CN114444523A - Portable off-line machine translation intelligent box - Google Patents
Portable off-line machine translation intelligent box Download PDFInfo
- Publication number
- CN114444523A CN114444523A CN202210124622.XA CN202210124622A CN114444523A CN 114444523 A CN114444523 A CN 114444523A CN 202210124622 A CN202210124622 A CN 202210124622A CN 114444523 A CN114444523 A CN 114444523A
- Authority
- CN
- China
- Prior art keywords
- machine translation
- model
- user
- translation
- development board
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/40—Processing or translation of natural language
- G06F40/58—Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/50—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems
- G06F21/52—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems during program execution, e.g. stack integrity ; Preventing unwanted data erasure; Buffer overflow
- G06F21/53—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems during program execution, e.g. stack integrity ; Preventing unwanted data erasure; Buffer overflow by executing in a restricted environment, e.g. sandbox or secure virtual machine
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Stored Programmes (AREA)
Abstract
The invention relates to the technical field of machine translation, in particular to a portable off-line machine translation intelligent box, wherein a hardware part comprises a basic carrier and an encapsulation improvement, a software part comprises a system frame and a model service, the model service comprises model distillation, marking generalization, user customization, marking replacement and model fine adjustment, and an application part comprises internal operation and external display. The system is more user-aware and better serves users.
Description
Technical Field
The invention relates to the technical field of machine translation, in particular to a portable off-line machine translation intelligent box.
Background
The machine translation technology is a technology for converting one language into another language, and occupies a very important professional position in the field of natural language processing and even in the field of artificial intelligence deep learning.
In the application scenario of the existing machine translation technology, one direction is to provide online machine translation services to users through web page terminals or application program related services by taking a computer host as a carrier, and the other direction is to provide offline or online machine translation services through mobile phone application programs by taking a mobile phone as a carrier.
However, in the above prior art, if a user is not covered by a network signal, or the user's usage scenario is confidential, and cannot access an external network, the online machine translation service cannot be used, and further, in the process of accessing a remote server, data transmission and interaction are inevitably generated, so that security cannot be guaranteed for private information of the user, where not only translation original texts and translations input by the user but also other sensitive information including the user IP in the network transmission process may be backed up in the remote server, which may cause property economic loss of the user.
Therefore, it is necessary to design a portable offline machine translation intelligent box, which can provide machine translation system services for users without online, and all operations and data are retained in the local device, and encrypted by a password, so that the security and privacy of user data are ensured to the maximum extent, and the portable offline machine translation intelligent box is not limited by a network environment and provides personalized customized operations for users.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a portable off-line machine translation intelligent box, can provide machine translation system service for a user without on-line, retains all operations and data in local equipment, encrypts through a password, ensures the security and privacy of user data to the greatest extent, is not limited by a network environment, and provides personalized customized operation for the user.
In order to achieve the above object, the present invention provides a portable off-line machine translation intelligent box, which comprises a hardware part, a software part and an application part;
hardware technology:
the device comprises a basic carrier and packaging improvement, wherein the basic carrier is a development board, the packaging improvement is to use plastic acrylic as an integral shell, spaces are reserved in a radiating fin, a power supply port and a network cable port of the development board to be connected, the radiating fin is a metal radiating fin and is provided with a fan, and a USB interface and an HDMI interface of the development board are removed;
[ software part ]:
the method comprises a system framework and a model service, wherein the system framework comprises the following steps:
s1: the system adopts a development board-based self-contained system or Windows, Mac OS;
s2: manually expanding the system exchange space of the development board, and installing swapfile;
s3: then adding a 6GB virtual exchange space for the development board;
s4: deleting irrelevant application programs, reserving a WEB visual interface, and setting a user name and a password which are only known by a developer for an operating system;
s5: basic language environment and configuration, including installation of basic programming language and deep learning framework;
the model service is a deep neural network model using a Transformer and is improved and optimized according to an application scene, and the method comprises the following steps:
s10: model distillation: compressing the machine translation model with large-scale parameter magnitude into a machine translation model with small-scale parameter magnitude, deploying and operating on an edge calculation box, and performing parameter learning layer by layer to enable each layer of parameters of the small model to learn several layers of parameters of the large model in a stepped manner;
s11: marking and generalization: discrete words are rewritten in a special mark mode, and input models are learned to achieve the purpose of data generalization, so that the text idea can be well reserved when the models meet the data input of the type;
s12: customizing by a user: when a professional dictionary or a contrast dictionary of special translation requirements is mined, customizing a machine translation model;
s13: the special words required by the user are marked and replaced by marking generalization processing and pointer network technical means in the S11;
s14: obtaining a certain number of parallel sentence pairs in the current field from an original training corpus and other parallel corpus resource platforms through a dictionary, and retraining a machine translation model in the trained general field on the part of professional corpus data to achieve model fine adjustment, so that the finally obtained machine translation model can be translated in a special field required by a user to be more accurately adapted;
[ application part ]:
including internal operation and external exposure, the steps of internal operation include:
s20: starting a machine translation system through local Web Service for data input and output;
s21: after receiving user input, performing word segmentation, sub-word, case and case conversion and marking preprocessing operation;
s22: transmitting the translation to a machine translation model for model reasoning and translation to obtain an original translation, marking and restoring, and combining words and then processing to obtain a target language translation required by a user;
the internal operation is developed and constructed based on a basic programming language, on the basis of realizing a machine translation model by a deep learning framework, the basic programming language is utilized to carry out data preprocessing and post-processing, and then data interaction is carried out with a Web Service interface, so that the one-time operation flow of a machine translation model system can be completed; the service is provided in the form of a webpage facing the user;
the external presentation includes: after the translation equipment is plugged in a power supply to be started, the machine translation system can be automatically started to operate, after the program operation is completed, a user can be connected with a host personal computer through a network cable, a network port is shared by the host, a corresponding local area network segment IP can be created, the user can use the machine translation system on the personal computer by accessing a network address corresponding to the IP, and after the use is completed, the network cable is pulled out to close the power supply.
The basic vector development board is JETSON NANO, Haisi, and the figure is a basic board.
The deep learning framework of the software part is MXNET, Tensorflow, Pytrch code framework.
The basic language environment is Python3, C/C + +/C #, Java, Php, JS.
The invention has the following beneficial effects:
the model algorithm is integrated in the offline equipment, so that the service can be provided under the scene without network coverage, and no matter the signal coverage or the intranet safety limit causes no influence on the machine translation service;
the invention provides service in an off-line mode, data interaction with other equipment such as a remote server and the like does not exist, all data information generated in the using process is stored in local equipment, and local protection is performed in a password mode, so that the safety and confidentiality of user information are ensured to the maximum extent;
the invention can provide service for users after the product packaging is finished, and keeps related expansion interfaces on the basis, can conveniently and quickly develop and design auxiliary functions on the existing functions, and has no problem that hardware equipment and a software algorithm are strongly bound to cause decoupling and upgrading;
the machine translation service provided by the invention can provide customized technical realization aiming at actual use scenes and professional contents of a user on the premise of meeting daily requirements of the public, so that smooth and accurate translation results are provided for language contents actually used by the user while the common translation effect is reasonable;
finally, the invention can also solve the problem that the enterprise unit orders the large-scale machine translation server equipment in the past, the large-scale machine translation server equipment in the past has large floor area and high maintenance cost, can not be moved and used at will, and can not meet the daily and convenient use requirements of users.
Compared with the prior art, the invention provides machine translation system service for users in an off-line service mode, ensures the safety and privacy of user data, uses an off-line edge computing box as a carrier, is used as the running basis of an integral scheme, is not limited by a network environment, enables the users to use the invention to carry out machine translation processing at any time and any place, provides personalized customization operation for the users on the basis of machine translation model parameters in the general field, enables the whole set of system to be more suitable for the private use scenes of the users, enables the system to understand the users and better serves the users.
Drawings
FIG. 1 is a schematic diagram of the specification parameters of a JETSON NANO base board according to the present invention.
FIG. 2 is a diagram of a Transformer architecture according to the present invention.
Fig. 3 is a flow chart of the internal operation of the present invention.
FIG. 4 is an externally presented flow chart of the present invention.
Detailed Description
The invention will now be further described with reference to the accompanying drawings.
Referring to fig. 1-4, the invention provides a portable off-line machine translation intelligent box, which comprises a hardware part, a software part and an application part;
[ hardware techniques ]:
the device comprises a basic carrier and packaging improvement, wherein the basic carrier is a development board, the packaging improvement is to use plastic acrylic as an integral shell, spaces are reserved in a radiating fin, a power supply port and a network cable port of the development board to be connected, the radiating fin is a metal radiating fin and is provided with a fan, and a USB interface and an HDMI interface of the development board are removed;
[ software part ]:
the method comprises a system framework and a model service, wherein the system framework comprises the following steps:
s1: the system adopts a development board-based self-contained system or Windows, Mac OS;
s2: manually expanding the system exchange space of the development board, and installing swapfile;
s3: then adding a 6GB virtual exchange space for the development board;
s4: deleting irrelevant application programs, reserving a WEB visual interface, and setting a user name and a password which are only known by a developer for an operating system;
s5: basic language environment and configuration, including installation of basic programming language and deep learning framework;
the model service is a deep neural network model using a Transformer and is improved and optimized according to an application scene, and the method comprises the following steps:
s10: model distillation: compressing the machine translation model with large-scale parameter magnitude into a machine translation model with small-scale parameter magnitude, deploying and operating on an edge calculation box, and performing parameter learning layer by layer to enable each layer of parameters of the small model to learn several layers of parameters of the large model in a stepped manner;
s11: marking and generalization: discrete words are rewritten in a special mark mode, and a model is input for learning, so that the purpose of data generalization is achieved, and the text idea can be well reserved when the model is subjected to data input of the type;
s12: customizing by a user: when a professional dictionary or a contrast dictionary of special translation requirements is mined, customizing a machine translation model;
s13: the special words required by the user are marked and replaced by marking generalization processing and pointer network technical means in the S11;
s14: obtaining a certain number of parallel sentence pairs in the current field from an original training corpus and other parallel corpus resource platforms through a dictionary, and retraining a machine translation model in the trained general field on the part of professional corpus data to achieve model fine adjustment, so that the finally obtained machine translation model can be translated in a special field required by a user to be more accurately adapted;
[ application part ]:
including internal operation and external exposure, the steps of internal operation include:
s20: starting a machine translation system through local Web Service for data input and output;
s21: after receiving user input, performing word segmentation, sub-word, case and case conversion and marking preprocessing operation;
s22: transmitting the translation to a machine translation model for model reasoning and translation to obtain an original translation, marking and restoring, and combining words and then processing to obtain a target language translation required by a user;
the internal operation is developed and constructed based on a basic programming language, on the basis of realizing a machine translation model by a deep learning framework, the basic programming language is utilized to carry out data preprocessing and post-processing, and then data interaction is carried out with a Web Service interface, so that the one-time operation flow of a machine translation model system can be completed; the service is provided in the form of a webpage facing the user;
the external presentation includes: after the translation equipment is plugged in a power supply to be started, the machine translation system can be automatically started to operate, after the program operation is completed, a user can be connected with a host personal computer through a network cable, a network port is shared by the host, a corresponding local area network segment IP can be created, the user can use the machine translation system on the personal computer by accessing a network address corresponding to the IP, and after the use is completed, the network cable is pulled out to close the power supply.
The basic vector development board is JETSON NANO, Haisi, and the figure is a basic board.
The deep learning framework of the software part is MXNET, Tensorflow and Pythrch code framework.
The basic language environment is Python3, C/C + +/C #, Java, Php, JS.
The embodiment is as follows:
the invention selects NANO in JETSON series development board produced by Yingweida company in the development process of hardware technology,
the development board of the JETSON NANO is a basic style in a series of invida, is small and exquisite in size, and meanwhile, as an edge computing AI development platform, the development board can support parallel acceleration of models such as a neural network and the like, has good performance and power consumption characteristics, can conveniently and quickly realize CV tasks such as object detection, video processing and the like, and natural language processing tasks such as speech recognition, machine translation and the like, a processor is a four-core CORTEX-a57 of an ARM framework, a GPU is a 128-core Maxwell of NVIDIA, a memory is a 4-GB 64-bit LPDDR4, and it is noted that a 4-GB memory of the JETSON NANO is shared by the CPU and the GPU, so that storage limitation may occur when the neural network model parameters are loaded in a use process, and specific parameters are shown in fig. 1.
The machine translation edge computing box is a product which is secondarily developed on the basis of the JETSON NANO development board. Because the development board is slightly not matched with the practical application scene of the invention in the development and use process, the invention is realized by correspondingly improving the package based on the development board.
The development board is packaged by a shell, plastic acrylic is used as the whole package, and space is reserved for access at the radiating fins, the power supply port, the network cable port and the like.
Secondly, the problem of interface retention, which is required to retain the power interface because the computing box of the present invention requires separate power supplies. In addition, the heat dissipation aspect is provided with a separate metal heat sink, the fan solution space on the basis of the heat sink is reserved, and the basic version is not provided with a fan for the sake of convenience and beauty, and can be additionally installed if a user thinks that the fan is necessary. In addition, the invention also reserves a network wire access port, because the machine translation system provides service for users in the form of personal offline local area network, the network wire port is necessary to reserve an interface space, and other interfaces including USB interface, HDMI interface and the like are not open to users in finished products, on one hand, the system is damaged in order to prevent misoperation of users or plagiarism attack of other people, on the other hand, the aesthetic property and the packaging property are considered, for the users, the fewer the interfaces are, the more concise the operation is, the better the operation is.
The development board is a JETSON NANO development board out of NVIDIA, has an adaptive operating system, is a UBUNTU-based LINUX system, can develop the performance advantages of the development board to the maximum extent, and is friendly in the development process of Linux. But a series of modifications are needed in the application scene of the invention in the using process.
The first problem is that the CPU and the GPU share 4GB of memory, and because the parameter level of the machine translation model is generally huge, under the condition of limited memory, the CPU and the GPU share the problem that the overall performance of the product is reduced and even the product is halted. The solution is to manually expand the system exchange space of the development board, install swapfile, and add 6GB virtual exchange space for the development board, so that the general conditions of machine translation model operation reasoning can be met, and the performance can be better improved after model parameters are further compressed by means of later stage matched with model distillation and the like.
And secondly, when the box really provides services, a visual component of an operating system is not required to be provided, as long as the machine translation system can be operated and a WEB visual interface is provided, so that the visual component of the system is deleted and reduced, and similarly, other irrelevant application programs such as document editing and the like are also included, and all irrelevant software is cut, so that all performances of the development board are fully exerted for the machine translation system, and the smooth applicability of the system is ensured.
Then, in order to ensure the security and privacy of the machine translation system and the edge computing box, the operating system running the machine translation system must be provided with a user name and a password which are only known by the developer. Therefore, on one hand, the system can be prevented from being crashed due to misoperation of the user, and on the other hand, the data safety of the user in the using process is ensured.
Finally, the basic environment and configuration are carried out for the machine translation system of the invention, including the installation of Python3 and MXNET and the configuration installation of other related dependent programs, so that the whole machine translation system can be normally and stably operated.
The Transformer creatively uses a self-attention mechanism to construct an encoder and a decoder, builds a sequence-to-sequence generation model, and the overall architecture is as shown in fig. 2.
Firstly, model distillation is carried out, and the machine translation model of the invention needs to operate reasoning on an edge computing box and is limited by hardware equipment, and the memory of the edge computing box is small, so that large-scale machine translation model parameters cannot be deployed on the edge computing box. The distillation method adopts the step-by-step parameter learning, each layer of parameters of the small model is used for learning several layers of parameters of the large model in a step-by-step mode, for example, three layers of transformers are used for learning six layers of transformers, namely, 0, 1 and 2 layers of parameters of the small model are used for learning 2, 4 and 6 layers of parameters of the large model, and the difference between the parameters is introduced into a loss function of the model for training together, so that the small model with the effect being equal to that of the large model and the parameter quantity being greatly reduced can be obtained. Therefore, on one hand, the performance of the model can be improved, the normal operation of the edge computing assembly is ensured, on the other hand, the quality of the machine translation model can be stabilized, and the daily use requirement of a user can be met.
Secondly, marking generalization, in the training process of the model, because of the limitation of the model structure and the inherent training process problem of the model, the machine translation model can only refer to the existing dictionary appearing in the training data when carrying out reasoning translation after the training is finished, therefore, the updating speed is high, when discretized words such as data of date, time, mailbox addresses and the like are processed, the problems of missing, wrong and the like caused by incomplete identification are easily caused, therefore, the invention carries out marking generalization treatment on the training data when training the machine translation model, namely, the discretization words are rewritten in a special mark form, and input into a model for learning to achieve the aim of data generalization, therefore, the text idea can be well preserved when the model encounters data input of the type, and the whole translation effect and experience sense are better.
Finally, the most important point is user customization, the use scene of the invention is a personal daily machine translation system of the user, and special optimization and improvement based on the professional field and special usage under the use scene of the user are necessary. The invention can carry out customization processing on the machine translation model by digging a professional dictionary, particularly when a user has a contrast dictionary with special translation requirements. On one hand, the special words required by the user can be marked and replaced by the technical means of marking generalization treatment, pointer network and the like, so that complete and accurate translation of a user dictionary is guaranteed, on the other hand, a certain number of parallel sentence pairs in the current field can be obtained from the original training corpus and other parallel corpus resource platforms through the dictionary, and the machine translation model in the trained general field is retrained on the part of professional corpus data to achieve the effect of model fine adjustment, so that the finally obtained machine translation model can be translated in the special field required by the user to be more accurately adapted, and further accord with the daily translation habit of the user, and the product positioning and using requirements based on user customization are met.
As shown in fig. 3, the solution of the present invention operating inside the translation device is to start a machine translation system through a local Web Service for data input and output, perform preprocessing operations such as word segmentation, sub-word, case-case conversion, and labeling after receiving user input, and then transmit the preprocessed operations to a machine translation model for model inference translation, so as to obtain an original translated text, and then perform post-processing operations such as label reduction and word merging to obtain a target language translation required by the user.
The overall system is developed and constructed based on Python, on the basis that the MXNET framework realizes the machine translation model, the Python is used for preprocessing and post-processing data, and then data interaction is carried out with a Web Service interface, so that one-time operation flow of the machine translation model system can be completed. And the service is provided in the form of a webpage when facing the user, and the most convenient use experience is provided for the user by simple page design.
The solution of the present invention is that after the translation device is plugged into a power supply and started, the machine translation system therein automatically starts up and runs, after the program runs, the user can connect with the host personal computer through the network cable, and after the host shares the network port, the corresponding local area network segment IP can be created, and the user can use the machine translation system of the present invention on the personal computer by accessing the website corresponding to the IP. After the use is finished, the network cable is pulled off to turn off the power supply. The solution is selected in consideration of convenience and safety of use of a user, a driver or other connecting means is not required to be independently installed, complexity of operation and unsafety of program implantation are avoided, operation perception of the user on the plug-and-play system is reduced to the greatest extent, and the plug-and-play daily use target is achieved. The flow of the externally presented operation is shown in fig. 4.
The working principle is as follows:
after the translation equipment is plugged in a power supply to start, a machine translation system in the translation equipment can automatically start to run, after a program runs, a user can be connected with a host personal computer through a network cable, a corresponding local area network segment IP can be created when the host shares a network port, and the user can use the machine translation system on the personal computer by accessing a website corresponding to the IP. After the use is finished, the network cable is pulled off to turn off the power supply.
The above are only preferred embodiments of the present invention, and are only used to help understanding the method and the core idea of the present application, the scope of the present invention is not limited to the above embodiments, and all technical solutions belonging to the idea of the present invention belong to the scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may occur to those skilled in the art without departing from the principle of the invention, and are considered to be within the scope of the invention.
The invention integrally solves the problems of high translation dependence network degree, easy leakage of private information and insufficient personalization in the interaction process in the prior art, realizes the characteristics of off-line translation service, local storage, password storage and personalized customization service through unique hardware, software design and application design, has smart integral structure and convenient use, does not need to independently install a driver or other connecting means, avoids the complexity of operation and unsafety of program implantation, reduces the operation perception of a user on the invention to the greatest extent, and realizes the daily use target of plug and play.
Claims (4)
1. A portable off-line machine translation intelligent box is characterized by comprising a hardware part, a software part and an application part;
hardware technology:
the device comprises a basic carrier and packaging improvement, wherein the basic carrier is a development board, the packaging improvement is to use plastic acrylic as an integral shell, a space is reserved for a radiating fin, a power supply port and a network cable port of the development board to be connected, the radiating fin is a metal radiating fin and is provided with a fan, and a USB interface and an HDMI interface of the development board are removed;
[ software part ]:
the method comprises a system framework and a model service, wherein the system framework comprises the following steps:
s1: the system adopts a development board-based self-contained system or Windows, Mac OS;
s2: manually expanding the system exchange space of the development board, and installing swapfile;
s3: then adding a 6GB virtual exchange space for the development board;
s4: deleting irrelevant application programs, reserving a WEB visual interface, and setting a user name and a password which are only known by a developer for an operating system;
s5: basic language environment and configuration, including installation of basic programming language and deep learning framework; the model service is a deep neural network model using a Transformer and is improved and optimized according to an application scene, and the method comprises the following steps:
s10: model distillation: compressing the machine translation model of the large-scale parameter magnitude into a machine translation model of the small-scale parameter magnitude, deploying and operating on the edge calculation box, and performing parameter learning layer by layer, wherein each layer of parameters of the small model learns several layers of parameters of the large model in a stepped manner;
s11: marking and generalization: discrete words are rewritten in a special mark mode, and input models are learned to achieve the purpose of data generalization, so that the text idea can be well reserved when the models meet the data input of the type;
s12: customizing by a user: when a professional dictionary or a contrast dictionary of special translation requirements is mined, customizing a machine translation model;
s13: the special words required by the user are marked and replaced by marking generalization processing and pointer network technical means in S11;
s14: obtaining a certain number of parallel sentence pairs in the current field from an original training corpus and other parallel corpus resource platforms through a dictionary, and retraining a machine translation model in the trained general field on the part of professional corpus data to achieve model fine adjustment, so that the finally obtained machine translation model can be translated in a special field required by a user to be more accurately adapted;
[ application part ]:
comprising an internal operation and an external presentation, the step of internal operation comprising:
s20: starting a machine translation system through local Web Service for data input and output;
s21: after receiving user input, performing word segmentation, sub-word, case and case conversion and marking preprocessing operation;
s22: transmitting the translation to a machine translation model for model reasoning and translation to obtain an original translation, marking and restoring, and combining words and then processing to obtain a target language translation required by a user;
the internal operation is developed and constructed based on a basic programming language, on the basis of realizing a machine translation model by a deep learning framework, the basic programming language is utilized to carry out data preprocessing and post-processing, and then data interaction is carried out with a Web Service interface, so that the one-time operation flow of a machine translation model system can be completed; the service is provided in the form of a webpage when facing the user;
the external presentation includes: after the edge box is plugged with a power supply and started, the machine translation system can be automatically started to operate, after the program operation is completed, a user can be connected with a host personal computer through a network cable, a network port is shared by the host, a corresponding local area network segment IP can be created, the user can use the machine translation system on the personal computer by accessing a network address corresponding to the IP, and after the use is completed, the network cable is pulled out and the power supply is closed.
2. The portable offline machine translation smart box of claim 1, wherein said basic carrier development board is JETSON NANO, haisi, and is a basic board.
3. The portable offline machine translation intelligence box of claim 1, wherein said deep learning framework of said software component is MXNET, tensrflow, Pytorch code framework.
4. The portable offline machine translation intelligence box of claim 1, wherein said basic language environment is Python3, C/C + +/C #, Java, Php, JS.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210124622.XA CN114444523A (en) | 2022-02-10 | 2022-02-10 | Portable off-line machine translation intelligent box |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210124622.XA CN114444523A (en) | 2022-02-10 | 2022-02-10 | Portable off-line machine translation intelligent box |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114444523A true CN114444523A (en) | 2022-05-06 |
Family
ID=81372544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210124622.XA Pending CN114444523A (en) | 2022-02-10 | 2022-02-10 | Portable off-line machine translation intelligent box |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114444523A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506353A (en) * | 2017-10-17 | 2017-12-22 | 深圳市沃特沃德股份有限公司 | Translate box and translation system |
| CN109670191A (en) * | 2019-01-24 | 2019-04-23 | 语联网（武汉）信息技术有限公司 | Calibration optimization method, device and the electronic equipment of machine translation |
| CN110245356A (en) * | 2019-06-24 | 2019-09-17 | 付金龙 | Offline translation machine |
| CN112347763A (en) * | 2020-12-03 | 2021-02-09 | 云知声智能科技股份有限公司 | Knowledge distillation method, device and system for pre-training language model BERT |
| CN113505883A (en) * | 2021-05-31 | 2021-10-15 | 华为技术有限公司 | Neural network training method and device |
| CN113887253A (en) * | 2021-11-10 | 2022-01-04 | 北京有竹居网络技术有限公司 | Method, apparatus, and medium for machine translation |
-
2022
- 2022-02-10 CN CN202210124622.XA patent/CN114444523A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506353A (en) * | 2017-10-17 | 2017-12-22 | 深圳市沃特沃德股份有限公司 | Translate box and translation system |
| CN109670191A (en) * | 2019-01-24 | 2019-04-23 | 语联网（武汉）信息技术有限公司 | Calibration optimization method, device and the electronic equipment of machine translation |
| CN110245356A (en) * | 2019-06-24 | 2019-09-17 | 付金龙 | Offline translation machine |
| CN112347763A (en) * | 2020-12-03 | 2021-02-09 | 云知声智能科技股份有限公司 | Knowledge distillation method, device and system for pre-training language model BERT |
| CN113505883A (en) * | 2021-05-31 | 2021-10-15 | 华为技术有限公司 | Neural network training method and device |
| CN113887253A (en) * | 2021-11-10 | 2022-01-04 | 北京有竹居网络技术有限公司 | Method, apparatus, and medium for machine translation |
Non-Patent Citations (1)
| Title |
|---|
| 赵旭: "基于ULR与元学习策略的蒙汉神经机器翻译的研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20190294676A1 (en) | Written-modality prosody subsystem in a natural language understanding (nlu) framework | |
| CN109801347B (en) | Method, device, equipment and medium for generating editable image template | |
| US10033835B2 (en) | System, method, and device for remotely operating a server | |
| CN102713848A (en) | Methods for interfacing with a virtualized computing service over a network using a lightweight client | |
| KR20190125863A (en) | Multilingual translation device and multilingual translation method | |
| CN111290778B (en) | AI model packaging method, platform and electronic equipment | |
| US10565314B2 (en) | Disambiguating concepts in natural language | |
| CN108415939B (en) | Dialog processing method, device and equipment based on artificial intelligence and computer readable storage medium | |
| CN112966824A (en) | Deployment method and device of inference library and electronic equipment | |
| CN108701045A (en) | Client operating system screenshot method and device in computer equipment | |
| CN114444523A (en) | Portable off-line machine translation intelligent box | |
| CN112015374A (en) | Cross-programming-language micro-service integration system based on natural language | |
| CN116911572A (en) | Page automation assembly method, system, computer equipment and storage medium | |
| US20220382985A1 (en) | Relation extraction exploiting full dependency forests | |
| CN114092608B (en) | Expression processing method and device, computer readable storage medium and electronic equipment | |
| US9141342B2 (en) | Programmatic conversion of support documentation into executable programs | |
| CN114389936A (en) | Cross-cloud multi-cluster deployment operation and maintenance method, system, processor and storage medium | |
| CN113742104A (en) | AI-based RPA (resilient packet Access) body-separating function generation method and device | |
| US20240176614A1 (en) | Automation of generating robotic process automation from automation domain specific languages | |
| US20240069912A1 (en) | Identifying the tranlatability of hard-coded strings in source code via pos tagging | |
| US20240184534A1 (en) | Experience Based Dispatch of Regulated Workloads in a Cloud Environment | |
| US20240170007A1 (en) | Self-supervised speech representations by disentangling speakers | |
| US11977830B2 (en) | Demand-based deployment of font server to an edge device | |
| US20240127801A1 (en) | Domain adaptive speech recognition using artificial intelligence | |
| US20240096012A1 (en) | Optimizing computer-based generation of three-dimensional virtual objects |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220506 |