Computer system and text generation control method

Abstract

A server system (1100) generates a first detailed text (scenario) that serves as an original by using a generation unit, and generates a summary text obtained by summarizing the first detailed text. Furthermore, the server system develops / expands a story from the summary so as to generate a scene, while keeping the content of the summary unchanged, and then generates a new detailed text (a second detailed text). The server system (1100) evaluates the new detailed text, replaces the detailed text to be summarized with the new detailed text, and repeats summarization and expansion until acceptable.

Description

Computer System and Text Generation Control Method

[0001] The present invention relates to a computer system and the like.

[0002] Patent Document 1 describes a scenario plot generation device that generates a scenario plot of a story consisting of a plurality of phases (corresponding to the beginning, middle, turning point, and end of a story). The scenario plot generation device of Patent Document 1 has a database in which blocks, which are samples of each phase of the scenario plot, are tagged and recorded in advance. The scenario plot generation device receives an input of tags from a user, reads out the blocks corresponding to the input tags, and generates a scenario plot.

[0003] Japanese Patent Application Laid-Open No. 2019-220039

[0004] As a technology different from that of Patent Document 1, a method of generating scenario data using a generation unit typified by generative AI (Artificial Intelligence) is known. For example, in the field of game production, a method for using the scenario generated by the generation unit in game production is being explored.

[0005] In order to cause the generation unit to generate data, it is necessary to give the generation unit information (generation instruction information) for a generation instruction called a prompt. The generation instruction information includes a plurality of types of contexts that describe what to generate, under what preconditions (generation conditions) to generate, and what data to refer to for generation.

[0006] The user creates generation instruction information so that a scenario (product) with the desired specifications and content can be obtained and gives this to the generation unit. However, the generation unit does not always generate the scenario that the user has envisioned.

[0007] In particular, as the amount of text increases, the generated scenarios tend to contain inconsistencies, excessive leaps in the story's development (so-called "super-realistic plot developments"), and unnecessary or boring plot points. Users have to read through long scenarios, and if the specifications and content do not match what they envisioned, they revise the generation instructions and feed them back to the generation unit to restart the generation process. Users have to repeat this process, which is one reason why production efficiency does not improve as much as desired despite utilizing the generation unit.

[0008] Furthermore, these problems are not limited to screenplays for films, plays, and games, but also apply to other forms of writing such as novels.

[0009] The problem that this invention aims to solve is to provide a new technology that enables improved efficiency in generating texts such as scenarios.

[0010] The first disclosure is a computer system comprising: a summary text information acquisition control unit that controls the acquisition of a first summary text information for a given detailed text information using a summary generation unit that can generate a summary, abstract, outline, overview, or digest (hereinafter collectively referred to as "summary") for a given detailed text information; and a detailed text information acquisition control unit that controls the acquisition of a second detailed text information for the first summary text information using an extension generation unit that can generate detailed text information by extending the given summary text information.

[0011] According to the first disclosure, the computer system can generate a new document based on the first detailed document by summarizing the original first detailed document and generating a second detailed document by expanding from the summary. This makes it possible to improve the efficiency of document generation.

[0012] Even if the initial detailed text contains inconsistencies, abrupt plot developments, or unnecessary and boring sections, summarizing it is highly likely that one or more of these flaws will be eliminated. Then, the story is developed and expanded based on this summary to generate new text. This makes it possible to generate a better text with fewer flaws compared to generating new text from scratch.

[0013] Of course, some inconsistencies may remain during the summarization process, but the expansion will be carried out based on those remaining inconsistencies. Therefore, parts that didn't make sense before summarization may become consistent, and parts that seemed like absurd developments are likely to be corrected into more appropriate developments.

[0014] The second disclosure is a computer system further comprising a repeat control unit that performs a cyclical repeat control, which repeatedly causes the summary text information acquisition control unit to acquire new first summary text information for the second detailed text information as new first detailed text information, and then causes the detail text information acquisition control unit to acquire new second detailed text information for the new first summary text information.

[0015] According to the second disclosure, the computer system will be able to repeatedly summarize and expand. Therefore, by repeatedly summarizing and expanding, the original first detailed text will be gradually revised to remove inconsistencies and illogical developments, and transformed into a better text.

[0016] Furthermore, the third disclosure is a computer system further comprising a detailed text information evaluation unit that evaluates the second detailed text information, wherein the repeat control unit executes the loop repeat control based on the evaluation result of the detailed text information evaluation unit.

[0017] According to the third disclosure, the computer system can evaluate the second detailed text information and, based on the evaluation results, perform summarization and expansion again. The computer system can automatically perform summarization and expansion multiple times until it can evaluate that the original first detailed text has become a better text.

[0018] The fourth disclosure is a computer system in which, in the above-described computer system, the summary text information acquisition control unit variably controls the amount of summary in the summary generation unit based on the number of repetitions of the loop repeat control.

[0019] The summarization amount refers to the length of the first summary text, indicating how much detail from the original first detailed text is omitted. On the other hand, the repetitive cycle control revises the original first detailed text to generate the second detailed text. However, if the second detailed text is summarized again with the same summarization amount, there is a possibility that even the parts that have become good sentences will be cut out. Therefore, according to the fourth disclosure, the computer system can suppress the waste of the results obtained by the repetitive cycle control by variably controlling the summarization amount based on the number of repetitions of the repetitive cycle control.

[0020] The fifth disclosure is a computer system in which, in the above-mentioned computer system, the repeat control unit performs a redo control that restarts from either acquiring new first summary text information or acquiring new second detail text information.

[0021] According to the fifth disclosure, the computer system can restart from either acquiring new first summary text information or acquiring new second detail text information.

[0022] The sixth disclosure is a computer system in which the repeat control unit performs the repeat control based on given repeat instruction information.

[0023] According to the sixth disclosure, the computer system will be able to perform retry control based on retry instruction information.

[0024] The seventh disclosure is a computer system comprising the above-mentioned computer system, further comprising a generation instruction information receiving unit that receives given generation instruction information for generating new detailed text information, the summary generation unit having a generation unit capable of generating the summary, the summary text information acquisition control unit generating summary instruction information to be given to the generation unit of the summary generation unit based on the generation instruction information, the extended generation unit having a generation unit capable of performing the extended generation to generate the detailed text information, and the detailed text information acquisition control unit generating extended instruction information to be given to the generation unit of the extended generation unit based on the generation instruction information.

[0025] Furthermore, the eighth disclosure is a computer system in which the summary instruction information includes information that indicates the summarization amount.

[0026] According to the seventh or eighth disclosure, the computer system will be able to generate extension instruction information based on the generation instruction information. For example, by generating extension instruction information that instructs the system to expand to the amount of text required by the generation instruction information, it will be possible to obtain second detailed text information with a text amount similar to that of the first detailed text information.

[0027] The ninth disclosure is a computer system further comprising: a generation instruction information receiving unit that receives given generation instruction information for generating new detailed text information; a summary generation unit having a generation unit capable of generating the summary; a summary text information acquisition control unit that generates summary instruction information to be given to the generation unit of the summary generation unit based on the generation instruction information; an extended generation unit having a generation unit capable of performing the extended generation information; a detailed text information acquisition control unit that generates extended instruction information to be given to the generation unit of the extended generation unit based on the generation instruction information; and a cycle repeat count determination unit that determines the number of repetitions by the cycle repeat control based on the generation instruction information.

[0028] According to the ninth disclosure, the computer system will be able to determine the number of repetitions by loop repeat control based on the generation instruction information.

[0029] The tenth disclosure is a computer system further comprising: a generation instruction information receiving unit that receives given generation instruction information for generating new detailed text information; a summary generation unit having a generation unit capable of generating the summary; a summary text information acquisition control unit that generates summary instruction information to be given to the generation unit of the summary generation unit based on the generation instruction information; an extended generation unit having a generation unit capable of performing the extended generation to generate the detailed text information; a detailed text information acquisition control unit that generates extended instruction information to be given to the generation unit of the extended generation unit based on the generation instruction information; and a feedback control unit that variably controls the summary instruction information in the loop repeat control based on the evaluation result of the detailed text information evaluation unit.

[0030] According to the tenth disclosure, the computer system can modify the summary instruction information in the loop repeat control based on the evaluation results of the detailed text information evaluation unit.

[0031] The eleventh disclosure is a computer system further comprising a retained portion designation unit that designates a specific portion of the first detailed text information to be retained in the summary, wherein the summary text information acquisition control unit performs control to acquire the first summary text information while retaining the specific portion designated by the retained portion designation unit.

[0032] According to the 11th disclosure, the computer system can retain a specific portion of the first detailed text information in the summary text information, and as a result, it can generate a second detailed text information that includes the said specific portion.

[0033] The twelfth disclosure is a computer system comprising, in the above-mentioned computer system, a retained portion designation unit that designates a specific portion of the second detailed text information to be retained in the summary, wherein the summary text information acquisition control unit performs control in the loop repeat control to acquire the new first summary text information while retaining the specific portion designated by the retained portion designation unit.

[0034] According to the twelfth disclosure, the computer system can retain a specific portion of the first detailed text information in the summary text information during loop repeat control, and as a result, it can generate a second detailed text information that includes the said specific portion.

[0035] The thirteenth disclosure is a computer system further comprising a highly expanded portion designation unit that designates a specific portion of the first summary text information to have a relatively high degree of expansion, wherein the detailed text information acquisition control unit performs the expansion with a relatively high degree of expansion on the specific portion designated by the highly expanded portion designation unit to acquire the second detailed text information.

[0036] According to the 13th disclosure, the computer system can perform a relatively high degree of expansion on a specific portion of the first summary text information and convert it into second detailed text information.

[0037] The fourteenth disclosure is a computer system in which the summary text information acquisition control unit has a summary generation condition setting unit that sets summary generation conditions, and acquires the first summary text information by causing the summary generation unit to generate a summary based on the summary generation conditions.

[0038] According to the 14th disclosure, a computer system can set summary generation conditions and generate a summary based on these conditions.

[0039] The fifteenth disclosure is a computer system in which the summary generation condition setting unit sets the summary generation conditions based on the number of repetitions of the loop repeat control.

[0040] According to the 15th disclosure, the computer system will be able to set summary generation conditions based on the number of repetitions of the loop repeat control.

[0041] Also, the 16th disclosure is a computer system in which, in the above computer system, the summary generation condition setting unit sets the summary generation conditions based on the first detailed text information.

[0042] According to the 16th disclosure, the computer system can set summary generation conditions based on the first detailed text information.

[0043] The 17th disclosure is a computer system in which, in the above computer system, the summary text information acquisition control unit acquires the first summary text information as text information in list form.

[0044] According to the 17th disclosure, the computer system can acquire the first summary text information as text information in list form.

[0045] The 18th disclosure is a computer system in which, in the above computer system, the expansion generation unit performs the expansion by adding new list-form information to the list-form information of the first summary text information.

[0046] According to the 18th disclosure, the computer system can acquire second detailed text information that adds to the first summary text information.

[0047] Also, the 19th disclosure is a computer system in which, in the above computer system, the detailed text information acquisition control unit has an expansion generation condition setting unit that sets expansion generation conditions, and causes the expansion generation unit to perform the expansion based on the expansion generation conditions to acquire the second detailed text information.

[0048] According to the 19th disclosure, the computer system can set expansion generation conditions and acquire second detailed text information based on these conditions.

[0049] The 20th disclosure is a computer system in which, in the above computer system, the expansion generation condition setting unit sets the expansion generation conditions based on the first detailed text information.

[0050] According to the 20th disclosure, the computer system will be able to set extended generation conditions based on the first detailed text information.

[0051] The 21st disclosure is a computer system further comprising a detailed repeat control unit that performs detailed repeat control to cause the detailed text information acquisition control unit to repeatedly acquire the second detailed text information.

[0052] According to the 21st disclosure, the computer system will be able to repeatedly retrieve the second detailed text information.

[0053] The 22nd disclosure is a computer system further comprising a detailed text information evaluation unit that evaluates the second detailed text information, wherein the repeat control unit executes the detailed repeat control based on the evaluation result of the detailed text information evaluation unit.

[0054] According to the 22nd disclosure, the computer system will be able to evaluate the second detailed text information and perform detailed repeat control based on the evaluation results.

[0055] The 23rd disclosure is a computer system in which the above-mentioned computer system further comprises an extended randomness control unit that generates the detailed text information using a given extended randomness control parameter value for controlling the randomness related to the extension, and controls the extended randomness control parameter value variably based on the number of repetitions of the detailed repeat control and / or the evaluation result of the detailed text information evaluation unit.

[0056] According to Disclosure No. 23, the computer system will be able to control the randomness related to the expansion and the degree of creativity related to the second detailed text information.

[0057] The 24th disclosure is a computer system further comprising a summary repeat control unit that performs summary repeat control to cause the summary text information acquisition control unit to repeatedly acquire the first summary text information.

[0058] According to Disclosure No. 24, the computer system will be able to repeatedly retrieve the first summary text information.

[0059] The 25th disclosure is a computer system further comprising a summary text information evaluation unit that evaluates the first summary text information, wherein the repeat control unit executes the summary repeat control based on the evaluation result of the summary text information evaluation unit.

[0060] According to the 25th disclosure, the computer system will be able to perform summary repeat control based on the evaluation results of the summary text information evaluation unit.

[0061] The 26th disclosure is a computer system in which the summary generation unit generates a summary using a given summary randomness control parameter value for controlling the randomness related to the generation of the summary, and further comprises a summary randomness control unit that variably controls the summary randomness control parameter value based on the number of repetitions of the summary repeat control and / or the evaluation result of the summary text information evaluation unit.

[0062] According to Disclosure No. 26, the computer system can change the creativity and diversity in the summarization process based on the number of repetitions of the summary repeat control and / or the evaluation results of the summary text information.

[0063] The 27th disclosure is a computer system in which the detailed text information acquisition control unit acquires the second detailed text information for the first summary text information selected from a plurality of first summary text information acquired by the summary repeat control based on given conditions.

[0064] According to Disclosure No. 27, a computer system becomes capable of obtaining second detailed text information for a first summary text information selected based on given conditions.

[0065] The 28th disclosure is a computer system in which, in the above-mentioned computer system, the detailed text information acquisition control unit is instructed to acquire a plurality of second detailed texts from the first summary text information, and the repeat control unit, in the loop repeat control, instructs the summary text information acquisition control unit to acquire a new first summary text information, which is a new first detailed text information, selected from the plurality of second detailed texts based on given conditions.

[0066] According to the disclosure in paragraph 28, the computer system can obtain multiple second detailed texts from a first summary text, select one of the second detailed texts from among them as the new first detailed text, and obtain a new first summary text.

[0067] The 29th disclosure is a computer system in which the first detailed text information is information from a gameplay log of a given game, and the second detailed text information is a computer system capable of acquiring new gameplay log information.

[0068] According to Disclosure No. 29, the computer system can generate a new gameplay log based on the original gameplay log by summarizing the original gameplay log and expanding from the summary.

[0069] The 30th disclosure is a text generation control method that includes: controlling the acquisition of a first summary text information for a first detailed text information using a summary generation unit capable of generating a summary, abstract, outline, overview, or digest (hereinafter collectively referred to as "summary") for a given detailed text information; and controlling the acquisition of a second detailed text information for the first summary text information using an extension generation unit capable of generating detailed text information by extending the given summary text information.

[0070] According to the 30th invention, it becomes possible to realize a text generation control method that produces the same effects as the computer system described above.

[0071] A system configuration diagram showing an example of the editing system's configuration. A diagram showing an example of the user interface. A diagram explaining the overview of automatic editing. A diagram explaining the flow of automatic editing, the data used, and the AI ​​used. A diagram explaining the flow of automatic editing, the data used, and the AI ​​used. A diagram explaining the machine learning of the detailed text generation AI. A diagram showing an example of origin generation instruction information written in natural language. A diagram explaining the machine learning of the summary generation AI. A diagram showing an example of summary instruction information written in natural language. A diagram showing an example of the structure of summary text information. A diagram explaining the machine learning of the summary evaluation AI. A diagram showing an example of summary evaluation instruction information written in natural language. A diagram explaining the machine learning of the extended generation AI. A diagram showing an example of extended instruction information written in natural language. A diagram showing an example of the structure of the second detailed text information. A diagram explaining the machine learning of the extended evaluation AI. A diagram showing an example of extended evaluation instruction information written in natural language. A diagram explaining the machine learning of the text classification AI. A diagram showing an example of classification instruction information written in natural language. A diagram explaining the machine learning of the detailed text evaluation AI. A diagram showing an example of detailed text evaluation instruction information described in natural language. A diagram showing an example of programs and data stored by the server system. A diagram showing an example of the functional configuration of the server processing unit. A flowchart explaining the flow of the editing process. A flowchart continuing from Figure 24. A flowchart continuing from Figure 25. A diagram explaining a modified version of the editing system. A flowchart explaining the flow in a modified version of the editing process (editing process β). A flowchart continuing from Figure 28. A flowchart continuing from Figure 29. A diagram showing an example of extended instruction information in a modified version of the editing process. A diagram showing an example of detailed text evaluation instruction information in a modified version of the editing process. A diagram showing a modified version of the machine learning of the summary generation AI. A diagram showing a modified version of the machine learning of the extended generation AI. A diagram showing a modified version of the machine learning of the detailed text evaluation AI.

[0072] Examples of embodiments of the present invention will be described below, but it goes without saying that the embodiments to which the present invention can be applied are not limited to the following embodiments.

[0073] Figure 1 is a system configuration diagram showing an example of the configuration of the editing system according to this embodiment. The editing system 1000 is a computer system that generates and edits text data and provides it to user 2. In this embodiment, a scenario is used as an example of text for explanation.

[0074] The proofreading system 1000 is a computer system that includes a server system 1100 and user terminals 1500 for each user, all of which are connected via a network 9 to enable data communication.

[0075] Network 9 refers to a communication path capable of data communication. In other words, Network 9 includes not only LANs (Local Area Networks) using dedicated lines (dedicated cables) for direct connection or Ethernet (registered trademark), but also telephone communication networks, cable networks, and the Internet.

[0076] The server system 1100 is a computer system that performs various processes such as managing and controlling registered user information and controlling editing.

[0077] The server system 1100 has a control board 1150 mounted on the main unit 1101. The control board 1150 is equipped with various microprocessors such as a CPU (Central Processing Unit) 1151, a GPU (Graphics Processing Unit), and a DSP (Digital Signal Processor), various IC memories 1152 such as VRAM, RAM, and ROM, and a communication device 1153. Some or all of the functions mounted on the control board 1150 may be implemented by an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a SoC (System on a Chip).

[0078] In Figure 1, the server system 1100 is depicted as a single server device, but it may also be configured with multiple devices. For example, the server system 1100 may be configured with multiple server devices, each responsible for a specific function, connected to each other via an internal bus or network 9 for data communication. Furthermore, the server system 1100 may include a database and online storage.

[0079] The server system 1100 has a generation unit called a generation AI 10. The AI ​​model of the generation AI 10 is a Large Language Model (LLM) realized by deep learning based on hardware employing a multi-core architecture (for example, a group of GPUs and memory, a group of AI chips, etc.).

[0080] The generation AI 10 generates text data for the source scenario based on the given generation instruction information. The generation instruction information is, for example, information called a prompt. The source scenario is referred to as the "first detailed document".

[0081] The generation AI 10 may be implemented with a single AI model, or it may be configured using multiple sub-AI models. In this embodiment, the generation unit, generation AI 10, is described as having the following sub-AI models, which are sub-generation units: detailed text generation AI 11, summary generation AI 12, summary evaluation AI 13, extended generation AI 14, extended evaluation AI 15, text classification AI 16, and detailed text evaluation AI 17. All of these sub-AI models are constructed as LLMs. The construction of the AI ​​model will be described later.

[0082] The detailed text generation AI 11 generates text data of the scenario, which is the first detailed text information, according to the given generation instruction information. The text format of the scenario and the elements included in the story follow the generation conditions specified in the generation instruction information.

[0083] The summary generation AI 12 generates text data that summarizes the given first detailed text information, which is the scenario, according to the given summary instruction information. The data containing the generated text is called "summary text information".

[0084] The summary evaluation AI 13 evaluates whether the given summary text information was generated as a summary of the first detailed text information in accordance with the summary evaluation instruction information, and determines whether the summary is acceptable or not.

[0085] The extended generation AI 14 generates new scenario text data by treating the given summary text information as a scenario plot, according to the given extended instruction information. In other words, it extends the summary text information to generate new detailed text that inherits the core of the first detailed text information. The data containing the generated text is called the "second detailed text information." The second detailed text information can be described as the result of adding to the summary text information.

[0086] The extended evaluation AI 15 evaluates the extension from the given summary text information to the second detailed text information according to the given extended evaluation instruction information, and determines whether it passes or fails.

[0087] The text classification AI 16 classifies the text according to the given classification instruction information. The classification items are set in correspondence with the prompt items (items that can be specified by prompt) of the generation instruction information given to the generation AI 10. Other classification items and classification options within classification items follow the given classification instruction information.

[0088] The detailed text evaluation AI 17 evaluates the "quality" and "quality of content" of the given detailed text information according to the given detailed text evaluation instruction information. Specifically, it evaluates the second detailed text information.

[0089] Furthermore, the generated AI 10 is not limited to the server system 1100. For example, the server system 1100 may be implemented as an external device that can be connected via the network 9, and may be an external system that can be used by customers other than those using the editing system 1000.

[0090] The user terminal 1500 serves as the Man-Machine Interface (MMIF) for the correction system 1000. Although only one terminal is depicted in Figure 1, in actual operation, it is typical for multiple user terminals 1500 to communicate with the server system 1100 simultaneously.

[0091] The user terminal 1500 is a computer system that can connect to the network 9, such as a personal computer, smartphone, wearable computer, portable game console, home game console, or tablet computer.

[0092] The user terminal 1500 is a computer comprising an operation input device, an image display device, a communication device, and a control board 1550 for performing calculations. Examples of the operation input device include a touch panel 1506, a keyboard, a game controller, and a mouse. Examples of the image display device include a touch panel 1506, a head-mounted display, and a glasses-type display.

[0093] The control board 1550 is equipped with a CPU 1551, various microprocessors such as a GPU and DSP, various IC memories 1552 such as VRAM, RAM, and ROM, and a communication module 1553 connected to the network 9. These elements mounted on the control board 1550 are electrically connected via bus circuits and the like, enabling data reading and writing, and signal transmission and reception. Part or all of the control board 1550 may be an ASIC, FPGA, or SoC.

[0094] The control board 1550 stores programs and various data necessary to realize the functions of the user terminal 1500 in the IC memory 1552. The user terminal 1500 realizes its functions as a terminal by executing a predetermined application program (for example, a client program) on the CPU 1551.

[0095] User 2 performs a predetermined request operation on the user terminal 1500 to cause the server system 1100 to generate a base scenario (first detailed text information) and obtains the text data of the desired scenario.

[0096] Figure 2 shows an example of a user interface (UI) related to request operations. When the server system 1100 detects that a predetermined request function has been invoked on the user terminal 1500, it displays the request setting UI 5 (5a, 5b, ...) for each setting item on the user terminal 1500.

[0097] The settings items mentioned here are not limited to the example shown in Figure 2. Also, while Request Settings UI5 is illustrated as a pull-down menu, the UI design is not limited to this. For example, Request Settings UI5 could be a text input field that allows natural language input for each request item.

[0098] Request setting UI5 is provided for each prompt item that describes the generation conditions (prerequisites), quality, supplementary information, etc., in the generation instruction information given to the generation AI 10. For example, request setting UI5 (5a, 5b, ...) is provided separately for "genre," "plot keywords," "character settings," "dialogue," "plot twists," "story length," "user-specified plot," etc. Multiple options are presented for selection in a single request setting UI5. Depending on the item, options may also be included that the user inputs in their own words (natural language).

[0099] The user terminal 1500 transmits the input result in the request setting UI 5 to the server system 1100 along with a predetermined request signal. The server system 1100 generates origin generation instruction information based on the received input result in the request setting UI 5 and provides this to the detailed text generation AI 11. The generation AI 11 then generates and outputs the original first detailed text information. Next, it automatically edits this information.

[0100] Figure 3 is a diagram illustrating the overview of automated text correction. In Figure 3, text is shown as a horizontal strip, and its length represents the amount of text.

[0101] The automated proofreading process consists of the following four stages: (1) The first stage presents the detailed text to be summarized to user 2 and accepts instructions from user 2 to select specific parts (specific part a, specific part b, ...) that they wish to keep from the detailed text. (2) The second stage generates a summarized text from the detailed text to be summarized, retaining the specific parts. (3) The third stage generates a new detailed text from the generated summarized text by creating and expanding the content of the summary while inheriting the retained specific parts. (4) The fourth stage evaluates the newly generated detailed text and determines whether it is acceptable as an output text that meets the request.

[0102] The resulting structure of the new detailed text after the third process depends on how the generating AI 10 generates the text during that process. In other words, whether the new detailed text has a four-part structure (introduction, development, turn, conclusion) or a five-part structure (introduction, development, turn, turn, conclusion), it will be an extension of the summary text.

[0103] If the detailed document evaluated in the fourth process passes, the server system 1100 determines that the editing is complete and outputs it as the output document in response to the request. If the detailed document evaluated in the fourth process fails, the server system 1100 sets that detailed document as the new detailed document to be summarized and repeats the first to fourth processes.

[0104] In this embodiment, the text presented in the first process, i.e., the first detailed text to be summarized, is the original first detailed text initially generated by the detailed text generation AI 11 in response to the request. The text generated by the extended generation AI 14 in the third process is the second detailed text.

[0105] In Figure 3, each time the second detailed text is presented to user 2 as the detailed text to be summarized next, the designation of specific parts gradually increases to "specific part a," "specific part b," "specific part c," and "specific part d." Note that although the bands representing the summarized text in Figure 3 have additional bands for specific parts (specific part a, specific part b, ...) attached to their ends, this is a conceptual diagram, and the actual structure of the summarized text is not limited to what is shown in Figure 3.

[0106] Even if the initially generated original first detail text contains inconsistencies in context, abrupt plot developments, or unnecessary and uninteresting sections, summarization eliminates one or more of these flaws, compressing the text to its core essence. Then, by developing and expanding the story based on this summary to generate the second detail text, the original first detail text is revised into a better piece of writing.

[0107] Even if there are flaws that weren't removed during the summarization process, the expansion will be based on those remaining flaws. Therefore, parts that didn't make sense before summarization may become consistent, and parts that seemed like absurd developments are likely to be corrected to a more appropriate development. Unnecessary and tedious developments are likely to be cut out.

[0108] Any minor flaws that may remain even after summarization and expansion will be corrected as the process repeats from the first to the fourth stage, ultimately resulting in a well-written document. The number of iterations required to produce the second detailed document depends on the content of the request and the performance of the AI ​​model of Generator AI 10.

[0109] Figures 4 and 5 are diagrams that provide a more detailed explanation of the flow of automatic editing, the data used, and the AI ​​used. As shown in Figure 4, when the server system 1100 receives a request signal and the content of the request from the user terminal 1500 [Communication A], it generates origin generation instruction information 600 to be given to the detailed text generation AI 11.

[0110] Figure 6 is a diagram illustrating the machine learning of the detailed text generation AI 11. The detailed text generation AI 11 is pre-trained to take learning origin generation instruction information and annotations as input and output detailed text information, and it supports generation instructions in natural language.

[0111] Figure 7 shows an example of the origin generation instruction information 600 described in natural language. The origin generation instruction information 600 is information that has been adapted to a format that the detailed text generation AI 11 can accept, based on the content of the request from user 2. The content of the request itself can also be said to correspond to the generation instruction information to the generation AI 10. Therefore, the process by which the server system 1100 receives the content of the request from the user terminal 1500 and generates the origin generation instruction information 600 can be summarized as the process of receiving the generation instruction information to the generation AI 10.

[0112] Returning to Figure 4, the server system 1100 provides the origin generation instruction information 600 to the detailed text generation AI 11, causing it to generate and output the original first detailed text information 602.

[0113] Next, the server system 1100 prompts the user terminal 1500 to present the original first detailed document [Communication B]. During this presentation, the server system accepts the user 2's input indicating whether the original first detailed document is acceptable or unacceptable as the requested document, as well as a specific part specification operation.

[0114] If the original first detail document is satisfactory, that first detail document will be output as the output document, and the response to the request will end.

[0115] If the original first detailed document fails, the user terminal 1500 sends information about the pass / fail judgment input and the operation of specifying a particular part to the server system 1100 [Communication C]. If the detailed document to be summarized (the original first detailed document) fails, the server system 1100 generates summary instruction information 604 to be given to the summary generation AI 12.

[0116] Figure 8 is a diagram illustrating the machine learning process of the summary generation AI 12. The summary generation AI 12 is pre-trained to take detailed text information to be summarized, specific part specification information, and annotations as input and output first summary text information, and it supports generation instructions in natural language.

[0117] Figure 9 shows an example of a summary instruction information 604 described in natural language. The summary instruction information 604 includes the number of sentences after summarization, specific partial information, and a summary randomness parameter value as summary generation conditions (prerequisites for summarization). The summary randomness parameter value is a parameter value that provides diversity in which parts of the detailed text of the summary sentence are summarized. This corresponds to the parameter value commonly referred to as "Temperature." The number of sentences after summarization and the summary randomness parameter value are set according to the number of times the summary is repeated (details will be described later).

[0118] Figure 10 shows an example of the structure of summary text information 606. Summary text information 606 is text data, and is information in the form of a list of summary sentences. Blank spaces and blank lines are set between each summary sentence in the list. For summary sentences about specific parts, the text of the specific part is written after the summary sentence, and blank spaces and blank lines are set. The method of writing the specific part is not limited to this. In addition, summary text information 606 includes the generation date and time as metadata.

[0119] Moving to Figure 5, the server system 1100 provides the summary instruction information 604 to the summary generation AI 12 to generate the summary text information 606. Then, it generates summary evaluation instruction information 608 to cause the summary evaluation AI 13 to evaluate the summary text information 606, and provides this to the summary evaluation AI 13 to perform the evaluation.

[0120] Figure 11 is a diagram illustrating the machine learning of the summary evaluation AI 13. The summary evaluation AI 13 is pre-trained to take detailed text information of the text to be summarized, summary generation instruction information, summary text information, and annotations as input, and to output evaluation results and pass / fail judgment results for the summary text information, and it supports generation instructions in natural language.

[0121] Depending on the performance of the AI ​​model of the summary generation AI 12, the summarized text information 606 may not necessarily be exactly as specified in the summary instruction information 604. The summary evaluation AI 13 determines whether the result is a pass or fail based on the summary instruction information 604.

[0122] Figure 12 shows an example of the summary evaluation instruction information 608 described in natural language.

[0123] Returning to Figure 5, if the summary text indicated by the summary text information 606 is unacceptable, the server system 1100 again provides the summary instruction information 604 to the summary generation AI 12 to regenerate the summary text information 606. This is called "summary repeat".

[0124] In addition, the summary instruction information 604 given to the summary generation AI 12 during summary repetition is modified so that the randomness of the summary randomness parameter value increases as the number of summary repetitions increases.

[0125] If the summary text indicated by the summary text information 606 is acceptable, the server system 1100 displays the summary text as text on the user terminal 1500 [Communication D] and accepts instructions for the highly extended portion of the summary text.

[0126] The instruction operation for the highly extended portion is an operation in which user 2 specifies the part of the newly generated and extended text that they would like to be further extended or added to, and is an operation to select a range from the displayed text. The result of the instruction operation for the highly extended portion is transmitted from the user terminal 1500 to the server system 1100 [communication E]. When the server system 1100 receives the result of the instruction operation for the highly extended portion, it generates extension instruction information 612 to be given to the extension generation AI 14.

[0127] Figure 13 is a diagram illustrating the machine learning of the augmented generative AI 14. The augmented generative AI 14 is pre-trained to take origin generation instruction information, summary text information, and annotations as input, and to output detailed text information (second detailed text) created by expanding on the summary, and it supports generation instructions in natural language.

[0128] Figure 14 shows an example of the extended instruction information 612 described in natural language. The extended instruction information 612 includes an extended generation condition (prerequisite for extension) that sets the degree of extension to the highly extended portion relatively high (underlined portion in Figure 14), and an extended randomness parameter value.

[0129] The extended randomness parameter value is a parameter value that provides diversity to the generated detail text (second detail text). This is the parameter value commonly referred to as "Temperature." Note that the extended randomness parameter value is set according to the number of times the extension is repeated (details below).

[0130] Figure 15 shows an example of the structure of the second detailed text information 614. The second detailed text information 614 is text data and includes an extended text that describes a scene based on the summary text, and metadata (including, for example, information on the date and time of creation). The extended text based on the summary text of a specific part includes the text of that specific part. In this case, the words used at the beginning and end of the specific part may be changed from the original specific part so that the flow of the text is smoothly connected.

[0131] The writing style of the detailed text in the second detailed text information 614 is basically the same as that of the detailed text in the first detailed text information 602. When compared side by side with the writing style of the summary text in the summary text information 606, it can be said that the second detailed text information 614 is an extension of the summary text in the summary text information 606, with additions made between the list-like summary texts.

[0132] Returning to Figure 5, the server system 1100 provides the extended instruction information 612 to the extended generation AI 14 to generate the second detailed text information 614. Then, it generates extended evaluation instruction information 616 to cause the extended evaluation AI 15 to evaluate the second detailed text information 614, and provides this to the extended evaluation AI 15 to perform the evaluation.

[0133] Figure 16 is a diagram illustrating the machine learning of the augmentation evaluation AI 15. The augmentation evaluation AI 15 is pre-trained to take the following inputs—summary text information to be used as the basis for the augmentation, augmentation generation instruction information, detailed text information to be evaluated (second detailed text information), and annotations—as input, and to output an evaluation of the augmentation content and a pass / fail judgment for the augmentation, and it supports generation instructions in natural language.

[0134] Depending on the quality of the AI ​​model of the extended generation AI 14, the second detailed text information 614 may not necessarily be exactly as specified in the extended instruction information 612. The extended evaluation AI 15 determines whether the second detailed text information 614 is a pass or fail as a result of responding to the extended instruction information 612.

[0135] Figure 17 shows an example of extended evaluation instruction information 616 described in natural language. Extended evaluation instruction information 616 defines one or more evaluation items and evaluation criteria in the extended evaluation conditions (prerequisites for evaluation). For example, evaluation items may include "no overlap in content in the extended scenes," "match in content between the summary and the extended scenes," and "inheritance of specific parts." Of course, other evaluation items may also be set as appropriate.

[0136] Returning to Figure 5, if the extended evaluation result is unsuccessful, the server system 1100 again provides the extended instruction information 612 to the extended generation AI 14 to regenerate the second detailed text information 614. This is referred to as "detailed repeat".

[0137] In addition, the extended instruction information 612 given to the extended generation AI 14 during detailed repetition is modified by a given value to change the extended randomness parameter value so that the randomness improves as the number of detailed repetitions increases.

[0138] If the extended evaluation result is satisfactory, the server system 1100 has successfully obtained the second detailed document information 614, which has been appropriately extended from the summary document information 606. Conversely, regarding the summary repetition, the summary repetition is repeated until a satisfactory summary document information 606 is generated, and each time, unsatisfactory summary document information 606 is accumulated. Therefore, obtaining the second detailed document information 614, for which the extended evaluation result is satisfactory, corresponds to obtaining the second detailed document information 614 for the summary document information 606 selected based on given conditions.

[0139] If the extended evaluation result is satisfactory, the server system 1100 generates classification instruction information 620 to be given to the text classification AI 16 and detailed text evaluation instruction information 622 to be given to the detailed text evaluation AI 17.

[0140] Figure 18 is a diagram illustrating the machine learning process of the text classification AI 16. The text classification AI 16 is pre-trained to take detailed text information and annotations prepared for training as input, perform text classification, and output evaluation scores for each classification, and it supports generation instructions in natural language.

[0141] The text classification is determined by analyzing the settings in Request Setting UI 5 (see Figure 2) and identifying which of the options in those settings corresponds to the text. As mentioned above, the settings in Request Setting UI 5 form the basis of the prompt items included in the generation conditions of Origin Generation Instruction Information 600. The evaluation points for each classification are the match rate and similarity rate between the specified prompt items included in the generation conditions of Origin Generation Instruction Information 600 and the classification result of the analysis item corresponding to those prompt items.

[0142] Figure 19 shows an example of classification instruction information 620 described in natural language. The classification instruction information 620 includes classification conditions, and the description of the classification conditions defines the analysis item and the classification type within that analysis item.

[0143] Figure 20 is a diagram illustrating the machine learning process of the detailed text evaluation AI 17. The detailed text evaluation AI 17 is pre-trained to take input from origin generation instruction information, detailed text information to be evaluated, detailed classification information of the detailed text, and annotations, and to output evaluation scores for each evaluation item and whether the detailed text is pass or fail, and it supports generation instructions in natural language. The pass or fail status of the detailed text is the pass or fail status of the output text that is output in response to a request.

[0144] Figure 21 shows an example of detailed text evaluation instruction information 622 written in natural language. Detailed text evaluation instruction information 622 includes evaluation conditions (prerequisites for evaluation). The evaluation conditions describe the criteria for evaluation points for each evaluation item.

[0145] Returning to Figure 5, the server system 1100 outputs the second detailed document as a response to the request if the total evaluation score of the second detailed document reaches a given passing score [Communication H]. If the passing score is not reached, the second detailed document is displayed as text on the user terminal 1500, and additional operations on specific parts are accepted [Communication F].

[0146] The operation to add a specific section is an operation to specify a part of the second detailed document that is otherwise unacceptable but which user 2 deems partially acceptable and wishes to keep. The result of the operation to add a specific section is transmitted from the user terminal 1500 to the server system 1100 [communication G].

[0147] When the server system 1100 receives the result of an additional operation on a specific part, it considers the second detailed document that failed to pass the review as a target for summarization and repeats the process of summarization → evaluation of the summary → expansion → evaluation of the expansion → document classification and evaluation of the detailed document. This is called a "cycle repeat".

[0148] Specifically, in controlling the loop repeat, the server system 1100 uses the summary instruction information 604 previously given to the summary generation AI 12 to designate the second detailed text that failed the previous cycle as the detailed text to be summarized. It then generates summary instruction information 604 instructing the AI ​​12 to include a specific part in the summary information and provides it to the summary generation AI 12.

[0149] Furthermore, in controlling the loop repeat, the server system 1100 generates summary instruction information 604 for loop repeat based on the extension instruction information 612 previously given to the extension generation AI 14. The summary instruction information 604 for loop repeat includes information specifying the highly extended portion and extension generation conditions that specify strengthening the extension to the highly extended portion (for example, doubling the number of scenes generated by expanding and extending from the summary). The server system 1100 then provides the extension instruction information 612 for loop repeat to the extension generation AI 14.

[0150] Each time the loop is repeated, the detailed text of the first detailed text information 602 is revised to meet the request, resulting in a better text than the original that was initially generated, and then provided to user 2.

[0151] Figure 22 shows an example of programs and data stored by the server system 1100. The server system 1100 stores the server program 501 and the distribution client program 503, which is the original client program to be provided to the user terminal 1500, in the IC memory 1152. The server program 501 may include one or more generation AI programs 502 to realize the function of generation AI 10. Alternatively, the generation AI programs 502 may be stored separately from the server program 501.

[0152] Furthermore, the server system 1100 stores UI definition data 510 of the request setting UI 5, a trained generation AI model 512, request information 514 indicating the result of the request setting operation, and specific part specification information 516.

[0153] Furthermore, the server system 1100 stores origin generation instruction information 600, first detailed text information 602, summary instruction information 604, summary text information 606, summary evaluation instruction information 608, high-extension part designation information 610, and extension instruction information 612. The summary text information 606 is stored if it passes the summary evaluation and is discarded when the editing is completed.

[0154] Furthermore, the server system 1100 stores the second detailed document information 614, the extended evaluation instruction information 616, the classification instruction information 620, and the detailed document evaluation instruction information 622. The second detailed document information 614 is stored if it passes the extended evaluation and is discarded when the editing is completed.

[0155] Furthermore, the server system 1100 stores the summary repeat count 630, the detailed repeat count 632, the loop repeat count 634, and the upper limit repetition count 636, which corresponds to the upper limit of the loop repeat count. Of course, other data, such as evaluation result data, may also be stored as appropriate.

[0156] The server system 1100 performs the server program 501 and calculates the results on the CPU 1151, thereby realizing the function of the server processing unit 200s shown in Figure 23.

[0157] The server processing unit 200s performs various controls related to the automatic editing of text, and in this embodiment, it performs various controls from generating a scenario using AI in response to a request operation from user 2 until it is output.

[0158] Specifically, the server processing unit 200s includes a generation instruction information receiving unit 204, a detailed text information acquisition control unit 210, a detailed text generation unit 220, a remaining portion designation unit 222, a summary text information acquisition control unit 224, and a summary generation unit 226.

[0159] Furthermore, the server processing unit 200s includes a summary randomness control unit 228, a summary text information evaluation unit 230, an extended generation unit 232, a highly extended portion designation unit 234, a text classification unit 236, and a detailed text information evaluation unit 238.

[0160] Furthermore, the server processing unit 200s includes a repeat control unit 240, a cycle repeat count determination unit 242, and a feedback control unit 244.

[0161] The generation instruction information receiving unit 204 receives given generation instruction information for generating new detailed text information. This includes receiving request information from the user terminal 1500 and generating the origin generation instruction information 600.

[0162] The detailed text information acquisition control unit 210 uses an expansion generation unit 232 capable of generating detailed text information by expanding on a given summary text information 606 to perform control to acquire detailed text information (second detailed text information 614) about the summary text information 606.

[0163] Furthermore, the detailed text information acquisition control unit 210 includes a summary instruction information generation unit 211 that generates summary instruction information 604, and an extended instruction information generation unit 214 that generates extended instruction information 612.

[0164] The summary instruction information generation unit 211 includes a summary generation condition setting unit 212 that sets summary generation conditions based on the first detailed text information 602.

[0165] The extended instruction information generation unit 214 includes an extended randomness control unit 216 and an extended generation condition setting unit 218 that sets extended generation conditions based on the first detailed text information 602.

[0166] The extended randomness control unit 216 controls the extended randomness control parameter values ​​variably based on the number of repetitions of the detailed repeat control and / or the evaluation result of the detailed text information evaluation unit.

[0167] The detailed text generation unit 220 generates detailed text information based on the origin generation instruction information 600. The detailed text generation unit 220 has a detailed text generation AI 11.

[0168] The retained portion designation unit 222 designates specific portions of the first detailed text information 602 to be retained in the summary. This corresponds to receiving the operation to designate specific portions on the user terminal 1500 and obtaining the result of the reception (see Figure 4).

[0169] The summary text information acquisition control unit 224 uses a summary generation unit 226 capable of generating a summary of given detailed text information to perform control for acquiring summary text information 606 for the first detailed text information 602. At that time, the summary text information acquisition control unit 224 acquires the summary text information 606 while retaining a specific portion designated by the remaining portion designation unit 222.

[0170] The summary generation unit 226 is capable of generating summary text information 606 for given detailed text information. The summary generation unit 226 has a summary generation AI 12.

[0171] The summary randomness control unit 228 controls the summary randomness control parameter values ​​variably based on the number of repetitions of the summary repeat control and / or the evaluation result of the summary text information evaluation unit 230.

[0172] The summary text information evaluation unit 230 evaluates the summary text information 606.

[0173] The extended generation unit 232 extends the given summary text information 606 to generate second detailed text information 614. The extended generation unit 232 has an extended generation AI 14.

[0174] The highly expanded portion designation unit 234 designates a specific portion of the summary text information 606 that has a relatively high degree of expansion. As a result of the highly expanded portion designation operation, the summary text is displayed on the user terminal 1500, so the highly expanded portion designation unit 234 can be said to also function as a summary text notification unit 235 that notifies the user 2 of the summary text.

[0175] The document classification unit 236 classifies the detailed documents to be classified. The document classification unit 236 has a document classification AI 16.

[0176] The detailed text information evaluation unit 238 evaluates the second detailed text information 614. The detailed text information evaluation unit 238 has a detailed text evaluation AI 17.

[0177] The repeat control unit 240 performs a redo control that restarts the process by either acquiring new first summary text information or acquiring new second detailed text information.

[0178] Specifically, the repeat control unit 240 performs summary repeat control, which causes the summary text information acquisition control unit 224 to repeatedly acquire the summary text information 606 based on the evaluation result of the summary text information evaluation unit 230.

[0179] Furthermore, the repeat control unit 240 performs detailed repeat control, causing the detailed text information acquisition control unit 210 to repeatedly acquire the second detailed text information 614 based on the evaluation results of the detailed text information evaluation unit 238.

[0180] Furthermore, the repeat control unit 240, based on the evaluation result of the detailed text information evaluation unit 238, makes the second detailed text information 614 a new first detailed text information (new detailed text information to be summarized) and causes the unit to acquire new summarized text information 606 for the second detailed text information. The repeat control unit 240 also causes the detailed text information acquisition control unit 210 to acquire the new second detailed text information 614 for the new summarized text information 606. The repeat control unit 240 performs a looping repeat control that repeats these steps.

[0181] The loop repeat count determination unit 242 determines the upper limit of the number of repetitions 636 by the loop repeat control based on the origin generation instruction information 600. For example, the upper limit of the number of repetitions 636 is determined so that the larger the amount of text specified in the origin generation instruction information 600, the greater the number of repetitions.

[0182] The feedback control unit 244 variably controls the summary instruction information 604 in the loop repeat control based on the evaluation result of the detailed text information evaluation unit 238. For example, this includes the control of adding a specific part to the summary instruction information 604 for loop repeat.

[0183] Figures 24 to 26 are flowcharts illustrating the flow of the editing process performed by the server system 1100. The server system 1100 receives a request operation from user 2 at the user terminal 1500 (step S10) and receives the number of request outputs, which is the number of scenarios to be output in relation to the request (step S12). This information is stored as request information 514.

[0184] Next, the server system 1100 generates origin generation instruction information 600 in response to the request operation (step S14), and provides the generated origin generation instruction information 600 to the detailed text generation AI 11 to obtain the first detailed text information 602 (step S16).

[0185] The server system 1100 displays the generated first detailed document information 602 on the user terminal 1500 and accepts the user's judgment to determine whether the first detailed document information 602 is correct or incorrect, and to specify a particular part of the first detailed document (step S18).

[0186] If the user judgment result is unsuccessful (NO in step S30), the server system 1100 initializes the number of repetitions 634 to "0" and sets the upper limit of the number of repetitions 636 according to the amount of text in the first detailed document (step S32). For example, the upper limit of the number of repetitions 636 is set to be larger as the amount of text increases.

[0187] Next, the server system 1100 initializes the summary repeat count 630 to "0" (step S40), generates summary instruction information 604 (step S42), and provides this to the summary generation AI 12 to obtain summary text information 606 (step S44).

[0188] Next, the server system 1100 generates summary evaluation instruction information 608 and provides it to the summary evaluation AI 13 to evaluate the summary text of the summary text information 606 (step S46).

[0189] If the evaluation result of the summary text is unsatisfactory (NO in step S50), the server system 1100 discards the summary text information 606 that was just acquired and increases the summary repeat count 630 by "1" (step S52). Then, it sets the summary randomness parameter value according to the summary repeat count 630 (step S54) and returns to step S42. In other words, the summary instruction information 604 is set according to the summary repeat count 630. The summary instruction information 604 after step S54 corresponds to the first retry instruction information that executes the summary repeat control as the first retry control. Note that an upper limit on the number of retry operations may be set as appropriate.

[0190] If the evaluation result of the summary text is satisfactory (YES in step S50), the process moves to Figure 25, where the server system 1100 displays the summary text of the summary text information 606 as text on the user terminal 1500 and accepts the operation to specify the highly extended portion (step S60).

[0191] Then, extended instruction information 612 is generated (step S62), and if a highly extended portion was specified in step S60 (step S64), a description of the highly extended portion is added or modified to the extended generation conditions of the extended instruction information 612 (step S66).

[0192] Next, the server system 1100 initializes the detailed repeat count 632 to "0" (step S70), presents the extended instruction information 612 on the user terminal 1500, and accepts modifications from user 2 (step S72).

[0193] Then, the server system 1100 provides the extended instruction information 612 to the extended generation AI 14 to obtain the second detailed text information 614 (step S74). Then, it generates extended evaluation instruction information 616 (step S76) and provides this to the extended evaluation AI 15 to perform an evaluation of the extension (step S78).

[0194] If the extended evaluation result is unsuccessful (NO in step S80), the server system 1100 discards the second detailed text information 614 that was just acquired and increases the detailed repeat count 632 by "1" (step S82). Then, it changes the extended randomness control parameter value according to the detailed repeat count 632 of the extended instruction information 612 (step S84).

[0195] Furthermore, extended evaluation conditions are added to the extended instruction information 612 (step S86) that instruct improvement for evaluation items that do not meet a predetermined standard among the evaluation points for each evaluation item in the extended evaluation (see Figure 17). Then, the process returns to step S72. In other words, the extended instruction information 612 after step S86 corresponds to second retry instruction information that executes detailed repeat control as a second retry control.

[0196] Steps S84 and S86 are referred to as "feedback control related to detailed repeat".

[0197] If the extended evaluation result is a pass (YES in step S80), the server system 1100 generates classification instruction information 620 (step S100) and provides it to the document classification AI 16 to classify the second detailed document information 614 (step S102).

[0198] Next, the server system 1100 generates detailed text evaluation instruction information 622 (step S104), and provides this to the detailed text evaluation AI 17 to evaluate the second detailed text information 614 (step S106).

[0199] Moving to Figure 26, if the second detailed text information 614 is unacceptable (NO in step S120), it is determined whether the number of cycle repeats 634 has reached the upper limit of the number of repetitions 636 (step S122).

[0200] If the upper limit of repetitions 636 has not been reached (NO in step S122), the server system 1100 displays the second detailed text of the second detailed text information 614 as text on the user terminal 1500. At this time, already specified specific parts are identified and displayed, and operations to deselect and add specific parts are accepted.

[0201] Then, if there are any deletions or additions to a specific part (step S126), the server system 1100 updates the summary instruction information 604 according to the specified (step S128).

[0202] Next, the server system 1100 increases the number of cycle repeats 634 by "1" (step S136). Then, it changes the setting of the summarization amount in the summary generation conditions of the summary instruction information 604, for example, the setting of "number of sentences after summarization," according to the summary instruction information 604 for cycle repeats (step S138).

[0203] The number of sentences after summarization is one context that determines the summarization volume of the summary. In addition to the number of sentences after summarization, the number of characters after summarization may also be set and modified as part of the summary volume setting in the summary generation conditions.

[0204] Next, the detailed text information to be summarized in the summary instruction information 604 is replaced with the second detailed text information 614 (step S140), and the process returns to step S72. In other words, loop repeat control is performed. Steps S138 and S140 are referred to as "feedback control related to loop repeat."

[0205] If the second detailed information 614 is approved (YES in step S120), the server system 1100 counts the number of approved second detailed information 614s. If the number of approved second detailed information 614s does not reach the number of requested outputs received in step S12 (NO in step S150), the system returns to step S40. In other words, it performs "forced repeat control".

[0206] If the number of approved second detailed document information 614s reaches the number of requested outputs received in step S12 (YES in step S150), the server system 1100 uses them as output document information corresponding to the request. It then sends them to the user terminal 1500 and provides / outputs them to user 2 (step S152).

[0207] As described above, this embodiment makes it possible to provide a new technology that enables improved efficiency in text generation. Specifically, the editing system 1000 can generate better text by summarizing the first detailed text and expanding from the summary to generate a second detailed text.

[0208] Even if the first detailed text is generated by the generation unit and contains inconsistencies or unexpected plot developments, summarizing eliminates these flaws. Then, by expanding and enhancing the text based on this summary, it is automatically rewritten into a better text and provided to user 2.

[0209] [Modifications] Although examples of embodiments to which the present invention is applied have been described above, the forms to which the present invention can be applied are not limited to the above forms, and it is possible to add, omit, or change the components as appropriate.

[0210] (Modification 1) For example, although the proofreading system 1000 was exemplified as a client-server type, multiple user terminals 1500 may be implemented in a P2P (Peer to Peer) architecture. In this case, programs and data corresponding to the division of functions are stored in the user terminals 1500, and the functions corresponding to the server processing unit 200s in the above embodiment are distributed and implemented by the user terminals 1500 which become P2P nodes. The same effects as in the above embodiment can be obtained in this configuration as well.

[0211] (Modification 2) In the above embodiment, the server system 1100 was exemplified as having a generation AI 10 (generation unit), but it is not limited to this. As shown in the correction system 1000B in Figure 27, the user terminal 1500B may be configured to have these generation AI 10 (generation unit). In this configuration, the server system 1100B transmits various instruction information to the AI ​​of the user terminal 1500B via the network 9 to acquire data.

[0212] (Modification 3) In the above embodiment, the proofreading system 1000 may be implemented not as a client-server type, but solely as the computer system that was the user terminal 1500 in the above embodiment.

[0213] Specifically, the user terminal 1500B in Figure 27 stores all the data that the server system 1100 in the above embodiment is supposed to store (see Figure 22). However, instead of the server program 501 and the distribution client program 503, an editing program is provided as an application program for the user terminal 1500.

[0214] In the above embodiment, the editing program implements all of the functional units of the server system 1100 (see Figure 23) on the user terminal 1500B. In this modified example, the editing program is executed on the user terminal 1500B. The processing flow in the above embodiment (see Figures 24 to 26) can be interpreted by replacing the execution entity from the server system 1100 to the user terminal 1500B.

[0215] (Modification 4) In the above embodiment, the detailed text to be edited was generated by the detailed text generation AI 11, but the user 2 may also use a detailed text that has been created separately as the subject of editing. For example, based on the above embodiment, the editing process β shown in Figures 28 to 30 is executed instead of the editing process. In editing process β, the server system 1100 executes steps S20 to S26 instead of steps S10 to S18.

[0216] Specifically, as shown in Figure 28, the server system 1100 inputs the data of the detailed document prepared in advance by user 2 as the first detailed document information 602 (step S20) and accepts the number of requested outputs (step S22).

[0217] Next, the server system 1100 generates classification instruction information 620 to classify the input first detailed text information 602, and provides this to the text classification AI 16 to classify the first detailed text information 602 (step S24). Then, it generates detailed text evaluation instruction information 622 to designate the first detailed text information 602 as the detailed text to be evaluated, and provides this to the detailed text evaluation AI 17 to evaluate the detailed text of the first detailed text information 602 (step S26). This evaluation result is later referenced in the evaluation of the second detailed text information 614 in the editing process β.

[0218] Furthermore, in the editing process β, step S42B is executed instead of step S42 in the above embodiment. That is, the server system 1100 sets the summarization conditions of the summarization instruction information 604 (see Figure 9) generated in step S42B based on the detailed text to be summarized, i.e., the input first detailed text information 602. For example, the number of sentences after summarization in the summarization conditions is set to increase as the amount of text in the input first detailed text information 602 increases.

[0219] Furthermore, in the editing process β, step S62B is executed instead of step S62 in the above embodiment. That is, the server system 1100 generates extended instruction information 612B as shown in Figure 31. The extended instruction information 612B includes a description in the extended generation conditions that sets the amount of text to approximately the same amount as that of the first detailed text information 602 in the editing process β. The underlined portion corresponds to this. In other words, in the editing process β, extended generation conditions based on the first detailed text information 602 are generated.

[0220] Furthermore, in the editing process β, step S104B is executed instead of step S104 in the above embodiment. That is, the server system 1100 generates detailed text evaluation instruction information 622B as shown in Figure 32. The evaluation conditions of the detailed text evaluation instruction information 622B are described in a comparative format between the classification result in step S24 and the evaluation result in step S26. The underlined portion corresponds to this.

[0221] According to the configuration of this modified example, the same effects as those of the above embodiment can be obtained.

[0222] (Modification #5) In Modification #4, the detailed text prepared in advance by User 2 is not limited to a scenario. For example, it could be a gameplay log. The "gameplay log" referred to here is text data that describes events in the game in chronological order, and can be, for example, in prose form, novel-like text data, etc. The gameplay log may be generated by a generative AI that has been trained to take so-called replay data (a set of data that can reproduce gameplay) as input and output a gameplay log.

[0223] If user 2 uses the game play log as the detailed text provided in advance, it is assumed that the generating AI 10 has been trained to use the game play log as the detailed text and that the AI ​​model has been constructed accordingly.

[0224] For example, as shown in Figure 33, the summary generation AI 12C in this configuration is pre-trained on a training dataset that takes game play logs to be summarized as input. Also, as shown in Figure 34, the extended generation AI 14C in this configuration is pre-trained to output game play logs as detailed text information. Also, as shown in Figure 35, the detailed text evaluation AI 17C in this configuration takes game play logs as detailed text information as input and is pre-trained to output evaluation scores for each item related to the fun factor of the game as evaluation score by evaluation item.

[0225] Of course, not only for AI but also for various types of instruction information, this configuration uses the game play log as the detailed text in the above embodiment and is configured to generate the game play log.

[0226] With this configuration, user 2 can obtain a different gameplay log with a different storyline based on the gameplay log they initially entered. For example, they can obtain a new document that serves as a sequel to the gameplay log they initially entered.

[0227] (Modification 6) In the above embodiment, the term "summary" was used to describe the technical details, but embodiments may be constructed in which "abstract," "outline," "overview," or "digest" are used instead of "summary."

[0228] (Variation 7) You may set an appropriate upper limit on the number of times summary repeats and detailed repeats occur. For example, you may set a higher upper limit the more generation conditions there are for the origin generation instruction information 600.

[0229] (Modification #8) In the above embodiment, the return position of the first retry control, which is the summary repeat control, and the return position of the second retry control, which is the detail repeat control, were described as fixed. However, these return positions may be changed according to the situation.

[0230] Specifically, if summarization fails (NO in step S50 of Figure 24), the process is restarted from immediately before summarization (step S42), but this is not limited to this. For example, after step S50, the server system 1100 determines whether the number of cycle repeats has reached a predetermined value (e.g., "5" or more). If affirmative, the first detailed text information 602 is overwritten with the second detailed text information 614, whose generation date and time is one step earlier (meaning it has passed the extended evaluation up to the previous cycle), and the latest summary instruction information 604, whose generation date and time is the most recent, is deleted. Then, the process may return to step S42.

[0231] By doing this, it becomes possible, with a low probability, to discard the result of the previous loop repeat, i.e., the process of the previous expansion, and rewind to the expansion before that. This makes it possible to escape the situation in the summarization process where one is trapped in a meaningless loop that continues to produce the same unsatisfactory summary.

[0232] Furthermore, if the expansion fails (NO in step S80 of Figure 25), the process is restarted from immediately before the expansion (step S72), but this is not limited to that. For example, after step S80, the server system 1100 determines whether the number of cycle repeats has reached a predetermined value (e.g., "5" or more). If affirmative, the first detailed text information 602 is overwritten with the second detailed text information 614, whose generation date and time is one step earlier (meaning it has passed the expansion evaluation up to the previous cycle), and the summary instruction information 604, whose generation date and time is the most recent, is deleted. Then, the process may return to step S72. Alternatively, the process may return to step S42.

[0233] By doing this, it becomes possible, with a low probability, to discard the result of the previous loop repeat, i.e., the previous expansion process, and rewind to the expansion before that. This makes it possible to escape from a situation where the expansion process is stuck in a meaningless loop that continues to output the same unsuccessful expansion.

[0234] (Modification #9) When expanding the list-format summary text information 606 to the second detailed text information 614, for example, one may be created and expanded one by one for each of the list-like summary texts in the summary text information 606, and the expansion may be evaluated. Specifically, steps S70 to S86 may be executed for each summary text.

[0235] This allows for editing each summary text individually, for example. This configuration also makes it possible to escape a situation where the extension process becomes trapped in a meaningless loop of continuously outputting unacceptable extensions.

[0236] (Modification 10) In the above embodiment, an example was shown in which one second detailed document information 614 is generated from one summary document information 606, but the embodiment is not limited to this. For example, a configuration may be used in which multiple second detailed document information 614 are generated from one summary document information 606.

[0237] Specifically, in step S74, multiple second detailed document information 614s are generated, and steps S76 to S106 are executed for each of the generated second detailed document information 614s.

[0238] Furthermore, in step S120, if all of the second detailed document information 614 are unsatisfactory, the evaluation determined for each in step S106 may be referenced, and a loop may be executed only for the unsatisfactory second detailed document information 614 with the highest evaluation.

[0239] As another example for step S120, suppose three second detailed text information 614s, A pattern, B pattern, and C pattern, are generated, but all are unsatisfactory. Alternatively, the second detailed text information 614 of pattern C, which received the highest evaluation, may be used as the first detailed text information 602, and a loop may be executed.

[0240] As another example for step S120, suppose three second detailed document information 614s, A pattern, B pattern, and C pattern, are generated, but all are unsatisfactory, and the evaluations of B pattern and C pattern are not the predetermined minimum evaluation, but only A pattern is the predetermined minimum evaluation. In this case, a loop may be executed for B pattern and C pattern.

[0241] 10...Generation AI 11...Detailed text generation AI 12...Summary generation AI 13...Summary evaluation AI 14...Extended generation AI 15...Extended evaluation AI 17...Detailed text evaluation AI 200s...Server processing unit 204...Generation instruction information receiving unit 210...Detailed text information acquisition control unit 211...Summary instruction information generation unit 212...Summary generation condition setting unit 214...Extended instruction information generation unit 216...Extended randomness control unit 218...Extended generation condition setting unit 220...Detailed text generation unit 222...Remaining part specification unit 224...Summary text information acquisition control unit 226...Summary generation unit 228...Summary randomness control unit 230...Summary text information evaluation unit 232...Extended generation unit 234...High-extension part specification unit 238...Detailed text information evaluation unit 240...Repeat control unit 242...Circle repeat count determination unit 244...Feedback control unit 501...Server program 512...Trained generative AI model 516...Specific part specification information 600...Origin generation instruction information 602...First detailed text information 604...Summarization instruction information 606...Summarized text information 608...Summarization evaluation instruction information 610...High extension part specification information 612...Extension instruction information 614...Second detailed text information 616...Extension evaluation instruction information 622...Detailed text evaluation instruction information 636...Number of repetitions 1000...Correction system 1100...Server system 1500...User terminal

Claims

1. A computer system comprising: a summary text information acquisition control unit that controls the acquisition of a first summary text information for a first detailed text information using a summary generation unit that can generate a summary, abstract, outline, overview, or digest (hereinafter collectively referred to as "summary") for a given detailed text information; and a detailed text information acquisition control unit that controls the acquisition of a second detailed text information for the first summary text information using an extension generation unit that can generate detailed text information by extending the given summary text information.

2. The computer system according to claim 1, further comprising: a repeat control unit that performs a cyclical repeat control, which repeatedly causes the summary text information acquisition control unit to acquire new first summary text information for the second detailed text information as new first detailed text information, and causes the detail text information acquisition control unit to acquire new second detailed text information for the new first summary text information.

3. The computer system according to claim 2, further comprising a detailed text information evaluation unit for evaluating the second detailed text information, wherein the repeat control unit executes the cycle repeat control based on the evaluation result of the detailed text information evaluation unit.

4. The computer system according to claim 2 or 3, wherein the summary text information acquisition control unit variably controls the amount of summary generated by the summary generation unit based on the number of repetitions of the circular repeat control.

5. The computer system according to any one of claims 2 to 4, wherein the repeat control unit performs a redo control that restarts from either acquiring new first summary text information or acquiring new second detailed text information.

6. The computer system according to claim 5, wherein the repeat control unit performs the retry control based on given retry instruction information.

7. A computer system according to any one of claims 1 to 6, further comprising: a generation instruction information receiving unit that receives given generation instruction information for generating new detailed text information; the summary generation unit having a generation unit capable of generating the summary; the summary text information acquisition control unit generating summary instruction information to be given to the generation unit of the summary generation unit based on the generation instruction information; the extended generation unit having a generation unit capable of performing the extended generation information; and the detailed text information acquisition control unit generating extended instruction information to be given to the generation unit of the extended generation unit based on the generation instruction information.

8. The computer system according to claim 7, wherein the summary instruction information includes information indicating the amount of summary.

9. A computer system according to any one of claims 2 to 6, further comprising: a generation instruction information receiving unit that receives given generation instruction information for generating new detailed text information; a summary generation unit having a generation unit capable of generating the summary; a summary text information acquisition control unit that generates summary instruction information to be given to the generation unit of the summary generation unit based on the generation instruction information; an extended generation unit having a generation unit capable of performing the extended generation information; a detailed text information acquisition control unit that generates extended instruction information to be given to the generation unit of the extended generation unit based on the generation instruction information; and a cycle repeat count determination unit that determines the number of repetitions by the cycle repeat control based on the generation instruction information.

10. The computer system according to claim 3, further comprising: a generation instruction information receiving unit that receives given generation instruction information for generating new detailed text information; the summary generation unit having a generation unit capable of generating the summary; the summary text information acquisition control unit generating summary instruction information to be given to the generation unit of the summary generation unit based on the generation instruction information; the extended generation unit having a generation unit capable of performing the extended generation information; the detailed text information acquisition control unit generating extended instruction information to be given to the generation unit of the extended generation unit based on the generation instruction information; and a feedback control unit that variably controls the summary instruction information in the loop repeat control based on the evaluation result of the detailed text information evaluation unit.

11. The computer system according to any one of claims 1 to 10, further comprising a retained portion designation unit for designating specific portions of the first detailed text information to be retained in the summary, wherein the summary text information acquisition control unit performs control to acquire the first summary text information while retaining the specific portions designated by the retained portion designation unit.

12. The computer system according to claim 4, further comprising: a retained portion designation unit that designates a specific portion of the second detailed text information to be retained in the summary, wherein the summary text information acquisition control unit performs control in the loop repeat control to acquire the new first summary text information while retaining the specific portion designated by the retained portion designation unit.

13. The computer system according to any one of claims 1 to 12, further comprising a high-extension portion designation unit that designates a specific portion of the first summary text information to have a relatively high degree of extension, wherein the detailed text information acquisition control unit performs the extension with a relatively high degree of extension on the specific portion designated by the high-extension portion designation unit to acquire the second detailed text information.

14. The computer system according to any one of claims 2 to 6, wherein the summary text information acquisition control unit comprises a summary generation condition setting unit for setting summary generation conditions, and acquires the first summary text information by causing the summary generation unit to generate a summary based on the summary generation conditions.

15. The computer system according to claim 14, wherein the summary generation condition setting unit sets the summary generation conditions based on the number of repetitions of the loop repeat control.

16. The computer system according to claim 14 or 15, wherein the summary generation condition setting unit sets the summary generation conditions based on the first detailed text information.

17. The computer system according to any one of claims 1 to 16, wherein the summary text information acquisition control unit acquires the first summary text information as information in list format.

18. The computer system according to claim 17, wherein the extension generation unit performs the extension by adding new list-format information to the list-format information of the first summary text information.

19. The computer system according to any one of claims 1 to 18, wherein the detailed text information acquisition control unit comprises an extended generation condition setting unit for setting extended generation conditions, and the second detailed text information is acquired by causing the extended generation unit to perform the extended generation conditions.

20. The computer system according to claim 19, wherein the extended generation condition setting unit sets the extended generation conditions based on the first detailed text information.

21. The computer system according to any one of claims 1 to 20, further comprising: a detailed repeat control unit that performs detailed repeat control to cause the detailed text information acquisition control unit to repeatedly acquire the second detailed text information.

22. The computer system according to claim 21, further comprising: a detailed text information evaluation unit for evaluating the second detailed text information, wherein the repeat control unit executes the detailed repeat control based on the evaluation result of the detailed text information evaluation unit.

23. The computer system according to claim 22, further comprising: an extended randomness control unit that generates the detailed text information using a given extended randomness control parameter value for controlling the randomness related to the extended; and an extended randomness control unit that variably controls the extended randomness control parameter value based on the number of repetitions of the detailed repeat control and / or the evaluation result of the detailed text information evaluation unit.

24. The computer system according to any one of claims 1 to 23, further comprising: a summary repeat control unit that performs summary repeat control to cause the summary text information acquisition control unit to repeatedly acquire the first summary text information.

25. The computer system according to claim 24, further comprising: a summary text information evaluation unit for evaluating the first summary text information, wherein the repeat control unit executes the summary repeat control based on the evaluation result of the summary text information evaluation unit.

26. The computer system according to claim 25, further comprising: a summary generation unit, a summary randomness control unit that generates a summary using a given summary randomness control parameter value for controlling the randomness related to the generation of the summary, and a summary randomness control unit that variably controls the summary randomness control parameter value based on the number of repetitions of the summary repeat control and / or the evaluation result of the summary text information evaluation unit.

27. The computer system according to any one of claims 24 to 26, wherein the detailed text information acquisition control unit acquires second detailed text information for a first summary text information selected from a plurality of first summary text information acquired by the summary repeat control based on a given condition.

28. The computer system according to any one of claims 2 to 6, wherein the detailed text information acquisition control unit is instructed to acquire a plurality of second detailed texts from the first summary text information, and the repeat control unit, in the loop repeat control, instructs the summary text information acquisition control unit to acquire new first summary text information by selecting the second detailed texts from the plurality of second detailed texts based on given conditions as new first detailed texts.

29. The computer system according to any one of claims 1 to 28, wherein the first detailed text information is information from a gameplay log of a given game, and new gameplay log information can be obtained as the second detailed text information.

30. A text generation control method comprising: controlling the acquisition of a first summary text information for a first detailed text information using a summary generation unit capable of generating a summary, abstract, outline, overview, or digest (hereinafter collectively referred to as "summary") for a given detailed text information; and controlling the acquisition of a second detailed text information for the first summary text information using an extension generation unit capable of generating detailed text information by extending the given summary text information.

Images