Estimation program for causing computer to estimate quality of conversation

Abstract

An estimation device according to one aspect of the present invention estimates the quality of a conversation on the basis of the amplitude value, of brain waves of each of two subjects having the conversation, of one or more predetermined components with respect to words included in the conversation.

Description

A computer-generated estimation program that estimates the quality of conversations

[0001] The present invention relates to an estimation program that causes a computer to estimate the quality of a conversation.

[0002] Previously, it has been reported that synchronization of brain activity is observed between two people conversing. For example, Non-Patent Document 1 below reports that increased synchronization of activity in the left inferior frontal cortex occurs between two familiar people conversing face-to-face. Non-Patent Document 1 also reports that synchronization of brain activity is not observed when people talk to themselves, even when they are face-to-face. Furthermore, Non-Patent Document 2 reports that synchronization of temporal-parietal comma band activity occurs between couples during natural conversation.

[0003] Various attempts to estimate or improve the quality of a conversation (e.g., a person's satisfaction with a conversation (dialogue)) are also known. For example, Patent Literature 1 discloses a dialogue system that estimates a user's satisfaction level based on a user's input voice, estimates a dialogue state with the user based on the estimated satisfaction level, and determines behavior toward the user based on the estimated dialogue state. Furthermore, Patent Literature 2 discloses a conversation satisfaction estimation device that estimates a speech intention from text of utterances in a conversation between multiple people and estimates a specific speaker's satisfaction level in the conversation from a transition of the speech intentions of the multiple people's utterances.

[0004] Jiang, J., Dai, B., Peng, D., Zhu, C., Liu, L., & Lu, C. (2012). “Neural synchronization during face-to-face communication”. Journal of Neuroscience, 32(45), 16064-16069. (Published November 7, 2012, doi: 10.1523 / JNEUROSCI.2926-12.2012) Kinreich, S., Djalovski, A., Kraus, L., Louzoun, Y., & Feldman, R. (2017). “Brain-to-brain synchrony during naturalistic social interactions”. Scientific reports, 7(1), 17060. (Published December 6, 2017, doi: 10.1038 / s41598-017-17339-5)

[0005] JP 2019-086679 A JP 2018-169506 A

[0006] However, none of the above-mentioned conventional technologies discloses or suggests the technical idea of ​​objectively estimating the quality of a conversation between two subjects from their respective electroencephalographic responses to words contained in the conversation.

[0007] In one aspect, the present invention has been made in consideration of the above circumstances, and its purpose is to provide an estimation program or the like that can objectively estimate the quality of a conversation between two subjects in a conversation from the electroencephalogram responses of each of the subjects to words contained in the conversation.

[0008] In order to solve the above-mentioned problems, the present invention employs the following configuration.

[0009] The estimation program relating to the first aspect causes a computer to execute an electroencephalogram acquisition step of acquiring electroencephalograms of each of two subjects who are having a conversation, an onset information acquisition step of acquiring onset information indicating the start time of each of a plurality of words included in the conversation, and an estimation step of estimating the quality of the conversation from the amplitude values ​​of one or more predetermined components of the electroencephalograms of each of the two subjects for the words.

[0010] In this configuration, the estimation program causes a computer to estimate the quality of the conversation from the amplitude values ​​of one or more predetermined components of the electroencephalograms of each of the two subjects in response to the words. The present inventors have discovered through experiments that there is a correlation between the amplitude values ​​of one or more predetermined components of the electroencephalograms of two people engaged in a conversation in response to words included in the conversation and the quality of the conversation, such as the level of interest in the conversation, emotional valence, and satisfaction of the two people. Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation from the electroencephalogram responses of each of the two people engaged in a conversation to words included in the conversation.

[0011] The estimation program according to a second aspect is the estimation program according to the first aspect, wherein the quality of the conversation includes the degree of satisfaction of each of the two subjects with the conversation, and in the estimation step, the computer may estimate the degree of satisfaction of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects, and may estimate the degree of satisfaction of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects.

[0012] In this configuration, the estimation program causes a computer to estimate the conversation partner's level of satisfaction with the conversation from the amplitude values ​​of the one or more predetermined components of the electroencephalograms of each of the two subjects. The present inventors have confirmed through experiments that "when the amplitude values ​​of one or more predetermined components of the electroencephalograms of one of two people in a conversation are large in response to words included in the conversation, the other person's level of satisfaction with the conversation is high." Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation, including the level of satisfaction with the conversation, from the electroencephalogram responses of each of the two people in a conversation to words included in the conversation.

[0013] The estimation program according to a third aspect may be the estimation program according to the first or second aspect, wherein the one or more predetermined components include at least one of a P2 component and an N400 component. In this configuration, the estimation program causes a computer to estimate the conversation partner's level of satisfaction with the conversation from the amplitude values ​​of at least one of a P2 component and an N400 component of the electroencephalograms of each of the two subjects in response to a word. The present inventors have experimentally confirmed that "when the amplitude values ​​of at least one of a P2 component and an N400 component of the electroencephalograms of one of two people engaged in a conversation are large in response to a word included in the conversation, the other person's level of satisfaction with the conversation is high." Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation, including the level of satisfaction with the conversation, from the amplitude values ​​of at least one of a P2 component and an N400 component of the electroencephalograms of each of the two people engaged in a conversation in response to a word included in the conversation.

[0014] An estimation program according to a fourth aspect is an estimation program according to any one of the first to third aspects, wherein the quality of the conversation includes the emotional valence of each of the two subjects toward the conversation, and in the estimation step, the computer may estimate the emotional valence of one of the two subjects from the amplitude values ​​of one or more specified components of the electroencephalograms of the other of the two subjects, and may estimate the emotional valence of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects.

[0015] In this configuration, the estimation program causes a computer to estimate the emotional valence of the other person in the conversation from the amplitude values ​​of the one or more predetermined components of the electroencephalograms of each of the two subjects. The present inventors have confirmed through experiments that "the amplitude values ​​of one or more predetermined components of the electroencephalograms of one of two people in a conversation in response to words included in the conversation were larger when the other person evaluated the content of the conversation negatively." Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation, including the emotional valence of the conversation, from the electroencephalogram responses of each of the two people in a conversation to words included in the conversation.

[0016] A fifth aspect of the present invention relates to the estimation program of the fourth aspect, wherein the one or more predetermined components may include an LPP (Late Positive Potential) component. In this configuration, the estimation program causes a computer to estimate the valence of the conversation partner's response to the conversation from the amplitude values ​​of the LPP components of the electroencephalograms of each of the two subjects for the words. The present inventors have experimentally confirmed that "the amplitude values ​​of the LPP components of one of two people engaged in a conversation for words included in the conversation were larger when the other person evaluated the content of the conversation negatively." Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation, including the valence of the conversation, from the amplitude values ​​of the LPP components of each of the two people engaged in a conversation for words included in the conversation.

[0017] An estimation program according to a sixth aspect is an estimation program according to any one of the first to fifth aspects, wherein the quality of the conversation includes the degree of interest of each of the two subjects in the conversation, and in the estimation step, the computer may estimate the degree of interest of one of the two subjects from the amplitude value of one or more specified components of the electroencephalogram of one of the two subjects, and estimate the degree of interest of the other of the two subjects from the amplitude value of the one or more specified components of the electroencephalogram of the other of the two subjects.

[0018] In this configuration, the estimation program causes a computer to estimate the level of interest of each of the two subjects in the conversation from the amplitude values ​​of the one or more predetermined components of the electroencephalograms of each of the two subjects. The present inventors have confirmed through experiments that "the amplitude values ​​of one or more predetermined components of the electroencephalograms of one of two people engaged in a conversation in response to words included in the conversation were large when the one person had a high (strong) interest in the content of the conversation." Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation, including the level of interest in the conversation, from the electroencephalogram responses of each of the two subjects engaged in a conversation to words included in the conversation.

[0019] According to a seventh aspect, the estimation program of the sixth aspect may further include a P2 component, wherein the one or more predetermined components include a P2 component. In this configuration, the estimation program causes a computer to estimate the level of interest of each of the two subjects in the conversation from the amplitude values ​​of the P2 component of the electroencephalograms of each of the two subjects in response to the words. The present inventors have experimentally confirmed that "the amplitude value of the P2 component of one of two people engaged in a conversation in response to a word included in the conversation was larger when the person in question had a high (strong) interest in the content of the conversation." Therefore, the estimation program causes a computer to objectively estimate the quality of the conversation, including the level of interest, from the amplitude values ​​of the P2 component of each of the two people engaged in a conversation in response to a word included in the conversation.

[0020] An estimation program according to an eighth aspect is an estimation program according to any one of the first to seventh aspects, wherein the quality of the conversation includes a mutual satisfaction level, which is the degree of satisfaction of both of the two subjects with the conversation, and in the estimation step, the computer may estimate the mutual satisfaction level from the amplitude values ​​of one or more specified components of the electroencephalograms of both of the two subjects.

[0021] In this configuration, the estimation program causes a computer to estimate the mutual satisfaction level, which is the degree of satisfaction of both the two subjects with the conversation, from the amplitude values ​​of the one or more predetermined components of the electroencephalograms of both the two subjects. The present inventors have confirmed through experiments that there is a relationship between "amplitude values ​​of one or more predetermined components of the electroencephalograms of both the two people in a conversation in response to words included in the conversation" and "mutual satisfaction level, which is the degree of satisfaction of both the two people with the conversation." Therefore, the estimation program causes a computer to objectively estimate the quality of a conversation, including the mutual satisfaction level, which is the degree of satisfaction of both the two subjects with the conversation, from the electroencephalogram responses of both the two subjects in a conversation to words included in the conversation.

[0022] A ninth aspect of the present invention relates to the estimation program of the eighth aspect, wherein the one or more predetermined components may include at least one of a P2 component and an LPP component. In this configuration, the estimation program causes a computer to estimate the mutual satisfaction level from amplitude values ​​of at least one of a P2 component and an LPP component of the electroencephalograms of both of the two subjects in response to a word. The present inventors have confirmed through experiments that "when the amplitude values ​​of the P2 component of the electroencephalograms of both of the two people engaged in a conversation in response to a word included in the conversation are large (e.g., both are large) and when the amplitude values ​​of the LPP component are small (e.g., both are small), the mutual satisfaction level, which is the degree of satisfaction of both of the two people with the conversation, is high." Therefore, the estimation program enables a computer to objectively estimate the quality of a conversation, including the mutual satisfaction, which is the degree of satisfaction that both of the two subjects have with the conversation, from the amplitude values ​​of at least one of the P2 component and LPP component of the electroencephalograms of both of the two subjects in a conversation in response to words included in the conversation.

[0023] According to the present invention, it is possible to provide an estimation program or the like that can objectively estimate the quality of a conversation between two subjects from the electroencephalogram responses of each of the subjects to words contained in the conversation.

[0024] FIG. 1 is a diagram illustrating an outline of an experiment conducted by the present inventors. FIG. 2 illustrates an example of a time response function of each subject's electroencephalogram (EEG) for words included in a conversation, as observed in the experiment of FIG. 1. FIG. 3 is a graph showing that, in the time response function illustrated in FIG. 2, the amplitude value of the P2 component increases as the subject becomes more interested in the content of the conversation. FIG. 4 is a graph showing that, in the time response function illustrated in FIG. 2, the amplitude value of the LPP component increases as the content of the conversation becomes more negative for the conversation partner. FIG. 5 schematically illustrates an example of a hardware configuration of an estimation device according to an embodiment. FIG. 6 schematically illustrates an example of a software configuration of an estimation device according to an embodiment. FIG. 7 illustrates an example of a processing procedure of an estimation device according to an embodiment.

[0025] An embodiment according to one aspect of the present invention (hereinafter also referred to as "the present embodiment") will be described below with reference to the drawings. However, the present embodiment described below is merely an example of the present invention in all respects. Needless to say, various improvements and modifications can be made without departing from the scope of the present invention. In other words, when implementing the present invention, specific configurations according to the embodiment may be appropriately adopted. Note that, although data appearing in the present embodiment are described in natural language, more specifically, they are specified using pseudo-language, commands, parameters, machine language, etc. that can be recognized by a computer.

[0026] §1.1. Experimental Overview The present inventors have investigated a method for objectively estimating the conversation quality (CQ) of a conversation (dialogue) between two people, and conducted an experiment illustrated in Figure 1. In this embodiment, the "conversation quality (CQ)" includes at least one of each person's satisfaction with the conversation, each person's emotional valence with the conversation, each person's level of interest in the conversation, and mutual satisfaction, which is the satisfaction of both people with the conversation.

[0027] The present inventors measured the electroencephalograms EE of two subjects Sb who were having a conversation using an electroencephalograph EG. In addition, each subject Sb was fitted with an eye tracker (gaze measurement device, gaze camera) ET to measure the gaze of each subject Sb during the conversation, and each subject Sb's speech UT was collected using a microphone (sound collector) MP worn by each subject Sb.

[0028] In the experiment, the electroencephalograms (EE) of two subjects Sb who were close friends and engaged in a natural conversation were simultaneously measured while the subjects Sb were engaged in the conversation. Specifically, two people who were husband and wife or a couple were considered to be the two subjects Sb engaged in the conversation, and the electroencephalograms (EE) of each subject Sb were measured using an electroencephalogram (EG) while the two subjects Sb were engaged in a natural conversation. "Natural conversation" refers to so-called everyday conversation, for example, conversation about everyday topics. In the experiment, the electroencephalograms (EE) of 20 pairs of two subjects Sb, a total of 40 people, were measured using an electroencephalogram (EG) during conversation.

[0029] FIG. 1 is a diagram illustrating an outline of an experiment conducted by the present inventors. In the experiment, each subject Sb was asked to wear an electroencephalograph EG. In the illustrated example, the electroencephalograph EG(1) was used to measure the brain waves EE(1) of subject Sb(1) during conversation, and the electroencephalograph EG(2) was used to measure the brain waves EE(2) of subject Sb(2) during conversation. Similarly, the eye tracker ET(1) was used to measure the gaze of subject Sb(1) during conversation, and the eye tracker ET(2) was used to measure the gaze of subject Sb(2) during conversation. Furthermore, the microphone MP(1) was used to collect the speech UT(1) of subject Sb(1) during conversation, and the microphone MP(2) was used to collect the speech UT(2) of subject Sb(2) during conversation. In the experiment, the recording sites for electroencephalogram (EE) were Fz, Cz, and Pz of the 10-20 method, with the reference electrode set to the right mastoid and grounded to the left mastoid.

[0030] §1.2. Processing of Measured EEGs The present inventors performed preprocessing such as noise removal and filtering on the EEGs EE (particularly the EEGs EE during conversation) of two subjects Sb who were engaged in a conversation, as measured in the experiment illustrated in Figure 1. Then, for each EEG EE, a temporal response function (TRF) was calculated for each word WD included in the conversation by linear regression with a stimulus sequence created based on annotation information of the conversational voice collected by the microphone MP. The annotation information indicates the start time (onset time) of each of the multiple words WD included in the conversation, and is also referred to as onset information OS.

[0031] For example, suppose that subjects Sb(1) and Sb(2) shown in Fig. 1 have had the following conversation: Subject Sb(1): "What did you do last night?" Subject Sb(2): "We were playing a board game, weren't we?" Subject Sb(1): "Yeah. I wonder who won?" Subject Sb(2): "Yamada-san, right? Until halfway through, it was Sato-san." In the example shown above, the words WD are separated by a "slash ( / )" to distinguish between the multiple words WD included in the conversation.

[0032] In the above example, the onset information OS indicates the start time of each of the multiple words WD(1) included in the utterance UT(1) of the subject Sb(1), such as "yesterday," "night," "what," "were we doing?", "that's right," "who," and "did I win?". That is, the onset information OS indicates the start time of each of the multiple words WD(1) included in the utterance UT(1) of the subject Sb(1). The onset information OS also indicates the start time of each of the multiple words WD(2) included in the utterance UT(2) of the subject Sb(2), such as "we were playing a board game," "Yamada-san, right?", "halfway through," and "it was Sato-san, though." That is, the onset information OS indicates the start time of each of the multiple words WD(2) included in the utterance UT(2) of the subject Sb(2). The onset information OS indicates the start time of each of the multiple words WD included in the utterance UT of each subject Sb.

[0033] The present inventors used this onset information OS (annotation information) to calculate the time response function TRF of the electroencephalogram EE of each subject Sb during conversation with respect to the words WD included in the conversation. In the above example, the time response function TRF(2) of the electroencephalogram EE(2) of subject Sb(2) with respect to the words WD(1) was calculated using the start times of each of the multiple words WD(1) included in the utterance UT(1) of subject Sb(1) indicated by the onset information OS. Similarly, the time response function TRF(1) of the electroencephalogram EE(1) of subject Sb(1) with respect to the words WD(2) was calculated using the start times of each of the multiple words WD(2) included in the utterance UT(2) of subject Sb(2) indicated by the onset information OS. That is, the time response function TRF(1) is the time response function of the electroencephalogram EE(1) of the subject Sb(1) to the word WD(2) contained in the speech UT(2) of the subject Sb(2). Also, the time response function TRF(2) is the time response function of the electroencephalogram EE(2) of the subject Sb(2) to the word WD(1) contained in the speech UT(1) of the subject Sb(1).

[0034] The inventors of the present invention identified the electroencephalogram components P2 component, N400 component, and late positive potential (LPP) component from the calculated time response function TRF, and calculated the amplitude value of each component.

[0035] 2 is a graph illustrating the time response function TRF of the electroencephalograms EE of a certain subject Sb observed in the above-mentioned experiment in response to words WD included in the conversation. In FIG. 2, the dashed-dotted line shows the time response function TRF of the electroencephalograms EE recorded at Fz in response to words WD included in the conversation, the solid line shows the time response function TRF of the electroencephalograms EE recorded at Cz in response to words WD included in the conversation, and the dotted line shows the time response function TRF of the electroencephalograms EE recorded at Pz in response to words WD included in the conversation.

[0036] §1.3. Collection of Subjective Evaluations of Conversations Furthermore, in the experiment illustrated in FIG. 1, the present inventors asked each of the two subjects Sb who had engaged in the conversation to subjectively evaluate their interest, valence, and satisfaction with the content of the conversation after the conversation. Specifically, each subject Sb was asked to indicate their level of interest (interest level) in the content of the conversation on one of several levels (e.g., seven levels from "extremely high" to "extremely low"). Furthermore, each subject Sb was asked to indicate their valence toward the conversation (the content of the conversation) on one of several levels (e.g., seven levels from "extremely positive" to "extremely negative"). Furthermore, each subject Sb was asked to indicate their level of satisfaction (satisfaction level) with the conversation (the content of the conversation) on one of several levels (e.g., seven levels from "extremely satisfied" to "extremely dissatisfied"). Furthermore, the "mutual satisfaction" which is the degree of satisfaction with the conversation between the two subjects Sb was determined from the degree of satisfaction of each subject Sb.

[0037] §1.4. Experimental Results The present inventors investigated the relationship between each EEG component of the time response function TRF and subjective ratings (interest, valence, satisfaction, and mutual satisfaction). For example, the present inventors investigated the relationship between each EEG component and interest and valence. For example, the present inventors evaluated each EEG component using a linear mixed model with interest, valence, and gender as the dependent variable, and subject Sb as the random effect. The present inventors also performed an ordered logistic regression analysis to investigate which EEG components correlate with satisfaction with the conversation and mutual satisfaction. As a result, the present inventors confirmed the following relationships between each EEG component of the time response function TRF and subjective ratings (interest, valence, satisfaction, and mutual satisfaction):

[0038] First, the relationship between each electroencephalogram component of the time response function TRF and subjective ratings of the conversation was evaluated using the linear mixed model described above. As a result, as shown in Figure 3, the amplitude of the P2 component, which is an electroencephalogram component reflecting attention, was large when interest in the conversation content was strong (high). In other words, in the graph shown in Figure 3, the greater the interest in the conversation content (the larger the "Interest Score"), the greater the amplitude of the P2 component of the time response function TRF ("P2 Amplitude").

[0039] For example, the amplitude value of the P2 component in the time response function TRF(1) of subject Sb(1) was larger when subject Sb(1) was more interested in the conversation with subject Sb(2). Furthermore, the amplitude value of the P2 component in the time response function TRF(2) of subject Sb(2) was larger when subject Sb(2) was more interested in the conversation with subject Sb(1). In other words, the present inventors have experimentally confirmed that, for two people having a conversation, the higher the level of interest in the conversation, the larger the amplitude value of the P2 component in the time response function TRF of each person's electroencephalogram EE to the word WD included in the conversation.

[0040] Second, the amplitude value of the LPP component, which is an electroencephalogram component that reflects emotional processing, was larger when the conversation partner evaluated the content of the conversation negatively, as shown in Figure 4. In other words, in the graph shown in Figure 4, the more negatively the conversation partner evaluated the content of the conversation (the smaller the "Valence Score"), the larger the amplitude value of the LPP component of the time response function TRF ("LPP Amplitude").

[0041] For example, the amplitude value of the LPP component in the time response function TRF(1) of subject Sb(1) was larger when the conversation partner, subject Sb(2), evaluated the content of the conversation negatively. Furthermore, the amplitude value of the LPP component in the time response function TRF(2) of subject Sb(2) was larger when the conversation partner, subject Sb(1), evaluated the content of the conversation negatively. In other words, the present inventors have experimentally confirmed that, for each of two people having a conversation, the more negative the emotional valence of the conversation partner is toward the conversation, the larger the amplitude value of the LPP component in the time response function TRF of each person's electroencephalogram EE to the word WD included in the conversation.

[0042] Third, the above-mentioned ordinal logistic regression was performed to examine the relationship between satisfaction with the conversation and each EEG component of the time response function TRF. As a result, the satisfaction level was high when the amplitude values ​​of the P2 component and the N400 component of the conversation partner were large. For example, when the amplitude values ​​of the P2 component and the N400 component in the time response function TRF(1) of subject Sb(1) were large, the satisfaction level of the conversation partner, subject Sb(2), was high. Furthermore, when the amplitude values ​​of the P2 component and the N400 component in the time response function TRF(2) of subject Sb(2) were large, the satisfaction level of the conversation partner, subject Sb(1), was high. In other words, the inventors of the present invention have experimentally confirmed that, for two people having a conversation, the greater the satisfaction that the other person has with the conversation, the greater the amplitude values ​​of the P2 component and the N400 component in the time response function TRF of each person's electroencephalogram EE in response to a word WD included in the conversation.

[0043] Fourth, mutual satisfaction, which is "the degree of satisfaction of both two subjects Sb with the conversation," was high when the amplitude values ​​of the P2 components in the time response functions TRF of each of the two subjects Sb were large (e.g., both large) and the amplitude values ​​of the LPP components were small (e.g., both small). In other words, mutual satisfaction was high when the amplitude values ​​of the P2 components in the time response functions TRF of each of the two subjects Sb were large (e.g., both large) and the amplitude values ​​of the LPP components were small (e.g., both small). "Mutual satisfaction" was highest when both two subjects Sb were "satisfied with the conversation" and lowest when both two subjects Sb were "dissatisfied (unsatisfied) with the conversation." Furthermore, if one of the two subjects Sb was satisfied but the other was not, "mutual satisfaction" was lower than when both subjects Sb were satisfied with the conversation.

[0044] For example, the amplitude value of the P2 component in the time response function TRF(1) of subject Sb(1) and the amplitude value of the P2 component in the time response function TRF(2) of subject Sb(2) increased as the mutual satisfaction level increased. Furthermore, the amplitude value of the LPP component in the time response function TRF(1) of subject Sb(1) and the amplitude value of the LPP component in the time response function TRF(2) of subject Sb(2) decreased as the mutual satisfaction level increased. In other words, the present inventors have experimentally confirmed that the higher the "mutual satisfaction level" of two people engaged in a conversation, the larger the amplitude values ​​of the P2 components in the time response functions TRF of the electroencephalograms of both people to the words included in the conversation (e.g., both become larger) and the smaller the amplitude values ​​of the LPP components (e.g., both become smaller).

[0045] The experiment further confirmed that the more similar (closer) the time intervals TI between the utterances UT of each subject Sb are to each other, the higher the level of mutual satisfaction. For example, it was confirmed that the more similar "the time interval TI(1-2) between the utterance UT(1) of subject Sb(1) and the utterance UT(2) of subject Sb(2) in response to the utterance UT(1)" and "the time interval TI(2-1) between the utterance UT(2) of subject Sb(2) and the utterance UT(1) of subject Sb(1) in response to the utterance UT(2)" are to each other, the higher the level of mutual satisfaction. In the above example, the more similar the time interval TI(1(1)-2(2)) between the utterance UT(1-1) of subject Sb(1) "What were you doing last night?" and the utterance UT(2-2) of subject Sb(2) "We were playing board games, weren't we?" and the more similar the time interval TI(2(2)-1(3)) between the utterance UT(2-2) and the utterance UT(1-3) of subject Sb(1) "That's right. I wonder who won?" are to each other, the higher the "mutual satisfaction," which is the satisfaction of both subjects Sb(1) and Sb(2) with the conversation.

[0046] The inventors have confirmed that by taking into consideration the amplitude values ​​of the P2 component and the LPP component in the time response functions TRF of both people having a conversation, as well as the similarity in the time interval TI of each of the utterances UT of the two people, it is possible to more accurately estimate the "mutual satisfaction" of both people, which is their level of satisfaction with the conversation.

[0047] Using the above-mentioned confirmation results (experimental results), the present inventors have realized an estimation device 1 (estimation program 121) according to this embodiment that "objectively estimates the quality CQ of a conversation between two subjects TA from their electroencephalogram responses to words included in the conversation." Details of the estimation device 1 and the like will be described below with reference to FIGS. 5 to 7.

[0048] §2 Configuration Example [Hardware Configuration] Fig. 5 schematically illustrates an example of the hardware configuration of the estimation device 1 according to this embodiment. As shown in Fig. 5, the estimation device 1 according to this embodiment is a computer to which a control unit 11, a storage unit 12, a communication interface 13, an external interface 14, an input device 15, an output device 16, and a drive 17 are electrically connected. Note that in Fig. 5, the communication interface and the external interface are referred to as a "communication I / F" and an "external I / F."

[0049] The control unit 11 includes a hardware processor such as a central processing unit (CPU), a random access memory (RAM), and a read-only memory (ROM), and is configured to execute information processing based on programs and various data. The CPU is an example of a processor resource. A graphics processing unit (GPU) may be used as the processor resource instead of or in addition to the CPU. The storage unit 12 is an example of a memory resource, and is configured, for example, with a hard disk drive or a solid-state drive. In this embodiment, the storage unit 12 stores various information such as an estimation program 121 and an estimation model 123.

[0050] The estimation program 121 is a program for causing the estimation device 1 to execute information processing ( FIG. 7 ), which will be described later, for performing a predetermined estimation task for at least the amplitude values ​​of one or more predetermined components of the electroencephalogram EE (particularly, the time response function TRF). Details will be described later, but in this embodiment, the estimation device 1 executes the above-described information processing using, for example, an estimation model 123. However, it is not essential for the estimation device 1 to estimate the conversation quality CQ using the estimation model 123, and the estimation device 1 may estimate the conversation quality CQ on a rule-based basis. The estimation program 121 includes a series of instructions for the information processing.

[0051] The estimation model 123 is a trained machine learning model (in other words, a learned statistical model) that is configured to perform an estimation task of estimating conversation quality CQ (in other words, to output an output value corresponding to the result of performing the estimation task) when given at least the amplitude values ​​of one or more predetermined components of the electroencephalogram EE (particularly, the time response function TRF).

[0052] Each training data set used for machine learning of the estimation model 123 is composed of a combination of "amplitude values ​​of one or more predetermined components of the electroencephalograms EE (particularly, time response functions TRF) of two people having a conversation" and "conversation quality CQ (interest level, valence, satisfaction, and mutual satisfaction)." As described above, the "time response functions TRF" are time response functions of the electroencephalograms EE of two people having a conversation to words WD included in the conversation. The "one or more predetermined components" include at least one of the electroencephalogram components P2 component, N400 component, and LPP component. The conversation quality CQ indicates the correct answer to the estimation task for "amplitude values ​​of one or more predetermined components of the electroencephalograms EE (particularly, time response functions TRF)." For example, the conversation quality CQ includes at least one of (1) interest level in the conversation, (2) valence level in the conversation, and (3) satisfaction level in the conversation of each of the two people having a conversation. The conversation quality CQ may also include (4) "mutual satisfaction," which is the degree of satisfaction that both of the two people having the conversation have with the conversation.

[0053] In this embodiment, performing machine learning includes the following training step: training the estimation model 123 so that, for each of the above-described learning data sets, when "amplitude values ​​of one or more predetermined components of electroencephalogram EE (particularly, time response function TRF)" are provided to the estimation model 123, a result of the estimation task of the estimation model 123 conforms to (matches) "conversation quality CQ (interest level, emotional valence, satisfaction, and mutual satisfaction)."

[0054] The estimation model 123 may be constructed by, for example, a random forest. However, the learning method (machine learning method) of the estimation model 123 is not limited to this. As an algorithm for constructing the estimation model 123, a known algorithm such as an algorithm used for machine learning can be used. Examples of machine learning algorithms include, in addition to the random forest, a support vector machine with a linear kernel (SVM linear), a support vector machine with an rbf kernel (SVM rbf), a neural network, a generalized linear model, a regularized linear discriminant analysis, and a regularized logistic regression.

[0055] Details will be described later using FIG. 6 , but in this embodiment, the estimation model 123 includes at least one of a first estimation model 1231, a second estimation model 1233, a third estimation model 1235, and a fourth estimation model 1237. The first estimation model 1231 estimates the level of interest of each subject TA in the conversation from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. The second estimation model 1233 estimates the emotional valence of each subject TA's conversation partner with respect to the conversation from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. The third estimation model 1235 estimates the level of satisfaction of each subject TA's conversation partner with the conversation from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. The fourth estimation model 1237 estimates the degree of mutual satisfaction, which is "the degree of satisfaction of both of the two subjects TA with the conversation," from the amplitude values ​​of one or more predetermined components of the time response functions TRF of both of the two subjects TA. In other words, the fourth estimation model 1237 estimates the above-mentioned degree of mutual satisfaction from the amplitude values ​​of one or more predetermined components of the two time response functions TRF, each of which is the time response function TRF of each subject TA.

[0056] The communication interface 13 is, for example, a wired LAN (Local Area Network) module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network. The estimation device 1 may use this communication interface 13 to perform data communication with other information processing devices via a network. The external interface 14 is, for example, a USB (Universal Serial Bus) port, a dedicated port, or the like, and is an interface for connecting to an external device. The type and number of external interfaces 14 may be selected appropriately depending on the type and number of external devices to be connected. The estimation device 1 may be connected to an electroencephalograph EG via at least one of the communication interface 13 and the external interface 14, which measures the electroencephalograms EE of two subjects TA during a conversation. The estimation device 1 can acquire the electroencephalograms EE of each subject TA during a conversation from each electroencephalograph EG. The estimation device 1 may be connected to an external device via at least one of the communication interface 13 and the external interface 14. The estimation device 1 may acquire from the external device onset information OS (annotation information) indicating the start time of each of a plurality of words WD included in the conversation between the two subjects TA.

[0057] The input device 15 is a device for inputting, for example, a mouse, a keyboard, etc. The output device 16 is a device for outputting, for example, a display, a speaker, etc. An operator such as a user can operate the estimation device 1 by using the input device 15 and the output device 16.

[0058] The drive 17 is, for example, a CD drive, a DVD drive, or the like, and is a drive device for reading various information, such as programs, stored in the storage medium 91. The storage medium 91 is a medium that stores information, such as programs, electrically, magnetically, optically, mechanically, or chemically, so that a computer or other device, machine, or the like can read the stored information. At least one of the estimation program 121 and the estimation model 123 may be stored in the storage medium 91. The estimation device 1 may acquire at least one of the estimation program 121 and the estimation model 123 from the storage medium 91. Note that FIG. 5 illustrates a disk-type storage medium, such as a CD or a DVD, as an example of the storage medium 91. However, the type of the storage medium 91 is not limited to a disk-type storage medium and may be other than a disk-type storage medium. Examples of storage media other than a disk-type storage medium include semiconductor memories, such as flash memories. The type of the drive 17 may be selected arbitrarily depending on the type of the storage medium 91.

[0059] Note that, with regard to the specific hardware configuration of the estimation device 1, components may be omitted, replaced, or added as appropriate depending on the embodiment. For example, the processor resource may include multiple hardware processors. The hardware processor may be configured with a microprocessor, a field-programmable gate array (FPGA), a digital signal processor (DSP), or the like. The storage unit 12 may be configured with RAM and ROM included in the control unit 11. At least one of the communication interface 13, the external interface 14, the input device 15, the output device 16, and the drive 17 may be omitted. The estimation device 1 may be configured with multiple computers. In this case, the hardware configurations of the computers may or may not be the same. Furthermore, the estimation device 1 may be an information processing device designed specifically for the service provided, as well as a general-purpose server device, a PC (Personal Computer), or the like.

[0060] [Software Configuration] Fig. 6 schematically illustrates an example of the software configuration of the estimation device 1 according to this embodiment. The control unit 11 of the estimation device 1 loads the estimation program 121 stored in the storage unit 12 onto the RAM. The control unit 11 then controls each component by interpreting and executing instructions included in the estimation program 121 loaded onto the RAM using the CPU. As a result, the estimation device 1 according to this embodiment operates as a computer including, as software modules, the EEG acquisition unit 110, OS information acquisition unit 120, TRF generation unit 130, estimation unit 140, and output unit 150 illustrated in Fig. 6. That is, in this embodiment, each software module of the estimation device 1 is realized by the control unit 11 (CPU).

[0061] The electroencephalogram acquiring unit 110 acquires the electroencephalograms EE of two subjects TA who are having a conversation. For example, the electroencephalogram acquiring unit 110 acquires the electroencephalograms EE(1) and EE(2) of subjects TA(1) and TA(2) who are having a conversation while they are having the conversation.

[0062] The OS information acquisition unit 120 acquires onset information OS indicating the start time of each of a plurality of words WD included in a conversation between two subjects TA. The onset information OS indicates the "start time of each of a plurality of words WD(1) included in the utterance UT(1) of subject TA(1)" and the "start time of each of a plurality of words WD(2) included in the utterance UT(2) of subject TA(2)."

[0063] The TRF generation unit 130 generates a time response function TRF for each subject TA with respect to words WD included in a conversation, based on the electroencephalogram EE of each subject TA acquired by the electroencephalogram acquisition unit 110 and the onset information OS acquired by the OS information acquisition unit 120. That is, the TRF generation unit 130 generates a time response function TRF for each of the electroencephalograms EE of two subjects TA engaged in a conversation with respect to the words WD included in the conversation. For example, the TRF generation unit 130 generates a time response function TRF(2) for word WD(1) of the electroencephalogram EE(2) of subject TA(2) using the "start times of each of the multiple words WD(1) included in the utterance UT(1) of subject TA(1)" indicated by the onset information OS. Similarly, the TRF generation unit 130 uses the "start time points of each of the multiple words WD(2) included in the utterance UT(2) of the subject TA(2)" indicated by the onset information OS to generate a time response function TRF(1) of the electroencephalogram EE(1) of the subject TA(1) for the word WD(2).

[0064] The estimation unit 140 estimates the quality CQ of the conversation between two subjects TA from the electroencephalograms EE of each subject TA, particularly from the amplitude values ​​of one or more predetermined components in the time response function TRF of each subject TA. That is, the estimation unit 140 estimates the quality CQ of the conversation between the two subjects TA from the amplitude values ​​of one or more predetermined components of the electroencephalograms EE of each of the two subjects TA for a word WD included in the conversation between the two subjects TA. In this embodiment, the estimation unit 140 estimates, as the quality CQ of the conversation, (1) the degree of interest in the conversation, (2) the emotional valence of the conversation, and (3) the level of satisfaction with the conversation for each of the two subjects TA. Furthermore, the estimation unit 140 estimates, as the quality CQ of the conversation, (4) mutual satisfaction, which is "the level of satisfaction of both subjects TA who are having the conversation." However, it is not essential for the estimation unit 140 to estimate all four of the above-mentioned conversation qualities CQ (interest level, valence, satisfaction level, and mutual satisfaction level); it is sufficient to estimate at least one of these four from the amplitude value of one or more predetermined components in the time response function TRF of each subject TA. In the illustrated example, the estimation unit 140 includes a first extraction unit 1412, a first estimation unit 1414, a second extraction unit 1422, a second estimating unit 1424, a third extraction unit 1432, a third estimating unit 1434, a fourth extraction unit 1442, and a fourth estimating unit 1444.

[0065] (Estimation of Degree of Interest in Conversation) The first extraction unit 1412 extracts one or more predetermined components (predetermined electroencephalogram components) of the time response function TRF of each subject TA. For example, the first extraction unit 1412 extracts the P2 component of the time response function TRF of each subject TA and identifies the amplitude value of the P2 component of each extracted time response function TRF. The first extraction unit 1412 notifies the first estimation unit 1414 of the identified "amplitude value of the P2 component of the time response function TRF of each subject TA." In other words, the first extraction unit 1412 identifies the "amplitude value of the P2 component for words WD included in the conversation" for the electroencephalograms EE of each subject TA and notifies the first estimation unit 1414 of the identified "amplitude value of the P2 component of the electroencephalograms EE of each subject TA."

[0066] The first estimation unit 1414 estimates the degree of interest of each subject TA in the conversation from the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the first extraction unit 1412. For example, the first estimation unit 1414 estimates the degree of interest of subject TA(1) in the conversation with subject TA(2) from the amplitude values ​​of one or more predetermined components of the time response function TRF(1) of subject TA(1). Furthermore, the first estimation unit 1414 estimates the degree of interest of subject TA(2) in the conversation with subject TA(1) from the amplitude values ​​of one or more predetermined components of the time response function TRF(2) of subject TA(2). In other words, the first estimation unit 1414 estimates the level of interest of the subject TA(1) in the conversation from the "amplitude values ​​of one or more predetermined components" of the electroencephalogram EE(1) of the subject TA(1) (particularly, the "amplitude values ​​of one or more predetermined components" for the word WD(2) included in the utterance UT(2) of the subject TA(2). In addition, the first estimation unit 1414 estimates the level of interest of the subject TA(2) in the conversation from the "amplitude values ​​of one or more predetermined components" of the electroencephalogram EE(2) of the subject TA(2) (particularly, the "amplitude values ​​of one or more predetermined components" for the word WD(1) included in the utterance UT(1) of the subject TA(1).

[0067] In this embodiment, the first estimation unit 1414 estimates the degree of interest of each subject TA in the conversation from the "amplitude value of the P2 component of the time response function TRF of each subject TA" identified by the first extraction unit 1412. For example, the first estimation unit 1414 estimates the degree of interest of subject TA(1) in the conversation with subject TA(2) from the amplitude value of the P2 component of the time response function TRF(1) of subject TA(1). Furthermore, the first estimation unit 1414 estimates the degree of interest of subject TA(2) in the conversation with subject TA(1) from the amplitude value of the P2 component of the time response function TRF(2) of subject TA(2).

[0068] For example, the first estimation unit 1414 acquires the first estimation model 1231 by referring to the storage unit 12. Then, the first estimation unit 1414 inputs the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the first extraction unit 1412 into the acquired first estimation model 1231, thereby estimating the degree of interest in the conversation of each subject TA.

[0069] The first estimation model 1231 is a trained machine learning model that is trained to output “the level of interest of each of the two people in the conversation” from “the amplitude values ​​of the P2 components of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation.” Each training data set used for the machine learning of the first estimation model 1231 is composed of, for example, a combination of “the amplitude values ​​of the P2 components of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation” and “the level of interest of each of the two people in the conversation.” Then, for each of the training data sets, the first estimation model 1231 is trained so that when the “amplitude values ​​of the P2 components of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation” are provided to the first estimation model 1231, the result of performing the estimation task of the first estimation model 1231 conforms to (matches) “the level of interest of each of the two people in the conversation.”

[0070] The first estimation unit 1414 estimates the degree of interest of each subject TA in the conversation by inputting the “amplitude value of the P2 component of the time response function TRF of each subject TA” into the first estimation model 1231.

[0071] In the above example, the first extraction unit 1412 and the first estimation unit 1414 have been described as two different functional units. However, the first extraction unit 1412 and the first estimation unit 1414 may be configured as an integrated unit, for example, as a neural network (neural network module). Such a neural network may, for example, extract the amplitude value of the P2 component as a feature value from the time response function TRF of each subject TA, and estimate the degree of interest in the conversation of each subject TA from the extracted amplitude value of the P2 component.

[0072] Note that the first estimation unit 1414 does not necessarily need to use a trained machine learning model (first estimation model 1231) to estimate each subject TA's level of interest in the conversation from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. The first estimation unit 1414 may rule-based estimate each subject TA's level of interest in the conversation from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. As described above, the higher the level of interest in the conversation between two people having a conversation, the larger the amplitude value of the P2 component in the time response function TRF of each person's electroencephalogram EE for a word WD included in the conversation. Therefore, by utilizing the relationship between the amplitude value of the P2 component and the level of interest in the conversation, the first estimation unit 1414 may estimate each subject TA's level of interest in the conversation from the amplitude value of one or more predetermined components (e.g., the P2 component) of the time response function TRF of each subject TA.

[0073] The estimation unit 140 estimates the quality CQ of the conversation that the two subjects TA have had (are having) in accordance with the "degree of interest in the conversation" of each subject TA estimated by the first estimation unit 1414. For example, if the "degree of interest in the conversation" of each subject TA estimated by the first estimation unit 1414 is both high, the estimation unit 140 may estimate that the quality CQ of the conversation that the two subjects TA have had is "high." Alternatively, if the "degree of interest in the conversation" of each subject TA estimated by the first estimation unit 1414 is both low, the estimation unit 140 may estimate that the quality CQ of the conversation that the two subjects TA have had is "low."

[0074] (Estimation of Emotional Valence for Conversation) The second extraction unit 1422 extracts one or more predetermined components of the time response function TRF of each subject TA. For example, the second extraction unit 1422 extracts the LPP component of the time response function TRF of each subject TA and identifies the amplitude value of the LPP component of each extracted time response function TRF. The second extraction unit 1422 notifies the second estimation unit 1424 of the identified "amplitude value of the LPP component of the time response function TRF of each subject TA." In other words, the second extraction unit 1422 identifies the "amplitude value of the LPP component for words WD included in the conversation" for the electroencephalogram EE of each subject TA and notifies the second estimation unit 1424 of the identified "amplitude value of the LPP component of the electroencephalogram EE of each subject TA."

[0075] The second estimation unit 1424 estimates the valence of the conversation partner of each subject TA from the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the second extraction unit 1422. For example, the second estimation unit 1424 estimates the valence of subject TA(2) from the conversation with subject TA(1) from the amplitude values ​​of one or more predetermined components of the time response function TRF(1) of subject TA(1). Furthermore, the second estimation unit 1424 estimates the valence of subject TA(1) from the conversation with subject TA(2) from the amplitude values ​​of one or more predetermined components of the time response function TRF(2) of subject TA(2). In other words, the second estimation unit 1424 estimates the emotional valence of the subject TA(2) with respect to the conversation from the "amplitude values ​​of one or more predetermined components" of the electroencephalogram EE(1) of the subject TA(1) (particularly, the "amplitude values ​​of one or more predetermined components" for the word WD(2) included in the utterance UT(2) of the subject TA(2). In addition, the second estimation unit 1424 estimates the emotional valence of the subject TA(1) with respect to the conversation from the "amplitude values ​​of one or more predetermined components" of the electroencephalogram EE(2) of the subject TA(2) (particularly, the "amplitude values ​​of one or more predetermined components" for the word WD(1) included in the utterance UT(1) of the subject TA(1).

[0076] In this embodiment, the second estimation unit 1424 estimates the valence of the conversation of each subject TA with respect to the conversation of the other party in the conversation from the "amplitude value of the LPP component of the time response function TRF of each subject TA" identified by the second extraction unit 1422. For example, the second estimation unit 1424 estimates the valence of subject TA(2) with respect to the conversation with subject TA(1) from the amplitude value of the LPP component of the time response function TRF(1) of subject TA(1). Furthermore, the second estimation unit 1424 estimates the valence of subject TA(1) with respect to the conversation with subject TA(2) from the amplitude value of the LPP component of the time response function TRF(2) of subject TA(2).

[0077] For example, the second estimation unit 1424 acquires the second estimation model 1233 by referring to the storage unit 12. Then, the second estimation unit 1424 inputs the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the second extraction unit 1422 into the acquired second estimation model 1233, thereby estimating the emotional valence of the conversation partner of each subject TA.

[0078] The second inference model 1233 is a trained machine learning model that is trained to output "the emotional valence of the conversation partner with respect to the conversation" from "the amplitude values ​​of the LPP components of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation." Each training data set used for the machine learning of the second inference model 1233 is composed of, for example, a combination of "the amplitude values ​​of the LPP components of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation" and "the emotional valence of the conversation partner with respect to the conversation." Then, for each of the above-mentioned training data sets, the second inference model 1233 is trained so that when "the amplitude values ​​of the LPP components of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation" are provided to the second inference model 1233, the result of performing the estimation task of the second inference model 1233 conforms to (matches) "the emotional valence of the conversation partner with respect to the conversation."

[0079] The second estimation unit 1424 estimates the emotional valence of the conversation partner of each subject TA by inputting the ``amplitude value of the LPP component of the time response function TRF of each subject TA'' into the second estimation model 1233.

[0080] In the above example, the second extraction unit 1422 and the second estimation unit 1424 have been described as two different functional units. However, the second extraction unit 1422 and the second estimation unit 1424 may be configured as an integrated unit, for example, as a neural network (neural network module). Such a neural network may, for example, extract the amplitude value of the LPP component as a feature value from the time response function TRF of each subject TA, and estimate the emotional valence of the conversation partner of each subject TA from the amplitude value of the extracted LPP component.

[0081] Note that it is not essential for the second estimation unit 1424 to "use a trained machine learning model (second estimation model 1233)" to estimate the valence of the conversation partner of each subject TA from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. The second estimation unit 1424 may rule-based estimate the valence of the conversation partner of each subject TA from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. As described above, for each of two people having a conversation, the more negative the valence of the conversation partner is toward the conversation, the larger the amplitude value of the LPP component in the time response function TRF of each person's electroencephalogram EE for words WD included in the conversation. Therefore, by utilizing the relationship between the amplitude value of the LPP component and the "emotional valence of the conversation partner toward the conversation," the second estimation unit 1424 may estimate the emotional valence of the conversation partner of each subject TA toward the conversation from the amplitude value of one or more specified components (e.g., LPP components) of the time response function TRF of each subject TA.

[0082] The estimation unit 140 estimates the conversation quality CQ of the conversation that the two subjects TA have had (are having) in accordance with the "emotional valence toward the conversation" of each subject TA estimated by the second estimation unit 1424. For example, if the "emotional valence toward the conversation" of each subject TA estimated by the second estimation unit 1424 is both high (positive), the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "high." Alternatively, if the "emotional valence toward the conversation" of each subject TA estimated by the second estimation unit 1424 is both low (negative), the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "low."

[0083] (Estimation of Satisfaction with Conversation) The third extraction unit 1432 extracts one or more predetermined components of the time response function TRF of each subject TA. For example, the third extraction unit 1432 extracts the P2 component and the N400 component of the time response function TRF of each subject TA and identifies the amplitude values ​​of the P2 component and the N400 component of each extracted time response function TRF. The third extraction unit 1432 notifies the third estimation unit 1434 of the identified "amplitude values ​​of the P2 component and the N400 component of the time response function TRF of each subject TA." That is, the third extraction unit 1432 identifies "amplitude values ​​of the P2 component and the N400 component for words WD included in the conversation" for the electroencephalogram EE of each subject TA and notifies the third estimation unit 1434 of the identified "amplitude values ​​of the P2 component and the N400 component of the electroencephalogram EE of each subject TA."

[0084] The third estimation unit 1434 estimates the degree of satisfaction of each subject TA's conversation partner with the conversation from the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the third extraction unit 1432. For example, the third estimation unit 1434 estimates the degree of satisfaction of subject TA(2) with the conversation with subject TA(1) from the amplitude values ​​of one or more predetermined components of the time response function TRF(1) of subject TA(1). Furthermore, the third estimation unit 1434 estimates the degree of satisfaction of subject TA(1) with the conversation with subject TA(2) from the amplitude values ​​of one or more predetermined components of the time response function TRF(2) of subject TA(2). In other words, the third estimation unit 1434 estimates the subject TA(2)'s satisfaction with the conversation from the "amplitude values ​​of one or more predetermined components" of the electroencephalogram EE(1) of the subject TA(1) (particularly, the "amplitude values ​​of one or more predetermined components" for the word WD(2) included in the utterance UT(2) of the subject TA(2). In addition, the third estimation unit 1434 estimates the subject TA(1)'s satisfaction with the conversation from the "amplitude values ​​of one or more predetermined components" of the electroencephalogram EE(2) of the subject TA(2) (particularly, the "amplitude values ​​of one or more predetermined components" for the word WD(1) included in the utterance UT(1) of the subject TA(1).

[0085] In this embodiment, the third estimation unit 1434 estimates the satisfaction level of each subject TA's conversation partner with the conversation from the "amplitude values ​​of the P2 component and the N400 component of the time response function TRF of each subject TA" identified by the third extraction unit 1432. For example, the third estimation unit 1434 estimates the subject TA(2)'s satisfaction level with the conversation with subject TA(1) from the amplitude values ​​of the P2 component and the N400 component of the time response function TRF(1) of subject TA(1). Furthermore, the third estimation unit 1434 estimates the subject TA(1)'s satisfaction level with the conversation with subject TA(2) from the amplitude values ​​of the P2 component and the N400 component of the time response function TRF(2) of subject TA(2).

[0086] For example, the third estimation unit 1434 acquires the third estimation model 1235 by referring to the storage unit 12. Then, the third estimation unit 1434 inputs the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the third extraction unit 1432 into the acquired third estimation model 1235, thereby estimating the satisfaction level of the conversation partner of each subject TA with respect to the conversation.

[0087] The third estimation model 1235 is a trained machine learning model that is trained to output "the conversation partner's level of satisfaction with the conversation" from "the amplitude values ​​of the P2 component and the N400 component of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation." Each training data set used for the machine learning of the third estimation model 1235 is composed of, for example, a combination of "the amplitude values ​​of the P2 component and the N400 component of the electroencephalograms EE of each of the two people having a conversation in response to a word WD included in the conversation" and "the conversation partner's level of satisfaction with the conversation." Then, for each of the above-mentioned learning data sets, when the third estimation model 1235 is given "the amplitude values ​​of the P2 component and the N400 component of the electroencephalogram EE of each of two people having a conversation in response to a word WD included in the conversation," the third estimation model 1235 is trained so that the result of performing the estimation task of the third estimation model 1235 conforms to (matches) "the satisfaction level of the conversation partner with the conversation."

[0088] The third estimation unit 1434 estimates the satisfaction level of the conversation partner of each subject TA by inputting the "amplitude value of the P2 component and the amplitude value of the N400 component of the time response function TRF of each subject TA" into the third estimation model 1235.

[0089] In the above example, the third extraction unit 1432 and the third estimation unit 1434 have been described as two different functional units. However, the third extraction unit 1432 and the third estimation unit 1434 may be configured as an integrated unit, for example, as a neural network (neural network module). Such a neural network may, for example, extract the amplitude values ​​of the P2 component and the N400 component as feature quantities from the time response function TRF of each subject TA, and estimate the satisfaction level of the conversation partner of each subject TA from the extracted amplitude values ​​of the P2 component and the N400 component.

[0090] Note that it is not essential for the third estimation unit 1434 to "use a trained machine learning model (third estimation model 1235)" to estimate the conversation partner's satisfaction with each subject TA from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. The third estimation unit 1434 may rule-based estimate the conversation partner's satisfaction with each subject TA from the amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA. As described above, for each of two people having a conversation, the higher the conversation partner's satisfaction with the conversation, the larger the amplitude values ​​of the P2 component and the N400 component of the time response function TRF of each person's electroencephalogram EE to words WD included in the conversation. Therefore, by utilizing the relationship between the "amplitude values ​​of the P2 component and the N400 component" and the "conversational partner's level of satisfaction with the conversation," the third estimation unit 1434 may estimate the conversational partner's level of satisfaction with the conversation from the amplitude values ​​of one or more specified components (e.g., the P2 component and the N400 component) of the time response function TRF of each subject TA.

[0091] The estimation unit 140 estimates the conversation quality CQ of the conversation that the two subjects TA have had (are having) in accordance with the "satisfaction with the conversation" of each subject TA estimated by the third estimation unit 1434. For example, if the "satisfaction with the conversation" of each subject TA estimated by the third estimation unit 1434 is both high, the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "high." Furthermore, if the "satisfaction with the conversation" of each subject TA estimated by the third estimation unit 1434 is both low, the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "low."

[0092] (Estimation of Mutual Satisfaction Degree with Conversation) The fourth extraction unit 1442 extracts one or more predetermined components of the time response function TRF of each subject TA. For example, the fourth extraction unit 1442 extracts the P2 component and the LPP component of the time response function TRF of each subject TA and identifies the amplitude values ​​of the P2 component and the LPP component of each extracted time response function TRF. The fourth extraction unit 1442 notifies the fourth estimation unit 1444 of the identified "amplitude values ​​of the P2 component and the LPP component of the time response function TRF of each subject TA." That is, the fourth extraction unit 1442 identifies "amplitude values ​​of the P2 component and the LPP component for words WD included in the conversation" for the electroencephalogram EE of each subject TA and notifies the fourth estimation unit 1444 of the identified "amplitude values ​​of the P2 component and the LPP component of the electroencephalogram EE of each subject TA."

[0093] The fourth estimation unit 1444 estimates the degree of mutual satisfaction, which is "the degree of satisfaction of both two subjects TA with the conversation," from the "amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA" identified by the fourth extraction unit 1442. For example, the fourth estimation unit 1444 estimates the degree of mutual satisfaction, which is "the degree of satisfaction of both subjects TA(1) and TA(2) with the conversation," from the amplitude values ​​of one or more predetermined components of the time response function TRF(1) of subject TA(1) and the amplitude values ​​of one or more predetermined components of the time response function TRF(2) of subject TA(2). In other words, the fourth estimation unit 1444 estimates the above-mentioned mutual satisfaction level from the "amplitude value of one or more predetermined components" of the electroencephalogram EE(1) of the subject TA(1) (particularly, the "amplitude value of one or more predetermined components" for the word WD(2) included in the utterance UT(2) of the subject TA(2)) and the "amplitude value of one or more predetermined components" of the electroencephalogram EE(2) of the subject TA(2) (particularly, the "amplitude value of one or more predetermined components" for the word WD(1) included in the utterance UT(1) of the subject TA(1)).

[0094] In the present embodiment, the fourth estimation unit 1444 estimates the degree of mutual satisfaction, which is "the degree of satisfaction of both two subjects TA with the conversation," from both "the amplitude values ​​of the P2 component and the LPP component of the time response function TRF of each subject TA" identified by the fourth extraction unit 1442. For example, the fourth estimation unit 1444 estimates the degree of mutual satisfaction, which is "the degree of satisfaction of both subjects TA(1) and TA(2) with the conversation," from the amplitude values ​​of the P2 component and the LPP component of the time response function TRF(1) of the subject TA(1) and the amplitude values ​​of the P2 component and the LPP component of the time response function TRF(2) of the subject TA(2).

[0095] For example, the fourth estimation unit 1444 acquires the fourth estimation model 1237 by referring to the storage unit 12. Then, the fourth estimation unit 1444 estimates the mutual satisfaction degree by inputting both of the “amplitude values ​​of one or more predetermined components of the time response function TRF of each subject TA” identified by the fourth extraction unit 1442 into the acquired fourth estimation model 1237.

[0096] The fourth estimation model 1237 is a trained machine learning model that is trained to output "mutual satisfaction, which is the 'satisfaction of both two people having a conversation with the conversation,'" from "amplitude values ​​of the P2 component and the LPP component of the electroencephalograms EE of both two people having a conversation in response to a word WD included in the conversation." Each training dataset used for the machine learning of the fourth estimation model 1237 is composed of, for example, a combination of "amplitude values ​​of the P2 component and the LPP component of the electroencephalograms EE of both two people having a conversation in response to a word WD included in the conversation" and "mutual satisfaction, which is the 'satisfaction of both two people having a conversation with the conversation.'" Then, for each of the above-mentioned training data sets, the fourth estimation model 1237 is trained so that when "amplitude values ​​of the P2 component and amplitude values ​​of the LPP component of the electroencephalograms EE of both of two conversing people in response to a word WD included in the conversation" are provided to the fourth estimation model 1237, the result of performing the estimation task of the fourth estimation model 1237 conforms to (matches) the above-mentioned mutual satisfaction level. For example, when both amplitude values ​​of the P2 component and amplitude values ​​of the LPP component of the electroencephalograms EE of two conversing people A and B in response to a word WD included in the conversation are provided to the fourth estimation model 1237, the fourth estimation model 1237 is trained so that the result of performing the estimation task of the fourth estimation model 1237 conforms to (matches) the mutual satisfaction level, which is "the satisfaction level of both of the two people A and B with the conversation."

[0097] The fourth estimation unit 1444 estimates mutual satisfaction, which is "the degree of satisfaction of both of the two subjects TA with the conversation," by inputting "the amplitude values ​​of the P2 component and the LPP component of the time response functions TRF of both of the two subjects TA" into the fourth estimation model 1237.

[0098] Here, as described above, the greater the similarity of the time intervals TI between the utterances UT of two people having a conversation (the closer the time intervals TI are to each other), the higher the "mutual satisfaction" that is the level of satisfaction that both of the two people have with the conversation. In other words, the greater the similarity between the "time interval TI(AB) from the utterance of one person A to the utterance that the other person B makes in response to that utterance" and the "time interval TI(BA) from the utterance of the other person B to the utterance that the other person A makes in response to that utterance," the higher the "mutual satisfaction" that is the level of satisfaction that both people A and B have. Thus, each training data set used for machine learning of the fourth estimation model 1237 may be configured by a combination of, for example, “amplitude values ​​of the P2 component and the LPP component of the electroencephalograms EE of both of the two people having a conversation in response to a word WD included in the conversation,” “similarity between the time intervals TI of the utterances UT of each of the two people having a conversation,” and “mutual satisfaction, which is the degree of satisfaction of both of the two people having a conversation with the conversation.” Then, for each of the above-mentioned training data sets, the fourth estimation model 1237 may be trained so that when the “amplitude values ​​of the P2 component and the LPP component of the electroencephalograms EE of both of the two people having a conversation in response to a word WD included in the conversation” and “similarity between the time intervals TI of the utterances UT of each of the two people having a conversation” are provided to the fourth estimation model 1237, the fourth estimation model 1237 performs a training such that a result of performing an estimation task of the fourth estimation model 1237 conforms to (matches) the above-mentioned mutual satisfaction.

[0099] When each learning data set used for the machine learning of the fourth estimation model 1237 includes "the amplitude values ​​of the P2 component and the LPP component of the electroencephalograms EE of both of the two people having a conversation in response to words WD included in the conversation" and "mutual satisfaction, which is the degree of satisfaction of both of the two people having a conversation with the conversation," as well as "the similarity of the time intervals TI of the utterances UT of the two people having a conversation," the fourth estimation unit 1444 may input the following values ​​to the fourth estimation model 1237. That is, the fourth estimation unit 1444 may input "the similarity of the time intervals TI of the utterances UT of each subject TA" to the fourth estimation model 1237 in addition to "the amplitude values ​​of the P2 component and the LPP component of the time response functions TRF of both of the two subjects TA." With this input, the fourth estimation unit 1444 can use the above-described fourth estimation model 1237 to more accurately estimate mutual satisfaction, which is "the degree of satisfaction of both two subjects TA with the conversation." The "similarity of the time interval TI of the utterances UT of each subject TA" refers to the similarity of the following two time intervals TI. That is, it is the similarity between "the time interval TI(1-2) between the utterance UT(1) of subject TA(1) and the utterance UT(2) of subject TA(2) as a response to the utterance UT(1)" and "the time interval TI(2-1) between the utterance UT(2) of subject TA(2) and the utterance UT(1) of subject TA(1) as a response to the utterance UT(2)."

[0100] In the above example, the fourth extraction unit 1442 and the fourth estimation unit 1444 have been described as two different functional units. However, the fourth extraction unit 1442 and the fourth estimation unit 1444 may be configured as an integrated unit, for example, as a neural network (neural network module). Such a neural network may, for example, extract the amplitude values ​​of the P2 component and the LPP component as feature quantities from the time response function TRF of each subject TA, and estimate mutual satisfaction, which is "the degree of satisfaction of both subjects TA with the conversation," from the extracted amplitude values ​​of the P2 component and the LPP component.

[0101] Note that it is not essential for the fourth estimation unit 1444 to "use a trained machine learning model (fourth estimation model 1237)" to estimate the mutual satisfaction, which is "the degree of satisfaction of both two subjects TA with the conversation," from the amplitude values ​​of one or more predetermined components of the time response functions TRF of both two subjects TA. The fourth estimation unit 1444 may rule-based estimate the mutual satisfaction, which is "the degree of satisfaction of both two subjects TA with the conversation," from the amplitude values ​​of one or more predetermined components of the time response functions TRF of both two subjects TA. As described above, the higher the "mutual satisfaction," which is the degree of satisfaction of both two people having a conversation, the larger the amplitude value of the P2 component and the smaller the amplitude value of the LPP component of the time response functions TRF of the electroencephalograms of both two people to words included in the conversation. Therefore, by utilizing the relationship between the "amplitude value of the P2 component and the amplitude value of the LPP component" and the degree of mutual satisfaction, the fourth estimation unit 1444 may estimate the degree of mutual satisfaction, which is "the degree of satisfaction with the conversation of both of the two subjects TA," from the amplitude values ​​of one or more specified components (e.g., the P2 component and the LPP component) of the time response functions TRF of both of the two subjects TA.

[0102] The estimation unit 140 estimates the conversation quality CQ of the conversation that the two subjects TA have had (are having) in accordance with the mutual satisfaction level, which is "the degree of satisfaction of both subjects TA with the conversation," estimated by the fourth estimation unit 1444. For example, if the mutual satisfaction level estimated by the fourth estimation unit 1444 is high, the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "high." Furthermore, if the mutual satisfaction level estimated by the fourth estimation unit 1444 is low, the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "low."

[0103] The estimation unit 140 may estimate the conversation quality CQ of the conversation that the two subjects TA have had (are having) by combining the estimated interest level, valence, satisfaction level, and mutual satisfaction level. For example, if the estimated interest level, valence, satisfaction level, and mutual satisfaction level are all high, the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "high." Alternatively, if the estimated interest level, valence, satisfaction level, and mutual satisfaction level are all low, the estimation unit 140 may estimate that the conversation quality CQ of the conversation that the two subjects TA have had is "low."

[0104] The output unit 150 outputs the "quality CQ (interest level, emotional valence, satisfaction level, and mutual satisfaction level) of the conversation between the two subjects TA" estimated by the estimation unit 140 as an estimation result ER, and displays the estimation result ER on the output device 16, for example.

[0105] §3 Operation Example Figure 7 is a flowchart showing an example of the processing procedure of the estimation device 1 according to this embodiment. The processing procedure described below is an example of the processing procedure of an information processing method PM that "estimates the quality CQ of a conversation between two subjects TA from the electroencephalogram response of each subject TA to a word WD included in the conversation." However, the processing procedure described below is merely an example, and each step may be modified as much as possible. Furthermore, steps in the processing procedure described below may be omitted, replaced, or added as appropriate depending on the embodiment.

[0106] (Step S110, EEG Acquisition Step) In step S110, the control unit 11 operates as the EEG acquisition unit 110 and acquires the EEGs EE of each of the two subjects TA who are having a conversation (i.e., during a conversation). For example, the control unit 11 acquires the EEGs EE(1) and EE(2) of the subjects TA(1) and TA(2) who are having a conversation while the conversation is taking place.

[0107] (Step S120, Onset Information Acquisition Step) In step S120, the control unit 11 operates as the OS information acquisition unit 120 and acquires onset information OS indicating the start time of each of the multiple words WD included in the conversation between the two subjects TA. For example, the control unit 11 acquires onset information OS indicating the "start time of each of the multiple words WD(1) included in the utterance UT(1) of the subject TA(1)" and the "start time of each of the multiple words WD(2) included in the utterance UT(2) of the subject TA(2)."

[0108] (Step S130, Time Response Function Generation Step) In step S130, the control unit 11 operates as the TRF generation unit 130, and generates a time response function TRF for each subject TA with respect to words WD included in the conversation, based on the electroencephalograms EE of each subject TA acquired in step S110 and the onset information OS acquired in step S120. For example, the control unit 11 generates a time response function TRF(2) for words WD(1) of the electroencephalograms EE(2) of subject TA(2) using the "start times of each of the multiple words WD(1) included in the utterance UT(1) of subject TA(1)" indicated by the onset information OS. Similarly, the control unit 11 uses the "start time points of each of the multiple words WD(2) included in the utterance UT(2) of the subject TA(2)" indicated by the onset information OS to generate a time response function TRF(1) of the electroencephalogram EE(1) of the subject TA(1) for the word WD(2).

[0109] (Step S140, Estimation Step) In step S140, the control unit 11 operates as the estimation unit 140 and estimates the conversation quality CQ of the conversation between two subjects TA from the amplitude values ​​of one or more predetermined components of each time response function TRF generated in S130. For example, the control unit 11 operates as the first extraction unit 1412 and the first estimation unit 1414 and estimates the level of interest of each subject TA in the conversation from the amplitude value of the P2 component in the time response function TRF of each subject TA. For example, the control unit 11 operates as the second extraction unit 1422 and the second estimation unit 1424 and estimates the emotional valence of the conversation partner of each subject TA from the amplitude value of the LPP component in the time response function TRF of each subject TA. For example, the control unit 11 operates as a third extraction unit 1432 and a third estimation unit 1434, and estimates the degree of satisfaction of the conversation partner of each subject TA from the amplitude values ​​of the P2 component and the N400 component in the time response function TRF of each subject TA. For example, the control unit 11 operates as a fourth extraction unit 1442 and a fourth estimating unit 1444, and estimates mutual satisfaction, which is "the degree of satisfaction of both subjects TA with the conversation," from the amplitude values ​​of the P2 component and the LPP component in the time response function TRF of both of the two subjects TA.

[0110] [Features] As described above, in this embodiment, the estimation program 121 causes the estimation device 1 (computer) to execute S110 (EEG acquisition step), S120 (onset information acquisition step), and S140 (estimation step). In S110, the estimation device 1 (e.g., the EEG acquisition unit 110) acquires the EEGs EE of each of two subjects TA who are having a conversation. In S120, the estimation device 1 (e.g., the OS information acquisition unit 120) acquires onset information OS indicating the start time of each of multiple words WD included in the conversation between the two subjects TA. In S140, the estimation device 1 (e.g., the estimation unit 140) estimates the conversation quality CQ of the conversation from the amplitude values ​​of one or more predetermined components of the EEGs EE of each of the two subjects TA for the words WD.

[0111] In this configuration, the estimation program 121 causes the estimation device 1 to estimate the conversation quality CQ of a conversation from the amplitude values ​​of one or more predetermined components of the electroencephalograms EE of each of the two subjects TA who are having a conversation in response to words WD included in the conversation. In other words, the estimation device 1 objectively estimates the conversation quality CQ of a conversation from the electroencephalogram responses of each of the two subjects TA who are having a conversation to the words included in the conversation. Therefore, the estimation program 121 can objectively estimate the conversation quality CQ of a conversation from the electroencephalogram responses of each of the two subjects TA who are having a conversation to the words included in the conversation.

[0112] When conversing with others, paying appropriate attention to what the other person is saying is essential for smooth information exchange; insufficient information exchange can lead to misunderstandings and a decline in the quality of the conversation. Furthermore, conversation is not simply a means of transmitting information; it also plays an important role in building intimate relationships through the sharing of emotions. Without being able to properly understand and empathize with the other person's emotions, it becomes difficult to maintain an intimate relationship. While attention and emotional understanding are important elements for achieving high-quality conversation, evaluation of everyday casual conversations is carried out at an abstract level and is summarized in the subjective experience of satisfaction. The elements of attention, emotional understanding, and the aggregate experience of satisfaction are all internal psychological states of each individual, making them difficult to evaluate objectively.

[0113] If there were a method for objectively evaluating an individual's internal psychological state, such as attention, emotional understanding, and subjective satisfaction, from brain activity, it would be possible to provide effective training for achieving high-quality conversations and evaluate their effectiveness. Furthermore, such a method could potentially enable us to understand and improve what kind of robots and AI can attract people's attention, effectively convey emotional information, and achieve satisfying conversations. Particularly in today's world, where interactions with robots, AI, and other devices are increasing in our daily lives, there is a strong demand for establishing such a method. Furthermore, in order to develop effective training for improving conversation quality, evaluate the results of such training, and provide effective feedback to interactive robots and AI, indicators for objectively measuring specific psychological states are necessary.

[0114] The previously known phenomenon of "synchronization of neural activity among multiple people" is a promising tool for assessing the quality of social interactions, including conversations. However, because neural synchronization is generally assessed using data over a period of a few minutes, it is difficult to understand what causes enhanced neural synchronization during a conversation. This means that it is difficult to objectively assess specific psychological states, such as attention or emotional understanding, and it is also difficult to understand what psychological states lead to satisfaction or dissatisfaction.

[0115] As described above, the estimation device 1 can objectively estimate the conversation quality CQ of a conversation between two subjects TA based on their respective EEG responses to words contained in the conversation. By objectively estimating (evaluating) the conversation quality CQ of a conversation between two subjects TA based on the EEG responses of each subject TA, it is possible to understand the conditions necessary for realizing high-quality conversation between humans. Furthermore, the estimation device 1 may be used not only to improve communication between humans, but also to improve communication between humans and robots or AI, and ultimately to evaluate and improve the performance of robots and AI that communicate with humans. For example, by measuring a person's EEG when communicating with a robot or AI, it may be possible to objectively evaluate how much attention the person pays to information provided by the robot or AI and how well the person understands that information. This will enable the creation of robots and AI that are appropriately adjusted according to objectives, such as how to attract human attention or how easily emotional information can be conveyed, which will contribute to the creation of a society in which humans, robots, and AI can coexist in harmony.

[0116] §4 Modifications Although the embodiments of the present invention have been described above in detail, the above description is merely an example of the present invention in every respect. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. For example, the following modifications are possible. Note that, in the following, the same reference numerals are used for components similar to those in the above embodiment, and descriptions of similar points to those in the above embodiment are omitted where appropriate. The following modifications can be combined as appropriate.

[0117] So far, an example has been described in which the estimation device 1 estimates the quality CQ of a conversation between two subjects TA from the "amplitude values ​​of one or more predetermined components of the electroencephalograms EE of each subject TA for the words WD included in the conversation." However, when estimating the quality CQ of the conversation, the estimation device 1 may use information other than the "amplitude values ​​of one or more predetermined components of the electroencephalograms EE of each subject TA for the words WD included in the conversation." For example, the estimation device 1 may estimate the quality CQ of the conversation by using, as described above, the "similarity of the time intervals TI of the utterances UT of each subject TA," in addition to the "amplitude values ​​of one or more predetermined components of the electroencephalograms EE of each subject TA for the words WD included in the conversation." The estimation device 1 may also use the duration of the conversation between the two subjects TA (the duration of the utterances UT of each subject TA). The estimation device 1 may use the number of words WD included in the conversation between two subjects TA (the number of words WD included in the utterance UT of each subject TA).

[0118] 1...estimation device (computer), 121...estimation program, CQ...conversation quality, EE...brain waves, OS...onset information, S110...brain wave acquisition step, S120...onset information acquisition step, S130...transition step, TA...subject, WD...word

Claims

1. An estimation program for causing a computer to execute the following steps: an electroencephalogram acquisition step for acquiring the electroencephalograms of each of two subjects who are conversing; an onset information acquisition step for acquiring onset information indicating the start time of each of a plurality of words included in the conversation; and an estimation step for estimating the quality of the conversation from the amplitude values ​​of one or more predetermined components of the electroencephalograms of each of the two subjects for the words.

2. The estimation program of claim 1, wherein the quality of the conversation includes the degree of satisfaction of each of the two subjects with the conversation, and in the estimation step, the computer estimates the degree of satisfaction of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects, and estimates the degree of satisfaction of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects.

3. The estimation program according to claim 2, wherein the one or more predetermined components include at least one of a P2 component and an N400 component.

4. The estimation program of claim 1 or 2, wherein the quality of the conversation includes the emotional valence of each of the two subjects toward the conversation, and in the estimation step, the computer estimates the emotional valence of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects, and estimates the emotional valence of one of the two subjects from the amplitude values ​​of the one or more specified components of the electroencephalograms of the other of the two subjects.

5. The estimation program according to claim 4, wherein the one or more predetermined components include an LPP component (Late Positive Potential).

6. The estimation program of claim 1 or 2, wherein the quality of the conversation includes the degree of interest of each of the two subjects in the conversation, and in the estimation step, the computer estimates the degree of interest of one of the two subjects from the amplitude value of one or more specified components of the electroencephalogram of one of the two subjects, and estimates the degree of interest of the other of the two subjects from the amplitude value of one or more specified components of the electroencephalogram of the other of the two subjects.

7. The estimation program according to claim 6, wherein the one or more predetermined components include a P2 component.

8. The estimation program of claim 1 or 2, wherein the quality of the conversation includes mutual satisfaction, which is the degree of satisfaction of both of the two subjects with the conversation, and in the estimation step, the computer estimates the degree of mutual satisfaction from the amplitude values ​​of the one or more specified components of the electroencephalograms of both of the two subjects.

9. The estimation program according to claim 8, wherein the one or more predetermined components include at least one of a P2 component and an LPP component.

Images