System, information processing method, and program

Abstract

When a character is controlled by user input in a free-form input format, the character is made to perform actions appropriate to the situation. [Solution] The system includes a computer device and a user terminal. The computer device receives user input in a free-form text or voice input format from the user, converts the text or voice into text, matches the converted text against a matching list containing multiple keywords and / or multiple key sentences, and generates reaction information to cause a character to perform a predetermined reaction if a keyword or key sentence is detected in the text or converted text. The user terminal causes the character to perform a reaction based on the reaction information received from the computer device.

Description

Technical Field

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

Background Art

[0002] A large number of systems that allow users to interact with AI have emerged, such as chatbots that use language models including large-scale language models. However, if AI is allowed to interact freely, there are risks that, for example, AI will make utterances including violent expressions, sexual expressions, crime-related topics and the like, causing discomfort to users or encouraging crimes.

[0003] Non-Patent Document 1 discloses that the above risks are addressed by an "abusing detection model" that interrupts a dialogue when AI detects an inappropriate utterance, situation, or context, and a "keyword filtering system" that interrupts a dialogue when an inappropriate keyword is sensed during the dialogue.

Prior Art Documents

Non-Patent Document

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Non-Patent Document 1 only provides a general overview of how to address the above risks, but does not disclose any specific configurations for addressing these risks. Furthermore, Non-Patent Document 1 only describes how to prevent inappropriate dialogue, and does not describe how to control the AI's behavior in other situations.

[0006] An object of at least one embodiment of this disclosure is to make a character behave in a situational manner when the character is operated in response to user input in a free input format. [Means for solving the problem]

[0007] In a non-limiting view, a system according to one embodiment of the present disclosure is a system including a computer device and a user terminal that can communicate with the computer device, wherein the computer device includes: receiving means for receiving user input in a free input format in text or voice from the user via the user terminal; matching means for matching converted text obtained by converting the text or voice into text with a matching list including a plurality of keywords and / or a plurality of key sentences; and reaction information generating means for generating reaction information for causing a character to perform a predetermined reaction when the matching means detects the keywords or key sentences from the text or the converted text, wherein the reaction information is transmitted to the user terminal, and the user terminal includes execution means for causing the character to perform a reaction based on the reaction information received from the computer device.

[0008] In a non-limiting view, an information processing method according to one embodiment of the present disclosure is an information processing method performed in a system including a computer device and a user terminal that can communicate with the computer device, the method comprising: causing the computer device to perform a reception step of receiving user input in free input format in text or voice from the user via the user terminal; a matching step of matching converted text obtained by converting the text or voice into text with a matching list including a plurality of keywords and / or a plurality of key sentences; a reaction information generation step of generating reaction information for causing a character to perform a predetermined reaction if the matching means detects the keywords or key sentences from the text or the converted text; the reaction information being transmitted to the user terminal, and an execution step of causing the user terminal to perform a reaction based on the reaction information received from the computer device.

[0009] In a non-limiting view, a program according to one embodiment of the present disclosure is a program executed on a computer device, which enables the computer device to implement: a reception function for receiving user input in a free input format in the form of text or voice from a user; a matching function for matching converted text obtained by converting the text or voice into text with a matching list containing a plurality of keywords and / or a plurality of key sentences; and a reaction information generation function for generating reaction information to cause a character to perform a predetermined reaction when the matching function detects the keyword or key sentence from the text or the converted text. [Effects of the Invention]

[0010] Each embodiment of the present invention resolves one or more of the shortcomings. [Brief explanation of the drawing]

[0011] [Figure 1]This figure shows an example of a system configuration corresponding to at least one embodiment of the present disclosure. [Figure 2] This block diagram shows an example of a functional configuration of a system corresponding to at least one embodiment of the present disclosure. [Figure 3] This figure shows an example of a specific list corresponding to at least one embodiment of the present disclosure. [Figure 4] This figure shows an example of a second prohibited list corresponding to at least one embodiment of the present disclosure. [Figure 5] This figure shows an example of a determination item corresponding to at least one embodiment of the present disclosure. [Figure 6] This flowchart shows an example of an information processing method corresponding to at least one embodiment of the present disclosure. [Figure 7] This is a conceptual diagram showing an example of a determination corresponding to at least one embodiment of the present disclosure. [Figure 8] This figure shows an example of a character-user interaction corresponding to at least one embodiment of the present disclosure. [Modes for carrying out the invention]

[0012] Hereinafter, examples of embodiments of this disclosure will be described with reference to the drawings. The various components in each embodiment described below can be combined as appropriate, provided that no inconsistencies arise. Furthermore, details described in an example of one embodiment (for example, components with the same reference numerals or names as those already described) may be omitted in other embodiments. Also, operations and processes unrelated to the characteristic features of each embodiment may be omitted. Moreover, the order of the various processes constituting the flowcharts described below may be in any order, provided that no inconsistencies arise in the processing content, and may be executed in parallel. In this specification, the terms "first," "second," etc., are used merely for distinction and do not indicate the order of processing.

[0013] [First Embodiment] First, a first embodiment of this disclosure will be described. In the following, a system including a computer device and a user terminal capable of communicating with the computer device will be given as an example of the first embodiment.

[0014] Figure 1 shows an example of the configuration of System 5 corresponding to at least one embodiment of the present disclosure. As shown in Figure 1, System 5 includes a computer device 1 and a plurality of user terminals 2A, 2B, etc. (hereinafter collectively referred to as "user terminals 2"). The computer device 1 and user terminals 2 can communicate with each other via a network 4 and can send and receive data from each other. In this embodiment, the computer device 1 functions as a server that provides services such as games to the user terminals 2.

[0015] Computer device 1 is a computer that performs various information processing. Computer device 1 includes, for example, a processor 11, memory 12, storage 13, and a communication I / F 14, which are connected to each other via a communication bus. Computer device 1 may also include other conventionally known components.

[0016] The processor 11 performs various calculations and controls. The processor 11 may be a general-purpose processor such as a CPU (Central Processing Unit), or a dedicated processor such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). The memory 12 includes storage devices such as a ROM (Read Only Memory) that stores various programs, and a RAM (Random Access Memory) that provides a work area when various programs are executed.

[0017] The storage 13 includes, for example, a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The storage 13 only needs to have a storage area accessible by the computer device 1, and for example, may have a dedicated storage area outside the computer device 1. The computer device 1 executes various types of information processing by the processor 11 using data read out to the memory 12, for example, and stores the obtained processing results in the storage 13 as necessary. The communication I / F 14 includes various processing circuits, connection terminals, and the like for communicating with other devices via the network 4, and is configured to conform to various communication standards.

[0018] The user terminal 2 is an information terminal operated by a user. The user terminal 2 may be, for example, a portable information terminal such as a smartphone, a tablet, a PDA (Personal Digital Assistant), or a notebook personal computer, or may be a wearable device such as smart glasses, VR goggles, AR glasses, or AR contact lenses. Further, the user terminal 2 may be a stationary information terminal such as a desktop personal computer. The user terminal 2 may include only one type of the various information terminals described above, or may include a plurality of types.

[0019] The external device 3 includes a predetermined language model such as a large language model. The external device 3 and the computer device 1 can be communicatively connected to each other via the network 4, and can transmit and receive data to and from each other.

[0020] Figure 2 is a block diagram illustrating an example of a functional configuration of the system 5 corresponding to at least one of the embodiments of the present disclosure. The computer device 1 at least includes a storage unit 20 and a control unit 30. The storage unit 20 is configured by, for example, a storage device such as a ROM, an HDD, or an SDD. The storage unit 20 stores various programs, data, and the like for operating the computer device 1.

[0021] The storage unit 20 can store, for example, a program 21, a comparison list 22, judgment item information 26, and a user input history 27. The program 21 is, for example, a computer program for implementing various functions described later on the computer device 1. The comparison list 22, judgment item information 26, and user input history 27 will be described later.

[0022] The control unit 30 is composed of, for example, a processor and memory such as RAM. When program 21 is executed, the control unit 30 of the computer device 1 functions as, for example, a communication unit 31, a reception unit 32, a matching unit 33, a reaction information generation unit 34, a first prompt generation unit 35, a first acquisition unit 36, a second prompt generation unit 37, and a second acquisition unit 38.

[0023] The communication unit 31 has the function of transmitting and receiving data between, for example, the user terminal 2 and the external device 3. The receiving unit 32 has the function of receiving user input in a free-form text or voice format from the user via the user terminal 2. The transmission of data related to user input from the user terminal 2 may be performed sequentially during input, or it may be performed in response to the operation of a button or the like for transmitting user input content on the user terminal 2.

[0024] When user input is provided by voice, for example, it is preferable that the user terminal 2 sequentially transmits the user's voice information to the computer device 1, and the receiving unit 32 receives the user input as sequential input. The sequentially input voice information is converted to text in real time using, for example, speech recognition technology, and sequentially received by the receiving unit 32 as user input. The text conversion may be performed on the user terminal 2 or on the computer device 1. Even when user input is provided by text, for example, the receiving unit 32 may receive the user input as sequential input in the same manner as described above.

[0025] The matching unit 33 has the function of matching, for example, the converted text (hereinafter also referred to as "text, etc.") obtained by converting the user's text or voice received by the reception unit 32 into text, with a matching list 22 that includes multiple keywords and / or multiple key sentences (hereinafter also referred to as "keywords, etc.").

[0026] The reaction information generation unit 34 has the function of generating reaction information to cause a character to perform a predetermined reaction when, for example, the matching unit 33 detects keywords, etc., included in the matching list 22 from the text, etc. The generated reaction information is transmitted to, for example, the user terminal 2.

[0027] Specifically, the reaction information generation unit 34 can generate reaction information to cause the character to perform a specific reaction corresponding to the detected specific keyword, etc., if it detects a specific keyword, etc., included in the specific list 23, from the user's text, etc. Furthermore, the reaction information generation unit 34 can generate reaction information to cause the character to perform a rejection reaction that rejects interaction with the user if the result of the matching by the matching unit 33 or the result of the determination by the language model 60, described later, satisfies predetermined rejection conditions.

[0028] Matching list 22 is a list containing multiple keywords, etc. Matching list 22 includes, for example, a specific list 23, a first prohibition list 24, and a second prohibition list 25. Each of these lists may be managed as a single list or as separate lists.

[0029] Specific List 23 is a list containing, for example, specific keywords, etc., that cause a character to perform a specific action. In Specific List 23, for example, multiple keywords, etc., are associated with at least one reaction that the character should perform.

[0030] Specific words and reactions are defined primarily to cause characters to perform actions appropriate to their assigned personality or the situation they are in. If there are multiple characters, the contents of specific list 23 may differ for each character, or they may be the same.

[0031] Figure 3 shows an example of a specific list 23 corresponding to at least one embodiment of the present disclosure. In the example in Figure 3, a "specific word / specific sentence" (a specific keyword etc. included in the specific list; hereinafter also referred to as "specific word etc.") and a "specific reaction" are stored in association with a character ID. If a specific word etc. is detected from the matched text etc., the character will perform the reaction associated with the detected specific word etc.

[0032] In Figure 3, "AR1~AR3," which are "specific reactions," are stored in association with "AW1," which is a specific word or phrase for character A. If the matched text or phrase contains the specific sentence shown in "AW1," reaction information is generated to cause character A to perform either "AR1" or "AR2." The choice between "AR1" and "AR2" is determined, for example, based on predetermined rules. There are no particular restrictions on the predetermined rules, but examples include random selection, selection based on the selection rate set for each option, and following the order or selection conditions set for each option.

[0033] "AR3," which is a "specific reaction," is selected, for example, when "AR1" or "AR2" is selected a predetermined number of times (for example, 3 times) or more. There is no particular limit to the predetermined number of times; it can be set appropriately depending on the character, specific word, etc. If the matched text, etc. contains the specific word shown in "AW2," reaction information is generated to cause character A to perform the specific reaction "AR4," for example.

[0034] Furthermore, "BR1~BR3," which are "specific reactions," are stored in association with "BW1," which is a specific word associated with character B. If the matched text contains the specific word indicated by "BW1," reaction information is generated to cause character B to perform one of the "BR1~BR3" specific reactions. Which of "BR1~BR3" is selected may be random, in a predetermined order, or based on other predetermined rules.

[0035] If the matched text contains a specific word indicated in "BW2", reaction information is generated to cause character B to perform a specific reaction, such as "BR4". Here, "BR4" is a reaction that refuses to continue the conversation on the topic the user has spoken about. "BR4" is a reaction set based on the relationship between character B and character A.

[0036] Furthermore, each specific word can be toggled on or off, for example. When "on," that specific word is subject to matching, and when "off," that specific word is not subject to matching. This configuration makes it possible to ensure that characters take appropriate actions even when their situation changes (for example, the progress of the game). The "on / off" toggle can be set to switch automatically, for example, according to the progress of the story. Alternatively, instead of toggling "on / off," a specific list 23 could be created for each situation the character may be in.

[0037] Returning to the explanation of Figure 2, the first forbidden list 24 is a list containing first forbidden words / first forbidden sentences (hereinafter also referred to as "first forbidden words, etc."), which are keywords, etc., used to cause a character to perform a rejection reaction that refuses to interact with the user. The first forbidden words, etc., are stored, for example, in association with a predetermined ID.

[0038] The first set of prohibited words and other terms are primarily intended to detect inappropriate expressions that are deemed inappropriate for continuing a conversation with a user, such as violent expressions (e.g., "kill," "hit," etc.), discriminatory expressions (e.g., "fat," "idiot," etc.), sexual expressions, and crime-related topics (e.g., "methamphetamine"), and to prevent the continuation of conversations that are presumed to contain inappropriate content.

[0039] For example, at least one rejection reaction is set for each of the first prohibited words, etc. If multiple rejection reactions are set, one of the rejection reactions may be selected based on a predetermined rule, for example, as explained with reference to Figure 3. Furthermore, the first prohibited list 24 may be created for each character, for example, as explained with reference to the specific list 23, or each of the prohibited words, etc. may be switchable on / off, or a list may be created for each situation in which a character may be placed.

[0040] The second prohibited list 25 is a list that includes second prohibited words / second prohibited sentences (hereinafter also referred to as "second prohibited words, etc."), such as keywords used to cause a character to perform a rejection reaction. Second prohibited words, etc. are, for example, those in which the continuation of the dialogue is prohibited or not depending on the context.

[0041] The second forbidden word, etc., is associated with and stored a prohibition condition that prohibits the continuation of the dialogue. The prohibition condition is, for example, the meaning or intent in that context. If the second forbidden word, etc., is detected from the matched text, etc., the language model 60 described later will be used to determine from the context whether or not it falls under the prohibition condition. If it is determined that it falls under the prohibition condition, the character will perform a rejection reaction. The rejection reaction can be determined, for example, in the same way as when the first forbidden word, etc., is detected.

[0042] The second set of prohibited words, similar to the first set of prohibited words, is primarily intended to detect inappropriate expressions such as violent language, discriminatory language, sexual language, and crime-related topics that are deemed inappropriate for continuing a conversation with a user, and to prevent the continuation of conversations that are presumed to contain inappropriate content.

[0043] Figure 4 shows an example of a second prohibited list 25 corresponding to at least one embodiment of the present disclosure. In the example in Figure 4, the "second prohibited word / second prohibited sentence" and the "prohibition condition" are stored in association with the word ID.

[0044] In Figure 4, the "prohibition condition" associated with the second prohibited word, "ice," is that "ice" means methamphetamine. If the language model 60, described later, determines that "ice" means methamphetamine, the character will perform a rejection reaction, for example. On the other hand, if it is determined that it means ice cream, then, for example, no dialogue rejection will occur based on the fact that "ice" is included in the matched text.

[0045] The "prohibition conditions" associated with the second prohibited word, "drug," are that "drug" refers to illegal drugs and that it is used with the intention of asking about how to obtain or use them. If the language model 60, described later, determines that these prohibition conditions are met, the character will, for example, perform a rejection reaction. On the other hand, if it is determined that "drug" refers to legally available over-the-counter drugs, or if it is used with the intention of asking about the chemical structure, rather than how to obtain or use them, then, for example, dialogue rejection based on the presence of "drug" in the matched text will not occur.

[0046] The "prohibition condition" associated with the second prohibited word, "suicide," is that "suicide" is used with the intention of asking about the method. If the language model 60, described later, determines that this prohibition condition is met, the character will perform a rejection reaction, for example. On the other hand, if it is determined that there is no intention to ask about the method, for example, no dialogue rejection will occur based on the fact that the matched text contains "suicide."

[0047] For example, at least one rejection reaction is set for each of the second prohibited words, etc. If multiple rejection reactions are set, one of the rejection reactions may be selected based on predetermined rules, for example, as explained with reference to Figure 3. Furthermore, the second prohibited list 25 may be created for each character, for example, as explained with reference to the specific list 23, or each of the prohibited words, etc. may be switchable on / off, or a list may be created for each situation in which a character may be placed.

[0048] Returning to the explanation of Figure 2, the matching unit 33 compares the text received by the receiving unit 32 with the various lists described above. The matching unit 33 may be executed in real time while the receiving unit 32 is receiving user input sequentially. In this case, if a keyword or key sentence included in the matching list 22 is detected in the text, the reaction information generation unit 34 may generate reaction information even while the receiving unit 32 is still receiving user input, or the execution unit 55, described later, may cause the character to perform a reaction.

[0049] In particular, if a specific word or the like included in the specific list 23 is detected during the matching process performed in real time by the matching unit 33, or if a first prohibited word or the like included in the first prohibited list 24 is detected, it is preferable to interrupt the acceptance of user input (user utterance) and have the character perform a specific reaction or a rejection reaction.

[0050] The first prompt generation unit 35 has the function of generating a first prompt that causes a predetermined language model to perform a judgment based on at least one judgment item for the user's text, etc. The first acquisition unit 36 ​​has the function of acquiring the result of the judgment from the predetermined language model using the first prompt, for example.

[0051] In this embodiment, the predetermined language model is the language model 60 stored in the external device 3. The external device 3 uses the language model 60 to create various response data in response to various prompts sent from the computer device 1. The created response data is transmitted to the computer device 1 as appropriate.

[0052] The language model 60 is not particularly limited as long as it is capable of generating response data in response to various prompts received from the computer device 1. Examples of language models 60 include machine learning models such as large-scale language models, medium-scale language models, and small-scale language models.

[0053] In this embodiment, multiple language models may be used. External device 3 may have multiple language models, or it may be configured to further utilize other language models included in other external devices. The language models may also be provided by computer device 1. Furthermore, computer device 1 may include a machine learning model that generates response data for prompts other than the first and second prompts.

[0054] The first prompt generation unit 35 generates a first prompt to cause the user to determine the meaning or intent of the detected second prohibited word, etc., from the context, when the matching unit 33 detects the second prohibited word, etc., from the user's text, etc. In this case, the first prompt may include, for example, an instruction to determine whether the meaning or intent of the detected second prohibited word, etc., falls under the prohibition conditions associated with the second prohibited word, etc. The meaning or intent of the detected second prohibited word, etc., falling under the prohibition conditions associated with the second prohibited word, etc., is, for example, one of the predetermined rejection conditions.

[0055] Furthermore, the criteria used by the language model 60 for determination may include, for example, an item for determining whether the content of the user's text falls under at least one predetermined prohibited topic. For example, the content of the user's text falling under a prohibited topic is one of the predetermined rejection conditions.

[0056] Furthermore, prohibited topics may have permitted words and / or permitted sentences (hereinafter also referred to as "permitted words, etc.") set to allow them. If the user's text, etc. contains permitted words, etc., the determination of whether or not it falls under a prohibited topic corresponding to those permitted words, etc. may be excluded from the determination items. Permitted words, etc. may be set for each prohibited topic, for example. Note that the first prompt is for determining the meaning, intent, topic, etc. from the context, so for example, the first prompt may include not only the text etc. received from the user this time, but also the history of recent user input.

[0057] Figure 5 shows an example of determination item information 26 corresponding to at least one embodiment of the present disclosure. In the example of Figure 5, a "determination item," a "detail item," and an "allowed word / allowed sentence" are stored in association with a determination item ID. The first prompt may include, for example, instructions to determine whether the topic indicated in the "detail item" is used with the intent corresponding to the "determination item."

[0058] The judgment item based on judgment item ID "G011" is, for example, whether or not the user's text discusses political ideology. The first prompt may include judgment items based on all judgment item IDs where the "On / Off" field is set to "On". The switching of "On / Off" can be explained using the same method described for specific list 23, for example, so the explanation is omitted here. Note that the "On / Off" setting may differ depending on the character, and the judgment items included in judgment item information 26 may also differ.

[0059] Even if the "On / Off" field is set to "On," if the user's text contains permitted words associated with that judgment item ID, for example, the first prompt will not include judgment items related to that judgment item ID. In the example in Figure 5, if the user's text contains the word "old," the judgment items based on judgment item ID "G011" will not be included in the first prompt. Note that the detection of permitted words in the user's text can be performed, for example, by the matching unit 33.

[0060] The first prompt may be a single prompt containing all judgment items, or it may be multiple prompts, each containing at least one judgment item. Furthermore, if the first prompt consists of multiple prompts, for example, each prompt may be sent sequentially to the external device 3 as soon as it is generated. If even one of the multiple judgment items included in the first prompt is determined to be a prohibited topic, it may be determined that the predetermined rejection conditions are met.

[0061] The criteria for judgment may include criteria other than those shown in Figure 5. Specifically, the criteria may include what topics are included in the user's text, etc. In this case, the denial condition may be that the topics identified by the language model 60 are included in the prohibited topics, etc.

[0062] Returning to the explanation of Figure 2, the user input history 27 includes, for example, the history of user input by at least one user. The user input history may be managed on a unit basis, for example, a single utterance by the user (from when the user starts speaking until they stop speaking to wait for the character's response), or on a unit basis, a series of utterances (context) by the user. In addition, the user input history 27 may store information associated with each past user input, for example, the results of each matching and each judgment described above. Furthermore, the user input history 27 may store information associated with each past user input, for example, the reaction performed by the character.

[0063] The control unit 30 may have a function to determine the similarity between the text, etc., that is the target of generating the first prompt and the text, etc., included in the user input history. The similarity determination may be performed on a unit of a single utterance, or on a unit of multiple sets of utterances. If a text, etc., included in the user input history is detected that has a high similarity to the text, etc., that is the target of generating the first prompt, the first prompt generation unit 35 may not generate the first prompt, and instead, the system may be configured to generate reaction information based on the result or reaction information associated with the detected text, etc. Conventional known techniques such as embedding can be used to determine the similarity. Furthermore, if the user input history 27 includes the user input histories of multiple users, the similarity of a user's user input with the user input histories of multiple users may be determined.

[0064] The second prompt generation unit 37 has the function of generating a second prompt for a predetermined language model to generate a character's response to user text, etc. The second prompt may include character information such as the character's personality and situation.

[0065] The second acquisition unit 38 has the function of acquiring a character's response from a predetermined language model, for example, using a second prompt. The reaction information generation unit 34 generates reaction information to execute a response reaction based on the character's response acquired by the second acquisition unit 38, if a predetermined rejection condition is not met.

[0066] Furthermore, the processes for obtaining the judgment results and the processes for obtaining the character's response can be executed in parallel, for example. Specifically, the series of processes following the generation of the first prompt by the first prompt generation unit 35 and the series of processes following the generation of the second prompt by the second prompt generation unit 37 can be executed in parallel.

[0067] The user terminal 2 comprises a storage unit 50 and a control unit 52. The storage unit 50 is composed of, for example, a storage device such as a ROM, HDD, or SSD included in the user terminal 2. The storage unit 50 stores a program 51. The program 51 is, for example, an information processing program for implementing various functions described later in the user terminal 2.

[0068] The control unit 52 is composed of, for example, a processor and RAM included in the user terminal 2. When program 51 is executed by the processor, the control unit 52 functions as, for example, a communication unit 53, a receiving unit 54, and an execution unit 55.

[0069] The communication unit 53 has the function of transmitting and receiving data to and from the computer device 1, for example. The receiving unit 54 has the function of receiving user input in text or voice format from the user via a predetermined input interface such as a keyboard, touchscreen, or microphone.

[0070] The user input received by the reception unit 54 is transmitted to the computer device via the communication unit 53, for example, sequentially, or in response to operations performed by the user to transmit the user input to the computer device 1. The communication unit 53 also receives, for example, reaction information transmitted from the communication unit 31 of the computer device 1.

[0071] The execution unit 55 controls the character to perform a reaction based on reaction information received from the computer device 1, for example. The user terminal 2 outputs the image and / or sound of the character controlled by the execution unit 55 to an output device (not shown), such as a display, on the user terminal 2, and presents it to the user.

[0072] System 5 may be implemented by including a native application downloaded to the user terminal 2, by including a web application, or by being implemented as a cloud-based service such as IaaS.

[0073] Next, an information processing method using the computer device 1 according to the first embodiment will be described. Figure 6 is a flowchart showing an example of an information processing method corresponding to at least one embodiment of the present disclosure. The following flowchart shows a series of processes from causing a character to perform a predetermined reaction based on user input.

[0074] The information processing methods described below can be applied, for example, to chatbots or to the control of characters appearing in games. There are no particular restrictions on the game genre; it can be broadly applied to games featuring NPCs, such as role-playing games, simulation games, and adventure games.

[0075] In step S1, user terminal 2 receives user input from the user. User input may be text input or voice input, but the following explanation will describe the case where voice input is used. The information entered by the user may be transmitted sequentially from user terminal 2 to computer device 1, or it may be transmitted all at once after the input is completed, but the following explanation will describe the case where it is transmitted sequentially.

[0076] In step S2, computer device 1 sequentially acquires the voice information input by the user. The acquired voice information is then sequentially converted into text information. Note that the conversion from voice information to text information may also be performed on user terminal 2.

[0077] In step S3, the computer device 1 compares the text obtained by converting the user's voice with various lists included in the comparison list 22. The comparison is performed in real time, for example, while the text is being acquired sequentially. The comparison with the various lists may also be performed in parallel, for example.

[0078] If a specific word or other element included in the specific list 23 is detected from the text (YES in step S4), the process proceeds to step S5. In step S5, the computer device 1 generates specific reaction information to cause the character to perform a specific reaction. The specific reaction is the reaction associated with the detected specific word or other element. The specific reaction information is sent to the user terminal 2. In step S6, the user terminal 2 causes the character to perform a specific reaction based on the specific reaction information and then terminates. The process in step S6 may be performed even if the user terminal 2 is in the process of accepting user input, interrupting the user input.

[0079] If a first forbidden word or the like included in the first forbidden list 24 is detected in the text (YES in step S7), the process proceeds to step S8. In step S8, computer device 1 generates rejection reaction information to cause the character to perform a rejection reaction that refuses to interact with the user. The rejection reaction information is sent to user terminal 2. In step S9, user terminal 2 causes the character to perform a rejection reaction based on the rejection reaction information and terminates. The process in step S9 may be performed even if user input is being accepted at user terminal 2, interrupting user input.

[0080] If a second prohibited word or the like included in the second prohibited list 25 is detected in the text (YES in step S10), the process proceeds to step S11. In step S11, the computer device 1 adds a determination item to determine whether the meaning or intent of the detected second prohibited word or the like falls under the prohibited conditions associated with the second prohibited word or the like. Specifically, it generates a first prompt that includes such determination items.

[0081] The first prompt generated in step S11 may be sent to the external device 3 immediately after generation, independently of other first prompts, or it may be sent to the external device 3 as a single first prompt together with other determination items. After step S11, the process proceeds to, for example, S12 and S18.

[0082] If no keywords or other elements included in the various lists of the matching list 22 are found in the text (all of steps S4, S7, and S10 are NO), the process proceeds to steps S12 and S18. Steps S12 and S18, and the subsequent series of processes, may be executed in parallel.

[0083] First, the series of processes from steps S12 to S17 will be explained. In step S12, the computer device 1 creates a first prompt related to the judgment items stored in the judgment item information 26. The judgment items differ depending on, for example, the character or the situation in which the character is placed. Also, if the user's text contains exclusion words, the judgment items corresponding to those exclusion words are excluded from the judgment target. The first prompt is generated, for example, when it is determined that the user's utterance has ended (the situation is one where the user is waiting for a response from the character). The generated first prompt is sent to the external device 3.

[0084] In step S13, the external device 3 inputs a first prompt to the language model 60 and causes the language model 60 to output the result of the judgment for each judgment item. If the first prompt consists of multiple independent prompts, the results of the judgments may be sent sequentially to the computer device 1.

[0085] In step S14, the computer device 1 obtains the result of the determination. If the result of the determination satisfies the predetermined rejection conditions as described above (YES in step S15), the process proceeds to step S16. In step S16, the computer device 1 generates rejection reaction information. The rejection reaction information is sent to the user terminal 2. In step S17, the user terminal 2 has the character perform a rejection reaction based on the rejection reaction information and then terminates.

[0086] Before step S12, the similarity between the text to be used to generate the first prompt and the text included in the user input history is determined. If a high degree of similarity is detected, for example, steps S12 to S15 do not need to be performed. In this case, reaction information can be generated in the same way as for previously detected text. Specifically, if a rejection reaction was given to previous text, a rejection reaction should be given again this time. If a response reaction was given to previous text, a response reaction given in the past may be used, or a response reaction based on the response generated in step S19 described later may be used.

[0087] Here, the processing of steps S12 to S17 will be specifically explained using Figures 7 and 8. Figure 7 is a conceptual diagram showing an example of a determination corresponding to at least one embodiment of the present disclosure. In the example of Figure 7, the first prompt asks, for example, whether the topic being discussed in user input 71 corresponds to any of the various items included in the determination item information 26. In Figure 7, only a single sentence spoken by the user is shown as user input 71, but the most recent dialogue history between the user and the character may also be included in user input 71.

[0088] Here, user input 71 includes the word "past," which is a permitted word for "item (judgment item): politics, Check (detailed item): ideology." In this case, "item: politics, Check: ideology" is excluded from the judgment items. In other words, the first prompt does not include instructions to determine whether user input 71 corresponds to "item: politics, Check: ideology."

[0089] In Figure 7, three language models 60 are used: language model A, language model B, and language model C. Furthermore, the judgment item information 26 specifies which language model 60 to use for each "Check (detailed item)". In Figure 7, "Item: Discrimination, Check: Body" is judged by language model A, and "Item: Discrimination, Check: Environment" is judged by language model B. In other words, there are multiple first prompts.

[0090] The judgment result 72 indicates that the topic of user input 71 corresponds to "Item: Politics, Check: Party Name". In this case, the rejection condition is met, so rejection reaction information is generated, and the character refuses to continue the conversation on the topic of user input 71.

[0091] Figure 8 shows an example of a character-user dialogue corresponding to at least one embodiment of the present disclosure. The user's utterance in dialogue field 81 is ethically sound as a single sentence, but when judged from the context, it is an inappropriate topic. In dialogue field 81, for example, after the user's utterance "It's a person," the entire dialogue field 81 (the most recent dialogue history) determines that the user's topic falls under the prohibited topic of "crime," and character A refuses to continue the topic.

[0092] Suppose the user then changes the topic, as shown in dialogue box 82. In this case, because an inappropriate context remains in the dialogue history, even if the user tries to have a normal conversation as shown in dialogue box 82, the contextual judgment of the user's utterance will still result in an inappropriate dialogue. Therefore, as shown in dialogue box 83, the character will again make an utterance rejecting the topic, and the user will be unable to return to a normal conversation.

[0093] To prevent such situations, if a character makes an utterance that avoids the topic, it is preferable that, for example, contextual judgments be made only for utterances made after that point. Specifically, for utterances made after boundary D1 in the dialogue box 84, utterances made before boundary D1 are not used as the basis for determining whether they are inappropriate or not. That is, when judging utterances made after boundary D1, the first prompt includes instructions such as, for example, that utterances made before boundary D1 are not included, or that utterances made before boundary D1 are not used as the basis for the judgment. Such processing is performed in the dialogue box 84, allowing the user to return to the dialogue.

[0094] On the other hand, as shown in dialogue box 85, if the user continues to make inappropriate remarks after the character refuses to continue the topic, the character may issue a strong warning. Also, if the number of inappropriate remarks reaches a predetermined number, the dialogue with the character or the service providing the dialogue (e.g., a game) may be forcibly terminated.

[0095] Let's return to the explanation of Figure 6. We will now describe the series of processes from steps S18 to S22. In step S18, computer device 1 generates a second prompt to generate a character response to the user's utterance. The second prompt is generated, for example, when it is determined that the user's utterance has ended. The generated second prompt is sent to external device 3.

[0096] In step S19, the external device 3 inputs a second prompt to the language model 60, causing the language model 60 to output the character's response to the user's utterance.

[0097] In step S20, computer device 1 obtains the character's response generated by the language model 60 from external device 3. In step S21, computer device 1 generates response reaction information to cause the character to perform a response reaction based on the obtained response. The response reaction information is sent to user terminal 2. In step S22, user terminal 2 causes the character to perform a response reaction based on the response reaction information and then terminates.

[0098] Here, if step S20 is executed before step S14, that is, if computer device 1 obtains information about the character's response before the judgment result, the response reaction information will not be sent to user terminal 2 until it is determined that the judgment result does not satisfy the rejection conditions. On the other hand, if step S14 is executed before step S20 and the judgment result satisfies the rejection conditions, the response reaction information will be sent to user terminal 2 without waiting for the arrival of information about the character's response, that is, regardless of the status of obtaining information about the character's response.

[0099] Furthermore, after step S20, the system may check whether the character's response contains any inappropriate expressions. In this case, for example, the system may check the character's response against the first prohibited list 24, or a separate comparison list 22 for checking the character's response may be prepared. If a prohibited word or the like is detected in the character's response, the computer device 1 generates a second prompt again and causes the language model 60 to generate the character's response again. This second prompt may include, for example, the reason why the previous response was not accepted. In this case, the system may also control the character to perform a reaction (a so-called filler reaction) such as "No, that's not right..." as if the character is reconsidering their response.

[0100] This concludes the explanation of the flowchart. However, at least some of the various processes performed by the computer device 1 described above may be performed by the user terminal 2. Alternatively, the computer device 1 may be configured to include a language model 60, so that at least some of the various processes performed by the external device 3 described above are performed by the computer device 1.

[0101] In the first embodiment, a so-called server-client type system including a computer device 1 and a user terminal 2 was illustrated, but this disclosure may also be implemented as a standalone type. In that case, at least some of the various functions of the computer device 1 described above will be provided by the user terminal 2, and at least some of the various processes performed by the computer device 1 described above will be performed by the user terminal 2.

[0102] Furthermore, in the first embodiment, "character" is not particularly limited, but may be, for example, a fictional character or a character based on a real person, animal, or plant. "Reaction" is not particularly limited, but may include, for example, the content of the character's speech, actions, and emotional expressions. "Character information" is, for example, information that defines the character's personality. "Character information" may include, for example, the character's personality, profile, hobbies, preferences, way of speaking, current situation, potential topics, prohibited topics, etc.

[0103] Although the embodiments described herein have been described in detail above, the present invention is not limited to the embodiments described above, and can be modified, improved, etc., as appropriate. The present invention includes all modifications within the meaning and scope of the claims and equivalents thereof.

[0104] [Note] The above description of the embodiments is written in such a way that a person with ordinary skill in the art to which the invention pertains can implement the invention described below.

[0105] [1] A system including a computer device and a user terminal capable of communicating with the computer device, The aforementioned computer device A receiving means for receiving user input in a free-form text or voice input format from the user via the user terminal, A matching means for matching the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences, The system includes a reaction information generation means that, when the matching means detects the keyword or key sentence from the text or the converted text, generates reaction information to cause the character to perform a predetermined reaction, The reaction information is transmitted to the user terminal. The aforementioned user terminal is The system includes an execution means that causes the character to perform a reaction based on the reaction information received from the computer device. system. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format.

[0106] [2] The receiving means receives the user input as sequential input, The verification by the verification means is performed in real time while the user input is being received. If the keyword or key sentence is detected in the text or converted text, even if the receiving means is in the process of receiving user input, the reaction information generation means generates the reaction information, and the execution means causes the character to perform the reaction. The system described in [1] above. This makes it possible, for example, to have the character perform situation-appropriate actions even while the user is speaking. In a conversation between humans, one person may interrupt the other to speak, such as by making a comment or asking a question. [2] According to this configuration, it is possible to reproduce situations that can occur in conversations between humans in conversations between the user and the character.

[0107] [3] Each of the aforementioned multiple keywords and / or the aforementioned multiple key sentences is associated with at least one reaction. The reaction information generation means generates reaction information corresponding to the detected keyword or key sentence. The system described in [1] or [2] above. This makes it possible, for example, to have a character perform appropriate actions in line with the system operator's intentions in response to a user's utterance.

[0108] [4] At least one of the aforementioned keywords and / or the aforementioned key sentence is associated with a plurality of reactions selected based on predetermined rules, The aforementioned predetermined rule is one or more of the following: random selection, drawing based on selection rates set for each reaction, selection based on the order set for each reaction, and selection based on selection conditions set for each reaction. The system described in [3] above. This makes it possible to reduce the unnaturalness of characters repeating the same reactions, for example, while also enabling characters to perform actions that are more in line with the system operator's intentions in response to user utterances.

[0109] [5] The matching list above allows you to change whether or not to include multiple keywords and / or multiple key sentences in the matching process. The system described in any of the above [1] to [4]. This makes it possible, for example, to make a character perform different actions depending on the character and the situation in which the character is located, even if the user's utterance is the same.

[0110] [6] The aforementioned matching list is set for each situation in which the character is placed. The system described in any of the above [1] to [5]. This makes it possible, for example, to make a character perform different actions depending on the situation, even if the user's speech is the same. As a result, it becomes easier to make character actions appropriate to the progress of the game.

[0111] [7] There are multiple such characters. The aforementioned matching list is set for each character. The system described in any of the above [1] to [6]. This makes it possible, for example, to make each character behave differently even if the user's speech is the same. As a result, it becomes easier to express the individuality of each character.

[0112] [8] The aforementioned computer device further, A first prompt generation means generates a first prompt for causing a predetermined language model to perform a determination based on at least one determination item for the text or the converted text, The system includes a first acquisition means for obtaining the result of the determination from a predetermined language model using the first prompt, The reaction information generation means generates reaction information to perform a reaction that rejects interaction with the user when the result of the matching or the result of the determination satisfies predetermined rejection conditions. The aforementioned matching list includes the first prohibited list. The predetermined rejection conditions include the detection by the matching means of a keyword or key sentence included in the first prohibited list from the text or the converted text, The aforementioned matching list includes the second prohibition list. The keywords or key sentences included in the aforementioned second prohibited list are such that whether or not the continuation of the dialogue is prohibited depends on the context, and the conditions under which the continuation of the dialogue is prohibited are stored in association with them. If the matching means detects a keyword or sentence included in the second prohibited list from the text or the converted text, the first prompt generation means generates a first prompt that causes the meaning or intent of the keyword or sentence included in the text or the converted text to be determined from the context. The predetermined rejection conditions include the result of the determination obtained by the first acquisition means indicating that a keyword or sentence included in the second prohibited list detected from the text or the converted text is used in a sense or with intent that prohibits the continuation of the dialogue. The system described in any of the above [1] through [7]. This makes it possible to prevent the character from performing inappropriate actions or providing undesirable information to the user by refusing to interact with the user if the user makes an inappropriate remark. Furthermore, it becomes possible to have the character respond flexibly to words and phrases that are difficult to judge as inappropriate at first glance, such as so-called slang, by refusing to interact with the user if they have an inappropriate meaning, and continuing to interact with the user if they do not.

[0113] [9] The aforementioned receiving means accepts the user input as sequential input, The matching using the first prohibited list is performed in real time while the user input is being received. If the first prohibited keyword or the first prohibited key sentence is detected in the text or the converted text, even if the receiving means is in the process of receiving user input, the reaction information generation means generates reaction information that rejects interaction with the user, and the execution means causes the character to perform the reaction. The system described in [8] above. This makes it possible, for example, to immediately have the character reject the conversation if the user makes a clearly inappropriate statement.

[0114]

[10] The determination items include items for determining whether the content of the text or the converted text falls under at least one of the predetermined prohibited topics. The aforementioned predetermined rejection conditions include the content of the text or the converted text falling under the prohibited topics, Multiple characters exist, and the judgment criteria differ depending on the character. The system described in [8] or [9] above. This makes it possible, for example, to have a character refuse to engage in conversation if the user's topic is inappropriate, even if the user hasn't uttered any specific words. It also makes it easier to express a character's personality, for example, by having the character refuse to engage in conversation about topics they dislike.

[0115]

[11] The aforementioned prohibited topics have permitted words and / or permitted sentences set to allow them. If the text or the converted text contains the permitted word or the permitted sentence, the determination of whether or not the permitted word or the permitted sentence contained in the text or the converted text falls under a prohibited topic is excluded from the determination items. The system described in

[10] above. This allows, for example, not to uniformly prohibit any prohibited topics, but to allow characters to continue dialogue even if the topic is prohibited in certain cases. As a result, it becomes possible to judge whether a topic is inappropriate more flexibly, allowing characters to respond appropriately on the spot and broadening the range of character expression.

[0116]

[12] The aforementioned determination items can be changed to determine whether or not they are subject to determination. The system described in any of the above [8] to

[11] . This makes it possible, for example, to make a character perform different actions depending on the character and the situation in which the character is located, even if the user's utterance is the same.

[0117]

[13] The aforementioned judgment criteria are set for each situation in which the character is placed. The system described in any of the above [8] to

[12] . This makes it possible, for example, to make a character perform different actions depending on the situation, even if the user's speech is the same. As a result, it becomes easier to make character actions appropriate to the progress of the game.

[0118]

[14] There are multiple instances of the aforementioned character. The aforementioned determination items are set for each character. The system described in any of the above [8] through

[13] . This makes it possible, for example, to make each character behave differently even if the user's speech is the same. As a result, it becomes easier to express the individuality of each character.

[0119]

[15] The aforementioned computer device further, A second prompt generation means for generating a second prompt to cause a predetermined language model to generate a character response to the text or the converted text, The system includes a second acquisition means for obtaining the character's response from a predetermined language model using the second prompt, If the predetermined rejection conditions are not met, the reaction information generation means generates reaction information to perform a reaction based on the character's response obtained by the second acquisition means. The process for obtaining the result of the judgment and the process for obtaining the character's response are executed in parallel. The system described in any of the above [8] to

[14] . This allows, for example, the system to reject inappropriate conversations if the user is attempting to engage in them, while simultaneously having the character react in various ways to non-inappropriate conversations, and to shorten the time it takes for the character to react.

[0120]

[16] The aforementioned computer device If the result of the determination is obtained before the character's response, and the result of the determination satisfies the predetermined rejection conditions, the reaction information generation unit generates reaction information to perform a reaction that rejects interaction with the user, regardless of the status of obtaining the character's response. The system described in

[15] above. This allows for a reduction in the time it takes for a character to react, for example, when the context indicates that a user is attempting to engage in inappropriate conversation.

[0121]

[17] The matching list further includes a character list containing multiple keywords and / or key sentences for the character, The matching means further compares the character's response with the character's list, If the matching means detects a keyword or key sentence included in the character list from the character's response, The reaction generation means does not generate reaction information based on the response in which the keyword or key sentence is detected, but rather generates reaction information to cause the character to perform a filler animation. The acquisition of the character's response by the second prompt generation means and the second acquisition means is performed again. The system described in

[15] or

[16] above. This can, for example, prevent characters from engaging in inappropriate dialogue. It can also prevent unnatural pauses before characters respond.

[0122]

[18] The aforementioned computer device It stores the history of user input by at least one user, The user input history is associated with the result of the determination or the reaction information for the user input history. Furthermore, the system includes similarity determination means for determining the similarity between the text or the converted text and the text included in the user input history. If the similarity exceeds a predetermined value, the reaction information generation means generates reaction information based on the result of the determination on the user input history or the reaction information. The system described in either [8] or

[17] above. This allows, for example, the system to reject inappropriate conversations if the user is attempting to engage in them, while simultaneously having the character react in various ways to non-inappropriate conversations, and to shorten the time it takes for the character to react.

[0123]

[19] An information processing method performed in a system including a computer device and a user terminal capable of communicating with the computer device, The aforementioned computer device, A reception step in which user input in a free-form text or voice input format is received from the user via the user terminal, A matching step of matching the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences, If the matching means detects the keyword or key sentence from the text or the converted text, the system performs a reaction information generation step which generates reaction information to cause the character to perform a predetermined reaction. The reaction information is transmitted to the user terminal. On the aforementioned user terminal, An execution step of causing the character to perform a reaction based on the reaction information received from the computer device, An information processing method that includes causing the execution of [a certain action]. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format.

[0124]

[20] A program that runs on a computer device, The aforementioned computer device, A reception function that accepts user input in the form of free-form text or voice input from the user. A matching function that matches the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences. If the matching function detects the keyword or key sentence from the text or the converted text, a reaction information generation function generates reaction information to cause the character to perform a predetermined reaction. A program that makes this a reality. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format. [twenty one] A non-temporary computer-readable medium on which the program described in

[20] above is stored. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format. [twenty two] The computer device in a system including a computer device and a user terminal capable of communicating with the computer device, A receiving means for receiving user input in a free-form text or voice input format from the user via the user terminal, A matching means for matching the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences, The system includes a reaction information generation means that, when the matching means detects the keyword or key sentence from the text or the converted text, generates reaction information to cause the character to perform a predetermined reaction, Computer device. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format.

[0125] [twenty three] A program that runs on a computer device, The aforementioned computer device, A reception function that accepts user input in the form of free-form text or voice input from the user. A matching function that matches the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences. If the matching function detects the keyword or key sentence from the text or the converted text, a reaction information generation function generates reaction information to cause the character to perform a predetermined reaction. A program that implements an execution means for causing the character to perform a reaction based on the reaction information. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format.

[0126] [twenty four] A reception means for receiving user input in a free-form text or voice input format from the user, A matching means for matching the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences, When the matching means detects the keyword or key sentence from the text or the converted text, the matching means generates reaction information to cause the character to perform a predetermined reaction, An execution means that causes the character to perform a reaction based on the reaction information, A computer device equipped with the following features. This makes it possible to make characters behave in a way that is appropriate to the situation when the character is controlled by user input in a free-form input format. [Explanation of Symbols]

[0127] 1: Computer device 2: User terminal 60: Language Models

Claims

1. A system including a computer device and a user terminal capable of communicating with the computer device, The aforementioned computer device, A receiving means for receiving user input in a free-form text or voice input format from the user via the user terminal, A matching means for matching the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences, The system includes a reaction information generation means that, when the matching means detects the keyword or key sentence from the text or the converted text, generates reaction information to cause the character to perform a predetermined reaction, The reaction information is transmitted to the user terminal. The aforementioned user terminal is The system includes an execution means that causes the character to perform a reaction based on the reaction information received from the computer device. system.

2. The receiving means receives the user input as sequential input, The verification by the verification means is performed in real time while the user input is being received. If the keyword or key sentence is detected in the text or converted text, even if the receiving means is in the process of receiving user input, the reaction information generation means generates the reaction information, and the execution means causes the character to perform the reaction. The system according to claim 1.

3. Each of the aforementioned multiple keywords and / or the aforementioned multiple key sentences is associated with at least one reaction. The reaction information generation means generates reaction information corresponding to the detected keyword or key sentence. The system according to claim 1 or 2.

4. The aforementioned computer device further, A first prompt generation means generates a first prompt for a predetermined language model to perform a determination based on at least one determination item with respect to the text or the converted text, The system includes a first acquisition means for obtaining the result of the determination from a predetermined language model using the first prompt, The reaction information generation means generates reaction information to perform a reaction that rejects interaction with the user when the result of the matching or the result of the determination satisfies predetermined rejection conditions. The aforementioned matching list includes the first prohibited list, The predetermined rejection conditions include the detection by the matching means of a keyword or key sentence included in the first prohibited list from the text or the converted text, The aforementioned matching list includes a second prohibited list. The keywords or key sentences included in the second prohibited list are such that whether or not the continuation of the dialogue is prohibited depends on the context, and the conditions under which the continuation of the dialogue is prohibited are stored in association with them. If the matching means detects a keyword or sentence included in the second prohibited list from the text or the converted text, the first prompt generation means generates a first prompt that causes the meaning or intent of the keyword or sentence included in the text or the converted text to be determined from the context. The predetermined rejection conditions include the result of the determination obtained by the first acquisition means indicating that a keyword or sentence included in the second prohibited list detected from the text or the converted text is used in a sense or with the intention of prohibiting the continuation of the dialogue. The system according to claim 1.

5. The determination items include items for determining whether the content of the text or the converted text falls under at least one of the predetermined prohibited topics. The aforementioned predetermined rejection conditions include the content of the text or the converted text falling under the prohibited topics, Multiple characters exist, and the judgment criteria differ depending on the character. The system according to claim 4.

6. The aforementioned prohibited topics have permitted words and / or permitted sentences set to allow them. If the text or the converted text contains the permitted word or the permitted sentence, the determination of whether or not the permitted word or the permitted sentence contained in the text or the converted text falls under a prohibited topic is excluded from the determination items. The system according to claim 5.

7. The aforementioned computer device further, A second prompt generation means for generating a second prompt to cause a predetermined language model to generate a response of the character to the text or the converted text, The system includes a second acquisition means for obtaining the character's response from a predetermined language model using the second prompt, If the predetermined rejection conditions are not met, the reaction information generation means generates reaction information to perform a reaction based on the character's response obtained by the second acquisition means. The process for obtaining the result of the judgment and the process for obtaining the character's response are executed in parallel. The system according to claim 4 or 5.

8. An information processing method performed in a system including a computer device and a user terminal capable of communicating with the computer device, The aforementioned computer device, A reception step in which user input in a free-form text or voice input format is received from the user via the user terminal, A matching step of matching the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences, If the matching means detects the keyword or key sentence from the text or the converted text, the system performs a reaction information generation step which generates reaction information to cause the character to perform a predetermined reaction. The reaction information is transmitted to the user terminal. On the aforementioned user terminal, An execution step of causing the character to perform a reaction based on the reaction information received from the computer device, An information processing method that includes causing the execution of [a certain action].

9. A program that runs on a computer device, The aforementioned computer device, A reception function that accepts user input in the form of free-form text or voice input from the user. A matching function that matches the converted text obtained by converting the aforementioned text or audio into text with a matching list containing multiple keywords and / or multiple key sentences. If the matching function detects the keyword or key sentence from the text or the converted text, a reaction information generation function generates reaction information to cause the character to perform a predetermined reaction. A program that makes this a reality.

Images