Information processing device and information processing method

Abstract

An information processing device according to the present disclosure comprises: a role presentation unit (111) that presents, to a first user, a role set for the first user; an acquisition unit (112) that acquires first data inputted by the first user, in response to the input of the first data; a detection unit (103) that detects, as contradiction information, information contradictory to the role in the first data acquired by the acquisition unit; and a contradiction presentation unit (110) that presents a message based on the contradiction information detected from the first data by the detection unit to the first user.

Description

[Technical Field]

[0001] The present disclosure relates to an information processing apparatus and an information processing method. [Background Art]

[0002] In recent years, demand for automatic dialogue systems that interact with users using artificial intelligence has been increasing. Further, in such automatic dialogue systems, setting a specific persona (e.g., a profile) for artificial intelligence makes it possible to realize more natural conversation between a user and the artificial intelligence. For example, in an automatic dialogue system, artificial intelligence carries out dialogue with a user using a model pre-trained according to a specific persona.

[0003] In order to train a model used by artificial intelligence in an automatic dialogue system, a dataset based on a large amount of dialogue data from more natural dialogues is required. Although this dataset can be collected manually, this approach leads to increased costs. Therefore, methods for automatically collecting datasets for training such models have been proposed (e.g., Patent Document 1, Non-Patent Document 1).

[0004] For example, in Patent Document 1, the system presents a specific response agent, a first user posts a question for the response agent to the system, and a second user impersonates the response agent to post the response agent's answer to the question to the system. Accordingly, the system collects dialogue data between any person and the response agent.

[0005] Further, in Non-Patent Document 1, a persona sentence as an introduction of a target person is presented to each crowdsourcing worker, and for example, two crowdsourcing workers impersonate the person corresponding to each persona sentence and carry out a dialogue. The system collects the text of this dialogue as dialogue data. [Prior Art Documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2011-103063 [Non-patent literature]

[0007] [Non-Patent Document 1] Saizheng Zhang, et al., "Personalizing Dialogue Agents: I have a dog, do you have pets too?", "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics ", (Australia), Association for Computational Linguistics, July 2018, Volume 1: Long Papers, p.2204-2213 [Overview of the project] [Problems that the invention aims to solve]

[0008] Conventional methods for collecting dialogue data have made it difficult to develop natural dialogues, making it challenging to collect dialogue data through such conversations. For example, Patent Document 1 obtains question-and-answer pairs, making it difficult to collect dialogue data through more natural conversations consisting of multiple turns. Non-Patent Document 1 attempts to force the inclusion of persona information within the dialogue, making it difficult to establish a natural conversation.

[0009] This disclosure aims to provide an information processing device and information processing method that can easily collect dialogue data through more natural conversations. [Means for solving the problem]

[0010] The information processing device relating to this disclosure includes: a role presentation unit that presents a role to be assigned to a first user; an acquisition unit that acquires first data input by the first user in conjunction with the input of the first data; a detection unit that detects information inconsistent with the role contained in the first data acquired by the acquisition unit as inconsistent information; and an inconsistency presentation unit that presents a message to the first user based on the inconsistent information detected from the first data by the detection unit. [Brief explanation of the drawing]

[0011] [Figure 1] This is a schematic diagram illustrating the configuration of an example of a data collection system applicable to the embodiments of this disclosure. [Figure 2] This is a schematic diagram showing an example of a dialogue screen applicable to the embodiments of this disclosure. [Figure 3] This is an example functional block diagram illustrating the functions of a server according to the embodiments of this disclosure. [Figure 4] This block diagram shows an example hardware configuration of a server applicable to the embodiment. [Figure 5] This is an example of a functional block diagram illustrating the functions of a worker terminal according to an embodiment. [Figure 6] This is a schematic diagram showing an example of persona information presented to worker X by a display area in the dialogue data collection screen according to the embodiment. [Figure 7] This is a schematic diagram showing an example of a warning display for inconsistencies in the content of a spoken sentence, using the inconsistency display unit according to the embodiment. [Figure 8] This is a schematic diagram showing an example of a warning display for inconsistencies in the style of spoken sentences by the inconsistency display unit according to the embodiment. [Figure 9] This is a schematic diagram illustrating an example in which a set of persona information consisting of multiple patterns of persona information is presented according to the embodiment. [Figure 10] This is a schematic diagram illustrating an example of a warning display for a contradiction in the content of a spoken sentence in case (A) according to an embodiment. [Figure 11]It is a schematic diagram illustrating an example of a warning display for a contradiction related to the content of an utterance sentence in case (B) according to the embodiment. [Figure 12] It is a schematic diagram illustrating an example where a set of persona information based on multiple patterns of persona information is presented according to the embodiment. [Figure 13] It is a schematic diagram illustrating an example of a warning display for a contradiction related to the content of an utterance sentence in case (C) according to the embodiment. [Figure 14] It is a schematic diagram illustrating an example of a warning display for a contradiction related to the content of an utterance sentence in case (D) according to the embodiment. [Figure 15] It is an example flowchart illustrating contradiction detection processing according to the embodiment. [Figure 16] It is a schematic diagram illustrating results of contradiction determination according to the contents of an utterance sentence and persona information according to the embodiment. [Figure 17] It is a schematic diagram illustrating an example of a subjective evaluation input screen according to the embodiment. [Figure 18] It is a schematic diagram illustrating an example of persona information and dialogue data evaluated as having high dialogue quality by a persona combination evaluation unit according to the embodiment. [Figure 19] It is a schematic diagram illustrating an example of persona information and dialogue data evaluated as having low dialogue quality by a persona combination evaluation unit according to the embodiment. [Figure 20] It is a schematic diagram illustrating an example of a graph presented by the persona combination evaluation unit for presenting the basis for evaluation of a combination of persona information according to the embodiment. [Figure 21A] It is an example flowchart illustrating combination evaluation processing of persona information in an initial state according to the embodiment. [Figure 21B] It is an example flowchart illustrating combination evaluation processing of persona information when the system has been sufficiently operated according to the embodiment. [Figure 22] It is a schematic diagram illustrating an example of changes in persona information accompanying updating of weights for a combination of persona information according to the embodiment. [Figure 23A]This is a schematic diagram for more specifically illustrating the process according to the first modification of the embodiment. [Figure 23B] This is a schematic diagram for more specifically illustrating the process according to the first modification of the embodiment. [Figure 24] This is a flowchart illustrating an example of the inconsistency detection process related to the first modified embodiment. [Figure 25] This is a schematic diagram illustrating a first example of an image generation process relating to a second modified embodiment. [Figure 26] This is a schematic diagram illustrating a second example of image generation processing according to a second modified embodiment. [Figure 27] This is a schematic diagram illustrating a third example of image generation processing related to a second modified embodiment. [Modes for carrying out the invention]

[0012] The embodiments of this disclosure will be described in detail below with reference to the drawings. In the following embodiments, the same parts will be denoted by the same reference numerals, and redundant descriptions will be omitted.

[0013] The embodiments of this disclosure will be described below in the following order. 1. Outline of this disclosure 1-1. Outline configuration of the data collection system related to this disclosure 1-2. Comparison with existing technologies 2. Configurations applicable to embodiments of the present disclosure 3. Embodiments of the Disclosure 3-1. Inconsistency detection according to the embodiment 3-1-1. Warnings regarding inconsistencies in the content of spoken sentences 3-1-2. Warnings regarding inconsistencies in speech style. 3-1-3. Warning display for inconsistencies in the image content of spoken sentences. 3-1-4. Example of setting multiple persona information patterns 3-1-5. Flow of the inconsistency detection process 3-1-6. Relationship between contradiction detection results and spoken text and persona information 3-2. Dialogue evaluation according to the embodiment 3-2-1. Specific Examples of Dialogue Evaluation 3-2-2. Flowchart for evaluating combinations of persona information 3-2-3. Changes in Persona Information Due to Weight Updates 4. First Modification of an Embodiment of the Present Disclosure 5. Second Modification of an Embodiment of the Present Disclosure

[0014] [1. Summary of this disclosure] This disclosure relates to the collection of dialogue data in text format. Prior to describing this disclosure, a brief overview of the dialogue data collection system applicable to this disclosure will be provided for ease of understanding.

[0015] In a dialogue data collection system applicable to this disclosure, one user engaging in a dialogue acts out a persona (role) presented by the system, either an image or profile information. This user then engages in a dialogue using text data with another user, who acts out a different role, based on a scenario presented by the system. The dialogue data collection system collects the dialogue data from these two users as a dialogue log.

[0016] The dialogue logs collected in this way can be used, for example, as training data and evaluation data for training models used by artificial intelligence in automated dialogue systems. Furthermore, by widely recruiting users involved in dialogue activities using this dialogue data collection system, for example through the internet, it is possible to collect a large amount of dialogue data and dialogue logs relatively easily.

[0017] Furthermore, in order to recruit a large number of users to participate in such dialogue tasks, a process called crowdsourcing can be used, in which tasks are outsourced to an unspecified number of users via the internet. For example, the operator of a dialogue data collection system may refer to users who are outsourced with dialogue tasks as crowdworkers. In the following, unless otherwise specified, users involved in dialogue tasks by a dialogue data collection system will be referred to as "workers."

[0018] The dialogue data collection system described in this disclosure detects inconsistencies between the dialogue content and the role presented to the worker when dialogue data is input by the worker, and presents the detected inconsistencies to the worker. The worker can correct the dialogue data in response to the presentation of these inconsistencies when transmitting the dialogue data to the system. This improves the quality of the dialogue data and enables the collection of higher-quality dialogue logs. Therefore, it becomes easier to collect dialogue data from more natural conversations.

[0019] Furthermore, the data collection system described in this disclosure requests subjective evaluations of their own and their partner's roles from the worker who engaged in the dialogue, and also performs quantitative evaluations of the dialogue based on the dialogue data. Through these subjective and quantitative evaluations, for example, the combination of images and profile information in the roles is evaluated, and these combinations of images and profile texts are improved. This makes it possible to update the roles presented to workers to make dialogue easier, thereby improving the quality of the dialogue data. Consequently, it becomes possible to easily collect dialogue data from more natural conversations.

[0020] (1-1. Outline configuration of the data collection system related to this disclosure) Next, the general configuration of the data collection system related to this disclosure will be explained using Figures 1 and 2. In the following, for example, it will be assumed that the operator of the data collection system employs two workers, Worker X and Worker Y, as workers who conduct conversations to collect dialogue data using the data collection system.

[0021] Figure 1 is a schematic diagram illustrating the configuration of an example of a dialogue data collection system applicable to embodiments of this disclosure. In Figure 1, the dialogue data collection system 1 includes a server 10, which is connected to a worker terminal 20a used by worker X and a worker terminal 20b used by worker Y via a network 2 such as the Internet.

[0022] Worker terminals 20a and 20b can be information processing devices such as personal computers, tablet computers, and smartphones. Furthermore, although server 10 is shown in Figure 1 as being composed of a single information processing device, this is not an exhaustive example. For example, server 10 may be configured in a distributed manner with multiple information processing devices connected to each other via a cloud network or the like.

[0023] In the data collection system, server 10 displays a dialogue screen for worker terminals 20a and 20b to facilitate interaction. Here, the dialogue data collection system 1 sets a specific role for worker X that is the target of dialogue log collection, and sets an arbitrary role for worker Y, who interacts with worker X. Furthermore, the dialogue data collection system 1 sets the situation in which worker X and worker Y interact.

[0024] Figure 2 is a schematic diagram showing an example of a dialogue screen applicable to embodiments of the present disclosure. Figure 2 shows an example of a dialogue data collection screen 30 displayed on a worker terminal 20a used by worker X. In the example of Figure 2, the dialogue data collection screen 30 includes display areas 31-33 and 36, an input area 34, and a send button 35.

[0025] Display area 31 displays persona information indicating the role assigned to worker X. Here, "persona" is a concept equivalent to a person's profile in general terms. More specifically, the persona information of a person (role) includes objective matters such as gender, age, and appearance, as well as subjective matters concerning that person (role), such as hobbies, preferences, way of thinking, and experiences. In other words, persona information can be said to be attribute information that indicates the attributes of that person (role).

[0026] In the example shown in Figure 2, the display area 31 shows a persona image 310 representing the appearance of a person, and profile information 311 that describes the person's profile in text, as persona information set for worker X. In other words, the persona image 310 is an image used to represent the person or character shown in the persona information that includes the persona image 310. In this example, the persona image 310 shows a woman holding a microphone and wearing a hair accessory. For example, an illustration may be used for the persona image 310. However, it is not limited to this, and a photograph may also be used as the persona image 310. The dialogue data collection system 1 can select the persona image 310 to present to worker X from, for example, a plurality of pre-prepared images in a predetermined manner.

[0027] In the example in Figure 2, the profile information 311 includes five items, from #1 to #5. In this example, items #1 to #4 are objective details about the person included in the persona image 310. Specifically, item #1 is the gender (female), status (high school sophomore), and age (17 years old) assigned to the person. Item #2 is the speech style assigned to the person (speaks in a feminine manner). Item #3 is the occupation of the person (works as an idol). Note that "occupation" here is not limited to activities that provide a livelihood, but may also include activities that the person is said to be focusing on. Item #4 is the family structure assigned to the person (has an older brother).

[0028] On the other hand, item #5 is defined as a subjective item associated with the person in question. Specifically, item #5 is defined as the person's hobby (cooking is their hobby). The dialogue data collection system 1 can select the profile information 311 to present to worker X from, for example, a set of pre-prepared profile information in a predetermined manner.

[0029] In the following, the persona image 310 and profile information 311 will sometimes be collectively referred to as "persona information."

[0030] Display area 32 displays information indicating the situation of the dialogue taking place on the dialogue data collection screen 30. This situation information is set, for example, by the dialogue data collection system 1 and displayed in display area 32. In the example shown in the figure, the information indicating the situation is displayed in display area 32 as "They are talking about each other's occupations." This instructs workers X and Y to have a dialogue about each other between the people they have each been assigned roles to.

[0031] Furthermore, the dialogue data collection system 1 displays the same screen on worker terminal 20b as the dialogue data collection screen 30 displayed on worker terminal 20a of worker X. Specifically, on the dialogue data collection screen 30 displayed on worker terminal 20b, the persona image 310 and profile information 311 are images and profiles representing the person assigned as a role for worker Y, respectively. Also, on worker terminal 20b, the same situation as the one presented to worker X is displayed in the display area 32.

[0032] The display area 33 displays the dialogue data of worker X and worker Y in chronological order. In the example in Figure 2, time progresses from top to bottom in the display area 33, with the dialogue data at the bottom being the most recent data. The display area 33 is also displayed in common on worker terminals 20a and 20b.

[0033] Specifically, when worker X inputs a dialogue utterance into the input area 34 and the send button 35 is operated, the utterance is posted to the dialogue data collection system 1 and displayed in the display area 33 as dialogue data 330a (shown enclosed in a solid line) by worker X. The dialogue data collection system 1 registers the posted dialogue data 330a as a dialogue log of the dialogue by worker X and collects the dialogue log.

[0034] Furthermore, the utterance posted by worker Y is displayed in the display area 33 as dialogue data 330b by worker Y (shown enclosed by a dotted line). The dialogue data collection system 1 can register the posted dialogue data 330b as a dialogue log of the dialogue by worker Y. Note that if a specific role for which dialogue log collection is to be targeted is set for worker X, the dialogue data collection system 1 may omit the registration of dialogue data 330b by worker Y.

[0035] In the display area 33, an icon 331a representing a person assigned to worker X is displayed adjacent to the dialogue data 330a, indicating that the dialogue data 330a consists of utterances posted by worker X. Similarly, an icon 331b representing a person assigned to worker Y is displayed adjacent to the dialogue data 330b, indicating that the dialogue data 330b consists of utterances posted by worker X. These icons 331a and 331b may be created, for example, based on the persona images 310 of worker X and worker Y, respectively. Furthermore, worker X and worker Y can learn, for example, the appearance of the persona image 310 of their conversation partner from these icons 331a and 331b.

[0036] A dialogue is formed between worker X and worker Y as worker X and worker Y take turns posting utterances. Therefore, in the display area 33, dialogue data 330a and dialogue data 330b are displayed alternately in chronological order.

[0037] It is possible for either Worker X or Worker Y to post an utterance without waiting for a response from the other party. In such cases, the two utterances, which do not have a chronological response in between, will be treated as a single utterance. Furthermore, Worker X's post and Worker Y's post may be made immediately in response to one post, as in a normal conversation, or there may be a certain amount of time elapsed between the two posts.

[0038] The display area 36 displays system messages from the dialogue data collection system 1. For example, if the dialogue data collection system 1 detects a contradiction between the content of the text entered in the input area 34 and the profile information 311, it will display information indicating the detected contradiction in the display area 36. The dialogue data collection system 1 can also generate a suggested correction that can resolve the detected contradiction from the text entered in the input area 34 and display it in the display area 36.

[0039] When a dialogue between worker X and worker Y is completed, the dialogue data collection system 1 requests subjective evaluations from, for example, worker X and worker Y, or from worker X, regarding whether the responses were easy to follow according to the assigned roles, and whether the other party's responses made the dialogue easy to follow. The dialogue data collection system 1 also performs a quantitative evaluation of the dialogue based on the collected dialogue logs and dialogue data. Based on these subjective and quantitative evaluations, the dialogue data collection system 1 evaluates, for example, the combination of persona image 310 and profile information 311 in the roles, and improves these combinations of persona image 310 and profile information 311.

[0040] (1-2. Comparison with existing technologies) As described above, the dialogue data collection system 1 according to the disclosed embodiment displays information indicating inconsistencies with the profile information 311, detected from the content of the text entered in the input area 34, on the dialogue data collection screen 30. Furthermore, after the dialogue ends, the dialogue data collection system 1 improves the combination of the persona image 310 and the profile information 311 based on subjective evaluations obtained from the worker and quantitative evaluations based on the dialogue data. Therefore, by applying the dialogue data collection system 1 according to the disclosed embodiment, the quality of the dialogue data can be improved, and more natural and high-quality dialogue logs can be collected.

[0041] In contrast, Patent Document 1 describes a system in which the system presents a specific responding entity. For example, a first user posts a question to the responding entity into the system, and a second user posts the responding entity's answer to that question, pretending to be the responding entity. In this way, the system collects dialogue data between any person and the responding entity.

[0042] However, Patent Document 1 obtains question-and-answer pairs, making it difficult to collect dialogue data through natural conversations consisting of multiple turns. Furthermore, in Patent Document 1, if the system presents the responding party, only those who know the target person as the responding party in advance can become workers, severely limiting the number of users who can be employed as workers.

[0043] The dialogue data collection system 1 according to the embodiment of this disclosure detects inconsistencies with the profile information of utterances input as dialogue and presents suggested corrections based on the detected inconsistencies. Therefore, according to the dialogue data collection system 1 according to the embodiment of this disclosure, it is possible to easily collect dialogue data from natural conversations consisting of multiple turns. Furthermore, since the person assigned as a role to the worker is presented to the worker using persona information that uses images and profile information, it is possible to recruit users who do not know the person in question beforehand as workers.

[0044] On the other hand, Non-Patent Document 1 presents each worker with a persona consisting of five sentences that serve as an introduction to the target person, and two workers, for example, engage in a dialogue, each taking on the role of the person described in their respective persona. The system collects the text of this dialogue as dialogue data.

[0045] However, in Non-Patent Literature 1, the attempt to force persona information into the dialogue makes it difficult to create a natural conversation. For example, in Non-Patent Literature 1, the image of the character is expanded upon based on a short persona statement of about five sentences, which is generally difficult, and no scene setting is provided. Therefore, in Non-Patent Literature 1, it is difficult for the worker to come up with topics other than those disclosed in the persona statement.

[0046] Furthermore, Non-Patent Document 1 does not consider the quality of the combination of persona information presented in the persona statement. For example, a persona statement may contain information that makes it difficult to understand the person's attributes and preferences (e.g., "I own a XX car from company XX," or "I own 1000 hats"). In such cases, the combination of persona statements becomes unnatural, making it difficult to develop an image of the person described by the persona statement.

[0047] The dialogue data collection system 1 according to the embodiment of this disclosure presents persona information, including images and profile information, to the worker, making it easy to develop an image of the person. Furthermore, it improves the combination of persona information through subjective and quantitative evaluations. Therefore, the dialogue data collection system 1 according to the embodiment of this disclosure can easily collect dialogue data from natural conversations consisting of multiple turns.

[0048] [2. Configurations applicable to embodiments of this disclosure] Next, configurations applicable to embodiments of this disclosure will be described. Figure 3 is an example functional block diagram illustrating the functions of server 10 according to an embodiment of this disclosure.

[0049] In Figure 3, the server 10 includes a dialogue data collection screen generation unit 100, a persona information DB (database) 101, a dialogue data DB 102, a contradiction detection unit 103, a persona image generation unit 104, an automatic dialogue evaluation unit 105, and a persona combination evaluation unit 106.

[0050] Of these, the dialogue data collection screen generation unit 100, the inconsistency detection unit 103, the persona image generation unit 104, the automatic dialogue evaluation unit 105, and the persona combination evaluation unit 106 are configured by having an information processing program according to the embodiment executed on the CPU (Central Processing Unit) of the server 10. However, some or all of the dialogue data collection screen generation unit 100, the inconsistency detection unit 103, the persona image generation unit 104, the automatic dialogue evaluation unit 105, and the persona combination evaluation unit 106 may be configured by hardware circuits that work together. Furthermore, the persona information DB 101 and the dialogue data DB are configured in a predetermined storage area of ​​a storage device owned by or connected to the server 10.

[0051] The dialogue data collection screen generation unit 100 includes a contradiction display unit 110, a persona information display unit 111, a dialogue collection unit 112, and a subjective evaluation collection unit 113. The dialogue data collection screen generation unit 100 generates display control information for displaying the dialogue data collection screen 30 shown in Figure 2, according to the processing performed by each unit included in the dialogue data collection screen generation unit 100, and transmits the generated display control information to worker terminals 20a and 20b. Worker terminals 20a and 20b display the dialogue data collection screen 30 on their displays according to the display control information transmitted from the dialogue data collection screen generation unit 100.

[0052] In the following, unless otherwise specified, the process by which the dialogue data collection screen generation unit 100 generates display control information for displaying the dialogue data collection screen 30 and transmits it to the worker terminals 20a and 20b to display the dialogue data collection screen 30 will be described as a process performed by each part of the dialogue data collection screen generation unit 100. Furthermore, in the following, it will be assumed that the person (role) to be the target of dialogue data collection will be set for worker X using the worker terminal 20a, and each part of the server 10 will execute processing on the worker terminal 20a.

[0053] The dialogue data collection screen generation unit 100 includes a contradiction display unit 110, a persona information display unit 111, a dialogue collection unit 112, and a subjective evaluation collection unit 113.

[0054] Persona Information DB101 stores multiple persona images 310 and multiple profile information 311 as persona information. Profile information may be stored as a single profile information 311 with multiple items grouped together, or each item may be stored separately.

[0055] The persona information display unit 111 retrieves persona information from the persona information database 101 and displays it in the display area 31 of the dialogue data collection screen 30. The persona information display unit 111 may randomly retrieve persona images 310 and profile information 311, each of which is stored in multiple quantities in the persona information database 101, or it may retrieve them according to a predetermined combination of persona images 310 and profile information 311. The persona information display unit 111 presents the role assigned to the first user to the first user, and presents the role assigned to the second user to the second user. In other words, the persona information display unit 111 functions as a role presentation unit that presents the role assigned to the first user to the first user.

[0056] The dialogue collection unit 112 acquires the utterances entered into the input area 34 of the dialogue data collection screen 30 as they are entered. In other words, the dialogue collection unit 112 functions as an acquisition unit that acquires the first data entered by the first user as it is entered. The dialogue collection unit 112 also acquires dialogue data from each utterance posted from worker terminals 20a and 20b in response to the operation of the send button 35 on the dialogue data collection screen 30.

[0057] The dialogue collection unit 112 collects dialogue logs by registering the acquired dialogue data in the dialogue data database 102. At this time, the dialogue collection unit 112 can associate identification information such as time information and serial numbers with the dialogue data 330a and 330b. The dialogue collection unit 112 can register only the dialogue data transmitted from worker terminal 20a by worker X, who is set as the person (role) to be collected, in the dialogue data database 102, and omit the registration of dialogue data transmitted from worker terminal 20b by worker Y in the dialogue data database 102.

[0058] Furthermore, the dialogue collection unit 112 transmits the acquired dialogue data from worker X and worker Y to worker terminals 20a and 20b, and displays the text from the dialogue data from worker X and worker Y in chronological order on the display area 33 of the dialogue data collection screen 30 of worker terminals 20a and 20b.

[0059] The inconsistency detection unit 103 detects inconsistencies between the dialogue data from worker X acquired by the dialogue collection unit 112 and the persona information displayed by the persona information display unit 111 acquired from the persona information DB 101, and passes information indicating the detected inconsistencies to the inconsistency display unit 110. The inconsistency display unit 110 transmits the information indicating the inconsistencies received from the inconsistency detection unit 103 to the worker terminal 20a used by worker X, and displays the information indicating the inconsistencies in the display area 36 of the dialogue data collection screen 30 of the worker terminal 20a.

[0060] In other words, the inconsistency detection unit 103 functions as a detection unit that detects information in the first data acquired by the acquisition unit that contradicts the role, as inconsistent information. The inconsistency display unit 110 functions as an inconsistency display unit that presents a message to the first user based on the inconsistent information detected from the first data by the detection unit.

[0061] Furthermore, the contradiction display unit 110 can generate proposed corrections to contradictions based on the information indicating the contradiction and the persona information displayed by the persona information display unit 111 obtained from the persona information DB 101. The contradiction display unit 110 can transmit the generated proposed corrections to the worker terminal 20a, and the proposed corrections can be displayed in the display area 36 of the dialogue data collection screen 30 of the worker terminal 20a.

[0062] The subjective evaluation collection unit 113 generates display control information for displaying a subjective evaluation input screen (described later) for worker X to input subjective evaluations. The subjective evaluation collection unit 113 transmits the generated display control information to the worker terminal 20a used by worker X, causing the worker terminal 20a to display the subjective evaluation input screen. The worker terminal 20a transmits information indicating the subjective evaluations entered in accordance with the subjective evaluation input screen to the server 10. In the server 10, the subjective evaluation collection unit 113 collects the information indicating the subjective evaluations transmitted from the worker terminal 20a.

[0063] The automated dialogue evaluation unit 105 automatically performs a quantitative evaluation of the dialogue regarding "naturalness of conversation" using a predetermined evaluation method based on the dialogue log collected by the dialogue collection unit 112. The automated dialogue evaluation unit 105 can apply known evaluation methods that utilize evaluation indicators in general dialogue systems and persona dialogue systems, such as Distinct-1 / 2, MaxBLEU, Persona-Recall / Predict / F1, and Persona-Cover, as the dialogue evaluation method, or it can apply its own evaluation method.

[0064] The persona combination evaluation unit 106 evaluates combinations of persona information based on at least one of the information indicating subjective evaluations by worker X collected by the subjective evaluation collection unit 113 and the information indicating quantitative evaluations automatically evaluated by the dialogue automatic evaluation unit 105. Here, combinations of persona information include, for example, combinations of persona images 310 and profile information 311 included in the persona information, or combinations of each item in the profile information 311.

[0065] Furthermore, the subjective evaluation collection unit 113 may acquire the subjective evaluation of worker Y in the same manner as the collection of the subjective evaluation of worker X. The persona combination evaluation unit 106 can evaluate the combination of persona information based on at least one of the subjective evaluation of worker X and the subjective evaluation of worker Y.

[0066] The persona combination evaluation unit 106 adjusts the weights of the persona information combinations based on the evaluation results so that combinations that are difficult for worker X to work with are less likely to be selected. The persona combination evaluation unit 106 reflects the weights of the persona information combinations in the persona information stored in the persona information DB 101. In other words, the persona combination evaluation unit 106 functions as an evaluation unit that evaluates the combination of information contained in the persona information based on the first data and the second data entered by the second user.

[0067] The persona image generation unit 104 generates a persona image 310 in the persona information and stores the generated persona image 310 in the persona information DB 101. The persona image generation unit 104 can, for example, generate a new persona image 310 based on profile information 311 included in the persona information. Alternatively, the persona image generation unit 104 can, for example, generate a new persona image 310 based on a persona image 310 stored in the persona information DB 101. The persona image generation unit 104 can apply, for example, GAN (Generative Adversarial Networks) as a method for generating a new persona image 310. However, it is not limited to this, and the persona image generation unit 104 can also use an image selected from the persona images 310 stored in the persona information DB 101 as a new persona image 310. In other words, the persona image generation unit 104 functions as an image generation unit that generates new image information based on a combination of image information and attribute information, or a combination of information contained in the attribute information.

[0068] Figure 4 is a block diagram showing an example of the hardware configuration of a server 10 applicable to the embodiment. In Figure 4, the server 10 includes a CPU 1000, a ROM (Read Only Memory) 1001, a RAM (Random Access Memory) 1002, a storage device 1003, a data I / F (interface) 1004, and a communication I / F 1005, all of which are communicated together by a bus 1010.

[0069] The storage device 1003 is a non-volatile storage medium such as a hard disk drive or flash memory. The storage device 1003 is not limited to being built into the server 10, but may be externally connected to the server 10, for example.

[0070] The CPU 1000 controls the operation of the server 10 using the RAM 1002 as work memory, according to the programs stored in the storage device 1003 and the ROM 1001.

[0071] Data I / F 1004 is an interface for sending and receiving data between server 10 and external devices. Data I / F 1004 can use wired communication interfaces such as USB (Universal Serial Bus) or wireless communication interfaces such as Bluetooth (registered trademark). Communication I / F 1005 is an interface for communication via network 2.

[0072] Furthermore, the server 10 can be equipped with input devices such as a keyboard, and display devices such as a display control unit and a display, as needed.

[0073] In the server 10, the CPU 1000 executes an information processing program for realizing the dialogue data collection system 1 according to the embodiment, thereby configuring the above-mentioned dialogue data collection screen generation unit 100, inconsistency detection unit 103, persona image generation unit 104, dialogue automatic evaluation unit 105, and persona combination evaluation unit 106 as modules, for example, on the main memory area of ​​the RAM 1002.

[0074] The information processing program can be obtained from an external source (e.g., another server device not shown) via a network (not shown) through communication via the communication I / F 1005, for example, and installed on the server 10. However, the information processing program may also be provided stored on a removable storage medium such as a CD (Compact Disk), DVD (Digital Versatile Disk), or USB (Universal Serial Bus) memory.

[0075] Figure 5 is an example functional block diagram illustrating the functions of worker terminals 20a and 20b according to the embodiment. Since the functions of worker terminals 20a and 20b according to the embodiment are common, in Figure 5, worker terminals 20a and 20b are shown together as worker terminal 20.

[0076] The worker terminal 20 can be configured as a typical computer, including, for example, a CPU, ROM, RAM, storage device, data interface, communication interface, input device, and display device. Therefore, a detailed explanation of the hardware configuration is omitted here.

[0077] The worker terminal 20 includes a control unit 200, a communication unit 201, an input unit 202, and a display unit 203. These control unit 200, communication unit 201, input unit 202, and display unit 203 are configured, for example, by a predetermined program running on a CPU. However, the configuration is not limited to this; some or all of the control unit 200, communication unit 201, input unit 202, and display unit 203 may be configured by hardware circuits that work together.

[0078] The control unit 200 controls the overall operation of the worker terminal 20. The communication unit 201 performs communication via the network 2. The input unit 202 accepts input operations from the user (worker) to the input device. The display unit 203 controls the display of the screen on the display device. More specifically, the display unit 203 displays the dialogue data collection screen 30 on the display device according to display control information transmitted from the server 10, for example. The input unit 202 also accepts input of character information (speech) to the input area 34 in the dialogue data collection screen 30 displayed on the display device by the display unit 203, for example.

[0079] Of these, the input unit 202 and the display unit 203 can be implemented using the functions of a general-purpose browser application 210. For example, the server 10 sends a program to the worker terminal 20 to implement the input unit 202 and the display unit 203 on the browser application 210. On the worker terminal 20, the browser application 210 configures the input unit 202 and the display unit 203 on the worker terminal 20 according to the program sent from the server 10. However, the input unit 202 and the display unit 203 may also be implemented using a program specifically for this dialogue data collection system 1.

[0080] [3. Embodiments of the Disclosure] Next, embodiments of this disclosure will be described. In the following, when worker X and worker Y engage in dialogue, a specific person (role) to be subject to dialogue log collection is set for worker X, and an arbitrary person (role) is set for worker Y who engages in dialogue with worker X.

[0081] (3.1. Inconsistency detection according to the embodiment) First, let's explain the inconsistency detection according to the embodiment. Specifically, the dialogue data collection system 1 according to the embodiment detects inconsistencies between the persona information indicating the role set for worker X and the dialogue data input by worker X.

[0082] Figure 6 is a schematic diagram showing an example of persona information presented to worker X by the display area 31 in the dialogue data collection screen 30 according to the embodiment. The example in Figure 6 corresponds to the example in the display area 31 shown in Figure 2 above, and the persona image 310a is an illustration of a woman holding a microphone and wearing a hair accessory. The profile information 311 includes five items, items #1 to #5. The contents of items #1 to #5 are the same as those explained using Figure 2. Item #1 is the gender (female), status (high school sophomore), and age (17 years old) set for the person. Item #2 is the speaking style set for the person (speaks in a feminine way). Item #3 is the occupation of the person (works as an idol). Item #4 is the family structure set for the person (has an older brother). Item #5 is the hobby set for the person (cooks as a hobby).

[0083] In the initial stages of dialogue log collection, the dialogue data collection system 1 can randomly determine combinations of persona information, such as a combination of persona image 310a and profile information 311. However, the dialogue data collection system 1 may also use persona information with predetermined combinations in the initial stages. As dialogue log collection progresses and dialogue logs are accumulated in the dialogue data DB 102, the dialogue data collection system 1 can determine the persona information to present to worker X, taking into account the weights of the persona information combinations adjusted by the persona combination evaluation unit 106.

[0084] Worker X interacts with Worker Y using Worker Terminal 20a via the Dialogue Data Collection Screen 30. More specifically, Worker X inputs utterances as text data into the input area 34 on the Dialogue Data Collection Screen 30. Once Worker X has confirmed the input content for the input area 34, Worker X operates the send button 35. In response to this operation of the send button 35, the utterances as text data entered in the input area 34 are sent from Worker Terminal 20a to Server 10. On Server 10, the Dialogue Collection Unit 112 stores the utterances as text data sent from Worker Terminal 20a as dialogue data 330a in the Dialogue Data DB 102 and collects the dialogue log.

[0085] In this embodiment, the interaction between worker X and worker Y is limited to dialogue using text data, and direct dialogue using voice or other means is not supported. However, it is possible to convert the dialogue content input by voice into text data using speech recognition processing or the like and send it to server 10.

[0086] Worker X can input a response utterance to the dialogue data 330b from Worker Y, or input the first utterance of the dialogue. Worker terminal 20a sends the utterance entered in the input area 34 to server 10 in conjunction with that input. The utterance sent here is sent without any operation being performed on the send button 35.

[0087] As an example, in the worker terminal 20a, the input unit 202 can sequentially transmit information indicating an input operation to the input area 34 and the text data entered into the input area 34 to the server 10 via the communication unit 201. In the server 10, the dialogue collection unit 112 sequentially constructs a spoken sentence based on the information indicating the input operation and the text data transmitted sequentially from the worker terminal 20, and passes it to the contradiction detection unit 103.

[0088] On server 10, the inconsistency detection unit 103 detects inconsistencies between the spoken text sent from the worker terminal 20 and the persona information displayed by the persona information display unit 111. The inconsistency detection unit 103 passes information indicating the detected inconsistencies to the inconsistency display unit 110.

[0089] The contradiction display unit 110 can present a warning to worker X based on the contradiction-indicating information received from the contradiction detection unit 103. In this warning, the contradiction display unit 110 includes information indicating the location of the contradiction in the utterance, along with the persona information of the target that forms the basis of the contradiction. Furthermore, the contradiction display unit 110 can present proposed revisions to the utterance to resolve the contradiction.

[0090] (3-1-1. Warning display regarding inconsistencies in the content of spoken sentences) Figure 7 is a schematic diagram showing an example of a warning display for a contradiction in the content of a spoken sentence by the contradiction display unit 110 according to the embodiment. Section (a) of Figure 7 shows an example of a spoken sentence entered into the input area 34 by worker X according to the persona information and situation shown in Figure 6. In this example, the spoken sentence is "I'm a third-year high school student," and the part "third-year high school student" is a contradiction point 340 that contradicts "second-year high school student" in item #1 of profile information 311 in the persona information. The contradiction display unit 110 highlights the display of this contradiction point 340 in the input area 34 (bold and underlined in the example shown in the figure) to present a warning about the existence of the contradiction.

[0091] The contradiction display unit 110 may also tag the detected contradiction in the utterance with the classification class to which the detected contradiction belongs.

[0092] Section (b) of Figure 7 shows an example of the presentation of the basis for the contradiction and proposed correction regarding the contradiction location 340 shown in Section (a) of Figure 7 by the contradiction display unit 110. The contradiction display unit 110 displays the basis for the contradiction 351 and the proposed correction 352 in the display area 36. In the example shown in the figure, the contradiction display unit 110 highlights the description of "second-year high school student" in item #1 of the profile information 311, which corresponds to the contradiction shown in the contradiction location 340, in area 36a of the display area 36, ​​as the basis for the contradiction 351.

[0093] Furthermore, the contradiction display unit 110 generates a revised version 352 for the contradiction shown in the contradictory location 340. The contradiction display unit 110 can generate a revised version 352 ("second-year high school student") based on a predetermined rule, for example. In the example shown in the figure, for example, a rule could be to replace the description of the contradictory location 340 with the description presented as the basis for the contradiction 351. However, the contradiction display unit 110 may also convert the description of the contradictory location 340 into the description of the revised version 352 based on a model that has been predeterminedly learned by machine learning. In the example shown in the figure, the contradiction display unit 110 displays the generated revised version 352 in area 36b, directly below area 36a, in the display area 36.

[0094] Worker X can modify the utterance entered in the input area 34 according to the suggested correction 352 displayed in the display area 36 (area 36b). After the utterance entered in the input area 34 has been modified by Worker X according to the suggested correction 352, the send button 35 can be operated by worker terminal 20a to send the modified utterance according to the suggested correction 352 to server 10.

[0095] In server 10, the dialogue collection unit 112 stores the modified utterance sent from worker terminal 20a as dialogue data 330a by worker X in the dialogue data DB 102 and collects the dialogue log. The dialogue collection unit 112 also displays the dialogue data 330a with the modified utterance in the display area 33.

[0096] Furthermore, the worker terminal 20a can also send the uncorrected (including contradictory) utterance to the server 10 by operating the send button 35 without correcting the utterance entered in the input area 34. In addition, the dialogue collection unit 112 can display, for example, a message in the display area 36 if the utterance entered in the input area 34 contains prohibited words. Not limited to this, the dialogue collection unit 112 can also prevent the utterance from being sent to the server 10 by disabling the function of the send button 35 if the utterance contains prohibited words. Prohibited words are set in advance by, for example, the operator or administrator of the dialogue data collection system 1.

[0097] (3-1-2. Warning about inconsistencies in speech style) In the example in Figure 7, the inconsistency detection unit 103 detects inconsistencies related to the content of the utterance. The inconsistency detection unit 103 can detect inconsistencies not only in the content of the utterance but also in the style of the utterance. For example, the inconsistency detection unit 103 detects inconsistencies related to style when the style of the utterance is explicitly specified in the profile information 311. The style of the utterance refers to things like first-person pronouns and sentence endings.

[0098] In the example of profile information 311 in Figure 6, item #2 states "Speaks in a feminine manner," specifying that the person described in this persona information uses what is commonly known as feminine language in conversation. Since this style specification may change depending on the relationship with the person being spoken to, multiple style specifications may be described in a single profile information 311.

[0099] Figure 8 is a schematic diagram showing an example of a warning display for a contradiction in the style of an utterance by the contradiction display unit 110 according to the embodiment. Section (a) of Figure 8 shows an example of an utterance entered into the input area 34 by worker X according to the persona information and situation shown in Figure 6. In this example, the utterance is "I'm a third-year high school student," and the first-person pronoun "ore" and the ending "da" are contradictory points 341a and 341b, which contradict item #2 of profile information 311 in the persona information, "speaks in a feminine manner." The contradiction display unit 110 highlights the display of contradictory points 341a and 341b in the input area 34 (bold and underlined in the example shown in the figure) to present a warning about the existence of the contradiction.

[0100] Section (b) of Figure 8 shows an example of the presentation of the basis for the contradiction and proposed corrections for the contradictions 341a and 341b shown in Section (a) of Figure 8 by the contradiction display unit 110. The contradiction display unit 110 displays the basis for the contradiction 353 and proposed corrections 354a and 354b in the display area 36. In the example shown in the figure, the contradiction display unit 110 highlights the description "speaks in a feminine manner" in item #2 of the profile information 311, which corresponds to the contradiction shown in the contradictions 341a and 341b, in area 36a of the display area 36 as the basis for the contradiction 353.

[0101] Furthermore, the inconsistency display unit 110 generates corrected versions 354a and 354b for the inconsistency shown at the inconsistency location 340, based on predetermined rules or a model learned by machine learning. In the example shown in the figure, the inconsistency display unit 110 displays the generated corrected versions 354a and 354b in the display area 36, ​​in area 36b, directly below area 36a. In this case as well, as described above, worker X can modify the utterance entered in the input area 34 according to the corrected versions 354a and 354b displayed in the display area 36 (area 36b) and send it to the server 10.

[0102] (3-1-3. Warning display regarding inconsistencies in the image content of spoken sentences) The inconsistency detection unit 103 can detect inconsistencies not only in the profile information 311 in the persona information, but also in the spoken sentence with respect to the persona image 310a. If metadata about the persona image 310a is associated with the persona image 310a, the inconsistency detection unit 103 can detect inconsistencies in the spoken sentence with respect to that metadata. Not limited to this, the inconsistency detection unit 103 can also perform image recognition on the persona image 310a and detect inconsistencies in the spoken sentence based on the results of the image recognition.

[0103] For example, the contradiction detection unit 103 detects a contradiction between the persona image 310a and the spoken statement if the persona image 310a shows a person with black hair, but the spoken statement is "My hair is brown." The contradiction display unit 110 displays a warning in the display area 36 of the dialogue data collection screen 30 indicating that the spoken statement is inconsistent with the persona image 310a, according to the result of this contradiction detection. The method for performing image recognition on the persona image 310a displayed as persona information is not particularly limited.

[0104] Furthermore, the inconsistency display unit 110 can generate suggested corrections for detected inconsistencies based on predetermined rules or models learned through machine learning, in the same manner as described above. In this case as well, the worker X can modify the utterance entered in the input area 34 according to the suggested corrections displayed in the display area 36 and send it to the server 10.

[0105] (3-1-4. Example of setting multiple persona information patterns) Next, we will explain a case where multiple patterns of persona information are set as sets of persona information for a single character (let's call him Person A) played by Worker X. For example, different persona information may be associated with Person A depending on the situation (e.g., workplace and home).

[0106] Figure 9 is a schematic diagram illustrating an example in which a set of persona information consisting of multiple patterns of persona information is presented according to the embodiment. Section (a) of Figure 9 shows how two patterns of persona information, consisting of a first persona information and a second persona information, are displayed in the display area 31a of the dialogue data collection screen 30. The first persona information is a combination of a persona image 310a and profile information 311a. The second persona information is a combination of a persona image 310b and profile information 311b.

[0107] The first persona information is set for person A "at work," and persona image 310a is an illustration of a woman holding a microphone and wearing a hair accessory, similar to Figure 6. Profile information 311a includes five items, from #1 to #5. Item #1 is the gender (female), status (high school sophomore), and age (17 years old) set for the person. Item #2 is the speaking style set for the person (speaks in a feminine manner). Item #3 is the occupation of the person (works as an idol). Item #4 is the preference set for the person (likes the color pink). Item #5 is the special skill set for the person (good at dancing and talking).

[0108] The second persona information is set for person A "when at school," and persona image 310b is an illustration of a woman wearing a high school uniform and no hair accessories. Profile information 311b includes five items, from #1 to #5. Item #1 is the gender (female), status (second year of high school), and age (17 years old) set for the person. Item #2 is the speech style set for the person (speaks in a feminine manner). Item #3 is the occupation of the person (works as an idol). Item #4 is the family structure set for the person (has an older brother). Item #5 is the hobby set for the person (enjoys cooking).

[0109] For a dialogue with person A, for whom the first and second persona information are set, a scenario is presented. In this example, as shown in section (b) of Figure 9, the scenario presented is "talking about family with friends at school." Worker X selects the appropriate persona information from the first and second persona information according to this scenario and engages in dialogue with worker Y.

[0110] In this case, a warning about a contradiction is issued if worker X's utterance falls into any of the following four cases (A) to (D). (A) This contradicts the persona information common to all patterns. (B) The presented situation contradicts the selected persona information. (C) When the presented situation does not depend on a specific pattern, persona information for a certain pattern is mentioned. In this case, this fact is notified and confirmation is requested. (D) Persona information that is independent of each other and presents in different patterns is mixed together.

[0111] Figure 10 is a schematic diagram showing an example of a warning display for a contradiction in the content of the utterance in case (A) described above, according to the embodiment. It is assumed that worker X is presented with the first and second persona information shown in section (a) of Figure 9 and the situation shown in section (b) of Figure 9.

[0112] Section (a) of Figure 10 shows an example of a speech sentence entered into the input area 34 by worker X. In this example, the speech sentence is "I am 18 years old," and the "18 years old" part is a contradiction 342 that contradicts "17 years old" in item #1 of the profile information 311a and 311b in the first and second persona information. The description of "17 years old" in item #1 is persona information common to both profile information 311a and 311b. The contradiction display unit 110 highlights the display of the contradiction 342 in the input area 34 (bold and underlined in the example in the figure) to warn about the existence of the contradiction.

[0113] Section (b) of Figure 10 shows an example of the presentation of the basis for the contradiction and proposed correction regarding the contradiction location 342 shown in Section (a) of Figure 10 by the contradiction display unit 110. The contradiction display unit 110 displays the basis for the contradiction 355 and the proposed correction 356 in the display area 36. In the example of Figure 10, the contradiction display unit 110 highlights the description of "17 years old" in item #1 of the profile information 311, which corresponds to the contradiction shown in the contradiction location 342, in area 36a of the display area 36 as the basis for the contradiction 355.

[0114] Furthermore, the contradiction display unit 110 generates a revised version 356 ("17 years old") for the contradiction shown at the contradictory location 342. In the example shown in the figure, the contradiction display unit 110 displays the generated revised version 356 in area 36b, directly below area 36a, in the display area 36.

[0115] Figure 11 is a schematic diagram showing an example of a warning display for a contradiction in the content of the utterance in case (B) described above, according to the embodiment. It is assumed that worker X is presented with the first and second persona information shown in section (a) of Figure 9 and the situation shown in section (b) of Figure 9.

[0116] Section (a) of Figure 11 shows an example of a spoken sentence entered into input area 34 by worker X. In this example, the spoken sentence is "I'm good at dancing and talking," and the part "dancing and talking" is flagged as a contradiction 343.

[0117] In other words, as shown in section (b) of Figure 9, the situation presented is a conversation with a friend at school, and the second persona information, "when at school," should be selected. On the other hand, the utterance in question is thought to refer to item #5, "good at dancing and talking," in the first persona information, which contradicts the presented situation. Therefore, the contradiction display unit 110 highlights the "dancing and talking" portion in this input area 34 as a contradiction point 343 (in bold and underlined in the example of Figure 11), providing a warning about the existence of a contradiction.

[0118] Section (b) of Figure 11 shows an example of the presentation of the basis for the contradiction and the proposed correction regarding the contradiction location 343 shown in Section (a) of Figure 11 by the contradiction display unit 110. The contradiction display unit 110 displays the basis for the contradiction 357 and the proposed correction 358 in the display area 36. In the example shown in the figure, the contradiction display unit 110 highlights in area 36g within the display area 36 that the content shown in the contradiction location 343 contradicts the "situation" as the basis for the contradiction 357.

[0119] Furthermore, the contradiction display unit 110 generates a revised version 358 ("He has an older brother") for the contradiction shown at the contradictory location 343. In the example shown in the figure, the contradiction display unit 110 displays the generated revised version 358 in the display area 36h, directly below area 36g.

[0120] Using Figures 12 and 13, we will explain an example of a warning display for a contradiction in the content of the utterance in case (C) described above.

[0121] Figure 12 is a schematic diagram showing an example in which a set of persona information consisting of multiple patterns of persona information is presented according to the embodiment. Section (a) of Figure 12 shows how two patterns of persona information, consisting of first persona information and second persona information, are displayed in the display area 31a of the dialogue data collection screen 30. The first persona information and the second persona information are the same as the first and second persona information explained using Figure 9, so their explanation is omitted here.

[0122] For a dialogue with person A, for whom the first and second persona information is set, a situation is presented. In this example, as shown in section (b) of Figure 12, the situation presented is "talking about special skills." This situation, "talking about special skills," does not depend on the first and second persona information shown in section (a).

[0123] Figure 13 is a schematic diagram showing an example of a warning display for a contradiction in the content of the utterance in case (C) described above, according to the embodiment. It is assumed that worker X is presented with the first and second persona information shown in section (a) of Figure 12 and the situation shown in section (b) of Figure 12.

[0124] Section (a) of Figure 13 shows an example of a spoken sentence entered into the input area 34 by worker X. In this example, the spoken sentence is "I am good at cooking," and the "cooking" part corresponds to "I am good at cooking" in item #5 of the profile information 311b in the second persona information, thus referring to the second persona information. The contradiction display unit 110 highlights the mention location 344 in this input area 34 (bold and underlined in the example in the figure).

[0125] Section (b) of Figure 13 shows an example of the presentation of evidence by the contradiction display unit 110 regarding the reference 344 shown in Section (a) of Figure 13. In the example shown, the contradiction display unit 110 indicates to the display area 36 that a reference to the second persona information ("when at school") has been detected in a situation that does not depend on the first and second persona information.

[0126] Figure 13 is a schematic diagram showing an example of a warning display for a contradiction in the content of the utterance in case (C) described above, according to the embodiment. It is assumed that worker X is presented with the first and second persona information shown in section (a) of Figure 12 and the situation shown in section (b) of Figure 12.

[0127] Section (a) of Figure 13 shows an example of a spoken sentence entered into the input area 34 by worker X. In this example, the spoken sentence is "I am good at cooking," and the "cooking" part corresponds to "I am good at cooking" in item #5 of the profile information 311b in the second persona information, thus referring to the second persona information. The contradiction display unit 110 highlights the mention location 344 in this input area 34 (bold and underlined in the example in the figure).

[0128] Section (b) of Figure 13 shows an example of the presentation of evidence by the contradiction display unit 110 regarding the reference 344 shown in Section (a) of Figure 13. In the example shown, the contradiction display unit 110 indicates to the display area 36 that a reference to the second persona information ("when at school") has been detected in a situation that does not depend on the first and second persona information.

[0129] Figure 14 is a schematic diagram showing an example of a warning display for a contradiction in the content of the utterance in case (D) described above, according to the embodiment. Worker X is presented with the first and second persona information shown in section (a) of Figure 12 and the situation shown in section (b) of Figure 12.

[0130] Section (a) of Figure 14 shows an example of an utterance entered into input area 34 by worker X. In this example, the utterance is "I am good at cooking and dancing," and the part "cooking and dancing" is flagged as a location 345 where different persona information is mixed.

[0131] In other words, as shown in section (a) of Figure 12, the first persona information is described as "good at dancing" in item #5 of profile information 311a. The second persona information is described as "cooking is a hobby" in item #5 of profile information 311b. These two pieces of persona information, "good at dancing" and "cooking is a hobby," are independent of each other.

[0132] Section (b) of Figure 14 shows an example of the presentation of the basis and proposed corrections for the mixed location 345 shown in Section (a) of Figure 14 by the inconsistency display unit 110. In order to show the basis for the mixed location 345, the inconsistency display unit 110 presents in area 36a of the display area 36 that information 360 in the first persona information and information 361 in the second persona information are mixed, as shown in Section (b) of Figure 14.

[0133] Furthermore, the contradiction display unit 110 generates a revised version 362 ("cooking") for the description shown in the mixed section 345. In the example shown in the figure, the contradiction display unit 110 generates the revised version 362 based on the second persona information and displays it in the display area 36b, directly below area 36a.

[0134] Furthermore, the dialogue data collection system 1 can retain a history of the results of the inconsistency detection described using Figures 6 to 14. Based on this history of inconsistency detection results, the dialogue data collection system 1 can provide feedback to, for example, an operator using the dialogue data collection system 1 regarding the setting of persona information that is prone to inconsistencies.

[0135] (3-1-5. Flow of the inconsistency detection process) Figure 15 is a flowchart illustrating an example of the inconsistency detection process according to the embodiment. The flowchart in Figure 15 focuses on the processing between the server 10 and the worker terminal 20a used by worker X.

[0136] In step S100, the server 10 retrieves persona information from the persona information DB 101 randomly or in a predetermined combination using the persona information display unit 111, transmits the retrieved persona information to the worker terminal 20a, and presents the persona information to worker X. In step S110, the worker terminal 20a displays the persona information transmitted from the server 10 using the display unit 203.

[0137] In practice, the server 10 sends display control information to the worker terminal 20a to display the dialogue data collection screen 30, which includes, for example, acquired persona information and situational information. The worker terminal 20a, using its display unit 203, displays the dialogue data collection screen 30 on a display or the like according to the display control information sent from the server 10. The display processing on the worker terminal 20a is performed by updating the dialogue data collection screen 30 on the server 10, except for the input processing for the input area 34, but here we will explain it as if these processes are performed on the worker terminal 20a.

[0138] Worker X inputs the utterance into the input area 34 of the dialogue data collection screen 30 displayed on worker terminal 20a, and the utterance is acquired by the input unit 202 (step S111). Worker terminal 20a sends the acquired utterance to server 10. The utterance is received by server 10 and acquired by dialogue collection unit 112.

[0139] In step S101, the server 10 compares the utterance acquired by the dialogue collection unit 112 with the persona information presented to worker X in step S100 using the inconsistency detection unit 103, and detects inconsistencies between the utterance and the persona information. In the next step S102, the server 10 uses the inconsistency display unit 110 to determine whether or not an inconsistency was detected in step S101. If it is determined that an inconsistency was detected (step S102, "Yes"), the process moves to step S103.

[0140] On the other hand, if server 10 determines that no inconsistency was detected in step S102 (step S102, "No"), it skips the processing in step S103.

[0141] In step S103, the server 10 generates proposed corrections for the detected inconsistencies using the inconsistency display unit 110.

[0142] After processing in step S103, or if no inconsistencies are detected in step S102 and step S103 is skipped, the server 10 sends information including the inconsistency detection results and, if any, the proposed corrections, to the worker terminal 20a.

[0143] In step S112, the worker terminal 20a determines whether the information sent from the server 10 includes proposed revisions, etc. If it determines that it does include proposed revisions, etc. (step S112, "Yes"), it proceeds to step S113. On the other hand, if the worker terminal 20a determines in step S112 that it does not include proposed revisions, etc. (step S112, "No"), it skips the processing in step S113.

[0144] In step S113, the worker terminal 20a displays the suggested revisions sent from the server 10 in the display area 36 of the dialogue data collection screen 30 using the display unit 203. Based on the displayed suggested revisions, worker X corrects the utterance displayed in the input area 34 as necessary.

[0145] In the next step S114, the worker terminal 20a sends the utterance entered in the input area 34 to the server 10 in response to the operation of the send button 35. If the utterance was modified in step S113, the modified utterance is sent to the server 10. In addition, the worker terminal 20a sends to the server 10 transmission operation information indicating that the send button 35 was operated, along with the utterance.

[0146] In this example, the worker terminal 20a sends the utterance to the server 10 triggered by an operation on the send button 35, but this is not limited to this example. For example, the worker terminal 20a may also send the utterance triggered by a predetermined operation in the input area 34 (for example, an operation in combination with a special key).

[0147] In step S104, the server 10 receives the utterance and transmission operation information sent from the worker terminal 20a. In the next step S105, the server 10 uses the dialogue collection unit 112 to register the received utterance as dialogue data 330a from worker X in the dialogue data DB 102 and collects the dialogue log. In other words, the server 10 can determine that the utterance is a finalized utterance, including any modifications, because the transmission operation information is sent along with the utterance from the worker terminal 20a, and therefore it can be registered in the dialogue data DB 102.

[0148] (3-1-6. Relationship between contradiction detection results and spoken text and persona information) Next, the relationship between the contradiction determination result and the utterance and persona information according to the embodiment will be explained. The result of the contradiction determination by the contradiction detection unit 103 changes depending on whether or not there is an inclusion relationship between the utterance and the persona information, and whether or not the description content (value) of the utterance and the persona information is unique.

[0149] Figure 16 is a schematic diagram showing the results of contradiction determination according to the content of the utterance and persona information according to the embodiment. In Figure 16, each column indicates the inclusion relationship between the utterance and persona information as "semantic inclusion relationship exists," "partial semantic inclusion relationship exists," and "no semantic inclusion relationship," respectively. In addition, each row indicates the description content (value) of the utterance and persona information as "unique value" and "can take multiple values," respectively.

[0150] In Figure 16, (1) there is no warning for items with a semantic inclusion relationship and unique descriptions (values), (2) there is no warning for items with a semantic inclusion relationship and multiple possible descriptions (values), (3) there is a warning for items with a partial semantic inclusion relationship and unique descriptions (values), (4) there is a warning for items with a partial semantic inclusion relationship and multiple possible descriptions (values), and (5) there is a warning for items without a semantic inclusion relationship and unique descriptions (values), and (6) there is a warning for items without a semantic inclusion relationship and multiple possible descriptions (values).

[0151] Figures (1) to (6) in Figure 16 will be explained using specific examples.

[0152] (1) There is a semantic inclusion relationship, and the description (value) is unique. For example, if the profile information in the persona data includes the description "17 years old, female," and the statement is "I am a 17-year-old female," then there is a semantic inclusion relationship between the profile information and the statement, and the profile information has a unique value. In this case, the inconsistency detection unit 103 determines that the statement is not inconsistent with the profile information, and no warning is issued.

[0153] (2) There is a semantic inclusion relationship, and multiple descriptive contents (values) can be taken. For example, if the profile information in the persona data includes the statement "My hobby is cooking," and the spoken statement is "My hobby is cooking," then there is a semantic inclusion relationship between the profile information and the spoken statement. Furthermore, the profile information can take on multiple values ​​in terms of "cooking." In this case, the contradiction detection unit 103 determines that the spoken statement is not contradictory to the profile information, and no warning is issued.

[0154] (3) There is a partial semantic inclusion relationship, and the description content (value) is unique. For example, if the profile information in the persona data includes the description "left-handed," and the statement reads "I'm left-handed, but I write with my right hand," then there is a partial semantic inclusion relationship between the profile information and the statement ("left-handed"). In this example, the principle is "left-handed," and this value is unique. Furthermore, according to the statement, there is a specific case (when writing) that contradicts the principle. In this case, the contradiction detection unit 103 does not detect a contradiction, but the contradiction display unit 110 can issue a warning.

[0155] (4) There is a partial semantic inclusion relationship, and multiple descriptive contents (values) can be taken. For example, if a persona's profile information states that "cooking is their hobby," and their statement says, "I'm not good at making fried rice," then there is a partial semantic inclusion relationship between the profile information and the statement ("(Although cooking is their hobby,) they are not good at making fried rice"). In this example, the principle is "cooking is their hobby," and "hobby" can take on multiple values. Furthermore, according to the statement, the specific case (making fried rice) contradicts the principle. In this case, the contradiction detection unit 103 does not detect a contradiction, but the contradiction display unit 110 can issue a warning.

[0156] (5) There is no semantic inclusion relationship, and the description (value) is unique. For example, if the profile information in the persona information states "high school student" and the statement is "I am a university student," there is no semantic inclusion relationship between the profile information and the statement. In this example, the value "high school student" in the profile information is unique. The contradiction detection unit 103 detects a contradiction if there is no semantic inclusion relationship between persona information that can only take unique values ​​and the statement. In this case, the contradiction display unit 110 issues a warning that there is a contradiction.

[0157] (6) Items that do not have a semantic inclusion relationship and can take on multiple descriptive content (values). For example, if the profile information in the persona data states that "cooking is one of my hobbies," and the spoken statement says, "I also enjoy playing sports," then there is no semantic relationship between the profile information and the spoken statement. In this example, the principle is "cooking is one of my hobbies," but according to the spoken statement, "sports are also one of my hobbies." Since "hobbies" is an item that can take multiple values, the inconsistency detection unit 103 does not detect an inconsistency, but the inconsistency display unit 110 can issue a warning.

[0158] The contradiction detection unit 103 classifies the persona information and statements into the aforementioned classes (1) to (6) and performs contradiction detection. Therefore, it is possible to change the presence or absence of warnings and the content of the warning messages for each system, depending on the result of the contradiction detection.

[0159] (3-2. Regarding the dialogue evaluation according to the embodiment) Next, the dialogue evaluation according to the embodiment will be described. In the embodiment, a subjective evaluation is performed by the worker to assess the "ease of dialogue," and a quantitative evaluation is performed to assess the quality of the dialogue data. In the embodiment, based on these subjective and quantitative evaluations, the weight of the combination of persona information that can be determined to be factors in low-quality dialogue data is reduced to make it less likely to be selected.

[0160] First, the subjective evaluation according to the embodiment will be explained. After the dialogue between worker X and worker Y has finished, the dialogue data collection system 1 according to the embodiment presents, for example, a subjective evaluation input screen to worker X and requests worker X to input a subjective evaluation. Using the subjective evaluation input screen, the dialogue data collection system 1 requests worker X to evaluate two points: whether it was easy for worker X to create a response in the role of worker X, and whether it was easy for worker X to create a response to the response from worker Y.

[0161] Figure 17 is a schematic diagram showing an example of a subjective evaluation input screen 40 according to the embodiment. In the server 10, the subjective evaluation collection unit 113 generates display control information for displaying the subjective evaluation input screen 40 and transmits it to the worker terminal 20a. The worker terminal 20a displays the subjective evaluation input screen 40 using the display unit 203 according to the display control information transmitted from the server 10.

[0162] In Figure 17, the subjective evaluation input screen 40 includes a dialogue display area 400, an evaluation input area 41, and a send button 42.

[0163] The dialogue display area 400 is an area where information about the dialogue to be evaluated is displayed. The dialogue display area 400 includes a display area 31 that displays persona information including a persona image 310 and profile information 311, a display area 32 that displays a situation, and a display area 401 that displays the dialogue data 330a and 330b to be evaluated. Display area 401 displays, for example, the dialogue data 330a and 330b to be evaluated extracted from the dialogue log information stored in the dialogue data DB 102. In the example shown in the figure, by operating the scroll bar 402, the dialogue data 330a and 330b that are overflowing the screen can be displayed within the display area 401.

[0164] The evaluation input area 41 includes input area 410a and input area 410b.

[0165] Input area 410a is an area for inputting whether it was easy for worker X to create a response in the role of worker X. More specifically, input area 410a includes an input means for worker X to input the difficulty level when creating a response as the person (role) assigned to them. In the example in the figure, the difficulty level of response creation by worker X is input in five stages, from the most difficult stage [1] "very difficult to create" to the easiest stage [5] "very easy to create". In the example in the figure, worker X refers to the display content of the dialogue display area 400, for example, and uses the checkbox 411a as an input means to exclusively input one of the stages [1] to [5].

[0166] Input area 410b is an area for inputting whether it was easy to create a response to worker Y's response. In the example shown in the figure, in input area 410b, worker X evaluates the interaction with worker Y to determine whether it was easy to create a response to worker Y's response. More specifically, input area 410b includes an input means for worker X to input the ease of interaction of worker Y's response. In the example shown in the figure, the ease of interaction of worker X with worker Y is input in five stages, from the lowest stage [1] "very difficult to talk to" to the highest stage [5] "very easy to talk to". In the example shown in the figure, worker X, for example, refers to the display content of the interaction display area 400 and uses the checkbox 411b as an input means to exclusively input one of the stages [1] to [5].

[0167] The send button 42 transmits the input results from input areas 410a and 410b to the server 10 as information indicating the subjective evaluation by worker X. In the server 10, the subjective evaluation collection unit 113 acquires the subjective evaluation information transmitted from the worker terminal 20a. The subjective evaluation collection unit 113 calculates a score based on the acquired subjective evaluation information and passes the information indicating the calculated score to the persona combination evaluation unit 106.

[0168] On server 10, the automatic dialogue evaluation unit 105 obtains dialogue data 330a and 330b to be evaluated from the dialogue log information stored in the dialogue data DB 102. Based on the obtained dialogue data 330a and 330b, the automatic dialogue evaluation unit 105 performs a quantitative evaluation of the dialogue using evaluation indicators from known evaluation methods, such as Distinct-1 / 2, MaxBLEU, Persona-Recall / Predict / F1, and Persona-Cover, as described above. However, the automatic dialogue evaluation unit 105 can also perform a quantitative evaluation of the dialogue using its own evaluation method.

[0169] In this embodiment, the automated dialogue evaluation unit 105 performs a quantitative evaluation of the dialogue based on, for example, whether there are too many topic changes in the dialogue (whether the topic transitions are abrupt) and whether there are any logical gaps in the dialogue (whether the conversation makes sense). The automated dialogue evaluation unit 105 calculates a score based on the results of the quantitative dialogue table and passes information indicating the calculated score to the persona combination evaluation unit 106.

[0170] The persona combination evaluation unit 106 determines the quality of the dialogue between worker X and worker Y based on the subjective evaluation score received from the subjective evaluation collection unit 113 and the quantitative evaluation score received from the automated dialogue evaluation unit 105. The persona combination evaluation unit 106 also evaluates the combination of persona information based on the subjective evaluation score and the quantitative evaluation score, and assigns weights to the combination.

[0171] Note that while Figure 17 shows the persona image 310 and profile information 311 being evaluated together, this is not an exhaustive example. For example, it is also possible to configure the system to evaluate each item of the persona image 310 and profile information 311 individually.

[0172] (3-2-1. Specific examples of dialogue evaluation) Figure 18 is a schematic diagram showing an example of persona information and dialogue data that were evaluated as having high dialogue quality by the persona combination evaluation unit 106 according to the embodiment.

[0173] Section (a) of Figure 18 shows an example of persona information. In the example in the figure, persona image 310a is an illustration of a woman holding a microphone and wearing a hair accessory, similar to the example in Figure 6 described above. Profile information 311b includes five items, #1 to #5, similar to profile information 311 in Figure 6. Item #1 is the gender (female), status (high school sophomore), and age (17 years old) set for the person. Item #2 is the speaking style set for the person (speaks in a feminine manner). Item #3 is the occupation of the person (works as an idol). Item #4 is the family structure set for the person (has an older brother). Item #5 is the hobby set for the person (enjoys cooking).

[0174] Section (b) also shows examples of situations set for the persona information presented in Section (a). In the example in the diagram, the situation is specified as discussing each other's occupations.

[0175] Section (c) of Figure 18 shows an example of a dialogue between worker X and worker Y, based on the persona information shown in section (a) and the situation shown in section (b). In this example, worker X responds to dialogue data 330b-1 from worker Y with dialogue data 330a-1, worker Y responds to that with dialogue data 330b-2, and so on, with utterances being posted alternately between worker X and worker Y, thus facilitating the dialogue.

[0176] In this example, we can see that the responses of workers X and Y in this dialogue were well-coordinated, indicating a high-quality conversation.

[0177] In such high-quality dialogues, both the subjective evaluation by worker X and the quantitative evaluation by the system yield high scores, suggesting that the combination of persona information is favorable. Therefore, the persona combination evaluation unit 106 maintains the weights of the persona information shown in section (a) of Figure 18. Alternatively, depending on the scores of the subjective and quantitative evaluations, the persona combination evaluation unit 106 increases the weight of the persona information to improve its likelihood of selection.

[0178] Figure 19 is a schematic diagram showing an example of persona information and dialogue data that were evaluated as having low dialogue quality by the persona combination evaluation unit 106 according to the embodiment.

[0179] Section (a) of Figure 19 shows an example of persona information. In the example in the figure, persona image 310c is an illustration of a woman wearing a headscarf, a hat, and work clothes. Profile information 311c includes five items, #1 to #5. Item #1 specifies the person's gender (female), status (high school sophomore), and age (17 years old). Item #2 specifies the person's speaking style (speaks in a feminine manner). Item #3 specifies the person's occupation (works as an idol). Item #4 specifies the person's family structure (has two children). Item #5 specifies the person's hobbies (works out).

[0180] Section (b) also shows examples of situations set for the persona information presented in Section (a). In the example in the diagram, the situation is specified as discussing each other's occupations.

[0181] Section (c) of Figure 19 shows an example of a dialogue between worker X and worker Y, based on the persona information shown in section (a) and the situation shown in section (b). Icon 331c represents worker X based on persona image 310c. In this example, worker X responds to dialogue data 330a-10 with dialogue data 330b-10 from worker Y, and worker Y responds to that with dialogue data 330b-11, and so on, with utterances being posted alternately between worker X and worker Y, thus creating a dialogue.

[0182] In this example, in this dialogue, Worker X's dialogue data 330a-10, stating that "I am a high school student and I am an idol," is criticized by Worker Y's dialogue data 330b-11 for being inconsistent with the persona image 310c. Furthermore, Worker X's dialogue data 330a-11, stating that "I have two children," is criticized by Worker Y's dialogue data 330b-12 for being inconsistent with profile information 311c, such as items #1 to #3. Thus, in this example, the responses of Worker X and Y do not mesh well with each other, indicating a low-quality dialogue.

[0183] In such low-quality dialogues, both the subjective evaluation by worker X and the quantitative evaluation by the system result in low scores, suggesting that the combination of persona information is not good. Therefore, the persona combination evaluation unit 106 reduces the weight of the persona information shown in section (a) of Figure 19, thereby decreasing the likelihood of that persona information being selected.

[0184] The dialogue data collection system 1 according to this embodiment can present the basis for the evaluation (weighting) of the combination of persona information based on subjective and quantitative evaluations. For example, the persona combination evaluation unit 106 can present the basis for the evaluation by representing the relationship between the combination of persona information and the quality of the dialogue using a graph.

[0185] Figure 20 is a schematic diagram showing an example of a graph for presenting the basis for evaluating combinations of persona information, as presented by the persona combination evaluation unit 106 according to this embodiment. Here, as shown in sections (a) to (f) of Figure 20, the persona combination evaluation unit 106 generates scatter plots of subjective evaluation scores and quantitative evaluation scores for each combination of persona information A to F.

[0186] In the example in Figure 20, the combinations of persona information A, B, and D shown in sections (a), (b), and (d) are combinations in which both the subjective evaluation score and the quantitative evaluation score are high (for example, each is 0.5 or higher). On the other hand, the combination of persona information F shown in section (f) is a combination in which both the subjective evaluation score and the quantitative evaluation score are low (for example, each includes scores below 0.5). The scatter plots of these sections (a) to (f) serve as the basis for the quality of the persona information combinations A to F.

[0187] By weighting combinations of persona information based on subjective and quantitative evaluations, it becomes easier to select high-quality combinations. However, there is a risk that similar combinations will be selected repeatedly even after the evaluation results have been accumulated. To avoid this, it is preferable to generate combinations that resemble "mutations" of persona information, similar to a genetic algorithm.

[0188] (3-2-2. Flowchart for evaluating combinations of persona information) The persona information combination evaluation process according to the embodiment will be explained using Figures 21A and 21B. Figure 21A is a flowchart showing an example of the persona information combination evaluation process in the initial state according to the embodiment.

[0189] In step S200, the dialogue data collection system 1, using the persona information display unit 111, selects one or more persona information to present to workers X and Y from the persona information stored in the persona information DB 101, either randomly or based on settings. The persona information display unit 111 transmits the selected persona information to worker terminals 20a and 20b. Worker terminals 20a and 20b display the dialogue data collection screen 30, which includes the persona information transmitted from the persona information display unit 111.

[0190] Workers X and Y interact according to the persona information displayed on their respective worker terminals 20a and 20b, and send the interaction data 330a and 330b to the server 10.

[0191] In the next step S201, the dialogue data collection system 1 acquires dialogue data 330a and 330b from workers X and Y using the dialogue collection unit 112 and stores them in the dialogue data DB 102. The dialogue data collection system 1 also acquires information indicating worker X's subjective evaluation using the subjective evaluation collection unit 113.

[0192] Furthermore, the dialogue data collection system 1, using the dialogue automatic evaluation unit 105, acquires information indicating a quantitative evaluation of the dialogue based on the acquired dialogue data 330a and 330b.

[0193] In the next step, S202, the persona combination evaluation unit 106 calculates a score based on the subjective evaluation information obtained by the subjective evaluation collection unit 113. The persona combination evaluation unit 106 also calculates a score based on the quantitative evaluation information obtained by the automated dialogue evaluation unit 105. Based on these calculated scores, the persona combination evaluation unit 106 evaluates the combination of persona information selected in step S200.

[0194] In the next step, S203, the persona combination evaluation unit 106 updates the weights for the combination of persona information selected in step S200, based on the evaluation results from step S202.

[0195] Figure 21B is a flowchart illustrating an example of the persona information combination evaluation process when the system is fully operational, according to the embodiment.

[0196] In step S210, the dialogue data collection system 1, using the persona information display unit 111, selects one or more persona information items to present to workers X and Y from the persona information stored in the persona information DB 101, taking into account the weight conditions for combinations of persona information. The weight conditions for combinations of persona information are specified to the dialogue data collection system 1, for example, by the operator of the dialogue data collection system 1.

[0197] The persona information display unit 111 transmits the selected persona information to the worker terminals 20a and 20b. The worker terminals 20a and 20b display the dialogue data collection screen 30, which includes the persona information transmitted from the persona information display unit 111.

[0198] The processing from step S210 onward is the same as the processing in steps S201 to S203 in the flowchart of Figure 21A, so the explanation here will be omitted.

[0199] (3-2-3. Changes in Persona Information Due to Weight Updates) As described above, in step S203 of the flowcharts in Figures 21A and 21B, the weights for the combinations of persona information are updated. In the dialogue data collection system 1, as the collection of dialogue data 330a and 330b progresses and the weights for the combinations of persona information are repeatedly updated, a good set of persona information combinations remains, taking into account the ease of work for workers X and Y.

[0200] Figure 22 is a schematic diagram illustrating an example of how persona information changes with the updating of weights for combinations of persona information according to the embodiment. More specifically, Figure 22 shows how the weights of persona information change as dialogue data is collected from step S1 to step S5, and how the persona information is updated accordingly.

[0201] In Figure 22, the changes in persona information are shown by calculating weights for each item of the persona image 310 and profile information 311 in both subjective and quantitative evaluations. The weights take values ​​between [0] and [1.0], and the dialogue data collection system 1 acquires persona information from the persona information DB 101 according to the probability based on the weights.

[0202] In the following explanation, we will assume that the persona information is fixed in steps S1 to S5, with item #1 "Female / High School Sophomore / 17 years old" and item #2 "Active as an idol" being fixed as profile information.

[0203] Step S1 shows an example of initial persona information. In Step S1, the dialogue data collection system 1 extracts persona image 310s-1 and profile information 311s-1 from the persona information DB 101 randomly or according to a predetermined setting, and presents them as persona information.

[0204] In this example, persona image 310s-1 is an illustration of a woman with a blank expression and no accessories, and its weight 50s-1 is set to [0.45]. Profile information 311s-1 includes five items, #1 to #5. Item #1 specifies the person's gender (female), status (high school sophomore), and age (17 years old). Item #2 specifies the person's occupation (working as an idol). Item #3 specifies the person's primary family structure (has an older brother). Item #4 specifies the person's secondary family structure (has two children). Item #5 specifies the person's hobby (works out). Furthermore, items #1 to #5 have weights of [1.0], [1.0], [0.98], [0.32], and [0.65], respectively, as shown on the far right.

[0205] In step S2, persona image 310s-1 from step S1 is updated to persona image 310s-2, which is an unadorned but expressive illustration of a woman. Persona image 310s-2 has a weight of 50s-2 that is greater than that of persona image 310s-1 [0.49]. Additionally, in profile information 311s-2, items #4 and #5, which had small weights in step S1, are updated to item #4 "Prone to shyness" (weight [0.52]) and item #5 "Good at studying" (weight [0.85]).

[0206] In step S3, persona image 310s-2 from step S2 is updated to persona image 310s-3, which is an illustration of a woman with a more emphasized facial expression and holding items in both hands. Persona image 310s-3 has a weight of 50s-3 which is greater than that of persona image 310s-2 [0.7]. In addition, in profile information 311s-3, item #4, which had a small weight in step S2, is updated to item #4 "Likes pink" (weight [0.95]). Furthermore, in profile information 311s-3, the weight of item #5 "Good at studying" is set to [0.8], which is smaller compared to step S2.

[0207] In step S4, persona image 310s-3 from step S3 is updated to persona image 310s-4, which is an illustration of a woman wearing a hair accessory and holding a microphone. Persona image 310s-4 has a weight of 50s-4 which is greater than that of persona image 310s-3 [0.97]. Also, in profile information 311s-4, the weight of item #2 "is an idol" is set to [0.99], which is smaller than in step S3. Item #3, which indicates family structure "has an older brother" in profile information 311s-3, has been mutated and updated to "speaks in a feminine manner" (weight [0.98]), which indicates speech style. Furthermore, in profile information 311s-4, the weight of item #5 "good at studying" is set to [0.7], which is smaller than in step S3.

[0208] In step S5, persona image 310s-4 from step S4 is used as persona image 310s-5, and the weight 50s-5 is also the same as persona image 310s-4 [0.97]. In addition, in profile information 311s-5, item #5 "Good at studying," which had the smallest weight in step S4, has been updated to item #5 "Hobbies cooking" (weight [0.92]), which has a larger weight. Items #1 to #4 in profile information 311s-5 remain unchanged from step S4.

[0209] Thus, in this embodiment, as the collection of dialogue data progresses, the persona information is updated with more important information. This improves the ease of dialogue by the worker.

[0210] [4. First Modification of an Embodiment of the Present Disclosure] Next, a first modification of the embodiment of this disclosure will be described. In the embodiment described above, inconsistencies in the dialogue with respect to persona information were detected and warnings were issued. In contrast, the first modification of the embodiment detects inconsistencies in information outside the scope of persona information included in the dialogue. More specifically, in the first modification of the embodiment, the dialogue data collection system 1 refers to past dialogue data for the utterance input by the worker and detects inconsistencies between the information included in the utterance and the information included in the past dialogue data.

[0211] As an example, consider a scenario where a dialogue takes place between worker X, who is presented with specific persona information that is the target of dialogue log collection, and worker Y, who is presented with other arbitrary persona information. It should be assumed that this specific persona information is available at a point when dialogue log collection has progressed sufficiently. Furthermore, it should be assumed that prior to the dialogue between worker X and worker Y, this specific persona information was presented to worker X' (a different worker from worker X), and that a dialogue took place between worker X' and the other worker, with dialogue logs being collected.

[0212] Here, the dialogue log of the conversation between worker X and worker Y is collected in addition to the dialogue log of the conversation between worker X', to whom the specific persona information was presented, and other workers. In this case, it is possible that there may be a contradiction between the information outside of the persona information contained in the dialogue data of worker X' and the information contained in the utterance entered by worker X after the conversation of worker X'.

[0213] Figures 23A and 23B are schematic diagrams to more specifically explain the processing according to the first modification of the embodiment. Figure 23A shows an example of dialogue log information collected with respect to specific persona information and registered in the dialogue data DB 102. In this example, the dialogue log information is registered with numbered entries in chronological order, representing dialogue data from a worker (let's call him worker X') to whom the specific persona information was presented. Specifically, in the example in Figure 23A, dialogue data numbered [1] includes the description "a female high school sophomore," and dialogue data numbered [2] includes the description "she is an idol." Also, dialogue data numbered [3] includes the description "her favorite subject is Japanese," and dialogue data numbered [4] includes the description "she ranked second in her grade on a test."

[0214] Figure 23B shows an example of the specific persona information, the utterance entered by the worker (let's call it Worker X) to whom the specific persona information was presented, and the warning display for the utterance. In Figure 23B, section (a) shows an example of the specific persona information. The persona information shown in section (a) is the same as the persona information explained using Figure 6, so its explanation is omitted here.

[0215] In Figure 23B, section (b) shows an example of a speech sentence entered by worker X into input area 34 in accordance with the persona information shown in section (a). In this example, the entered speech sentence is "I'm not good at Japanese," and the part "I'm not good at Japanese" is identified as contradiction point 363, which contradicts the description "My strong subject is Japanese" in the dialogue log information numbered [3] shown in Figure 23A.

[0216] In this case, the classification classes and inconsistency criteria can be those described using (1) to (6) in Figure 16 in the embodiment.

[0217] The dialogue data collection system 1 uses a contradiction detection unit 103 to refer to the dialogue data DB 102 and detects contradictions in spoken sentences with respect to the dialogue data included in the dialogue log. If a contradiction is detected, the contradiction display unit 110 displays in the display area 36 that the description at the contradiction point 363 contradicts the description "My favorite subject is Japanese" in the dialogue data number [3] included in the dialogue log information, "I'm studying hard! My favorite subject is Japanese." At this time, it is preferable for the contradiction display unit 110 to highlight the part 364 of "My favorite subject is Japanese."

[0218] As a first example of inconsistency detection according to the first modification of the embodiment, the dialogue data collection system 1 can warn of the existence of an inconsistency if only one inconsistency of a similar category is detected in previously collected dialogue log information. For example, referring to section (a) of Figure 23B, suppose the utterance contains a description of the category "favorite subject" which is not present in the profile information 311 of the presented persona information. In this case, the inconsistency display unit 110 can warn if the description of the category "favorite subject" contained in the utterance contradicts the previously registered information of the category "favorite subject" in the utterance.

[0219] As a second example of inconsistency detection related to the first modification of the embodiment, the dialogue data collection system 1 can reflect the majority trend when multiple inconsistent descriptions outside the scope of persona information accumulate. For example, utterances containing similar content are classified into classes using the X-means method, and the majority trend is determined when the number of utterances in a cluster exceeds a threshold. In the above example, if multiple descriptions of "good at Japanese" and "bad at Japanese" are collected in the dialogue log information, the one with the larger number of occurrences among "good at Japanese" and "bad at Japanese" is set to true. The inconsistency display unit 110 warns of a contradiction when an input utterance contains a description that contradicts this true description.

[0220] Of the first and second examples described above, it is preferable to adopt the second example in principle.

[0221] Figure 24 is a flowchart illustrating an example of a contradiction detection process related to the first modified embodiment. The flowchart in Figure 24 focuses on the processing between the server 10 and the worker terminal 20a used by worker X. Furthermore, this section describes the case where the second example of contradiction detection, one of the first and second examples of contradiction detection related to the first modified embodiment described above, is applied.

[0222] In step S300, the server 10 retrieves persona information from the persona information DB 101 randomly or in a predetermined combination using the persona information display unit 111, transmits the retrieved persona information to the worker terminal 20a, and presents the persona information to worker X. In step S310, the worker terminal 20a displays the persona information transmitted from the server 10 using the display unit 203.

[0223] In practice, the server 10 sends display control information to the worker terminal 20a to display the dialogue data collection screen 30, which includes, for example, acquired persona information and situational information. The worker terminal 20a, using its display unit 203, displays the dialogue data collection screen 30 on a display or the like according to the display control information sent from the server 10. The display processing on the worker terminal 20a is performed by updating the dialogue data collection screen 30 on the server 10, except for the input processing for the input area 34, but here we will explain it as if these processes are performed on the worker terminal 20a.

[0224] Worker X inputs the utterance into the input area 34 of the dialogue data collection screen 30 displayed on worker terminal 20a, and the utterance is acquired by the input unit 202 (step S311). Worker terminal 20a sends the acquired utterance to server 10. The utterance is received by server 10 and acquired by dialogue collection unit 112.

[0225] In step S301, the server 10 uses the inconsistency detection unit 103 to refer to the dialogue log information stored in the dialogue data DB 102 based on the utterances acquired by the dialogue collection unit 112. The inconsistency detection unit 103 detects inconsistencies in the utterances related to the dialogue data included in the dialogue log information that fall outside the scope of the persona information. In the next step S302, the server 10 uses the inconsistency display unit 110 to determine whether an inconsistency was detected in step S301. If it is determined that an inconsistency was detected (step S302, "Yes"), the process proceeds to step S303.

[0226] On the other hand, if server 10 determines that no inconsistency was detected in step S302 (step S302, "No"), it skips the processing in step S303.

[0227] In step S303, the server 10 generates a warning corresponding to the detected inconsistency using the inconsistency display unit 110.

[0228] After processing step S303, or if step S303 is skipped because no inconsistencies are detected in step S302, server 10 sends information to worker terminal 20a including the inconsistency detection results and any warnings that have been created.

[0229] In step S312, the worker terminal 20a determines whether the information sent from the server 10 contains a warning. If it determines that a warning is included (step S312, "Yes"), it proceeds to step S313. On the other hand, if the worker terminal 20a determines in step S312 that no corrections or other suggestions are included (step S312, "No"), it skips the processing in step S313 and proceeds to step S314.

[0230] In step S313, the worker terminal 20a displays the warning sent from the server 10 in the display area 36 of the dialogue data collection screen 30 using the display unit 203, and proceeds to step S314.

[0231] In step S314, worker terminal 20a sends the utterance entered in input area 34 to server 10 in response to the operation of the send button 35. Worker X can modify the utterance entered in input area 34 in response to the warning displayed in step S313. If the utterance has been modified, the modified utterance is sent to server 10. In addition, worker terminal 20a sends transmission operation information to server 10 along with the utterance, indicating that the send button 35 has been operated.

[0232] In step S304, the server 10 receives the utterance and transmission operation information sent from the worker terminal 20a. In the next step S305, the server 10 uses the dialogue collection unit 112 to determine whether or not to register the received utterance as dialogue data 330a by worker X in the dialogue data DB 102. The dialogue collection unit 112, for example, refers to the dialogue data DB 102 and determines to register the utterance if the majority of the content in the utterance received from the worker terminal 20a is inconsistent.

[0233] If the dialogue collection unit 112 determines in step S305 to register (step S305, "Yes"), it moves the process to step S306 and registers the utterance received in step S304 as dialogue data 330a by worker X in the dialogue data DB 102. On the other hand, if the dialogue collection unit 112 determines in step S305 not to register (step S305, "No"), it moves the process to step S307 and discards the utterance received in step S304. However, even if it determines in step S305 not to register, the dialogue data 330a related to the utterance may be registered in the dialogue data DB 102.

[0234] In addition, in the first modified embodiment, it is also possible to present new candidate persona information by applying the contradiction determination result. In the example described above, it is conceivable to present candidate profile information 311 to be included in the persona information based on the utterance determined to be registered in step S305 (for example, "My strong subject is Japanese" or "I'm not good at Japanese"). In this way, for example, it is possible to present utterances that are infrequent but produce high-quality dialogue.

[0235] Whether the persona information, including the presented candidates, is easy for worker X to embody and easy for worker Y to converse with can be determined by performing a subjective evaluation as described in the embodiment.

[0236] Furthermore, in the first modified embodiment, even if there is a majority tendency for descriptions outside the scope of persona information, the tendency for "ease of dialogue by worker Y, the dialogue partner" may be low (for example, the evaluation value related to said evaluation is below a predetermined level) based on the results of subjective evaluations, etc. In this case, the dialogue data corresponding to said description can be excluded from the data to be registered. This prevents inappropriate utterances or settings from being notified or registered due to attacks by malicious workers.

[0237] Thus, according to the first modification of the embodiment, even if utterances outside the scope of the presented persona information are input to dialogue data from different workers who have been presented with the same persona information, the consistency of the persona information can be maintained. This makes it possible to collect dialogue data that is easy to handle after the dialogue log information is collected.

[0238] [5. Second Modification of an Embodiment of the Present Disclosure] Next, a second modification of the embodiment of the present disclosure will be described. The second modification of the embodiment is an example in which images to be included in persona information are generated based on persona information. More specifically, in the second modification of the embodiment, images to be included in persona information are generated based on combinations of profile information 311 included in persona information that are determined to be of good quality.

[0239] Figure 25 is a schematic diagram illustrating a first example of the image generation process according to a second modification of the embodiment. In the first example, as shown in section (a) of Figure 25, only profile information 511a exists in the persona information, and no profile image exists. The profile information 511a is assumed to be a combination of items #1 to #5 that has been determined to be a sufficiently good combination by subjective evaluation as described in the embodiment.

[0240] In the dialogue data collection system 1, the persona image generation unit 104 generates a persona image 510b, as exemplified in section (b) of Figure 25, based on each of items #1 to #5 of the profile information 511a.

[0241] For example, as shown in section (a), in profile information 511a, item #1 is set as the person's gender (female), status (high school sophomore), and age (17 years old). Item #2 is set as the person's speaking style (speaks in a feminine manner). Item #3 is set as the person's occupation (works as an idol). Item #4 is set as the person's preference (likes the color pink). Item #5 is set as the person's hobby (enjoys cooking).

[0242] The persona image generation unit 104 generates a persona image 510b based on information such as "female," "second year high school student," "17 years old," "feminine way of speaking," "idol activities," "pink color," and "cooking as a hobby," which are included in each of the items #1 to #5. In this case, the persona image generation unit 104 can use, for example, GAN (Generative Adversarial Networks) to generate the persona image 510b. However, the persona image generation unit 104 may also select an image from a plurality of persona images pre-stored in the persona information DB 101 that corresponds to each of the above-mentioned pieces of information included in each of the items #1 to #5 of the profile information 511a, and use that as the persona image 510b.

[0243] Furthermore, in the example shown in Figure 25, when the persona image 510b is generated, the profile information 511a is updated to generate profile information 511b corresponding to the persona image 510b. The persona information is composed of these persona image 510b and profile information 511b. The persona image generation unit 104 stores the generated persona image 510b and profile information 511b in the persona information DB 101.

[0244] Figure 26 is a schematic diagram illustrating a second example of the image generation process according to a second modification of the embodiment. In the first example, as shown in section (a) of Figure 26, the persona information is composed of a persona image 510c and profile information 511c. In this case, as shown in section (b) of Figure 26, the persona image generation unit 104 can generate a new persona image 510d based on the persona image 510c and each of items #1 to #5 of the profile information 511c.

[0245] In this example in Figure 26, as in the example in Figure 25 described above, the profile information 511c is updated when the persona image 510d is generated, thereby generating profile information 511d corresponding to the persona image 510d. The persona information is composed of these persona image 510d and profile information 511d.

[0246] Figure 27 is a schematic diagram illustrating a third example of the image generation process according to a second modified embodiment. This third example is an example of generating a persona image in a set of persona information consisting of multiple patterns of persona information. As shown in section (a) of Figure 27, the set of persona information includes first persona information consisting of a persona image 510e-1 and profile information 511e-1, and second persona information consisting of profile information 511e-2. The first persona information represents the persona "at work," and the second persona information represents the persona "at school." In the example shown in the figure, the second persona information does not include a persona image.

[0247] In this case, the persona image generation unit 104 can generate the persona image 510f-2 shown in section (b) of Figure 26, based on the profile information 511e-2 that is the target of persona image generation and the persona image 510e-1 included in the first persona information, among other things.

[0248] Let me explain in more detail. In section (a) of Figure 27, persona image 510e-1 shows a woman holding a microphone and wearing a hair accessory. In profile information 511e-1, item #1 is the gender (female) and age (17 years old) set for the person. Item #2 is the speech style set for the person (speaks in a feminine manner). Item #3 is the occupation of the person (works as an idol). Item #4 is the preference set for the person (likes the color pink). Item #5 is the hobby set for the person (good at dancing and talking).

[0249] On the other hand, in profile information 511e-2, item #1 is the gender (female) and age (17 years old) set for the person. Item #2 is the speech style set for the person (speaks in a feminine manner). Item #3 is the occupation of the person (a second-year high school student). Item #4 is the family structure set for the person (has an older brother). Item #5 is the hobby set for the person (cooking is their hobby).

[0250] The persona image generation unit 104 may, for example, modify the persona image 510e-1 based on the profile information 511e-2 that is the target of the persona image generation. As an example, the persona image generation unit 104 may, based on the information of items #3 to #5 of the profile information 511e-2 that are different from those of the profile information 511e-1, remove the microphone and hair accessories from the persona image 510e-1, add a high school uniform, and generate the persona image 510f-2.

[0251] The persona image generation unit 104 may also select a persona image corresponding to profile information 511e-1 from the persona images stored in the persona information DB 101 and use it as persona image 510f-2.

[0252] In the example shown in Figure 27, the persona image generation unit 104 uses the persona image 510e-1 and profile information 511e-1 as they are, as persona image 510f-1 and profile information 511f-1. Similarly, the persona image generation unit 104 uses the profile information 511e-2 as it is, as profile information 511f-2.

[0253] The persona image generation unit 104 stores the generated persona image 510f-2 in the persona information DB 101, associating it with the profile information 511f-2.

[0254] As described above, according to the second modification of the embodiment, the dialogue data collection system 1 generates a persona image based on a combination of profile information determined to be high-quality. Therefore, by applying the second modification of the embodiment, the persona image generated based on the profile information included in the persona information can compensate for the imagination of the worker who conducts the dialogue.

[0255] It should be noted that the effects described in the present specification are merely illustrative and not restrictive, and other effects may be obtained.

[0256] Furthermore, the present technology can also adopt the following configurations. (1) a role presentation unit that presents a role set for a first user to the first user; an acquisition unit that acquires first data input by the first user in association with the input of the first data; a detection unit that detects information conflicting with the role, which is included in the first data acquired by the acquisition unit, as contradiction information; a contradiction presentation unit that presents a message based on the contradiction information detected from the first data by the detection unit to the first user; comprising: An information processing apparatus. (2) the contradiction presentation unit generates a correction proposal for the first data based on the contradiction information and the role, and presents the generated correction proposal to the first user; The information processing apparatus according to (1). (3) the detection unit detects the contradiction based on at least one of a content indicated by the first data and a style in the first data; The information processing apparatus according to (1) or (2). (4) the role presentation unit The role is presented to the first user by persona information which includes at least one of the character's image information and the character's attribute information. An information processing device as described in any of (1) to (3) above. (5) The aforementioned role presentation unit is, The system presents the first user with a number of roles that correspond one-to-one with each of the multiple pieces of persona information. The information processing device described in (4) above. (6) An evaluation unit evaluates the combination of information included in the persona information based on the first data and the second data entered by the second user, who is the conversation partner of the first user. Furthermore, The evaluation unit, The combination of information in the persona information is changed according to the evaluation. The information processing device described in (4) or (5) above. (7) The evaluation unit, The combination of the image information and attribute information included in the persona information is evaluated. The information processing device described in (6) above. (8) The evaluation unit, The combination of information included in the aforementioned attribute information is evaluated. The information processing device described in (6) above. (9) The evaluation unit, The evaluation is performed based on at least one of the following: a subjective evaluation based on the subjective opinions of at least one of the first user and the second user in accordance with the first data and the second data; and a quantitative evaluation based on an index calculated based on the first data and the second data. The information processing device described in any of (6) to (8) above. (10) The detection unit is The discrepancy is detected based on at least one of the image information and the attribute information. An information processing device as described in any of (4) to (9) above. (11) An image generation unit that generates new image information based on a combination of the aforementioned image information and the aforementioned attribute information, or a combination of information contained in the aforementioned attribute information. Furthermore, The aforementioned role presentation unit is, The persona information, which combines the attribute information and the new image information, is used to present the role to the first user. An information processing device as described in any of (4) to (10) above. (12) A registration unit that registers the aforementioned first data as log information in the log information storage unit. Furthermore, The aforementioned registration unit is When the first data from which the inconsistent information has been detected by the detection unit is corrected, the corrected first data is registered as log information in the log information storage unit. An information processing device as described in any of (1) to (11) above. (13) The aforementioned log information is, Including data entered by a user different from the first user, to whom the aforementioned role has been presented in common with the first user, The information processing device described in (12) above. (14) The detection unit is The acquisition unit further detects any information in the first data acquired that contradicts the log information, as the contradictory information. The information processing device described in (12) or (13) above. (15) The aforementioned role presentation unit is, The role is presented to the first user by persona information which includes at least one of the character's image information and the character's attribute information. Based on information inconsistent with the log information detected by the detection unit, new persona information is generated, and the role based on the generated persona information is presented to the first user. The information processing apparatus according to (14) above. (16) The detection unit is: when information that belongs to a category common to each other and contradicts the log information is detected in a predetermined amount or more, detecting, as the contradiction information, information that contradicts information contradictory to the log information, which is included in the first data, The information processing apparatus according to any one of (12) to (15) above. (17) The registration unit is: not registering information that contradicts the log information in the log information storage unit when the detection unit detects a predetermined amount or more of information that belongs to a mutually common category and contradicts the log information, and an evaluation of a dialogue between the first user and a second user who is the dialogue partner of the first user, which relates to the information that contradicts the log information, is equal to or less than a predetermined value, The information processing apparatus according to any one of (12) to (16) above. (18) The role presentation unit further: presents a role set for a second user, who is a dialogue partner of the first user, to the second user, The information processing apparatus according to any one of (1) to (17) above. (19) executed by a processor, a role presentation step of presenting a role set for a first user to the first user; an obtaining step of obtaining first data input by the first user in association with the input of the first data; a detection step of detecting, as contradiction information, information that contradicts the role and is included in the first data obtained in the obtaining step; a contradiction presentation step of presenting a message based on the contradiction information detected from the first data in the detection step to the first user; and having An information processing method. Description of Reference Numerals

[0257] 1. Dialogue Data Collection System 10 servers 20, 20a, 20b Worker terminals 30 Dialogue Data Collection Screen 31,31a,32,33,36,401 Display area 34,410a,410b Input Area 35,42 Send button 40 Subjective evaluation input screen 41 Evaluation Input Area 100 Dialogue Data Collection Screen Generation Unit 101 Persona Information Database 102 Dialogue Data Database 103 Inconsistency detection unit 104 Persona Image Generation Unit 105 Automatic Dialogue Evaluation Unit 106 Persona Combination Evaluation Department 110 Contradiction display section 111 Persona Information Display Section 112 Dialogue Collection Department 113 Subjective Evaluation Collection Department 202 Input Section 203 Display section 310, 310a, 310b, 310c, 310s-1, 310s-2, 310s-3, 310s-4, 310s-5, 510b, 510c, 510d, 510e-1, 510f-1, 510f-2 Persona Images 311, 311a, 311b, 311c, 311s-1, 311s-2, 311s-3, 311s-4, 311s-5, 511a, 511b, 511c, 511d, 511e-1, 511e-2, 511f-1, 511f-2 Profile Information Dialogue data for 330a, 330a-1, 330a-10, 330a-11, 330b, 330b-1, 330b-2, 330b-10, 330b-11, and 330b-12. 400 Interactive display area

Claims

1. A role presentation unit that presents the roles to be assigned to the first user, An acquisition unit that acquires the first data entered by the first user in conjunction with the input of the first data, A detection unit detects information in the first data acquired by the acquisition unit that contradicts the role, as contradictory information. The system includes a contradiction presentation unit that presents a message to the first user based on the contradiction information detected from the first data by the detection unit, The aforementioned contradiction presentation unit is, Based on the aforementioned contradictory information and the aforementioned roles, a proposed correction for the first data is generated, and the generated proposed correction is presented to the first user. Information processing device.

2. The detection unit is The discrepancy is detected based on at least one of the content indicated by the first data and the style in the first data. The information processing apparatus according to claim 1.

3. The aforementioned role presentation unit is, The role is presented to the first user by persona information which includes at least one of the character's image information and the character's attribute information. The information processing apparatus according to claim 1.

4. The aforementioned role presentation unit is, The system presents the first user with multiple roles that correspond one-to-one with each of the multiple pieces of persona information. The information processing apparatus according to claim 3.

5. An evaluation unit evaluates the combination of information included in the persona information based on the first data and the second data entered by the second user, who is the conversation partner of the first user. Furthermore, The evaluation unit, The combination of information in the persona information is changed according to the evaluation. The information processing apparatus according to claim 3.

6. The evaluation unit, The combination of the image information and attribute information included in the persona information is evaluated. The information processing apparatus according to claim 5.

7. The evaluation unit, The combination of information included in the aforementioned attribute information is evaluated. The information processing apparatus according to claim 5.

8. The evaluation unit, The evaluation is performed based on at least one of the following: a subjective evaluation based on the subjective opinions of at least one of the first user and the second user in accordance with the first data and the second data; and a quantitative evaluation based on an index calculated based on the first data and the second data. The information processing apparatus according to claim 5.

9. The detection unit is The discrepancy is detected based on at least one of the image information and the attribute information. The information processing apparatus according to claim 3.

10. An image generation unit that generates new image information based on a combination of the aforementioned image information and the aforementioned attribute information, or a combination of information contained in the aforementioned attribute information. Furthermore, The aforementioned role presentation unit is, The persona information, which combines the attribute information and the new image information, is used to present the role to the first user. The information processing apparatus according to claim 3.

11. A role presentation unit that presents a role to be assigned to a first user, An acquisition unit that acquires the first data entered by the first user in conjunction with the input of the first data, A detection unit detects information in the first data acquired by the acquisition unit that contradicts the role, as contradictory information. A contradiction presentation unit presents a message to the first user based on the contradiction information detected from the first data by the detection unit, The system includes a registration unit that registers the aforementioned first data as log information in a log information storage unit, The aforementioned registration unit is When the first data from which the inconsistent information has been detected by the detection unit is corrected, the corrected first data is registered as log information in the log information storage unit. Information processing device.

12. The aforementioned log information is, Including data entered by a user different from the first user, to whom the aforementioned role has been presented in common with the first user, The information processing apparatus according to claim 11.

13. The detection unit is The acquisition unit further detects any information in the first data acquired that contradicts the log information, as the contradictory information. The information processing apparatus according to claim 11.

14. The aforementioned role presentation unit is, The role is presented to the first user by persona information which includes at least one of the character's image information and the character's attribute information. Based on information inconsistent with the log information detected by the detection unit, new persona information is generated, and the role based on the generated persona information is presented to the first user. The information processing apparatus according to claim 13.

15. The detection unit is When a predetermined number of pieces of information belonging to a common category and which contradict the log information are detected, the information included in the first data that contradicts the information contradicting the log information is detected as the contradictory information. The information processing apparatus according to claim 11.

16. The aforementioned registration unit is If the detection unit detects more than a predetermined amount of information in common categories that contradicts the log information, and the evaluation of the dialogue between the first user and the second user related to the information contradicting the log information is below a predetermined amount, the information contradicting the log information is not registered in the log information storage unit. The information processing apparatus according to claim 11.

17. The aforementioned role presentation unit further, To the second user, who is the conversation partner of the first user, the role assigned to the second user is presented. The information processing apparatus according to claim 1.

18. Executed by the processor, A role presentation step in which the role to be assigned to the first user is presented to the first user, The acquisition step involves acquiring the first data entered by the first user in conjunction with the input of the first data, A detection step that detects information in the first data acquired in the acquisition step that is inconsistent with the role as inconsistent information, The method includes a contradiction presentation step which presents a message to the first user based on the contradiction information detected from the first data in the detection step, The aforementioned contradiction presentation step is, Based on the aforementioned contradictory information and the aforementioned roles, a proposed correction for the first data is generated, and the generated proposed correction is presented to the first user. Information processing methods.

19. Executed by a processor, A role presentation step in which the role to be assigned to the first user is presented to the first user, The acquisition step involves acquiring the first data entered by the first user in conjunction with the input of the first data, A detection step that detects information in the first data acquired in the acquisition step that is inconsistent with the role as inconsistent information, A contradiction presentation step in which a message based on the contradiction information detected from the first data in the detection step is presented to the first user, A registration step in which the first data is registered as log information in the log information storage unit, It has, The aforementioned registration step is, If the first data from which the inconsistent information was detected in the detection step is corrected, the corrected first data is registered as log information in the log information storage unit. Information processing methods.

Images