Multilingual machine translation method and device based on dynamic routing

Through the multilingual machine translation method of dynamic routing, multiple machine translation models are automatically bridged and the optimal translation path is determined, which solves the problems of poor and high cost of existing multilingual translation, and achieves higher quality and lower cost translation Effect.

CN119067135APending Publication Date: 2024-12-03IOL WUHAN INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411239918.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing multilingual machine translation methods have problems with poor translation effect and high cost, especially the multilingual machine translation model is inferior to the single language model in performance, and the new language translation direction requires retraining the model, resulting in resource consumption and Management is difficult.

Method used

Using a multilingual machine translation method based on dynamic routing, we use the construction of a translation routing table, automatically bridge multiple machine translation models, determine the optimal translation path, realize support for more language directions and reduce the cost of translation model development.

Benefits of technology

It improves the quality and reliability of multilingual translation, integrates existing resources, reduces the cost of translation model development, and solves the problem of poor translation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067135A_ABST
    Figure CN119067135A_ABST
Patent Text Reader

Abstract

The invention provides a multi-language machine translation method and device based on dynamic routing, which are applied to the technical field of machine translations, and the method comprises the following steps: obtaining a to-be-translated source language and a to-be-generated target language of a to-be-translated task; traversing a translation routing table based on the to-be-translated source language and the to-be-generated target language, and determining a target translation path, the starting point of the target translation path being the to-be-translated source language, and the end point of the target translation path being the to-be-generated target language; determining at least one machine translation model corresponding to the target translation path; sequentially executing the to-be-translated tasks according to the target translation path based on the at least one machine translation model to obtain a translation result; existing machine translation model resources can be integrated, and the translation model development cost is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of machine translation, and in particular to a multilingual machine translation method and device based on dynamic routing. Background Art

[0002] With the rapid development of technology, machine translation technology has played a huge role in more and more occasions. The current mainstream machine translation technology is the end-to-end neural network machine translation model, that is, given a source language sentence, the machine translation model can automatically give the corresponding target language translation.

[0003] However, machine translation models are closely related to language direction. An English-Chinese machine translation model can only translate English into Chinese, and cannot handle translation tasks from Chinese to English or other languages. Although there are multilingual machine translation models, their performance is often inferior to that of single-language machine translation models. After the multilingual machine translation model is trained, if you want to add a new language translation direction, you will need to spend a lot of energy and resources to retrain the multilingual machine translation model.

[0004] It can be seen that the multilingual machine translation method in the related art has technical problems of poor translation effect and high cost. Summary of the invention

[0005] The present invention provides a multilingual machine translation method and device based on dynamic routing, which are used to solve the defect of poor translation effect in the multilingual machine translation method in the prior art, integrate the existing machine translation model resources, and reduce the cost of translation model development.

[0006] The present invention provides a multilingual machine translation method based on dynamic routing, comprising the following steps: obtaining a source language to be translated and a target language to be generated of a task to be translated; traversing a translation routing table based on the source language to be translated and the target language to be generated, and determining a target translation path, wherein the starting point of the target translation path is the source language to be translated, and the end point of the target translation path is the target language to be generated; determining at least one machine translation model corresponding to the target translation path; and sequentially executing the task to be translated according to the target translation path based on the at least one machine translation model to obtain a translation result.

[0007] According to a multilingual machine translation method based on dynamic routing provided by the present invention, before obtaining the source language to be translated and the target language to be generated of the task to be translated, the method further includes: determining a translation estimation score of each machine translation model in a plurality of preset machine translation models, wherein the translation estimation score is used to represent an estimation of the score of the machine translation model translating the source language into the target language; and constructing a translation routing table based on a plurality of source languages ​​of the plurality of machine translation models and a plurality of target languages ​​of the plurality of machine translation models, wherein the nodes of the translation routing table are the source language and the target language, respectively, and the weight of the directed path of the translation routing table is the translation estimation score.

[0008] According to a multilingual machine translation method based on dynamic routing provided by the present invention, the determining of the translation estimation score of each of the preset multiple machine translation models includes: obtaining source language sets corresponding to the preset multiple machine translation models respectively; calling the multiple machine translation models to translate the source language sets respectively to obtain a target translation set; calling a preset translation evaluation model to perform score estimation based on the source language set and the target translation set to obtain a translation estimation score corresponding to each of the multiple machine translation models.

[0009] According to a multilingual machine translation method based on dynamic routing provided by the present invention, after constructing a translation routing table based on multiple source languages ​​of the multiple machine translation models and multiple target languages ​​of the multiple machine translation models, the method further includes: updating the translation routing table when an additional machine translation model is added or an existing machine translation model is deleted.

[0010] According to a multilingual machine translation method based on dynamic routing provided by the present invention, the step of traversing a translation routing table based on the source language to be translated and the target language to be generated to determine a target translation path comprises: traversing a translation routing table based on the source language to be translated and the target language to be generated to determine a path score for each translation path; and taking the translation path with the highest path score as the target translation path.

[0011] According to a multilingual machine translation method based on dynamic routing provided by the present invention, the traversing the translation routing table based on the source language to be translated and the target language to be generated to determine the path score of each translation path includes: determining the source language node and the target node respectively corresponding to the source language to be translated and the target language to be generated in the translation routing table; determining the path parameters of each translation path between the source language node and the target node, wherein the path parameters include a path weight and a path length; and determining the path score of each translation path based on the path weight, the path length and a preset length penalty factor.

[0012] The present invention also provides a multilingual machine translation device based on dynamic routing, comprising the following modules: an acquisition module, used to acquire a source language to be translated and a target language to be generated of a task to be translated; a first determination module, used to traverse a translation routing table based on the source language to be translated and the target language to be generated, and determine a target translation path, wherein the starting point of the target translation path is the source language to be translated, and the end point of the target translation path is the target language to be generated; a second determination module, used to determine at least one machine translation model corresponding to the target translation path; and an execution module, used to sequentially execute the task to be translated according to the target translation path based on the at least one machine translation model to obtain a translation result.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the multilingual machine translation method based on dynamic routing as described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the multi-language machine translation method based on dynamic routing as described above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned multilingual machine translation methods based on dynamic routing.

[0016] The multilingual machine translation method and device based on dynamic routing provided by the present invention obtain the source language to be translated and the target language to be generated of the task to be translated; take the source language to be translated as the starting point and the target language to be generated as the end point to traverse the translation routing table to determine the target translation path; determine at least one machine translation model corresponding to the target translation path, thereby realizing more language direction support by automatically bridging multiple machine translation models, and determining the optimal path on multiple translation paths to ensure the reliability of translation quality; based on at least one machine translation model, the task to be translated is executed in sequence according to the target translation path to obtain the translation result, thereby achieving the effect of integrating existing resources and reducing the cost of translation model development, thereby solving the technical problem of poor translation effect in the multilingual machine translation method in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of the multilingual machine translation method based on dynamic routing provided by the present invention.

[0019] Figure 2 It is a schematic diagram of the process of constructing a translation routing table provided by the present invention.

[0020] Figure 3 It is a schematic diagram of a process for determining the estimated translation score of each machine translation model provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of a multilingual machine translation device based on dynamic routing provided by the present invention.

[0022] Figure 5 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not 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 are within the scope of protection of the present invention.

[0024] With the rapid development of technology, machine translation technology has played a huge role in more and more occasions, greatly promoting information exchange and international cooperation in the era of globalization. The current mainstream machine translation technology is the end-to-end neural network machine translation model. Given a source language sentence, the machine translation model can automatically give the corresponding target language translation. However, the machine translation model is closely related to the language direction. An English-Chinese machine translation model can only translate English into Chinese, and cannot handle the translation tasks from Chinese to English and other languages. Although there are also multilingual machine translation models, such as multilingual to Chinese machine translation models, Chinese to multilingual machine translation models, and even multilingual to multilingual machine translation models. However, multilingual machine translation models are often inferior to single-language machine translation models in performance. On the other hand, once the multilingual machine translation model is trained, it takes a lot of effort and resources to retrain the multilingual machine translation model to add a new language translation direction. In other words, in actual business, it is often necessary to customize different machine translation models according to the language translation direction, which causes great resource consumption and management difficulties.

[0025] The embodiment of the present invention proposes a multilingual machine translation method based on dynamic routing. The method can integrate multiple machine translation models into a new machine translation system, realize more language direction support by automatically bridging multiple translation models within the system, and automatically select the optimal path on multiple translation paths to ensure the reliability of translation quality, thereby achieving the effect of integrating existing resources and reducing the cost of translation model development.

[0026] Optionally, the multilingual machine translation method based on dynamic routing in the embodiment of the present application can be executed by a server, or by a terminal device, or by both a server and a terminal device, taking the multilingual machine translation method in the embodiment executed by a server as an example.

[0027] refer to Figure 1 , Figure 1 FIG. 1 is a flow chart of a multilingual machine translation method based on dynamic routing provided by the present invention. Figure 1 As shown, the method includes the following steps.

[0028] Step 101: Obtain a source language to be translated and a target language to be generated of a task to be translated.

[0029] In the embodiment of the present invention, the source language to be translated and the target language to be generated of the task to be translated are different. For example, in the task to be translated in English translation, the source language to be translated is English, and the target language to be generated is Chinese.

[0030] Step 102, traversing the translation routing table based on the source language to be translated and the target language to be generated, to determine a target translation path, wherein the starting point of the target translation path is the source language to be translated and the end point of the target translation path is the target language to be generated.

[0031] In the embodiment of the present invention, the translation routing table is a directed graph constructed according to all existing machine translation models, and the source language and the target language in all machine translation models are used as a node on the graph.

[0032] In the embodiment of the present invention, the source language to be translated is taken as the starting point and the target language to be generated is taken as the end point, and the translation routing table is traversed to determine all translation paths; and according to the path parameters of each translation path, the corresponding translation path score is determined, so that the translation path with the highest path score is taken as the target translation path.

[0033] Step 103: Determine at least one machine translation model corresponding to the target translation path.

[0034] In the embodiment of the present invention, since the target translation path may have multiple nodes and the paths between each node correspond to different machine translation models, it is necessary to determine at least one machine translation model corresponding to the target translation path.

[0035] For example, the starting point in the target translation path is Chinese and the end point is German. The preset multiple machine translation models are Chinese to English translation model and English to German translation model. At this time, the Chinese to English translation model and the English to German translation model need to be called sequentially.

[0036] Step 104: based on at least one machine translation model and in accordance with the target translation path, sequentially execute the tasks to be translated to obtain a translation result.

[0037] In the embodiment of the present invention, based on the source language and target language of the translation task and according to the target translation path given by the translation routing table, the corresponding machine translation model is called to perform sequential translation to obtain the final translation result.

[0038] For example, the starting point in the target translation path is Chinese and the end point is German. The preset multiple machine translation models are Chinese to English translation model and English to German translation model. At this time, the Chinese to English translation model is first called to translate Chinese into English; then the English to German translation model is called to translate English into German to obtain the translation result.

[0039] Here, the translation result at least includes the text content in the source language, the text content in the target language, and the comparative translation of each word.

[0040] Through the above steps of the embodiment of the present invention, the source language to be translated and the target language to be generated of the task to be translated are obtained; based on the source language to be translated and the target language to be generated, the translation routing table is traversed to determine the target translation path, wherein the starting point of the target translation path is the source language to be translated and the end point of the target translation path is the target language to be generated; at least one machine translation model corresponding to the target translation path is determined; based on the at least one machine translation model, the task to be translated is sequentially executed according to the target translation path to obtain a translation result; thereby, the effect of integrating existing resources and reducing the cost of developing a translation model can be achieved, thereby solving the technical problem of poor translation effect in the multilingual machine translation method in the related art.

[0041] refer to Figure 2 , Figure 2 The present invention provides a flow chart of constructing a translation routing table, which specifically includes the following steps.

[0042] According to a multilingual machine translation method based on dynamic routing provided by the present invention, before obtaining the source language to be translated and the target language to be generated of the task to be translated, the method further includes: Step 201 : determining a translation estimation score of each machine translation model among a plurality of preset machine translation models, wherein the translation estimation score is used to represent an estimation score of the machine translation model when translating a source language into a target language.

[0043] Step 202: construct a translation routing table based on multiple source languages ​​of multiple machine translation models and multiple target languages ​​of multiple machine translation models, wherein the nodes of the translation routing table are the source languages ​​and the target languages, respectively, and the weights of the directed paths of the translation routing table are the translation estimation scores.

[0044] In the embodiment of the present invention, the translation score estimation is realized by manual (or existing automatic translation evaluation model) based on the source language set and the target translation set, and the score M of the model translating the source language i into the target language j is obtained. ij , M ij The rating range is between 0-100.

[0045] In the embodiment of the present invention, the principle of constructing the translation routing table is: construct a directed graph based on all existing machine translation models, and use the source language and target language in all machine translation models as a node N on the graph. j , N j denotes the node representing language j in the graph.

[0046] For each machine translation model, if the model supports translation from source language i to target language j, a line from node N is built in the graph. i Points to node N j The directed edge ofij , the weight of this edge is M ij (Has been scored manually or by the model.) When there are multiple identical directed edges (directed paths) in the graph (the same source language and the same target language, that is, there are multiple machine translation models that support translation from the source language to the target language), only the edge with the largest weight is a valid edge, and the other edges are ignored.

[0047] refer to Figure 3 , Figure 3 It is a flow chart of determining the estimated translation score of each machine translation model provided by the present invention, which specifically includes the following steps.

[0048] According to a multilingual machine translation method based on dynamic routing provided by the present invention, determining a translation estimation score of each of a plurality of preset machine translation models includes: Step 301: Obtain source language sets corresponding to a plurality of preset machine translation models.

[0049] Step 302: Call multiple machine translation models to translate the source language set respectively to obtain a target translation set.

[0050] Step 303: Call a preset translation evaluation model to perform score estimation based on the source language set and the target translation set to obtain a translation estimation score corresponding to each machine translation model in the multiple machine translation models.

[0051] In an embodiment of the present invention, determining the translation estimation score of each machine translation model in a plurality of preset machine translation models specifically includes the following steps.

[0052] Step a: prepare a corresponding source language set (source language test text set) for each machine translation model.

[0053] Step b: calling the machine translation model to translate the text in the source language set corresponding to the model in step a into the corresponding target language text to obtain a target translation set.

[0054] Step c: manually (or through existing automatic translation evaluation models) to estimate the translation score based on the source language set and the target translation set, and obtain the score M of the machine translation model translating the source language i into the target language j ij , M ij The rating range is between 0-100.

[0055] According to a multilingual machine translation method based on dynamic routing provided by the present invention, after constructing a translation routing table based on multiple source languages ​​of multiple machine translation models and multiple target languages ​​of multiple machine translation models, the method further includes: When an additional machine translation model is added or an existing machine translation model is deleted, the translation routing table is updated.

[0056] In the embodiment of the present invention, whenever a machine translation model is added or deleted, the translation routing table is recalculated.

[0057] According to a multilingual machine translation method based on dynamic routing provided by the present invention, a translation routing table is traversed based on a source language to be translated and a target language to be generated to determine a target translation path, including: Traversing the translation routing table based on the source language to be translated and the target language to be generated, determining a path score for each translation path; The translation path with the highest path score is used as the target translation path.

[0058] In the embodiment of the present invention, the source language to be translated is taken as the starting point and the target language to be generated is taken as the end point, and the translation routing table is traversed to determine all translation paths; and according to the path parameters of each translation path, the corresponding translation path score is determined, so that the translation path with the highest path score is taken as the target translation path.

[0059] Through the embodiments of the present invention, multiple machine translation models are automatically bridged to achieve support for more language directions, and the reliability of translation quality is ensured by determining the optimal path on multiple translation paths.

[0060] According to a multilingual machine translation method based on dynamic routing provided by the present invention, based on traversing a translation routing table of a source language to be translated and a target language to be generated, a path score of each translation path is determined, including: Determine the source language node and the target node corresponding to the source language to be translated and the target language to be generated in the translation routing table; Determine path parameters of each translation path between a source language node and a target node, wherein the path parameters include a path weight and a path length; The path score of each translation path is determined based on the path weight, path length, and a preset length penalty factor.

[0061] In the embodiment of the present invention, a directed graph is constructed as a translation routing table based on all existing machine translation models, and then an acyclic path (target translation path) is found on this directed graph (translation routing table), and the score of this path is calculated. The specific steps are as follows: Step a: All source languages ​​and target languages ​​in the machine translation model are treated as a node N on the graph. j , N j denotes the node representing language j in the graph.

[0062] Step b: For each machine translation model, if the model supports translation from source language i to target language j, a node N is constructed in the graph. i Points to node N j The directed edge of ij , the weight of this edge is M ij (Already scored manually or by the model.) When there are multiple identical directed edges (directed paths) in the graph (the same source language and the same target language, that is, there are multiple machine translation models that support translation from the source language to the target language), only the edge with the largest weight is a valid edge, and the other edges are ignored.

[0063] Step c: Use the breadth-first traversal algorithm to find all directed acyclic paths (there are no two identical languages ​​on the path) P in the graph.

[0064] The path score of path P is calculated using the following formula (1): (1)

[0065] in, Used to indicate translation path The path score, is an edge in the path, is the weight of the edge, Indicates the path The length of (number of sides), Represents the preset length penalty factor. The longer the edge, the more computing resources are required. Therefore, this parameter is used to encourage the use of shorter paths.

[0066] Step d: construct a routing table using all found paths and path scores, where the starting node of the path is the source language in the routing table and the end point of the path is the target language in the routing table.

[0067] In step e, whenever a machine translation model is added or deleted, the routing table is automatically recalculated according to step ad.

[0068] The following describes a specific embodiment of a multilingual machine translation method based on dynamic routing provided by an embodiment of the present invention in an actual application scenario.

[0069] The invention provides a multilingual machine translation engine management method based on dynamic routing. The method comprises three parts: a data preprocessing module, a translation task execution module, and a scheduling module based on dynamic routing.

[0070] The data preprocessing module is responsible for regularizing the input text according to the established rules.

[0071] The translation task execution model is responsible for completing the translation task according to the translation path given by the scheduling model and returning the final translation.

[0072] The scheduling module based on dynamic routing includes two main aspects: on the one hand, based on the source language and target language of the translation task, the best translation path is given according to the routing table; on the other hand, the routing table is automatically generated according to the existing multiple machine translation models, and the routing table is automatically updated when a new translation model is added or deleted.

[0073] For example, the preset machine translation models include: English to Chinese translation model, Chinese to English translation model, Chinese to French translation model, English to German translation model, and German to French translation model.

[0074] The translation routing table generated by the scheduling system can refer to the following Table 1, where Table 1 includes the source language, the target language, the path, and the path score.

[0075] Table 1

[0076] When the source language to be translated of the task to be translated is English, and the target language to be generated of the task to be translated is French, referring to Table 1, although the existing model does not have an English-to-French translation model, two English-to-French translation paths can be found according to the routing table generated by the scheduling system, namely 1-3 and 4-5. Because the path score 90 of 1-3 is greater than the path score of 4-5, the system gives a command: first call model 1 to translate English into Chinese, and then call model 3 to translate Chinese into French, thereby completing the translation task from English to French.

[0077] The following is a description of a multilingual machine translation apparatus based on dynamic routing provided by the present invention. The multilingual machine translation apparatus based on dynamic routing described below and the multilingual machine translation method based on dynamic routing described above can be referred to each other.

[0078] refer to Figure 4 , Figure 4 4 is a schematic diagram of the structure of a multilingual machine translation device based on dynamic routing provided by the present invention, which includes an acquisition module 401, a first determination module 402, a second determination module 403 and an execution module 404.

[0079] An acquisition module 401 is used to acquire a source language to be translated and a target language to be generated of a task to be translated; A first determination module 402 is used to traverse the translation routing table based on the source language to be translated and the target language to be generated, and determine a target translation path, wherein the starting point of the target translation path is the source language to be translated, and the end point of the target translation path is the target language to be generated; A second determination module 403, configured to determine at least one machine translation model corresponding to the target translation path; The execution module 404 is used to sequentially execute the tasks to be translated based on the at least one machine translation model and according to the target translation path to obtain a translation result.

[0080] Specifically, the multilingual machine translation device based on dynamic routing provided by the present invention can implement all the method steps implemented by the multilingual machine translation method embodiment based on dynamic routing, and can achieve the same technical effects. The parts and beneficial effects that are the same as those in the method embodiment will not be described in detail herein.

[0081] Figure 5 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a multilingual machine translation method based on dynamic routing, the method comprising: obtaining a source language to be translated and a target language to be generated of a task to be translated; traversing a translation routing table based on the source language to be translated and the target language to be generated, and determining a target translation path, wherein the starting point of the target translation path is the source language to be translated, and the end point of the target translation path is the target language to be generated; determining at least one machine translation model corresponding to the target translation path; and sequentially executing the tasks to be translated according to the target translation path based on at least one machine translation model to obtain a translation result.

[0082] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0083] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multilingual machine translation method based on dynamic routing provided by the above methods, the method including: obtaining a source language to be translated and a target language to be generated of the task to be translated; traversing a translation routing table based on the source language to be translated and the target language to be generated, and determining a target translation path, wherein the starting point of the target translation path is the source language to be translated, and the end point of the target translation path is the target language to be generated; determining at least one machine translation model corresponding to the target translation path; and sequentially executing the tasks to be translated according to the target translation path based on the at least one machine translation model to obtain a translation result.

[0084] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the multilingual machine translation method based on dynamic routing provided by the above-mentioned methods, the method comprising: obtaining a source language to be translated and a target language to be generated of a task to be translated; traversing a translation routing table based on the source language to be translated and the target language to be generated, and determining a target translation path, wherein the starting point of the target translation path is the source language to be translated, and the end point of the target translation path is the target language to be generated; determining at least one machine translation model corresponding to the target translation path; and sequentially executing the tasks to be translated according to the target translation path based on the at least one machine translation model to obtain a translation result.

[0085] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0086] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multilingual machine translation method based on dynamic routing, It is characterized in that include: Obtaining a source language to be translated and a target language to be generated for a task to be translated; Traversing a translation routing table based on the source language to be translated and the target language to be generated, determining a target translation path, wherein a starting point of the target translation path is the source language to be translated and an end point of the target translation path is the target language to be generated; determining at least one machine translation model corresponding to the target translation path; Based on the at least one machine translation model and in accordance with the target translation path, the tasks to be translated are sequentially executed to obtain a translation result.

2. The multilingual machine translation method based on dynamic routing according to claim 1, It is characterized in that Before obtaining the source language to be translated and the target language to be generated of the task to be translated, the method further includes: Determining a translation estimation score of each of a plurality of preset machine translation models, wherein the translation estimation score is used to represent an estimation score of the machine translation model for translating a source language into a target language; Based on the multiple source languages ​​of the multiple machine translation models and the multiple target languages ​​of the multiple machine translation models, a translation routing table is constructed, wherein the nodes of the translation routing table are the source languages ​​and the target languages, respectively, and the weights of the directed paths of the translation routing table are the translation estimation scores.

3. The multilingual machine translation method based on dynamic routing according to claim 2, It is characterized in that The step of determining the estimated translation score of each of the preset multiple machine translation models includes: Obtaining source language sets corresponding to a plurality of preset machine translation models respectively; Calling the multiple machine translation models to translate the source language set respectively to obtain a target translation set; A preset translation evaluation model is called to perform score estimation based on the source language set and the target translation set to obtain a translation estimation score corresponding to each machine translation model in the multiple machine translation models.

4. The multilingual machine translation method based on dynamic routing according to claim 2, It is characterized in that After constructing the translation routing table based on the multiple source languages ​​of the multiple machine translation models and the multiple target languages ​​of the multiple machine translation models, the method further includes: When an additional machine translation model is added or an existing machine translation model is deleted, the translation routing table is updated.

5. The multilingual machine translation method based on dynamic routing according to claim 1, It is characterized in that The traversing the translation routing table based on the source language to be translated and the target language to be generated to determine the target translation path includes: Traversing the translation routing table based on the source language to be translated and the target language to be generated, determining a path score for each translation path; The translation path with the highest path score is used as the target translation path.

6. The multilingual machine translation method based on dynamic routing according to claim 5, It is characterized in that The traversing the translation routing table based on the source language to be translated and the target language to be generated to determine the path score of each translation path includes: Determine the source language node and the target node corresponding to the source language to be translated and the target language to be generated in the translation routing table respectively; Determine path parameters of each translation path between the source language node and the target node, wherein the path parameters include path weight and path length; A path score of each translation path is determined based on the path weight, the path length, and a preset length penalty factor.

7. A multilingual machine translation device based on dynamic routing, It is characterized in that include: An acquisition module is used to acquire a source language to be translated and a target language to be generated of a task to be translated; A first determination module, configured to traverse a translation routing table based on the source language to be translated and the target language to be generated, and determine a target translation path, wherein a starting point of the target translation path is the source language to be translated, and an end point of the target translation path is the target language to be generated; A second determination module, configured to determine at least one machine translation model corresponding to the target translation path; An execution module is used to sequentially execute the tasks to be translated based on the at least one machine translation model and according to the target translation path to obtain a translation result.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the multilingual machine translation method based on dynamic routing as claimed in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the multilingual machine translation method based on dynamic routing as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the multilingual machine translation method based on dynamic routing as claimed in any one of claims 1 to 6 is implemented.