Information processing method and related apparatus

Abstract

This application discloses an information processing method and related apparatus. When a new translation method for target information emerges, prompt information for identifying the new translator can be constructed. During training, multiple sample information for various translation methods corresponding to the target information is provided. The sample information and prompt information are simultaneously input into an initial translation model to generate a pending translation result. Using the prompt information, the initial translation model can quickly locate the target information in the sample information and translate the target information through information substitution. By utilizing the difference between the pending translation result and the sample translation result, the initial translation model can learn how to select the correct translation method to translate the target information. Thus, this method enables the initial translation model to execute new translation methods using prompt information with minimal impact on its own translation knowledge for information translation.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to information processing methods and related apparatus. Background Technology

[0002] Translation is one of the main processing methods in the field of information processing. Through translation, information can be translated into multiple languages ​​for application in a wide range of information domains.

[0003] Some information domains may have niche translation methods. For example, some information domains may have niche terms, and translations of these terms may only occur within that specific information domain. In related technologies, when training models for information translation, corresponding training samples are constructed for these niche translation methods. Based on these training samples, the model learns these niche information translation methods.

[0004] However, in the construction of training samples, the related technologies only provide the translation results of information translated using this information translation method for the model to learn. This can easily lead the model to only learn this information translation method and ignore other information translation methods that may be involved in the same information domain. As a result, the model over-translates the information, leading to low final translation accuracy. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides an information processing method that enables the model to accurately translate specific information without affecting its normal translation capabilities, thereby optimizing the translation effect.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, embodiments of this application disclose an information processing method, the method comprising:

[0008] Obtain first sample information and second sample information. Both the first sample information and the second sample information include target information. The target information corresponds to first information in the first sample translation result corresponding to the first sample information and to second information in the second sample translation result corresponding to the second sample information. The first information and the second information are different.

[0009] An initial translation model is obtained, which is used to translate the target information into the second information;

[0010] Using the initial translation model, the first sample information is translated according to the prompt information corresponding to the first sample information and the target information to generate a first pending translation result, and the second sample information is translated according to the second sample information and the prompt information to generate a second pending translation result, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0011] The model parameters corresponding to the initial translation model are adjusted based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain a translation model. The translation model is used to generate a translation result corresponding to the undetermined translation information based on the undetermined translation information including the target information and the prompt information.

[0012] Secondly, embodiments of this application disclose an information processing method, the method comprising:

[0013] Obtain the information to be translated;

[0014] Identify target information with corresponding prompt information in the information to be translated, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0015] The translation model generates a translation result corresponding to the information to be translated based on the information to be translated and the prompt information. The translation model is used to translate the information to be translated, and when translating the information to be translated, it determines whether to use the first information as the translation result corresponding to the target information based on the prompt information.

[0016] Thirdly, embodiments of this application disclose an information processing apparatus, which includes a first acquisition unit, a second acquisition unit, a first generation unit, and an adjustment unit:

[0017] The first acquisition unit is used to acquire first sample information and second sample information. Both the first sample information and the second sample information include target information. The target information corresponds to first information in the first sample translation result corresponding to the first sample information and to second information in the second sample translation result corresponding to the second sample information. The first information and the second information are different.

[0018] The second acquisition unit is used to acquire an initial translation model, which is used to translate the target information into the second information;

[0019] The first generation unit is configured to translate the first sample information according to the prompt information corresponding to the first sample information and the target information using the initial translation model, to generate a first pending translation result, and to translate the second sample information according to the second sample information and the prompt information, to generate a second pending translation result, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0020] The adjustment unit is used to adjust the model parameters corresponding to the initial translation model according to the difference between the first sample translation result and the first undetermined translation result, and the difference between the second sample translation result and the second undetermined translation result, to obtain a translation model. The translation model is used to generate a translation result corresponding to the undetermined translation information according to the undetermined translation information including the target information and the prompt information.

[0021] In one possible implementation, the second acquisition unit is specifically used for:

[0022] Obtain third sample information including the target information, wherein the target information corresponds to the second information in the third translation result of the sample corresponding to the third sample information;

[0023] The third sample information is translated using the initial model to generate a pending third translation result;

[0024] Based on the difference between the third translation result of the sample and the third translation result to be determined, the model parameters corresponding to the initial model are adjusted to obtain the initial translation model.

[0025] In one possible implementation, the second acquisition unit is specifically used for:

[0026] Based on the difference between the third translation result of the sample and the third translation result to be determined, the model parameters corresponding to the initial model are adjusted to obtain the translation module;

[0027] The initial translation model is constructed based on the translation module and the initial replacement module;

[0028] The first generation unit is specifically used for:

[0029] The initial replacement module determines whether to replace the target information in the first sample information according to the prompt information based on the first sample information, and outputs first pending information; and determines whether to replace the target information in the second sample information according to the prompt information based on the second sample information, and outputs second pending information.

[0030] The translation module translates the untranslated information in the first pending information to generate the first pending translation result, and translates the untranslated information in the second pending information to generate the second pending translation result.

[0031] The adjustment unit is specifically used for:

[0032] The model parameters corresponding to the initial replacement module are adjusted based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain a replacement module. The translation module and the replacement module are used to constitute the translation model.

[0033] In one possible implementation, the first sample information and the second sample information each have corresponding information domain tags, the information domain tags being used to identify the information domain to which the corresponding information belongs, and the first generation unit is specifically used for:

[0034] Using the initial translation model, the first sample information is translated based on the first sample information, the information domain label corresponding to the first sample information, and the prompt information corresponding to the target information to generate a first pending translation result; and the second sample information is translated based on the second sample information, the information domain label corresponding to the second sample information, and the prompt information to generate a second pending translation result.

[0035] In one possible implementation, the first acquisition unit is specifically used for:

[0036] Multiple sample information is acquired, and each of the multiple sample information has a corresponding sample translation result;

[0037] Key information is identified from the multiple sample information to determine the key information included in each of the multiple sample information. The key information has a greater representational effect on the information content of the sample information than the information in the sample information other than the key information.

[0038] The first sample information is determined as the sample information in which the target information is included in the corresponding key information among the plurality of sample information, and the target information corresponds to the first information in the corresponding sample translation result. The first sample information is determined as the sample information in which the target information is included in the corresponding key information among the plurality of sample information, and the target information corresponds to the second information in the corresponding sample translation result.

[0039] In one possible implementation, the target sample information is any one of the plurality of sample information, the target sample information is composed of a plurality of sub-information, and the target sub-information is any one of the plurality of sub-information; the first acquisition unit is specifically used for:

[0040] Based on the target sample information, a combination of one or more weights is determined for the target sub-information, including position weight, word frequency weight, similarity weight, and information number weight. The position weight is used to characterize the position of the target sub-information in the target sample information, the word frequency weight is used to characterize the frequency of the target sub-information in the target sample information, the similarity weight is used to characterize the information similarity between the target sub-information and other information in the target sample information, and the information number weight is used to characterize the proportion of sample information including the target sub-information in the multiple sample information.

[0041] Based on a combination of one or more weights, key parameters corresponding to the target sub-information are determined, and the key parameters are used to characterize the criticality of the target sub-information;

[0042] Based on the key parameters corresponding to the multiple sub-information items, the key information included in the target sample information is determined.

[0043] In one possible implementation, the information domain corresponding to the target sample information is the game domain, and the combination of one or more weights includes the similarity weight and the word frequency weight. The first acquisition unit is specifically used for:

[0044] A first weight adjustment is performed on the similarity weight, and a second weight adjustment is performed on the word frequency weight. The first weight adjustment is used to increase the influence of the similarity weight on the key parameters corresponding to the target sub-information, and the second weight adjustment is used to reduce the influence of the word frequency weight on the key parameters corresponding to the target sub-information.

[0045] Based on the adjusted combination of one or more weights, the key parameters corresponding to the target sub-information are determined.

[0046] Fourthly, embodiments of this application disclose an information processing apparatus, which includes a third acquisition unit, a first determination unit, and a second generation unit:

[0047] The third acquisition unit is used to acquire the information to be translated;

[0048] The first determining unit is used to determine target information with corresponding prompt information in the information to be translated, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0049] The second generation unit is configured to generate a translation result corresponding to the information to be translated based on the information to be translated and the prompt information using a translation model. The translation model is configured to translate the information to be translated and, when translating the information to be translated, determine whether to use the first information as the translation result corresponding to the target information based on the prompt information.

[0050] In one possible implementation, the translation model includes a replacement module and a translation module, wherein the second generation unit is specifically used for:

[0051] The replacement module determines whether to replace the target information in the information to be translated according to the prompt information, and outputs pending information.

[0052] The translation module translates the untranslated information in the pending information and generates the translation result corresponding to the information to be translated.

[0053] In one possible implementation, the device further includes a second determining unit:

[0054] The second determining unit is used to determine the information domain label corresponding to the information to be translated, wherein the information domain label is used to identify the information domain corresponding to the information to be translated;

[0055] The second generation unit is specifically used for:

[0056] The translation model generates a translation result corresponding to the information to be translated based on the information to be translated, the information domain label, and the prompt information. The translation model is used to determine whether to use the first information as the translation result corresponding to the target information based on the prompt information when translating the information to be translated, according to the information to be translated and the information domain label.

[0057] In one possible implementation, the first determining unit is specifically used for:

[0058] The key information to be translated is identified, and the key information corresponding to the key information is determined. The key information has a greater representational effect on the information content of the information to be translated than the information other than the key information in the information to be translated.

[0059] Obtain an information set, which includes multiple candidate information items, each of which has a corresponding prompt message;

[0060] Among the multiple candidate information, the candidate information that exists in the key information is determined as the target information.

[0061] In one possible implementation, the first determining unit is specifically used for:

[0062] Determine the target information domain corresponding to the information to be translated;

[0063] Obtain the information set corresponding to the target information domain, and use the prompt information corresponding to the multiple candidate information to identify the translation relationship of the multiple candidate information under the target information domain.

[0064] In one possible implementation, the apparatus further includes a fourth acquisition unit, an extraction unit, a third determination unit, a fourth determination unit, and a fifth determination unit:

[0065] The fourth acquisition unit is used to acquire multiple pieces of information to be analyzed, all of which include the target information. The target information corresponds to the first information in the translation results that are respectively corresponding to the multiple pieces of information to be analyzed.

[0066] The extraction unit is used to extract benchmark features that characterize the common information features of the multiple pieces of information to be analyzed.

[0067] The third determining unit is used to determine the similarity between the information features corresponding to the information to be translated and the benchmark features;

[0068] The fourth determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked based on the fact that the translation result corresponding to the information to be translated includes the first information and the similarity is less than the similarity threshold.

[0069] The fifth determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked based on the fact that the translation result corresponding to the information to be translated does not include the first information and the similarity is not less than the similarity threshold. The translation result to be checked is used to provide the checking party for verification.

[0070] In one possible implementation, the translation relationship between the target information and the first information is a translation relationship existing in the target information domain, and the device further includes a sixth determining unit, a seventh determining unit, and an eighth determining unit:

[0071] The sixth determining unit is used to determine the information domain corresponding to the information to be translated;

[0072] The seventh determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked, based on the information domain corresponding to the information to be translated being the target information domain and the translation result corresponding to the information to be translated not including the first information.

[0073] The eighth determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked based on the fact that the information domain corresponding to the information to be translated is not the target information domain and the translation result corresponding to the information to be translated includes the first information. The translation result to be checked is used to provide the verification party for verification.

[0074] Fifthly, embodiments of this application disclose a computer device, the computer device including a processor and a memory:

[0075] The memory is used to store computer programs and to transfer the computer programs to the processor;

[0076] The processor is configured to execute the information processing method described in any one of the first aspects, or to execute the information processing method described in any one of the second aspects, according to the instructions in the computer program.

[0077] Fourthly, embodiments of this application disclose a computer-readable storage medium for storing a computer program, the computer program being used to execute the information processing method described in any one of the first aspects, or to execute the information processing method described in any one of the second aspects;

[0078] Fifthly, embodiments of this application disclose a computer program product including a computer program, which, when run on a computer device, causes the computer device to execute the information processing method described in any one of the first aspects, or to execute the information processing method described in any one of the second aspects.

[0079] As can be seen from the above technical solution, in order for the initial translation model, which already possesses translation capabilities, to learn new translation methods for target information, multiple sample information including the target information can be acquired first. The translation results corresponding to the target information differ in different sample information. Specifically, translating the target information into second information represents the existing translation capability of the initial translation model, while translating the target information into first information represents a translation capability not yet possessed by the initial translation model. When inputting multiple sample information into the initial translation model for training, prompt information identifying the translation relationship between the target information and the first information can be simultaneously input. Thus, the initial translation model can quickly locate the target information in the sample information based on this prompt information, replace the target information with the first information based on the prompt information to complete the translation of the target information, and translate other untranslated information based on its own information translation capabilities, obtaining the pending translation results corresponding to each of the multiple sample information. Based on the differences between sample translation results and the results to be translated, the initial translation model can learn which types of information require the translation relationship indicated by the prompt information to replace the target information, and which types of information require its own information translation capabilities to complete the translation. Thus, with minimal impact on the initial translation model's own information translation capabilities, the initial translation model can accurately translate the target information into the first information in reasonable scenarios. Consequently, the resulting translation model can accurately process various translation methods of the target information based on the information to be translated and the prompt information, enabling the translation model to accurately translate various types of information including the target information. This facilitates the application of the translation model in a wider range of translation scenarios. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0081] Figure 1 A schematic diagram illustrating an information processing method in a practical application scenario provided by an embodiment of this application;

[0082] Figure 2 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0083] Figure 3 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0084] Figure 4A schematic diagram illustrating an information processing method provided in an embodiment of this application;

[0085] Figure 5 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0086] Figure 6 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0087] Figure 7 A schematic diagram illustrating an information processing method provided in an embodiment of this application;

[0088] Figure 8 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0089] Figure 9 A schematic diagram illustrating an information processing method provided in an embodiment of this application;

[0090] Figure 10 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0091] Figure 11 A schematic diagram illustrating an information processing method provided in an embodiment of this application;

[0092] Figure 12 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0093] Figure 13 A schematic diagram illustrating an information processing method provided in an embodiment of this application;

[0094] Figure 14 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0095] Figure 15 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0096] Figure 16 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0097] Figure 17 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0098] Figure 18 A schematic diagram illustrating an information processing method provided in an embodiment of this application;

[0099] Figure 19 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0100] Figure 20A flowchart illustrating an information processing method provided in an embodiment of this application;

[0101] Figure 21 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0102] Figure 22 A flowchart illustrating an information processing method in a practical application scenario provided in this application embodiment;

[0103] Figure 23 A schematic diagram illustrating an information processing method in a practical application scenario provided by an embodiment of this application;

[0104] Figure 24 A structural block diagram of an information processing device provided in an embodiment of this application;

[0105] Figure 25 A structural block diagram of an information processing device provided in an embodiment of this application;

[0106] Figure 26 A structural diagram of a terminal provided in an embodiment of this application;

[0107] Figure 27 This is a structural diagram of a server provided in an embodiment of this application. Detailed Implementation

[0108] The embodiments of this application will now be described with reference to the accompanying drawings.

[0109] To achieve efficient information translation, researchers have trained various translation models for automated translation. Simply input the information to be translated into the translation model to obtain the corresponding translation result.

[0110] In related technologies, to train the translation capabilities of a translation model, sample information can be input into the model to be trained. The model parameters are then adjusted based on the difference between the translation output of the model and the corresponding sample translation, enabling the model to learn how to accurately translate the sample information. When a new and less common translation method appears for a given piece of information, researchers need to construct new training samples for this new translation method and train the model based on these samples, allowing the model to learn the new translation method.

[0111] However, this training method can lead to changes in the model's existing knowledge for information translation when learning new translation methods. Since the new translation methods may differ significantly from the old ones—for example, the new translation method may be a unique translation method for a niche information domain—the adjustment of model parameters may be too drastic. This can cause the model to lose some of its original knowledge, making it easy to translate all information using the new translation method, resulting in over-translation and inaccurate translation. It is also impossible to combine multiple translation methods to accurately translate information in multiple information domains.

[0112] To address the aforementioned technical problems, this application provides an information processing method. When a new translation method for target information emerges, a computer device can construct prompting information to identify the translation relationship between the target information and the translation result obtained through the new translation method. During training, multiple sample information for various translation methods corresponding to the target information and the corresponding sample translation results are provided. The sample information and prompting information are simultaneously input into an initial translation model to generate a pending translation result. Using the prompting information, the initial translation model can quickly locate the target information in the sample information and translate the target information through information substitution. By utilizing the difference between the pending translation result and the sample translation result, the initial translation model can be trained to determine in what information the prompting information is needed to translate the target information, and in what information it needs to rely on its own translation knowledge to translate the target information. Thus, this method enables the initial translation model to accurately translate based on the new translation method with minimal impact on its own translation knowledge used for information translation. The resulting translation model can accurately translate target information in various types of information by combining multiple translation methods for the target information.

[0113] Understandably, this method can be applied to computer devices capable of information processing, such as terminal devices or servers. The method can be executed independently by a terminal device or server, or it can be applied to network scenarios where a terminal device and a server communicate, executing in cooperation. Terminal devices can be mobile phones, tablets, laptops, desktop computers, etc. Terminal devices can also include various virtual reality devices, such as augmented reality (AR) devices like AR glasses and AR screens, and virtual reality (VR) devices like VR headsets. The server can be understood as an application server or a web server. In actual deployment, the server can be a standalone server, a cluster server, or a cloud server, etc.

[0114] To facilitate understanding of the technical solution provided in this application, the information processing method provided in this application will be introduced next in conjunction with a practical application scenario.

[0115] See Figure 1 , Figure 1 This is a schematic diagram of an information processing method in a practical application scenario provided by an embodiment of this application. In this practical application scenario, the computer device is a server 101 with information processing function.

[0116] The target information can be any type of information. The initial translation model is a model with information translation capabilities. Through the initial translation model, the target information can be translated into a second type of information. Translating the target information into the first type of information is a new translation method that needs to be learned during this model training. Server 101 can obtain the first sample information and the second sample information, both of which contain the target information. The first sample information corresponds to the first sample translation result, in which the target information is translated into the first type of information. The second sample information corresponds to the second sample translation result, in which the target information is translated into the second type of information.

[0117] To enable the initial translation model to learn which information should be translated into the first information, server 101 can input the first and second sample information as training samples into the initial translation model, and simultaneously input prompt information to the initial translation model. This prompt information is used to identify the translation relationship between the target information and the first information. Thus, through this prompt information, the initial translation model can locate the target information in the input information and directly translate the target information by replacing it with the first information in the prompt information, without needing to understand the content of the target information, and therefore without needing to understand how to translate the target information. Combining the above information, the initial translation model can generate a first undetermined translation result corresponding to the first sample information and a second undetermined translation result corresponding to the second sample information.

[0118] Since all information except the target information can be accurately translated using the initial translation model's own translation capabilities, the differences between the first pending translation result and the first sample translation result, and between the second pending translation result and the second sample translation result, can characterize whether the initial translation model can accurately determine whether to replace the target information with prompts or to translate the target information using existing model translation knowledge. Based on these differences, parameter adjustments can be made to the initial translation model. By narrowing these differences, the initial translation model can learn which types of target information require replacement with prompts and which types require translation using its own model translation knowledge. This allows the initial translation model to acquire new translation capabilities for accurately translating the target information into the first information with minimal impact on its original information translation capabilities. Ultimately, the resulting translation model can accurately translate target information in different types of information using matching translation methods, optimizing the model's translation capabilities for information with multiple translation results while ensuring translation accuracy.

[0119] Next, the technical solution provided in this application will be described in detail with reference to the accompanying drawings.

[0120] See Figure 2 , Figure 2 A flowchart of an information processing method provided in this application embodiment. In this embodiment, the computer device can be any type of computer device with information processing function. The method includes:

[0121] S201: Obtain the information of the first sample and the second sample.

[0122] To enable the model to learn how to accurately translate information with multiple translation methods, the computer can construct training samples for each translation method. Taking target information as an example, the target information can be any type of information with multiple translation methods. These multiple translation methods include translating the target information into a first type of information and translating the target information into a second type of information, where the first and second types of information are different.

[0123] The computer device can acquire first sample information and second sample information for the target information. Both the first and second sample information include the target information. The target information corresponds to the first information in the first sample translation result corresponding to the first sample information, and to the second information in the second sample translation result corresponding to the second sample information. Thus, the two sample information can represent two translation methods of the target information. The languages ​​corresponding to the sample information and the sample translation results can include multiple languages, which are not limited here.

[0124] S202: Obtain the initial translation model.

[0125] The initial translation model is a model that already possesses certain translation capabilities. The model knowledge within the initial translation model can be used for information translation. For the target information, the initial translation model can be used to translate the target information into second information—that is, translating the target information into second information is a capability already possessed by the initial translation model—while translating the target information into first information is a capability not yet possessed by the initial translation model.

[0126] S203: Using the initial translation model, translate the first sample information based on the prompt information corresponding to the first sample information and the target information to generate a first pending translation result, and translate the second sample information based on the second sample information and the prompt information to generate a second pending translation result.

[0127] Computer devices can construct prompts corresponding to target information. These prompts identify the translation relationship between the target information and the first information, indicating that the target information can be translated into the first information. For example, the prompt could be "The target information can be translated into the first information."

[0128] If no prompts are provided to the initial translation model, training the initial translation model with the first sample information and the first sample translation result will affect the model parameters used to translate the target information in the initial translation model. These model parameters were originally intended to accurately translate the target information into the second information. If these model parameters are adjusted to make the initial translation model translate the target information into the first information, the model will be forced to learn two different translation methods through the same set of model parameters. This can easily cause the model to lose its original ability to translate the target information into the second information, resulting in confused model knowledge and an inability to accurately translate either translation method.

[0129] To avoid the aforementioned technical problems, when training a model using both types of training samples, the computer device provides the prompt information as input to the initial translation model. This allows the initial translation model to output the desired translation result based on both the sample information and the prompt information. Figure 1As shown in the figure. The advantage of this approach is that, through the prompt information, the initial translation model can quickly locate the target information in the sample information and directly replace the target information based on the first information in the prompt information to complete the translation of the target information. This process only involves information replacement processing and does not involve information translation processing. The model parameters used to perform information replacement and the model parameters used to perform information translation can be independent of each other. Therefore, the purpose of this application is to adjust the model parameters for performing information replacement processing so that the model can accurately analyze whether it is necessary to replace the target information with the first information based on the prompt information without adjusting the model parameters used for performing information translation processing in the initial translation model. This reduces the impact on the original model capability of the initial translation model for information translation and effectively preserves the initial translation model's ability to translate the target information into the second information.

[0130] Specifically, for the first sample information, the initial translation model can combine the prompt information to obtain the first pending translation result. For the second sample information, it can combine the prompt information to obtain the second pending translation result. In these two translation processes, the initial translation model can use its own model parameters to decide whether to translate the target information through the prompt information. Before adjusting the model parameters, the initial translation model may use the prompt information incorrectly, which may lead to a difference between the pending translation result and the sample translation result.

[0131] S204: Adjust the model parameters corresponding to the initial translation model based on the differences between the translation results of the first sample and the first undetermined translation results, as well as the differences between the translation results of the second sample and the second undetermined translation results, to obtain the translation model.

[0132] Since the prompt information does not involve information other than the target information, all information in the sample information except the target information can be translated using the model knowledge of the initial translation model itself. When constructing the sample information, this information does not involve multiple translation methods. Therefore, the difference between the pending translation result and the sample translation result has a small impact, so the difference between the pending translation result and the sample translation result can be mainly focused on the translation result for the target information.

[0133] Based on this, the difference between the translation result of the first sample and the first undetermined translation result can characterize whether the initial translation model reasonably utilizes the prompt information to translate the target information in the first sample information; similarly, the difference between the translation result of the second sample and the second undetermined translation result can characterize whether the initial translation model reasonably utilizes the prompt information to translate the target information in the second sample information. Therefore, by adjusting the model parameters of the initial translation model through these differences, the initial translation model can learn which information needs to be directly replaced with the prompt information and which information needs to be translated using the model's own translation knowledge. This ultimately yields a translation model used to generate the translation result corresponding to the information to be translated based on the information to be translated, including the target information and the prompt information.

[0134] As can be seen from the above, this training method allows the translation model to effectively retain the original information translation capabilities of the initial translation model, while enabling it to rationally choose to translate using prompts and its own translation capabilities for different information. This allows it to accurately generate multiple different translation results for the same information, ensuring translation accuracy while improving the comprehensiveness of the translation model's capabilities, thus enabling the translation model to be applied in a wider range of application scenarios.

[0135] Next, we will introduce in detail the application process of the above translation model.

[0136] See Figure 3 , Figure 3 A flowchart illustrating an information processing method provided in this application embodiment. In this embodiment, the computer device can be any type of computer device with information processing capabilities. The method includes...

[0137] S301: Obtain the information to be translated.

[0138] The information to be translated can be any kind of information that needs to be translated, and the language and type of information involved can be multiple, without any limitation here.

[0139] S302: Identify the target information in the information to be translated that has corresponding prompt information.

[0140] As can be seen from the above model training method, this application can construct corresponding prompt information for information with multiple translation methods to suggest one or more translation methods for that information. After obtaining the information to be translated, the computer device can first determine the target information with corresponding prompt information. This target information is the information that needs to provide prompt information during the translation process so that the translation model can select a translation method for translation. The prompt information is used to identify the translation relationship between the target information and the first information, and the translation method from the target information to the first information is the translation method that the translation model needs to implement through the prompt information.

[0141] S303: Using a translation model, generate the translation result corresponding to the information to be translated based on the information to be translated and the prompt information.

[0142] like Figure 4 As shown, the computer device can input the information to be translated and the prompt information into the translation model to generate the translation result corresponding to the information to be translated. The translation model is used to translate the information to be translated, and during the translation process, it determines whether to use the prompt information as the translation result corresponding to the target information based on the information to be translated. That is, the translation model can analyze the information to be translated, determining whether to use the prompt information to replace the target information in the information to be translated, or to use model parameters related to information translation to translate the target information to generate other translation results besides the first information; for information parts in the information to be translated that do not have corresponding prompt information, it can directly use its own model parameters for information translation to complete the information translation, ultimately obtaining the translation result corresponding to the information to be translated. Therefore, it can be seen that, through this method, only prompt information related to the information to be translated is required to achieve accurate translation of information with multiple translation methods, which helps to improve the accuracy and comprehensiveness of information translation.

[0143] As mentioned above, the initial translation model used in this application during the training process has the ability to translate information. The following section will describe in detail how to train this initial translation model.

[0144] In one possible implementation, see Figure 5 When executing step S202, the computer device may execute steps S2021-S2023, where steps S2021-S2023 are a possible implementation of step S202, including:

[0145] S2021: Obtain third sample information including target information.

[0146] In this embodiment, the target information corresponds to the second information in the third translation result of the sample corresponding to the third sample information. In this embodiment, training samples with multiple large differences in translation results involving the same information will not be provided for training, so as to prevent the model from having difficulty learning effective translation knowledge for the same information. In the case of multiple large differences in translation methods for the same information, training can be carried out through the above-mentioned prompt information.

[0147] S2022: Translate the third sample information using the initial model to generate a pending third translation result.

[0148] The initial model can be any model with a model architecture that can be used for information translation. The undetermined third translation result is the translation result obtained by the initial model translating the information of the third sample.

[0149] S2023: Based on the difference between the third translation result of the sample and the third translation result to be determined, adjust the model parameters corresponding to the initial model to obtain the initial translation model.

[0150] The third sample translation result is the accurate translation result corresponding to the third sample information. Therefore, the difference between the third sample translation result and the pending third translation result can characterize the accuracy of the initial model in translating the third sample information. This accuracy is inversely correlated with the difference. Thus, the computer device can adjust the model parameters of the initial model based on this difference, making the pending third translation result output by the initial model gradually approach the third sample translation result. This allows the initial model to learn how to accurately translate the third sample information, that is, how to accurately translate the target information into second information, resulting in an initial translation model with information translation capabilities.

[0151] Next, we will introduce in detail the model's translation process of information, and how to adjust the model parameters for each part of the model.

[0152] First, in one possible implementation, see [link to relevant documentation]. Figure 6 When executing step S2023, the computer device may execute steps S20231-S20232, where steps S20231-S20232 are one possible implementation of step S2023, including:

[0153] S20231: Based on the difference between the third translation result of the sample and the third translation result to be determined, adjust the model parameters corresponding to the initial model to obtain the translation module.

[0154] As can be seen from the above, this translation module has a certain information translation capability, such as translating the target information in the information into second information.

[0155] S20232: The initial translation model is constructed based on the translation module and the initial replacement module.

[0156] like Figure 7 As shown, the computer device can add an initial replacement module to the translation module to form an initial translation model. The translation module is used to perform a process of directly translating the information when translating the information, and the initial replacement module is used to perform a process of determining whether to replace the information based on the prompt information when translating the information.

[0157] When executing step S203, the computer device may execute steps S2031-S2032, where steps S2031-S2032 are one possible implementation of step S203, including:

[0158] S2031: The initial replacement module determines whether to replace the target information in the first sample information according to the prompt information based on the first sample information, and outputs the first pending information; and determines whether to replace the target information in the second sample information according to the prompt information based on the second sample information, and outputs the second pending information.

[0159] When translating the first and second sample information, the initial replacement module can determine whether to replace the target information in the sample information based on the input information and output the processed pending information. Taking the first pending information as an example, the first pending information can be the first sample information itself without replacement processing, or it can be the information after replacing the target information in the first sample information with the first information. The second pending information is similar.

[0160] S2032: Through the translation module, translate the untranslated information in the first pending information to generate the first pending translation result, and translate the untranslated information in the second pending information to generate the second pending translation result.

[0161] Specifically, if the target information in the sample information is replaced by the first information by the initial replacement module, the first information in the pending information will be considered as already translated by the translation module, and the translation module will translate the other information besides the first information. If the initial replacement module does not replace the target information in the sample information, the entire pending information will be considered as untranslated by the translation module.

[0162] When performing step S204, the computer device may execute step S2041, where step S2041 is a possible implementation of step S204, including:

[0163] S2041: Adjust the model parameters corresponding to the initial replacement module based on the differences between the translation results of the first sample and the first undetermined translation results, as well as the differences between the translation results of the second sample and the second undetermined translation results, to obtain the replacement module.

[0164] Since the translation module has acquired relatively accurate and effective information translation capabilities through the aforementioned training process, the difference between the sample translation result and the pending translation result is mainly affected by the accuracy of the initial replacement module's processing of the target information, i.e., whether it performs reasonable information replacement for the target information in the sample information. Therefore, the computer device can adjust the model parameters corresponding to the initial replacement module based on the aforementioned differences in translation results, enabling the initial replacement module to learn how to accurately determine whether the target information in the information needs to be replaced through prompts, thus obtaining the replacement module. The translation module and the replacement module constitute the translation model. Because this method does not require adjustment of the model parameters corresponding to the translation module, it has a smaller impact on the information translation capability of the translation module. This allows the translation model to retain the information translation capability of the translation module, while the trained replacement module enables the translation model to acquire the ability to translate new information translation methods for the target information. Consequently, the translation model can accurately combine multiple information translation methods to translate the target information.

[0165] Next, we will introduce how to apply the translation model with the above model architecture.

[0166] See Figure 8 In one possible implementation, when executing step S303, the computer device may execute steps S3031-S3032, where steps S3031-S3032 are a possible implementation of step S303, including:

[0167] S3031: The replacement module determines whether to replace the target information in the information to be translated according to the prompt information, and outputs the pending information.

[0168] like Figure 9 As shown, the replacement module can analyze the semantics, context, and information architecture of the information to be translated to determine whether the target information should be replaced using the translation method indicated by the prompt information, and output the pending information. If the replacement module determines that replacement is required, the pending information is the information that replaces the target information in the information to be translated with the first information; if not, the pending information can be the information to be translated itself.

[0169] S3032: The translation module translates the untranslated information in the pending information and generates the translation result corresponding to the information to be translated.

[0170] The translation module can translate the untranslated portions of the given information, thus obtaining a translation result of the entire information. This approach combines the substitution module's accurate judgment of when to utilize prompts with the translation module's accurate translation capabilities, enabling the translation model to accurately translate target information across different types of information using various translation methods. This makes the translation model's translation function more accurate and comprehensive.

[0171] Understandably, one reason why information can be translated in different ways is that the same information may have different meanings in different information domains. Therefore, accurate translation requires accurately expressing the meaning of information in the corresponding information domain. For example, in the gaming field, there are various gaming terms with specific meanings, while in other information domains, these terms usually correspond to the meaning of the information itself. Consequently, the translation results of these gaming terms differ between the gaming and other fields.

[0172] Based on this, in one possible implementation, in order to enable the translation model to more accurately determine which information translation method to use for translating information with multiple translation methods, the computer device can suggest the information domain corresponding to the information to the model.

[0173] In this implementation, the first sample information and the second sample information each have corresponding information domain labels. These information domain labels identify the information domain to which the corresponding information belongs. The information domain can be any domain, such as gaming, sports, news, or academia. See also... Figure 10 When executing step S203, the computer device may execute step S2033, where step S2033 is a possible implementation of step S203, including:

[0174] S2033: Using the initial translation model, the first sample information is translated based on the first sample information, the information domain label corresponding to the first sample information, and the prompt information corresponding to the target information to generate a first pending translation result; and the second sample information is translated based on the second sample information, the information domain label corresponding to the second sample information, and the prompt information to generate a second pending translation result.

[0175] See Figure 11By providing information domain labels as input to the initial translation model during training, the model can analyze the information domain of the sample information and determine the most appropriate translation method for the target information within the sample. By comparing the output of the proposed translation with the sample translation, the initial model learns which information domains are more suitable for replacing the target information with prompts, and which are more suitable for direct translation. This allows the initial model to learn how to accurately select the translation method by combining the information domain and the information to be translated. Consequently, the translation model can utilize the information domain labels of the information to be translated for more accurate translation, accurately reflecting the characteristics of information translation where the same information requires different translation methods in different information domains.

[0176] Correspondingly, in one possible implementation, when using a translation model for information translation, see [link to relevant documentation]. Figure 12 The computer device can first execute step S1201, including:

[0177] S1201: Determine the information domain label corresponding to the information to be translated.

[0178] Information domain labels are used to identify the information domain corresponding to the information to be translated. The methods for determining information domain labels can be varied, such as manually labeling the information domain, or training a corresponding information analysis model to generate the labels; no specific limitations are specified here.

[0179] When executing step S303, the computer device may execute step S3033, where step S3033 is one possible implementation of step S303, including:

[0180] S3033: Using a translation model, generate the translation result corresponding to the information to be translated based on the information to be translated, the information domain label, and the prompt information.

[0181] The process of determining the translation result is as follows: Figure 13 As shown, the translation model can be used to determine whether to use the first piece of information as the target information for translation based on the information to be translated and the information domain label, according to the prompt information. Then, it combines its own model knowledge for information translation to translate the untranslated parts of the information to be translated. In this way, the information dimension used by the translation model when judging the translation method of the information to be translated can be increased, thereby enabling the translation model to combine more information to select the translation method and improve the accuracy of the translation method selection.

[0182] Next, we will explain in detail how to identify the information portion of the message that contains prompts.

[0183] Understandably, in information translation, the accuracy of translating key information that plays a crucial role in representing the content usually has the primary impact on the overall translation effect. For example, information typically consists of key information that represents the content, along with some less relevant modal particles, auxiliary words, and adjectives. In the information "The equipment system in a game is a relatively important system," the words "game" and "equipment system" are key information representing the content, and their accurate translation directly determines the translation effect. Words like "important" and "relatively" have a relatively smaller representational effect on the content; these words are usually easier to translate, have a lower probability of translation errors, and have a smaller impact on the translation effect.

[0184] Based on this, in one possible implementation, when training the model on information with multiple translation methods, the computer device can train the model based only on this key information, thereby enabling the model to have a relatively effective translation capability with a smaller training load.

[0185] See Figure 14 When executing step S201, the computer device may execute steps S2011-S2013, where steps S2011-S2013 are a possible implementation of step S201, including:

[0186] S2011: Obtain information from multiple samples.

[0187] Each sample information has a corresponding sample translation result, which is the accurate translation result corresponding to the sample information.

[0188] S2012: Identify key information from multiple sample information and determine the key information included in each sample information.

[0189] The key information identification process is used to identify key information within the information. The principle behind this application for key information identification is to analyze the representational effect of each component of the information on the information content, and identify information with a significant representational effect as key information. Various methods can be used for key information identification based on this principle, which will be described in detail below and will not be elaborated here. Through key information identification, the key information included in the sample information can be determined. The representational effect of key information on the information content of the sample information is greater than the representational effect of other information in the sample information.

[0190] S2013: Among multiple sample information, the sample information in which the corresponding key information includes target information and the target information corresponds to the first information in the corresponding sample translation result is determined as the first sample information; and among multiple sample information, the sample information in which the corresponding key information includes target information and the target information corresponds to the second information in the corresponding sample translation result is determined as the first sample information.

[0191] The processing equipment can analyze the key information corresponding to the sample information to see if the aforementioned target information with multiple translation methods exists. If it does, it indicates that the target information has a significant role in representing the content of the sample information, and therefore the accuracy of translating the target information directly affects the translation effect of the sample information. The computer equipment can train the initial translation model based on this sample information, enabling the initial translation model to accurately translate the target information in this part of the information. This ensures the translation effect while reducing the amount of sample information required for model training, thus improving model training efficiency.

[0192] The computer device can determine the sample information in which the corresponding key information includes target information and the target information corresponds to the first information in the sample translation result as the first sample information, and determine the sample information in which the corresponding key information includes target information and the target information corresponds to the first information in the sample translation result as the second sample information, so as to meet the training requirements for multiple translation methods of target information.

[0193] Next, we will introduce in detail a method for identifying key information.

[0194] It is understandable that the representational effect of information on information content can be manifested from multiple dimensions. For example, generally speaking, the earlier a piece of information appears in a complete set of information, the more frequently it appears, and the greater its difference from other information in the complete set of information, the greater its representational effect on the content of the complete set of information is likely to be. Based on this, in one possible implementation, computer devices can combine multiple analytical dimensions to analyze the content representation effect of information.

[0195] In this implementation, the target sample information can be any one of multiple sample information sets. The target sample information is composed of multiple sub-information sets, each of which is a portion of the target sample information. When dividing the sub-information, the length of the sub-information can vary based on different actual needs or the length characteristics of key information in different information domains. For example, it can contain only a single word or a phrase composed of multiple words; no limitation is made here. The target sub-information can be any one of multiple sub-information sets; see [link to relevant documentation]. Figure 15When executing step S2012, the computer device may execute steps S20121-S20123, where steps S20121-S20123 are a possible implementation of step S2012, including:

[0196] S20121: Based on the target sample information, determine a combination of one or more of the following weights: position weight, word frequency weight, similarity weight, and information number weight corresponding to the target sub-information.

[0197] Among these, position weight represents the position of the target sub-information within the target sample information. Generally, the position of information influences its representation of content; information appearing earlier usually has a greater representational effect and is therefore more crucial. Frequency weight represents the frequency of the target sub-information within the target sample information. A higher frequency indicates a larger proportion of the target sub-information's representation within the overall target sample information, thus indicating a greater representational effect. Similarity weight represents the similarity between the target sub-information and other information within the target sample information. Lower similarity indicates a more unique semantic meaning, thus increasing the likelihood of its crucial representational effect. Information quantity weight represents the proportion of sample information containing the target sub-information within multiple samples. A higher proportion indicates more samples containing the target sub-information, suggesting a more important role for the target sub-information in composing the information, and thus increasing the likelihood of its crucial representational effect. Therefore, the weights of these multiple dimensions can, to some extent, represent the criticality of the target sub-information in representing the content of the target sample information.

[0198] S20122: Determine the key parameters corresponding to the target sub-information based on a combination of one or more weights.

[0199] Computer equipment can combine one or more of the above weights to analyze the representation effect of target sub-information on the information content of target sample information, obtaining the key parameters corresponding to the target sub-information. The key parameters characterize the criticality of the target sub-information, and the criticality characterizes its representation effect on the information content of the target sample information. The greater the criticality, the more critical the information, and the greater its representation effect on the information content of the target sample information. Combining multiple weights allows for a more accurate analysis of the criticality of target sub-information.

[0200] For example, the formula for determining the key parameter S(w) can be as follows:

[0201]

[0202] The meanings of each parameter are as follows:

[0203] W Rel For similarity weights, W Position These are positional weights, both of which are inversely correlated with importance, and are placed at the position of the molecule. W case To address the case sensitivity of English information, W Freq For word frequency weighting, W DifSentence These three weights, which are positively correlated with the degree of importance, are placed in the denominator. Thus, the key parameters determined in this way are inversely correlated with the degree of importance and inversely correlated with the content representation effect of sub-information on sample information.

[0204] S20123: Determine the key information included in the target sample information based on the key parameters corresponding to multiple sub-information items.

[0205] Based on the key parameters corresponding to multiple sub-information items, the computer device can analyze the representational effect of each sub-information item on the target sample information, thereby identifying the sub-information items with the largest representational effect as the target information. For example, the target information can be selected from the n sub-information items whose key parameters are greater than preset parameter values ​​or whose key parameters are the largest; no limitation is imposed here. This method allows for the quantitative analysis of the representational effect of each sub-information item on the sample information from multiple dimensions, thus accurately identifying the key information with the largest content representational effect and improving the efficiency of sample information construction. Furthermore, this method identifies key information based solely on the information characteristics of the information itself, therefore it is less affected by the information domain and does not require auxiliary information outside the information to be identified. It has wide applicability, requires no complex preliminary preparation work, and has high information identification efficiency.

[0206] Since the characteristics of information may differ in different information domains, computer devices may also use different methods to identify key information in different information domains in order to improve the accuracy of computer devices in identifying key information in each information domain.

[0207] For example, in the gaming industry, key information often consists of gaming terminology such as "xx system," "xx attribute," and "xx gameplay." Because gaming terminology is typically unique and not commonly used in everyday language, it rarely appears repeatedly in a single piece of information, and its differences from other non-gaming terms within the information are usually significant. Therefore, in one possible implementation, for gaming-related information, computer devices can specifically adjust word frequency weights and similarity weights when determining key parameters to better suit the characteristics of key information in the gaming domain.

[0208] Taking target sample information as an example, in this implementation, the information domain corresponding to the target sample information is the game domain. The combination of one or more weights mentioned above can include similarity weight and word frequency weight. See [link to relevant documentation]. Figure 16 When executing step S20122, the computer device can execute steps S201221-S201222. Steps S201221-S201222 are one possible implementation of step S20122, including:

[0209] S201221: Perform the first weight adjustment on the similarity weight and the second weight adjustment on the word frequency weight.

[0210] The first weight adjustment is used to increase the influence of similarity weight on the key parameters corresponding to the target sub-information, and the second weight adjustment is used to reduce the influence of word frequency weight on the key parameters corresponding to the target sub-information. This weight adjustment method can effectively match the information characteristics of game terms as key information in the game field, thereby avoiding the inability to accurately identify game terms as key information due to their low word frequency, and highlighting the unique information characteristics of game terms, thus improving the accuracy of game term identification.

[0211] S201222: Determine the key parameters corresponding to the target sub-information based on the adjusted combination of one or more weights.

[0212] For example, in the formula for determining key parameters, similarity weight is positively correlated with the key parameters and inversely correlated with the representational effect of the content, while word frequency weight is inversely correlated with the key parameters and positively correlated with the representational effect of the content. Based on this, computer devices can reduce the similarity weight corresponding to game domain information based on the determined similarity weight, and reduce the word frequency weight corresponding to game domain information based on the determined word frequency weight, ultimately achieving accurate identification of key information in the game domain.

[0213] Correspondingly, the following section will introduce how to utilize key information to improve information translation efficiency on the model application side.

[0214] First, in one possible implementation, to further improve the efficiency of information translation, see [link to relevant documentation]. Figure 17 When executing step S302, the computer device may execute steps S3021-S3023, where steps S3021-S3023 are a possible implementation of step S302, including:

[0215] S3021: Identify key information in the information to be translated and determine the key information corresponding to the information to be translated.

[0216] Among these, key information plays a greater role in representing the content of the information to be translated than other information in the information to be translated. The methods for identifying key information are as described above and will not be repeated here. As can be seen from the above, the accuracy of the translation of key information has a significant impact on the overall accuracy of the information to be translated. Therefore, to improve the efficiency of information translation, computer equipment can only search for information with multiple translation methods within this key information and perform targeted translation, without analyzing the overall translation methods of the information to be translated.

[0217] S3022: Obtain the information set.

[0218] In order to identify information with multiple translation methods in the information to be translated, the computer device can pre-build an information set corresponding to multiple translation methods. This information set includes multiple candidate information, and each candidate information has corresponding prompt information. That is, the multiple candidate information are all information with multiple translation methods, and the translation methods of the candidate information need to be analyzed in combination with the prompt information.

[0219] S3023: Among multiple candidate information, the candidate information that exists in the key information is identified as the target information.

[0220] Computer equipment can compare key information with candidate information to determine the key information that includes the candidate information. The comparison can be performed in various ways, such as... Figure 18 In this process, computer devices store information sets in hash tables. By leveraging the hash table's ability to quickly insert and look up data, they can quickly determine whether each piece of key information exists in the hash table. This allows them to determine whether the key information includes candidate information, thereby improving the efficiency of identifying target information in the information to be translated.

[0221] As mentioned above, the analysis of information translation characteristics shows that the translation method is usually closely related to the information domain. Based on this, in one possible implementation, in order to further improve the translation accuracy, in addition to providing information domain labels, the computer device can also select the information domain prompt information corresponding to the information to be translated when providing prompt information, so that the translation model can translate the information in that information domain more accurately.

[0222] See Figure 19 When executing step S3022, the computer device can execute steps S30221-S30222. Steps S30221-S30222 are one possible implementation of step S3022, including:

[0223] S30221: Determine the target information domain corresponding to the information to be translated.

[0224] There are many ways to determine the information domain corresponding to information. For example, computer devices can identify the target information domain corresponding to the information to be translated by recognizing some key information that can characterize the information domain, or they can use a model for recognizing the information domain. This is not limited here.

[0225] S30222: Obtain the information set corresponding to the target information domain.

[0226] The prompts corresponding to multiple candidate pieces of information are used to identify the translation relationships between these candidate pieces of information within the target information domain. That is, within the target information domain, candidate pieces of information may be translated in the manner indicated by the prompts. Therefore, translating information based on the prompts corresponding to this set of information improves the accuracy and effectiveness of the prompts provided to the translation model, enabling the model to more accurately determine the translation methods for information within the target information domain.

[0227] For example, when the target information domain is the gaming domain, the computer device can obtain a gaming terminology list from the gaming domain. This gaming terminology list records some gaming terms that serve as key information, along with their corresponding translations. The computer device can then use this gaming terminology list as an information set to determine the target information that contains prompts.

[0228] The above content mainly introduces the information translation process. In order to further improve the translation effect, this application also provides a verification function for automatically verifying the translation results, as shown below.

[0229] First, it's understandable that the method of information translation is usually closely related to the context information corresponding to that information. When the context information corresponding to the same information in different contexts is similar, it indicates that the semantics expressed by the information in different contexts are quite similar, and therefore the translation method of that information in different contexts is usually consistent. Based on this, in one possible implementation, computer equipment can use this translation characteristic to verify whether the translation result of the information translation model is accurate.

[0230] See Figure 20 The computer device can execute steps S2001-S2005, including:

[0231] S2001: Obtain multiple pieces of information to be analyzed.

[0232] In this context, multiple pieces of information to be analyzed all include target information, and the target information corresponds to the first information in the respective translation results of each of the multiple pieces of information to be analyzed. That is, when translating multiple pieces of information to be analyzed, the target information must be translated into the first information. The information to be analyzed can be any information that meets the above requirements; no limitation is imposed here.

[0233] S2002: Extract benchmark features to characterize the common information features of multiple pieces of information to be analyzed.

[0234] Since multiple pieces of information to be analyzed require translating target information into first information, the commonalities in the features of these multiple pieces of information can characterize the information characteristics that need to be translated into first information. Furthermore, when the information features corresponding to the information are close to the baseline features, it indicates that the information is relatively close to these pieces of information to be analyzed, thus it can be judged that the target information in the information has a high probability of being translated into first information. The information features corresponding to the information can be multi-dimensional features, such as semantic features, information structure features, etc., as long as they can represent the information; there are no restrictions here.

[0235] S2003: Determine the similarity between the information features corresponding to the information to be translated and the baseline features.

[0236] Based on the above, computer equipment can extract information features corresponding to the information to be translated and analyze the similarity between the information features and the benchmark features. This similarity can characterize the information similarity between the information to be translated and multiple pieces of information to be analyzed.

[0237] S2004: Based on the fact that the translation result corresponding to the information to be translated includes the first information and the similarity is less than the similarity threshold, the translation result corresponding to the information to be translated is determined as the translation result to be checked.

[0238] Computer equipment can preset a similarity threshold, which measures whether two features have a high degree of similarity. If the similarity between the information feature corresponding to the information to be translated and the benchmark feature is less than the similarity threshold, it indicates that the similarity between the information to be translated and multiple pieces of information to be analyzed is low. That is, the contextual information of the target information differs greatly between the information to be translated and the multiple pieces of information to be analyzed. Under normal circumstances, the target information will not be translated as the first information. Therefore, if the similarity is less than the similarity threshold, and the translation result corresponding to the information to be translated still includes the first information, it is highly likely that the target information has been incorrectly translated as the first information. The computer equipment can use this translation result as the translation result to be checked, and the translation result to be checked is provided to the checker for verification.

[0239] S2005: Based on the fact that the translation result corresponding to the information to be translated does not include the first information and the similarity is not less than the similarity threshold, the translation result corresponding to the information to be translated is determined as the translation result to be checked.

[0240] Similarly, if the similarity is not less than the similarity threshold, it indicates a high degree of information similarity between the information to be translated and multiple pieces of information to be analyzed. The contextual information of the target information within these pieces of information is relatively close, therefore, the target information is more likely to be translated as the first piece of information. Thus, if the translation result corresponding to the information to be translated does not include the first piece of information, there is a high probability of an incorrect translation of the target information. The computer device can also determine the translation result corresponding to the information to be translated as the translation result to be checked by the verification party.

[0241] By analyzing the commonalities among multiple pieces of information corresponding to a particular translation method, computer equipment can predict the accurate translation method for the information to be translated. This effectively identifies potentially abnormal translation results, allowing for verification and further ensuring translation accuracy. Furthermore, the entire anomaly identification process can be completed automatically by the computer equipment without human intervention, thus improving the efficiency of translation result verification.

[0242] In addition to verifying based on commonalities of information, as mentioned above, the method of information translation is closely related to the information domain. Therefore, computer equipment can also use the information domain corresponding to the information to be translated as one of the dimensions for judging the accuracy of the translation result.

[0243] In this implementation, the translation relationship between the target information and the first information is a translation relationship existing within the target information domain. That is, within the target information domain, the probability of the target information being translated into the first information is higher; in other information domains, the probability of the target information being translated into the first information is lower. Based on this, see [link to relevant documentation]. Figure 21 The computer device can perform steps S2101-S2103, including:

[0244] S2101: Determine the information domain corresponding to the information to be translated.

[0245] The methods for determining the information domain have been introduced above and will not be repeated here.

[0246] S2102: Based on the information domain corresponding to the information to be translated being the target information domain, and the translation result corresponding to the information to be translated not including the first information, the translation result corresponding to the information to be translated is determined as the translation result to be checked.

[0247] If the information to be translated is within the target information domain, it indicates a high probability that the target information within the information to be translated will be translated as the first information. In this case, if the translation result corresponding to the information to be translated does not include the first information, there is a high probability of an incorrect translation of the target information. In this situation, the computer device can determine the translation result corresponding to the information to be translated as the translation result to be checked, and this translation result is provided to the checker for verification.

[0248] S2103: Based on the fact that the information domain corresponding to the information to be translated is not the target information domain, and the translation result corresponding to the information to be translated includes the first information, the translation result corresponding to the information to be translated is determined as the translation result to be checked.

[0249] Similarly, if the information to be translated is not within the domain of the target information, it means that the probability of the target information in the information to be translated into the first information is relatively low. In this case, if the translation result corresponding to the information to be translated includes the first information, there is a high probability that an incorrect translation of the target information has occurred. At this point, the computer device can determine the translation result corresponding to the information to be translated as the translation result to be checked.

[0250] By combining the information domain corresponding to the information to be translated with the corresponding translation method of the target information in the translation results, computer equipment can accurately analyze the accuracy of the translation results, thereby effectively filtering out translation results that may contain abnormalities, so that the verification party can conduct verification and further ensure the accuracy of information translation.

[0251] To facilitate understanding of the technical solution provided in this application, the information processing method provided in this application will be described in detail below, taking into account a practical application scenario.

[0252] See Figure 22 , Figure 22 This application provides a flowchart of an information processing method in a practical application scenario. In this scenario, the computer device can be any type of computer device with information processing capabilities, such as a mobile phone, computer, server, or other device capable of information processing. The method includes:

[0253] S2201: Obtain multiple sample information including target information.

[0254] The target information can be any type of information with multiple translation methods. The translation results corresponding to multiple sample information pieces each include multiple translation methods corresponding to the target information. For example, in some sample information pieces, the translation result corresponding to the target information is the first type of information; in other sample information pieces, the translation result corresponding to the target information is the second type of information, and so on. When acquiring sample information, the computer device can first identify the key information in the sample information using a key information recognition method, and then determine the sample information that can be used for model training based on the key information. In this implementation method, the target information can be information from the game domain, such as game terminology.

[0255] As shown in steps S22011-S22015, steps S22011-S22015 are one possible implementation of step S2201, including:

[0256] S22011: Obtain information on multiple candidate samples.

[0257] S22012: Determine whether the candidate sample information contains key information.

[0258] If key information exists, proceed to step S22013 to determine whether the key information includes the target information; if no key information exists, the candidate sample information can be discarded directly, or proceed to step S22015 to use the candidate sample information as sample information for training the translation module in the initial translation model. This part of the sample information will not provide prompts when inputting, so that the model does not need to learn multiple translation methods for the information in these sample information.

[0259] S22013: Determine whether the key information includes the target information.

[0260] The computer device can match key information and target information to determine whether the key information includes the target information. If it does, step S22014 can be executed to use the candidate sample information as the sample information corresponding to the target information. If it does not, the candidate sample information can be discarded, or step S22015 can be executed.

[0261] S22014: Use candidate sample information as the sample information corresponding to the target information.

[0262] S22015: Use candidate sample information as sample information for training the translation module.

[0263] S2202: Construct the prompt information corresponding to the target information.

[0264] This prompt is used to identify the translation relationship between the target information and the first information. Translating the target information into the first information is the new information translation method that the initial translation model needs to learn through model training.

[0265] S2203: Train the initial translation model using prompt information and multiple sample information to obtain the translation model.

[0266] The final information input to the initial translation model can be as follows: Figure 23 As shown, the specific model training method has been detailed above and will not be repeated here. This information allows adjustment of the model parameters corresponding to the replacement module in the initial translation model, which is used for information replacement based on prompts. This enables the translation model to accurately determine which target information needs to be translated using prompts and which needs to be translated using its own information translation capabilities. In addition to the above information, the computer device can also provide the information domain labels corresponding to the sample information as input to the initial translation model, allowing the initial translation model to learn to more accurately determine the information translation method based on the information domain identified by the information domain labels.

[0267] S2204: Obtain the information to be translated.

[0268] The information to be translated can be any information that needs to be translated. The translation of information in this application can involve multiple languages, and the source of the information to be translated can also include multiple sources, such as information extracted from text, information extracted from images, videos, etc. There are no limitations here.

[0269] S2205: Identify the key information corresponding to the information to be translated.

[0270] Among them, key information refers to information that plays a significant role in representing the content of the information to be translated. Its identification methods can be as shown above, or other language tools (such as language models) can be used to identify key information, which is not limited here.

[0271] S2206: Determine the target information domain corresponding to the information to be translated.

[0272] In this practical application scenario, the target information domain can be the gaming domain.

[0273] S2207: Obtain the information set corresponding to the target information domain.

[0274] This information set includes multiple candidate information corresponding to the target information domain. For example, it can be multiple game terms in the game domain. These candidate information all correspond to multiple translation methods and have corresponding prompts. The prompts are used to identify the translation results of these candidate information in the target information domain. For example, it can be a terminology list that can identify the correspondence between game terms and their corresponding translation results.

[0275] S2208: Identify the candidate information existing in the key information corresponding to the information to be translated as the target information.

[0276] Computer devices can construct hash tables based on information sets and use these hash tables to match candidate information within key information. This candidate information is the target information that needs to be considered during the translation process, and the translation model needs to accurately determine how to translate this target information. It is important to emphasize that the target information here can be the same as or different from the target information used in the model training process. Furthermore, the target information can be a single piece of information or multiple pieces of information; this is not limited here.

[0277] S2209: Input the information to be translated, the prompt information corresponding to the target information, and the information domain label corresponding to the information to be translated into the translation model to generate the translation result corresponding to the information to be translated.

[0278] The translation model can decide whether to replace the target information based on the prompt information or to translate the target information using its own information translation capabilities, based on the information to be translated and the information domain labels. In addition, the translation model can translate the untranslated information parts of the information to be translated other than the target information using its own information translation capabilities, and finally obtain the translation result corresponding to the information to be translated.

[0279] S2210: Based on the similarity between the information features corresponding to the information to be translated and the benchmark features, determine whether the translation result is the translation result to be checked.

[0280] This baseline feature is used to characterize the commonalities of information features among multiple pieces of information that translate target information into first information. The computer device can determine the probability of an anomaly in the translation method presented in the translation result based on the similarity between information features, thereby determining whether the translation result is a translation result that needs to be checked. If yes, step S2211 is executed, providing the translation result to the checker for verification and correction; if no, step S2212 is executed, completing the translation of the information to be translated.

[0281] S2211: The verification party checks and corrects the translation results to be checked.

[0282] The verification party can be of various types, such as the computer equipment itself, other computer equipment, or a human verification party, etc., and is not limited here.

[0283] S2212: Complete the translation of the information to be translated.

[0284] As can be seen from the above solutions, this application has the following technical effects:

[0285] 1. The model training method described in this application enables the translation model to effectively retain the original information translation capabilities of the initial translation model. On the other hand, it enables the translation model to rationally select and use prompt information and its own translation capabilities for different information, thereby accurately generating multiple different translation results for the same information. While ensuring translation accuracy, it improves the comprehensiveness of the translation model's translation capabilities, enabling the translation model to be applied in a wider range of application scenarios.

[0286] 2. This application innovatively proposes a model architecture for a translation model, which can reduce the impact on the model's translation knowledge while enabling the model to accurately translate the same information by combining multiple translation methods.

[0287] 3. This application can integrate multiple types of information on the input side, such as information domain labels and prompts, to further improve the model's translation performance.

[0288] 4. This application can quickly locate information in the information to be translated that has multiple translation methods and plays a significant role in representing the content of the information to be translated through key information identification. This enables the model to accurately translate this information, improving the efficiency of information translation. Furthermore, this application can tailor the key information identification method to different information domains, making it more suitable for the characteristics of key information in each domain, thus ensuring accurate identification of key information in all domains.

[0289] 5. This application can analyze the accuracy of the translation results through various methods, thereby filtering out translation results that may have abnormalities for the checker to verify, further ensuring the accuracy of information translation and improving the efficiency of checking translation results.

[0290] Based on the information processing method related to model training provided in the above embodiments, this application also provides an information processing apparatus, see below. Figure 24 , Figure 24 This is a structural block diagram of an information processing device provided in an embodiment of this application. The device 2400 includes a first acquisition unit 2401, a second acquisition unit 2402, a first generation unit 2403, and an adjustment unit 2404.

[0291] The first acquisition unit 2401 is used to acquire first sample information and second sample information. Both the first sample information and the second sample information include target information. The target information corresponds to first information in the first sample translation result corresponding to the first sample information and to second information in the second sample translation result corresponding to the second sample information. The first information and the second information are different.

[0292] The second acquisition unit 2402 is used to acquire an initial translation model, which is used to translate the target information into the second information;

[0293] The first generation unit 2403 is configured to translate the first sample information according to the prompt information corresponding to the first sample information and the target information using the initial translation model, to generate a first pending translation result, and to translate the second sample information according to the second sample information and the prompt information, to generate a second pending translation result, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0294] The adjustment unit 2404 is used to adjust the model parameters corresponding to the initial translation model according to the difference between the first sample translation result and the first undetermined translation result, and the difference between the second sample translation result and the second undetermined translation result, to obtain a translation model. The translation model is used to generate a translation result corresponding to the undetermined translation information according to the undetermined translation information including the target information and the prompt information.

[0295] In one possible implementation, the second acquisition unit 2402 is specifically used for:

[0296] Obtain third sample information including the target information, wherein the target information corresponds to the second information in the third translation result of the sample corresponding to the third sample information;

[0297] The third sample information is translated using the initial model to generate a pending third translation result;

[0298] Based on the difference between the third translation result of the sample and the third translation result to be determined, the model parameters corresponding to the initial model are adjusted to obtain the initial translation model.

[0299] In one possible implementation, the second acquisition unit 2402 is specifically used for:

[0300] Based on the difference between the third translation result of the sample and the third translation result to be determined, the model parameters corresponding to the initial model are adjusted to obtain the translation module;

[0301] The initial translation model is constructed based on the translation module and the initial replacement module;

[0302] The first generation unit 2403 is specifically used for:

[0303] The initial replacement module determines whether to replace the target information in the first sample information according to the prompt information based on the first sample information, and outputs first pending information; and determines whether to replace the target information in the second sample information according to the prompt information based on the second sample information, and outputs second pending information.

[0304] The translation module translates the untranslated information in the first pending information to generate the first pending translation result, and translates the untranslated information in the second pending information to generate the second pending translation result.

[0305] The adjustment unit 2404 is specifically used for:

[0306] The model parameters corresponding to the initial replacement module are adjusted based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain a replacement module. The translation module and the replacement module are used to constitute the translation model.

[0307] In one possible implementation, the first sample information and the second sample information each have corresponding information domain tags, the information domain tags being used to identify the information domain to which the corresponding information belongs, and the first generation unit 2403 is specifically used for:

[0308] Using the initial translation model, the first sample information is translated based on the first sample information, the information domain label corresponding to the first sample information, and the prompt information corresponding to the target information to generate a first pending translation result; and the second sample information is translated based on the second sample information, the information domain label corresponding to the second sample information, and the prompt information to generate a second pending translation result.

[0309] In one possible implementation, the first acquisition unit 2401 is specifically used for:

[0310] Multiple sample information is acquired, and each of the multiple sample information has a corresponding sample translation result;

[0311] Key information is identified from the multiple sample information to determine the key information included in each of the multiple sample information. The key information has a greater representational effect on the information content of the sample information than the information in the sample information other than the key information.

[0312] The first sample information is determined as the sample information in which the target information is included in the corresponding key information among the plurality of sample information, and the target information corresponds to the first information in the corresponding sample translation result. The first sample information is determined as the sample information in which the target information is included in the corresponding key information among the plurality of sample information, and the target information corresponds to the second information in the corresponding sample translation result.

[0313] In one possible implementation, the target sample information is any one of the plurality of sample information, the target sample information is composed of a plurality of sub-information, and the target sub-information is any one of the plurality of sub-information; the first acquisition unit 2401 is specifically used for:

[0314] Based on the target sample information, a combination of one or more weights is determined for the target sub-information, including position weight, word frequency weight, similarity weight, and information number weight. The position weight is used to characterize the position of the target sub-information in the target sample information, the word frequency weight is used to characterize the frequency of the target sub-information in the target sample information, the similarity weight is used to characterize the information similarity between the target sub-information and other information in the target sample information, and the information number weight is used to characterize the proportion of sample information including the target sub-information in the multiple sample information.

[0315] Based on a combination of one or more weights, key parameters corresponding to the target sub-information are determined, and the key parameters are used to characterize the criticality of the target sub-information;

[0316] Based on the key parameters corresponding to the multiple sub-information items, the key information included in the target sample information is determined.

[0317] In one possible implementation, the information domain corresponding to the target sample information is the game domain, and the combination of one or more weights includes the similarity weight and the word frequency weight. The first acquisition unit 2401 is specifically used for:

[0318] A first weight adjustment is performed on the similarity weight, and a second weight adjustment is performed on the word frequency weight. The first weight adjustment is used to increase the influence of the similarity weight on the key parameters corresponding to the target sub-information, and the second weight adjustment is used to reduce the influence of the word frequency weight on the key parameters corresponding to the target sub-information.

[0319] Based on the adjusted combination of one or more weights, the key parameters corresponding to the target sub-information are determined.

[0320] Based on the information processing method related to model applications provided in the above embodiments, this application also provides an information processing apparatus, see [link to previous document]. Figure 25 , Figure 25 This application provides a structural block diagram of an information processing device 2500, which includes a third acquisition unit 2501, a first determination unit 2502, and a second generation unit 2503.

[0321] The third acquisition unit 2501 is used to acquire the information to be translated;

[0322] The first determining unit 2502 is used to determine target information with corresponding prompt information in the information to be translated, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0323] The second generation unit 2503 is used to generate a translation result corresponding to the information to be translated based on the information to be translated and the prompt information through a translation model. The translation model is used to translate the information to be translated, and when translating the information to be translated, to determine whether to use the first information as the translation result corresponding to the target information based on the prompt information.

[0324] In one possible implementation, the translation model includes a replacement module and a translation module, and the second generation unit 2503 is specifically used for:

[0325] The replacement module determines whether to replace the target information in the information to be translated according to the prompt information, and outputs pending information.

[0326] The translation module translates the untranslated information in the pending information and generates the translation result corresponding to the information to be translated.

[0327] In one possible implementation, the device further includes a second determining unit:

[0328] The second determining unit is used to determine the information domain label corresponding to the information to be translated, wherein the information domain label is used to identify the information domain corresponding to the information to be translated;

[0329] The second generation unit 2503 is specifically used for:

[0330] The translation model generates a translation result corresponding to the information to be translated based on the information to be translated, the information domain label, and the prompt information. The translation model is used to determine whether to use the first information as the translation result corresponding to the target information based on the prompt information when translating the information to be translated, according to the information to be translated and the information domain label.

[0331] In one possible implementation, the first determining unit 2502 is specifically used for:

[0332] The key information to be translated is identified, and the key information corresponding to the key information is determined. The key information has a greater representational effect on the information content of the information to be translated than the information other than the key information in the information to be translated.

[0333] Obtain an information set, which includes multiple candidate information items, each of which has a corresponding prompt message;

[0334] Among the multiple candidate information, the candidate information that exists in the key information is determined as the target information.

[0335] In one possible implementation, the first determining unit 2502 is specifically used for:

[0336] Determine the target information domain corresponding to the information to be translated;

[0337] Obtain the information set corresponding to the target information domain, and use the prompt information corresponding to the multiple candidate information to identify the translation relationship of the multiple candidate information under the target information domain.

[0338] In one possible implementation, the apparatus further includes a fourth acquisition unit, an extraction unit, a third determination unit, a fourth determination unit, and a fifth determination unit:

[0339] The fourth acquisition unit is used to acquire multiple pieces of information to be analyzed, all of which include the target information. The target information corresponds to the first information in the translation results that are respectively corresponding to the multiple pieces of information to be analyzed.

[0340] The extraction unit is used to extract benchmark features that characterize the common information features of the multiple pieces of information to be analyzed.

[0341] The third determining unit is used to determine the similarity between the information features corresponding to the information to be translated and the benchmark features;

[0342] The fourth determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked based on the fact that the translation result corresponding to the information to be translated includes the first information and the similarity is less than the similarity threshold.

[0343] The fifth determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked based on the fact that the translation result corresponding to the information to be translated does not include the first information and the similarity is not less than the similarity threshold. The translation result to be checked is used to provide the checking party for verification.

[0344] In one possible implementation, the translation relationship between the target information and the first information is a translation relationship existing in the target information domain, and the device further includes a sixth determining unit, a seventh determining unit, and an eighth determining unit:

[0345] The sixth determining unit is used to determine the information domain corresponding to the information to be translated;

[0346] The seventh determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked, based on the information domain corresponding to the information to be translated being the target information domain and the translation result corresponding to the information to be translated not including the first information.

[0347] The eighth determining unit is used to determine the translation result corresponding to the information to be translated as the translation result to be checked based on the fact that the information domain corresponding to the information to be translated is not the target information domain and the translation result corresponding to the information to be translated includes the first information. The translation result to be checked is used to provide the verification party for verification.

[0348] This application also provides a computer device; please refer to [link to relevant documentation]. Figure 26 As shown, the computer device can be a terminal device; for example, a mobile phone can be used as a terminal device.

[0349] Figure 26 This diagram illustrates a partial structural representation of a mobile phone related to the terminal device provided in this embodiment. (Reference) Figure 26 The mobile phone includes components such as a radio frequency (RF) circuit 710, a memory 720, an input unit 730, a display unit 740, a sensor 750, an audio circuit 760, a wireless Fidelity (WiFi) module 770, a processor 780, and a power supply 790. Those skilled in the art will understand that... Figure 26 The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0350] The following is combined with Figure 26 A detailed introduction to each component of a mobile phone:

[0351] RF circuit 710 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with processor 780; additionally, it transmits uplink data to the base station. Typically, RF circuit 710 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), and a duplexer. Furthermore, RF circuit 710 can also communicate wirelessly with networks and other devices. The aforementioned wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, and Short Messaging Service (SMS).

[0352] The memory 720 can be used to store software programs and modules. The processor 780 executes various mobile phone functions and data processing by running the software programs and modules stored in the memory 720. The memory 720 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 720 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0353] The input unit 730 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the mobile phone. Specifically, the input unit 730 may include a touch panel 731 and other input devices 732. The touch panel 731, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 731), and drive the corresponding connected devices according to a pre-set program. Optionally, the touch panel 731 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 780, and can also receive and execute commands sent by the processor 780. In addition, the touch panel 731 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 731, the input unit 730 may also include other input devices 732. Specifically, other input devices 732 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0354] The display unit 740 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 740 may include a display panel 741, which may optionally be configured as a Liquid Crystal Display (LCD), Organic Light-Emitting Diode (OLED), or similar display panel. Further, a touch panel 731 may cover the display panel 741. When the touch panel 731 detects a touch operation on or near it, it transmits the information to the processor 780 to determine the type of touch event. Subsequently, the processor 780 provides corresponding visual output on the display panel 741 based on the type of touch event. Although in Figure 26 In this embodiment, the touch panel 731 and the display panel 741 are two separate components to realize the input and output functions of the mobile phone. However, in some embodiments, the touch panel 731 and the display panel 741 can be integrated to realize the input and output functions of the mobile phone.

[0355] The mobile phone may also include at least one sensor 750, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 741 according to the ambient light level, and the proximity sensor can turn off the display panel 741 and / or backlight when the phone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, taps), etc. Other sensors that may be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0356] Audio circuit 760, speaker 761, and microphone 762 provide an audio interface between the user and the mobile phone. Audio circuit 760 converts received audio data into electrical signals and transmits them to speaker 761, where speaker 761 converts them into sound signals for output. On the other hand, microphone 762 converts collected sound signals into electrical signals, which are received by audio circuit 760, converted into audio data, and then processed by processor 780 before being transmitted via RF circuit 710 to, for example, another mobile phone, or the audio data can be output to memory 720 for further processing.

[0357] WiFi is a short-range wireless transmission technology. Through the WiFi module 770, mobile phones can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access. Although Figure 26 The WiFi module 770 is shown, but it is understood that it is not an essential component of a mobile phone and can be omitted as needed without changing the essence of the invention.

[0358] The processor 780 is the control center of the mobile phone, connecting various parts of the phone through various interfaces and lines. It executes software programs and / or modules stored in the memory 720, and calls data stored in the memory 720 to perform various functions and process data, thereby performing overall detection of the phone. Optionally, the processor 780 may include one or more processing units; preferably, the processor 780 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 780.

[0359] The mobile phone also includes a power supply 790 (such as a battery) that supplies power to various components. Preferably, the power supply can be logically connected to the processor 780 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0360] Although not shown, mobile phones may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0361] In this embodiment, the processor 780 included in the terminal device also has the following functions:

[0362] Obtain first sample information and second sample information. Both the first sample information and the second sample information include target information. The target information corresponds to first information in the first sample translation result corresponding to the first sample information and to second information in the second sample translation result corresponding to the second sample information. The first information and the second information are different.

[0363] An initial translation model is obtained, which is used to translate the target information into the second information;

[0364] Using the initial translation model, the first sample information is translated according to the prompt information corresponding to the first sample information and the target information to generate a first pending translation result, and the second sample information is translated according to the second sample information and the prompt information to generate a second pending translation result, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0365] The model parameters corresponding to the initial translation model are adjusted based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain a translation model. The translation model is used to generate a translation result corresponding to the undetermined translation information based on the undetermined translation information including the target information and the prompt information.

[0366] In this embodiment, the processor 780 included in the terminal device also has the following functions:

[0367] Obtain the information to be translated;

[0368] Identify target information with corresponding prompt information in the information to be translated, wherein the prompt information is used to identify the translation relationship between the target information and the first information;

[0369] The translation model generates a translation result corresponding to the information to be translated based on the information to be translated and the prompt information. The translation model is used to translate the information to be translated, and when translating the information to be translated, it determines whether to use the first information as the translation result corresponding to the target information based on the prompt information.

[0370] This application also provides a server; please refer to [link / reference]. Figure 27 As shown, Figure 27 This is a structural diagram of a server 800 provided in an embodiment of this application. The server 800 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 822 (e.g., one or more processors) and a memory 832, and one or more storage media 830 (e.g., one or more mass storage devices) for storing application programs 842 or data 844. The memory 832 and storage media 830 can be temporary or persistent storage. The program stored in the storage media 830 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 822 may be configured to communicate with the storage media 830 and execute the series of instruction operations in the storage media 830 on the server 800.

[0371] Server 800 may also include one or more power supplies 826, one or more wired or wireless network interfaces 850, one or more input / output interfaces 858, and / or one or more operating systems 841, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0372] The steps performed by the server in the above embodiments can be based on Figure 27 The server structure shown.

[0373] This application also provides a computer-readable storage medium for storing a computer program that executes any one of the information processing methods described in the foregoing embodiments.

[0374] This application also provides a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform any of the information processing methods described in the above embodiments.

[0375] It is understood that in the specific implementation of this application, data related to user information (such as information to be translated) is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0376] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium can be at least one of the following media: read-only memory (ROM), RAM, magnetic disk, or optical disk, etc., and other media capable of storing program code.

[0377] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0378] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An information processing method, characterized in that, The method includes: Obtain first sample information and second sample information. Both the first sample information and the second sample information include target information. The target information corresponds to first information in the first sample translation result corresponding to the first sample information and to second information in the second sample translation result corresponding to the second sample information. The first information and the second information are different. Obtain an initial translation model, which is used to translate the target information into the second information; Using the initial translation model, the first sample information is translated according to the prompt information corresponding to the first sample information and the target information to generate a first pending translation result, and the second sample information is translated according to the second sample information and the prompt information to generate a second pending translation result, wherein the prompt information is used to identify the translation relationship between the target information and the first information; The model parameters corresponding to the initial translation model are adjusted based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain a translation model. The translation model is used to generate a translation result corresponding to the undetermined translation information based on the undetermined translation information including the target information and the prompt information.

2. The method according to claim 1, characterized in that, The process of obtaining the initial translation model includes: Obtain third sample information including the target information, wherein the target information corresponds to the second information in the third translation result of the sample corresponding to the third sample information; The third sample information is translated using the initial model to generate a pending third translation result; Based on the difference between the third translation result of the sample and the third translation result to be determined, the model parameters corresponding to the initial model are adjusted to obtain the initial translation model.

3. The method according to claim 2, characterized in that, The step of adjusting the model parameters corresponding to the initial model based on the difference between the third translation result of the sample and the third translation result to be determined, to obtain the initial translation model, includes: Based on the difference between the third translation result of the sample and the third translation result to be determined, the model parameters corresponding to the initial model are adjusted to obtain the translation module; The initial translation model is constructed based on the translation module and the initial replacement module; The step of translating the first sample information based on the prompt information corresponding to the first sample information and the target information using the initial translation model to generate a first pending translation result, and translating the second sample information based on the second sample information and the prompt information to generate a second pending translation result, includes: The initial replacement module determines whether to replace the target information in the first sample information according to the prompt information based on the first sample information, and outputs first pending information; and determines whether to replace the target information in the second sample information according to the prompt information based on the second sample information, and outputs second pending information. The translation module translates the untranslated information in the first pending information to generate the first pending translation result, and translates the untranslated information in the second pending information to generate the second pending translation result. The step of adjusting the model parameters corresponding to the initial translation model based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain the translation model, includes: The model parameters corresponding to the initial replacement module are adjusted based on the differences between the first sample translation result and the first undetermined translation result, and the differences between the second sample translation result and the second undetermined translation result, to obtain a replacement module. The translation module and the replacement module are used to constitute the translation model.

4. The method according to claim 1, characterized in that, The first sample information and the second sample information each have corresponding information domain labels, which are used to identify the information domain to which the corresponding information belongs. The step of translating the first sample information according to the prompt information corresponding to the first sample information and the target information using the initial translation model to generate a first pending translation result, and translating the second sample information according to the second sample information and the prompt information to generate a second pending translation result, includes: Using the initial translation model, the first sample information is translated based on the first sample information, the information domain label corresponding to the first sample information, and the prompt information corresponding to the target information to generate a first pending translation result; and the second sample information is translated based on the second sample information, the information domain label corresponding to the second sample information, and the prompt information to generate a second pending translation result.

5. The method according to claim 1, characterized in that, The acquisition of the first sample information and the second sample information includes: Multiple sample information is acquired, and each of the multiple sample information has a corresponding sample translation result; Key information is identified from the multiple sample information to determine the key information included in each of the multiple sample information. The key information has a greater representational effect on the information content of the sample information than the information in the sample information other than the key information. The first sample information is determined as the sample information in which the target information is included in the corresponding key information among the plurality of sample information, and the target information corresponds to the first information in the corresponding sample translation result. The first sample information is determined as the sample information in which the target information is included in the corresponding key information among the plurality of sample information, and the target information corresponds to the second information in the corresponding sample translation result.

6. The method according to claim 5, characterized in that, The target sample information is any one of the plurality of sample information, which is composed of multiple sub-information pieces, and the target sub-information is any one of the plurality of sub-information pieces. The step of identifying key information from the plurality of sample information and determining the key information included in each of the plurality of sample information includes: Based on the target sample information, a combination of one or more weights is determined for the target sub-information, including position weight, word frequency weight, similarity weight, and information number weight. The position weight is used to characterize the position of the target sub-information in the target sample information, the word frequency weight is used to characterize the frequency of the target sub-information in the target sample information, the similarity weight is used to characterize the information similarity between the target sub-information and other information in the target sample information, and the information number weight is used to characterize the proportion of sample information including the target sub-information in the multiple sample information. Based on a combination of one or more weights, key parameters corresponding to the target sub-information are determined, and the key parameters are used to characterize the criticality of the target sub-information; Based on the key parameters corresponding to the multiple sub-information items, the key information included in the target sample information is determined.

7. The method according to claim 6, characterized in that, The target sample information corresponds to the game domain, and the combination of one or more weights includes the similarity weight and the word frequency weight. Determining the key parameters corresponding to the target sub-information based on the combination of the one or more weights includes: A first weight adjustment is performed on the similarity weight, and a second weight adjustment is performed on the word frequency weight. The first weight adjustment is used to increase the influence of the similarity weight on the key parameters corresponding to the target sub-information, and the second weight adjustment is used to reduce the influence of the word frequency weight on the key parameters corresponding to the target sub-information. Based on the adjusted combination of one or more weights, the key parameters corresponding to the target sub-information are determined.

8. An information processing method, characterized in that, The method includes: Obtain the information to be translated; Identify target information with corresponding prompt information in the information to be translated, wherein the prompt information is used to identify the translation relationship between the target information and the first information; The translation model generates a translation result corresponding to the information to be translated based on the information to be translated and the prompt information. The translation model is used to translate the information to be translated, and when translating the information to be translated, it determines whether to use the first information as the translation result corresponding to the target information based on the prompt information.

9. The method according to claim 8, characterized in that, The translation model includes a replacement module and a translation module. The step of generating a translation result corresponding to the information to be translated based on the information to be translated and the prompt information using the translation model includes: The replacement module determines whether to replace the target information in the information to be translated according to the prompt information, and outputs pending information. The translation module translates the untranslated information in the pending information and generates the translation result corresponding to the information to be translated.

10. The method according to claim 8, characterized in that, The method further includes: Determine the information domain label corresponding to the information to be translated, wherein the information domain label is used to identify the information domain corresponding to the information to be translated; The step of generating a translation result corresponding to the information to be translated based on the information to be translated and the prompt information using a translation model includes: The translation model generates a translation result corresponding to the information to be translated based on the information to be translated, the information domain label, and the prompt information. The translation model is used to determine whether to use the first information as the translation result corresponding to the target information based on the prompt information when translating the information to be translated, according to the information to be translated and the information domain label.

11. The method according to claim 8, characterized in that, The step of determining the target information with corresponding prompt information in the information to be translated includes: The key information to be translated is identified, and the key information corresponding to the key information is determined. The key information has a greater representational effect on the information content of the information to be translated than the information other than the key information in the information to be translated. Obtain an information set, which includes multiple candidate information items, each of which has a corresponding prompt message; Among the multiple candidate information, the candidate information that exists in the key information is determined as the target information.

12. The method according to claim 11, characterized in that, The acquired information set includes: Determine the target information domain corresponding to the information to be translated; Obtain the information set corresponding to the target information domain, and use the prompt information corresponding to the multiple candidate information to identify the translation relationship of the multiple candidate information under the target information domain.

13. The method according to claim 8, characterized in that, The method further includes: Multiple pieces of information to be analyzed are obtained, each of which includes the target information. The target information corresponds to the first information in the translation results that are respectively corresponding to the multiple pieces of information to be analyzed. Extract benchmark features that characterize the common information features of the multiple pieces of information to be analyzed; Determine the similarity between the information features corresponding to the information to be translated and the benchmark features; Based on the fact that the translation result corresponding to the information to be translated includes the first information, and the similarity is less than the similarity threshold, the translation result corresponding to the information to be translated is determined as the translation result to be checked; Based on the fact that the translation result corresponding to the information to be translated does not include the first information, and the similarity is not less than the similarity threshold, the translation result corresponding to the information to be translated is determined as the translation result to be checked, and the translation result to be checked is provided to the checker for verification.

14. The method according to claim 8, characterized in that, The translation relationship between the target information and the first information is a translation relationship existing in the domain of the target information, and the method further includes: Determine the information domain corresponding to the information to be translated; Based on the fact that the information domain corresponding to the information to be translated is the target information domain, and the translation result corresponding to the information to be translated does not include the first information, the translation result corresponding to the information to be translated is determined as the translation result to be checked; Based on the fact that the information domain corresponding to the information to be translated is not the target information domain, and the translation result corresponding to the information to be translated includes the first information, the translation result corresponding to the information to be translated is determined as the translation result to be checked, and the translation result to be checked is provided to the checker for verification.

15. An information processing device, characterized in that, The device includes a first acquisition unit, a second acquisition unit, a first generation unit, and an adjustment unit: The first acquisition unit is used to acquire first sample information and second sample information. Both the first sample information and the second sample information include target information. The target information corresponds to first information in the first sample translation result corresponding to the first sample information and to second information in the second sample translation result corresponding to the second sample information. The first information and the second information are different. The second acquisition unit is used to acquire an initial translation model, which is used to translate the target information into the second information; The first generation unit is configured to translate the first sample information according to the prompt information corresponding to the first sample information and the target information using the initial translation model, to generate a first pending translation result, and to translate the second sample information according to the second sample information and the prompt information, to generate a second pending translation result, wherein the prompt information is used to identify the translation relationship between the target information and the first information; The adjustment unit is used to adjust the model parameters corresponding to the initial translation model according to the difference between the first sample translation result and the first undetermined translation result, and the difference between the second sample translation result and the second undetermined translation result, to obtain a translation model. The translation model is used to generate a translation result corresponding to the undetermined translation information according to the undetermined translation information including the target information and the prompt information.

16. An information processing device, characterized in that, The device includes a third acquisition unit, a first determination unit, and a second generation unit: The third acquisition unit is used to acquire the information to be translated; The first determining unit is used to determine target information with corresponding prompt information in the information to be translated, wherein the prompt information is used to identify the translation relationship between the target information and the first information; The second generation unit is configured to generate a translation result corresponding to the information to be translated based on the information to be translated and the prompt information using a translation model. The translation model is configured to translate the information to be translated and, when translating the information to be translated, determine whether to use the first information as the translation result corresponding to the target information based on the prompt information.

17. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to execute the information processing method according to any one of claims 1-7, or the information processing method according to any one of claims 8-14, according to instructions in the computer program.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the information processing method according to any one of claims 1-7, or for performing the information processing method according to any one of claims 8-14.

19. A computer program product comprising a computer program, which, when run on a computer device, causes the computer device to perform the information processing method according to any one of claims 1-7, or to perform the information processing method according to any one of claims 8-14.

Images