Language intelligent translation method and device for automobile repair data

Abstract

The application discloses a language intelligent translation method and device for automobile maintenance data, and the method comprises the following steps: encoding each source language object after segmentation by using a trained source language encoding model to obtain the source language encoding corresponding to each source language object; encoding and decoding the source language encoding sequence according to all target language corpus alignment templates corresponding to all determined source language objects to obtain the target language decoding sequence corresponding to the source language encoding sequence, and converting the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence. It can be seen that the source language can be intelligently converted into the target language by implementing the application, and a large amount of time is not needed for one-by-one conversion, which is beneficial to improving the translation efficiency of the source language, and further beneficial to quickly repairing the automobile according to the translated automobile maintenance data, so that the automobile repair work can be smoothly carried out.

Description

Technical Field

[0001] This invention relates to the field of language translation technology, and in particular to a method and apparatus for intelligent language translation of automotive repair data. Background Technology

[0002] With the continuous development of the transportation industry, the auto repair industry has also presented a diverse and vibrant landscape. As automobiles are constantly being updated and replaced, auto repair technology is also continuously being updated and improved, spanning the entire lifecycle of a vehicle. Therefore, auto repair technology is crucial for the normal use of a car.

[0003] Currently, automotive repair requires extensive reference to automotive repair manuals for technical guidance. However, these manuals are written in a wide variety of languages ​​and use different grammars, causing technicians to spend considerable time translating them, which is detrimental to repair work. Therefore, providing a method to improve the efficiency of translating automotive repair manuals is particularly important. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and apparatus for intelligent translation of automotive repair data, which is beneficial to improving the translation efficiency of the source language, thereby facilitating the repair of automobiles based on the translated automotive repair data, and thus enabling the automotive repair work to proceed smoothly.

[0005] To address the aforementioned technical problems, the first aspect of this invention discloses a method for intelligent translation of automotive repair data, the method comprising:

[0006] The source language statement that needs to be translated into the target language is segmented to obtain the source language sequence corresponding to the source language statement; the source language sequence includes at least one source language object arranged in a preset order, and the source language object is a source language character or source language word;

[0007] By using a pre-trained source language encoding model, an encoding operation is performed on each source language object to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence.

[0008] Based on the source language code corresponding to each source language object, multiple target language corpus alignment templates corresponding to each source language object are determined from a preset target language corpus alignment library; each target language corpus alignment template includes multiple reference source language objects and a reference target language object corresponding to each reference source language object, and there is a preset positional relationship between each reference source language object and its corresponding reference target language object;

[0009] Based on the alignment templates of all the target language corpora corresponding to all the source language objects, an encoding and decoding operation is performed on the source language encoding sequence corresponding to the source language sequence to obtain the target language decoding sequence corresponding to the source language encoding sequence, and a decoding-translation conversion is performed on the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence.

[0010] As an optional implementation, in the first aspect of the present invention, before determining the plurality of target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object, the method further includes:

[0011] Based on the source language code corresponding to each source language object, a target language translation code tag grid corresponding to each source language object is matched from a preset target language translation code tag library; the target language translation code tag grid includes multiple target language translation code tags that match the corresponding source language object;

[0012] The step of determining multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object includes:

[0013] Based on the source language encoding corresponding to each source language object, multiple undetermined target language corpus alignment templates corresponding to each source language object are matched from a preset target language corpus alignment library;

[0014] For each source language object, based on the target language translation encoding label grid corresponding to the source language object, all target language corpus alignment templates that match any target language translation encoding in the corresponding target language translation encoding label grid are determined from all the target language corpus alignment templates corresponding to the source language object, and are used as the target language corpus alignment templates corresponding to the source language object.

[0015] As an optional implementation, in the first aspect of the present invention, the step of performing an encoding-decoding operation on the source language encoding sequence corresponding to the source language sequence based on the alignment templates of all the target language corpora corresponding to all the source language objects, to obtain the target language decoding sequence corresponding to the source language encoding sequence, includes:

[0016] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, determine the object position of the reference target language object corresponding to all reference source language objects in each target language corpus alignment template, and use it as the recombination position corresponding to all reference source language objects in each target language corpus alignment template;

[0017] For each reference source language object in each target language corpus alignment template, based on the recombination position corresponding to the reference source language object, determine the statistical quantity of multiple decodes of the reference target language object that match the recombination position corresponding to the reference source language object;

[0018] Based on the statistics of each decoded code of the reference target language object that matches the recombination position corresponding to the reference source language object, the decoded codes whose statistics are greater than or equal to a preset statistical threshold are determined from all the decoded codes and used as the target decoded codes of the reference target language object that match the recombination position corresponding to the reference source language object.

[0019] Based on the target decoding in the recombination position corresponding to each reference source language object in each target language corpus alignment template and the preset decoding contribution value calculation formula, calculate the target decoding contribution value corresponding to all reference source language objects in each target language corpus alignment template;

[0020] From all the target language corpus alignment templates, determine the target language output reconstruction template whose target decoding contribution value is greater than or equal to a preset contribution value threshold, and determine the target language decoding sequence corresponding to the source language encoding sequence based on the target decoding in the reconstruction position corresponding to each reference source language object in the target language output reconstruction template.

[0021] As an optional implementation, in the first aspect of the present invention, determining the object position of the reference target language object corresponding to all reference source language objects in each target language corpus alignment template, based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, as the recombination position corresponding to all reference source language objects in each target language corpus alignment template, includes:

[0022] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, the correlation degree between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template is calculated; the greater the correlation degree, the more similar the sequence content of the source language encoding sequence corresponding to the source language sequence is to the template content of the target language corpus alignment template.

[0023] The recombination order of each target language corpus alignment template is determined based on the degree of correlation between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template.

[0024] Based on the recombination order of each target language corpus alignment template, the preset positional relationship between each reference source language object and the corresponding reference target language object in each target language corpus alignment template, and the object position of each source language object in the source language encoding sequence corresponding to the source language sequence, the object positions of all reference source language objects in each target language corpus alignment template are determined sequentially, and these positions are used as the recombination positions of all reference source language objects in each target language corpus alignment template.

[0025] As an optional implementation, in the first aspect of the present invention, before segmenting the source language statement to be translated into the target language to obtain the source language sequence corresponding to the source language statement, the method further includes:

[0026] The source language training sample set and the target language training sample set corresponding to the source language training sample set are segmented to obtain a first list corresponding to the source language training sample set and a second list corresponding to the target language training sample set; both the first list and the second list include a number of corresponding training sample objects, wherein the training sample objects are training sample characters or training sample words.

[0027] Multiple training operations are performed on the preset language model to be trained using the first list and the second list.

[0028] During each training operation, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set are generated based on the first list and the second list. The encoding contribution value of the first word combination sequence and the encoding contribution value of the second word combination sequence corresponding to the previous training operation are updated based on the first word combination sequence and the second word combination sequence.

[0029] As an optional implementation, in a first aspect of the present invention, generating a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set based on the first list and the second list includes:

[0030] Calculate the association matrix parameters between the first list and the second list;

[0031] Based on the correlation matrix parameters, analyze the object alignment relationship between all training sample objects corresponding to the first list and all training sample objects corresponding to the second list;

[0032] Based on the object alignment relationship, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set are generated.

[0033] The association matrix parameter between the first list and the second list is:

[0034]

[0035] ceM(i,j) is the parameter of the correlation matrix, p(a i p(b) represents the probability of occurrence of the training object corresponding to the first list. j p(a) represents the probability of occurrence of the training object corresponding to the second list. i │b j p(b) represents the probability of the first list corresponding to the training object appearing given that the training object corresponding to the second list appears. j │a i ) represents the probability of the second list corresponding to the training object appearing given that the training object corresponding to the first list appears.

[0036] As an optional implementation, in the first aspect of the present invention, updating the encoding contribution value of the first phrase combination sequence and the encoding contribution value of the corresponding second phrase combination sequence corresponding to the previous training operation based on the first phrase combination sequence and the second phrase combination sequence includes:

[0037] Determine the statistical value of each first word group in the first word group combination sequence and the statistical value of each second word group in the second word group combination sequence, and determine the statistical growth direction of each first word group and the statistical growth direction of each second word group based on the statistical value of each first word group and the statistical value of each second word group.

[0038] Based on the statistical growth direction of each first word group and the statistical growth direction of each second word group, determine the word group code of each first word group and the word group code of each second word group, which serve as the encoding sequence corresponding to the first word group combination sequence and the encoding sequence corresponding to the second word group combination sequence.

[0039] Based on the encoding sequence corresponding to the first word combination sequence and the encoding sequence corresponding to the second word combination sequence, update the encoding sequence corresponding to the first word combination sequence and the encoding sequence corresponding to the second word combination sequence in the previous training operation.

[0040] Based on the updated encoding sequence of the first phrase combination sequence and the updated encoding sequence of the second phrase combination sequence, calculate the encoding contribution value corresponding to the phrase encoding of each first phrase and the encoding contribution value corresponding to the phrase encoding of each second phrase.

[0041] Based on the encoding contribution values ​​corresponding to the word codes of all the first word groups and the encoding contribution values ​​corresponding to the word codes of all the second word groups, update the encoding contribution values ​​of the first word group combination sequence and the encoding contribution values ​​of the corresponding second word group combination sequence corresponding to the previous training operation;

[0042] Among them, the coding contribution value corresponding to the word group code is:

[0043]

[0044] S x n is the code contribution value corresponding to the word group code. x T represents the frequency of occurrence corresponding to the word group encoding. x N represents the total number of word group codes corresponding to the word groups in all preset libraries, and C represents the number of all preset libraries.

[0045] A second aspect of this invention discloses a language intelligent translation device for automotive repair data, the device comprising:

[0046] The segmentation module is used to segment the source language statement that needs to be translated into the target language to obtain the source language sequence corresponding to the source language statement; the source language sequence includes at least one source language object arranged in a preset order, and the source language object is a source language character or a source language word;

[0047] The encoding module is used to perform encoding operations on each source language object using a pre-trained source language encoding model to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence.

[0048] The determining module is used to determine multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object; each target language corpus alignment template includes multiple reference source language objects and a reference target language object corresponding to each reference source language object, and there is a preset positional relationship between each reference source language object and the corresponding reference target language object;

[0049] The encoding / decoding module is used to perform encoding / decoding operations on the source language encoding sequence corresponding to the source language sequence according to the alignment template of all the target language corpora corresponding to all the source language objects, so as to obtain the target language decoding sequence corresponding to the source language encoding sequence.

[0050] The decoding and translation module is used to perform decoding and translation on the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence.

[0051] As an optional implementation, in a second aspect of the invention, the apparatus further includes:

[0052] The matching module is used to match a target language translation encoding tag grid corresponding to each source language object from a preset target language corpus alignment library before the determining module determines multiple target language corpus alignment templates corresponding to each source language object based on the source language encoding corresponding to each source language object; the target language translation encoding tag grid includes multiple target language translation encoding tags that match the corresponding source language object;

[0053] Specifically, the method by which the determining module determines multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object is as follows:

[0054] Based on the source language encoding corresponding to each source language object, multiple undetermined target language corpus alignment templates corresponding to each source language object are matched from a preset target language corpus alignment library;

[0055] For each source language object, based on the target language translation encoding label grid corresponding to the source language object, all target language corpus alignment templates that match any target language translation encoding in the corresponding target language translation encoding label grid are determined from all the target language corpus alignment templates corresponding to the source language object, and are used as the target language corpus alignment templates corresponding to the source language object.

[0056] As an optional implementation, in the second aspect of the present invention, the encoding / decoding module performs encoding / decoding operations on the source language encoding sequence corresponding to the source language sequence based on the alignment templates of all the target language corpora corresponding to all the source language objects, and obtains the target language decoding sequence corresponding to the source language encoding sequence in the following specific manner:

[0057] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, determine the object position of the reference target language object corresponding to all reference source language objects in each target language corpus alignment template, and use it as the recombination position corresponding to all reference source language objects in each target language corpus alignment template;

[0058] For each reference source language object in each target language corpus alignment template, based on the recombination position corresponding to the reference source language object, determine the statistical quantity of multiple decodes of the reference target language object that match the recombination position corresponding to the reference source language object;

[0059] Based on the statistics of each decoded code of the reference target language object that matches the recombination position corresponding to the reference source language object, the decoded codes whose statistics are greater than or equal to a preset statistical threshold are determined from all the decoded codes and used as the target decoded codes of the reference target language object that match the recombination position corresponding to the reference source language object.

[0060] Based on the target decoding in the recombination position corresponding to each reference source language object in each target language corpus alignment template and the preset decoding contribution value calculation formula, calculate the target decoding contribution value corresponding to all reference source language objects in each target language corpus alignment template;

[0061] From all the target language corpus alignment templates, determine the target language output reconstruction template whose target decoding contribution value is greater than or equal to a preset contribution value threshold, and determine the target language decoding sequence corresponding to the source language encoding sequence based on the target decoding in the reconstruction position corresponding to each reference source language object in the target language output reconstruction template.

[0062] As an optional implementation, in the second aspect of the present invention, the encoding / decoding module determines the object positions of all reference source language objects corresponding to reference source language objects in each target language corpus alignment template based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object. The specific method for determining the recombination positions corresponding to all reference source language objects in each target language corpus alignment template is as follows:

[0063] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, the correlation degree between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template is calculated; the greater the correlation degree, the more similar the sequence content of the source language encoding sequence corresponding to the source language sequence is to the template content of the target language corpus alignment template.

[0064] The recombination order of each target language corpus alignment template is determined based on the degree of correlation between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template.

[0065] Based on the recombination order of each target language corpus alignment template, the preset positional relationship between each reference source language object and the corresponding reference target language object in each target language corpus alignment template, and the object position of each source language object in the source language encoding sequence corresponding to the source language sequence, the object positions of all reference source language objects in each target language corpus alignment template are determined sequentially, and these positions are used as the recombination positions of all reference source language objects in each target language corpus alignment template.

[0066] As an optional implementation, in a second aspect of the invention, the segmentation processing module is further configured to:

[0067] Before segmenting the source language statement to be translated into the target language to obtain the source language sequence corresponding to the source language statement, the source language training sample set in the target training sample set used for training and the target language training sample set corresponding to the source language training sample set are segmented to obtain a first list corresponding to the source language training sample set and a second list corresponding to the target language training sample set; both the first list and the second list include a number of corresponding training sample objects, wherein the training sample objects are training sample characters or training sample words.

[0068] The device also includes:

[0069] The training module is used to perform multiple training operations on a preset language model to be trained using the first list and the second list.

[0070] The sequence generation module is used to generate, in each training operation, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set, based on the first list and the second list.

[0071] The update module is used to update the encoding contribution value of the first word combination sequence and the encoding contribution value of the corresponding second word combination sequence in the previous training operation based on the first word combination sequence and the second word combination sequence.

[0072] As an optional implementation, in a second aspect of the present invention, the sequence generation module generates a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set based on the first list and the second list, specifically in the following manner:

[0073] Calculate the association matrix parameters between the first list and the second list;

[0074] Based on the correlation matrix parameters, analyze the object alignment relationship between all training sample objects corresponding to the first list and all training sample objects corresponding to the second list;

[0075] Based on the object alignment relationship, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set are generated.

[0076] The association matrix parameter between the first list and the second list is:

[0077]

[0078] ceM(i,j) is the parameter of the correlation matrix, p(a i p(b) represents the probability of occurrence of the training object corresponding to the first list. j p(a) represents the probability of occurrence of the training object corresponding to the second list. i │b j p(b) represents the probability of the first list corresponding to the training object appearing given that the training object corresponding to the second list appears. j │a i ) represents the probability of the second list corresponding to the training object appearing given that the training object corresponding to the first list appears.

[0079] As an optional implementation, in a second aspect of the present invention, the updating module updates the encoding contribution value of the first phrase combination sequence and the encoding contribution value of the corresponding second phrase combination sequence based on the first phrase combination sequence and the second phrase combination sequence in the manner specifically as follows:

[0080] Determine the statistical value of each first word group in the first word group combination sequence and the statistical value of each second word group in the second word group combination sequence, and determine the statistical growth direction of each first word group and the statistical growth direction of each second word group based on the statistical value of each first word group and the statistical value of each second word group.

[0081] Based on the statistical growth direction of each first word group and the statistical growth direction of each second word group, determine the word group code of each first word group and the word group code of each second word group, which serve as the encoding sequence corresponding to the first word group combination sequence and the encoding sequence corresponding to the second word group combination sequence.

[0082] Based on the encoding sequence corresponding to the first word combination sequence and the encoding sequence corresponding to the second word combination sequence, update the encoding sequence corresponding to the first word combination sequence and the encoding sequence corresponding to the second word combination sequence in the previous training operation.

[0083] Based on the updated encoding sequence of the first phrase combination sequence and the updated encoding sequence of the second phrase combination sequence, calculate the encoding contribution value corresponding to the phrase encoding of each first phrase and the encoding contribution value corresponding to the phrase encoding of each second phrase.

[0084] Based on the encoding contribution values ​​corresponding to the word codes of all the first word groups and the encoding contribution values ​​corresponding to the word codes of all the second word groups, update the encoding contribution values ​​of the first word group combination sequence and the encoding contribution values ​​of the corresponding second word group combination sequence corresponding to the previous training operation;

[0085] Among them, the coding contribution value corresponding to the word group code is:

[0086]

[0087] S x n is the code contribution value corresponding to the word group code. x T represents the frequency of occurrence corresponding to the word group encoding. x N represents the total number of word group codes corresponding to the word groups in all preset libraries, and C represents the number of all preset libraries.

[0088] A third aspect of this invention discloses another intelligent language translation device for automotive repair data, the device comprising:

[0089] Memory containing executable program code;

[0090] A processor coupled to the memory;

[0091] The processor calls the executable program code stored in the memory to execute the intelligent language translation method for automotive repair data disclosed in the first aspect of the present invention.

[0092] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute the intelligent language translation method for automotive repair data disclosed in the first aspect of the present invention.

[0093] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0094] In this embodiment of the invention, the source language statements to be translated into the target language are segmented to obtain the source language sequences corresponding to the source language statements. Using a pre-trained source language encoding model, encoding is performed on each source language object to obtain the source language code corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence. Based on the source language code corresponding to each source language object, multiple target language corpus alignment templates corresponding to each source language object are determined from a preset target language corpus alignment library. Based on all target language corpus alignment templates corresponding to all source language objects, encoding and decoding operations are performed on the source language encoding sequences corresponding to the source language sequences to obtain the target language decoding sequences corresponding to the source language encoding sequences. Furthermore, the target language decoding sequences corresponding to the source language encoding sequences are decoded and translated to obtain the target language translation sequences corresponding to the source language encoding sequences. Therefore, implementing this invention can intelligently convert the source language into the target language without spending a significant amount of time on individual conversions. While ensuring the grammatical accuracy of each language during the translation process, it is beneficial to improve the translation efficiency of the source language, thereby facilitating rapid vehicle repair based on the translated automotive repair data, and enabling smooth vehicle repair work. Attached Figure Description

[0095] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0096] Figure 1 This is a schematic diagram of a target language translation encoding tag grid corresponding to a source language object, as disclosed in an embodiment of the present invention.

[0097] Figure 2 This is a schematic diagram of a target language corpus alignment template corresponding to a source language object, as disclosed in an embodiment of the present invention.

[0098] Figure 3 This is a schematic diagram illustrating the calculation of the correlation matrix parameters between a first list and a second list, as disclosed in an embodiment of the present invention.

[0099] Figure 4 This is a schematic diagram illustrating the process of determining the word group encoding of a corresponding word group according to an embodiment of the present invention;

[0100] Figure 5 This is a flowchart illustrating a method for intelligent language translation of automotive repair data disclosed in an embodiment of the present invention;

[0101] Figure 6 This is a flowchart illustrating another intelligent language translation method for automotive repair data disclosed in an embodiment of the present invention;

[0102] Figure 7 This is a schematic diagram of the structure of a language intelligent translation device for automobile repair data disclosed in an embodiment of the present invention;

[0103] Figure 8 This is a schematic diagram of the structure of another intelligent language translation device for automotive repair data disclosed in an embodiment of the present invention;

[0104] Figure 9 This is a schematic diagram of the structure of another intelligent language translation device for automotive repair data disclosed in an embodiment of the present invention. Detailed Implementation

[0105] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0106] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0107] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0108] This invention discloses a method and apparatus for intelligent translation of automotive repair data, which can improve the translation efficiency of the source language, thereby facilitating rapid automotive repair based on the translated data and ensuring smooth automotive repair work. Detailed descriptions follow.

[0109] Example 1

[0110] Please see Figure 5 , Figure 5 This is a flowchart illustrating a method for intelligent language translation of automotive repair data, as disclosed in an embodiment of the present invention. Figure 5 The described intelligent language translation method for automotive repair data can be applied to translate data on various types of repair work containing common words or professional terms, such as vehicle repair, electrical repair, equipment repair, etc., and this embodiment of the invention is not limited thereto. Optionally, this method can translate different types of source languages ​​into other target languages ​​(e.g., translating Chinese or English into French), or it can translate source languages ​​into different types of target languages ​​(e.g., translating Chinese into French or English). Further optionally, this method can be implemented by a language translation system, which can be integrated into a language translation device, or it can be a local server or cloud server used to process the language translation process, etc., and this embodiment of the invention is not limited thereto. Figure 5 As shown, the intelligent language translation method for automotive repair data can include the following operations:

[0111] 101. Segment the source language statements that need to be translated into the target language to obtain the source language sequence corresponding to the source language statements.

[0112] In this embodiment of the invention, the source language sequence includes at least one source language object arranged in a preset order. Optionally, the source language object is a source language character or a source language word. Specifically, the segmentation process of the source statement can include regular segmentation and full-mode segmentation. Regular segmentation divides the source language statement into individual characters and words, such as segmenting "oxidation catalytic converter" into "oxidation", "catalytic", and "converter". Full-mode segmentation divides the sequence of individual characters and words into a sequence composed of characters, words, phrases, and characters, such as segmenting "oxidation", "catalytic", and "converter" into "Oxidation / Oxidation catalyticconverter", "Catalytic / catalytic converter", and "converter".

[0113] 102. Using a pre-trained source language encoding model, perform encoding operations on each source language object to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence.

[0114] In this embodiment of the invention, each source language object in the source language sequence that needs to be translated into the target language is converted into a code. Optionally, the encoding base can be binary, decimal, hexadecimal, etc., and this embodiment of the invention does not limit it. For example, the source language object "oxidation" is converted into the binary code "1101".

[0115] 103. Based on the source language encoding corresponding to each source language object, determine multiple target language corpus alignment templates corresponding to each source language object from the preset target language corpus alignment library.

[0116] In this embodiment of the invention, each target language corpus alignment template includes multiple reference source language objects and a corresponding reference target language object for each reference source language object, and a preset positional relationship exists between each reference source language object and its corresponding reference target language object. For example, such as Figure 2 As shown, Figure 2 This is a schematic diagram of a target language corpus alignment template corresponding to a source language object, as disclosed in an embodiment of the present invention. Figure 2As shown, template 79484 includes the following reference source language objects: "right", "catalytic converter", "exhaust", "temperature", and "sensor", while the corresponding reference target language objects are "right side", "catalytic converter", "exhaust gas", "temperature", and "sensor". There is a one-to-one positional relationship between the reference source language objects and the reference target language objects, as shown by the arrow (for ease of understanding, the corresponding codes have been converted into source and target languages).

[0117] 104. Based on the alignment templates of all target language corpora corresponding to all source language objects, perform encoding and decoding operations on the source language encoding sequences corresponding to the source language sequences to obtain the target language decoding sequences corresponding to the source language encoding sequences. Then, perform decoding and translation conversion on the target language decoding sequences corresponding to the source language encoding sequences to obtain the target language translation sequences corresponding to the source language encoding sequences.

[0118] In this embodiment of the invention, the source language encoding sequence is first converted into the target language encoding sequence by aligning with the corresponding target language corpus template, and then the target language encoding sequence is converted into the target language translation.

[0119] Furthermore, as an optional implementation, the method may further include:

[0120] After obtaining the target language translation sequence corresponding to the source language encoding sequence, each target language translation object in the target language translation sequence is merged to obtain all target language translation objects after merging.

[0121] Determine the padding positions and corresponding padding characters in all merged target language translation objects, and perform character padding operations on all merged target language translation objects based on these positions and characters to update all merged target language translation objects.

[0122] For example, after obtaining the target language translation sequence "exhaust gas", "temperature", "sensor", and "voltage", performing a merging operation and a character padding operation can yield "exhaust gas temperature sensor, voltage".

[0123] As can be seen, implementing the embodiments of the present invention can intelligently convert the source language into the target language without spending a lot of time on step-by-step conversion. While ensuring the grammatical accuracy of each language during the translation process, it helps improve the efficiency of translating the source language, which in turn facilitates rapid vehicle repair based on the translated automotive repair data, thus enabling smooth vehicle repair work.

[0124] In an optional embodiment, before segmenting the source language statement to be translated into the target language in step 101 above to obtain the source language sequence corresponding to the source language statement, the method further includes:

[0125] The source language training sample set and the target language training sample set corresponding to the source language training sample set are segmented in the target training sample set used for training to obtain a first list corresponding to the source language training sample set and a second list corresponding to the target language training sample set.

[0126] Multiple training operations are performed on the preset language model to be trained using the first list and the second list.

[0127] During each training operation, based on the first list and the second list, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set are generated. Based on the first word combination sequence and the second word combination sequence, the encoding contribution value of the first word combination sequence and the encoding contribution value of the corresponding second word combination sequence in the previous training operation are updated.

[0128] In this optional embodiment, both the first list and the second list include corresponding training sample objects. Optionally, the training sample objects are training sample characters or training sample words. Specifically, the source language training sample set and the target language training sample set have a complete alignment relationship, that is, they are one-to-one sentence combinations. More specifically, during each training operation, the encoding of the first word combination sequence and the encoding of the corresponding second word combination sequence corresponding to the previous training operation can be updated according to the first word combination sequence and the second word combination sequence. Even more specifically, after completing multiple training operations on the language model to be trained, a language encoding model can be obtained, which can be applied to encode source language sequences.

[0129] As can be seen, this optional embodiment can intelligently train the language model to be trained using the source language training sample set and the corresponding target language training sample set to obtain a language encoding model that can be applied to the encoding of the source language sequence. In this way, the language encoding model can be obtained reliably and accurately, which not only helps to improve the subsequent encoding efficiency of the source language sequence, thereby quickly obtaining the encoding sequence corresponding to the source language sequence, but also helps to improve the reliability and accuracy of the subsequent encoding of the source language sequence, thereby accurately obtaining the encoding sequence corresponding to the source language sequence.

[0130] In another optional embodiment, the step of generating a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set based on the first list and the second list, includes:

[0131] Calculate the parameters of the association matrix between the first list and the second list;

[0132] Based on the correlation matrix parameters, analyze the object alignment relationship between all training sample objects corresponding to the first list and all training sample objects corresponding to the second list.

[0133] Based on the object alignment relationship, generate the first word combination sequence corresponding to the source language training sample set and the second word combination sequence corresponding to the target language training sample set.

[0134] In this optional embodiment, specifically, the association matrix parameters between the first list and the second list are:

[0135]

[0136] Where ceM(i,j) is the correlation matrix parameter, p(a i p(b) represents the probability of occurrence of the training object corresponding to the first list. j p(a) represents the probability of occurrence of the training object corresponding to the second list. i │b j p(b) represents the probability of the first list's corresponding training object appearing given that the second list's corresponding training object appears. j │a i Let be the probability of the second training object appearing given that the training object in the first list has appeared. For example, if... Figure 3 As shown, Figure 3 This is a schematic diagram of the calculation of the correlation matrix parameters between a first list and a second list disclosed in an embodiment of the present invention. In the table, the decimal part is the specific correlation parameter value of the correlation matrix parameters between the first list and the second list, while the integer part is the statistical quantity corresponding to each training object. The stepped shaded part in the middle is the analyzed object alignment relationship, which is analyzed by the maximum specific correlation parameter value under each row and column.

[0137] As can be seen, this optional embodiment can analyze the object alignment relationship between each training sample object by calculating the correlation matrix parameters, thereby generating a first word combination sequence and a second word combination sequence with corresponding relationships. In this way, the reliability and accuracy of the obtained word combination sequences can be improved, and the encoding corresponding to each word combination sequence can be obtained reliably and accurately, so that the trained language coding model can be reliably and effectively applied.

[0138] In another optional embodiment, updating the encoding contribution value of the first phrase combination sequence and the encoding contribution value of the corresponding second phrase combination sequence in the previous training operation based on the first phrase combination sequence and the second phrase combination sequence includes:

[0139] Determine the statistical value of each first word group in the first word group combination sequence and the statistical value of each second word group in the second word group combination sequence. Based on the statistical values ​​of each first word group and each second word group, determine the statistical growth direction of each first word group and the statistical growth direction of each second word group.

[0140] Based on the statistical growth direction of each first word group and the statistical growth direction of each second word group, determine the word group code of each first word group and the word group code of each second word group, which serve as the coding sequence corresponding to the first word group combination sequence and the coding sequence corresponding to the second word group combination sequence.

[0141] Based on the encoding sequences corresponding to the first and second word combination sequences, update the encoding sequences corresponding to the first and second word combination sequences of the previous training operation.

[0142] Based on the updated encoding sequences of the first and second word combination sequences, calculate the encoding contribution value corresponding to the word code of each first word and the encoding contribution value corresponding to the word code of each second word.

[0143] Based on the encoding contribution values ​​corresponding to the word codes of all first word groups and the word codes of all second word groups, update the encoding contribution values ​​of the first word group combination sequence and the corresponding second word group combination sequence corresponding to the previous training operation.

[0144] In this optional embodiment, for example, such as Figure 4 As shown, Figure 4 This is a schematic diagram of the process for determining the word group code of a corresponding word group disclosed in an embodiment of the present invention. The quantity value in the circle is the statistical quantity corresponding to each word group. Based on the statistical growth direction of the statistical quantity corresponding to each word group (as shown by the arrow), the corresponding code can be determined. For example, the code corresponding to "actuator" is "100" and the code corresponding to "circuit" is "1011". At the same level, the word group with the smaller statistical quantity is located to the left of the leaf node.

[0145] Specifically, the coding contribution value corresponding to the word group code is:

[0146]

[0147] Among them, S xn is the code contribution value corresponding to the word group code. x T represents the frequency of occurrence corresponding to the word group encoding. x N represents the total number of word group codes corresponding to the word groups in all preset libraries, and C represents the number of all preset libraries.

[0148] As can be seen, this optional embodiment can determine the encoding sequence corresponding to each word combination sequence by the statistical growth direction of each word group, and calculate the encoding contribution value to update the encoding sequence and encoding contribution value corresponding to each word combination sequence in the previous training operation. This is beneficial to improving the reliability and accuracy of determining the encoding sequence corresponding to each word combination sequence, and also beneficial to improving the reliability and accuracy of calculating the encoding contribution value corresponding to each word combination sequence. This, in turn, is beneficial to improving the reliability, accuracy and effectiveness of updating the encoding sequence and encoding contribution value corresponding to each word combination sequence in each training process.

[0149] Example 2

[0150] Please see Figure 6 , Figure 6 This is a flowchart illustrating a method for intelligent language translation of automotive repair data, as disclosed in an embodiment of the present invention. Figure 6 The described intelligent language translation method for automotive repair data can be applied to translate data on various types of repair work containing common words or professional terms, such as vehicle repair, electrical repair, equipment repair, etc., and this embodiment of the invention is not limited thereto. Optionally, this method can translate different types of source languages ​​into other target languages ​​(e.g., translating Chinese or English into French), or it can translate source languages ​​into different types of target languages ​​(e.g., translating Chinese into French or English). Further optionally, this method can be implemented by a language translation system, which can be integrated into a language translation device, or it can be a local server or cloud server used to process the language translation process, etc., and this embodiment of the invention is not limited thereto. Figure 6 As shown, the intelligent language translation method for automotive repair data can include the following operations:

[0151] 201. Segment the source language statements that need to be translated into the target language to obtain the source language sequence corresponding to the source language statements.

[0152] 202. Using a pre-trained source language encoding model, perform encoding operations on each source language object to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence.

[0153] 203. Based on the source language code corresponding to each source language object, match the target language translation code tag grid corresponding to each source language object from the preset target language translation code tag library.

[0154] In this embodiment of the invention, the target language translation encoding tag grid includes multiple target language translation encoding tags that match the corresponding source language object, i.e., the encoding of the translated words that match the source language object. For example, such as Figure 1 As shown, Figure 1 This is a schematic diagram of a target language translation encoding tag grid corresponding to a source language object, as disclosed in an embodiment of the present invention. Figure 1 As shown, the target language translation encoding tag grid for “oxidation” includes “oil oxidation catalysis”, “oxidation catalysis”, “oil oxidation”, “oxidation”, “catalysis”, and “oxygen” (for ease of understanding, the corresponding codes have been converted into the source and target languages).

[0155] Furthermore, as an optional implementation, the method may further include:

[0156] After matching the target language translation encoding tag grid corresponding to each source language object, for each source language object, the first semantic feature information of the source language object in the source language sentence and the second semantic feature information of the source language object in other source language sentences adjacent to the source language sentence are analyzed.

[0157] Based on the first semantic feature information and the second semantic feature information, the relevance of all target language translation codes corresponding to the source language object is determined, and all target language translation codes with a relevance greater than or equal to a preset relevance threshold are identified from all target language translation codes, so as to update the target language translation code label grid corresponding to the source language object.

[0158] 204. Based on the source language encoding corresponding to each source language object, match multiple pending target language corpus alignment templates corresponding to each source language object from the preset target language corpus alignment library.

[0159] In this embodiment of the invention, the goal is to match and align templates for all pending target language corpora containing source language objects.

[0160] 205. For each source language object, based on the target language translation encoding label grid corresponding to the source language object, determine all the target language corpus alignment templates that match any target language translation encoding in the corresponding target language translation encoding label grid from all the target language corpus alignment templates corresponding to the source language object, and use them as the target language corpus alignment templates corresponding to the source language object.

[0161] In this embodiment of the invention, the goal is to determine all target language corpus alignment templates that have key translated words that match the source language object from all undetermined target language corpus alignment templates containing the source language object.

[0162] 206. Based on the alignment templates of all target language corpora corresponding to all source language objects, perform encoding and decoding operations on the source language encoding sequence corresponding to the source language sequence to obtain the target language decoding sequence corresponding to the source language encoding sequence, and perform decoding and translation conversion on the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence.

[0163] In this embodiment of the invention, for other descriptions of steps 201, 202 and 206, please refer to the detailed description of steps 101, 102 and 104 in Embodiment 1. These descriptions will not be repeated in this embodiment of the invention.

[0164] As can be seen, implementing the embodiments of the present invention can determine all target language corpus alignment templates corresponding to the source language object based on the target language translation encoding label grid that matches the source language object. In this way, not only can the amount of data analysis required by the language translation system be reduced, thereby accelerating the translation process from the source language to the target language, but also the reliability and accuracy of the determined target language corpus alignment templates can be improved, thereby enabling targeted and precise encoding and decoding operations to be performed on the encoding sequence corresponding to the source language sequence.

[0165] In an optional embodiment, step 206 above, which aligns templates with all target language corpora corresponding to all source language objects and performs encoding-decoding operations on the source language encoding sequence corresponding to the source language sequence to obtain the target language decoding sequence corresponding to the source language encoding sequence, includes:

[0166] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in the alignment template of each target language corpus corresponding to each source language object, determine the object position of the reference target language object corresponding to all reference source language objects in the alignment template of each target language corpus, and use it as the recombination position corresponding to all reference source language objects in the alignment template of each target language corpus;

[0167] For each reference source language object in the alignment template of each target language corpus, based on the recombination position corresponding to the reference source language object, determine the statistics of multiple decodes of the reference target language object that match the recombination position corresponding to the reference source language object;

[0168] Based on the statistics of each decoded code of the reference target language object that matches the recombination position corresponding to the reference source language object, the decoded codes with statistics greater than or equal to a preset statistical threshold are determined from all decoded codes and used as the target decoded codes of the reference target language object that match the recombination position corresponding to the reference source language object.

[0169] Based on the target decoding in the recombination position corresponding to each reference source language object in the determined alignment template of each target language corpus and the preset decoding contribution value calculation formula, calculate the target decoding contribution value corresponding to all reference source language objects in the alignment template of each target language corpus.

[0170] From all target language corpus alignment templates, target language output reconstruction templates with target decoding contribution values ​​greater than or equal to a preset contribution value threshold are determined. Based on the target decoding in the reconstruction position corresponding to each reference source language object in the target language output reconstruction template, the target language decoding sequence corresponding to the source language encoding sequence is determined.

[0171] In this optional embodiment, the preset decoding contribution value calculation formula is the encoding contribution value calculation formula corresponding to the word group encoding used in the training process of the language model to be trained. Specifically, the recombination positions corresponding to all reference source language objects in each target language corpus alignment template can be understood as the translation position of the target language decoding required for each source language object; and, since a source language object has multiple corresponding target language decodings, the calculation of the target decoding contribution value can be understood as the specific decoding of the target language that should be translated into for each source language object.

[0172] As can be seen, this optional embodiment can determine the target language decoding sequence corresponding to the source language encoding sequence based on the recombination positions of all reference source language objects in the target language corpus alignment template and the calculated target decoding contribution value. This can improve the reliability and accuracy of the determined target language decoding sequence, thereby improving the reliability, accuracy and effectiveness of subsequent conversion of the target language decoding sequence, and thus accurately obtaining the target language translation sequence.

[0173] In another optional embodiment, the step described above, determining the object positions of all reference source language objects in each target language corpus alignment template based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, as the recombination positions corresponding to all reference source language objects in each target language corpus alignment template, includes:

[0174] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in the alignment template of each target language corpus corresponding to each source language object, calculate the correlation degree between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template.

[0175] The reassembly order of each target language corpus alignment template is determined based on the degree of correlation between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template.

[0176] Based on the recombination order of each target language corpus alignment template, the preset positional relationship between each reference source language object and the corresponding reference target language object in each target language corpus alignment template, and the object position of each source language object in the source language encoding sequence corresponding to the source language sequence, the object positions of all reference source language objects in each target language corpus alignment template are determined sequentially, and these positions are used as the recombination positions of all reference source language objects in each target language corpus alignment template.

[0177] In this optional embodiment, the greater the correlation, the more similar the sequence content of the source language encoded sequence corresponding to the source language sequence is to the template content of the alignment template of the target language corpus. For example, such as Figure 2 As shown, each target language corpus alignment template is recombined in order of relevance, resulting in the recombined position shown in the Chinese section. The shaded part represents the content of the target language corpus alignment template that matches the sequence content of the source language encoding sequence (for ease of understanding, the corresponding encoding has been converted into the source and target languages).

[0178] As can be seen, this optional embodiment can sequentially and one by one reorganize the position of each target language corpus alignment template according to the correlation between the source language encoding sequence and each target language corpus alignment template. This is beneficial for reliably and accurately determining the target language decoding and translation position corresponding to each source language object, thereby improving the reliability and accuracy of the translation of the source language object, and enabling the translation of each source language sentence to proceed smoothly.

[0179] Example 3

[0180] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a language intelligent translation device for automotive repair data disclosed in an embodiment of the present invention. Figure 7 As shown, the intelligent language translation device for automotive repair data may include:

[0181] The segmentation processing module 301 is used to segment the source language sentences that need to be translated into the target language to obtain the source language sequence corresponding to the source language sentences;

[0182] Encoding module 302 is used to perform encoding operations on each source language object through a pre-trained source language encoding model to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence.

[0183] The determination module 303 is used to determine multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object.

[0184] The encoding / decoding module 304 is used to perform encoding / decoding operations on the source language encoding sequence corresponding to the source language sequence according to the template alignment of all target language corpora corresponding to all source language objects, and to obtain the target language decoding sequence corresponding to the source language encoding sequence.

[0185] The decoding module 305 is used to perform decoding and translation on the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence.

[0186] In this embodiment of the invention, the source language sequence includes at least one source language object arranged in a preset order, and the source language object is a source language character or source language word; each target language corpus alignment template includes multiple reference source language objects and a reference target language object corresponding to each reference source language object, and there is a preset positional relationship between each reference source language object and the corresponding reference target language object.

[0187] It is evident that implementation Figure 7 The described intelligent language translation device for automotive repair documents can intelligently convert the source language into the target language, eliminating the need for time-consuming step-by-step conversion. While ensuring grammatical accuracy in each language during the translation process, it improves translation efficiency, thereby facilitating rapid vehicle repair based on the translated documents and ensuring smooth operation of the repair work.

[0188] In an optional embodiment, the device further includes:

[0189] The matching module 306 is used to match the target language translation code tag grid corresponding to each source language object from the preset target language translation code tag library according to the source language code corresponding to each source language object before the determining module 303 determines multiple target language corpus alignment templates corresponding to each source language object from the preset target language corpus alignment library according to the source language code corresponding to each source language object.

[0190] Specifically, the method by which the determining module 303 determines multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object is as follows:

[0191] Based on the source language encoding corresponding to each source language object, multiple undetermined target language corpus alignment templates are matched from the preset target language corpus alignment library for each source language object;

[0192] For each source language object, based on the target language translation encoding label grid corresponding to the source language object, determine all the target language corpus alignment templates that match any target language translation encoding in the corresponding target language translation encoding label grid from all the target language corpus alignment templates corresponding to the source language object, and use them as the target language corpus alignment templates corresponding to the source language object.

[0193] In this optional embodiment, the target language translation encoding tag grid includes multiple target language translation encoding tags that match the corresponding source language objects.

[0194] It is evident that implementation Figure 8 The described intelligent language translation device for automotive repair data can determine the alignment templates of all target language corpora corresponding to the source language object based on the target language translation encoding label grid that matches the source language object. This not only reduces the amount of data analysis required by the language translation system, thereby accelerating the translation process from the source language to the target language, but also improves the reliability and accuracy of the determined target language corpus alignment templates, thus enabling targeted and precise encoding and decoding operations on the encoding sequences corresponding to the source language sequences.

[0195] In another optional embodiment, the encoding / decoding module 304 performs encoding / decoding operations on the source language encoding sequence corresponding to the source language sequence based on the alignment templates of all target language corpora corresponding to all source language objects, and obtains the target language decoding sequence corresponding to the source language encoding sequence in the following specific way:

[0196] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in the alignment template of each target language corpus corresponding to each source language object, determine the object position of the reference target language object corresponding to all reference source language objects in the alignment template of each target language corpus, and use it as the recombination position corresponding to all reference source language objects in the alignment template of each target language corpus;

[0197] For each reference source language object in the alignment template of each target language corpus, based on the recombination position corresponding to the reference source language object, determine the statistics of multiple decodes of the reference target language object that match the recombination position corresponding to the reference source language object;

[0198] Based on the statistics of each decoded code of the reference target language object that matches the recombination position corresponding to the reference source language object, the decoded codes with statistics greater than or equal to a preset statistical threshold are determined from all decoded codes and used as the target decoded codes of the reference target language object that match the recombination position corresponding to the reference source language object.

[0199] Based on the target decoding in the recombination position corresponding to each reference source language object in the determined alignment template of each target language corpus and the preset decoding contribution value calculation formula, calculate the target decoding contribution value corresponding to all reference source language objects in the alignment template of each target language corpus.

[0200] From all target language corpus alignment templates, target language output reconstruction templates with target decoding contribution values ​​greater than or equal to a preset contribution value threshold are determined. Based on the target decoding in the reconstruction position corresponding to each reference source language object in the target language output reconstruction template, the target language decoding sequence corresponding to the source language encoding sequence is determined.

[0201] It is evident that implementation Figure 8 The described intelligent language translation device for automotive repair data can determine the target language decoding sequence corresponding to the source language encoding sequence based on the recombination positions of all reference source language objects in the target language corpus alignment template and the calculated target decoding contribution value. This can improve the reliability and accuracy of the determined target language decoding sequence, thereby improving the reliability, accuracy and effectiveness of subsequent conversion of the target language decoding sequence, and thus accurately obtaining the target language translation sequence.

[0202] In another optional embodiment, the encoding / decoding module 305 determines the object positions of all referenced source language objects in each target language corpus alignment template based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object. The specific method for determining the recombination positions of all referenced source language objects in each target language corpus alignment template is as follows:

[0203] Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in the alignment template of each target language corpus corresponding to each source language object, calculate the correlation degree between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template.

[0204] The reassembly order of each target language corpus alignment template is determined based on the degree of correlation between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template.

[0205] Based on the recombination order of each target language corpus alignment template, the preset positional relationship between each reference source language object and the corresponding reference target language object in each target language corpus alignment template, and the object position of each source language object in the source language encoding sequence corresponding to the source language sequence, the object positions of all reference source language objects in each target language corpus alignment template are determined sequentially, and these positions are used as the recombination positions of all reference source language objects in each target language corpus alignment template.

[0206] In this optional embodiment, the greater the correlation, the more similar the sequence content of the source language encoding sequence corresponding to the source language sequence is to the template content of the alignment template of the target language corpus.

[0207] It is evident that implementation Figure 8 The described intelligent language translation device for automotive repair data can sequentially and one by one reorganize the position of each target language corpus alignment template based on the correlation between the source language encoding sequence and each target language corpus alignment template. This facilitates the reliable and accurate determination of the target language decoding translation position corresponding to each source language object, thereby improving the reliability and accuracy of the translation of the source language object and enabling the translation of each source language sentence to proceed smoothly.

[0208] In yet another optional embodiment, the segmentation processing module 301 is further configured to:

[0209] Before segmenting the source language statement to be translated into the target language to obtain the source language sequence corresponding to the source language statement, the source language training sample set and the target language training sample set corresponding to the source language training sample set in the target training sample set are segmented to obtain the first list corresponding to the source language training sample set and the second list corresponding to the target language training sample set.

[0210] The device also includes:

[0211] Training module 307 is used to perform multiple training operations on a preset language model to be trained using a first list and a second list.

[0212] The sequence generation module 308 is used to generate, in each training operation, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set, based on the first list and the second list.

[0213] The update module 309 is used to update the encoding contribution value of the first word combination sequence and the encoding contribution value of the second word combination sequence corresponding to the previous training operation based on the first word combination sequence and the second word combination sequence.

[0214] In this optional embodiment, both the first list and the second list include a number of corresponding training sample objects, which are training sample characters or training sample words.

[0215] It is evident that implementation Figure 8 The described intelligent language translation device for automotive repair data can intelligently train the language model to be trained using a source language training sample set and a corresponding target language training sample set to obtain a language encoding model that can be applied to the encoding of source language sequences. This allows for the reliable and accurate acquisition of the language encoding model, which not only improves the efficiency of subsequent encoding of source language sequences, thus quickly obtaining the encoding sequence corresponding to the source language sequence, but also enhances the reliability and accuracy of subsequent encoding of source language sequences, thereby accurately obtaining the encoding sequence corresponding to the source language sequence.

[0216] In another optional embodiment, the sequence generation module 308 generates a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set based on the first list and the second list, specifically as follows:

[0217] Calculate the parameters of the association matrix between the first list and the second list;

[0218] Based on the correlation matrix parameters, analyze the object alignment relationship between all training sample objects corresponding to the first list and all training sample objects corresponding to the second list.

[0219] Based on the object alignment relationship, generate the first word combination sequence corresponding to the source language training sample set and the second word combination sequence corresponding to the target language training sample set.

[0220] In this optional embodiment, the association matrix parameters between the first list and the second list are:

[0221]

[0222] ceM(i,j) is the parameter of the correlation matrix, p(a i p(b) represents the probability of occurrence of the training object corresponding to the first list. j p(a) represents the probability of occurrence of the training object corresponding to the second list. i │b j p(b) represents the probability of the first list's corresponding training object appearing given that the second list's corresponding training object appears.j │a i ) represents the probability of the second training object appearing given that the training object in the first list appears.

[0223] It is evident that implementation Figure 8 The language intelligent translation device for automotive repair data described herein can analyze the object alignment relationship between each training sample object by calculating the correlation matrix parameters, thereby generating a first word group combination sequence and a second word group combination sequence with corresponding relationships. This can improve the reliability and accuracy of the obtained word group combination sequences, and thus reliably and accurately obtain the encoding corresponding to each word group combination sequence, so that the trained language encoding model can be reliably and effectively applied.

[0224] In another optional embodiment, the update module 309 updates the encoding contribution value of the first phrase combination sequence and the encoding contribution value of the corresponding second phrase combination sequence based on the first phrase combination sequence and the second phrase combination sequence in the specific way as follows:

[0225] Determine the statistical value of each first word group in the first word group combination sequence and the statistical value of each second word group in the second word group combination sequence. Based on the statistical values ​​of each first word group and each second word group, determine the statistical growth direction of each first word group and the statistical growth direction of each second word group.

[0226] Based on the statistical growth direction of each first word group and the statistical growth direction of each second word group, determine the word group code of each first word group and the word group code of each second word group, which serve as the coding sequence corresponding to the first word group combination sequence and the coding sequence corresponding to the second word group combination sequence.

[0227] Based on the encoding sequences corresponding to the first and second word combination sequences, update the encoding sequences corresponding to the first and second word combination sequences of the previous training operation.

[0228] Based on the updated encoding sequences of the first and second word combination sequences, calculate the encoding contribution value corresponding to the word code of each first word and the encoding contribution value corresponding to the word code of each second word.

[0229] Based on the encoding contribution values ​​corresponding to the word codes of all first word groups and the word codes of all second word groups, update the encoding contribution values ​​of the first word group combination sequence and the corresponding second word group combination sequence corresponding to the previous training operation.

[0230] In this optional embodiment, the encoding contribution value corresponding to the phrase encoding of the corresponding phrase is:

[0231]

[0232] S x n is the code contribution value corresponding to the word group code. x T represents the frequency of occurrence corresponding to the word group encoding. x N represents the total number of word group codes corresponding to the word groups in all preset libraries, and C represents the number of all preset libraries.

[0233] It is evident that implementation Figure 8 The intelligent language translation device for automotive repair data described herein can determine the encoding sequence corresponding to each word combination sequence by analyzing the statistical growth direction of each word group, and calculate the encoding contribution value to update the encoding sequence and encoding contribution value corresponding to each word combination sequence in the previous training operation. This helps to improve the reliability and accuracy of determining the encoding sequence corresponding to each word combination sequence, and also helps to improve the reliability and accuracy of calculating the encoding contribution value corresponding to each word combination sequence. As a result, it helps to improve the reliability, accuracy and effectiveness of updating the encoding sequence and encoding contribution value corresponding to each word combination sequence in each training process.

[0234] Example 4

[0235] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of another intelligent language translation device for automotive repair data disclosed in an embodiment of the present invention. (See diagram below.) Figure 9 As shown, the intelligent language translation device for automotive repair data may include:

[0236] Memory 401 storing executable program code;

[0237] Processor 402 coupled to memory 401;

[0238] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the intelligent language translation method for automobile repair data described in Embodiment 1 or Embodiment 2 of the present invention.

[0239] Example 5

[0240] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute the steps in the intelligent language translation method for automotive repair data described in Embodiment 1 or Embodiment 2 of this invention.

[0241] Example 6

[0242] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the intelligent language translation method for automotive repair data described in Embodiment 1 or Embodiment 2.

[0243] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0244] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0245] Finally, it should be noted that the language intelligent translation method and device for automobile repair data disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent translation of automotive repair data, characterized in that, The method includes: The source language statement that needs to be translated into the target language is segmented to obtain the source language sequence corresponding to the source language statement; the source language sequence includes at least one source language object arranged in a preset order, and the source language object is a source language character or source language word; By using a pre-trained source language encoding model, an encoding operation is performed on each source language object to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence. Based on the source language code corresponding to each source language object, a target language translation code tag grid corresponding to each source language object is matched from a preset target language translation code tag library; the target language translation code tag grid includes multiple target language translation code tags that match the corresponding source language object; Based on the source language code corresponding to each source language object, multiple target language corpus alignment templates corresponding to each source language object are determined from a preset target language corpus alignment library; each target language corpus alignment template includes multiple reference source language objects and a reference target language object corresponding to each reference source language object, and there is a preset positional relationship between each reference source language object and its corresponding reference target language object; Based on the alignment templates of all target language corpora corresponding to all source language objects, an encoding and decoding operation is performed on the source language encoding sequence corresponding to the source language sequence to obtain the target language decoding sequence corresponding to the source language encoding sequence, and a decoding-translation conversion is performed on the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence. The method further includes: After matching the target language translation encoding tag grid corresponding to each source language object, for each source language object, the first semantic feature information of the source language object in the source language sentence and the second semantic feature information of the source language object in other source language sentences adjacent to the source language sentence are analyzed; For each source language object, the relevance of all target language translation codes corresponding to the source language object is determined based on the first semantic feature information and the second semantic feature information of the source language object. Then, all target language translation codes with a relevance greater than or equal to a preset relevance threshold are determined from all the target language translation codes to update the target language translation code label grid corresponding to the source language object.

2. The intelligent language translation method for automotive repair data according to claim 1, characterized in that, The step of determining multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object includes: Based on the source language encoding corresponding to each source language object, multiple undetermined target language corpus alignment templates corresponding to each source language object are matched from a preset target language corpus alignment library; For each source language object, based on the target language translation encoding label grid corresponding to the source language object, all target language corpus alignment templates that match any target language translation encoding in the corresponding target language translation encoding label grid are determined from all the target language corpus alignment templates corresponding to the source language object, and are used as the target language corpus alignment templates corresponding to the source language object.

3. The intelligent language translation method for automotive repair data according to claim 1 or 2, characterized in that, The step of performing encoding-decoding operations on the source language encoding sequence corresponding to the source language sequence based on the alignment templates of all the target language corpora corresponding to all the source language objects, to obtain the target language decoding sequence corresponding to the source language encoding sequence, includes: Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, determine the object position of the reference target language object corresponding to all reference source language objects in each target language corpus alignment template, and use it as the recombination position corresponding to all reference source language objects in each target language corpus alignment template; For each reference source language object in each target language corpus alignment template, based on the recombination position corresponding to the reference source language object, determine the statistical quantity of multiple decodes of the reference target language object that match the recombination position corresponding to the reference source language object; Based on the statistics of each decoded code of the reference target language object that matches the recombination position corresponding to the reference source language object, the decoded codes whose statistics are greater than or equal to a preset statistical threshold are determined from all the decoded codes and used as the target decoded codes of the reference target language object that match the recombination position corresponding to the reference source language object. Based on the target decoding in the recombination position corresponding to each reference source language object in each target language corpus alignment template and the preset decoding contribution value calculation formula, calculate the target decoding contribution value corresponding to all reference source language objects in each target language corpus alignment template; From all the target language corpus alignment templates, determine the target language output reconstruction template whose target decoding contribution value is greater than or equal to a preset contribution value threshold, and determine the target language decoding sequence corresponding to the source language encoding sequence based on the target decoding in the reconstruction position corresponding to each reference source language object in the target language output reconstruction template.

4. The intelligent language translation method for automotive repair data according to claim 3, characterized in that, The step of determining the object positions of all reference source language objects in each target language corpus alignment template based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, as the recombination positions corresponding to all reference source language objects in each target language corpus alignment template, includes: Based on the sequence content of the source language encoding sequence corresponding to the source language sequence and the template content in each target language corpus alignment template corresponding to each source language object, the correlation degree between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template is calculated; the greater the correlation degree, the more similar the sequence content of the source language encoding sequence corresponding to the source language sequence is to the template content of the target language corpus alignment template. The recombination order of each target language corpus alignment template is determined based on the degree of correlation between the source language encoding sequence corresponding to the source language sequence and each target language corpus alignment template. Based on the recombination order of each target language corpus alignment template, the preset positional relationship between each reference source language object and the corresponding reference target language object in each target language corpus alignment template, and the object position of each source language object in the source language encoding sequence corresponding to the source language sequence, the object positions of all reference source language objects in each target language corpus alignment template are determined sequentially, and these positions are used as the recombination positions of all reference source language objects in each target language corpus alignment template.

5. The intelligent language translation method for automotive repair data according to claim 1 or 2, characterized in that, Before segmenting the source language statement to be translated into the target language to obtain the source language sequence corresponding to the source language statement, the method further includes: The source language training sample set and the target language training sample set corresponding to the source language training sample set are segmented to obtain a first list corresponding to the source language training sample set and a second list corresponding to the target language training sample set; both the first list and the second list include a number of corresponding training sample objects, wherein the training sample objects are training sample characters or training sample words. Multiple training operations are performed on the preset language model to be trained using the first list and the second list. During each training operation, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set are generated based on the first list and the second list. The encoding contribution value of the first word combination sequence and the encoding contribution value of the second word combination sequence corresponding to the previous training operation are updated based on the first word combination sequence and the second word combination sequence.

6. The intelligent language translation method for automotive repair data according to claim 5, characterized in that, The step of generating a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set based on the first list and the second list includes: Calculate the association matrix parameters between the first list and the second list; Based on the correlation matrix parameters, analyze the object alignment relationship between all training sample objects corresponding to the first list and all training sample objects corresponding to the second list; Based on the object alignment relationship, a first word combination sequence corresponding to the source language training sample set and a second word combination sequence corresponding to the target language training sample set are generated. The association matrix parameter between the first list and the second list is: ； ceM(i,j) is the parameter of the correlation matrix, p(a i p(b) represents the probability of occurrence of the training object corresponding to the first list. j p(a) represents the probability of occurrence of the training object corresponding to the second list. i │b j p(b) represents the probability of the first list corresponding to the training object appearing given that the training object corresponding to the second list appears. j │a i ) represents the probability of the second list corresponding to the training object appearing given that the training object corresponding to the first list appears.

7. The intelligent language translation method for automotive repair data according to claim 6, characterized in that, The step of updating the encoding contribution value of the first phrase combination sequence and the encoding contribution value of the corresponding second phrase combination sequence based on the first phrase combination sequence and the second phrase combination sequence in the previous training operation includes: Determine the statistical value of each first word group in the first word group combination sequence and the statistical value of each second word group in the second word group combination sequence, and determine the statistical growth direction of each first word group and the statistical growth direction of each second word group based on the statistical value of each first word group and the statistical value of each second word group. Based on the statistical growth direction of each first word group and the statistical growth direction of each second word group, determine the word group code of each first word group and the word group code of each second word group, which serve as the encoding sequence corresponding to the first word group combination sequence and the encoding sequence corresponding to the second word group combination sequence. Based on the encoding sequence corresponding to the first word combination sequence and the encoding sequence corresponding to the second word combination sequence, update the encoding sequence corresponding to the first word combination sequence and the encoding sequence corresponding to the second word combination sequence in the previous training operation. Based on the updated encoding sequence of the first phrase combination sequence and the updated encoding sequence of the second phrase combination sequence, calculate the encoding contribution value corresponding to the phrase encoding of each first phrase and the encoding contribution value corresponding to the phrase encoding of each second phrase. Based on the encoding contribution values ​​corresponding to the word codes of all the first word groups and the encoding contribution values ​​corresponding to the word codes of all the second word groups, update the encoding contribution values ​​of the first word group combination sequence and the encoding contribution values ​​of the corresponding second word group combination sequence corresponding to the previous training operation; Among them, the coding contribution value corresponding to the word group code is: ； S x n is the code contribution value corresponding to the word group code. x T represents the frequency of occurrence corresponding to the word group encoding. x N represents the total number of word group codes corresponding to the word groups in all preset libraries, and C represents the number of all preset libraries.

8. A language intelligent translation device for automotive repair data, characterized in that, The apparatus is used to perform the intelligent language translation method for automotive repair data as described in any one of claims 1-7, and the apparatus comprises: The segmentation module is used to segment the source language statement that needs to be translated into the target language to obtain the source language sequence corresponding to the source language statement; the source language sequence includes at least one source language object arranged in a preset order, and the source language object is a source language character or a source language word; The encoding module is used to perform encoding operations on each source language object using a pre-trained source language encoding model to obtain the source language encoding corresponding to each source language object, which serves as the source language encoding sequence corresponding to the source language sequence. The matching module is used to match the target language translation encoding tag grid corresponding to each source language object from a preset target language translation encoding tag library based on the source language encoding corresponding to each source language object; the target language translation encoding tag grid includes multiple target language translation encoding tags that match the corresponding source language object; The determining module is used to determine multiple target language corpus alignment templates corresponding to each source language object from a preset target language corpus alignment library based on the source language encoding corresponding to each source language object; each target language corpus alignment template includes multiple reference source language objects and a reference target language object corresponding to each reference source language object, and there is a preset positional relationship between each reference source language object and the corresponding reference target language object; The encoding / decoding module is used to perform encoding / decoding operations on the source language encoding sequence corresponding to the source language sequence according to the alignment template of all the target language corpora corresponding to all the source language objects, so as to obtain the target language decoding sequence corresponding to the source language encoding sequence. The decoding and translation module is used to perform decoding and translation on the target language decoding sequence corresponding to the source language encoding sequence to obtain the target language translation sequence corresponding to the source language encoding sequence; Furthermore, the matching module is also used for: After matching the target language translation encoding tag grid corresponding to each source language object, for each source language object, the first semantic feature information of the source language object in the source language sentence and the second semantic feature information of the source language object in other source language sentences adjacent to the source language sentence are analyzed; For each source language object, the relevance of all target language translation codes corresponding to the source language object is determined based on the first semantic feature information and the second semantic feature information of the source language object. Then, all target language translation codes with a relevance greater than or equal to a preset relevance threshold are determined from all the target language translation codes to update the target language translation code label grid corresponding to the source language object.

9. A language intelligent translation device for automotive repair data, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the intelligent language translation method for automotive repair data as described in any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the intelligent language translation method for automotive repair data as described in any one of claims 1-7.

Images