Method for translating texts in batches
By building an automotive translation vocabulary database and a deep learning language model, and optimizing the machine translation engine, the problems of low efficiency and poor quality in automotive industry text translation have been solved, achieving efficient and accurate batch translation.
Patent Information
- Application Number
- CN202510883040.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-28
- Publication Date
- 2025-11-07
AI Technical Summary
Existing technologies for text translation in the automotive industry are inefficient and of poor quality. Human translation is time-consuming and labor-intensive, while machine translation is inaccurate, especially in the translation of technical terms, which affects production and sales.
We build an automotive translation vocabulary database, optimize the machine translation engine, perform multi-round checks using a deep learning language model, standardize text format and set translation priorities, and process files using a batch translation module.
It improves the efficiency and quality of text translation in the automotive industry, ensures the accuracy of technical terms, reduces errors, and meets the diverse needs of enterprises.
Smart Images
Figure CN120911486A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of automobiles, in particular to a method for batch translation of text. BACKGROUND
[0002] In the globalized automobile industry environment, automobile enterprises need to deal with a large number of technical documents, user manuals, market promotion materials and other texts from different language regions. When automobile manufacturers introduce advanced technology from abroad, they need to translate the relevant technical documents into Chinese so that domestic engineers can understand and apply them. When exporting automobiles, product descriptions and other materials need to be translated into the target market language. Currently, the existing translation methods mainly have the following problems: first, manual translation is inefficient. In the face of a large amount of text, manual translation takes a long time and costs a lot, and is prone to human error. Second, although existing machine translation software can quickly translate, it is not accurate for the translation of automobile industry professional terms. The automobile industry has a large number of specific professional terms and industry expression habits, and general translation software is difficult to accurately match, resulting in poor translation quality and affecting the production, sales and technical exchange of automobile enterprises. In view of this, we propose a method for batch translation of text. SUMMARY
[0003] To solve the above technical problems, a method for batch translation of text is provided, which solves the problems of low translation efficiency and poor translation quality.
[0004] To achieve the above purpose, the technical scheme adopted by the present application is as follows: a method for batch translation of text, the translation steps are:
[0005] S1, obtaining automobile industry professional terms to construct an automobile translation vocabulary bank;
[0006] S2, performing format unification, recognizing title information and error correction processing on the text to be translated, and judging the content characteristics of the text;
[0007] S3, selecting an optimal machine translation engine according to the characteristics of the text to be translated and the target language, optimizing the selected translation engine, connecting the automobile translation vocabulary bank with the translation engine, and prioritizing the professional vocabulary of the automobile translation vocabulary bank in the translation process of the translation engine for translation processing;
[0008] S4, based on the language model, checking the coherence and correctness of the translated text, using a second selected translation engine for re-translation when there is an error translation, and re-checking until there is no error;
[0009] S5, based on the batch translation module, batch importing a plurality of files to be translated, setting translation task parameters including target language and translation priority, and performing translation processing on the text in sequence according to the above steps, and displaying the task progress and translation results in real time.
[0010] Preferably, the automobile professional terms obtained in the S1 step are obtained by consulting automobile industry materials, browsing automobile brand official websites, industry information platforms, and communicating with automobile engineers and maintenance technicians in the industry.
[0011] When building the automobile translation vocabulary library, the obtained terms are classified, including power system, vehicle body structure, and electronic device category. For each term, preliminary translation is performed through professional translation software, manual correction is performed in combination with dictionaries and industry translation materials, and annotations are added to each term to explain the meaning and application scenarios. The vocabulary library is built using Excel tools and is updated regularly to include newly emerging industry terms.
[0012] Preferably, in the S2 step, the text font and size are pre-specified, and the paragraph format is unified. The title information is identified by preliminarily screening the title based on the title format characteristics and identifying the title information based on the text content logic. The spelling errors are corrected based on the spelling check tool, and the syntax check tool finds the sentence structure and word collocation syntax problems.
[0013] Preferably, in the S2 step, the text content characteristics are determined by word frequency analysis. The high-frequency words in the text are extracted, and the content characteristics of the text are determined based on the high-frequency words. The word frequency analysis sorts the word frequency list from high to low by counting the frequency of each word in the text. The formula is:
[0014]
[0015] wherein TF i，j is the word frequency of word i in document j, n i，j is the number of times the word appears in document j, is the total number of times all words appear in document j. The high-frequency words in the document are calculated by the formula.
[0016] Preferably, in the S3 step, based on the text feature analysis result and the target language for translation, the best translation engine is selected from the machine translation engine. After selecting the translation engine, the parameter settings of the engine are adjusted, including the language model and translation preferences. In translation, the translation engine preferentially retrieves and matches professional terms in the automobile translation vocabulary library. When a matching term is identified in the vocabulary library, the translation is processed according to the established translation rules and optimized settings. The matching term is represented as for each word w in the text to be translated, a search is performed in the automobile translation vocabulary library v. If , the translation result T(w) = v(w), otherwise, T(w) is determined according to other translation rules, wherein T(w) represents the final translation result of the word w.
[0017] Preferably, the language model checks the coherence and correctness of the text in the S4 step by a deep learning language model, which captures the contextual relationship between words in a sentence through a self-attention mechanism, evaluates the rationality of the sentence, is trained based on a masked language model task, and in training, BERT randomly masks some words in the sentence and predicts these masked words. When checking the translated text, the text is input into the pre-trained BERT model, and for each word in the text, the model calculates its prediction probability distribution in the given context. If the prediction probability of a word in its context is low, it means that the word is not coherent or correct in the context. Specifically, for each word wi in the sentence, the model outputs a probability distribution P(w, a) on the vocabulary, where a is the rest of the sentence except wi. If the probability value of the actual wi is lower than a certain threshold, it is considered that there is an error at the position of the word.
[0018] Preferably, when the translated text is checked by the language model in the S4 step, and it is found that there is an error in the translation, the secondary translation engine intervention mechanism is started, the sentence fragment marked as incorrect by the language model is identified and extracted, and transmitted to the pre-selected secondary translation engine. The text with problems is retranslated, the corpus resources are combined, the original text is reinterpreted and translated, and the language model is checked again. As in the first check, it is considered from the aspects of sentence coherence, grammatical correctness and semantic accuracy. If errors are still detected, the secondary translation engine is called again for a new round of translation according to the established process, and the check is performed again. This cycle is repeated until the language model confirms that there is no coherence or correctness problem in the translated text.
[0019] Preferably, in the S5 step, when the user processes a large number of files to be translated, a plurality of files are imported into the system in batches, and the system supports formats including PDF, DOCX, PPTX and TXT. After importing the files, the user sets priorities for different files according to the importance and urgency of the files, and translates the text in sequence. Resources are automatically allocated, and translation tasks are allocated to different translation engines, and the translation status is updated in real time.
[0020] Compared with the prior art, the present application has the following advantages:
[0021] The present application collects, screens, arranges, stores and updates multi-source data to ensure the accuracy of the terminology, uses deep learning language models and rule evaluation, and realizes efficient and accurate batch translation of automobile industry text through multiple refined steps, professional terminology libraries and various technical means, improves translation quality and processing efficiency, meets the needs of automobile enterprises in various aspects, improves the efficiency of batch translation, reduces translation errors and inaccurate translation, and improves translation quality. BRIEF DESCRIPTION OF DRAWINGS
[0022] Figure 1 A translation step flow chart is provided for the present application. DETAILED DESCRIPTION
[0023] The following description is provided to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only examples, and other obvious modifications can be made by those skilled in the art.
[0024] Referring to Figure 1 As shown, a method for batch translating text, the translation step is:
[0025] S1, obtaining professional terms in the automobile industry, and constructing an automobile translation vocabulary library;
[0026] S2, performing format unification, recognizing title information, and error correction processing on the text to be translated, and judging the content characteristics of the text;
[0027] S3, according to the characteristics of the text to be translated and the target language, selecting an optimal machine translation engine, optimizing the selected translation engine, and connecting the automobile translation vocabulary library with the translation engine. The professional vocabulary of the automobile translation vocabulary library is preferentially matched in the translation process of the translation engine, and the translation processing is performed;
[0028] S4, for the translated text, based on the language model, checking the coherence and correctness of the text, if there is an error translation, using the second selected translation engine to translate again, and checking again until there is no error;
[0029] S5, based on the batch translation module, a plurality of files to be translated are batch imported, and translation task parameters are set, including target language and translation priority, and the text is translated in sequence according to the above steps, and the task progress and translation result are displayed in real time.
[0030] The automobile industry terminology in the present application is complex and professional, such as "turbocharging" and "dual clutch transmission". The professional vocabulary library can provide accurate term correspondence for translation, avoid errors caused by broad interpretation of the general vocabulary library, ensure the accuracy of the translation in the professional context, and unify the vocabulary library to make the translation of the same automobile terminology consistent by different translators, which is beneficial to form standardized industry translation, facilitate communication and reference among industry insiders, and improve the stability of the overall automobile industry translation quality;
[0031] The unified format can reduce the processing obstacles caused by format differences in the translation process, such as different fonts and paragraph spacing problems. Recognizing the title information helps to quickly grasp the text structure and key points, and the error correction processing excludes errors in the text in advance, so that the translation engine can run more smoothly, and the overall translation time is shortened;
[0032] Prioritize the matching of automobile translation vocabulary library, so that the translation engine can call accurate professional vocabulary in the first time when processing automobile text, greatly improve the accuracy of professional term translation, and output translation text conforming to industry standards;
[0033] The language model can deeply check the translation text from the aspects of coherence and correctness, check whether the sentence conforms to the grammar rules and the semantics is coherent, and through repeated correction, the translation errors are minimized to provide high-quality and flawless translation text for users. The batch import function allows multiple files to be processed at a time, saving time for importing and translating one by one.
[0034] In the S1 step, the automobile professional terms are obtained by referring to automobile industry materials, browsing automobile brand official websites, industry information platforms, and communicating with automobile engineers and maintenance technicians in the industry.
[0035] When building the automobile translation vocabulary library, the obtained terms are classified, including power system, vehicle body structure, and electronic device category. For each term, preliminary translation is performed through professional translation software, manual correction is performed in combination with dictionaries and industry translation materials, and annotations are added to each term to explain the meaning and application scenarios. The vocabulary library is built using Excel tools and is updated regularly to include new terms in the industry.
[0036] Referring to automobile industry materials can obtain systematic and authoritative professional terms, such as term definitions in automobile engineering textbooks and industry standard documents, which are accurate and standardized. Browsing automobile brand official websites can help understand the technical terms and product description vocabulary specific to each brand, such as some exclusive terms in the automatic driving field of Tesla. Industry information platforms can provide real-time and cutting-edge terms reflecting the current hotspots and trends in the automobile industry, such as the latest terms in new energy automobile battery technology. Communicating with automobile engineers and maintenance technicians can obtain commonly used colloquial and practical terms in practice, which enriches the diversity of term sources.
[0037] The terms are divided into power system, vehicle body structure, and electronic device category, like classifying and placing books in a library. When translating related texts, translators can quickly locate the required term category, improving the search efficiency. For example, when translating automobile engine-related documents, terms can be directly searched in the power system category, saving time and improving translation speed.
[0038] The automobile industry develops rapidly, and new terms emerge continuously. Regularly updating the vocabulary library can ensure that it reflects the latest industry trends, such as new terms in the new energy automobile and intelligent driving fields. This makes the translation work always keep pace with the industry development, ensuring the timeliness and professionalism of the translation, and providing strong support for information exchange and technology dissemination in the automobile industry.
[0039] In the S2 step, the font and size of the pre-standardized text are specified, and the paragraph format is unified; the title information is identified by preliminarily screening the title based on the title format characteristics, and the title information is identified in combination with the text content logic; the error is corrected based on the spelling check tool to correct the spelling errors of words, and the grammar check tool finds the syntax problems of sentence structure and word collocation.
[0040] The unified font and size of the present application can avoid visual interference caused by format differences in the translation process, so that the translator can focus more on the text content itself. The standardized paragraph format, such as uniform line spacing and indentation, can make the text layout clear, facilitate the translator to quickly locate and understand the content, reduce the reading obstacles caused by format chaos, and thus greatly improve the translation speed. In processing long automobile technical documents, the clear format can quickly find the starting position of each chapter for the translator, saving time.
[0041] In the S2 step, the text content characteristics are determined by word frequency analysis method. The high-frequency words in the text are extracted, and the content characteristics of the text are determined in combination with the high-frequency words. The word frequency analysis method sorts the word frequency list from high to low by counting the frequency of each word in the text. The calculation formula is:
[0042]
[0043] wherein TF i，j is the word frequency of word i in document j, n i，j is the number of times the word appears in document j, is the total number of times all words appear in document j; the high-frequency words in the document are calculated by the formula.
[0044] The present application calculates the high-frequency words by the word frequency analysis method, so that the translator can quickly understand the focus of the text. In an automobile technical report, if the words "battery", "mileage", and "charging speed" appear frequently, the translator can intuitively judge that the text focuses on the energy-related problems of electric vehicles, laying a foundation for accurate translation. This helps the translator to focus on the key content during translation and avoid deviating from the core due to long text length and complex content;
[0045] After determining the high-frequency words, the translator can adopt different translation strategies for different words. For high-frequency professional words, more effort can be invested in accurate translation, and professional materials and industry standards can be referred to to ensure the accuracy of the translation. After determining the content characteristics of the text based on the high-frequency words, the translator can select translation resources accordingly.
[0046] In S3, based on the text feature analysis result and the target language of translation, the best translation engine is selected from the machine translation engine. After selecting the translation engine, the parameter settings of the engine are adjusted, including the language model and translation preference. In translation, the translation engine preferentially searches and matches the professional vocabulary in the automobile translation vocabulary library. When the matching vocabulary is identified, the translation is processed according to the established translation rules and the optimized settings. The matching vocabulary is represented as follows: for each word w in the text to be translated, search in the automobile translation vocabulary library v. If , the translation result T(w) = v(w), otherwise, determine T(w) according to other translation rules, where T(w) represents the final translation result of the word w.
[0047] Through the analysis of text features such as the field to which the text belongs and the style of the text, the most suitable translation engine can be selected. For automobile professional technical papers, the translation engine that is good at processing academic and professional texts is more suitable, which can accurately grasp complex terminology and logical structure, and ensure that the translated text accurately conveys the information of the original text.
[0048] The optimized translation engine is faster and more accurate when processing text. In the processing of a large number of automobile document translations, the efficient translation engine can save time and cost, and the optimized parameter settings reduce the probability of incorrect translation, output high-quality translated texts, and meet the needs of enterprises and customers for translation.
[0049] In S4, the language model checks the coherence and correctness of the text sentence. The coherence and correctness of the text sentence are checked by a deep learning language model. The self-attention mechanism is used to capture the context relationship between words in the sentence, evaluate the rationality of the sentence, and train based on the mask language model task. During training, BERT randomly masks some words in the sentence and predicts these masked words. When checking the translated text, the text is input into the pre-trained BERT model. For each word in the text, the model calculates its prediction probability distribution in the given context. If the prediction probability of a word in its context environment is low, it means that the word is not coherent or correct in the context. Specifically, for each word wi in the sentence, the model outputs a probability distribution P(w, a) on the vocabulary, where a is the rest of the sentence except wi. If the probability value of the actual wi is lower than a certain threshold, it is considered that there is an error at the position of the word.
[0050] The self-attention mechanism of the deep learning language model of the application can comprehensively capture the complex context relationship between words in a sentence. In the translated text of the automobile, when describing the automobile electronic system, "the sensor detects data anomalies and makes corresponding adjustments", the model can understand the logical association between "sensor" and "controller" by means of the self-attention mechanism, so as to accurately judge whether the sentence is reasonable. This makes the checking of the coherence of the sentence no longer limited to the surface grammatical rules, but goes deep into the semantic level, ensuring that the translated text is semantically accurate and logically coherent.
[0051] Based on the mask language model task, the model learns language rules on a large amount of text data. In the automobile field, the model learns a large number of automobile technical documents and maintenance manual texts, and is familiar with the professional expressions and common sentence patterns in this field. Therefore, when checking the translated text, the model can accurately evaluate whether the sentence conforms to the language habits and logic of the automobile industry according to the learned knowledge. For the description of the automobile maintenance process, the model can judge whether the translated sentence conforms to the actual operation logic and professional expression.
[0052] In the S4 step, when the translated text is checked by the language model and it is found that there is an error translation, the secondary translation engine intervention mechanism is started, the sentence fragment marked as incorrect by the language model is identified and extracted, and is transmitted to the pre-selected secondary translation engine. The text with problems is retranslated, the original text is reinterpreted and translated in combination with the corpus resources, and is again subjected to the check of the language model. As in the first check, the sentence coherence, grammatical correctness and semantic accuracy are considered. If errors are still detected, the secondary translation engine is called again for a new round of translation according to the established process, and the check is performed again. This cycle is repeated until the language model confirms that there is no coherence or correctness problem in the translated text.
[0053] The application considers the sentence coherence, grammatical correctness and semantic accuracy from multiple dimensions to ensure that the translated text is not only accurate in vocabulary, smooth in sentence, but also logically coherent. In the text describing the automobile assembly process, the secondary engine retranslation can make the sentence connection more natural and the step description more consistent with the actual operation sequence, meeting the logical requirements of the automobile industry.
[0054] The error sentence fragment is automatically identified, transmitted to the secondary engine for retranslation, and checked again. The entire process has high automation, which not only saves the time cost of manual error screening and retranslation, but also reflects the efficiency and advancement of the intelligent translation system. The translator does not need to manually find errors, and the system automatically completes the error correction process, improving the work efficiency.
[0055] In the S5 step, when the user processes a large number of files to be translated, a plurality of files are batched into the system, and the system supports formats including PDF, DOCX, PPTX and TXT; after the files are imported, the user sets priorities for different files according to the importance and urgency of the files, and sequentially processes the translation of the text, automatically allocates resources, distributes the translation task to different translation engines, and updates the translation status in real time.
[0056] When setting the priority of a file, we will consider the importance and urgency of the file as two key factors, and the specific setting method is to give a certain proportion of weight to the importance and urgency, and the sum of the two weights is 1. If you think the urgency is more important, you can set the weight of the urgency to be higher, such as 0.6, and correspondingly, the weight of the importance is set to 0.4.
[0057] Suppose there is a file, after evaluation, its importance can be scored as 8 out of 10, and its urgency can be scored as 9, when calculating the priority of this file, first multiply the score of importance by its corresponding weight, that is, 8 times 0.4, get 3.2, then multiply the score of urgency by its corresponding weight, that is, 9 times 0.6, get 5.4, finally, add the two results, 3.2 plus 5.4, calculate the priority value of this file as 8.6.
[0058] In this way, each file can calculate a corresponding priority value according to its own importance and urgency, with these values, the system can arrange the translation order of the files according to the priority, and ensure that important and urgent files are processed first.
[0059] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, the above embodiments and descriptions in the specification are only the principles of the present application, and various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application.
Claims
1. A method of batch translating text, characterized by, The translation steps are: S1, acquire automobile industry professional terms, and build an automobile translation vocabulary database; S2, format the text to be translated, identify the title information, and correct errors, and determine the content characteristics of the text; S3, according to the characteristics of the text to be translated and the target language, select the preferred machine translation engine, optimize the selected translation engine, and connect the automobile translation vocabulary database with the translation engine. The translation engine prioritizes matching professional vocabulary in the automobile translation vocabulary database during the translation process and performs translation processing; S4, for the translated text, based on the language model, check the coherence and correctness of the text. If there is an error in translation, use the second selected translation engine to translate again, and check again until there is no error; S5, based on the batch translation module, import several files to be translated in batches, set the translation task parameters, including the target language and the translation priority, and translate the text according to the above steps in turn, and display the task progress and translation results in real time.
2. The method of claim 1, wherein, In step S1, the automobile professional terms are obtained by consulting automobile industry materials, browsing automobile brand official websites, industry information platforms, and communicating with automobile engineers and maintenance technicians in the industry; When building the automobile translation vocabulary database, the obtained terms are classified, including power system, vehicle body structure, and electronic device category. For each term, preliminary translation is performed through professional translation software, manual correction is performed in combination with dictionaries and industry translation materials, and annotations are added to each term to explain the meaning and application scenarios. The vocabulary database is built using Excel tools and is updated regularly to include new terms in the industry.
3. The method of claim 1, wherein, In step S2, the text font and size are pre-specified, the paragraph format is unified, the title information is identified by preliminary screening of the title format characteristics, and the content logic of the text is combined to identify the title information. Based on the spelling check tool, the spelling errors of the words are corrected, and the syntax check tool finds the sentence structure, word collocation, and syntax problems.
4. The method of claim 1, wherein, In step S2, the content characteristics of the text are determined by word frequency analysis. The high-frequency words in the text are extracted, and the content characteristics of the text are determined based on the high-frequency words. The word frequency analysis calculates the frequency of each word in the text and sorts the word frequency list from high to low. The formula is: where TF i，j is the term frequency of word i in document j, n i，j is the number of occurrences of the word in document j, is the sum of the number of occurrences of all words in document j; The high-frequency words in the document are calculated by the formula.
5. The method of claim 1, wherein, In step S3, based on the text feature analysis result and the target language for translation, the best translation engine is selected from the machine translation engine. After selecting the translation engine, the parameter settings of the engine are adjusted, including the language model and the translation preference; During translation, the translation engine prioritizes searching and matching professional vocabulary in the automobile translation vocabulary database. When matching vocabulary in the vocabulary database is identified, the translation is performed according to the established translation rules and optimized settings; where matching vocabulary is represented as for each word w in the text to be translated, a lookup is performed in the car translation vocabulary v, if then the translation result T(w) = v(w), otherwise, T(w) is determined according to other translation rules, where T(w) represents the final translation result of the word w.
6. The method of claim 1, wherein, In the S4 step, the language model checks the coherence and correctness of the text sentences. The coherence and correctness of the text sentences are checked by a deep learning language model. The self-attention mechanism is used to capture the context relationship between words in a sentence, evaluate the rationality of the sentence, and train based on the mask language model task. During training, BERT randomly masks some words in the sentence and predicts these masked words. When checking the translated text, the text is input into the pre-trained BERT model. For each word in the text, the model calculates its prediction probability distribution in the given context. If the prediction probability of a word in its context environment is low, it means that the word is not coherent or correct in the context. Specifically, for each word wi in the sentence, the model outputs a probability distribution P(w, a) on the vocabulary, where a is the rest of the sentence except wi. If the probability value of the actual wi is lower than a certain threshold, it is considered that there is an error in the position of the word.
7. The method of claim 1, wherein, In the S4 step, when the translated text is checked by the language model and it is found that there is an error in the translation, the secondary translation engine intervention mechanism is started. The sentence fragment marked as incorrect by the language model is extracted and transmitted to the pre-selected secondary translation engine. The text with problems is retranslated, the original text is reinterpreted and translated in combination with the corpus resources, and the language model is checked again. As in the first check, the coherence of the sentence, the correctness of the grammar, and the accuracy of the semantics are considered. If errors are still detected, the secondary translation engine is called again to perform a new round of translation according to the established process, and the check is performed again. This cycle continues until the language model confirms that there are no coherence or correctness problems in the translated text.
8. The method of claim 1, wherein, In the S5 step, when the user processes a large number of files to be translated, several files are imported into the system in batches. The system supports formats including PDF, DOCX, PPTX, and TXT. After importing the files, the user sets priorities for different files based on the importance and urgency of the files. The text is translated in sequence, resources are automatically allocated, translation tasks are allocated to different translation engines, and the translation status is updated in real time.