Information processing method, program, and information processing system

Abstract

To provide an information processing method capable of improving the convenience for users.SOLUTION: An information processing method to be executed by a computer includes: generating a model generation text from an input sentence, using a language model; acquiring a token sequence Otrg to be calibrated, which is a part of the model generation text that needs calibration; extracting a prefix token sequence Opre and a suffix token sequence Osuf from a part of the model generation text excluding the token sequence Otrg to be calibrated; and generating a substitute text to be substituted for a text of the token sequence Otrg to be calibrated, based on the input text, the prefix token sequence Opre and the suffix token sequence Osuf.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to an information processing method, a program, and an information processing system. [Background technology]

[0002] A conventional system is described in Patent Document 1 below. This system creates a summary text from a text sentence using a learning model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-16181 Summary of the Invention [Problem to be solved by the invention]

[0004] In a system such as that described in Patent Document 1, if the content of the created text is incorrect, it may be difficult to correct the error. An object of an embodiment of the present disclosure is to provide an information processing method, a program, and an information processing system that can improve user convenience. [Means for solving the problem]

[0005] An information processing method of one embodiment is an information processing method executed by a computer, and includes generating a second sentence from a first sentence using a language model, obtaining a portion of the second sentence that needs to be proofread, extracting a portion of the second sentence that is not to be proofread from a portion of the second sentence excluding the portion to be proofread, and generating an alternative sentence to replace the sentence in the portion that is to be proofread, based on the first sentence and the portion that is not to be proofread.

[0006] This method can improve user convenience. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a block diagram showing a schematic configuration of an information processing system according to an embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the operation of the information processing system according to the embodiment. [Figure 3] FIG. 10 is a diagram illustrating an example of information displayed on the terminal device according to the embodiment. [Figure 4] FIG. 1 is a block diagram showing a configuration of an information processing apparatus according to an embodiment. [Figure 5] FIG. 2 is a diagram schematically illustrating an example of the operation of a language model according to the embodiment. [Figure 6] FIG. 10 is a diagram illustrating an example of information displayed on the terminal device according to the embodiment. [Figure 7] FIG. 4 is a diagram showing an example of a model generation sentence according to the embodiment. [Figure 8] FIG. 2 is a diagram schematically illustrating an example of the operation of a language model according to the embodiment. [Figure 9] 4 is a sequence chart showing an example of the operation of the information processing system according to the embodiment. [Figure 10] FIG. 1 is a block diagram showing a hardware configuration of a computer according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, an embodiment of an information processing method, a program, and an information processing system will be described with reference to the drawings. To facilitate understanding of the description, the same components in each drawing are denoted by the same reference numerals as much as possible, and redundant description will be omitted.

[0009] <Embodiment> First, an overview of the information processing system of this embodiment will be described.

[0010] (Outline of information processing system) 1, an information processing system 10 of this embodiment includes a terminal device 20 and an information processing device 30. The terminal device 20 and the information processing device 30 are communicably connected to each other via a network line N. The network line N is a communication network capable of high-speed communication, such as a wired or wireless communication network such as the Internet, an intranet, or a LAN (Local Area Network).

[0011] In this information processing system 10, as shown in FIG. 2, when a user operates a terminal device 20 to input a first input sentence S10, the information of the first input sentence S10 is transmitted to an information processing device 30 via a network line N.

[0012] When the information processing device 30 of this embodiment receives information about a first input sentence S10 transmitted from the terminal device 20, it inputs the received first input sentence S10 into the language model 310. The language model 310 generates a first output sentence S20 from the first input sentence S10. FIG. 3 is an example of the first input sentence S10. The first input sentence S10 shown in FIG. 3 includes a task instruction sentence S11 and a target sentence S12. The task instruction sentence S11 is a sentence that instructs what processing should be performed on the target sentence S12. In this embodiment, the target sentence S12 is an example of a first sentence.

[0013] In this embodiment, when a first input sentence S10 as shown in FIG. 3 is input, the language model 310 extracts a task instruction S11 and a target sentence S12 from the first input sentence S10 and generates a first output sentence S20 by performing task processing on the target sentence S12 according to the extracted task instruction S11. For example, if the task instruction S11 instructs the creation of a summary as shown in FIG. 3, the language model 310 executes task processing to create a summary of the target sentence S12, thereby generating a first output sentence S20 including a summary Sm of the target sentence S12. Note that generating a summary is an example of task processing, and the language model 310 may also execute task processing related to any sentence generation, such as a simple sentence generation process or a translation process. Hereinafter, the sentence Sm generated by the task of the language model 310 in this manner will be referred to as a "model-generated sentence Sm." In this embodiment, the model-generated sentence Sm is an example of a second sentence. When the information processing device 30 generates the first output sentence S20 in this way, it transmits information about the first output sentence S20 to the terminal device 20.

[0014] When the terminal device 20 of this embodiment receives information about the first output sentence S20 transmitted from the information processing device 30, it displays the received first output sentence S20, for example, as shown in Fig. 3. In the first output sentence S20, the sentence "I conveyed my request to change the items covered by Article 9 and delete Article 12" corresponds to the model-generated sentence Sm. In this way, in the information processing system 10 of this embodiment, the user can easily obtain the model-generated sentence Sm displayed on the terminal device 20, in other words, a more appropriate sentence generated by the language model 310.

[0015] In this embodiment, an example will be described in which text exchange between a user and the language model 310 is carried out in the form of a text chat on the terminal device 20 as shown in FIG.

[0016] Comparing the first input sentence S10 and the model-generated sentence Sm shown in FIG. 3 reveals that the model-generated sentence Sm contains some errors. Specifically, the content of the model-generated sentence Sm, "Please delete Article 12," is consistent with the content S122 of the first input sentence S10. On the other hand, the content of the model-generated sentence Sm, "Changes to the items covered by Article 9," is slightly different from the content S121 of the first input sentence S10. When generating a sentence such as a summary using the language model 310, a sentence that is inaccurate or inappropriate, including grammatical errors, may be generated. Therefore, in practice, the sentence generated by the language model 310 must be manually checked to identify problematic parts and proofread (including rewriting and correction). However, this process requires time and leaves room for improvement. The information processing device 30 of this embodiment improves user convenience by facilitating proofreading of such sentences.

[0017] (Terminal Device Configuration) Next, the configuration of the terminal device 20 of this embodiment will be described.

[0018] The terminal device 20 of this embodiment is a device used by a user. The terminal device 20 is a personal computer, a smartphone, a tablet terminal, or the like. The user can launch a web browser or a predetermined application on the terminal device 20. In the web browser or the predetermined application, it is possible to input text to be sent to the information processing device 30 and to display text to be sent from the information processing device 30.

[0019] (Configuration of information processing device) Next, the configuration of the information processing device 30 of this embodiment will be described.

[0020] The information processing device 30 of this embodiment is a server device, etc. As shown in Fig. 4, the information processing device 30 includes a storage unit 31, a communication unit 32, and a control unit 33, etc. The storage unit 31, the communication unit 32, and the control unit 33 are electrically connected to each other.

[0021] The storage unit 31 of this embodiment stores various programs, various data, and the like for operating the information processing device 30. The storage unit 31 also stores a language model 310. The language model 310 is, for example, a large-scale language model. The large-scale language model is, for example, a natural language processing model in which a machine learning model with several hundred million or more parameters is trained with several hundred GB or more of training data. Examples of large-scale language models include GPT-3, GPT-4, BERT, and LLaMA.

[0022] The language model 310 may be provided in a server device or the like separate from the information processing device 30. In this case, the information processing device 30 uses the language model 310 by communicating with the separate server device or the like via a network line N.

[0023] The communication unit 32 of this embodiment performs various communications with, for example, the terminal device 20. The communication unit 32 acquires, for example, various pieces of input information input by a user to the terminal device 20. The communication unit 32 also transmits, to the terminal device 20, various pieces of information to be displayed on the terminal device 20.

[0024] The control unit 33 of this embodiment controls, for example, the information processing device 30. The control unit 33 includes, as functional components realized by executing a program stored in the storage unit 31, an acquisition unit 330, an output unit 331, a sentence generation unit 332, a proofreading target acquisition unit 333, and an alternative sentence generation unit 334.

[0025] The acquisition unit 330 of this embodiment acquires, from the terminal device 20, information on various sentences input by a user through a web browser or a predetermined application on the terminal device 20. For example, when a user inputs a first input sentence S10 as shown in FIG. 3 on the terminal device 20, the acquisition unit 330 acquires information on this first input sentence S10 from the terminal device 20.

[0026] The output unit 331 of this embodiment transmits information on various sentences generated by the language model 310 to the terminal device 20. Upon receiving the information on various sentences transmitted from the output unit 331, the terminal device 20 displays the various sentences using a web browser or a predetermined application. For example, when a first output sentence S20 as shown in FIG. 3 is generated by the language model 310, the output unit 331 transmits information on the first output sentence S20 to the terminal device 20. As a result, the first output sentence S20 as shown in FIG. 3 is displayed on the terminal device 20.

[0027] The sentence generation unit 332 of this embodiment generates a first output sentence S20 from a first input sentence S10 acquired by the acquisition unit 330, using the language model 310. Specifically, when the acquisition unit 330 acquires the first input sentence S10, the sentence generation unit 332 inputs the first input sentence S10 to the language model 310 as shown in FIG. 5 . As a result, the language model 310 generates the first output sentence S20 by performing task processing on the target sentence S12 in accordance with the task instruction sentence S11, based on the task instruction sentence S11 and the target sentence S12 included in the first input sentence S10. When the first output sentence S20 is generated by the language model 310 in this manner, the sentence generation unit 332 transmits information about the first output sentence S20 to the terminal device 20 via the output unit 331, thereby displaying the first output sentence S20 on the terminal device 20.

[0028] The proofreading target acquisition unit 333 of this embodiment acquires a proofreading target token sequence O, which is a set of tokens that need to be proofread in the model generation sentence Sm generated by the language model 310. trg For example, if the user determines that the model generation sentence Sm shown in FIG. 3 needs to be proofread, the user operates the terminal device 20 to input a second input sentence S30 as shown in FIG. 6. In this case, the acquisition unit 330 acquires information on the second input sentence S30 from the terminal device 20. The proofreading target acquisition unit 333 extracts the part "change of target item" from the second input sentence S30 acquired by the acquisition unit 330 and converts it into a proofreading target token string O. trgExtract as. More generally, given the token sequence of the model-generated sentence Sm generated by the language model 310 as O(=[o1,...,o n ), when the range to be corrected in the token sequence O is (stt, end), the correction target acquisition unit 333 uses [o stt ,...,o end-1 as the correction target token sequence O trg for acquisition. Here, "n", "stt", and "end" are natural numbers that satisfy "1≦stt<end≦n". In this embodiment, the part corresponding to the correction target token sequence O trg in the model-generated sentence Sm is an example of the correction target part.

[0029] The alternative sentence generation unit 334 of this embodiment generates candidates for the alternative sentence Sa that replaces the sentence of the correction target token sequence O trg . The alternative sentence is syntactically and semantically different from the sentence of the correction target token sequence O trg , and is also a correct sentence both grammatically and as the output of the task processing even when replaced with the sentence of the correction target token sequence O trg . For example, as shown in FIG. 7, the alternative sentence generation unit 334 extracts the token sequence [o1,...,o trg of the part described before the correction target token sequence O stt-1 from the token sequence O of the model-generated sentence Sm as the prefix token sequence O pre . Further, the alternative sentence generation unit 334 extracts the token sequence [o trg ,...,o end ,...,o n of the part described after the correction target token sequence O suf from the token sequence O of the model-generated sentence Sm as the suffix token sequence O pre . In this embodiment, the parts corresponding to the prefix token sequence O suf and the suffix token sequence O pre in the model-generated sentence Sm are examples of non-correction target parts. Also, the prefix token sequence O suf and the suffix token sequence O are examples of non-correction target token sequences. Furthermore, the prefix token sequence Opre is an example of a first token sequence, and the suffix token sequence O suf is an example of the second token sequence.

[0030] Then, as shown in FIG. 8, the alternative sentence generation unit 334 generates the extracted prefix token string O pre and suffix token sequence O suf , and the first input sentence S10 are input to the language model 310. As a result, the language model 310 generates a token sequence O trg The alternative sentence generation unit 334 generates a second output sentence S40 including one or more alternative sentences Sa that replace the sentence in question. After generating the second output sentence S40 including one or more alternative sentences Sa in this manner, the alternative sentence generation unit 334 transmits the second output sentence S40 to the terminal device 20 via the output unit 331, thereby causing the second output sentence S40 as shown in FIG. 6 to be displayed on the terminal device 20. The second output sentence S40 includes a model generation sentence Sm in addition to the multiple alternative sentences Sa1 to Sa3. The underlined parts in this model generation sentence Sm are parts that need to be proofread, in other words, the proofreading target token sequence O trg This is the part that corresponds to.

[0031] Next, an example of a method for generating the second output sentence S40 using the language model 310 will be described. A typical language model generates text by sequentially outputting the next word from the beginning of a sentence through a process called decoding. Beam search is one of the decoding methods for such language generation tasks. At each time t in the algorithm, beam search outputs the B candidates with the highest scores from the token string of length t that has been searched so far, called a beam. At the next time t+1, a beam for time t+1 is created by adding one token based on the beam at time t. This process is repeated until a token indicating the end of the sentence (eos) is added. The beam at the final time is the generated sentence of the model. Here, B is called the beam width. Also, at the start of processing at time t>1, the set of B beams held by beam search is called Y t-1 ={y 1,t-1 ,···,yB,t-1} where Y t-1 element y of i,t-1 is the token sequence generated up to time t-1. At time t, the beam search method is t =Y t-1 ×V, and the next time, the token sequence Y with the highest score is t where V is the vocabulary handled by the language model.

[0032] The beam search method finds a set of beams Y at the end of time t. t is selected according to the following formula f1.

[0033]

number

[0034] Next, a description will be given of the suffix consideration method, which is a method for generating a sentence by taking into account the following context and is used in the language model 310 of this embodiment.

[0035] As mentioned above, the normal beam search method outputs one token from the vocabulary V at each time or outputs (eos) tokens to finish sentence generation. Therefore, in the normal beam search method, the suffix token sequence O suf It is difficult to generate a sentence that naturally connects to (eos) in a grammatical sense. To solve this problem, the language model 310 of this embodiment uses a suffix-based method, which is an extension of the score function of the beam search method. The suffix-based method outputs one token from the vocabulary V at each time, or generates a suffix token string O instead of (eos).suf At this time, the suffix token sequence O suf Generate a score P(O suf |w <t ) is shown in the following equation f2.

[0036]

number

[0037] (Example of information processing system operation) Next, an example of the operation of the information processing system 10 of this embodiment will be described.

[0038] As shown in FIG. 9, in the information processing system 10 of this embodiment, first, when a user inputs a first input sentence S10 as shown in FIG. 3 into the terminal device 20 (step St10), the terminal device 20 transmits information about the first input sentence S10 to the information processing device 30 (step St11).

[0039] In the information processing device 30, when the acquisition unit 330 of the control unit 33 acquires a first input sentence S10 transmitted from the terminal device 20 (step St20), the sentence generation unit 332 of the control unit 33 inputs the first input sentence S10 to the language model 310 (step St21). As a result, the language model 310 generates a first output sentence S20 including a model generation sentence Sm based on the task instruction sentence S11 and the target sentence S12 contained in the first input sentence S10 (step St30), and outputs the generated first output sentence S20 to the control unit 33 (step St31). When the sentence generation unit 332 of the control unit 33 acquires the first output sentence S20 output from the language model 310 (step St22), it transmits the acquired first output sentence S20 to the terminal device 20 (step St23).

[0040] When the terminal device 20 receives the first output sentence S20 transmitted from the information processing device 30 (step St12), it displays the first output sentence S20 including the model-generated sentence Sm in the form shown in Fig. 3 (step St13). Suppose the user subsequently inputs the second input sentence S30 as shown in Fig. 6 in order to partially proofread the model-generated sentence Sm included in the first output sentence S20 (step St14), then the terminal device 20 transmits information about the second input sentence S30 to the information processing device 30 (step St15).

[0041] In the information processing device 30, when the second input sentence S30 transmitted from the terminal device 20 is received by the acquisition unit 330 of the control unit 33 (step St24), the proofreading target acquisition unit 333 of the control unit 33 extracts a proofreading target token sequence O from the second input sentence S30. trg (Step St25). Next, the alternative sentence generation unit 334 extracts the token string O trg Based on the model generation, prefix token sequence O is generated from sentence Sm. pre and suffix token sequence O suf (Step St26). The alternative sentence generation unit 334 also extracts the first input sentence S10, the prefix token string O pre , and a suffix token sequence O sufThe language model 310 is then requested to generate an alternative sentence Sa using the first input sentence S10, the prefix token string O, and the pre , and a suffix token sequence O suf Based on this, the alternative sentence generation unit 334 generates a second output sentence S40 including one or more alternative sentences Sa (step St32), and outputs the generated second output sentence S40 to the control unit 33 (step St33). When the alternative sentence generation unit 334 of the control unit 33 acquires the second output sentence S40 output from the language model 310 (step St28), the alternative sentence generation unit 334 transmits the acquired second output sentence S40 to the terminal device 20 (step St29).

[0042] When the terminal device 20 receives the second output sentence S40 transmitted from the information processing device 30 (step St16), it displays the second output sentence S40 including multiple alternative sentences Sa1 to Sa3 in a form such as that shown in FIG. 6 (step St17).

[0043] (Hardware configuration of document processing system) Next, an example of a hardware configuration in which the terminal device 20 and the information processing device 30 of this embodiment are realized by a computer 100 will be described with reference to Fig. 10. Fig. 10 is a diagram showing an example of the hardware configuration of the computer 100.

[0044] 10, the computer 100 includes, for example, a processor 101, a memory 102, a storage device 103, an input I / F 104, a data I / F 105, a communication I / F 106, and a display device 107. Note that the computer 100 may include multiple processors 101, multiple memories 102, multiple storage devices 103, multiple input I / Fs 104, multiple data I / Fs 105, multiple communication I / Fs 106, and multiple display devices 107.

[0045] Computer 100 may be, for example, a server computer, a personal computer (e.g., desktop, laptop, tablet, etc.), a media computing platform (e.g., cable, satellite set-top box, digital video recorder, etc.), a handheld computing device (e.g., PDA, email client, etc.), or any other type of computing or communications platform.

[0046] The processor 101 is a control unit that controls various processes in the computer 100 by executing programs stored in the memory 102, for example.

[0047] The memory 102 is a storage medium such as a RAM (Random Access Memory), etc. The memory 102 temporarily stores the program code of the program executed by the processor 101 and data required when the program is executed.

[0048] The storage device 103 is a non-volatile storage medium such as a hard disk drive (HDD), flash memory, etc. The storage device 103 stores an operating system and various programs for realizing the above-mentioned components.

[0049] The input I / F 104 is a device for receiving input from a user. The input I / F 104 may be, for example, a keyboard, a mouse, a touch panel, various sensors, a wearable device, etc. The input I / F 104 may be connected to the computer 100 via an interface such as a USB (Universal Serial Bus).

[0050] The data I / F 105 is a device for inputting data from outside the computer 100. The data I / F 105 is, for example, a drive device for reading data stored in various storage media. The data I / F 105 may be provided outside the computer 100. When the data I / F 105 is provided outside the computer 100, the data I / F 105 is connected to the computer 100 via an interface such as a USB.

[0051] The communication I / F 106 is a device for performing data communication via a network such as the Internet, either wired or wirelessly, with devices external to the computer 100. The communication I / F 106 may be provided outside the computer 100. When the communication I / F 106 is provided outside the computer 100, the communication I / F 106 is connected to the computer 100 via an interface such as a USB.

[0052] The display device 107 is a device for displaying various types of information. The display device 107 is, for example, a liquid crystal display, an organic EL (Electro-Luminescence) display, a display of a wearable device, or the like. The display device 107 may be provided outside the computer 100. When the display device 107 is provided outside the computer 100, the display device 107 is connected to the computer 100 via, for example, a display cable. Furthermore, when a touch panel is adopted as the input I / F 104, the display device 107 may be configured as an integral part of the input I / F 104.

[0053] (Actions and Effects of the Information Processing System of the Present Embodiment) As described above, the information processing system 10 of this embodiment includes an acquisition unit 330, a sentence generation unit 332, a proofreading target acquisition unit 333, and an alternative sentence generation unit 334. The acquisition unit 330 acquires the target sentence S12 (first sentence) of the first input sentence S10. The sentence generation unit 332 generates the model generation sentence Sm (second sentence) from the target sentence S12 using the language model 310. The proofreading target acquisition unit 333 acquires the proofreading target token sequence O (S12) which is the part of the model generation sentence Sm that needs proofreading. trg The alternative sentence generation unit 334 obtains the token sequence O (the part to be corrected) in the model generation sentence Sm. trg excluding prefix token sequence O pre and suffix token sequence O suf (non-proofreading target part), target sentence S12, prefix token sequence O pre , and a suffix token sequence O suf Generate an alternative sentence Sa based on the above.

[0054] According to this configuration, the token sequence O trg Since the alternative sentence Sa is automatically generated as a replacement for the sentence S12, it is possible to improve user convenience. pre , and a suffix token sequence O suf By generating the alternative sentence Sa based on the above, it is possible to generate a more appropriate alternative sentence Sa.

[0055] The alternative sentence generation unit 334 of this embodiment receives a target sentence S12, a prefix token sequence O pre , and a suffix token sequence O suf A plurality of alternative sentences Sa are generated based on the above.

[0056] According to this configuration, the user can select a more appropriate alternative sentence from among the multiple alternative sentences Sa, which can further improve user convenience.

[0057] The alternative sentence generation unit 334 of this embodiment causes the terminal device 20 to display the alternative sentence Sa together with the model-generated sentence Sm created by the language model 310.

[0058] According to this configuration, it is possible to compare the model generation sentence Sm with the alternative sentence Sa and determine whether the alternative sentence Sa is appropriate, thereby further improving user convenience.

[0059] (Other embodiments) The present disclosure is not limited to the above specific examples.

[0060] For example, when a plurality of alternative sentences Sa1 to Sa3 are displayed on the terminal device 20 as the second output sentence S40 as shown in Fig. 7, the alternative sentence generation unit 334 may accept a user operation to select one of the alternative sentences Sa1 to Sa3. In this case, the alternative sentence generation unit 334 generates a proofreading target token sequence O from the alternative sentence selected by the user. trg may be displayed on the terminal device 20. For example, when the user selects the alternative sentence Sa1, the alternative sentence generation unit 334 causes the terminal device 20 to display the following sentence as the model generation sentence Sm after proofreading: "I have conveyed my request to change the items excluded from the target of Article 9 and to delete Article 12."

[0061] The alternative sentence generator 334 generates a prefix token sequence O pre and suffix token sequence O suf Alternatively, the alternative sentence generation unit 334 may generate the alternative sentence Sa by using only one of the first input sentence S10 and the prefix token sequence O. pre Alternatively, the alternative sentence generation unit 334 may generate a second output sentence S40 including one or more alternative sentences Sa by inputting the first input sentence S10 and the suffix token string O suf is input to the language model 310 to generate a second output sentence S40 including one or more alternative sentences Sa.

[0062] The proofreading target acquisition unit 333 acquires a proofreading target token sequence O, which is a portion of the model generation sentence Sm that needs proofreading. trg The proofreading target acquisition unit 333 is not limited to acquiring the parts that need proofreading or rewriting based on a user's input operation, but may acquire the parts that need proofreading or rewriting from the model-generated sentence Sm using another method. For example, the proofreading target acquisition unit 333 may acquire the parts that need proofreading or rewriting from the model-generated sentence Sm using large language models (LLM). Furthermore, instead of a large language model, the proofreading target acquisition unit 333 may use a processing program or the like that can mechanically determine the content of a sentence using numerical values.

[0063] Design modifications made by a person skilled in the art to the above specific examples as appropriate are also included within the scope of the present disclosure as long as they comprise the features of the present disclosure. The elements of each of the above specific examples, as well as their arrangement, conditions, shape, etc., are not limited to those exemplified and can be modified as appropriate. The elements of each of the above specific examples can be combined as appropriate as long as no technical contradictions arise. [Explanation of symbols]

[0064] 10: Information processing system, 20: Terminal device, 100: Computer, 330: Acquisition unit, 332: Sentence generation unit, 333: Proofreading target acquisition unit, 334: Alternative sentence generation unit.

Claims

1. 1. A computer-implemented information processing method, comprising: generating a second sentence from the first sentence using a language model; acquiring a proofreading target portion of the second sentence that needs proofreading; extracting a non-proofreading target portion from a portion of the second sentence excluding the proofreading target portion; generating an alternative sentence to replace the sentence of the portion to be proofread based on the first sentence and the portion not to be proofread. Information processing methods.

2. a token sequence to be proofread, which is a token sequence of a portion of the second sentence that needs proofreading, is acquired as the portion to be proofread; As the non-proofreading target portion, a token sequence not to be proofread, which is a token sequence of a portion of the second sentence excluding the proofreading target portion, is acquired. The information processing method according to claim 1 .

3. extracting, from the second sentence, as the non-proofreading target portion, a second token sequence which is a token sequence of a portion written after the proofreading target portion; generating the alternative sentence from the first sentence and the second token sequence using the language model; The information processing method according to claim 2 .

4. extracting, from the second sentence, as the non-proofreading target portion, a first token sequence which is a token sequence of a portion written before the proofreading target portion; generating the alternative sentence from the first sentence, the first token sequence, and the second token sequence using the language model; The information processing method according to claim 3 .

5. The first sentence is acquired from the user's terminal device. The information processing method according to claim 1 .

6. On the computer, A process of generating a second sentence from the first sentence using a language model; A process of acquiring a proofreading target portion in the second sentence, which is a portion that needs proofreading; a process of extracting a non-proofreading target portion from a portion of the second sentence excluding the proofreading target portion; and generating a substitute sentence for the sentence of the portion to be proofreading based on the first sentence and the portion not to be proofreading. program.

7. a sentence generation unit that generates a second sentence from the first sentence using a language model; a proofreading target acquisition unit that acquires a proofreading target portion that is a portion of the second sentence that needs proofreading; an alternative sentence generation unit that extracts a non-proofreading target portion from a portion of the second sentence excluding the proofreading target portion, and generates an alternative sentence to replace the proofreading target portion based on the first sentence and the non-proofreading target portion. Information processing system.

Images