Information processing method, apparatus and electronic device

Abstract

The application is suitable for the technical field of data processing, and provides an information processing method and device and electronic equipment, including: obtaining first information to be translated; calling a translation model to perform translation processing on the first information to obtain second information, the translation model being a neural network model trained based on translation samples, the translation samples including first original information and target translation information corresponding to the first original information; calling a calibration model to calibrate the second information to obtain a calibration result, the calibration model being a neural network model trained based on calibration samples; in a case where the calibration result includes translation calibration information corresponding to the first information, taking the first information and the translation calibration information corresponding to the first information as new translation samples, wherein the new translation samples are used for optimizing training of the translation model. Through the above processing, the accuracy of the translation result output by the translation model can be improved.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, and in particular relates to information processing methods, apparatus, electronic devices, computer-readable storage media, and computer program products. Background Technology

[0002] Currently, products from different countries are circulating more and more frequently. Since different countries usually speak different languages, when using products from other countries, users often need to translate the text associated with these products into their own language.

[0003] Before performing language translation, a training set is usually determined first. Then, an initial translation model is trained using the training data from that set. Finally, the trained translation model is used to translate the language to be translated. However, even when using the above method for language translation, the translation results may still not meet the requirements. Summary of the Invention

[0004] This application provides information processing methods, apparatus, and electronic devices, which are beneficial for improving the accuracy of translation results.

[0005] In a first aspect, embodiments of this application provide an information processing method, including:

[0006] Obtain the first piece of information to be translated;

[0007] The translation model is invoked to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information.

[0008] The calibration model is invoked to calibrate the second information to obtain a calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, target translation information corresponding to the second original information, and translation calibration information corresponding to the second original information.

[0009] If the calibration result includes translation calibration information corresponding to the first information, the first information and the translation calibration information corresponding to the first information are used as new translation samples, wherein the new translation samples are used to optimize and train the translation model.

[0010] The beneficial effects of the embodiments in this application compared with the prior art are:

[0011] Since the translation calibration information for the first piece of information is output by the calibration model after calibrating the first and second pieces of information, and this calibration model is a trained neural network model, the translation calibration information for the first piece of information does not require manual calibration of the first and second pieces of information. Therefore, the efficiency of obtaining the translation calibration information for the first piece of information is improved. Simultaneously, since the new translation samples are used to optimize the training of the translation model, and the new translation samples are the first piece of information and the corresponding translation calibration information, training the translation model with the new translation samples is equivalent to using more accurate translation samples to reinforce the training of the translation model. This helps to improve the accuracy of the target translation information output by the reinforced translation model, that is, to improve the accuracy of the translation result output by the translation model.

[0012] Secondly, embodiments of this application provide an information processing apparatus, including:

[0013] The first information acquisition module is used to acquire the first information to be translated.

[0014] The translation model invocation module is used to invoke a translation model to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information.

[0015] The calibration model invocation module is used to invoke the calibration model to calibrate the second information and obtain the calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, target translation information corresponding to the second original information, and translation calibration information corresponding to the second original information.

[0016] A new translation sample determination module is used to, when the calibration result includes translation calibration information corresponding to the first information, use the first information and the translation calibration information corresponding to the first information as new translation samples, wherein the new translation samples are used to optimize the training of the translation model.

[0017] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect.

[0018] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0019] Fifthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to execute the method described in the first aspect above.

[0020] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0022] Figure 1 This is a flowchart illustrating an information processing method provided in an embodiment of this application;

[0023] Figure 2 This is a schematic flowchart of another information processing method provided in an embodiment of this application;

[0024] Figure 3 This is a schematic flowchart of another information processing method provided in an embodiment of this application;

[0025] Figure 4 This is a schematic diagram illustrating the implementation details of a translation model and a calibration model provided in one embodiment of this application;

[0026] Figure 5 This is a schematic diagram of a translation interface provided in another embodiment of this application;

[0027] Figure 6 This is a schematic diagram of a translation interface including positive feedback controls and negative feedback controls provided in an embodiment of this application;

[0028] Figure 7 This is a schematic diagram illustrating a user feedback window displayed on a translation interface, provided in another embodiment of this application.

[0029] Figure 8 This is an example diagram of a translation interface including error type selection options provided in an embodiment of this application;

[0030] Figure 9 This is a schematic diagram of a calibration interface provided in an embodiment of this application;

[0031] Figure 10 This is a schematic diagram of the training process of a translation model and a calibration model provided in an embodiment of this application;

[0032] Figure 11 This is a schematic diagram of the structure of an information processing device provided in an embodiment of this application;

[0033] Figure 12 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0034] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0035] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0036] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0037] Furthermore, in the description of this application and the appended claims, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0038] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0039] Currently, when text translation is required, a limited training set is usually determined first, and then an initial translation model is trained based on the fixed training data in the training set. Finally, the trained translation model is used to translate various texts to be translated.

[0040] As the amount of translation increases, such as with the increase in the types of information to be translated, the translation model may encounter texts to be translated that are completely different from the training data. At this time, the translation model cannot ensure that the output translation results meet the translation requirements, such as the correctness of the output translation format and semantics, resulting in low accuracy of the translation results output by the translation model.

[0041] To improve the accuracy of translation results output by translation models, embodiments of this application provide an information processing method. In this method, after initial training, the translation model can be further optimized using translation calibration information obtained by calibrating the first information input to the translation model and the target translation information output by the model, thereby improving the accuracy of the translation results output by the translation model.

[0042] The information processing method provided in the embodiments of this application is described below with reference to the accompanying drawings.

[0043] Figure 1 A flowchart illustrating an information processing method provided in an embodiment of this application is shown. This information processing method can be applied to electronic devices, such as electronic devices acting as clients, or electronic devices acting as servers, as detailed below:

[0044] S11, obtain the first information to be translated.

[0045] Here, the first piece of information can be any of the following: text information, image information, audio information, etc.

[0046] Specifically, the first information to be translated can be information obtained through direct interaction between the electronic device and the user. For example, the electronic device can use text information input by the user through an input device as the first information; or the electronic device can use audio information input by the user through a microphone as the first information. In addition to using information obtained through direct interaction between the electronic device and the user as the first information, information obtained through interaction between the electronic device and other electronic devices can also be used as the first information, which is not limited here.

[0047] S12, the translation model is invoked to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information.

[0048] The second information can be any type of information, such as text, image, or audio, and the language of the second information is different from the language of the first information. For example, if the second information is text, and the first information can be any type of information, such as text, image, or audio, and the language of the second information is different from the language of the first information, the translation process involves translating the text information in one language into translated information in another language. If the first information is an image, and the language of the second information is different from the language of the text contained in the image, the translation process involves translating the text extracted from the image into text information in another language. Similarly, if the first information is audio, and the language of the second information is different from the language of the text extracted from the audio, the translation process involves translating the text extracted from the audio into text information in another language. For another example, if the second information is an image, and the first information can also be an image, the translation process involves translating the text extracted from the image into text information in another language and generating the second information. In this case, the second information can be the image information obtained by replacing the text information in the first information (i.e., the original text) with the translated text information. It could be that the image information is obtained by attaching the translated text information on the basis of retaining the original text in the first information; for example, the second information is audio information, and the first information can be text information or audio information. If the first information is text information, and the language of the second information is different from the language corresponding to the first information, the translation process is to translate the text information extracted from the audio information into text information in another language. The second information is the audio information corresponding to the translated text information in another language. If the first information is audio information, and the language of the second information is different from the language corresponding to the first information, the translation process is to translate the text information extracted from the audio information into text information in another language and generate the second information. The second information is the audio information generated based on the translated text information.

[0049] Among the multiple pieces of first original information, there are languages ​​that are the same as the first information itself. Similarly, among the multiple pieces of target translation information, there are languages ​​that are the same as the second information. However, the languages ​​corresponding to the first original information and the target translation information are usually different. For example, when the language corresponding to the first original information is Chinese, the language corresponding to the target translation information can be English, Japanese, German, or other non-Chinese languages; when the language corresponding to the first original information is English, the language corresponding to the target translation information can be Chinese, Japanese, German, or other non-English languages.

[0050] In this embodiment, when the target translation information is marked as the expected value of the first original information, it indicates that the target translation information is the correct translation result corresponding to the first original information (e.g., the corpus of the first original information is correctly translated), meaning that the first original information and the target translation information corresponding to the first original information are equivalent to positive samples in the translation samples. Conversely, when the target translation information is marked as a non-expected value of the first original information, it indicates that the target translation information is not the correct translation result corresponding to the first original information, meaning that the first original information and the target translation information are equivalent to negative samples in the translation samples. Here, a positive sample refers to a data point belonging to the positive class (or category of interest), and a negative sample refers to a data point belonging to the negative class (or category of non-interest).

[0051] After obtaining the positive and negative samples of the translation sample, the neural network model used as the initial translation model can be trained based on these samples to obtain the trained translation model. It should be noted that the translation model used in this step can be either the translation model obtained after training the initial model, or a translation model obtained after subsequent optimization training of the trained model.

[0052] In this embodiment, after translating the first information using a translation model to obtain the second information, the second information can be output. For example, when the electronic device acts as a client, it can display the second information through its display interface; alternatively, it can output the second information through the client's voice output device, which is not limited here. As another example, when the electronic device acts as a server, it can send the second information to a client communicating with it, so that the client can display the second information through its display interface or output the second information through its voice output device.

[0053] S13, call the calibration model to calibrate the second information mentioned above, and obtain the calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information.

[0054] In this embodiment, the calibration model is trained based on calibration samples including second original information, target translation information corresponding to the second original information, and translation calibration information corresponding to the second original information. Specifically, the calibration samples may include positive samples and negative samples. For example, when the target translation information of the second original information is marked as expected translation information (i.e., the target translation information of the second original information is an accurate translation of the second original information), the translation calibration information corresponding to the second original information can be "none" or blank information, and the calibration model can be set not to generate translation calibration information corresponding to the second original information in this case. In this scenario, the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information are equivalent to positive samples in the calibration samples. Another example is when the target translation information of the second original information is marked as unexpected translation information (i.e., the target translation information of the second original information is an inaccurate translation of the second original information), the translation calibration information corresponding to the second original information can be accurate translation calibration information, and the accurate translation calibration information can be used to correct the errors. Accurate translation calibration information is marked as expected translation calibration information. In this scenario, the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information (i.e., expected translation calibration information) are equivalent to positive samples in the calibration samples. For example, when the target translation information of the second original information is marked as expected translation information (i.e., the target translation information of the second original information is an accurate translation of the second original information), the translation calibration information corresponding to the second original information can be inaccurate translation calibration information, and this inaccurate translation calibration information is marked as unexpected translation calibration information. In this scenario, the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information (i.e., unexpected translation calibration information) are equivalent to negative samples in the calibration samples.

[0055] After training a calibration model based on calibration samples, if the calibration model is invoked, it will receive first and second information sent by the translation model and calibrate the second information. Specifically, the calibration model calibrates the second information by determining its accuracy as a translation of the first information and obtaining the corresponding calibration result. The calibration model can determine the accuracy of the second information as a translation of the first information through at least one dimension: corpus, symbols, semantics, and context. Optionally, considering that users have higher expectations for the accurate translation of corpus (such as specific words, numbers, etc.) and symbols (such as "#", "=", etc.), the calibration model can first determine whether the second information can accurately translate the corpus and / or symbols of the first information. If it determines that the second information does not accurately translate the corpus and / or symbols of the first information, it does not need to continue judging the second information in the semantic and context dimensions, and instead directly generates a calibration result. Only after determining that the second information accurately translates the corpus and symbols of the first information does it continue to judge whether the second information accurately translates the semantics and / or context of the first information, and then generate a calibration result based on the corresponding judgment result.

[0056] The calibration results described above are used to characterize the accuracy of the second information. Specifically, this accuracy can be characterized by accuracy information. For example, when there is a difference between the semantics or context of the second information and the semantics or context of the first information, the accuracy information can be characterized as the accuracy of the second information being less than 100%. Optionally, the accuracy information can also be represented by a score; the higher the score, the higher the accuracy of the second information, and vice versa. For example, a score of 0 indicates that the second information does not retain the original meaning of the first information (e.g., the second information does not accurately translate the corpus and / or symbols of the first information), a score of 100 indicates that the second information fully retains the original meaning of the first information and is grammatically correct (e.g., the second information accurately translates the corpus, symbols, semantics, and context of the first information), and a score greater than 0 and less than 100 indicates insufficient accuracy of the second information, such as indicating that the second information accurately translates the corpus and symbols of the first information, but does not accurately translate the semantics and / or context of the first information.

[0057] Furthermore, when the accuracy of the second information is insufficient (e.g., the accuracy information is lower than 100 or lower than other preset values ​​less than 100 (e.g., any value from 0 to 95)), the above calibration result may also include translation calibration information corresponding to the first information. This translation calibration information is obtained by the above calibration model calibrating the second information based on the first information, such as calibrating the corpus and / or symbols of the second information based on the first information.

[0058] In this embodiment, the calibration model can be actively invoked to calibrate the second information after each translation model obtains the second information, or the calibration model can be actively invoked once after the translation model has performed N translation processes; no limitation is made here. Here, N is a natural number greater than 1.

[0059] S14, if the calibration result includes translation calibration information corresponding to the first information, the first information and the translation calibration information corresponding to the first information are used as new translation samples, wherein the new translation samples are used to optimize and train the translation model.

[0060] Specifically, if the calibration result includes translation calibration information corresponding to the first information, it indicates that the calibration model believes the corpus of the first information has not been correctly translated by the translation model. In this case, the translation calibration information corresponding to the first information is taken as the translation result of the first information, and the first information and its translation result are used as new translation samples for optimization training of the translation model. The initiation conditions for optimization training can be set according to the actual situation. For example, the translation model can be optimized after obtaining new translation samples; or it can be optimized when new translation samples are obtained and the current time is a preset time; and so on.

[0061] In this embodiment, since the translation calibration information of the first information is output by the calibration model after calibrating the first and second information, and the calibration model is a trained neural network model, the translation calibration information of the first information does not require manual calibration of the first and second information. Therefore, the efficiency of obtaining the translation calibration information of the first information is improved. Simultaneously, since new translation samples are used to optimize the training of the translation model, and the new translation samples are the first information and the corresponding translation calibration information, training the translation model with new translation samples is equivalent to reinforcing the translation model with more accurate translation samples. This helps improve the accuracy of the target translation information output by the reinforced translation model, i.e., improves the accuracy of the translation result output by the translation model.

[0062] In some embodiments, considering that training the translation model with translation samples containing both positive and negative samples is beneficial to increasing the probability that the translation model learns a wider range of features, thereby improving its robustness to new data, therefore, when determining new translation samples, the positive and negative samples in the new translation samples can be determined. In this case, the above-mentioned use of the first information and the translation calibration information corresponding to the first information as new translation samples includes:

[0063] A1. Take the first information and the translation calibration information corresponding to the first information as positive samples in the new translation samples mentioned above.

[0064] A2. Take the first information and the second information mentioned above as negative samples in the new translation samples.

[0065] Specifically, when the calibration result includes translation calibration information corresponding to the first information, it indicates that the second information obtained by the translation model from the translation processing of the first information is inaccurate. In this case, the translation calibration information of the first information obtained by correcting the first and second information is used as the target translation information corresponding to the first information. Its reliability is higher than that of using the second information as the target translation information corresponding to the first information. That is, using the first information and its translation calibration information as positive samples in the new translation samples is beneficial to improving the accuracy of the obtained positive samples. In addition, since the positive and negative samples in the new translation samples are obtained by calling the calibration model to calibrate the second information, that is, the first information in the new translation samples is data that the validated translation model cannot accurately translate. Therefore, determining the positive and negative samples of the new translation samples in the above manner is beneficial to improving the accuracy of the obtained positive and negative samples.

[0066] In some embodiments, after obtaining new translation samples, the decision to optimize the translation model can be made based on the number of new positive and negative samples. (See reference) Figure 2 , Figure 2 A flowchart illustrating another information processing method provided in an embodiment of this application is shown below, and details are as follows:

[0067] S21, Obtain the first information to be translated.

[0068] S22, the translation model is invoked to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information.

[0069] S23, call the calibration model to calibrate the second information mentioned above, and obtain the calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information.

[0070] S24, if the calibration result includes translation calibration information corresponding to the first information, the first information and the translation calibration information corresponding to the first information are used as new translation samples, wherein the new translation samples are used to optimize and train the translation model.

[0071] S25, if the number of positive and / or negative samples in the new translation samples is greater than a preset threshold, the translation model is optimized and trained using the positive and negative samples in the new translation samples.

[0072] The preset quantity threshold can be set according to the actual situation.

[0073] In this embodiment, after the initial translation model training is completed, or after each optimization training of the translation model, the number of positive and / or negative samples in the new translation samples (i.e., translation samples that have not participated in training) obtained after the training or optimization training is completed is accumulated. If it is determined that the accumulated number of positive and / or negative samples is greater than a preset threshold, then all new translation samples are used to optimize the training of the translation model. Since the translation model is optimized only after it is determined that the accumulated number of positive and / or negative samples is greater than the preset threshold, the training frequency of the translation model is reduced, thereby saving the resources consumed in training the translation model.

[0074] Optionally, considering that the probability of translation errors in the translation model decreases with increasing training iterations, and the severity of such errors also decreases accordingly, the aforementioned preset threshold can be dynamically adjusted based on the number of training iterations. For example, the preset threshold can be set to be positively correlated with the number of training iterations, meaning that the preset threshold increases as the number of training iterations increases. This setting helps to reduce the training frequency of the translation model while maintaining its translation performance.

[0075] In some embodiments, to improve the accuracy of translation processing, the translation model can be fine-tuned before translating the first information. (Reference) Figure 3 , Figure 3 A flowchart illustrating another information processing method provided in an embodiment of this application is shown below, and details are as follows:

[0076] S31, obtain the first information to be translated.

[0077] S32, supervise and fine-tune the translation model according to the application scenario of the translation model.

[0078] Supervised fine-tuning (SFT) is a commonly used technique in machine learning and natural language processing. It involves further training a pre-trained translation model using labeled data to improve its performance on specific tasks or domains.

[0079] For example, when the application scenario of the translation model is a game scenario, corresponding translation samples are generated based on the terms corresponding to the game scenario. The translation model is then trained based on the generated translation samples and corresponding annotations to achieve supervised fine-tuning of the translation model, thereby improving the accuracy of the translation results obtained by the adjusted translation model when translating the corpus of the game scenario.

[0080] It should be pointed out that, in Figure 3 In this context, supervised fine-tuning of the translation model is set after the step of obtaining the first information to be translated. In practice, supervised fine-tuning of the translation model can also be set before the step of obtaining the first information to be translated, as long as it is ensured that supervised fine-tuning of the translation model is performed before the translation model processes the first information.

[0081] S33, the supervised fine-tuned translation model is invoked to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information.

[0082] S34, call the calibration model to calibrate the second information mentioned above, and obtain the calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information.

[0083] S35, if the calibration result includes translation calibration information corresponding to the first information, the first information and the translation calibration information corresponding to the first information are used as new translation samples, wherein the new translation samples are used to optimize and train the translation model.

[0084] In this embodiment of the application, considering that the application scenario of the translation model is related to the corpus to be translated involved in the translation model, supervising and fine-tuning the translation model according to the application scenario of the translation model is beneficial to improving the accuracy of the translation results output by the adjusted translation model.

[0085] The above describes the supervised fine-tuning of the translation model before translation processing. In some embodiments, the calibration model can also be supervised fine-tuned before calibration to improve the accuracy of the calibration results. That is, before calling the calibration model to calibrate the second information mentioned above, the following steps are also included:

[0086] The calibration model is then fine-tuned under supervision based on its application scenarios.

[0087] Correspondingly, the above-mentioned calibration model is invoked to calibrate the second information, including:

[0088] The supervised fine-tuned calibration model is invoked to calibrate the second piece of information mentioned above.

[0089] The application scenarios of the calibration model are similar to those of the translation model, and the process of supervising the fine-tuning of the calibration model is similar to that of the translation model, so it will not be elaborated here.

[0090] In some embodiments, considering that the calibration results obtained from the calibration model may also be incorrect, it is necessary to determine whether the calibration results are correct before generating translation samples based on the translation calibration information. That is, when the calibration results include translation calibration information corresponding to the first information, using the first information and the translation calibration information corresponding to the first information as new translation samples includes:

[0091] If the above calibration result is correct and the above calibration result includes translation calibration information corresponding to the above first information, the above first information and the above translation calibration information corresponding to the above first information shall be used as the new above translation sample.

[0092] Specifically, the correctness of the calibration result can be determined by manually analyzing it, or the calibration model can be set to output the confidence level of the calibration result when outputting the calibration result, and the correctness of the calibration result can be determined based on the confidence level of the calibration result (for example, if the confidence level is less than 50%, the calibration result is considered incorrect, and vice versa).

[0093] In this embodiment of the application, if it is determined that the calibration result is correct and the calibration result includes translation calibration information corresponding to the first information, it indicates that the translation model's translation of the first information is incorrect. At this time, the first information and the translation calibration information corresponding to the first information are used as new translation samples. For example, the first information and the translation calibration information corresponding to the first information are used as positive samples in the new translation samples, which helps to improve the accuracy of the new translation samples.

[0094] In some embodiments, if the calibration result is determined to be incorrect, a new translation sample is not generated based on the first information, in order to reduce the probability of generating an incorrect translation sample.

[0095] In some embodiments, considering the possibility of erroneous calibration results, indicating that the calibration capability of the calibration model needs improvement, the information processing method provided in this application embodiment further includes the following steps to obtain new calibration samples:

[0096] B1. In the event of an error in the above calibration results, obtain the recommended translation calibration information corresponding to the first information mentioned above.

[0097] The recommended translation calibration information is the translation result corresponding to the first information, and compared with the translation calibration information output by the correction model, the recommended translation calibration information has higher reliability.

[0098] Specifically, if an error is detected in the calibration result, a manual correction is prompted. The electronic device sends a manual correction prompt to a designated terminal. This prompt includes first information, second information, and a description of the correction required for the second information. The designated terminal uses the information entered by the corrector in response to the manual correction prompt as recommended translation calibration information corresponding to the first information and sends this recommended translation calibration information to the electronic device.

[0099] B2. The first information, the second information, and the recommended translation calibration information mentioned above are used as new calibration samples. These new calibration samples are used to optimize and train the calibration model.

[0100] After receiving the recommended translation calibration information, the electronic device uses the first information, the second information, and the recommended translation calibration information as new calibration samples for subsequent optimization training of the calibration model.

[0101] Since an erroneous calibration result indicates low accuracy of the translation calibration information output by the calibration model, obtaining recommended translation calibration information corresponding to the first information, and using this recommended translation calibration information, the first information, and the second information as new calibration samples, is beneficial to improving the accuracy of the new calibration samples.

[0102] Optionally, the recommended translated calibration information, the first information, and the second information are used as positive samples in the new calibration samples, and the translated calibration information, the first information, and the second information are used as negative samples in the new calibration samples, so as to obtain a balanced ratio of positive and negative samples.

[0103] To more clearly describe the process of determining new translation samples and new calibration samples, the following section combines... Figure 4 Describe it. In Figure 4 middle:

[0104] S41, the translation model receives the first information, translates it to obtain the second information, and then sends the first and second information to the calibration model.

[0105] S42, the calibration model receives the first information and the second information, calibrates the second information, and obtains a calibration result. Specifically, the process of the translation model translating the first information and the process of the calibration model calibrating the second information based on the first information can be implemented with reference to the foregoing embodiments, and will not be repeated here.

[0106] S43: Review the first information, the second information, and the calibration results to determine whether the calibration model or the translation model needs optimized training. If the calibration model needs optimized training, proceed to S44; if the translation model needs optimized training, proceed to S47.

[0107] In this embodiment, the calibration result can be output to a designated device so that relevant personnel can review it. If the calibration result is judged manually based on the first and second information, and if the result is correct, it indicates that the translation model did not accurately translate the first information, meaning the model needs optimization training. If the result is incorrect, it indicates that the calibration model did not accurately calibrate the second information, meaning the model needs optimization training.

[0108] S44, when the calibration model needs to be optimized for training, obtain the recommended translation calibration information corresponding to the first information.

[0109] In this embodiment of the application, the first information can be translated manually, and the translated recommended translation calibration information can be input into an electronic device so that the electronic device can obtain the recommended translation calibration information.

[0110] S45, generate negative samples of the calibration model based on the first information, the second information, and the translated calibration information included in the calibration results, and generate positive samples of the calibration model based on the recommended translated calibration information corresponding to the first information, the first information, and the second information.

[0111] When the calibration model needs to be optimized for training, it indicates that the translation calibration information included in the calibration results has not accurately calibrated the second information. That is, the translation calibration information is determined to be an incorrect translation of the first information, but the recommended translation calibration information is determined to be an accurate translation of the first information. In this case, the first information, the second information, and the translation calibration information should be used as negative samples of the calibration model, and the obtained recommended translation calibration information, the first information, and the second information corresponding to the first information should be used as positive samples of the calibration model.

[0112] S46, store the generated positive samples and negative samples of the calibration model into the positive and negative sample library of the calibration model.

[0113] S47, when the translation model needs to be optimized during training, negative samples of the translation model are generated based on the first information and the second information, and positive samples of the translation model are generated based on the first information and the translation calibration information included in the calibration results.

[0114] When the translation model needs to be optimized during training, it indicates that the translation calibration information included in the calibration result has accurately calibrated the second information. That is, the translation calibration information is determined to be the accurate translation of the first information, and the second information is determined to be the incorrect translation of the first information. At this time, the first information and the second information are used as negative samples of the translation model, while the first information and the translation calibration information included in the calibration result are used as positive samples of the translation model.

[0115] S48. Store the generated positive and negative samples of the translation model into the positive and negative sample library of the translation model.

[0116] As described above, the information processing method of this application embodiment can be applied to both client and server. When the information processing method is applied to a client and the first information is text information, it can interact with the user through an interface. That is, before obtaining the first information to be translated, it further includes:

[0117] The translation interface includes a language selection control and an input area. The language selection control is used to determine the source language and the target language, and the input area is used to input the first information.

[0118] Correspondingly, obtaining the first information to be translated includes obtaining the first information input in the input area.

[0119] Correspondingly, the above-mentioned call to the translation model to translate the first information includes: calling the translation model to perform text translation processing on the first information according to the source language and target language indicated by the language selection control.

[0120] The translation interface can be a standalone page or a subpage within a standalone page, such as a subpage embedded in a browser page.

[0121] When the translation interface is a separate page, the interface diagram can be as follows: Figure 5 As shown. In Figure 5 In the translation interface 5, there are a language selection control 51 and an input area 52.

[0122] In the language selection control 51, language A represents the language of the first information (i.e., the source language), and language B represents the language of the second information (i.e., the target language). When the area containing language A (or language B) is clicked, all the languages ​​that can be selected for language A (or language B) can be displayed.

[0123] The input area 51 is located in an area that can receive information input by the input device of the electronic device, and the received information is used as the first information mentioned above.

[0124] In this embodiment, since a translation interface including a language selection control and an input area is displayed, users can intuitively and flexibly select the language of the information to be translated and input the information they wish to translate, thereby improving the user experience.

[0125] Optionally, after the translation model is invoked to translate the first information, the translation result—the second information—corresponding to the first information is displayed on the translation interface. It should be noted that the page containing the second information may be the same as or different from the page containing the first information; this is not limited here.

[0126] In the above description, the translation interface includes a language selection control and an input area. In some embodiments, the translation interface may also include a file receiving control. Specifically, when the file receiving control is clicked, the electronic device displays selectable file paths, and after the user selects the corresponding file according to the file path, the electronic device uses the information from the selected file as the aforementioned first information.

[0127] In this embodiment, since the entire file can be imported through the translation interface, the invoked translation model can translate the information in the entire file. That is, since the user does not need to enter the information to be translated line by line, translation efficiency is improved.

[0128] In the above description, the translation interface includes a language selection control and an input area; alternatively, the translation interface includes a language selection control, an input area, and a file receiving control. In some embodiments, the translation interface may also include a user feedback control associated with the second information mentioned above.

[0129] When the aforementioned user feedback control is triggered, a user feedback window is displayed. The data entered in the aforementioned user feedback window is used as the target translation information of the aforementioned first information, and the aforementioned first information and the target translation information of the aforementioned first information are used as the new aforementioned translation sample.

[0130] Specifically, when a user believes that the second piece of information is not a correct translation of the first piece of information, they can click the user feedback control. The electronic device will then display a user feedback window, where the user can input data. The electronic device will then use this data as the target translation information for the first piece of information, meaning the data input in the user feedback window will be considered the correct translation of the first piece of information. When the translation sample includes both positive and negative samples, the electronic device will use the first piece of information and its target translation information as the positive sample in the new translation sample, and the first and second pieces of information as the negative sample in the new translation sample.

[0131] Optionally, considering that users may be satisfied or dissatisfied with the second information, the user feedback control can be set to include a positive feedback control and a negative feedback control. When the positive feedback control is triggered, it indicates that the user is satisfied with the second information; conversely, when the negative feedback control is triggered, it indicates that the user is dissatisfied with the second information.

[0132] like Figure 6 As shown, positive feedback controls can be represented by a thumbs-up gesture, while negative feedback controls can be represented by an upside-down thumbs-up gesture. Figure 6 In this context, the first piece of information is in Chinese, and the second piece of information is in English. The prompt "Translate the following sentence from Chinese to English" is equivalent to a prompt from the electronic device to the translation model, indicating that the translation model should translate the first piece of information from Chinese to English.

[0133] exist Figure 6 In the diagram, L and the circle surrounding L represent the user, and H and the circle surrounding H represent the translation model or the electronic device. The first message is "cheering from Ya Ya," and the second message is "Emoji - Cheering Aya." Assuming the user finds the translation model's translation of the first message accurate, the user can click... Figure 6 The thumbs-up gesture in Chinese (i.e.) Figure 6 (The gesture diagram on the left) The electronic device can record the detected click operation, and it can also use the first and second information as positive samples in the new translation samples. Of course, if the user believes that the translation model's translation of the first information is incorrect, the user can click... Figure 6 The inverted thumb gesture (i.e.) Figure 6 (The gesture icon on the right) shows how electronic devices can display actions based on detected clicks. Figure 7 The user feedback window shown is 7. Figure 7 In the middle, the user feedback window 7 includes an area 71 for receiving user input data. After the user finishes entering data in this area 71, they can click... Figure 7 Clicking the "OK" button in the image indicates that the electronic device uses the data received in region 71 as the target translation information for the first information. Furthermore, the electronic device uses the first information and its target translation information as positive samples in the new translation samples, while using the first information and the second information as negative samples in the new translation samples, in order to more effectively optimize and train the translation model.

[0134] In the above description, the user feedback window includes an area 71 for receiving user input data. Considering that a translation result may have correct corpus but inconsistent usage habits with reality, in this case, the translation result of the translation model can be considered correct, but requires optimization. Therefore, error type selection options can be provided in the user feedback window to indicate the type of error the user believes exists in the second information. That is, in some embodiments, the user feedback window also includes at least two error type selection options. When an error type selection option is selected, it indicates that the second information contains the error type corresponding to the selected error type selection option.

[0135] Correspondingly, the above-mentioned use of the first information and the target translation information of the first information as the new translation sample includes: when the error type of the second information is the target error type, the first information and the target translation information of the first information are used as the new translation sample.

[0136] like Figure 8 As shown, the error type selection options above can include: translation errors and translation optimization.

[0137] exist Figure 8 In the user feedback window, the selection of the error type corresponding to the translation error indicates that the user believes the translation model's translation of the second information is incorrect. Since the user feedback window also includes at least two error type selections, it helps the electronic device determine the type of error present in the second information. Subsequently, the target translation information of either the second or first information can be processed accordingly to improve the accuracy of the processing. For example, when translation model optimization is needed, the target translation information corresponding to the "translation optimization" error type can be used as a new translation sample; in this case, the target error type includes "translation optimization." Conversely, when translation model optimization is not needed, the target translation information corresponding to the "translation error" error type is not used as a new translation sample; only the target translation information corresponding to the "translation error" error type is used. In this case, the target error type is "translation error," excluding "translation optimization."

[0138] The above describes how an electronic device displays first information and second information obtained by the translation model from the first information in a translation interface. In some embodiments, the electronic device may also display a calibration interface, which displays the first information, the second information, and the calibration result obtained by the calibration model after calibrating the second information.

[0139] Specifically, the calibration interface can be divided into two areas: a first area for displaying first information and second information, and a second area for displaying the calibration result. For example, when a user needs to calibrate information to be translated and translated information obtained by translating the information to be translated, the information to be translated can be used as the first information, and the translated information as the second information. The first and second information are then input into the first area of ​​the calibration interface. After the calibration model calibrates the first and second information, it outputs the corresponding calibration result in the second area.

[0140] It should be noted that the translated information mentioned above may be the translation result output by the translation model described in the embodiments of this application, or it may not be the translation result output by the translation model. For example, considering that the calibration model automatically calibrates the output of the translation model, the calibration interface can be displayed when the calibration model is being trained. At this time, the translated information displayed on the calibration interface is the translation result output by the translation model. However, when the user uses the calibration model to calibrate the acquired information to be translated and the translated information, the translated information displayed on the calibration interface is not the translation result output by the translation model.

[0141] In some embodiments, the second area of ​​the calibration interface described above, in addition to displaying calibration results, can also be used to display translation problem types, which reflect the translation problem types present in the second information.

[0142] To describe the calibration interface more clearly, the following will combine... Figure 9 Describe it.

[0143] Figure 9 A schematic diagram of a calibration interface is shown. Figure 9 The first area displays the translation type, source text, and translated text. The translation type reflects the language of the first and second information. The source text corresponds to the first information, and the translated text corresponds to the second information. Furthermore, in the calibration interface, the "I" and its surrounding circle indicate that the information to its left is the information requiring calibration, and the "J" and its surrounding circle indicate that the information to its right is the calibrated information.

[0144] exist Figure 9In the text, "cn==>en" indicates translation from Chinese to English. The information to be translated is "Basic Attack Damage (No Critical Hit)". The translated text is "Magic Attack Base Damage". After calibrating "Basic Attack Damage (No Critical Hit)" and "Magic Attack Base Damage", the calibration model obtained the following results: the translation is inaccurate, and the recommended translation "Normal Attack Base Damage (No Crit)" is also incorrect.

[0145] exist Figure 9 In the text, the translation problem type is "inaccurate translation". In reality, other types of translation problems may also occur, such as the translation not being translated according to the corpus or omissions.

[0146] When the translation problem type is that the translation is not based on the corpus or that there are omissions, the calibration results of the calibration model can indicate that "the translation is not based on the corpus" and provide a specific translation in the recommended translation, with the missing content added to the specific translation. Of course, the calibration results of the calibration model can also indicate "the translation is not based on the corpus" and "the translation contains missing content" respectively.

[0147] Optionally, the calibration results may include not only the specific translation problem type, but also solutions for that translation problem type. For example, when the translation problem type is "the translation does not follow the corpus," if the translation for the original text "life" is "heals" and the translation for the original text "reply" is "additional," then the calibration results will include not only "the translation does not follow the corpus," but also "Life: "Health," "Reply: "Restore," indicating that "Life" should be translated as "Health" and "Reply" should be translated as "Restore."

[0148] In some embodiments, the calibration interface may further include a positive feedback calibration area and a negative feedback calibration area. The positive feedback calibration area and the negative feedback calibration area can be referenced to the positive feedback control and negative feedback control described above, and will not be repeated here.

[0149] The above describes determining the first information by detecting information in the input area of ​​the translation interface. In practice, the data currently displayed and selected on the interface can also be used as the first information. For example, when an electronic device displays file X, and the user selects a word or sentence from file X, the electronic device uses the selected word or sentence as the aforementioned first information. That is, in some embodiments, the above information processing method is applied to the client, and obtaining the first information to be translated includes:

[0150] Upon receiving a translation trigger instruction, the selected data in the currently displayed interface of the client is obtained as the first information to be translated.

[0151] Specifically, a correlation is established beforehand between the selected action and the translation model. The selected data is determined by identifying the selected action, and the selected data is identified as the primary information based on the pre-established correlation.

[0152] In this embodiment of the application, when a translation trigger instruction is received, the translation model will be invoked and its translation function will be activated. In this way, after the first information is determined, the translation processing of the first information can be realized.

[0153] Optionally, the translation model can be associated with a specified application, and the translation model will start by default when the specified application starts. For example, the specified application can be a Word application. When the Word file is opened, it indicates that the Word application has started, and the translation model associated with the Word application will automatically start. When data in the Word file is selected, the translation model actively translates the selected data and displays the second information obtained from the translation process.

[0154] The above mainly introduced the implementation method of the information processing method when applied to the client. The following describes the implementation method when applied to the server. That is, in some embodiments, the above information processing method is applied to the server, and obtaining the first information to be translated includes:

[0155] Receive the first message to be translated sent by the client.

[0156] Correspondingly, after obtaining the second information as described above, the process also includes: sending the second information back to the client.

[0157] Specifically, when the information processing method is applied to the server, the server usually does not provide an interactive interface for user interaction. In this case, the first information obtained by the server can be obtained by receiving information sent by the client interacting with the user. After obtaining the second information output by the translation model, the server feeds back the second information to the client so that the user interacting with the client can know the second information.

[0158] To more clearly describe the information processing method provided in the embodiments of this application, the following is combined with... Figure 10 Describe it.

[0159] Figure 10 This is a schematic diagram illustrating the training process of a translation model and a calibration model provided in an embodiment of this application.

[0160] Training of the translation model (1001-1004):

[0161] 1001. Obtain the training corpus for the translation model. Here, the training corpus is equivalent to the first source information and the corresponding target translation information described above.

[0162] 1002. Clean the training corpus of the translation model.

[0163] Specifically, the correct translated source text and target translation text are selected from the training corpus of the translation model. That is, the first source information and the target translation information that corresponds to the first source information are identified as positive samples of the translation samples. Of course, to ensure that the translation samples have a certain number of negative samples, the first source information and the target translation information that does not correspond to the first source information can also be retained during the cleaning process.

[0164] 1003. Train the translation model.

[0165] Specifically, the initial translation model is trained using the cleaned training corpus of the translation model until the condition for stopping training is met, thus obtaining the translation model after initial training.

[0166] 1004. Supervised fine-tuning yields the translation model after supervised fine-tuning.

[0167] Specifically, the translation model after initial training is fine-tuned under supervision according to the application scenario of the translation model, resulting in a supervised fine-tuned translation model.

[0168] Training of the calibration model (1005-1008):

[0169] 1005. Obtain the training corpus for the calibration model. Here, the training corpus corresponds to the second original information, the corresponding target translation information, and the corresponding translation calibration information described above.

[0170] 1006. Training corpus for cleaning and calibration models.

[0171] Specifically, translation calibration information that accurately calibrates the second original information is selected from the training corpus of the calibration model. This second original information, the target translation information obtained by mistranslating the second original information, and the accurate translation calibration information corresponding to the second original information are used as positive samples for calibration. Of course, to ensure that the calibration samples have a certain number of negative samples, the second original information and the translation calibration information obtained by miscalibrating the second original information can also be retained during the cleaning process.

[0172] 1007. Training and calibrating the model.

[0173] Specifically, the initial calibration model is trained using the cleaned calibration model's training corpus until the conditions for stopping training are met, thus obtaining the calibration model after initial training.

[0174] 1008. Supervised fine-tuning yields the calibrated model after supervised fine-tuning.

[0175] The calibration model after initial training is used in the application scenario to perform supervised fine-tuning on the initially trained calibration model, resulting in a supervised fine-tuned calibration model.

[0176] Generation of positive and negative sample libraries:

[0177] 1009. Manual annotation for Web front-end.

[0178] Here, "human annotation" on the web front-end refers to the target translation information obtained from user feedback on the translation interface of the client, and the recommended translation calibration information obtained from user feedback on the calibration interface of the client.

[0179] 1010. Translation expert calibration data.

[0180] Here, the translation expert calibration data refers to the data obtained by calibrating the calibration results output by the calibration model through translation experts, which includes the recommended translation calibration information described above.

[0181] 1011. Generate positive and negative samples for the calibration model.

[0182] Positive samples for the calibration model are generated based on the recommended translation calibration information and the corresponding source and translation texts. Negative samples for the calibration model are generated based on the translation calibration information generated by the calibration model and the corresponding source and translation texts.

[0183] 1012. Generate positive and negative samples for the translation model.

[0184] Positive samples for the translation model are generated based on the translation calibration information output by the calibration model and the corresponding source text, while negative samples for the translation model are generated based on the translated text output by the translation model and the corresponding source text.

[0185] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0186] Corresponding to the information processing method described in the above embodiments, Figure 11 A structural block diagram of the information processing apparatus provided in the embodiments of this application is shown. For ease of explanation, only the parts related to the embodiments of this application are shown.

[0187] Reference Figure 11 The information processing device 11 includes: a first information acquisition module 111, a translation model invocation module 112, a calibration model invocation module 113, and a new translation sample determination module 114. Wherein:

[0188] The first information acquisition module 111 is used to acquire the first information to be translated.

[0189] Here, the first piece of information can be any of the following: text information, image information, audio information, etc.

[0190] The translation model calling module 112 is used to call the translation model to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information.

[0191] The second information can be any of the following: text information, image information, audio information, etc., and the language type (i.e., language) corresponding to the second information is different from the language corresponding to the first information.

[0192] The calibration model calling module 113 is used to call the calibration model to calibrate the second information and obtain the calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, the target translation information corresponding to the second original information, and the translation calibration information corresponding to the second original information.

[0193] The calibration results described above are used to characterize the accuracy of the second information. Specifically, this accuracy can be characterized by accuracy information. For example, when there is a difference between the semantics or context of the second information and the semantics or context of the first information, the accuracy information can be characterized as the accuracy of the second information being less than 100%. Optionally, the accuracy information can also be represented by a score; the higher the score, the higher the accuracy of the second information, and vice versa.

[0194] The new translation sample determination module 114 is used to take the first information and the translation calibration information corresponding to the first information as new translation samples when the calibration result includes translation calibration information corresponding to the first information. The new translation samples are used to optimize and train the translation model.

[0195] In this embodiment, since the translation calibration information of the first information is output by the calibration model after calibrating the first and second information, and the calibration model is a trained neural network model, the translation calibration information of the first information does not require manual calibration of the first and second information. Therefore, the efficiency of obtaining the translation calibration information of the first information is improved. Simultaneously, since new translation samples are used to optimize the training of the translation model, and the new translation samples are the first information and the corresponding translation calibration information, training the translation model with new translation samples is equivalent to reinforcing the translation model with more accurate translation samples. This helps improve the accuracy of the target translation information output by the reinforced translation model, i.e., improves the accuracy of the translation result output by the translation model.

[0196] In some embodiments, when the new translation sample determination module 114 uses the first information and the translation calibration information corresponding to the first information as the new translation sample, it is specifically used for:

[0197] The aforementioned first information and the aforementioned translation calibration information corresponding to the aforementioned first information are used as positive samples in the new aforementioned translation samples;

[0198] The first and second pieces of information mentioned above are used as negative samples in the new translation samples.

[0199] In some embodiments, the information processing apparatus 11 provided in this application further includes:

[0200] The translation model optimization training module is used to optimize and train the translation model using the positive and negative samples in the new translation samples after the first information and the translation calibration information corresponding to the first information are used as new translation samples, when the number of positive and / or negative samples in the new translation samples is greater than a preset number threshold.

[0201] Optionally, considering that the probability of translation errors in the translation model decreases as the number of training iterations increases, and the severity of such errors also decreases accordingly, the aforementioned preset quantity threshold can be dynamically adjusted based on the number of training iterations of the translation model. For example, the preset quantity threshold can be set to be positively correlated with the number of training iterations of the translation model, meaning that the preset quantity threshold increases as the number of training iterations of the translation model increases.

[0202] In some embodiments, the information processing apparatus 11 provided in this application further includes:

[0203] The translation model supervision and fine-tuning module is used to supervise and fine-tune the translation model according to the application scenario of the translation model before the translation model is called to process the first information.

[0204] Correspondingly, the translation model calling module 112 mentioned above is specifically used for:

[0205] The above-mentioned first information is translated by calling the supervised fine-tuned translation model.

[0206] Optionally, the information processing apparatus 11 provided in this application embodiment further includes:

[0207] The calibration model supervision and fine-tuning module is used to supervise and fine-tune the calibration model according to the application scenario of the calibration model.

[0208] Correspondingly, the calibration model calling module 113 mentioned above is specifically used for:

[0209] The supervised fine-tuned calibration model is invoked to calibrate the second piece of information mentioned above.

[0210] In some embodiments, the new translation sample determination module 114 described above is specifically used for:

[0211] If the above calibration result is correct and the above calibration result includes translation calibration information corresponding to the above first information, the above first information and the above translation calibration information corresponding to the above first information shall be used as the new above translation sample.

[0212] In some embodiments, the information processing apparatus 11 provided in this application further includes:

[0213] The recommended translation calibration information acquisition module is used to acquire recommended translation calibration information corresponding to the first information in the event of an error in the calibration result.

[0214] The calibration model optimization training module is used to take the first information, the second information, and the recommended translation calibration information as new calibration samples, and the new calibration samples are used to optimize and train the calibration model.

[0215] In some embodiments, the first information is text information; the information processing method is applied to a client, and the information processing apparatus 11 provided in this application embodiment further includes:

[0216] The translation interface display module is used to display a translation interface before obtaining the first information to be translated. The translation interface includes a language selection control and an input area. The language selection control is used to determine the source language and the target language, and the input area is used to input the first information.

[0217] Correspondingly, the first information acquisition module 111 is specifically used to: acquire the first information input in the input area;

[0218] Correspondingly, the translation model calling module is specifically used to: call the translation model to perform text translation processing on the first information according to the source language and target language indicated by the language selection control.

[0219] Optionally, the translation interface may also include a file receiving control.

[0220] In some embodiments, the translation interface further includes a user feedback control associated with the second information; the information processing device 11 provided in this application embodiment further includes:

[0221] The user feedback window display module is used to display a user feedback window when the aforementioned user feedback control is triggered. The data entered in the aforementioned user feedback window is used as the target translation information of the aforementioned first information, and the aforementioned first information and the target translation information of the aforementioned first information are used as new translation samples.

[0222] In some embodiments, the user feedback window further includes at least two error type selection options. When an error type selection option is selected, it indicates that the second information contains an error type corresponding to the selected error type selection option.

[0223] When the aforementioned user feedback window display module uses the aforementioned first information and the target translation information of the aforementioned first information as new translation samples, it is specifically used to: when the error type of the aforementioned second information is the target error type, use the aforementioned first information and the target translation information of the aforementioned first information as new translation samples.

[0224] In some embodiments, the information processing device 11 provided in this application is applied to a client, and the first information acquisition module 111 is specifically used for:

[0225] Upon receiving a translation trigger instruction, the selected data in the currently displayed interface of the aforementioned client is obtained as the first information to be translated.

[0226] In some embodiments, the information processing device 11 provided in this application is applied to a server, and the first information acquisition module 111 is specifically used for:

[0227] Receive the first message to be translated sent by the client;

[0228] The information processing device 11 provided in this application embodiment further includes:

[0229] The second information feedback module is used to provide feedback of the second information to the client after obtaining the second information.

[0230] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0231] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 12 As shown, the electronic device 12 of this embodiment includes: at least one processor 120 ( Figure 12 The diagram shows only one processor, a memory 121, and a computer program 122 stored in the memory 121 and executable on the at least one processor 120, wherein the processor 120 executes the computer program 122 to implement the steps in any of the above method embodiments.

[0232] The electronic device 12 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. This electronic device may include, but is not limited to, a processor 120 and a memory 121. Those skilled in the art will understand that... Figure 12This is merely an example of electronic device 12 and does not constitute a limitation on electronic device 12. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0233] The processor 120 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0234] In some embodiments, the memory 121 may be an internal storage unit of the electronic device 12, such as a hard disk or memory of the electronic device 12. In other embodiments, the memory 121 may be an external storage device of the electronic device 12, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 12. Furthermore, the memory 121 may include both internal and external storage units of the electronic device 12. The memory 121 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 121 can also be used to temporarily store data that has been output or will be output.

[0235] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0236] This application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0237] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps in the above-described method embodiments.

[0238] This application provides a computer program product that, when run on an electronic device, enables the electronic device to implement the steps described in the various method embodiments above.

[0239] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographic device / electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0240] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0241] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0242] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0243] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0244] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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. Such 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 this application, and should all be included within the protection scope of this application.

Claims

1. An information processing method, characterized in that, include: Obtain the first piece of information to be translated; The translation model is invoked to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information. The calibration model is invoked to calibrate the second information to obtain a calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, target translation information corresponding to the second original information, and translation calibration information corresponding to the second original information. If the calibration result includes translation calibration information corresponding to the first information, the first information and the translation calibration information corresponding to the first information are used as new translation samples, wherein the new translation samples are used to optimize and train the translation model.

2. The information processing method as described in claim 1, characterized in that, The step of using the first information and the translation calibration information corresponding to the first information as the new translation sample includes: The first information and the translation calibration information corresponding to the first information are used as positive samples in the new translation samples; The first information and the second information are used as negative samples in the new translation sample.

3. The information processing method as described in claim 2, characterized in that, After taking the first information and the translation calibration information corresponding to the first information as the new translation sample, the method further includes: If the number of positive and / or negative samples in the new translation samples is greater than a preset threshold, the translation model is optimized and trained using the positive and negative samples in the new translation samples.

4. The information processing method as described in claim 1, characterized in that, Before invoking the translation model to translate the first information, the method further includes: The translation model is supervised and fine-tuned according to the application scenario of the translation model. The process of calling the translation model to translate the first information includes: The first information is translated by invoking the supervised, fine-tuned translation model.

5. The information processing method according to any one of claims 1 to 4, characterized in that, When the calibration result includes translation calibration information corresponding to the first information, using the first information and the translation calibration information corresponding to the first information as the new translation sample includes: If the calibration result is correct and the calibration result includes translation calibration information corresponding to the first information, the first information and the translation calibration information corresponding to the first information are used as the new translation sample.

6. The information processing method according to any one of claims 1 to 5, characterized in that, Also includes: In the event of an error in the calibration result, obtain recommended translation calibration information corresponding to the first information; The first information, the second information, and the recommended translation calibration information are used as new calibration samples, which are then used to optimize and train the calibration model.

7. The information processing method according to any one of claims 1 to 6, characterized in that, The first information is text information; the information processing method is applied to the client and, before obtaining the first information to be translated, further includes: The translation interface includes a language selection control and an input area. The language selection control is used to determine the source language and the target language, and the input area is used to input the first information. The step of obtaining the first information to be translated includes: obtaining the first information input in the input area; The step of calling the translation model to translate the first information includes: calling the translation model to perform text translation processing on the first information according to the source language and target language indicated by the language selection control.

8. The information processing method as described in claim 7, characterized in that, The translation interface also includes a user feedback control associated with the second information; the information processing method further includes: When the user feedback control is triggered, a user feedback window is displayed, and the data entered in the user feedback window is used as the target translation information of the first information, and the first information and the target translation information of the first information are used as the new translation sample.

9. The information processing method as described in claim 8, characterized in that, The user feedback window also includes at least two error type selection options. When an error type selection option is selected, it indicates that the second information contains an error type corresponding to the selected error type selection option. The step of using the first information and the target translation information of the first information as the new translation sample includes: when the error type of the second information is the target error type, using the first information and the target translation information of the first information as the new translation sample.

10. The information processing method according to any one of claims 1 to 6, characterized in that, The information processing method is applied to the client, and the step of obtaining the first information to be translated includes: Upon receiving a translation trigger instruction, the selected data in the currently displayed interface of the client is obtained as the first information to be translated.

11. The information processing method according to any one of claims 1 to 6, characterized in that, The information processing method is applied to the server side, and the step of obtaining the first information to be translated includes: Receive the first message to be translated sent by the client; After obtaining the second information, the process further includes: feeding back the second information to the client.

12. An information processing device, characterized in that, include: The first information acquisition module is used to acquire the first information to be translated. The translation model invocation module is used to invoke a translation model to translate the first information to obtain the second information. The translation model is a neural network model trained based on translation samples. The translation samples include the first original information and the target translation information corresponding to the first original information. The calibration model invocation module is used to invoke the calibration model to calibrate the second information and obtain the calibration result. The calibration model is a neural network model trained based on calibration samples. The calibration samples include the second original information, target translation information corresponding to the second original information, and translation calibration information corresponding to the second original information. A new translation sample determination module is used to, when the calibration result includes translation calibration information corresponding to the first information, use the first information and the translation calibration information corresponding to the first information as new translation samples, wherein the new translation samples are used to optimize the training of the translation model.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 11.

14. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 11.

15. A computer program product, characterized in that, Includes a computer program, which, when run, causes the method as described in any one of claims 1 to 11 to be performed.

Images