JP2010186237A - Thought-evoking multi-party dialogue device, thought-evoking multi-party dialogue method, thought-evoking multi-party dialogue program, and computer-readable recording medium with the program recorded thereon - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a thought-evoking multi-party dialogue device determining the content or timing of the statement of an agent according to the situation of a dialog. <P>SOLUTION: The thought-evoking multi-party dialogue device includes: a user statement comprehension part 2 for storing a dialog control rule for selecting a dialog act of the agent according to a dialog situation and a statement generation rule for selecting the statement expressions of the agent according to the dialog situations, converting the user statement into the dialog act, and writing the dialog act in an statement queue 6 and a dialog state 5; a dialog control part 3 for selecting the dialog act related to the agent by applying the dialog control rule by referring to the history of the dialog act at the beginning of the statement queue 6 and the history of dialog act stored in the dialog state 5, sending the selected dialog act to the agent statement generation part 4, and removing the dialog act at the beginning of the statement queue 6; and an agent statement generation part 4 for generating the statement expressions of the agent based on the statement generation rule by referring to the history of the dialog act when acquiring the dialog act, presenting the generated statement expressions of the agent to a user, storing the dialog act into the dialog state 5, and adding the dialog act to the statement queue 6. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a technique for a multi-person dialogue system in which a plurality of users (humans) and a dialogue agent have a conversation in a natural language using input / output means such as text or voice.

  A dialogue agent (agent) is a computer program that communicates with humans autonomously using natural language using input / output means such as text or speech. The present invention relates to control of a multi-person dialogue between a plurality of users (humans) and a plurality of dialogue agents.

  In a multi-person dialogue in which a plurality of users participate, unlike a dialogue in which only one user participates, it can be expected that communication including interaction between human users is activated. Furthermore, the activation of communication in a multi-person dialogue in which a plurality of users participate can be expected to have an effect of strengthening human connections, closeness and solidarity.

  As a means for activating communication in dialogue, firstly, there is a thought-inducing dialogue. Thinking-inducing conversations do not give information directly to the user through dialogue with the user, but encourage the user's thoughts and associations so that the user can use the thoughts and associations to reach the information. It is a motivational dialogue. The thought-inducing dialogue has the effect of increasing the user's willingness to participate in dialogue and can be expected to activate communication. As an example of a thought-inducing conversation, a quiz-style thinking-inducing conversation is known (see Non-Patent Documents 1 and 2).

  In a quiz-style thought-inducing dialogue, the system presents a person quiz or place name quiz to the user and presents hints in order. In the technique described in Non-Patent Document 1, hints are automatically rearranged and presented in order of difficulty so that the user does not easily reach the correct answer. In response, the user answers the person or place name that he thinks is correct. Through this series of dialogues, the user experiences a thought process that tries to find the correct answer from the hint, and the user's willingness to participate in the dialogue increases. As a result, it is known that the user's understanding is improved in the quiz-style thought-inducing dialogue that presents these facts as hints compared to reading the facts about a person or place name to the user. (See Non-Patent Document 2).

  As described above, it has been proved that the quiz-type thought-inducing dialogue described in Non-Patent Documents 1 and 2 has an effect of stimulating the user's thought and improving the user's willingness to participate in the dialogue. However, these conventional quiz-style thought-stimulating dialogues are limited to one-to-one dialogues between one user and one dialogue agent that questions the quiz. In such a one-to-one dialogue, only one person participates, so no human communication occurs in the dialogue. If a plurality of people can participate in the dialogue, there is a possibility that communication including exchanges between people may occur, and it is expected that the communication will be further activated. In addition, if there is another dialogue agent that participates in the dialogue in the same position as a human user other than the dialogue agent that questions the quiz, the intervention of the dialogue agent increases the willingness to participate in the dialogue, Communication can be expected to become more active.

  There has been known research aimed at activating communication between humans through the intervention of a dialogue agent (robot) in an obfuscated game in which multiple users participate (see Non-Patent Document 3). This conventional obfuscated game involving multiple users and dialogue agents is an example of a thought-promoting multi-person dialogue in the sense that users can reach the correct answer by using their own thoughts on how to read kanji that are difficult to read Can be considered. In the technique of Non-Patent Document 3, a plurality of users and the dialogue agent can speak at a free timing asynchronously. In this way, when multiple users and the conversation agent speak asynchronously at free timing, in order for the conversation agent to speak to a specific user to have the intended effect, the conversation of the conversation that changes as the conversation progresses Depending on the situation, it is necessary to determine the appropriate content and timing of the agent's remarks. However, in the technology of Non-Patent Document 3, in a dialogue with a large number of people, appropriate content and timing of the speech according to the situation of the dialogue so as to ensure that the speech to the specific user by the dialogue agent has the intended effect. Does not have a function to determine. In addition, this conventional technology does not consider the emotional remarks by the dialogue agent as described below, and does not have a function for activating communication by the emotional remarks by the dialogue agent.

  Secondly, as means for activating communication in dialogue, there is means for influencing the user's psychological state by emotional remarks by the dialogue agent and making the communication more comprehensive. Emotional utterances can be classified into empathic utterances and self-centered utterances depending on whether the emotions are toward others or to themselves. The empathic speech is an emotional speech that shows compassion for the other person's situation, and the self-centered speech is an emotional speech made for the situation. For example, in Non-Patent Document 4, in a system for playing a blackjack game with a user, a computer agent that participates in the game from the same standpoint as the user is created. When the agent feels “care is being treated” and favorable feelings, and the agent makes a self-centered statement, the “sensation of being care” does not increase, but the experimental results show that favorable feelings increase. It is shown. Non-Patent Document 5 shows a study that reports that empathetic speech from computer agents in the same position as the user improves the user's learning effect in a system that supports the user's learning of English idioms. Has been.

  These Non-Patent Documents 4 and 5 indicate that the agent's emotional remarks may improve the impression of the agent felt by the user or increase the user's willingness. 5 can be expected to lead to activation of communication as a result. However, in Non-Patent Documents 4 and 5, in the first place, it is not directly assumed to be applied to a multi-person dialogue in which a plurality of users participate, and the timing at which a user or agent speaks is fixed. is there. Therefore, even if the systems disclosed in Non-Patent Documents 4 and 5 are applied to a multi-person dialogue in which a plurality of users participate, the changes are made every time a plurality of users speak asynchronously and freely in the dialogue. There is no provision for the agent to select the appropriate content and timing of the speech according to the situation of the conversation. For example, an agent who participates in the blackjack game disclosed in Non-Patent Document 4 speaks only when one hand of blackjack is determined to win or lose, and the user only participates in the blackjack game. Never speak. Further, in the system that supports the learning of the user's English idiom disclosed in Non-Patent Document 5, the user can speak only at a fixed timing of answering the question from the teacher as an agent.

Ryuichiro Higashinaka, Kohji Dohsaka, and Hideki Isozaki, “Learning to Rank Definitions to Generate Quizzes for Interactive Information Presentation”, Proceedings of the ACL-2007 (Poster Presentation), p.117-120, 2007 Ryuichiro Higashinaka, Kohji Dohsaka, Shigeaki Amano and Hideki Isozaki, “Effects of Quiz-style Information Presentation on User Understanding”, in Proceedings of Interspeech, p.2725-2728, 2007 Yoichi Matsuyama, Teru Taniyama, Shinya Fujie, Tetsunori Kobayashi: Development of robots for activating human-human communication, SIG-SLUD-A801-03, p.15-22, 2008 Scott Brave, Clifford Nass and Kevin Hutchinson, “Computers that care: investigating the effects of orientation of emotion represented by an embodied computer agent”, Internationa1 Journal of Human-Computer Studies, Volume 62, Issue 2, p.161-178, 2005 Heidy Maldonado, Jong-Eun Roselyn Lee, Scott Brave, Cliff Nass, Hiroshi Nakajima, Ryota Yamada, Kimihiko Iwamura and Yasunori Morishima, `` We learn better together: enhancing eLearning with emotional characters '', in Proceedings of 2005 conference on Computer support for collaborative learning, p.408-417, 2005

  First, in a multi-person thought-stimulated conversation in which multiple users and multiple conversation agents talk in natural language using input means such as text or voice, the users and the conversation agents can speak asynchronously and freely. In the conventional method, the satisfaction level of the dialogue that the user feels, the familiarity with the agent, the favorable feeling, the sense of care by the agent, etc. There is a problem that it cannot be improved and the amount of communication cannot be increased by empathic speech or self-centered speech of the dialogue agent.

  Secondly, in a multi-person thought-stimulated dialogue in which multiple users and multiple conversation agents talk in natural language using input means such as text or speech, the conventional method has the intended effect of the conversation agent speaking. As there is no function to determine the appropriate content and timing of the dialogue agent's utterance according to the situation of the dialogue so that it can be guaranteed, multiple users and dialogue agents can speak asynchronously and freely. If there is a problem, the content of the agent's speech may be misunderstood and transmitted, or the intended effect may not be provided to the user.

The first problem will be described.
As described in the background art, in the technique of Non-Patent Document 3, which is a conventional method related to a large number of thought-stimulating dialogues, by performing emotional utterances such as empathic utterances and self-centered utterances by dialogue agents. Does not have the ability to increase the amount of communication. In the technology of Non-Patent Document 3, even if sympathetic speech or self-centered speech is added to the repertoire of speech made by the dialogue agent, the technology of Non-Patent Literature 3 has the effect that the speech of the dialogue agent intended. The second problem described below cannot be solved because it does not have the function of determining the appropriate content and timing of the agent's remarks according to the situation of the dialogue so as to ensure that it is brought about. Therefore, in the technology of Non-Patent Document 3, the intended effect cannot be sufficiently brought to the user by the sympathetic utterance or the self-centered utterance of the dialogue agent. It cannot improve the feelings, likability, and the feeling of being cared for by the agent, and the amount of communication cannot be increased by the sympathetic or self-centered statements of the dialogue agent.

The second problem will be described.
Here, it demonstrates concretely through the conversation examples 1-5. These are examples of quiz-style multi-person thought-inducing dialogues. Hanako and Ryoko participate as users, and as a dialogue agent, the question agent "Foo-Foo" presents questions related to person quiz and presents hints, and the answer agent participates in the dialogue from the same position as the user Two mushroom agents are participating. In the dialogue examples 1 and 2, the reply agent “mushroom” does not appear. First, consider the following dialogue example 1.

<Dialogue Example 1>
(1.1) Fu-Foo: The question is a person quiz.
(1.2) Fu-Foo: A person who died due to a change in Honnoji.
(1.3) Hanako: Oda (1.3) Ryoko: Toyotomi (1.5) Hoohu: I'm sorry.
(1.6) Hanako: Toyotomi Hideyoshi (1.7) Fu-Foo: Not at all.

  The correct answer to this person quiz question is Nobunaga Oda. In (1.2), the questioning agent “Foo-Foo” presents hints. On the other hand, Hanako and Ryoko answered in (1.3) and (1.4), respectively. Since the user can reply at any time, it is not known in advance in what order and when the user answers. In this case, Hanako and Ryoko happened to make a series of responses before “Foo-Foo” responded to the response in (1.5). Actually, the statement (1.5) of “Foo-Foo” was in response to Hanako ’s statement of (1.3), but it was just “Sad”. It is not possible to determine whether it responded to Hanako's utterance (1.3) or Yoshiko (1.4). For this reason, Hanako misunderstood that “Foo-Foo” remarked that it was regrettable to Ryoko's statement “Toyotomi” in (1.4). He made his speech in the wrong direction. In order to make “Foo-Foo” not to be misunderstood in (1.5), instead of simply saying “I ’m regrettable”, “Is it Oda? It is necessary to adjust the content of the remarks. In other words, it can be said that the questioning agent “Foo-Foo” had to adjust the content of the statement as in (2.5) of the following dialogue example 2.

<Dialogue example 2>
(2.1) Fu-Foo: A question quiz for people.
(2.2) Fu-Foo: A person who died due to a change in Honnoji.
(2.3) Hanako: Oda (2.4) Ryoko: Toyotomi (2.5) Fu-Foo: Is it Oda? I'm sorry. / Hanako-san is regrettable.
(2.6) Hanako: Oda Nobunaga Kana (2.7) Foo: Per. The correct answer is Oda Nobunaga.

  If the questioning agent "Foo-Foo" always makes redundant statements such as "Is it Oda? I'm regrettable" or "Hanako-san, I'm regrettable", there will be no misunderstandings, but the dialogue itself is not good. It becomes necessary and redundant, and there is a possibility of impairing satisfaction and communication activity felt by the user. As described above, there is a problem that the dialogue agent must determine an appropriate content of the speech according to the situation of the dialogue. The technology of Non-Patent Document 3, which is an example of a multi-person thought-stimulating dialogue in the conventional technology, cannot cope with this problem.

  Next, a case where a dialogue agent makes a sympathetic or self-centered speech in a dialogue with many people is considered as an example.

  In dialogue example 3 shown below, two users Hanako and Ryoko said “I do n’t know” and “I do n’t know” in (3.2) and (3.3), respectively, and I feel that the problem is difficult. In response to this situation, the reply agent “Mushroom” makes a remark (3.4) “It's difficult”. Non-patent document 4 describes the experimental result that the agent's sympathetic remarks increase the sense of care of the agent that the user feels.

<Dialogue example 3>
(3.1) Fu-Foo: Born in 1534.
(3.2) Hanako: I can't think of it.
(3.3) Ryoko: I don't know.
(3.4) Mushrooms: That's difficult.

  In the following dialogue example 4, although the users Hanako and Ryoko feel that the problem is simple, the answer agent “Mushrooms” feels a difficult problem for them. Self-centered remarks (4.4) “It's difficult”. Here, self-centered speech refers to emotional speech directed to one's own situation, not to the other's situation. Non-patent document 4 describes the experimental result that the agent's self-centered remark increases the user's perception of the agent.

<Dialog example 4>
(4.1) Fu-fu: I built Azuchi Castle.
(4.2) Hanako: It's easy.
(4.3) Ryoko: It ’s Oda somehow.
(4.4) Mushrooms: It ’s difficult.

  Such sympathetic or self-centered statements by dialogue agents increase the user's feeling of “care being felt” and the likability of the agents, resulting in the activation of communication in multi-party dialogues. Expected to be useful. However, as already mentioned, the technology described in the conventional non-patent document 3 does not make sympathetic speech or self-centered speech by the dialogue agent, so that communication is activated by emotional speech of the dialogue agent. I can't.

  In the technique described in Non-Patent Document 3, a method of adding sympathetic speech or self-centered speech to the repertoire of speech of the dialog agent can be considered, but if the empathetic speech or self-centered speech by the dialog agent is unconditionally performed, It's not good. In the following dialogue example 5, the answer agent “mushroom” feels that the problem is difficult as in the dialogue example 3, and the self-centered remarks about his / her situation (5.4) “It is difficult”. It is carried out. However, since the users Hanako and Ryoko have said that the problem is difficult in (5.2) and (5.3), “Mushroom” (5.4) “It ’s difficult” is self-centered. There is a problem that Hanako and Ryoko may be regarded as sympathetic remarks for feeling that the problem is difficult.

<Dialogue example 5>
(5.1) Fu-Foo: Born in 1534.
(5.2) Hanako: I can't think of it.
(5.3) Ryoko: I don't know.
(5.4) Mushrooms: It ’s difficult.

  Based on the knowledge (see Non-Patent Document 4) that the user feels the best likability when the agent makes both sympathetic and self-centered utterances, the system designer ( In 5.4), “Mushroom” (5.4) is recognized as a sympathetic utterance even if it is intended to make “mushroom” a self-centered utterance rather than a sympathetic utterance. Therefore, there arises a problem that the effect of the self-centered remark of “mushroom” may not be brought about. If you want the answering agent to make a self-centered statement, you must not let the answering agent say “It's difficult” at the timing of (5.4). In this way, it is necessary to determine the appropriate timing of the dialogue agent's speech according to the situation of the dialogue so as to guarantee that the dialogue agent's sympathetic speech and self-centered speech have the intended effect. There is.

  Therefore, an object of the present invention is to solve the above-described problems and provide a technique capable of determining the content and timing of an agent's utterance in accordance with the state of dialogue in a multi-person thought-inducing dialogue.

  The present invention was devised to achieve the above object, and the multi-person thought-inducing interactive apparatus according to the present invention is a multi-person system comprising a plurality of human users and a plurality of agents implemented by computers. In a dialogue, a multi-person thought-inducing interactive device that arouses the user's thoughts by the agent's remarks, and a time parameter that defines a timing at which the agent speaks as a situation of the multi-person dialogue, and each agent A situation database storage means for storing emotion parameters for setting emotional conditions that prescribe utterances as sympathetic or self-centered utterances, and from each agent for each type of utterance content of participants in the multi-person dialogue The first dialogue action information describing the internal structure of the speech presented to the user is converted into the multi-person dialogue situation. A dialogue control rule storage means storing a dialogue control rule describing a rule to be selected, a representation of a statement presented from the agent to the user, a type of the content of each agent, the emotion condition, and the multiple Utterance generation rule storage means storing a utterance generation rule describing a rule to be selected according to the situation of the number of people dialog, the first dialog action information, and the second dialog action information describing the internal structure of the user's utterance And storing the first dialogue action information relating to the speech already presented to the user and the history of the second dialogue action information as a history of dialogue by the participants of the multi-person dialogue. And receiving text data or voice data indicating the remarks of each user, and converting them into the second interactive action information of the user. User speech understanding means that is added to the tail of the queue storage means and written to the dialogue state storage means, the first or second dialogue action information at the head of the queue storage means, and the dialogue state storage means By referring to the history of the stored first and second dialogue action information, and applying the dialogue control rule, the first dialogue action information related to the speech next presented to the user from the agent is selected, and The dialog control means for removing the first or second dialog action information at the head of the queue storage means and the first dialog action information selected by the dialog control means are acquired and stored in the dialog state storage means. Referring to the history of the first and second dialogue action information, and making a statement next to the acquired first dialogue action information based on the statement generation rule The agent's speech expression is generated, the generated speech expression is presented to the user, the first dialogue action information into which the speech expression is substituted is stored in the dialog state storage means, and at the end of the queue storage means And agent remark generation means to be added.

  In order to solve the above-mentioned problem, a multi-person thought-inducing dialogue method according to the present invention includes a user speech understanding unit, a dialogue control unit, an agent speech generation unit, and a storage unit, and includes a plurality of humans. In a multi-person dialogue between a user and a plurality of agents realized by a computer, a multi-person thought-stimulating dialogue method by a multi-person thought-stimulating dialogue device that arouses the user's thoughts by the remarks of the agent, the memory The means includes a time parameter that defines a timing at which the agent speaks, and an emotion parameter that sets an emotion condition that defines the speech of each agent as a sympathetic speech or a self-centered speech as the situation of the multi-person dialogue. A status database for storing information, and each age for each type of remarks of participants of the multi-person dialogue A dialogue control rule describing a rule for selecting the first dialogue action information describing the internal structure of the speech presented to the user from the agent according to the situation of the multi-person dialogue, and the agent presenting the user to the user. A speech generation rule that describes a rule for selecting an expression of a speech according to the type of speech content of each agent, the emotion condition, and the situation of the multi-person dialogue, and is stored in advance. A step of accepting text data or voice data indicating the speech of each user, a step of converting the accepted speech into second interactive action information describing an internal structure of the user's speech, in speech understanding means; (2) adding dialogue action information to the tail of the queue storage means, and as a history of dialogues by participants of the multi-person dialogue, Writing the second dialogue action information of the accepted utterance into the dialogue state storage means for storing the first dialogue action information regarding the presented utterance and the history of the second dialogue action information; The dialog control means refers to the first or second dialog action information at the head of the queue storage means and the history of the first and second dialog action information stored in the dialog state storage means, and the dialog Selecting first dialogue action information relating to a statement to be presented next to the user from the agent by applying a control rule; and sending the selected first dialogue action information to the agent statement generation means; Removing the first or second dialogue action information at the head of the queue storage means; and the agent message generation means, Obtaining the first dialogue action information selected by the speech control means, referring to the history of the first and second dialogue action information stored in the dialogue state storage means, and based on the statement generation rule, A step of generating a speech expression of the agent to speak next in response to the acquired first dialogue action information; a step of presenting the generated speech expression to the user; and a first dialog in which the speech expression is substituted And storing the action information in the dialogue state storage means and adding the action information to the tail of the queue storage means.

  According to the multi-person thought-inducing interactive device having such a configuration, or the multi-person thought-inducing interactive device having such a procedure, the multi-person thought inviting interactive device can enable the agent to output a sympathetic speech or a self-centered speech. The dialogue control rules that describe the rules for selecting which first dialogue action information to select according to the situation of the multi-person dialogue, the types and emotional conditions of each agent's remarks, and the situation of the multi-party dialogue Utterance generation rules describing rules for selecting which utterance expression to select for each agent according to the above. In the dialogue control rule, a rule for determining the timing of the speech is described in advance so as to ensure that the self-centered speech and the emotionless speech of the agent do not have the effect of empathizing the user. Further, in the message generation rule, a rule for determining the content of the message is described in advance so as to ensure that the content of the agent's message is transmitted without misunderstanding according to the situation of the conversation.

  Then, the multi-person thought arousing dialogue apparatus selects the first dialogue action information of the next utterance of the agent according to the dialogue control rule by the dialogue control means for the utterance of the agent, and by the agent utterance generation means, A speech expression corresponding to the first dialogue action information is generated based on the speech generation rule and presented to the user. Further, the multi-person thought-inducing dialogue device converts the user's speech into second dialogue action information similar to the agent's speech by the user speech understanding means for the user's speech. This multi-person thought-inducing dialogue device uses the dialogue control rules and the speech generation rules to determine the content and timing of the agent's speech according to the situation of the conversation, so that the user and the agent can asynchronously and freely. Even if it is allowed to speak, it is possible to output an agent's sympathetic utterance and self-intermediate utterance with appropriate content and timing according to the situation of the conversation.

  The multi-person thought-inducing dialogue device changes every moment when a user and an agent speak asynchronously and freely in a multi-person thought-stimulating dialogue between a plurality of people (users) and a plurality of agents. The state of the dialogue is expressed by the queue storage means and the dialogue state storage means. In other words, the multi-person thought-inducing dialogue device stores and manages the first and second dialogue action information in the queue storage means and the dialogue state storage means as the dialogue status related to the contents and timing of the statements of the agent and the user. Yes. Therefore, the content and timing of the agent's utterance can be appropriately determined according to the situation of the dialogue. As a result, it is possible to improve the degree of satisfaction of the dialogue felt by the user, the familiarity and likability of the agent, and the sense of care for the agent, and as a result, the amount of communication including human interaction can be increased. Furthermore, even if humans and dialogue agents are allowed to speak asynchronously and freely, the content of the dialogue agent's speech so that the dialogue agent's speech is conveyed without misunderstanding and the speech has the intended effect. And the timing can be determined.

  Further, the multi-person thinking and arousing dialogue apparatus according to the present invention is configured such that the plurality of agents are based on the selected first dialogue action information in order to conduct a quiz question-and-answer dialogue as the multi-person dialogue. The first dialogue action information, the dialogue control rule, and the statement generation rule are determined in advance so as to act as a question agent that questions the user and an answer agent that answers the question. Problem storage means for storing the question of the quiz, hint generation means for generating a hint sentence list in which a plurality of hints of the problem are arranged in descending order of difficulty, and a hint for storing the generated hint sentence list Storage means, correct / incorrect determination means for determining whether the user or the answer agent answers correctly, and when the answer is incorrect Similarity determination means for determining the degree of similarity between the answer and the answer, and when the agent speech generation means determines that the participant's answer is incorrect by the correctness determination means and the correct answer When it is determined that there is a message, a statement in accordance with the emotion condition with respect to the determination result is generated as a statement expression of the questioning agent or the answering agent.

  Further, in the multi-person thought-inducing dialogue method according to the present invention, in order to perform a quiz question-and-answer dialogue as the multi-person dialogue, the plurality of agents are based on the selected first dialogue action information. The first dialogue action information, the dialogue control rule, and the statement generation rule are determined in advance so as to act as a question agent that questions the user and an answer agent that answers the question. The questioning agent makes a statement in a series of processes for asking the question of the quiz so that a plurality of hints of the question are presented to the participant in descending order of difficulty. The correctness of the answer of the answer agent is determined. When the answer is incorrect, the degree of similarity between the correct answer and the answer is determined. If the answering agent or the answering agent gives an incorrect answer or a correct answer, the answering agent or the answering agent follows the emotional condition for the result. It is characterized by making remarks.

  According to the multi-person thought-inducing interactive device having such a configuration, or the multi-person thought-inducing interactive method having such a procedure, the multi-person thought inducing interactive device includes a plurality of human users and a plurality of computers. A quiz question-and-answer dialogue can be performed with an agent. Thereby, it is possible to activate communication between users, which was impossible in the conventional question-and-answer dialogue between one user and one agent.

  Further, the multi-person thought-stimulating dialogue program according to the present invention is a program for causing a computer to function as each means constituting one of the multi-person thought-stimulating interactive devices. By being configured in this way, a computer in which this program is installed can realize each function based on this program.

  Also, a computer-readable recording medium according to the present invention is characterized in that the multi-person thought-inducing dialogue program is recorded. By being configured in this way, a computer equipped with this recording medium can realize each function based on a program recorded on this recording medium.

  According to the present invention, in a multi-person thought-inducing dialogue in which a plurality of people and a plurality of dialogue agents participate, the dialogue status relating to the contents and timing of the utterances of the agent and the user is managed by the storage means, and predetermined. Based on the dialogue control rule and the speech generation rule, the agent can output a sympathetic speech or a self-centered speech with appropriate content and appropriate timing. Therefore, the degree of satisfaction that humans feel, the sense of familiarity and likability with agents, and the sense that they are being cared for can be improved, and the amount of communication, including interaction between humans, can be increased and communication can be activated. It becomes. As a result, the effect of strengthening the connection, closeness, and solidarity between humans participating in the multi-person dialogue can be expected.

It is a conceptual diagram which shows the structure of the multi-person thought arousing dialogue system which concerns on embodiment of this invention. It is a flowchart which shows operation | movement of the user speech understanding part shown in FIG. It is a flowchart which shows operation | movement of the dialogue control part shown in FIG. It is a flowchart which shows operation | movement (hint sentence) of the agent statement production | generation part shown in FIG. It is a flowchart which shows operation | movement (other than a hint sentence) of the agent statement production | generation part shown in FIG. It is a block diagram which shows the structural example of the dialogue control part shown in FIG. It is a block diagram which shows the structural example of the hint production | generation part shown in FIG. It is a flowchart which shows operation | movement of the hint production | generation part shown in FIG. It is a block diagram which shows the structural example of the question control means shown in FIG. It is a flowchart which shows operation | movement of the question control means shown in FIG.

  An embodiment (hereinafter referred to as “embodiment”) for carrying out the multi-person thought-stimulating dialogue apparatus and multi-person thought-stimulating dialogue method of the present invention will be described in detail with reference to the drawings. The following is an outline of a multi-person thought-aware dialogue system, details of the system configuration, details of the dialogue control unit, concrete examples of dialogue control rules, details of the agent speech generation unit, concrete examples of speech generation rules, concrete examples of question control means Examples will be described sequentially.

[Outline of multi-person thinking-aware dialogue system]
FIG. 1 is a conceptual diagram showing a configuration of a multi-person thought-stimulating dialogue system according to an embodiment of the present invention. The multi-person thought-stimulating dialogue system 1 (hereinafter also simply referred to as a system) arouses the user's thoughts by the agent's remarks in a multi-person dialogue between a plurality of human users and a plurality of agents realized by a computer. System. In the present embodiment, the system will be described as a system that performs a quiz-type multi-person thought-inducing dialogue. In this system, two users (humans) and two dialogue agents participate in the dialogue. Hanako and Ryoko participate in the dialogue as two users. As two interactive agents, a question-taking agent “Foo-Foo” that presents a question about a person quiz and presents hints in order, and an answer agent “Mushroom” who answers the quiz in the same position as a human participate in the dialogue. . The user and the dialogue agent are collectively called a dialogue participant (participant). The user inputs an utterance using the keyboard, and the utterances of the user and the dialogue agent are output as character strings on the display 10. Note that the user's terminal device and the multi-person thought-stimulating dialogue system 1 are communicably connected via a network such as a local area network (LAN) or the Internet.

  The dialogue proceeds as follows. First, the questioning agent “Foo-Foo” selects the name of the person who has a problem with the name guessing quiz and creates a quiz hint sentence. The hint sentence is generated in accordance with Non-Patent Document 1 by automatically rearranging difficult hints to easy hints so that the user does not easily reach the correct answer. The dialogue starts when the questioning agent starts a quiz problem with a statement such as "I will start a quiz for people." Subsequently, the questioning agent presents hints in order.

  Two users and two interactive agents can speak at any time. When the user or answering agent utters an answer to a quiz hint sentence, if the answer is incorrect, the questioning agent evaluates the answer (eg, “near” or “far”). Do. If the answer of the user or the answer agent is correct, the questioning agent informs that the answer is correct and ends the dialogue about the current problem. The questioning agent presents one hint sentence and presents the next hint when the user and the answering agent do not give a correct answer within a predetermined time. If you lose a hint that you have created in advance, tell it and end the conversation about the current problem. When the dialogue on the current problem ends, move to the dialogue on the next problem.

  In addition to the above interaction, the user and the dialogue agent expressed that they felt that the problem was difficult or easy, and responded to that statement. Or, give an impression of having ended up with an incorrect answer. In addition, the user and the dialogue agent can make sympathetic speech and self-centered speech.

(System configuration)
A multi-person thought-inducing dialogue system (multi-person thought-inspiring dialogue apparatus) 1 shown in FIG. 1 includes an arithmetic device such as a CPU, a storage device (storage means) such as a memory and a hard disk, and an external device such as a mouse and a keyboard. An input device that detects input of information, an interface device that transmits and receives various types of information to and from the outside, a computer that includes a display device such as an LCD (Liquid Crystal Display), and a program that is installed in the computer The

  The multi-person thought arousing interactive device 1 is realized by the hardware device and software cooperating to control the aforementioned hardware resources by a program, and as shown in FIG. 2, a dialogue control unit 3, an agent statement generation unit 4, a dialogue state (conversation state storage unit) 5, and a statement queue (queue storage unit) 6.

In order to control the emotion expression pattern of the system, a parameter called a system emotion parameter is given from the outside.
The system emotion parameter is an emotion parameter for setting an emotion condition that prescribes each agent's utterance as sympathetic utterance or self-centered utterance, and includes α and β described later.
Various time parameters are given from the outside in order to control the behavior of the system with respect to time. The time parameter defines the timing at which the agent speaks, and includes HT, ST, and WT described later.

  The contents of the speech of the user and the dialogue agent are expressed as dialogue actions (dialogue action information) and accompanying effects accompanying the dialogue actions in the system. The dialogue action (dialogue action information) is first dialogue action information (T201 to T216: refer to Table 4) that describes the internal structure of the utterance presented to the user from each agent for each kind of utterance contents of the participants of the multi-person dialogue. ) And second dialogue action information (T206 to T217: see Table 4) describing the internal structure of the user's remarks. Details of dialogue actions and incidental effects will be described later.

  In this embodiment, the user's speech is input by a keyboard. However, when a method for recognizing the user's speech is provided, the user's speech can be input by speech. In this embodiment, the speech of the user and the dialogue agent is output to the display 10 of each user. However, if the dialogue agent has a voice output function, the speech of the agent is the voice output of each agent. Using function, it is output by voice.

  2 is a flowchart showing the operation of the user speech understanding unit, FIG. 3 is a flowchart showing the operation of the dialogue control unit, and FIGS. 4 and 5 are flowcharts showing the operation of the agent speech generation unit. The following explanations and definitions of terms in the flowchart will be described with reference to FIG. 1 as appropriate, and then the operation of each part will be described.

[System configuration details]
(System emotion parameters)
System emotion parameters are given externally to control the emotional expression of the system.
The system emotion parameters include a sympathy parameter α and a self-centered parameter β, each having a value of “1” or “0”. The system emotion parameter is represented by a set of empathy parameter α and self-centered parameter β as follows.
<Empathetic parameters, self-centered parameters> = <α, β>

When the sympathy parameter α is “1”, the system can make a sympathetic speech, and when the sympathy parameter α is “0”, the system does not make a sympathetic speech.
When the self-center parameter β is “1”, the system can make a self-centered speech, and when the self-center parameter β is “0”, the system does not make a self-centered speech. Here, empathy is an emotion that the speaker considers for the other person's situation, and self-centered is an emotion that the speaker gives to the own situation.

As shown below, according to the value of the system emotion parameter, the three emotion conditions of the system called the empathy condition, self-centered condition, and no emotion condition are defined as follows.
(1) When the sympathy parameter α is “1”, it is called that the sympathy condition is satisfied, and when the sympathy parameter α is “0”, it is called that the sympathy condition is not satisfied.
(2) When the self-center parameter β is “1”, it is called that the self-center condition is satisfied, and when the self-center parameter β is “0”, it is called that the self-center condition is not satisfied.
(3) When both the empathy parameter α and the self-centered parameter β are “0”, it is said that the emotionless condition is satisfied. Otherwise, it is said that no emotional condition is established.
(4) When both the empathy parameter α and the self-center parameter β are “1”, both the empathy condition and the self-center condition are satisfied.

  The relationship between system emotion parameters (<sympathetic parameters, self-centered parameters>) and emotion conditions is summarized as follows.

(Time parameter)
In order to control the behavior of the system with respect to time, the following parameters are given externally.
(1) Hint interval time HT [seconds]
(2) Maximum silent time ST [seconds]
(3) Speech start waiting time WT [seconds]
The hint interval time HT is the time from when the questioning agent presents one hint to when the next hint is presented. After any dialog participant speaks, if the time during which no dialog participant speaks exceeds the maximum silent time ST, any dialog agent must make the next speech. In addition, after any dialogue participant speaks, in order for the dialogue agent to make the next speech, at least the speech start waiting time WT must be kept. The maximum silence time ST needs to be larger than the speech start waiting time WT. In this embodiment, as an example, the hint interval time is set to 30 [seconds], the maximum silence time is set to 4 [seconds], and the speech start waiting time is set to 0.5 [seconds].

(Speech type)
The statements made by the user and the dialogue agent are as follows. The utterances are classified according to utterance types (types of utterance contents of each agent). In this embodiment, as shown in Table 2 and Table 3, it was classified into 17 types such as T101 to T117. Descriptions of each message type and examples of message expressions are described in Tables 2 and 3, and thus the description thereof is omitted.

(Dialogue)
The content of the speech of the user and the dialogue agent is expressed as a dialogue act within the system. The dialogue act has the form of the following formula (1).
<Speech ID, Speak Time, Speaker, Speak Type, Speak Expression> ... (1)

Here, the speech time is the time when the speech is executed.
The speaker is an entity that executes a statement. In this embodiment, the speaker is any one of Hanako, Ryoko, the questioning agent “Foo-Foo”, and the answer agent “Mushroom”.
The utterance type represents the kind of utterance. The speech expression is a natural language character string.

  Dialogue actions may have accompanying effects. The incidental effect represents a proposition that supplements the content of the dialogue act or a proposition established after the dialogue act is executed. For example, the question starting type dialogue act in which the questioning agent “Foo-Foo” starts a person quiz for the person name PERSON has the following equations (2a) and (2b) with two accompanying effects. ), And is expressed as in equation (2c).

Dialogue action: <ID, TIME, foo, problem start, EXPR> ... Formula (2a)
Accompanying effects: <Problem object, ID, PERSON> Formula (2b)
<Current problem, ID> ... Formula (2c)

  In the expression (2a) which is a dialogue action, ID indicates “speech ID”, TIME indicates “speech time”, and EXPR indicates “speech expression”. Further, in the expression (2b) which is an accompanying effect, PERSON indicates “person name to be a problem”.

<Problem object, ID, PERSON> in the expression (2b), which is an accompanying effect, indicates that a quiz to hit the person name PERSON is started by a problem start type utterance having a “utterance ID”.
The <current problem, ID> in the expression (2c), which is an accompanying effect, indicates that the current quiz dialog has been started by a problem start type dialog act having a “speech ID”.

  The expression of the dialogue action and the accompanying effect in the present embodiment is as follows. Symbols starting with ID are “speech ID”, TIME is “speech time”, USER is “user”, AGENT is “dialogue agent”, symbols beginning with P (= PARTICIPANT) are “dialog participants”, EXPR Symbols that begin with “representation” respectively.

(Remark queue)
The speech queue 6 is a first-in-first-out queue composed of dialogue actions representing the content of speech by the user or the dialogue agent. When a new dialogue action is added to the message queue 6, it is added to the tail of the message queue 6. When the dialogue action is taken out from the speech queue 6, the first dialogue action is taken out. The message queue 6 indicates queue storage means for storing first dialog action information and second dialog action information (T206 to T217: see Tables 4 and 5) describing the internal structure of the user's message. .

(Interactive state)
In the dialogue state 5, a history of dialogue actions that have been executed by the agent statement generation unit 4 and their associated effects are written. In the dialogue state 5, a history of user utterances (the user's dialogue act and its accompanying effects) converted by the user utterance understanding unit 2 is also written. This dialogue state 5 shows dialogue state storage means for storing the history of the first and second dialogue action information relating to the speech already presented to the user as the dialogue history by the participants of the multi-person dialogue.

(User remark understanding department)
The user utterance understanding unit 2 receives each user's utterance input by a keyboard or a voice recognition device, and converts it into an interactive action. The user speech understanding unit 2 operates according to the procedure shown in FIG. The user speech understanding unit 2 acquires each user's speech (step S21), and converts the user's speech into a dialogue act (step S22). Then, the user's interactive action is added to the tail of the speech queue 6 (step S23), and the user's interactive action is written in the dialog state 5 (step S24). Then, the user speech understanding unit 2 repeats this series of processes (steps S21 to S24).

(Dialogue control section)
The dialogue control unit 3 has a function of selecting a dialogue action to be executed by the dialogue agent and sending the dialogue action and accompanying effects to the agent speech generation unit 4.

  When selecting a question start type interactive action by the questioning agent, hint sentences arranged in order of difficulty are automatically generated according to the method of Non-Patent Document 1, and the hint sentences are sent to the agent statement generation unit 4.

  The dialogue control unit 3 operates according to the procedure shown in FIG. The dialogue control unit 3 refers to the dialogue act at the head of the message queue 6 and the dialogue state 5 to determine a dialogue control rule (step S31), and the dialogue agent's dialogue act for executing the decided dialogue control rule. A candidate is selected (step S32). If the dialog control unit 3 refers to the dialog state 5, it can acquire information on who has spoken after someone has spoken. Each dialogue action candidate may have an accompanying effect. When there are a plurality of dialogue action candidates (step S33: Yes), the dialogue control unit 3 selects one of the dialogue actions according to the dialogue action selection rule described later (step S34), and the dialogue action of the selected agent and the accompanying action. The effect is input to the agent message generation unit 4 (step S35). On the other hand, when there is only one dialogue action candidate (step S33: No), the dialogue control unit 3 skips step S34 and proceeds to step S35. Subsequent to step S35, the dialogue control unit 3 removes the dialogue action at the head of the statement queue 6 from the statement queue 6 (step S36). The dialog control unit 3 returns to step S31 if the message queue 6 is not empty (step S37: No), and ends the process if the message queue 6 is empty (step S37: Yes).

(Dialogue action selection rules)
When there are a plurality of dialogue action candidates, the speech type dialogue action included in the next (A) is selected in preference to the speech type dialogue action contained in (B). Still, when a plurality of dialogue action candidates remain, one dialogue action and its accompanying effects are selected at random.
(A) Problem start, hint presentation, answer evaluation, correct answer end, incorrect answer end, answer evaluation response, correct answer impression-yourself, correct answer impression-others, incorrect answer end impression, difficulty expression response, easy Degree expression response (B) Difficulty expression, difficulty expression, prompt, answer, Aichi

(Agent remark generation part)
The agent speech generation unit 4 has two functions. One is a function for receiving and holding a hint sentence from the dialog control unit 3 (when the hint sentence is acquired), and the other is a dialog act and accompanying effect from the dialog control unit 3. This is a function (except when acquiring a hint sentence) that generates a message expression of an agent corresponding to the message and outputs it to the display 10. The agent statement generation unit 4 operates according to the procedure shown in FIGS.

  First, when the hint sentence is acquired, as shown in FIG. 4, the agent statement generation unit 4 receives the hint sentence from the dialogue control unit 3 (step S41), and stores the acquired hint sentence (step S41). Step S42).

  When the hint sentence is not acquired, as shown in FIG. 5, the agent speech generation unit 4 receives the agent's dialogue action and accompanying effects from the dialogue control unit 3 (step S51), and receives the dialogue state. 5, the speech expression (character string) corresponding to the dialogue action is generated based on the speech generation rule (step S52). Here, the statement generation rule is a rule for the dialogue control unit 3 to select an appropriate statement expression (character string) according to the dialogue state 5 with reference to the dialogue state 5. Details of the statement generation rule will be described later.

  Subsequently, the agent utterance generation unit 4 substitutes the generated utterance expression (character string) as the value of the utterance expression of the dialogue action, and substitutes the current time as the value of the utterance time of the dialogue action (step S53). Then, the agent message generation unit 4 outputs the generated message expression (character string) to the display 10 (step S54). Then, the agent statement generation unit 4 writes the agent's dialogue action and accompanying effects in the dialogue state 5 and updates the history of the dialogue state 5 (step S55). However, for the accompanying effect of the form <current problem, ID, Person>, if there is an accompanying effect of the same form in the dialog state 5, it is deleted and the accompanying effect sent from the dialog control unit 3 is written. Come on. As a result, the dialog state 5 holds only one quiz problem handled in the current dialog. Then, the agent message generation unit 4 adds the agent's dialogue action to the tail of the message queue 6. The agent message generation unit 4 repeats this series of processing (steps S51 to S56).

(Operation of the multi-person thinking-aware dialogue system 1 as a whole)
Here, referring back to FIG. 1, the overall operation of the multi-person thought-stimulating dialogue system 1 centering on the “dialogue state 5 and the speech queue 6” as the dialogue status will be outlined. First, regarding the user's remarks, each user (user A, user B) inputs a remark at a free timing (S1). The speech expression (character string) of this speech is displayed on the display 10 of each user (user A, user B) (S2). Further, the speech expression (character string) of the speech is converted into the user's dialogue action (second dialogue action information) by the user speech understanding unit 2, and the user's dialogue action is added to the speech queue 6 (S3). State 5 is written (S4).

  On the other hand, in the system, the dialogue control unit 3 refers to the message queue 6 (S5a), and refers to the dialogue state 5 (S5b), and selects the dialogue action (presentation of hint text) of the questioning agent according to the dialogue control rule. If so, the generated hint sentence (hint sentence list) is transmitted to the agent message generation unit 4 (S6a). Alternatively, when the dialog control unit 3 selects another dialog act (+ accompaniment effect) of the questioning agent or the answer agent according to the dialog control rule, the dialog controller 3 utters the selected other dialog act (+ accompaniment effect). It transmits to the production | generation part 4 (S6b). Then, the dialogue control unit 3 takes out the first dialogue action from the statement queue 6 (S7).

  In the system, when the agent message generation unit 4 acquires another dialogue action (+ accompaniment effect) while saving the hint sentence acquired from the dialogue control unit 3, the agent statement generation unit 4 refers to the dialogue state 5 (S8). ) In accordance with the statement generation rules, an agent's statement expression (character string) is generated and substituted into the acquired dialogue action (+ accompaniment effect), and each user (user A) uses the statement expression (character string) as the agent's statement. And display on the display 10 of the user B) (S9). Then, the agent speech generation unit 4 writes this dialogue action in the dialogue state 5 and updates the dialogue state 5 (S10). Further, it is also added to the message queue 6 (S11).

[Details of Dialogue Control Unit]
Details of the dialogue control unit will be described below. FIG. 6 is a block diagram illustrating a configuration example of the dialogue control unit illustrated in FIG. 1. Here, the dialogue control unit 3 includes the dialogue control rule 11, the situation DB 12, the statement type discrimination means 13, the dialogue action selection means 14, the situation condition discrimination means 15, the speech promotion condition discrimination means 16, and the question control. And means 17.

(Interaction control rules)
The dialogue control rule 11 indicates dialogue control rule storage means for storing a plurality (for example, 20) of dialogue control rules CR1 to CR20.
Each of the dialogue control rules CR1 to CR20 includes a condition part CR1a to CR20a and an action part CR1b to CR20b corresponding to the condition part. When all the conditions included in each condition part are satisfied, the rule is selected, and the dialogue action and accompanying effects described in the action part are selected as dialogue action candidates. Specifically, the speech type determination unit 13, the dialogue action selection unit 14, the situation condition determination unit 15, and the speech promotion condition determination unit 16 cooperate to refer to the dialogue control rule 11 and the situation DB 12. To select a dialogue action candidate.

  The situation DB 12 stores time information and emotional conditions as information relating to the current situation. The time information includes an elapsed time used for the above-described determination of HT, ST, and WT. The emotion condition indicates the current state such as the sympathy condition or the self-centered condition described above.

The speech type discriminating means 13 classifies the dialogue actions into a plurality (17) of speech types so that the dialogue action selection rule described above can be applied. Further, information indicating whether or not it is “answer evaluation type” is added so that it can be used in the processing of the agent message generation unit 4.
The dialogue action selection means 14 applies the aforementioned dialogue action selection rule.
The situation condition discriminating means 15 discriminates the following conditions (propositions) (1) to (4).
The speech promotion condition determination unit 16 determines whether the following (1) to (4) are satisfied.

  As a condition repeatedly used in the condition part of the dialog control rule, a condition called an answer agent utterance promotion condition is defined as follows.

(Responding agent speech promotion conditions)
The reply agent speech promotion condition is satisfied when all of the following propositions (1) to (4) are satisfied.
(1) In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
(2) The head of the message queue 6 is the next prompt type message.
<ID, TIME, USER, prompt, EXPR>
(3) USER is one of the users.
(4) In the dialog state 5, the following holds. That is, the target of prompting is an answer agent.
<Promotion object, ID, mushroom>

  The question control means 17 stores a question storage unit 21 for storing a quiz problem, a hint generation unit 22 for generating a hint sentence list in which a plurality of hints for a problem are arranged in descending order of difficulty, and a hint sentence list. A hint storage unit 25, a correctness / incorrectness determination unit 23 that determines whether the answer of the user or the answer agent is correct, and a similarity determination unit 24 that determines the degree of similarity between the correct answer and the answer when the answer is incorrect. ing.

  Since the question control means 17 includes the functions of both the question agent and the answer agent, the question control means 17 is described so as to straddle both the dialogue control unit 3 and the agent message generation unit 4 in FIG. . The question-taking agent speaks to present the participants with multiple hints of the problem in order from the most difficult in a series of questions to ask the question of the quiz. If the answer is an incorrect answer, the degree of similarity between the correct answer and the answer is determined, and a statement according to the degree of similarity at the time of the incorrect answer and a statement at the time of the correct answer are made. Also, the questioning agent or answering agent makes a statement according to the emotional condition when the answer of the participant is incorrect or correct. This question control means 17 can be realized by a known method, for example, the method described in Non-Patent Document 1. A specific example of the question control means 17 will be described later.

[Specific examples of dialog control rules]
Below, each dialog control rule in this embodiment is shown.
In the description of the dialog control rule, a symbol starting with ID is a variable representing a speech ID.
Symbols that start with TIME are variables that represent the time of speech.
Symbols that begin with EXPR are variables that represent speech expressions.
USER is a variable representing one of the users.
Symbols beginning with PARTICIPANT are variables that represent one of the dialog participants (user or dialog agent).
PERSON is a variable that represents the name of the person who has a quiz problem.

  Although the dialogue action candidate selected by the action unit of the dialogue control unit 3 includes variables representing the utterance time and the utterance expression, the values of these variables are not yet determined in the dialogue control unit 3. . The values of these variables are determined by the agent message generation unit 4.

In this embodiment, 20 dialogue control rules are prepared. There are five rules for selecting a dialogue act by the questioning agent and 15 rules for a dialogue act by the answering agent. The details of each dialog control rule will be described below.
First, the rules for selecting the dialogue action by the questioning agent are shown.

Rule 1 (Rule for starting the problem by the questioning agent: CR1)
Condition section (CR1a)
In dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the maximum silent time ST.
In the dialogue state 5, no question has been given yet (the dialogue has not started), or the correct answer end or the incorrect answer end dialogue action is executed for the question that has been asked recently.
Action part (CR1b)
The person name PERSON to be quized is selected (not disclosed to the user and the answer agent), a hint sentence relating to the person name is generated according to the method of Non-Patent Document 1, and the hint sentence is sent to the agent statement generation unit 4.
The next dialogue act and accompanying effect are selected as the question start type dialogue act of the questioning agent (T201).
<ID, TIME, foo, problem start, EXPR>
<Problem, ID, PERSON>
<Current problem, ID>

  As described above, the value of the dialogue action candidate variable selected by the action unit of the dialogue control unit 3 is determined by the agent statement generation unit 4, but for reference, an example of the variable is The expression R01 corresponds. "I will start a person quiz", "Then it will be the next problem"

Rule 2 (Rule for presenting hints by the questioning agent: CR2)
Condition part (CR2a)
Neither the correct answer nor the incorrect answer has been completed for the current problem.
In dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the maximum silent time ST.
In the dialog state 5, no hint has been presented yet, or the time from the latest hint presentation speech time to the present time exceeds the hint interval time HT.
The speech queue 6 is empty.
Action part (CR2b)
The next dialogue act and accompanying effect are selected as the hint presenting type dialogue act by the questioning agent (T202).
<ID, TIME, foo, hint presentation, EXPR>

Rule 3 (Rules for evaluating responses by the questioning agent: CR3)
Condition section (CR3a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
The head of the message queue 6 is an answer type interactive action by the user or the answer agent shown below (T206).
<ID1, TIME1, PARTICIPANT, answer, EXPR1>
PARTICIPANT is either a user or an answer agent.
Answer EXPR1 is an incorrect answer to the current problem.
Action part (CR3b)
The following dialogue action and accompanying effect are selected as the answer evaluation type dialogue action of the questioning agent (T203).
<ID2, TIME2, Hoofoo, answer evaluation, EXPR2>
<Response evaluation target, ID2, ID1, PARTICIPANT>

Rule 4 (Rule for correct answer end by the questioning agent: CR4)
Condition section (CR4a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the answer waiting time WT.
In the dialogue state 5, neither the correct answer end nor the incorrect answer end has been performed for the current problem.
The head of the message queue 6 is an answer type interactive action by the user or the answer agent shown below (T206).
<ID1, TIME1, PARTICIPANT, answer, EXPR1>
PARTICIPANT is either a user or an answer agent.
Answer EXPR1 is the correct answer to the current question.
Action part (CR4b)
The next dialogue act is selected as the correct answer end type dialogue act by the questioning agent (T204).
<ID2, TIME2, Foo, correct end, EXPR2>
<Correct answer completion target, ID2, ID1>

Rule 5 (Rule for terminating incorrect answer by questioning agent: CR5)
Condition section (CR5a)
In dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the maximum silent time ST.
In the dialogue state 5, neither the correct answer end nor the incorrect answer end has been performed for the current problem.
The speech queue 6 is empty.
In the dialog state 5, the time from the previous hint presentation time to the present time exceeds the hint interval time HT.
There are no hints to present.
Action part (CR5b)
The next dialogue action is selected as the correct answer end type dialogue action by the questioning agent (T205).
<ID, TIME, foo, incorrect answer termination, EXPR>

Next, the rules for interactive actions by answering agents are shown. First, after the questioning agent performs an answer evaluation on the answer of the user or the answer agent, a rule for the answer agent to respond to the answer evaluation (answer evaluation answer) is shown.
The answer evaluation response type dialogue action is selected based on different rules for the sympathetic condition, the self-centered condition, and the emotionless condition. That is, in the case of the sympathy condition, the response evaluation response is performed only for the response evaluation to the user's response. In the case of self-centered conditions, the response evaluation response is performed only for the response evaluation to the response of the response agent “Mushroom”. When the emotional condition is satisfied, an answer evaluation response is performed for an answer evaluation to the answer of the user or the answer agent.

Rule 6 (Rules for Response Evaluation by Agents under Sympathy Conditions: (CR6)
Condition section (CR6a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
Sympathy conditions are met.
The head of the message queue 6 is the following reply evaluation type interactive action (T203). In order to combine with T207 of the action part, “1” and “2” were exchanged in the condition part (the same applies hereinafter).
<ID1, TIME1, FOOO, answer evaluation, EXPR1>
In the dialog state 5, the following holds.
<Response evaluation target, ID1, ID2, USER>
USER is one of the users.
Action part (CR6b)
The next dialogue action is selected as the answer evaluation response type dialogue action by the answer agent (T207).
<ID3, TIME3, mushroom, response evaluation, EXPR3>
<Response evaluation response target, ID3, ID1, ID2, USER>

Rule 7 (Rule for Response Evaluation by Response Agents under Self-Centered Conditions: CR7)
Condition section (CR7a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
Self-centered condition is established.
The head of the message queue 6 is the following reply evaluation type interactive action (T203).
<ID1, TIME1, FOOO, answer evaluation, EXPR1>
In the dialog state 5, the following holds.
<Response evaluation target, ID1, ID2, mushroom>
Action part (CR7b)
The next dialogue action is selected as the answer evaluation response type dialogue action by the answer agent (T207). Note that P2 = P3 = answer agent.
<ID3, TIME3, mushroom, response evaluation, EXPR3>
<Respondents for answer evaluation, ID3, ID1, ID2, mushroom>

Rule 8 (Rule for answer evaluation by answer agent in emotionless condition: CR8)
Condition section (CR8a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
An emotionless condition is met.
The head of the message queue 6 is the following reply evaluation type interactive action (T203).
<ID1, TIME1, FOOO, answer evaluation, EXPR1>
In the dialog state 5, the following holds.
<Response evaluation target, ID1, ID2, PARTICIPANT>
PARTICIPANT is a user or answer agent “mushroom”.
Action part (CR8b)
The next dialogue action is selected as the answer evaluation response type dialogue action by the answer agent (T207). P2 is PARTICIPANT and P3 is an answer agent.
<ID3, TIME3, mushroom, response evaluation, EXPR3>
<Response evaluation response target, ID3, ID1, ID2, PARTICIPANT>

  Next, a rule for selecting an interactive action of an answer agent's impression at the end of correct answer-self, impression at the end of correct answer-others, impression at the end of incorrect answer is shown.

Rule 9 (Comments from the answering agent at the end of the correct answer-Rules for me: CR9)
Condition section (CR9a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
The head of the message queue 6 is a correct end type interactive action shown below (T204).
<ID1, TIME1, FOOO, correct end, EXPR1>
In the dialog state 5, the following holds (T206). That is, it is the answering agent itself that has made the correct answer.
<Correct answer completion target, ID1, ID2>
<ID2, TIME2, mushroom, answer, EXPR2>
Action part (CR9b)
The next dialogue action is selected as the answer at the end of the correct answer by the answering agent-white part type dialogue action (T212).
<ID3, TIME3, Mushroom, Impressions at the end of correct answer-yourself, EXPR3>
<Study object at the end of correct answer, ID3, ID1, ID2, mushroom>

Rule 10 (Comments from the answering agent at the end of the correct answer-Rules for others: CR10)
Condition section (CR10a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
The head of the message queue 6 is a correct end type interactive action shown below (T204).
<ID1, TIME1, FOOO, correct end, EXPR1>
In the dialog state 5, the following holds (T206).
<Correct answer completion target, ID1, ID2>
<ID2, TIME2, USER, answer, EXPR2>
USER is one of the users. That is, it is not the answer agent but the user who has made the correct answer.
Action part (CR10b)
Impression at the end of correct answer by answering agent-The next dialogue action is selected as the dialogue action of the other person type (T213).
<ID3, TIME3, Mushroom, Impression at the end of correct answer-Others, EXPR3>
<Impression target ID3, ID1, ID2, USER at the end of correct answer>

Rule 11 (Rule for Impression of Improper Answer End by Answer Agent: CR11)
Condition section (CR11a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
The head of the speech queue 6 is an incorrect answer end type interactive action shown below (T205).
<ID1, TIME1, FOOO, incorrect answer end, EXPR1>
Action part (CR11b)
The next dialogue act is selected as the dialogue action of the response type at the end of the incorrect answer by the answering agent (T214).
<ID2, TIME2, Mushroom, Impression at the end of incorrect answer, EXPR2>

Next, after the user performs difficulty level expression and difficulty level expression, a rule for the answer agent to perform difficulty level response and difficulty level response to the response agent is shown.
In the present embodiment, the rules are described so that both the difficulty expression response and the difficulty expression response are selected only when the sympathy condition is satisfied.
This is because, in the present embodiment, the speech expression for the difficulty expression response and the difficulty expression response prepared by the agent statement generation unit 4 always has a sympathetic effect that cares about the user. is there.
By describing the dialog control rules in this way, it is possible to avoid executing a dialog act that unintentionally conveys empathy to the user.

Rule 12 (Rules for responding to difficulty level of answering agent: CR12)
Condition section (CR12a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
Sympathy conditions are met.
The head of the speech queue 6 is the following difficulty level type interactive action by the user (T208).
<ID1, TIME1, USER, difficulty expression, EXPR1>
USER is one of the users.
Action part (CR12b)
The following dialogue action and accompanying effect are selected as the answering agent's response type interaction action (T210).
<ID2, TIME, mushroom, difficulty level response, EXPR2>
<Target, ID2, ID1>

Rule 13 (Rules for answering agent's difficulty level response: CR13)
Condition section (CR13a)
In the dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the speech start waiting time WT.
Sympathy conditions are met.
The head of the speech queue 6 is the following easy-to-express dialogue type user action (T209).
<ID1, TIME1, USER, difficulty expression, EXPR1>
USER is one of the users.
Action part (CR13b)
The following dialogue act and accompanying effect are selected as the answering agent interaction type dialogue act of the answering agent (T211).
<ID2, TIME, mushroom, difficulty level response, EXPR2>
<Service ID 2, ID1>

Next, a rule for selecting a difficulty expression and a difficulty expression dialogue action by an answer agent is shown.
The dialogue action of the difficulty expression type and the difficulty expression type is selected according to different dialogue control rules in the sympathy condition and in the self-centered condition or the emotionless condition.
That is, when the self-centered condition or the emotionless condition is satisfied, when the head of the speech queue 6 is the user's difficulty expression type or difficulty expression type interactive action, the difficulty agent expression type and difficulty by the answer agent Do not select an expression-type dialogue act.
In this way, when only the self-centered condition is satisfied, or when the emotionless condition is satisfied, the difficulty expression type or the difficulty expression type that unintentionally conveys empathy to the user It is possible to avoid the execution of a dialogue act, and it is possible to avoid a statement that conveys sympathy to the user unintentionally, such as the statement (5, 4) of the dialogue example 5 shown in the problem to be solved by the invention It becomes.

Rule 14 (Rule for difficulty level expression by answering agents under empathy conditions: CR14)
Condition section (CR14a)
In the dialogue state 5, the elapsed time from the speech of the previous dialogue participant to the present time exceeds the maximum silent time ST, or the answer agent speech promotion condition is satisfied.
Sympathy conditions are met.
Action part (CR14b)
The next dialogue action is selected as the difficulty level expression type dialogue action of the answering agent (T208).
<ID, TIME, mushroom, difficulty level expression, EXPR>

Rule 15 (Rule for expressing difficulty of answering agent under self-centered or emotionless condition: CR15)
Condition section (CR15a)
In the dialogue state 5, the elapsed time from the speech of the previous dialogue participant to the present time exceeds the maximum silent time ST, or the answer agent speech promotion condition is satisfied.
Self-centered or emotionless conditions are met.
The head of the speech queue 6 is not the user's difficulty expression type interactive action (T208).
Action part (CR15b)
The next dialogue action is selected as the difficulty level expression type dialogue action of the answering agent (T208).
<ID, TIME, mushroom, difficulty level expression, EXPR>

Rule 16 (Rule for expression of difficulty by answering agent under empathy conditions: CR16))
Condition section (CR16a)
In the dialogue state 5, the elapsed time from the speech of the previous dialogue participant to the present time exceeds the maximum silent time ST, or the answer agent speech promotion condition is satisfied.
Sympathy conditions are met.
Action part (CR16b)
The next dialogue action is selected as the dialogue action of the answer agent's difficulty expression type (T209).
<ID, TIME, mushroom, difficulty expression, EXPR>

Rule 17 (Rule for expression of difficulty of answering agent under self-centered or emotionless condition: CR17)
Condition section (CR17a)
In the dialogue state 5, the elapsed time from the speech of the previous dialogue participant to the present time exceeds the maximum silent time ST, or the answer agent speech promotion condition is satisfied.
Self-centered or emotionless conditions are met.
The head of the speech queue 6 is not the user's easy expression type interactive action (T209).
Action part (CR17b)
The next dialogue action is selected as the dialogue action of the answer agent's difficulty expression type (T209).
<ID, TIME, mushroom, difficulty expression, EXPR>

  Next, the rules for selecting the answer, prompting, and dialogue action by the answer agent are shown.

Rule 18 (Rule for answering by an answer agent: CR18)
Condition section (CR18a)
In the dialogue state 5, the elapsed time from the speech of the previous dialogue participant to the present time exceeds the maximum silent time ST, or the answer agent speech promotion condition is satisfied.
Action part (CR18b)
Generate a personal name PERSON that will answer the current question.
The next dialogue act is selected as the answer type dialogue act of the answer agent (T206).
<ID, TIME, mushroom, answer, EXPR>
<Respondent name, ID, PERSON>

Rule 19 (Rule for prompting by answering agent: CR19)
Condition section (CR19a)
In dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the maximum silent time ST.
Action part (CR19b)
The next dialogue action is selected as the prompt action of the answer agent (T215).
<ID, TIME, mushroom, prompt, EXPR>

Rule 20 (Rule for Answering by Answer Agent: CR20)
Condition section (CR20a)
In dialog state 5, the elapsed time from the speech of the previous dialog participant to the present exceeds the maximum silent time ST.
It is a condition of empathy.
The head of the speech queue 6 is another type of interactive action by the user.
Action part (CR20b)
The next dialogue action is selected as the answering action of the answering agent (T216).
<ID, TIME, Mushroom, Aichi, EXPR>

[Details of agent remark generation part]
In the present embodiment, as shown in FIG. 6, the agent statement generation unit 4 includes a statement generation unit 18 and a statement generation rule 19.
The utterance generation rule 19 indicates utterance generation rule storage means for storing utterance expressions R01 to R16b corresponding to plural types (for example, 34 types shown in Tables 6 to 8).
The speech generation means 18 selects the speech expressions R01 to R16b with reference to the speech generation rule 19 based on the dialogue action (+ accompaniment effect) acquired from the dialogue control unit 3.

[Specific example of statement generation rule]
In the present embodiment, the following statement generation rule is used.
The statement generation rule is different for an answer evaluation type dialogue action and other types of dialogue actions. By using the following statement generation rules for answer evaluation, even if the user and the dialogue agent are allowed to speak asynchronously at any time, the content of the dialogue agent's utterance can be conveyed without misunderstanding. This makes it possible to avoid misleading statements such as the statement (1.5) in Dialogue Example 1 shown in the problem to be solved by the invention.

(1) Statement Generation Rule for Answer Evaluation Type Dialog Action Assume that the dialog control unit 3 sends the following dialog action and accompanying effects as the answer evaluation type dialog action of the questioning agent.
<ID2, TIME2, AGENT, response evaluation, EXPR2>
<Response evaluation target, ID2, ID1, PARTICIPANT>
Further, it is assumed that the following holds in the dialog state 5 (T206).
<ID1, TIME1, PARTICIPANT, answer, EXPR1>
In other words, the user or the response agent PARTICIPANT performs the response type dialogue action ID1 with the speech expression EXPR1 at the utterance time TIME1, and the question agent AGENT performs the response evaluation type dialogue action ID2 for the answer type dialogue action ID1. I'm going.
At this time, if either of the following conditions (A) and (B) is satisfied, a redundant expression (R03b, R03d: see Table 6) is selected as the reply expression of the reply evaluation type. (R03a, R03c :: Refer to Table 6) is selected.

(A) In the dialogue state 5, another answer type dialogue act is executed between the comment time TIME1 of the answer ID1 and the current time.
(B) There is a dialogue action ID3 of another response type executed before the utterance time TIME1 of response ID1, and the response evaluation for response ID3 is performed between the other response ID3 and response ID1. Absent.

  The redundant and concise expressions corresponding to the answer evaluation type dialogue act are selected according to the following table according to the emotional conditions.

(2) Statement generation rules for dialogue actions other than the answer evaluation type For dialogue actions other than the answer evaluation type, a speech expression is selected according to emotional conditions according to the following table.

[Specific examples of question control means]
Since the question control means 17 shown in FIG. 6 is characterized by using the hint sentence list generated by the hint generator 22, first, the hint generator 22 will be described in detail with reference to FIG. In the following description, it is assumed that the correct answer is a person name, and a hint for hitting the person name is generated as a question.

(Configuration example of hint generation unit)
The hint generation unit 22 shown in FIG. 7 generates a correct answer and generates a plurality of hint sentences corresponding to the generated correct answer. The hint generator 22 is connected to the network data server 31 via a network N such as the Internet. The network data server 31 stores various information (text data group) used for the hint sentence. There are a plurality of network data servers 31.

  The hint generation unit 22 is connected to the question answering system 32. The question answering system 32 is a known QA (Question Answering) system, for example, stores a large number of document sets (database) such as newspaper articles in a storage means (not shown), and searches for answers to the questions from the database. Here, the questions are, for example, “What is the related word of XX?”, “Where is related to XX?”, “Who is related to XX?”, And the like.

  As shown in FIG. 7, the hint generation unit 22 includes a communication interface 41, storage means 42, correct answer generation means 43, hint expression acquisition means 44, hint sentence candidate generation means 45, and hint sentence list generation means 46. And.

The communication interface 41 inputs predetermined information (such as a hint expression such as a keyword or a phrase) from the network data server 31.
The storage means 42 includes a correct answer candidate database 421, a hint sentence candidate database 422, and a hint sentence list database 423.
The correct answer candidate database 421 stores a plurality of correct answer candidates, and is, for example, a person name list in which a plurality of person names are recorded.
The hint sentence candidate database 422 stores hint sentence candidates generated by the hint sentence candidate generating unit 45. Here, the hint sentence candidate is a sentence including hint expressions such as keywords or phrases.
The hint sentence list database 423 stores the hint sentence list generated by the hint sentence list generating unit 46. Here, the hint sentence list is composed of sentences (hint sentences) selected from the hint sentence candidates for questions and arranged in a predetermined order.

The correct answer generation means 43 selects a predetermined correct answer candidate from the correct answer candidate database 421, generates the selected correct answer candidate as a correct answer, and outputs it to the hint expression acquisition means 44.
For example, when the correct answer generation means 43 selects “Nobunaga Oda” as the name of the person, there is a problem of hitting “Nobunaga Oda”, and the user hits who the hints that are sequentially issued indicate. . Hereinafter, for example, a description will be given assuming that a hint sentence for assigning a person name “Oda Nobunaga” is generated. Note that the method for creating a correct answer is the same as the method for creating a keyword entry quiz by presenting a definition sentence in the dictionary, so a known problem creating method is used.

  The hint expression acquisition unit 44 acquires a plurality of hint expressions as hints corresponding to the correct answer generated by the correct answer generation unit 43 from the database 31 a stored in the network data server 31 via the network N. Here, the hint expression indicates a hint expressed by a keyword or a phrase. Hereinafter, a text data group that can be acquired via the network N is referred to as Web data. Web data includes, for example, a personal dictionary on the Internet. On the Internet, for example, in the case of “Oda Nobunaga”, there is currently a personal dictionary in which category information such as “Sengoku warlord” is added as metadata. Therefore, such data is assumed in the case of Web data.

The hint expression acquisition unit 44 acquires a first expression acquisition unit 441, a second expression acquisition unit 442, a third expression acquisition unit 443, and a fourth expression in order to acquire a hint expression that is considered to explain the correct answer. And an expression acquisition unit 444.
Although not shown, the first expression acquisition means 441 includes means for executing a well-known phrase search and means for executing a morphological analysis. Characteristic expressions (keywords) are obtained from Web data obtained from the network N. It is extracted as a hint expression. Specifically, a characteristic pattern for explaining a person is captured, and a hint expression “Sengoku Daimyo” is acquired from a sentence such as “Nobunaga Oda etc.”.

  The second expression acquisition unit 442 extracts a correct reading kana obtained from the morphological analysis by the first expression acquisition unit 441 as a hint expression. The second expression acquisition unit 442 creates a hint sentence from the correct answer (person name) notation (first character, last character). If the subject person name is analyzed by morphological analysis, the reading character can be acquired, so the first character and the last character are acquired.

  The third expression acquisition unit 443 extracts phrases from the Web data obtained from the network N as hint expressions. Here, the phrase is a verb phrase. The third expression acquisition unit 443 creates a hint sentence from a sentence including A (correct answer) without using a characteristic expression. In this case, if a related verb phrase of the expression “A is” is found, the verb phrase is used as a hint expression. For example, if “Oda Nobunaga opened Rakuichi Rakuza”, “Oda Nobunaga” opened “Rakuichi Rakuza”, so “Rakuichi Rakuza was opened” is used as a hint expression.

  The fourth expression acquisition unit 444 outputs a question to the question answering system 32 and acquires the response of the question answering system 32 as a hint expression. The fourth expression acquisition unit 444 generates three question sentences, “What is the related word of XX?”, “What is the place related to XX?” And “What is the person related to XX?” Is sent to the question answering system 32 and the response is used as a hint expression.

  The hint sentence candidate generating unit 45 generates a hint sentence candidate indicating a sentence including the hint expression acquired by the hint expression acquiring unit 44, and includes a first candidate generating unit 451, a second candidate generating unit 452, Third candidate generation means 453 and fourth candidate generation means 454 are provided.

  The first candidate generation unit 451 creates a hint sentence for the hint expression acquired by the first expression acquisition unit 441 by using a template of “XXX is,” “XX is,”. For example, when the category information related to the person is “Sengoku warlord”, the first candidate generating means creates a hint sentence “It ’s a warrior warrior”.

  The second candidate generation unit 452 creates a hint sentence using the template for the hint expression acquired by the second expression acquisition unit 442. The second candidate generation means 452 creates a hint sentence from the name of the person using the templates “the first character is ○” and “the last character is ○”.

  The third candidate generation unit 453 creates a hint sentence using a template for the hint expression acquired by the third expression acquisition unit 443. The third candidate generating means 453 applies “yo” after making the final verb a past form by applying a syntactic rule to the ending of the sentence. That is, a hint sentence is created using a template “OO Tayo”. For example, when the hint expression is “killed by Akechi Mitsuhide”, the hint sentence is “I was killed by Akechi Mitsuhide”.

  The fourth candidate generation unit 454 creates a hint sentence using the template for the hint expression acquired by the fourth expression acquisition unit 444. As with the third candidate generation unit 453, the fourth candidate generation unit 454 applies a syntactic rule to the end of the sentence to make the final verb a past form, It is given.

As shown in FIG. 7, the hint sentence list generating unit 46 includes a hint keyword extracting unit 461, a feature amount extracting unit 462, a score calculating unit 463, a sorting unit 464, and a storage unit 465.
The hint keyword extraction unit 461 extracts a keyword from each hint sentence candidate stored in the hint sentence candidate database 422. Hereinafter, the keyword extracted by the hint keyword extracting unit 461 is simply referred to as a keyword. The keyword is, for example, a noun as a result of morphological analysis. For example, when the hint sentence candidate is “A military commander”, “Wushou” is a keyword.

  The feature amount extraction unit 462 includes, in the Web data, the number of documents including the correct answer (for example, html document) E (feature amount related to the document) E and the number of documents including both the correct answer and the keyword (feature amount related to the document) F. And the number of all documents (features related to documents) G in the target Web data and the number of keyword hits (features related to documents) H in the target Web data. Note that the feature amount G may be replaced with a very large value.

The score calculation unit 463, based on each feature amount (E, F, G, H) extracted by the feature amount extraction unit 462, the co-occurrence frequency of the correct answer and the keyword in the Web data, and the general keyword in the Web data The score (first score) S ₁ for each hint sentence candidate is calculated based on the formula (1).

S ₁ = log (F / E) + log (H / G) (3)

In Expression (3), the degree of association F / E between the correct answer and the answer is used as the co-occurrence frequency of the correct answer and the keyword. A hint sentence candidate having a high degree of relevance F / E is considered to be one in which the correct answer can be easily understood by the user. However, there is a possibility that some of the keywords have a strong co-occurrence with the correct answer but are rarely used and difficult for the user to understand. Therefore, in formula (3), the generality of keywords in Web data is H / G. The larger the value of H / G, the more familiar the keyword is to the user. That is, equation (3) indicates that the greater the value of the first score S ₁ , the more difficult it is to understand the correct answer. For example, in the case of Oda Nobunaga, the hint that “I was killed by Mitsuhide Akechi” is more related to the hint than Oda Nobunaga, and it seems to be directly linked to the answer.

The sorting unit 464 determines the priority order of each hint sentence candidate based on the score (first score S ₁ ) of each hint sentence candidate, arranges the hint sentence candidates in the determined priority order, and generates a hint sentence list. To do. Here, sorting means 464 arranged above (the beginning) of about the first score S ₁ value is less hint sentence candidate preferentially hint statement list. By arranging in this way, it is possible to realize an arrangement in which correct answers are not immediately known.

Specifically, if there hint sentence candidate number, ranking lower hint sentence candidates (keywords first value of the score S ₁ is low), the relevance is low, generality is also considered to be low. Therefore, a hint sentence candidate with a low ranking is highly likely not necessary as a hint sentence. Therefore, the sorting unit 464 leaves only the one with the top J ranking (for example, J = 10) among the many hint sentence candidates. Then, the sorting unit 464 reverses the hint sentences up to the top J position in order, and forms a hint sentence list. The hint sentence list generated by the sorting unit 464 is stored in the hint sentence list database 423 by the storage unit 465 for each correct answer.

(Operation of the hint generator)
A series of operations of the hint generation unit 22 shown in FIG. 7 will be described with reference to FIG. 8 (refer to FIG. 7 as appropriate). FIG. 8 is a flowchart showing the operation of the hint generator shown in FIG.
The hint generation unit 22 selects the correct answer candidate by the correct answer generation means 43 to generate a correct answer (step S101), and acquires the hint expression corresponding to the correct answer by the hint expression acquisition means 44 (step S102). Then, the hint generation unit 22 generates a hint sentence candidate using the acquired hint expression by the hint sentence candidate generation unit 45 (step S103), and the hint keyword extraction unit 461 extracts a keyword from each hint sentence candidate (step S103). S104). Then, the hint generation unit 22 extracts a plurality of feature amounts related to the keyword from the Web data via the network N by the feature amount extraction unit 462 (step S105), and based on the feature amount extracted by the score calculation unit 463. The first score for each hint sentence candidate is calculated (step S106). Subsequently, the hint generation unit 22 generates a hint sentence list based on the score of each hint sentence candidate and the question rule by the sorting unit 464 (step S107), and the storage unit 465 sorts the hint sentence list by problem. Store (step S108).

(Concrete example)
Table 9 is a hint sentence list created for the correct answer “Nobunaga Oda” using Web data acquired using the search engine goo (registered trademark).

(Configuration example of question control means)
FIG. 9 is a block diagram showing a configuration example of the question control means. As shown in FIG. 9, the question control unit 17 includes a communication interface 41, a storage unit 42, a match determination unit 33, a search unit 34, a similarity calculation unit 35, a hint sentence selection unit 36, a response sentence. A generation unit 37, a question management unit 38, and an answer extraction unit 39 are provided.
The storage means 42 includes a correct answer candidate database 421, a hint sentence list database 423, and a response sentence database 424. The response sentence database 424 stores response sentences for answers. As the response sentence, a question start declaration sentence, a response sentence when the answer matches the correct answer, a response sentence when the answer does not match the correct answer, and the like are prepared.

  The coincidence determination means 33 determines whether or not the answer and the correct answer match when there is an answer (input). When the answer and the correct answer match, the match determining means 33 notifies the response sentence generating means 37 to that effect, and when the answer does not match the correct answer, the search means 34 and the similarity calculating means 35 are notified. To that effect.

  When the search means 34 is notified from the match determination means 33 that the answer and the correct answer do not match, in the Web data, both the data group containing the correct answer, the data group containing the answer, and both the answer and the correct answer The number of documents in each of the data group including “” is searched. The search result is output to the similarity calculation means 35. Specifically, the search means 34 uses, for example, a personal dictionary on the network N to determine how articles (explanatory pages) relating to two person names respectively indicating answers and correct answers from the articles of the personal dictionary. To see if it is referenced by a hyperlink. That is, for the article (explanation page) related to the person name indicating the answer and the article (explanation page) related to the person name indicating the correct answer, the search means 34 has a common article (not limited to the person name) referring to both. Search for it.

  The similarity calculation means 35 calculates the degree of co-occurrence between the answer and the correct answer in the network N (text data group) as the similarity between the answer and the correct answer based on the result searched by the search means 34. Is. The calculated similarity is output to the response sentence generation unit 37 and the question management unit 38. In the present embodiment, for example, the similarity calculation means 35 tests the co-occurrence level by the Fisher's exact test (Fisher's exact test, FET: Fisher's Exact Test). Calculate as degrees.

When the similarity calculation means 35 calculates the similarity by using an FET, for example, the correct answer is “Nobunaga Oda”, the answer is “Hideyoshi Toyotomi”, the number of pages of an article relating to “Nobunaga Oda”, and “Hideyoshi Toyotomi” When the page number of the article and the page number Q of the target person encyclopedia are obtained as shown in Table 10, the p-value of the FET is expressed by Expression (4). The FET is described in “Agresti A, (1992), A Survey of Exact Inference for Contingence Tables, Statical Science, 7, 131-153”. Here, the calculation of the similarity is not limited to the FET, but the chi-square value (χ ² ) used for the statistical test, and the value obtained by normalizing the chi-square value to a range of 0 to 1 ² etc. can be used.

The hint sentence selection means 36 selects hint sentences according to the priority order from the hint sentence list stored for each problem in the hint sentence list database 423.
When the hint sentence list is rearranged based on the instruction from the question management means 38, the hint sentence selecting unit 36 selects the hint sentence based on the rearranged order and rearranges the hint sentence list. The order information is notified to the question management means 38.

  For example, when the number of hint sentences recorded in the hint sentence list is 10 and 6 questions are given and then rearranged, the hint sentence that was initially “9” in question order is newly added to “7 After the “th” mark, the hint sentence that was originally “7th” is newly “8th”. Here, the rearrangement includes hollowing out. For example, the hint sentence which was originally “7” to “9” is omitted (skip), and the hint sentence which was initially “10” in the order of questions is newly set as “7”, and the hint is skipped. Cancel the statement.

The response sentence generation unit 37 generates a response sentence for the answer based on the similarity calculated by the similarity calculation unit 35. The response sentence in this case is for informing the user how close the answer is to the correct answer.
The response sentence generation unit 37 generates a response sentence for the answer when the answer and the correct answer match. The response sentence in this case is for informing the user that the answer is correct. The response sentence is generated according to a predetermined rule.
In the present embodiment, the response sentence generation unit 37 generates a response sentence by selecting a response sentence from the response sentence database 424. When notified from the question management means 38 that the hint sentence list has been rearranged, the response sentence generating means 37 generates a response sentence based on the notified information. The response sentence when the hint sentences are simply rearranged is, for example, “Can you understand this?”. In addition, as a result of the rearrangement of hint sentences, the response sentence when the last hint sentence is given after removing the middle part is, for example, “This is the last hint”.

  The question management unit 38 determines whether an answer has been input, and repeatedly questions different questions until the input answer matches the correct answer. The question management means 38 outputs the answer received from the input device M to the coincidence determination means 33, and outputs the response sentence generated by the response sentence generation means 37 and the hint sentence selected by the hint sentence selection means 36. Output to device D. Here, the input device M is, for example, a keyboard or a mouse, and the output device D is, for example, a liquid crystal display.

  When the quiz starts, the question management means 38 outputs an initial response sentence (question start declaration sentence) and then outputs the first hint sentence. After starting the quiz, if the answer is correct, the question management means 38 outputs a response sentence corresponding to the correct answer, and if it is incorrect, outputs a response sentence corresponding to the similarity and then outputs the next hint sentence To do. For this reason, the question management means 38 manages the number of questions (number of answers) and controls the response sentence generation means 37 and the hint sentence selection means 36 according to the number of questions.

  In the present embodiment, the question management unit 38 determines whether or not the similarity (p value) acquired from the similarity calculation unit 35 is equal to or greater than a predetermined value γ. The hint sentence selection means 36 is instructed to sort the hint sentence list. The predetermined value γ may be set to 0.01, for example. The question management unit 38 acquires the result of the rearrangement process from the hint sentence selection unit 36 and notifies the response sentence generation unit 37 of the result.

  The answer extracting means 39 is a function of an answer agent, and extracts answers from the correct answer candidate database 421 and outputs the extracted answers to the question management means 38. In this case, the question management unit 38 outputs the answer received from the answer extraction unit 39 to the match determination unit 33. The question management means 38 receives a predetermined command (start or end of a quiz, skip of a hint sentence, etc.) from the input device M.

(Operation of the question control means)
The operation of the question control unit shown in FIG. 9 will be described with reference to FIG. 10 (see FIG. 9 as appropriate). FIG. 10 is a flowchart showing the operation of the question control means shown in FIG.
The question control means 17 initializes the control variable i (i = 0) by the question management means 38 (step S111), and generates the question start declaration sentence (i = 0) by the response sentence generation means 37 (step S111). S112). Then, the question control unit 17 adds “1” to the current value of the control variable i by the question management unit 38 (step S113), and determines whether the similarity (p value) is equal to or greater than a predetermined value γ. It discriminate | determines (step S114). If the similarity (p value) is equal to or greater than the predetermined value γ (step S114: Yes), the question control means 17 selects the i-th hint sentence by the hint sentence selection means 36 (step S115), The hint text is displayed by the management means 38. Then, the question control means 17 determines whether or not there is an input by the question management means 38 (step S116). When there is an input (step S116: Yes), the question control unit 17 determines whether or not they match by the match determination unit 33 (step S117). When they do not match (Step S117: No), the question control means 17 uses the search means 34 to search the Web data with a data group including the correct answer, a data group including the answer, and a data group including both the answer and the correct answer. Each document number is searched (step S118). Then, the question control unit 17 calculates the similarity by the similarity calculation unit 35 (step S119), and generates a response sentence by the response sentence generation unit 37 (step S120).

  On the other hand, if they match in step S117 (step S117: Yes), the question control means 17 proceeds to step S120. Subsequent to step S120, the question control means 17 determines whether or not the quiz is ended by the question management means 38 (step S121). When it is the end of the quiz (step S121: Yes), the question control unit 17 ends the process. On the other hand, when it is not the end of the quiz (step S121: No), the question control means 17 returns to step S113.

  If the similarity (p value) is smaller than the predetermined value γ in step S114 (step S114: No), the question control means 17 performs a rearrangement process by the hint sentence selection means 36 (step S122). Then, a hint sentence is selected (step S123). If there is no input in step S116 (step S116: No), the question control means 17 determines whether or not the hint interval time HT has passed by the question management means 38 (step S124). When the hint interval time HT has elapsed (step S124: Yes), the question control unit 17 proceeds to step S120. On the other hand, when the hint interval time HT has not elapsed (Step S124: No), the question control means 17 returns to Step S116.

  Note that the multi-person thought-inducing interactive device 1 is operated by a program (a multi-person thought-inducing interactive program) that causes a general computer to function as each of the above-described means constituting the multi-person thought inducing interactive device 1. Can be realized. This program can be provided via a communication line, or can be written on a recording medium such as a CD-ROM and distributed.

  According to the present embodiment, in a multi-person thought-inducing dialogue in which a plurality of people and a plurality of dialogue agents participate, the dialogue status relating to the content and timing of the utterances of the agent and the user is managed by the storage means, and is determined in advance. Based on the dialogue control rule and the speech generation rule, the agent can output a sympathetic speech or a self-centered speech with appropriate content and appropriate timing. Therefore, the degree of satisfaction that humans feel, the sense of familiarity and likability with agents, and the sense that they are being cared for can be improved, and the amount of communication, including interaction between humans, can be increased and communication can be activated. It becomes. As a result, the effect of strengthening the connection, closeness, and solidarity between humans participating in the multi-person dialogue can be expected.

  As mentioned above, although embodiment of this invention was described, this invention is not limited to this, It can implement in the range which does not change the meaning. For example, the number of users and the number of agents can be changed as appropriate. For example, in the present embodiment, the personal name guessing quiz is described, but the present invention is not limited to this. For example, an organism name such as an animal name or a plant name, an inanimate name such as a place name, a product name, a learning term name, or an academic terminology name may be used. Also, the question control means 17 shown in FIG. 9 and the hint generator 22 shown in FIG. 7 are examples, and the present invention is not limited to this. In addition, the dialogue is not limited to the quiz dialogue format as long as it is a multi-person thought-inducing dialogue that arouses the user's thoughts by the remarks of the agent.

  In order to confirm the effect of the present invention, the performance of the multi-person thought-stimulating dialogue apparatus according to the present embodiment was confirmed by experiments. Experiment participants were recruited and system evaluation experiments were conducted. Thirty-two pairs were collected as two participants. Of the 32 pairs, 16 pairs were male and 16 pairs were female.

  (A) When there is no answer agent, (B) When there is an answer agent and the no emotion condition is satisfied, (C) When there is an answer agent and only the sympathy condition is satisfied, (D) When the answer agent exists and only the self-centered condition is satisfied, (E) five systems are prepared in which the answer agent exists and both the empathy condition and the self-centered condition are satisfied.

  Each pair of experimental participants interacted with the five systems in turn. For each pair, the order of interaction with the five systems was rearranged so that the influence of the interaction order with each system did not appear. In each dialogue, five quiz questions were presented. The degree of difficulty of the questions presented in the dialogue with each system is the same.

  After the dialogue with each system was completed, a questionnaire survey was conducted on the satisfaction with the dialogue and the impression of the answering agent. The questionnaire about the impression to the answer agent was done only for the system where the answer agent exists.

The following was found by analyzing the experimental results.
(1) Compared with the case where there is no answer agent, when the answer agent is present, the amount of speech of the user is statistically significantly increased, and the satisfaction degree of the dialogue is statistically significantly improved.
(2) When there is an answer agent, the amount of user's utterance increases statistically significantly when the sympathy condition is satisfied, and when the sympathy condition is not satisfied, the degree of satisfaction with the dialogue , The feeling of familiarity and favorableness to the answering agent and the sense of care by the answering agent were statistically significantly improved (p <0.01).
(3) When there is an answer agent, the preference for the answer agent is statistically significantly improved when the self-centric condition is satisfied (p <0) than when the self-centric condition is not satisfied. .01).
There was no statistically significant difference in the amount of user's utterance, the sense of familiarity with the answering agent, and the sense of care being taken by the answering agent, when the self-centered condition is satisfied and when it is not satisfied.

  From this result, it was found that the system of the present invention can activate communication in a dialogue with a large number of thoughts by allowing the dialogue agent to speak at an appropriate content and timing.

  In addition, empathy conditions are given from the outside as system emotion parameters, and by letting the dialogue agent make sympathetic remarks with appropriate content and timing according to the situation of the dialogue, the satisfaction of the dialogue, the familiarity / preference of the answering agent It can be seen that it is possible to improve the sensitivity, the sense of care by the answering agent, etc., and to activate communication including interaction between users.

  Furthermore, by giving a self-centered condition as a system emotion parameter from the outside, and letting the conversation agent speak self-centered with appropriate content and timing according to the state of the conversation, there is no increase in communication volume. , It was found that the user's preference for the answering agent can be improved.

1 Multi-person thinking-aware dialogue system (Multi-person thinking-aware dialogue device)
2 User remark understanding section (User remark understanding means)
3 Dialogue control unit (dialogue control means)
4 Agent statement generator (Agent statement generator)
5 Dialogue state (dialogue state storage means)
6 Talk queue (queue storage means)
10 Display 11 Dialog control rule (dialog control rule storage means)
12 Situation DB (situation DB storage means)
DESCRIPTION OF SYMBOLS 13 Statement type discrimination means 14 Dialogue action selection means 15 Situation condition discrimination means 16 Statement promotion condition discrimination means 17 Question control means 18 Statement generation means 19 Statement generation rule 21 Problem storage part (problem storage means)
22 Hint generator (Hint generator)
23 Correctness determination unit (correction determination means)
24 Similarity determination unit (similarity determination means)
25 Hint storage unit (Hint storage means)
31 Network Data Server 31a Database 32 Question Response System 33 Match Determination Unit 34 Search Unit 35 Similarity Calculation Unit 36 Hint Sentence Selection Unit 37 Response Sentence Generation Unit 38 Question Management Unit 39 Answer Extraction Unit 41 Communication Interface 42 Storage Unit 43 Correct Answer Generation Unit 44 hint expression acquisition means 45 hint sentence candidate generation means 46 hint sentence list generation means 421 correct answer candidate database 422 hint sentence candidate database 423 hint sentence list database 424 response sentence database 441 first expression acquisition means 442 second expression acquisition means 443 third Expression acquisition means 444 Fourth expression acquisition means 451 First candidate generation means 452 Second candidate generation means 453 Third candidate generation means 454 Fourth candidate generation means 461 Hint keyword extraction means 462 Feature amount extraction means 463 Score calculation means 464 Sort means 465 Storage means D Output device M Input device N Network

Claims (6)

In a multi-person dialogue between a plurality of human users and a plurality of agents realized by a computer, a multi-person thought-inducing interactive device that arouses the user's thoughts by the agent's remarks,
Situation for storing a time parameter that defines the timing at which the agent speaks and an emotion parameter that sets an emotion condition that defines the utterance of each agent as a sympathetic speech or a self-centered speech as the situation of the multi-person dialogue Database storage means;
Rules for selecting the first dialogue action information describing the internal structure of the utterances presented to the user from each agent for each type of utterance contents of participants in the multi-person dialogue according to the situation of the multi-person dialogue A dialog control rule storage means storing a dialog control rule describing
A utterance that stores a utterance generation rule that describes a rule for selecting an utterance expression presented to the user from the agent according to the type of utterance content of each agent, the emotion condition, and the situation of the multi-person conversation Generation rule storage means;
Queue storage means for storing the first dialog act information and second dialog act information describing an internal structure of the user's speech;
Dialog state storage means for storing the first dialogue action information related to the speech already presented to the user and the history of the second dialogue action information as a history of dialogue by the participants of the multi-person dialogue,
A user message that accepts text data or voice data indicating the message of each user, converts it into the second dialogue action information of the user, adds it to the tail of the queue storage means, and writes it to the dialogue state storage means Understanding means,
Referencing the first or second interactive action information at the head of the queue storage means and the history of the first and second interactive action information stored in the interactive state storage means, and applying the interactive control rule By means of the dialog control means for selecting the first dialog action information related to the speech to be presented next to the user from the agent and removing the first or second dialog action information at the head of the queue storage means;
When acquiring the first dialogue action information selected by the dialogue control means, referring to the history of the first and second dialogue action information stored in the dialogue state storage means, based on the statement generation rule, Corresponding to the acquired first dialogue action information, a speech expression of the agent that will speak next is generated, the generated speech expression is presented to the user, and the first conversation action information in which the speech expression is substituted is generated. Agent speech generation means for storing in the conversation state storage means and adding to the tail of the queue storage means;
A multi-person thought-inspiring interactive device characterized by comprising: In order to conduct a quiz question-and-answer format conversation as the multi-person dialogue, the plurality of agents, based on the selected first dialogue action information, give a question to the user, and the questions The first dialogue action information, the dialogue control rule, and the statement generation rule are predetermined so as to behave as an answer agent that answers a problem,
Problem storage means for storing the quiz problem;
Hint generating means for generating a hint sentence list in which a plurality of hints of the problem are arranged in descending order of difficulty;
Hint storage means for storing the generated hint sentence list;
Correctness determination means for determining correctness of the answer of the user or the answer agent;
Similarity determination means for determining the degree of similarity between the correct answer and the answer when the answer is incorrect;
Further comprising
The agent utterance generation means follows the emotion condition for the determination result when the correctness determination means determines that the participant's answer is incorrect and when it is determined to be correct. The multi-person thought-invoking interactive apparatus according to claim 1, wherein an utterance is generated as an utterance expression of the questioning agent or the answering agent. In a multi-party conversation between a plurality of users who are humans and a plurality of agents realized by a computer, comprising user speech understanding means, dialogue control means, agent speech generation means, and storage means. A multi-person thought-stimulating dialogue method by a multi-person thought-stimulating dialogue device that arouses the user's thoughts,
In the storage means,
Situation for storing a time parameter that defines the timing at which the agent speaks and an emotion parameter that sets an emotion condition that defines the utterance of each agent as a sympathetic speech or a self-centered speech as the situation of the multi-person dialogue A database,
Rules for selecting the first dialogue action information describing the internal structure of the utterances presented to the user from each agent for each type of utterance contents of participants in the multi-person dialogue according to the situation of the multi-person dialogue A dialogue control rule that describes
A speech generation rule describing rules for selecting a speech expression presented to the user from the agent according to the type of speech content of each agent, the emotion condition, and the situation of the multi-person dialogue Is remembered,
In the user speech understanding means,
Receiving text data or voice data indicating the remarks of each user;
Converting received speech into second interactive action information describing the internal structure of the user's speech;
Adding the second dialogue action information to the tail of the queue storage means;
With respect to the dialogue state storage means for storing the first dialogue act information related to the speech already presented to the user and the history of the second dialogue act information as the dialogue history by the participants of the multi-person dialogue, Writing the second dialogue action information of the accepted speech;
In the dialog control means,
Referencing the first or second interactive action information at the head of the queue storage means and the history of the first and second interactive action information stored in the interactive state storage means, and applying the interactive control rule Selecting first dialogue action information relating to a statement to be presented next to the user from the agent;
Sending the selected first dialogue action information to the agent statement generation means;
Removing the first or second interactive action information at the head of the queue storage means;
In the agent message generation means,
Obtaining first dialogue action information selected by the dialogue control means;
The agent that refers to the history of the first and second dialogue action information stored in the dialogue state storage means and makes a statement next in accordance with the acquired first dialogue action information based on the statement generation rule Generating a speech expression of
Presenting the generated speech expression to the user;
Storing the first dialogue action information into which the utterance expression is substituted in the dialogue state storage unit and adding the information to the tail of the queue storage unit;
A multi-person thought-aware dialogue method characterized by including and executing. In order to conduct a quiz question-and-answer format conversation as the multi-person dialogue, the plurality of agents, based on the selected first dialogue action information, give a question to the user, and the questions The first dialogue action information, the dialogue control rule, and the statement generation rule are predetermined so as to behave as an answer agent that answers a problem,
The questioning agent is
In a series of processes for asking the question of the quiz, a plurality of hints of the question are said to be presented to the participant in descending order of difficulty, and the answer of the user or the answer agent Determine correct / incorrect, if the answer is incorrect, determine the degree of similarity between the correct answer and the answer, make a statement according to the degree of similarity at the incorrect answer and say at the correct answer,
4. The answering agent according to claim 3, wherein the answering agent or the answering agent makes a statement according to the emotion condition for the result when the answer of the participant is incorrect or correct. The multi-person thought-aware dialogue method described.   A multi-person thought-stimulating interactive program for causing a computer to function as each means constituting the multi-person thought-stimulating interactive apparatus according to claim 1.   6. A computer-readable recording medium on which the multi-person thought-inducing dialogue program according to claim 5 is recorded.

Images