "Rivoli" in "Rue de Rivoli" in French), generating a standardized phoneme sequence of proper nouns (such as "Rivoli" → / / ), providing an accurate pronunciation benchmark for subsequent syllable segmentation mapping.

[0036] Specifically, in step S42, perform syllable segmentation on the phoneme sequence of the source language place name address proper nouns to obtain the syllable sequence of the source language place name address proper nouns. Specifically, since there are essential differences in syllable division rules for different languages (such as English allowing complex consonant clusters while Japanese strictly follows the CV structure), and mechanical equal division of phonemes easily leads to pronunciation distortion (such as mis-segmenting the Chinese "Xi'an" / / into / / ). Therefore, to achieve syllable boundary localization that conforms to speech perception, this application further analyzes the context dependence of each phoneme in the phoneme sequence of the source language place name address proper nouns to accurately identify the syllable boundary, thereby segmenting each syllable unit in the source language place name address proper nouns. Among them, FIG. 5 is a flowchart of sub-step S42 of the intelligent place name address translation method based on multi-language syllable segmentation according to an embodiment of this application. As shown in FIG. 5, step S42 includes the steps: S421, input the phoneme sequence of the source language place name address proper nouns into a Bi-LSTM model including an embedding layer to obtain a sequence of phoneme embedding correlation encoding vectors of the source language place name address proper nouns; S422, extract the phoneme embedding correlation encoding vector of the source language place name address proper nouns to be analyzed from the sequence of phoneme embedding correlation encoding vectors of the source language place name address proper nouns, and perform syllable boundary prediction on the phoneme embedding correlation encoding vector of the source language place name address proper nouns to be analyzed to obtain a syllable boundary annotation prediction result; S423, perform syllable segmentation on the phoneme sequence of the source language place name address proper nouns according to the syllable boundary prediction result to obtain the syllable sequence of the source language place name address proper nouns.

[0037] More specifically, in step S421, the phoneme sequence of the source language toponym address proper nouns is input into a Bi-LSTM model including an embedding layer to obtain a sequence of phoneme embedding correlation coding vectors of the source language toponym address proper nouns. It should be understood that since the original phoneme sequence obtained from the G2P model (for example, discrete phoneme symbols such as " / b / , / / , / r / , / k / , / l / , / i / ") itself lacks direct and rich context correlation information, and the syllable boundary division often highly depends on the combination environment before and after phonemes and the phonotactics of the language. Therefore, in order to fully understand the specific role and context dependence relationship of each phoneme in its sequence and thus accurately predict the syllable boundary, based on the principle of deep learning sequence modeling, this application first maps each discrete phoneme symbol into a low-dimensional dense real vector through an embedding layer (EmbeddingLayer), and preserves the phonetic similarity between phonemes in the vector space. Then, the bidirectional long short-term memory network (Bi-LSTM) is used to model the context dependence relationship between phonemes, capture the long-distance dependence features between phonemes, and obtain the deep-level correlation information between phonemes. Specifically, the Bi-LSTM layer uses two independent LSTM networks to traverse the phoneme sequence from the front and back respectively to independently capture the forward and backward dependence relationships of phonemes, and generates a context-aware representation of each phoneme by splicing the forward hidden state and the backward hidden state of each phoneme embedding feature, obtaining a sequence of phoneme embedding correlation coding vectors of the source language toponym address proper nouns. In this way, the complex interaction between phonemes can be understood more accurately, providing a strong basis for subsequent syllable boundary prediction.

[0038] More specifically, in step S422, the phoneme embedding correlation coding vector of the source language toponym address proper noun to be analyzed is extracted from the sequence of phoneme embedding correlation coding vectors of the source language toponym address proper nouns, and the syllable boundary prediction is performed on the phoneme embedding correlation coding vector of the source language toponym address proper noun to be analyzed to obtain the predicted result of the syllable boundary annotation. Specifically, since syllable segmentation needs to consider both the local phoneme combination legality (such as Japanese does not allow a consonant to follow "ん") and the global prosodic structure (such as the tendency of Chinese to be bisyllabic), and the feature extraction of a single scale is likely to lead to over-segmentation or under-segmentation (such as splitting the French word "aujourd'hui" / / The incorrect segmentation is into three syllables instead of the correct four syllables). In this regard, to achieve accurate syllable boundary prediction, the present application extracts the phoneme-embedded correlation coding vector of the source language place name and address proper noun to be analyzed as the analysis target from the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper noun, and measures its syllable boundary fitness at different granularities by mining its local neighborhood structure and global distribution characteristics in the phoneme embedding space, thereby realizing intelligent prediction of syllable boundaries. Among them, Figure 6 is a flowchart of sub-step S422 of the intelligent place name and address translation method based on multilingual syllable segmentation according to an embodiment of the present application. As Figure 6 shown, the step S422 includes the steps of: S4221, performing syllable boundary prediction based on global context association perception on the phoneme-embedded correlation coding vector of the source language place name and address proper noun to be analyzed to obtain the global fitness of the phoneme feature syllable boundary to be analyzed; S4222, performing syllable boundary prediction based on local neighborhood association perception on the phoneme-embedded correlation coding vector of the source language place name and address proper noun to be analyzed to obtain the local fitness of the phoneme feature syllable boundary to be analyzed; S4223, determining whether the phoneme corresponding to the phoneme-embedded correlation coding vector of the source language place name and address proper noun to be analyzed is a syllable boundary based on the global fitness of the phoneme feature syllable boundary to be analyzed and the local fitness of the phoneme feature syllable boundary to be analyzed.

[0039] In a specific example of the present application, the step S4221 includes: First, inputting the phoneme-embedded correlation coding vector of the source language place name and address proper noun to be analyzed and the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper noun into a global context fitness learning module based on a transformer structure to obtain a global context adaptation correlation coding vector of the phoneme feature syllable boundary to be analyzed, which is represented by the formula: ; ; Among them, represents the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper noun, , , and respectively represent the 1st, 2nd, th, and th phoneme-embedded correlation coding vectors of the source language place name and address proper noun in the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper noun, represents the phoneme-embedded correlation coding vector of the source language place name and address proper noun to be analyzed, represents the Transformer encoder, Represents the globally context-adapted association coding vector for the phoneme feature syllable boundary of the to-be-analyzed phoneme.

[0040] That is, through the Transformer architecture, the complex dependencies between the to-be-analyzed phonemes of the proper nouns in the source language place names and addresses and the overall sequence of phonemes of the proper nouns in the source language are captured, generating a globally context-aware globally context-adapted association coding vector for the phoneme feature syllable boundary of the to-be-analyzed phoneme. The globally context-adapted association coding vector for the phoneme feature syllable boundary of the to-be-analyzed phoneme not only contains the attribute information of the to-be-analyzed phoneme itself, but also encapsulates its structural role in the distribution of all phonemes of the proper noun, the interaction relationship with all other phonemes, and its position in the global phoneme ecosystem, providing an information basis for accurately dividing syllable boundaries and realizing cross-language syllable mapping subsequently.

[0041] Then, perform explicit decoding on the globally context-adapted association coding vector for the phoneme feature syllable boundary of the to-be-analyzed phoneme to obtain the global adaptation degree of the phoneme feature syllable boundary of the to-be-analyzed phoneme, which is expressed by the formula: ; Wherein, Represents the sigmoid activation function, Represents the transpose, Is a learnable weight parameter, Represents the bias term, Represents the global adaptation degree of the phoneme feature syllable boundary of the to-be-analyzed phoneme.

[0042] That is, through the explicit decoding process, the globally context-adapted association coding vector for the phoneme feature syllable boundary of the to-be-analyzed phoneme is projected from the high-dimensional space to the low-dimensional scalar value space, and the signal most relevant to the global adaptation of the syllable boundary is extracted and made explicit, so as to quantify the degree of fit between the to-be-analyzed phoneme feature and the macroscopic pattern or distribution defined by the set of all phonemes of the source language place names and addresses. Through this information compression and summary extraction, the global adaptation degree of the phoneme feature syllable boundary of the to-be-analyzed phoneme that directly reflects the rationality of syllable boundary division can be obtained, effectively guiding the subsequent syllable segmentation and mapping decisions, and improving the pronunciation restoration accuracy and cross-language speech mapping accuracy in the process of transliterating proper nouns.

[0043] Figure 7 Is a flowchart of sub-step S4222 of the intelligent place name and address translation method based on multi-language syllable segmentation according to an embodiment of the present application. As Figure 7As shown, the step S4222 includes steps: S42221, extracting local neighborhood context information of the phoneme-embedded correlation coding vector of the to-be-analyzed source language place name and address proper nouns in the sequence of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns to obtain a local neighborhood set of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns; S42222, inputting the phoneme-embedded correlation coding vector of the to-be-analyzed source language place name and address proper nouns and the local neighborhood set of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns into a local context adaptation degree learning module to obtain the local adaptation degree of the to-be-analyzed phoneme feature syllable boundary.

[0044] In a specific example of the present application, the step S42221 is expressed by the formula: ; wherein, represents the local neighborhood feature extraction radius, represents taking as the center at the position in the sequence of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns, and extracting the phoneme-embedded correlation coding vector of the source language place name and address proper nouns within the extraction radius , represents the local neighborhood set of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns, is the dimension of the vector in the local neighborhood set of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns, is the number of vectors in the local neighborhood set of the phoneme-embedded correlation coding vector of the source language place name and address proper nouns.

[0045] That is to say, it defines and isolates the microenvironment or small range most directly related to the phoneme of the to-be-analyzed source language place name and address proper nouns, provides a data basis for subsequent evaluation of its local adaptation degree with surrounding phonemes, and thus evaluates phoneme features in different local environments, enhancing the robustness of the evaluation. In this way, it can focus on the local context information around the to-be-analyzed phoneme, capture the short-distance dependence relationship and local speech pattern between phonemes, provide support for subsequent more accurate identification of syllable boundaries and realization of cross-language syllable mapping, and improve the accuracy and naturalness of the transliteration result.

[0046] In a specific example of the present application, the step S42222 is expressed by the formula: ; ; wherein, represents the a phoneme embedding correlation encoding vector for a source language geographical name and address proper noun, denote relative to the single phoneme association fitness of the syllable boundary, is a learnable weight parameter, 、 and are local neighborhood weight matrices, is an activation function, is and the feature scale of, denote the hyperbolic tangent function, denote the local fitness of the phoneme feature syllable boundary to be analyzed.

[0047] That is, the fitness degree of the phoneme feature and the syllable boundary is evaluated from the dimension of the local context. By quantifying and mining the single phoneme association fitness of the syllable boundary between the phoneme feature to be analyzed and each phoneme feature in each local neighborhood, and averaging and aggregating the context association fitness information of all local neighborhood phonemes, the speech feature association pattern of the phoneme in the local range can be captured, and the local fitness of the phoneme feature syllable boundary to be analyzed can be obtained, which helps to accurately identify the syllable boundary that conforms to the source language pronunciation rule, and effectively improves the accuracy of proper nouns in the syllable segmentation stage. At the same time, by combining the local context information, the subsequent syllable mapping process can better retain the pronunciation features of the source language, and finally enhance the recognizability and speech restoration degree of proper geographical name translation in a multilingual environment.

[0048] Specifically, in a preferred example of the present application, the step S42221 includes: taking the phoneme embedding correlation encoding vector of the source language geographical name and address proper noun to be analyzed as the center, and extracting a local neighborhood set of the phoneme embedding correlation encoding vector of the source language geographical name and address proper noun based on the initial local neighborhood feature extraction radius; based on the mixed state information entanglement intensity between the local neighborhood set of the phoneme embedding correlation encoding vector of the source language geographical name and address proper noun and the phoneme embedding correlation encoding vector of the source language geographical name and address proper noun to be analyzed, iteratively optimizing the initial local neighborhood feature extraction radius to obtain an optimized local neighborhood feature extraction radius; based on the optimized local neighborhood feature extraction radius, taking the phoneme embedding correlation encoding vector of the source language geographical name and address proper noun to be analyzed as the center, and extracting the local neighborhood set of the phoneme embedding correlation encoding vector of the source language geographical name and address proper noun.

[0049] When extracting the local neighborhood set corresponding to the phoneme embedding correlation coding vector of the proper nouns of the source language place names and addresses to be analyzed, although the local neighborhood set of the phoneme embedding correlation coding vector of the proper nouns of the source language place names and addresses defines the relevant microenvironment or small range, the phoneme embedding correlation coding vector of the proper nouns of the source language place names and addresses to be analyzed and its local neighborhood set still transition from the geometric eigenstate to the correlated coupling phase state. Therefore, when calculating the characteristic local fitness between them, measure correlation coupling is caused. Therefore, this application expects to iteratively optimize the initial local neighborhood feature extraction radius to improve the calculation accuracy of the local fitness of the phoneme feature syllable boundary to be analyzed.

[0050] Specifically, first, let , and use the adjacent manifold architecture based on the microscopic behavior pattern as the probability density reduction distribution, so as to obtain the mixed state functional index value of the pure state adaptation functional in the technical form of the global non-local entropy feature: ; Among them, represents the natural logarithm, represents relative to the pure state fitness, represents the global mixed state functional index value.

[0051] That is to say, when each is used as the pure state fitness distribution, based on the statistical mixture of each pure state fitness distribution, the global non-local uncertainty or mixing degree is formally measured in order to quantify the mixed state characterization under different degrees.

[0052] At the same time, when introducing the geometric regularization factor , calculate the manifold measure expansion value: ; Among them, represents the logarithm with base 2, represents the mixed state information entanglement intensity, represents the th geometric regularization factor.

[0053] That is to say, using the global non-negative eigenvalue decomposition of the pure state fitness distribution, based on the entropy correlation expansion paradigm, the mixed state characterization is expanded so that the information entropy is correlated with the entanglement degree. In this way, the initial local neighborhood feature extraction radius of the modulation neighborhood parameter can be used to make It is minimized, thereby reducing the measurement correlation coupling. Finally, based on the local neighborhood feature extraction radius after modulation optimization, the local neighborhood set of the source language place name address specific vocabulary phoneme embedded associated coding vector is re-extracted with the source language place name address specific vocabulary phoneme embedded associated coding vector as the center, and then the local neighborhood set of the source language place name address specific vocabulary phoneme embedded associated coding vector and the source language place name address specific vocabulary phoneme embedded associated coding vector are input into the local context adaptation learning module to calculate the local adaptation of the phoneme feature syllable boundary to be analyzed, thereby improving the calculation accuracy of the local adaptation of the phoneme feature syllable boundary to be analyzed.

[0054] In a specific example of the present application, the step S4223 includes: first, based on the global fitness of the phoneme feature syllable boundary to be analyzed and the local fitness of the phoneme feature syllable boundary to be analyzed, determining the comprehensive fitness of the phoneme feature syllable boundary to be analyzed, which is expressed as: ; ; in, and represent the weighted fusion weights of the global fitness of the syllable boundary of the phoneme feature to be analyzed and the local fitness of the syllable boundary of the phoneme feature to be analyzed, respectively. represents the normalized exponential function, represents the comprehensive adaptation weight parameter matrix, represents the comprehensive adaptation bias term, Indicates the comprehensive fitness of the syllable boundary of the phoneme feature to be analyzed.

[0055] That is, by integrating the global fitness of the syllable boundaries of the phoneme features to be analyzed and the local fitness of the syllable boundaries of the phoneme features to be analyzed, a more comprehensive and robust comprehensive evaluation standard is constructed to address the limitations of single-scale evaluation in the syllable segmentation of cross-language place names, addresses and proper nouns. Among them, the global fitness can capture the overall correlation pattern between phoneme features and syllable boundaries from a macro level, while the local fitness can explore the fine phonetic structure of phonemes in the local context. The combination of the two can give full play to the synergistic effect and accurately locate the syllable boundaries that conform to the pronunciation rules of the source language and take into account the phonetic habits of the target language. The generated comprehensive fitness of the syllable boundaries of the phoneme features to be analyzed can effectively overcome the problems of global analysis ignoring local phonetic details and local analysis lacking overall phonetic association, and significantly improve the accuracy and stability of syllable boundary recognition.

[0056] Then, based on the comparison between the comprehensive adaptation degree of the phoneme feature syllable boundary to be analyzed and a preset threshold, it is determined whether the phoneme corresponding to the phoneme embedding correlation coding vector of the proper noun of the source language place name address to be analyzed is a syllable boundary. That is, by comparing the comprehensive adaptation degree of the phoneme feature syllable boundary to be analyzed with the preset threshold, on the basis of comprehensively considering the global speech correlation and local pronunciation details, a clear decision basis is provided for whether each phoneme constitutes a syllable boundary. The binary result generated by this threshold decision mechanism can accurately filter out the redundant phoneme boundaries with weak speech feature correlation, retain the effective boundaries that conform to the pronunciation rules of multiple languages, and significantly improve the accuracy and efficiency of syllable segmentation of proper nouns. In the face of complex scenarios such as emerging landmarks or obscure place names, it can effectively reduce the transliteration deviation caused by syllable segmentation errors and enhance the semantic recognizability and information transmission accuracy in a cross-lingual environment.

[0057] More specifically, in step S423, according to the syllable boundary prediction result, syllable segmentation is performed on the phoneme sequence of the proper noun of the source language place name address to obtain the syllable sequence of the proper noun of the source language place name address. It should be understood that the syllable boundary prediction result reveals the end positions of each syllable unit in the phoneme sequence of the proper noun of the source language place name address. Based on this syllable boundary information, a segmentation operation is performed on the phoneme sequence of the proper noun of the source language place name address, so as to obtain a series of separated syllable units, that is, the syllable sequence of the proper noun of the source language place name address. For example, if the original phoneme sequence of the proper noun of the source language place name address is / / , and the predicted boundary annotation indicates that the end positions of the syllable units are / k / and / i / , then the segmentation operation will generate two syllables: the first syllable is / / , and the second syllable is / li / . In this way, the originally continuous phoneme stream is effectively and accurately converted into an ordered list or sequence composed of multiple syllables, such as [ / / , / li / ].

[0058] Specifically, in step S43, syllable mapping based on the syllable library of the target language is performed on each syllable of the syllable sequence of the proper noun of the source language place name address to obtain the syllable sequence of the proper noun of the target language place name address. It should be understood that since transliteration needs to take into account the pronunciation approximation and the orthographic norms of the target language (for example, Japanese katakana cannot directly represent French nasal vowels), and simple phoneme literal translation is likely to generate illegal syllables (such as mapping the English "Smith" / / to Korean needs to avoid the appearance of " ” and other invalid consonant clusters). Therefore, to achieve a balance between pronunciation fidelity and acceptability in the target language, this application is based on the principle of optimal syllable alignment. By constructing a syllable library for the target language and using a multi-level syllable mapping method to find the best mapping path. Specifically, first, load the syllable list of the target language (such as the Japanese JIS X4063 katakana syllable table, the Arabic ISO 233 transliteration rules), and then perform multi-candidate matching for each source syllable: for example, the Chinese syllable "zhang" / / When mapped to Russian, calculate its phoneme edit distance with candidates such as " " ( / / ), " " ( / / ), etc., and combine the historical translation frequency (such as the appearance rate of "Zhang→ " in the Russian official place name library reaches 93%). Select the optimal mapping by comprehensively considering the phoneme edit distance and historical translation frequency, and finally generate the syllable sequence of the target language place name and address proper nouns.

[0059] Specifically, in step S44, use the P2G model to convert the syllable sequence of the target language place name and address proper nouns into text to obtain the transliteration result of the place name and address proper nouns. That is, in order to convert the syllable sequence of the target language place name and address proper nouns into a text form that conforms to the writing norms of the target language, this application is based on the principle of Phoneme-to-Grapheme (P2G). By using the P2G model trained for the target language, the syllable sequence of the target language is converted into the final text output. Specifically, the P2G model is also a sequence-to-sequence model based on deep learning. It can select the most appropriate combination of characters or letters according to the syllable sequence, context information, and common vocabulary and writing habits of the target language, not only ensuring the accurate pronunciation of the transliteration result, but also being able to generate a text form that conforms to the writing norms of the target language, is easy to read and understand, and completes the complete conversion from the source language proper name to the high-quality transliterated text in the target language, so that the final place name and address translation result can not only accurately convey geographical information, but also meet the reading and usage habits of the target language users.

[0060] In the above intelligent geographical name and address translation method based on multi - language syllable segmentation, in step S5, a structural reconstruction is performed on the sequence of the translation results of the general geographical name and address vocabulary and the sequence of the transliteration results of the proper geographical name and address vocabulary to obtain the translation result of the target - language geographical name and address. It should be understood that this application takes into account the relatively high structural requirements of geographical name and address texts. Although individual words have been correctly translated or transliterated, if they are not recombined according to grammar or local habits, problems such as disordered word order and unclear expression are likely to occur. Therefore, in order to finally output a normalized and standardized target - language geographical name and address while comprehensively ensuring accurate meaning expression and readability, this application is based on a structural reconstruction engine to automatically splice and format the output of general translation words and transliterated proper words according to the word order, punctuation, and writing norms of the target language. In the specific implementation process, first, a target - language template library is constructed (including English "[Block] [Type], [City] [Suffix]", Japanese "[Prefecture] [City / District / Town / Village] [Block Number]", etc.). The transliteration results and dictionary translation results are sorted according to the target - language norms (English, and necessary prepositions and format symbols are inserted at the same time (such as English commas, black dots in Japanese). In this way, the translation results not only retain the source semantic information but also conform to the cognitive habits of target users.

[0061] In summary, the intelligent geographical name and address translation method based on multi - language syllable segmentation according to the embodiments of this application is elucidated. It makes an adaptive translation strategy decision for each lexical unit in the source - language geographical name and address text by introducing a deep - learning algorithm to automatically distinguish proper nouns that should be transliterated and general vocabulary that should be translated using a dictionary. Then, through fine - grained syllable - level segmentation and target - language syllable mapping of the proper nouns marked for transliteration, accurate transliteration of proper nouns is achieved; at the same time, for general vocabulary marked for dictionary translation, a dictionary - matching algorithm is used for accurate translation. Furthermore, through the normalized recombination of the transliteration results of proper nouns and the translation results of general vocabulary, the translation result of the geographical name and address is obtained. This method can effectively improve the translation effect of proper geographical names when mixed with general words and enhance the translation accuracy and flexibility of geographical names and addresses in a multi - language environment.

[0062] The basic principles of the present invention have been described above in combination 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. 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 - mentioned 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 - mentioned specific details for implementation.

[0063] In the above embodiments, the descriptions of the respective embodiments have their own focuses. 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 merely 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.

[0064] 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 without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. 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 encompassed by the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0065] In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0066] 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 and not 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. An intelligent geographical name and address translation method based on multi - language syllable segmentation, characterized in that, Including: Obtain the source language place name and address text and the target language specified by the user; Make translation strategy decisions for each lexical unit in the source language place name and address text to obtain a sequence of source language place name and address text words with translation strategy annotations, where the translation strategies include transliteration and dictionary translation; Based on the target language, call the target language general vocabulary translation dictionary to perform standard translation on the source language place name and address text words marked as dictionary translation in the sequence of source language place name and address text words with translation strategy annotations to obtain a sequence of general vocabulary translation results of place names and addresses; Based on the target language, perform syllable segmentation mapping on the source language place name and address text words marked as transliteration in the sequence of source language place name and address text words with translation strategy annotations to obtain a sequence of transliteration results of place name and address proper nouns; Perform structured reconstruction on the sequence of general vocabulary translation results of place names and addresses and the sequence of transliteration results of place name and address proper nouns to obtain the target language place name and address translation result.

2. The intelligent geographical name and address translation method based on multilingual syllable segmentation according to claim 1, characterized in that Making translation strategy decisions for each lexical unit in the source language place name and address text to obtain a sequence of source language place name and address text words with translation strategy annotations, where the translation strategies include transliteration and dictionary translation, includes: Perform word segmentation on the source language place name and address text to obtain a sequence of source language place name and address text words; Perform semantic embedding encoding based on the mBERT model on each source language place name and address text word in the sequence of source language place name and address text words to obtain a sequence of source language place name and address word-level semantic embedding encoding vectors; Input the sequence of source language place name and address word-level semantic embedding encoding vectors into a translation strategy classifier based on the Transformer architecture to obtain the sequence of source language place name and address text words with translation strategy annotations.

3. The intelligent geographical name and address translation method based on multi - language syllable segmentation according to claim 1, characterized in that, Performing syllable segmentation mapping on the source language place name and address text words marked as transliteration in the sequence of source language place name and address text words with translation strategy annotations to obtain a sequence of transliteration results of place name and address proper nouns, includes: Use the G2P model to perform phoneme conversion on the source language place name and address text words marked as transliteration to obtain a sequence of source language place name and address proper noun phonemes; Perform syllable segmentation on the sequence of source language place name and address proper noun phonemes to obtain a sequence of source language place name and address proper noun syllables; Perform syllable mapping on each source language place name and address proper noun syllable in the sequence of source language place name and address proper noun syllables based on the target language syllable library to obtain a sequence of target language place name and address proper noun syllables; Use the P2G model to convert the sequence of target language place name and address proper noun syllables into text to obtain the transliteration result of the place name and address proper nouns.

4. The intelligent geographical name and address translation method based on multi - language syllable segmentation according to claim 3, characterized in that, Performing syllable segmentation on the sequence of source language place name and address proper noun phonemes to obtain a sequence of source language place name and address proper noun syllables, includes: Input the sequence of source language place name and address proper noun phonemes into a Bi-LSTM model including an embedding layer to obtain a sequence of source language place name and address proper noun phoneme embedding correlation encoding vectors; Extract the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed from the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper nouns, and perform syllable boundary prediction on the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed to obtain the predicted result of syllable boundary annotation; Perform syllable segmentation on the phoneme sequence of the source language place name and address proper nouns according to the syllable boundary prediction result to obtain the syllable sequence of the source language place name and address proper nouns.

5. The intelligent geographical name and address translation method based on multi-language syllable segmentation according to claim 4, wherein Perform syllable boundary prediction on the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed to obtain the predicted result of syllable boundary annotation, including: Perform syllable boundary prediction based on global context correlation perception on the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed to obtain the global fitness of the syllable boundary of the phoneme features to be analyzed; Perform syllable boundary prediction based on local neighborhood correlation perception on the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed to obtain the local fitness of the syllable boundary of the phoneme features to be analyzed; Based on the global fitness of the syllable boundary of the phoneme features to be analyzed and the local fitness of the syllable boundary of the phoneme features to be analyzed, determine whether the phoneme corresponding to the phoneme-embedded correlation coding vector of the source language place name and address proper nouns to be analyzed is a syllable boundary.

6. The intelligent place name and address translation method based on multilingual syllable segmentation according to claim 5, characterized in that Perform syllable boundary prediction based on global context correlation perception on the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed to obtain the global fitness of the syllable boundary of the phoneme features to be analyzed, including: Input the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed and the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper nouns into the global context fitness learning module based on the Transformer structure to obtain the global context adaptation correlation coding vector of the syllable boundary of the phoneme features to be analyzed; Perform explicit decoding on the global context adaptation correlation coding vector of the syllable boundary of the phoneme features to be analyzed to obtain the global fitness of the syllable boundary of the phoneme features to be analyzed.

7. The intelligent geographical name and address translation method based on multi-language syllable segmentation according to claim 6, characterized in that Perform syllable boundary prediction based on local neighborhood correlation perception on the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed to obtain the local fitness of the syllable boundary of the phoneme features to be analyzed, including: Extract the local neighborhood context information of the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed in the sequence of phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to obtain the local neighborhood set of the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns; Input the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns to be analyzed and the local neighborhood set of the phoneme-embedded correlation coding vectors of the source language place name and address proper nouns into the local context fitness learning module to obtain the local fitness of the syllable boundary of the phoneme features to be analyzed.

8. The intelligent geographical name and address translation method based on multilingual syllable segmentation according to claim 7, wherein Extract the local neighborhood context information in the sequence of the source language toponym address specific vocabulary phoneme embedding correlation coding vectors, so as to obtain the local neighborhood set of the source language toponym address specific vocabulary phoneme embedding correlation coding vectors, including: Taking the source language toponym address specific vocabulary phoneme embedding correlation coding vector to be analyzed as the center, and based on the initial local neighborhood feature extraction radius, extract the local neighborhood set of the initial source language toponym address specific vocabulary phoneme embedding correlation coding vector; Based on the mixed state information entanglement intensity between the local neighborhood set of the initial source language toponym address specific vocabulary phoneme embedding correlation coding vector and the source language toponym address specific vocabulary phoneme embedding correlation coding vector to be analyzed, iteratively optimize the initial local neighborhood feature extraction radius to obtain the optimized local neighborhood feature extraction radius; Based on the optimized local neighborhood feature extraction radius, taking the source language toponym address specific vocabulary phoneme embedding correlation coding vector to be analyzed as the center, extract the local neighborhood set of the source language toponym address specific vocabulary phoneme embedding correlation coding vector.

9. The intelligent place name and address translation method based on multilingual syllable segmentation according to claim 7, characterized in that Based on the global adaptability of the phoneme feature syllable boundary to be analyzed and the local adaptability of the phoneme feature syllable boundary to be analyzed, determine whether the phoneme corresponding to the source language toponym address specific vocabulary phoneme embedding correlation coding vector to be analyzed is a syllable boundary, including: Based on the global adaptability of the phoneme feature syllable boundary to be analyzed and the local adaptability of the phoneme feature syllable boundary to be analyzed, determine the comprehensive adaptability of the phoneme feature syllable boundary to be analyzed; Based on the comparison between the comprehensive adaptability of the phoneme feature syllable boundary to be analyzed and a preset threshold, determine whether the phoneme corresponding to the source language toponym address specific vocabulary phoneme embedding correlation coding vector to be analyzed is a syllable boundary.

Images