Chapter error correction method, device, electronic device and storage medium

Abstract

The present disclosure provides a method, device, electronic device and storage medium for correcting passages. The method includes: being able to segment passage data to be processed and arrange them in passage order to obtain a plurality of sentences to be processed, so that the passage correction model can be used to correct the grammatical relationship between the sentences to be processed in the passage order, and the passage correction model can determine the correlation between the upper and lower sentences according to the order of the upper and lower sentences, so that the corrected sentences are more consistent with the correlation between the upper and lower sentences, and these corrected sentences are integrated to obtain the corrected passage, which can ensure that the corrected passage is more accurate.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a chapter error correction method, device, electronic device and storage medium. Background Art

[0002] In language applications, in order to ensure the grammatical correctness of the text, it is very important to correct the grammatical errors of the text. How to correct the grammatical errors of the text to make the language expression more appropriate, generally using a correction model that can correct sentences for grammatical correction processing.

[0003] The error correction model generally divides the entire chapter into several sentences, and then inputs them one by one into the error correction model for modification and correction. This fails to grasp the correlation between sentences in the entire chapter, and will cause contradictions in the modifications between sentences. Summary of the invention

[0004] In view of this, the purpose of the present disclosure is to propose a chapter error correction method, device, electronic device and storage medium to solve or partially solve the above-mentioned technical problems.

[0005] Based on the above purpose, the first aspect of the present disclosure provides a method for correcting errors in a passage, comprising:

[0006] Receiving the chapter data to be processed, performing sentence segmentation processing on the chapter data to be processed, and obtaining a plurality of sentences to be processed arranged in the order of the chapters;

[0007] Determine a plurality of input quantities corresponding to the plurality of sentences to be processed according to the sequence of the text, input the plurality of input quantities into a text error correction model, perform error correction processing according to the sequence of the text, and obtain a plurality of corrected sentences;

[0008] The multiple corrected sentences are integrated to obtain a corrected chapter.

[0009] Based on the same technical concept, the second aspect of the present disclosure proposes a paragraph error correction device, comprising:

[0010] A receiving module is configured to receive the chapter data to be processed, perform sentence segmentation processing on the chapter data to be processed, and obtain a plurality of sentences to be processed arranged in the order of the chapters;

[0011] An error correction processing module is configured to determine a plurality of corresponding input quantities for the plurality of sentences to be processed according to the chapter order, input the plurality of input quantities into a chapter error correction model, perform error correction processing according to the chapter order, and obtain a plurality of corrected sentences;

[0012] The integration module is configured to integrate the multiple corrected sentences to obtain a corrected chapter.

[0013] Based on the same technical concept, the third aspect of the present disclosure proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0014] Based on the same technical concept, the fourth aspect of the present disclosure proposes a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the method described in the first aspect.

[0015] From the above, it can be seen that the passage correction method, device, electronic device and storage medium provided by the present invention can segment the passage data to be processed and arrange them in the order of the passage to obtain multiple sentences to be processed. In this way, the passage correction model can be used to perform whole-passage correction on the grammatical relationship between each sentence to be processed according to the order of the passage. The passage correction model can determine the correlation between the upper and lower sentences according to the order of the upper and lower sentences, so that the corrected sentences are more in line with the correlation between the upper and lower sentences. After integrating these corrected sentences, the corrected passage can be obtained, which can ensure that the corrected passage is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A schematic diagram of an application scenario of an embodiment of the present disclosure;

[0018] Figure 2 A flowchart of a method for correcting errors in a chapter according to an embodiment of the present disclosure;

[0019] Figure 3 A flowchart of a process of obtaining a paragraph error correction model according to an embodiment of the present disclosure;

[0020] Figure 4 It is a structural block diagram of a text error correction device according to an embodiment of the present disclosure;

[0021] Figure 5 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0023] The principles and spirit of the present disclosure will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present disclosure, and are not intended to limit the scope of the present disclosure in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0024] It should be understood herein that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction rather than having any limiting meaning.

[0025] Based on the description of the above background technology, the related technology also has the following situations:

[0026] Generally speaking, sentence-based error correction models use sentences as processing units. Sentence-based error correction models include: generative error correction models and discriminative error correction models. In generative error correction models, the modification results of the sentence can be directly generated; in discriminative error correction models, the corresponding modifications of each word in the sentence can be predicted.

[0027] Chapter-level grammar correction is particularly common, such as essays and emails. Therefore, improving the performance of chapter-level grammar correction is of great value.

[0028] However, the sentence length that the sentence-based error correction model accepts is limited, and the large number of words in a passage (such as an article / email) can easily exceed the length limit of the model, which makes it difficult to directly generalize the sentence-based error correction model to the passage. Therefore, when processing passages, the passages have to be split into several sentences and input into the model one by one for modification. However, such a model cannot grasp the relationship between sentences, which can easily cause contradictions in the modification between sentences, especially in the context of language phenomena such as tense and reference. For example, the first sentence of the article specifies the tense background, and the second sentence of the article uses a different tense. Even if the second sentence itself is grammatically correct, it contradicts the first sentence; however, the sentence-based error correction model has no way to ensure the uniformity of modifications between different sentences.

[0029] Based on the above description, the principle and spirit of the present disclosure are explained in detail below with reference to several representative embodiments of the present disclosure.

[0030] The present disclosure provides a passage error correction method, device, electronic device and storage medium, which can segment passage data to be processed and arrange them in the order of the passage to obtain multiple sentences to be processed. In this way, the passage error correction model can be used to perform whole-passage error correction on the grammatical relationship between each sentence to be processed according to the order of the passage. The passage error correction model can determine the correlation between the upper and lower sentences according to the order of the upper and lower sentences, so that the corrected sentences are more consistent with the correlation between the upper and lower sentences. After integrating these corrected sentences to obtain a corrected passage, it can be ensured that the corrected passage is more accurate.

[0031] refer to Figure 1 , which is a schematic diagram of an application scenario of the chapter error correction method provided by an embodiment of the present disclosure. The application scenario includes a terminal device 101, a server 102, and a data storage system 103. Among them, the terminal device 101, the server 102 and the data storage system 103 can be connected through a wired or wireless communication network. The terminal device 101 includes but is not limited to a desktop computer, a mobile phone, a mobile computer, a tablet computer, a media player, a smart wearable device, a personal digital assistant (PDA) or other electronic devices capable of realizing the above functions. The server 102 and the data storage system 103 can both be independent physical servers, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0032] The server 102 is used to adjust the parameters of the obtained initial error correction model so that the adjusted initial paragraph error correction model can perform error correction processing based on the context of the sentence in the paragraph, and then train the initial error correction model using the obtained paragraph error correction sample set to obtain the paragraph error correction model. The data storage system 103 provides data storage support for the operation of the server 102, for example, for storing the paragraph error correction sample set required for training.

[0033] After obtaining the chapter error correction model, it can be divided into two situations for subsequent chapter error correction processing.

[0034] The first method: the server 102 sends the obtained chapter error correction model to the data storage system 103 for storage. In this way, after the terminal device 101 receives the chapter data to be processed, it can send the chapter data to be processed to the server 102. The server 102 performs sentence segmentation processing on the chapter data to be processed and arranges them in the order of the chapter to obtain multiple sentences to be processed. In this way, the chapter error correction model can be used to perform whole-chapter error correction processing on the grammatical relationship between each sentence to be processed in the order of the chapter. The chapter error correction model can determine the correlation between the upper and lower sentences according to the order of the upper and lower sentences, thereby making the corrected sentence more consistent with the correlation between the upper and lower sentences. These corrected sentences are integrated to obtain the corrected chapter, and then the corrected chapter is sent to the terminal device 101.

[0035] The second method: the server 102 sends the obtained chapter error correction model to the terminal device 101. After receiving the chapter data to be processed, the terminal device 101 can segment the chapter data to be processed and arrange them in the order of the chapter to obtain multiple sentences to be processed. In this way, the chapter error correction model can be used to perform whole-chapter error correction on the grammatical relationship between the sentences to be processed in the order of the chapter. The chapter error correction model can determine the correlation between the upper and lower sentences according to the order of the upper and lower sentences, so that the corrected sentences are more consistent with the correlation between the upper and lower sentences, and these corrected sentences are integrated to obtain the corrected chapter.

[0036] Combine the following Figure 1 The article error correction method according to the exemplary embodiment of the present disclosure is described by using the application scenario. It should be noted that the above application scenario is only shown to facilitate understanding of the spirit and principle of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0037] The present disclosure provides a method for correcting errors in a passage. Figure 2 As shown, the execution process of this method includes:

[0038] Step 201, receiving the chapter data to be processed, segmenting the chapter data to be processed, and obtaining a plurality of sentences to be processed arranged in chapter order.

[0039] In specific implementation, the chapter data to be processed can be an article, a student composition or an email, etc. Before error correction, the chapter data to be processed needs to be segmented (for example, each "." is segmented into a sentence). In this way, the chapter data to be processed will be segmented into multiple sentences to be processed.

[0040] Step 202, determining a plurality of corresponding input quantities for the plurality of sentences to be processed according to the chapter order, inputting the plurality of input quantities into a chapter error correction model, performing error correction processing according to the chapter order, and obtaining a plurality of corrected sentences.

[0041] In some embodiments, step 202 includes:

[0042] Step 2021, determining the input amount corresponding to each sentence to be processed according to the preset error correction direction includes: [previous sentence content, next sentence content, sentence content to be processed].

[0043] Step 2022: Utilize the paragraph error correction model to perform error correction processing on the multiple input amounts according to the rolling modification strategy to obtain multiple corrected sentences.

[0044] In specific implementation, if the error correction direction is from the previous sentence to the next sentence, the input includes: [the previous sentence content after error correction, the unprocessed next sentence content, and the content of the sentence to be processed], and error correction processing is performed starting from the beginning sentence of the chapter data to be processed according to the rolling modification strategy.

[0045] If the error correction direction is from the next sentence to the previous sentence, the input includes: [unprocessed previous sentence content, corrected next sentence content, and sentence content to be processed], and error correction processing is performed starting from the last sentence of the chapter data to be processed according to the rolling modification strategy.

[0046] If the error correction direction is to determine the sentence content according to the first predetermined probability (pa) and the second predetermined probability (pb), the probability of using the corrected previous sentence content for the previous sentence content in the input is pa, and the probability of using the unprocessed previous sentence content is 1-pa; the probability of using the corrected next sentence content for the next sentence content in the input is pb, and the probability of using the unprocessed next sentence content is 1-pb. The error correction process can be started from the beginning sentence of the chapter data to be processed, or from the end sentence.

[0047] Then, if there are H sentences in the chapter data to be processed, H rounds of rolling correction are required, and the input amount used in each round of correction is determined according to the above method. In this way, after H rounds of correction, H corrected sentences are obtained.

[0048] Step 203, integrating the plurality of corrected sentences to obtain a corrected chapter.

[0049] In specific implementation, the passage is corrected according to the above process, and each sentence of the passage can be corrected according to the grammar and contextual relationship of the previous and next sentences, and the results of the correction of each sentence are integrated to obtain the final corrected passage.

[0050] Among them, during the process of correcting the passage, after the above step 201, the numbers of each sentence to be processed will be placed in the database. After the correction of a sentence to be processed is completed, the number of the sentence to be processed will be deleted from the database until there is no sentence number in the database, which proves that the correction of all passages of the passage to be processed is completed.

[0051] For example, if the error correction direction is determined to be from the previous sentence to the next sentence, the corresponding input parameters are: the length of the previous sentence is 1, the length of the next sentence is 1, and the neighboring sentence overlap parameter of the current sentence is 1. Assume that the chapter has 5 sentences The subscript i indicates the sentence position, which is {1, 2, 3, 4, 5}; the superscript j indicates the result of the round of correction, which is {0, 1, 2, 3, ...}, where 0 indicates the original sentence. The five rounds of correction process corresponding to this chapter, as well as the corresponding context and model input and output are shown in Table 1:

[0052] Table 1

[0053]

[0054] In some embodiments, after the passage error correction is completed, it may be necessary to correct the words in the sentence. In this way, the above step 203 specifically includes:

[0055] Step 2031, in response to determining that the error correction processing of the plurality of sentences to be processed is all completed, converting the paragraph error correction model into an initial error correction model.

[0056] Step 2032: Use the initial error correction model to perform grammatical correction on the multiple corrected sentences respectively to obtain multiple word-corrected sentences.

[0057] Step 2033, integrating the multiple sentences after word error correction to obtain a corrected chapter.

[0058] In specific implementation, when there is no sentence number in the above database, the paragraph error correction model is converted into an initial error correction model. If the paragraph error correction model is obtained by adjusting the generative initial error correction model, the paragraph error correction model is adjusted back to the input of the generative initial error correction model. If the paragraph error correction model is obtained by adjusting the discriminative initial error correction model, the input and output of the paragraph error correction model are adjusted back to the input and output of the discriminative initial error correction model.

[0059] Then, the converted initial error correction model is used to perform grammatical correction on each of the corrected sentences obtained above, and the sentences after each word correction are integrated according to the order of the chapter to obtain the final corrected chapter.

[0060] Through the above scheme, while performing chapter correction on the chapter data to be processed, grammatical correction can also be performed on multiple corrected sentences after the chapter correction, which can further improve the accuracy of error correction.

[0061] In some embodiments, the process of obtaining the above-mentioned paragraph error correction model is as follows: Figure 3 As shown, including:

[0062] Step 301, adjusting the parameters of the obtained initial error correction model to obtain an initial paragraph error correction model, wherein the initial paragraph error correction model can perform error correction processing based on the context of a sentence in the paragraph.

[0063] In specific implementation, the parameters of the ready-made initial error correction model are adjusted to reduce the training time of the initial chapter error correction model and improve the training efficiency. The initial chapter error correction model can be used to input the sentences of the chapter and the upper and lower sentences of the sentence, and the sentence can be corrected based on the grammatical relationship between the upper and lower sentences and the sentence. However, the error correction accuracy of the newly adjusted initial chapter error correction model is not high enough, and it is necessary to improve the error correction accuracy through training.

[0064] In some embodiments, step 301 includes:

[0065] Step 3011, adjusting the input amount and / or output amount of the obtained initial error correction model to obtain a contextual error correction model.

[0066] In the specific implementation, the initial error correction model includes: a generative initial error correction model and a discriminative initial error correction model. The generative initial error correction model can predict the modification type corresponding to each word in the sentence, and then obtain the corresponding error correction processing result from the modification type. The contextual error correction model is recorded as the mapping relationship from the input sentence to the output sentence as model: X→Y, where X is the input quantity and Y is the output quantity.

[0067] In some embodiments, the initial error correction model is a generative initial error correction model, and step 3011 includes:

[0068] Step 3011a, in response to determining that the obtained initial error correction model is a generative initial error correction model, wherein an encoder is provided at the input side of the generative initial error correction model. After the encoder can directly encode the input amount, the error correction process is started, the corresponding error correction process result can be directly generated, and the error correction process result is decoded by the decoder to form an output amount for output.

[0069] In a specific implementation, an encoder is provided on the input side of the generative initial error correction model, and a decoder is provided on the output side.

[0070] Step 3011b, modify the input quantity of the encoder in the generative initial error correction model, and obtain the modified input quantity including: [previous sentence content, next sentence content, current sentence content], and obtain the contextual error correction model.

[0071] In the specific implementation, only the sentence X and its modified result Y in the generative initial error correction model are used. The generative initial error correction model is supplemented on the input side, and the corresponding upper and lower bureaus are added to the input of the encoder, becoming [Prev, Next, X], while the output of the model remains unchanged, where Prev refers to the content of the previous sentence, Next refers to the content of the next sentence, and X is the current sentence content. In addition, in order to ensure the performance of the generative initial error correction model, the input side of the generative initial error correction model can distinguish Prev, Next and X, and a corresponding distinguishing mark is added to each input.

[0072] The distinguishing mark includes at least one of the following: a different vector mark, a different color mark, and a different symbol mark.

[0073] Through the above scheme, for the generative initial error correction model, it is only necessary to adjust the input amount on the encoder side accordingly, and the output amount on the decoder side remains unchanged. In this way, the adjusted data is relatively small and the operation is simple and fast.

[0074] In some embodiments, the initial error correction model is a discriminant initial error correction model, and step 3011 includes:

[0075] Step 3011A, in response to determining that the obtained initial error correction model is a discriminant initial error correction model.

[0076] In the specific implementation, the corresponding error correction processing result can be directly generated. In the discriminant initial error correction model, no encoder and decoder are set, and the modification type corresponding to each word in the sentence can be predicted, and then the corresponding error correction processing result is obtained from the modification type.

[0077] Step 3011B, modify the input quantity of the discriminant initial error correction model, and obtain the modified input quantity including: [previous sentence content, next sentence content, current sentence content].

[0078] The corresponding input quantity can also be expressed as: [Prev, Next, X].

[0079] Step 3011C, connects the output layer corresponding to the current sentence content in the discriminant initial error correction model to the output side of the discriminant initial error correction model, so that the output layer corresponding to the current sentence content outputs the output result as an output quantity from the output side to obtain a contextual error correction model.

[0080] Through the above scheme, since there is no encoder and decoder in the discriminant initial error correction model, it is necessary to modify the input amount on the input side and adjust the output amount on the output side. It is only necessary to extract the result of the output layer corresponding to the current sentence content X as the output amount output. In this way, the context error correction model can be obtained based on the discriminant initial error correction model.

[0081] In some embodiments, step 3011 may further include:

[0082] Step 30111, pre-set various parameters of the input of the initial error correction model, including: the length of the previous sentence content, the length of the next sentence content and the adjacent sentence overlap parameter of the current sentence content, that is, the parameters corresponding to Prev, Next and X.

[0083] Step 30112, adjust the previous sentence content in the input according to the length of the previous sentence content, adjust the next sentence content in the input according to the length of the next sentence content, and adjust the current sentence content in the input according to the adjacent sentence overlap parameter of the current sentence content to obtain a contextual error correction model.

[0084] In specific implementation, the content of the previous sentence, the content of the next sentence, and the content of the current sentence can be adjusted according to various parameters of the input amount. Among them, the length of the content of the previous sentence and the length of the content of the next sentence can be the number of sentences; the neighboring sentence overlap parameter of the current sentence content is the number of sentences overlapping between X in the current number and X in the next input amount.

[0085] Through the above scheme, various parameters of the input quantity can be set according to actual needs, and then the various input quantities can be adjusted accordingly according to these parameters, thus meeting the diverse needs of users and improving user experience.

[0086] Step 3012, adjusting the error correction direction of the context error correction model to obtain the initial chapter error correction model.

[0087] Among them, the error correction direction includes: from the upper sentence to the lower sentence, that is, according to the error correction direction from left to right of the article, or from the lower sentence to the upper sentence, that is, according to the error correction direction from right to left of the article, or regardless of the direction of the upper and lower sentences, the error correction direction is controlled by probability.

[0088] In some embodiments, step 3012 includes:

[0089] Step 3012A, the correction direction of the contextual correction model is adjusted from the previous sentence to the next sentence to obtain the initial chapter correction model, and the specific adjustment includes: the previous sentence content in the input is the previous sentence content after correction, and the next sentence content is the unprocessed next sentence content.

[0090] In specific implementation, the error correction direction of the above steps is performed in accordance with the normal reading order (ie, from left to right), so that the error correction and modification can be continuously scrolled backward from the beginning of the chapter in accordance with the above error correction direction.

[0091] Alternatively, step 3012B, the correction direction of the contextual correction model is adjusted from the next sentence to the previous sentence to obtain the initial chapter correction model, and the specific adjustment includes: the previous sentence content in the input is the unprocessed previous sentence content, and the next sentence content is the next sentence content after correction.

[0092] In specific implementation, the error correction direction of the above steps is opposite to the normal reading order (ie, from right to left), so that the error correction and modification can be continuously scrolled forward from the end of the chapter.

[0093] Alternatively, in step 3012C, the previous sentence content in the input of the contextual error correction model is selected as the previous sentence content after error correction with a first predetermined probability, and the next sentence content in the input is selected as the next sentence content after error correction with a second predetermined probability, to obtain the initial chapter error correction model.

[0094] In specific implementation, the error correction direction of the above step is controlled according to the probability (i.e., random direction). For example, the probability pa (i.e., the first predetermined probability) of using the error correction result of the previous sentence and the probability pb (i.e., the second predetermined probability) of using the error correction result of the next sentence, that is, the previous sentence content in the input has a probability pa of using the grammatically correct previous sentence content after error correction, and a probability 1-pa of using the unprocessed previous sentence content; the next sentence content has a probability pb of using the grammatically correct next sentence content after error correction, and a probability 1-pb of using the unprocessed next sentence content. When pa=1, pb=0, it is the error correction direction from the previous sentence to the next sentence, and when pa=0, pb=1, it is the error correction direction from the next sentence to the previous sentence.

[0095] Among them, the values ​​of pa and pb can be set according to actual needs, and the value range of pa and pb is [0,1].

[0096] Step 302 , obtain multiple chapter data, annotate each chapter data with a corresponding error correction result, and obtain multiple chapter error correction samples as a chapter error correction sample set.

[0097] In order to ensure the training effect, the corresponding error correction results need to be annotated for each chapter data, where the annotated content includes the position of the corresponding corrected sentence and the content after the error correction (including: text content or symbol content). Each chapter can be annotated with one or more positions, or it can be left unannotated (i.e. the annotated content is empty). Chapters without annotations prove that the content of the chapter is correct and do not need to be corrected.

[0098] In some embodiments, step 302 specifically includes:

[0099] Step 302A, receiving the input error correction result corresponding to each chapter data, marking each error correction result in the corresponding chapter data, obtaining a first marked sample as the chapter error correction sample, and taking multiple chapter error correction samples as the chapter error correction sample set.

[0100] In specific implementation, this step describes that the user can input the sentence position of the corresponding error correction, and the corresponding modified content as the error correction result. After receiving the error correction result, the error correction result can be marked at the corresponding sentence position of the corresponding chapter data. In this way, each chapter data can be marked by manual marking, wherein the chapter data can be articles, students' essays, emails, etc. After the marking is completed, the chapter error correction sample set can be directly obtained, and the initial error correction model can be trained using the chapter error correction sample set. The manually marked error correction results are more accurate.

[0101] Alternatively, in step 302B, noise content is added to each of the chapter data according to a noise adding algorithm to obtain noisy chapter data, the chapter data is annotated as a correction result in the noisy chapter data, a second annotated sample is obtained as the chapter error correction sample, and a plurality of the chapter error correction samples are used as the chapter error correction sample set.

[0102] In specific implementation, the corresponding noise adding algorithm includes: noise adding based on grammatical rules or noise adding using the reverse translation model. This noise adding method can add noise to a large amount of chapter data, improve the annotation efficiency, and quickly obtain a large number of chapter error correction samples.

[0103] In some embodiments, based on the above steps 302A and 302B, the method of obtaining the passage error correction sample set may further include:

[0104] Step 302C1 , dividing a plurality of chapter data into a first chapter group and a second chapter group, wherein the number N1 of the chapter data in the first chapter group is less than the number N2 of the chapter data in the second chapter group.

[0105] Step 302C2: determine the first annotation sample corresponding to each chapter data in the first chapter group to obtain a first chapter sample group.

[0106] Step 302C3: determine the second annotation sample corresponding to each chapter data in the second chapter group to obtain a second chapter sample group.

[0107] Step 302C4: Use the first chapter sample group and the second chapter sample group as the chapter error correction sample set.

[0108] In specific implementation, the obtained multiple chapter data can be annotated respectively using the two methods of step 302A and step 302B mentioned above, and the number of annotations in the two methods can be set according to actual conditions. Since the manual annotation method of step 302A is relatively slow, the number of chapter data in the first chapter group N1 is set to be less than the number of chapter data in the second chapter group N2, where N1+N2=the total amount of chapter data.

[0109] Through the above scheme, the two annotation methods can be mixed, so that the chapter error correction sample set obtained by the above two methods is more effective in training the initial error correction model, and the chapter error correction model obtained after training has higher error correction accuracy.

[0110] Step 303: input the passage error correction sample set into the initial error correction model and perform training processing based on the annotated error correction results to obtain a passage error correction model.

[0111] In specific implementation, before training the initial error correction model, the parameters of the initial error correction model can be initialized and adjusted using the parameters of the trained initial error correction model, or the parameters of the general pre-trained model can be used to initialize and adjust the parameters of the initial error correction model. In this way, the initial error correction model can have a certain error correction accuracy before training, thereby reducing the training time and improving the error correction accuracy of the passage error correction model obtained after training.

[0112] In some embodiments, step 303 includes:

[0113] Step 3031, according to the error correction direction of the initial error correction model, determine the input amount of the initial error correction model corresponding to each passage error correction sample in the passage error correction sample set.

[0114] In specific implementation, for the previous sentence in the input, the sentence before error correction is recorded as Prev 0 , the corrected sentence is Prev 1 Similarly, for the next sentence in the input, the record before error correction is Next 0 , after error correction, it is Next 1 ; Determine the input amount of the initial error correction model according to the selected error correction direction.

[0115] Step 3032: input the input of the initial error correction model into the initial error correction model, process the passage error correction sample according to the error correction direction, and obtain the processing result of the passage error correction sample.

[0116] Step 3033, compare the processing result with the error correction result marked in the passage error correction sample, adjust the internal parameters of the initial error correction model according to the comparison result, and complete the training of the passage error correction sample.

[0117] In specific implementation, after the first chapter error correction sample is trained according to the above process, the trained initial error correction model is used to train the next chapter error correction sample according to the process of steps 3031 to 3033, and this process is continuously iterated until all chapter error correction samples in the chapter error correction sample set are trained.

[0118] Step 3034, in response to determining that the training of the passage error correction sample set is all completed, the trained initial error correction model is used as the passage error correction model.

[0119] During specific implementation, the initial error correction model finally obtained through training is used as the chapter error correction model, so that the chapter error correction model can perform error correction processing on the chapter.

[0120] In addition, a certain number of test samples can also be collected, and the method for obtaining the test samples can be the same as the method for obtaining the above-mentioned passage error correction samples. The passage error correction model is tested using the test samples, and the output accuracy of the passage error correction model during the test process is statistically analyzed. If the accuracy obtained after all the test samples are tested is greater than or equal to the accuracy threshold (for example, 98%, 97% or 95%, etc., not specifically limited here), it is proved that the passage error correction model can be put into use for passage error correction processing. If it is less than the accuracy threshold, it is proved that the passage error correction model still needs to be trained, and the passage error correction model is retrained according to the training method of the passage error correction model described in the above embodiment until the accuracy obtained by the test is greater than or equal to the accuracy threshold.

[0121] Through the scheme of the above-mentioned embodiment, the parameters of the obtained initial error correction model are adjusted so that the initial chapter error correction model obtained after the adjustment can perform error correction processing based on the context of the sentences in the chapter, and then the initial error correction model is trained using the obtained chapter error correction sample set to obtain the chapter error correction model. This can shorten the model training time and improve the training efficiency. The obtained chapter error correction model can correct the entire chapter with higher accuracy, ensuring that the chapter after error correction is more accurate.

[0122] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0123] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0124] Based on the same technical concept, this embodiment proposes a text error correction device, which corresponds to the text error correction method of any of the above embodiments, such as Figure 4 As shown, including:

[0125] The receiving module 41 is configured to receive the chapter data to be processed, perform sentence segmentation processing on the chapter data to be processed, and obtain a plurality of sentences to be processed arranged in the order of the chapters;

[0126] The error correction processing module 42 is configured to determine a plurality of corresponding input quantities for the plurality of sentences to be processed according to the chapter order, input the plurality of input quantities into the chapter error correction model, perform error correction processing according to the chapter order, and obtain a plurality of corrected sentences;

[0127] The integration module 43 is configured to integrate the plurality of corrected sentences to obtain a corrected chapter.

[0128] In some embodiments, the error correction processing module 42 is further configured to:

[0129] The input quantity determination determines the input quantity corresponding to each sentence to be processed according to a preset error correction direction, including: [previous sentence content, next sentence content, sentence content to be processed]; the multiple input quantities are corrected according to the rolling modification strategy using the chapter error correction model to obtain multiple corrected sentences.

[0130] In some embodiments, the integration module 43 is further configured to:

[0131] In response to determining that all the error correction processing of the multiple sentences to be processed has been completed, the chapter error correction model is converted into an initial error correction model; the multiple corrected sentences are respectively subjected to grammatical correction processing using the initial error correction model to obtain multiple word-corrected sentences; the multiple word-corrected sentences are integrated to obtain a corrected chapter.

[0132] The device also includes a model acquisition module, which is configured to train the neural network using the sample set to obtain a paragraph error correction model;

[0133] The model acquisition module includes:

[0134] An adjustment unit is configured to adjust the parameters of the obtained initial error correction model to obtain an initial chapter error correction model, wherein the initial chapter error correction model can perform error correction processing based on the context of a sentence in the chapter;

[0135] The sample acquisition unit is configured to acquire multiple chapter data, mark the corresponding error correction result for each chapter data, and obtain multiple chapter error correction samples as a chapter error correction sample set;

[0136] The training unit is configured to input the passage error correction sample set into the initial error correction model and perform training processing based on the annotated error correction results to obtain a passage error correction model.

[0137] In some embodiments, the adjusting unit comprises:

[0138] An input / output adjustment subunit is configured to adjust the input amount and / or output amount of the obtained initial error correction model to obtain a contextual error correction model;

[0139] The error correction direction adjustment subunit is configured to adjust the error correction direction of the context error correction model to obtain the initial chapter error correction model.

[0140] In some embodiments, the input / output adjustment subunit is further configured to:

[0141] In response to determining that the obtained initial error correction model is a generative initial error correction model, wherein an encoder is provided on the input side of the generative initial error correction model; the input amount of the encoder in the generative initial error correction model is modified, and the modified input amount includes: [previous sentence content, next sentence content, current sentence content], and a contextual error correction model is obtained.

[0142] In some embodiments, the input / output adjustment subunit is further configured to:

[0143] In response to determining that the obtained initial error correction model is a discriminative initial error correction model; modifying the input quantity of the discriminative initial error correction model, and obtaining a modified input quantity including: [previous sentence content, next sentence content, current sentence content]; connecting the output layer corresponding to the current sentence content in the discriminative initial error correction model to the output side of the discriminative initial error correction model, so that the output layer corresponding to the current sentence content outputs the output result as an output quantity from the output side, thereby obtaining a contextual error correction model.

[0144] In some embodiments, the input / output adjustment subunit is further configured to:

[0145] Various parameters of the input amount of the initial error correction model are pre-set, including: the length of the previous sentence content, the length of the next sentence content and the adjacent sentence overlap parameter of the current sentence content; the previous sentence content in the input amount is adjusted according to the previous sentence content length, the next sentence content in the input amount is adjusted according to the next sentence content length, and the current sentence content in the input amount is adjusted according to the adjacent sentence overlap parameter of the current sentence content, to obtain a contextual error correction model.

[0146] In some embodiments, the error correction direction adjustment subunit is further configured to:

[0147] The error correction direction of the context error correction model is adjusted according to the error correction direction from the previous sentence to the next sentence to obtain the initial chapter error correction model, and the specific adjustment includes: the previous sentence content in the input amount is the previous sentence content after error correction, and the next sentence content is the unprocessed next sentence content;

[0148] Alternatively, the correction direction of the contextual correction model is adjusted from the next sentence to the previous sentence to obtain the initial paragraph correction model, and the specific adjustment includes: the previous sentence content in the input is the unprocessed previous sentence content, and the next sentence content is the next sentence content after correction;

[0149] Alternatively, the previous sentence content in the input of the contextual error correction model is selected as the previous sentence content after error correction with a first predetermined probability, and the next sentence content in the input is selected as the next sentence content after error correction with a second predetermined probability, to obtain the initial chapter error correction model.

[0150] In some embodiments, the training unit includes:

[0151] An input determination subunit is configured to determine the input amount of the initial error correction model corresponding to each passage error correction sample in the passage error correction sample set according to the error correction direction of the initial error correction model;

[0152] A processing subunit is configured to input the input of the initial error correction model into the initial error correction model, process the passage error correction sample according to the error correction direction, and obtain a processing result of the passage error correction sample;

[0153] A comparison subunit is configured to compare the processing result with the error correction result marked in the passage error correction sample, and adjust the internal parameters of the initial error correction model according to the comparison result to complete the training of the passage error correction sample;

[0154] The model determination subunit is configured to, in response to determining that the entire passage error correction sample set has been trained, use the trained initial error correction model as the passage error correction model.

[0155] In some embodiments, the sample acquisition unit is further configured to:

[0156] Receiving the input error correction result corresponding to each chapter data, marking each error correction result in the corresponding chapter data, obtaining a first marked sample as the chapter error correction sample, and taking a plurality of the chapter error correction samples as the chapter error correction sample set;

[0157] Alternatively, noise content is added to each of the chapter data according to a noise adding algorithm to obtain noisy chapter data, the chapter data is annotated as a correction result in the noisy chapter data, a second annotated sample is obtained as the chapter error correction sample, and a plurality of the chapter error correction samples are used as the chapter error correction sample set.

[0158] In some embodiments, the sample acquisition unit is further configured to:

[0159] Divide multiple chapter data into a first chapter group and a second chapter group, wherein the number of chapter data in the first chapter group is less than the number of chapter data in the second chapter group; determine a first annotated sample corresponding to each chapter data in the first chapter group to obtain a first chapter sample group; determine a second annotated sample corresponding to each chapter data in the second chapter group to obtain a second chapter sample group; and use the first chapter sample group and the second chapter sample group as the chapter error correction sample set.

[0160] For the convenience of description, the above device is described by dividing it into various modules according to its functions. Of course, when implementing the present disclosure, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0161] The device of the above embodiment is used to implement the corresponding method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0162] Based on the same technical concept, corresponding to the passage error correction method of any of the above embodiments, the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the passage error correction method described in any of the above embodiments when executing the program.

[0163] Figure 5 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 510, a memory 520, an input / output interface 530, a communication interface 540, and a bus 550. The processor 510, the memory 520, the input / output interface 530, and the communication interface 540 are connected to each other in communication within the device through the bus 550.

[0164] The processor 510 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0165] The memory 520 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 520 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 520 and are called and executed by the processor 510.

[0166] The input / output interface 530 is used to connect the input / output module to realize information input and output. The input / output module can be configured as a component in the device (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0167] The communication interface 540 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0168] The bus 550 comprises a pathway that transmits information between the various components of the device (eg, the processor 510 , the memory 520 , the input / output interface 530 , and the communication interface 540 ).

[0169] It should be noted that, although the above device only shows the processor 510, the memory 520, the input / output interface 530, the communication interface 540 and the bus 550, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0170] The electronic device of the above embodiment is used to implement the corresponding passage error correction method in any of the above embodiments, and has the beneficial effects of the corresponding passage error correction method embodiment, which will not be repeated here.

[0171] Based on the same technical concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the chapter error correction method described in any of the above embodiments.

[0172] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0173] The computer instructions stored in the storage medium of the above embodiments are used to enable the computer to execute the method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0174] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Based on the concept of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present disclosure as described above, which are not provided in detail for the sake of simplicity.

[0175] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present disclosure difficult to understand, the known power / ground connections to the integrated circuit (IC) chips and other components may or may not be shown in the provided figures. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present disclosure difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure will be implemented (that is, these details should be fully within the scope of understanding of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present disclosure, it is apparent to those skilled in the art that the embodiments of the present disclosure can be implemented without these specific details or with changes in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0176] Although the present disclosure has been described in conjunction with specific embodiments of the present disclosure, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0177] The embodiments of the present disclosure are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A method for correcting errors in a passage, characterized in that: include: Receiving the chapter data to be processed, performing sentence segmentation processing on the chapter data to be processed, and obtaining a plurality of sentences to be processed arranged in the order of the chapters; Determine the input amount corresponding to each sentence to be processed according to the preset error correction direction, including: the content of the previous sentence, the content of the next sentence and the content of the sentence to be processed; Using the paragraph error correction model to correct multiple inputs according to the rolling modification strategy, multiple corrected sentences are obtained; The multiple corrected sentences are integrated to obtain a corrected chapter.

2. The method according to claim 1, characterized in that Integrating the multiple corrected sentences to obtain a corrected chapter includes: In response to determining that the error correction processing of the plurality of sentences to be processed is all completed, converting the paragraph error correction model into an initial error correction model; Using the initial error correction model, respectively perform grammatical error correction processing on the plurality of corrected sentences to obtain a plurality of word-corrected sentences; The sentences after the error correction of the multiple words are integrated to obtain the corrected chapter.

3. The method according to claim 1, characterized in that The process of obtaining the paragraph error correction model includes: Adjusting the parameters of the obtained initial error correction model to obtain an initial chapter error correction model, wherein the initial chapter error correction model can perform error correction processing based on the context of a sentence in the chapter; Acquire multiple chapter data, annotate corresponding error correction results for each chapter data, and obtain multiple chapter error correction samples as a chapter error correction sample set; The chapter error correction sample set is input into the initial error correction model and training processing is performed based on the annotated error correction results to obtain a chapter error correction model.

4. The method according to claim 3, characterized in that The parameters of the obtained initial error correction model are adjusted to obtain the initial chapter error correction model, including: Adjusting the input amount and / or output amount of the obtained initial error correction model to obtain a contextual error correction model; The error correction direction of the context error correction model is adjusted to obtain the initial chapter error correction model.

5. The method according to claim 4, characterized in that The step of adjusting the input amount and / or output amount of the obtained initial error correction model to obtain a contextual error correction model includes: In response to determining that the obtained initial error correction model is a generative initial error correction model, wherein an encoder is provided at an input side of the generative initial error correction model; The input amount of the encoder in the generative initial error correction model is modified to obtain a modified input amount including: the content of the previous sentence, the content of the next sentence and the content of the current sentence, thereby obtaining a contextual error correction model.

6. The method according to claim 4, characterized in that The step of adjusting the input amount and / or output amount of the obtained initial error correction model to obtain a contextual error correction model includes: In response to determining that the obtained initial error correction model is a discriminant initial error correction model; Modifying the input of the discriminant initial error correction model, so that the modified input includes: the previous sentence content, the next sentence content and the current sentence content; The output layer corresponding to the current sentence content in the discriminant initial error correction model is connected to the output side of the discriminant initial error correction model, so that the output layer corresponding to the current sentence content outputs the output result as an output quantity from the output side to obtain a contextual error correction model.

7. The method according to any one of claims 5 or 6, characterized in that: The step of adjusting the input amount and / or output amount of the obtained initial error correction model to obtain a contextual error correction model includes: Presetting various parameters of the input quantity of the initial error correction model, including: the length of the previous sentence content, the length of the next sentence content and the adjacent sentence overlap parameter of the current sentence content; The previous sentence content in the input is adjusted according to the length of the previous sentence content, the next sentence content in the input is adjusted according to the length of the next sentence content, and the current sentence content in the input is adjusted according to the adjacent sentence overlap parameter of the current sentence content to obtain a contextual error correction model.

8. The method according to claim 5 or 6, characterized in that: The step of adjusting the error correction direction of the context error correction model to obtain the initial chapter error correction model includes: The error correction direction of the context error correction model is adjusted according to the error correction direction from the previous sentence to the next sentence to obtain the initial chapter error correction model, and the specific adjustment includes: the previous sentence content in the input amount is the previous sentence content after error correction, and the next sentence content is the unprocessed next sentence content; or, The error correction direction of the context error correction model is adjusted according to the error correction direction from the next sentence to the previous sentence to obtain the initial chapter error correction model, and the specific adjustment includes: the previous sentence content in the input amount is the unprocessed previous sentence content, and the next sentence content is the next sentence content after error correction; Alternatively, the previous sentence content in the input of the contextual error correction model is selected as the previous sentence content after error correction with a first predetermined probability, and the next sentence content in the input is selected as the next sentence content after error correction with a second predetermined probability, to obtain the initial chapter error correction model.

9. The method according to claim 4, characterized in that The step of inputting the passage error correction sample set into the initial error correction model and performing training processing based on the annotated error correction results to obtain the passage error correction model includes: Determining the input amount of the initial error correction model corresponding to each passage error correction sample in the passage error correction sample set according to the error correction direction of the initial error correction model; Inputting the input of the initial error correction model into the initial error correction model, processing the passage error correction sample according to the error correction direction, and obtaining a processing result of the passage error correction sample; Comparing the processing result with the error correction result marked in the passage error correction sample, adjusting the internal parameters of the initial error correction model according to the comparison result, and completing the training of the passage error correction sample; In response to determining that the training of the passage error correction sample set is all completed, the trained initial error correction model is used as the passage error correction model.

10. The method according to claim 3, characterized in that: The step of marking each chapter data with a corresponding error correction result to obtain a plurality of chapter error correction samples as a chapter error correction sample set includes: Receiving the input error correction result corresponding to each chapter data, marking each error correction result in the corresponding chapter data, obtaining a first marked sample as the chapter error correction sample, and taking a plurality of the chapter error correction samples as the chapter error correction sample set; or, According to the noise adding algorithm, noise content is added to each of the chapter data to obtain the noisy chapter data, and the chapter data is annotated as the error correction result in the noisy chapter data to obtain a second annotated sample as the chapter error correction sample, and a plurality of the chapter error correction samples are used as the chapter error correction sample set.

11. The method according to claim 10, characterized in that The corresponding error correction results are marked for each chapter data to obtain a chapter error correction sample set, including: Dividing a plurality of chapter data into a first chapter group and a second chapter group, wherein the number of chapter data in the first chapter group is less than the number of chapter data in the second chapter group; Determine a first annotated sample corresponding to each chapter data in the first chapter group to obtain a first chapter sample group; Determine a second annotated sample corresponding to each chapter data in the second chapter group to obtain a second chapter sample group; The first chapter sample group and the second chapter sample group are used as the chapter error correction sample set.

12. A device for correcting errors in a passage, characterized in that: include: A receiving module is configured to receive the chapter data to be processed, perform sentence segmentation processing on the chapter data to be processed, and obtain a plurality of sentences to be processed arranged in the order of the chapters; The error correction processing module is configured to determine the input amount corresponding to each sentence to be processed according to a preset error correction direction, including: the content of the previous sentence, the content of the next sentence and the content of the sentence to be processed; use the chapter error correction model to perform error correction processing on the multiple input amounts according to the rolling modification strategy to obtain multiple corrected sentences; The integration module is configured to integrate the multiple corrected sentences to obtain a corrected chapter.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 11 is implemented.

14. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 11.

Images