Multi-language recognition inter-translation system based on offline double models

Abstract

The invention discloses a multi-language recognition inter-translation system based on offline double models, and belongs to the technical field of artificial intelligence. The objective of the invention is to solve the problem of poor accuracy of an existing translation system during special industry translation. The translator is adopted to translate original text information to be translated into a required language form, then the artificial intelligence adjustment module is adopted to perform word order and coherence adjustment on the translated text information, and the adjusted information is translated into original text language text information through the translation result and original text proofreading module. Comparing the translated original text language characters with the to-be-translated original text, judging whether there is a difference, extracting a difference fragment, translating again, replacing the translated difference fragment with a corresponding fragment of the translated original text language character information, then comparing the translated difference fragment with the to-be-translated original text again, and if there is still a difference, marking and outputting. The multi-language translation method is suitable for multi-language translation.

Description

Technical Field

[0001] The invention belongs to the technical field of artificial intelligence. Background Art

[0002] Artificial intelligence translation technology has made rapid progress in recent years, but it still exposes many shortcomings in special application scenarios in special industries. First, some industries involve a large number of professional terms, industry-specific terms, and obscure code words. Existing artificial intelligence translation technology often cannot accurately translate these professional terms and specific context expressions. For example, in cases involving cybercrime, ordinary artificial intelligence translation systems may not be able to give accurate translation results for professional terms such as "names of hacker attack methods" and "cryptocurrency crime terms", which affects the public security personnel's accurate understanding and analysis of case information. Secondly, the multilingual environment of the public security industry is extremely complex and dynamically changing. In transnational law enforcement operations, various niche languages, dialects, and mixed languages ​​may be encountered. Most of the existing artificial intelligence translation systems are trained and optimized for common mainstream languages, and have limited recognition and translation capabilities for these niche languages ​​and dialects. For example, in law enforcement activities in some border areas, local residents may use specific minority languages ​​or border dialects to communicate, and existing translation technologies may not be able to effectively recognize and translate these languages, resulting in obstructed information transmission and law enforcement work being in trouble. Furthermore, public security work has extremely high requirements for the accuracy, timeliness and confidentiality of information. Existing AI translation technology is prone to translation errors or semantic deviations in a more complex language environment, which may mislead the decision-making of public security personnel, delay the timing of case detection, and may even lead to serious safety accidents. At the same time, in terms of information confidentiality, some existing AI translation services cannot meet the public security industry's needs for strict confidentiality of sensitive case information. For example, in major cases involving national security, if the data is leaked or tampered with during the translation process, it will pose a serious threat to the security of the country and the people. Summary of the invention

[0003] The present invention aims to solve the problem of poor accuracy of existing translation systems when performing translation for special industries, and now provides a multi-language recognition and translation system based on an offline dual model.

[0004] The multi-language recognition and translation system based on offline dual models of the present invention comprises: an information input device, an input type judgment module, an audio signal processing module, a translator, an artificial intelligence adjustment module and a translation result and original text proofreading module;

[0005] The information input device obtains the information to be translated by voice input or text input, and transmits the input information to the input type determination module;

[0006] The input type judgment module is used to judge the type of the received information. When the received information is audio information, the received audio is sent to the audio signal processing module; if the received information is text information, the text information is transmitted to the translator and the translation result and original text proofreading module;

[0007] The audio signal processing module converts the received audio information into text of the corresponding language, and sends the converted original text information to the translator and the translation result and original text proofreading module at the same time;

[0008] The translator translates the received original text information into the required language and transmits the translation result to the artificial intelligence adjustment module;

[0009] The artificial intelligence adjustment module adjusts the word order and coherence of the received information and transmits the adjusted information to the translation result and original text proofreading module;

[0010] The translation result and original text proofreading module translates the adjusted information back to the original language text information, compares the original language text information translated back again with the received original text information, extracts the segments with differences, and sends the extracted difference segments to the translator, and the translator translates the difference segments and sends the translated segments to the artificial intelligence adjustment module;

[0011] The artificial intelligence adjustment module adjusts the word order and coherence of the translated segments, and sends the adjusted segments to the translation result and original text proofreading module;

[0012] The translation result and original text proofreading module replaces the corresponding segment in the original language text information with the adjusted segment, and the replaced text information is compared with the original text information again. If there is no difference, it is directly transmitted to the output end. If there is a difference, the difference segment is marked and transmitted to the output end.

[0013] Furthermore, in the present invention, the output end is a voice output device or a text display device.

[0014] Furthermore, the present invention also includes a database, wherein the database stores professional terms, commonly used communication terms and code words of different industries, and the translator uses the data in the database for translation.

[0015] The present invention adopts an artificial intelligence translation module combined with network terms and common professional terms, which effectively ensures the accuracy of language translation, avoids the problem of inaccurate translation due to some industry terms or code words, and effectively improves the accuracy of translation in some special industries. When multiple calibrations still cannot accurately translate, the inaccurate fragments are marked, which effectively avoids misunderstandings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a block diagram of the system principles of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0018] Specific implementation method 1: refer to Figure 1 Specifically describing this embodiment, the multi-language recognition and translation system based on offline dual models described in this embodiment includes: an information input device 1, an input type judgment module 2, an audio signal processing module 3, a translator 4, an artificial intelligence adjustment module 5 and a translation result and original text proofreading module 6;

[0019] The information input device 1 obtains the information to be translated by voice input or text input, and transmits the input information to the input type determination module 2;

[0020] The input type judgment module 2 is used to judge the type of the received information. When the received information is audio information, the received audio is sent to the audio signal processing module 3; if the received information is text information, the text information is transmitted to the translator 4 and the translation result and original text proofreading module 6;

[0021] The audio signal processing module 3 converts the received audio information into text of the corresponding language, and sends the converted original text information to the translator 4 and the translation result and original text proofreading module 6 at the same time;

[0022] The translator 4 translates the received original text information into the required language and transmits the translation result to the artificial intelligence adjustment module 5;

[0023] The artificial intelligence adjustment module 5 adjusts the word order and coherence of the received information, and transmits the adjusted information to the translation result and original text proofreading module 6;

[0024] The translation result and original text proofreading module 6 translates the adjusted information back to the original language text information, compares the translated original language text information with the received original text information, extracts the segments with differences, and sends the extracted difference segments to the translator 4, which translates the difference segments and sends the translated segments to the artificial intelligence adjustment module 5;

[0025] The artificial intelligence adjustment module 5 adjusts the word order and coherence of the translated segment, and sends the adjusted segment to the translation result and original text proofreading module 6;

[0026] The translation result and original text proofreading module 6 replaces the corresponding segment in the original language text information with the adjusted segment, and the replaced text information is compared with the original text information again. If there is no difference, it is directly transmitted to the output end. If there is a difference, the difference segment is marked and transmitted to the output end.

[0027] Furthermore, in the present invention, the output end is a voice output device or a text display device.

[0028] Furthermore, the present invention also includes a database, in which professional terms, commonly used communication terms and code words of different industries are stored, and the translator 4 uses the data in the database for translation.

[0029] When the present invention is applied to the international communication and law enforcement environment of the public security industry, the system also integrates a large model deployed on a single machine as an interactive interface in the form of a web page, and uses API technology to enable the translation results to be displayed in real time on relevant business pages, such as the foreign-related case record page. When the police enter a foreign language text or transfer a file, the translation is automatically filled in the corresponding position, and the format is unified with the original text record, which is convenient for archiving and subsequent reference.

[0030] Small models within 10B can be deployed on a single machine, and large models above 32B rely on the private cloud environment of the public security intranet to build a core model computing service cluster, which undertakes the heavy task of multi-language deep learning model training and large-scale data storage. It stores massive multi-language corpora, including professional data such as various global criminal terms and commonly used phrases for cross-border communication, which are regularly updated and synchronized to the local cache of the police communication system to ensure that language knowledge keeps pace with the times. The trained large model can be quantified and deployed on a single machine.

[0031] When a complex translation request is initiated, such as the translation of a long professional legal text or a mixture of multiple languages, the data is quickly uploaded to the public security network intranet with the help of a dedicated network and a network gateway device, and the large computing power of the computing server is used for processing. After the processing is completed, the results are quickly transmitted back to the client, reducing the computing pressure of the local device. The overall response time is controlled within 3 seconds to meet the real-time law enforcement needs. Adopt local cache optimization and data encryption transmission: The data interaction between the client and the server uses the SSL / TLS encryption protocol throughout the process to ensure that sensitive information such as audio, text and translation results is not stolen or tampered with during the transmission process. For example, in the cross-border pursuit information sharing scenario involving the transmission of multi-language clues of suspects, high-intensity encryption ensures that the data safely crosses the public network to prevent foreign forces from intercepting and cracking. The key is updated regularly and complies with the encryption standards of the public security industry. Access rights are set: the access rights of police communication equipment and personnel are strictly limited based on the identity authentication system. Only legitimate users who have passed the device fingerprint recognition, police identity password and dynamic token multi-factor authentication can enable the multi-language system function. Different police types are assigned different language resource access levels according to their duties. For example, immigration management police have broader foreign language permissions, and criminal investigation police focus on languages ​​related to cross-border crimes to prevent unauthorized operations and ensure that the system and data are safe and controllable. Security audit mechanism: Establish a comprehensive security audit system to record the operation log of the multi-language system of Police Communication in real time, covering key information such as login time, translation request content, data transmission source and destination. Regular audit analysis to promptly discover abnormal access and potential security vulnerabilities. If an unknown IP address is detected to frequently attempt to crack the login or an abnormal large amount of data is downloaded, an alarm will be triggered immediately and the source will be traced and blocked to maintain the safe and stable operation of the system and protect the security line of public security business data.

[0032] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that the features described in conjunction with a single embodiment may be used in other described embodiments.

Claims

1. A multi-language recognition and translation system based on offline dual models, characterized by: include: An information input device (1), an input type determination module (2), an audio signal processing module (3), a translator (4), an artificial intelligence adjustment module (5) and a translation result and original text proofreading module (6); The information input device (1) obtains information to be translated by voice input or text input, and transmits the input information to the input type determination module (2); The input type judgment module (2) is used to judge the type of the received information. When the received information is audio information, the received audio is sent to the audio signal processing module (3); if the received information is text information, the text information is transmitted to the translator (4) and the translation result and original text proofreading module (6); The audio signal processing module (3) converts the received audio information into text of the corresponding language, and sends the converted original text information to the translator (4) and the translation result and original text proofreading module (6) at the same time; The translator (4) translates the received original text information into the required language and transmits the translation result to the artificial intelligence adjustment module (5); The artificial intelligence adjustment module (5) adjusts the word order and coherence of the received information, and transmits the adjusted information to the translation result and original text proofreading module (6); The translation result and original text proofreading module (6) translates the adjusted information back to the original language text information, compares the translated original language text information with the received original text information, extracts the segments with differences, and sends the extracted difference segments to the translator (4). The translator (4) translates the difference segments and sends the translated segments to the artificial intelligence adjustment module (5); The artificial intelligence adjustment module (5) adjusts the word order and coherence of the translated segment, and sends the adjusted segment to the translation result and original text proofreading module (6); The translation result and original text proofreading module (6) replaces the corresponding segment in the translated original language text information with the adjusted segment, and the replaced text information is compared with the original text information again. If there is no difference, it is directly transmitted to the output end. If there is a difference, the difference segment is marked and then transmitted to the output end.

2. The multi-language recognition and translation system based on offline dual models according to claim 1 is characterized in that: The output end is a voice output device or a text display device.

3. The multi-language recognition and translation system based on offline dual models according to claim 1 is characterized in that: It also includes a database, in which professional terms, commonly used communication terms and code words of different industries are stored, and the translator (4) uses the data in the database for translation.

Images