Dialog reproduction method, dialog reproduction device, and program

The system generates dialogue scenarios to change user opinions by structuring discussions around opposing positions, enhancing user confidence in their initial stance through automated dialogues.

JP2024165969APending Publication Date: 2024-11-28NIPPON TELEGRAPH & TELEPHONE CORP +2
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2023082602
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-18
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Existing dialogue systems cannot effectively persuade users to change their opinions during discussions and lack clear methods to structure dialogues for opinion change.

Method used

A method and device that generate a dialogue scenario where information indicating opposing positions on a topic is input, and a computer executes dialog generation and output control to converge discussions on a specific position, using pre-created graph data to structure dialogues.

Benefits of technology

The system supports users in changing their opinions by generating dialogues that increase the confidence in their initial stance through automated discussions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024165969000001_ABST
    Figure 2024165969000001_ABST
Patent Text Reader

Abstract

To enable provision of assistance for changing the thinking of someone who holds an opinion based on a particular point of view regarding a certain argument.SOLUTION: A dialog reproduction method causes a computer to perform: an argument dialog generation step of inputting information indicating one of two opposing points of view regarding a theme of a certain argument to generate a dialog scenario in such a way that the argument converges to the one point of view; and an argument dialog output control step of outputting a dialog based on the scenario.SELECTED DRAWING: Figure 2
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a dialogue playback method, a dialogue playback device, and a program. [Background technology]

[0002] In a dialogue system, humans obtain various information and fulfill requests by interacting with a computer. There are also dialogue systems that not only accomplish specific tasks but also conduct everyday conversations, and these systems allow humans to achieve mental stability, satisfy their desire for recognition, and build trusting relationships. Types of dialogue systems are described in Non-Patent Document 1.

[0003] Research is also being conducted into using computers to engage in discussions, rather than just task completion or everyday conversation. Discussions play an important role for humans, as they change people's value judgments and organize their thoughts.

[0004] For example, in Non-Patent Document 2, a discussion dialogue is realized using graph data in which opinions are nodes. The graph data is manually created based on a preset discussion theme (for example, "If you are going to live permanently, the city is better than the countryside"). By using the discussion data created manually, it becomes possible to discuss a specific topic. A discussion is carried out by mapping user utterances to nodes, and returning nodes that are connected to the mapped nodes as utterances from the system to the user. It is also possible to prepare two dialogue systems and have the two robots discuss by having them utter nodes according to the graph. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Tatsuya Kawahara, "Evolution and Selection of Speech Dialogue Systems - History and Recent Technological Trends", Journal of the Japanese Society for Artificial Intelligence, Vol. 28, No. 1, pp45-51, 2013 [Non-Patent Document 2] Ryuichiro Higashinaka et al., "Argumentative dialogue system based on argumentation structures", Proceedings of The 21st Workshop on the Semantics and Pragmatics of Dialogue, pp154-155, 2017 [Non-Patent Document 3] Hiroki Itahara et al., "The impact of consensus building based on dialogue between robots on human thinking," Proceedings of Information Processing Society of Japan Interaction 2022, pp. 225-230, 2022 Summary of the Invention [Problem to be solved by the invention]

[0006] By using the method of Non-Patent Document 2, it is possible to have the two robots generate any argument, but it is not clear how to construct a dialogue in order to persuade the user who is listening to the argument or to change the user's opinion. As a related study, Non-Patent Document 3 reports the results of an investigation into the impression that an argument between two robots gives to users (for example, whether the robots' argument was fair or accurate). However, although findings on how to change users' impressions of the robots' argument have been reported, findings on how to change the users' own opinions have not been the subject of the investigation, and are therefore not clear.

[0007] The present invention has been made in view of the above points, and has an object to enable support for changing the thinking of those who hold a particular position regarding a certain debate. [Means for solving the problem]

[0008] In order to solve the above problem, a computer executes a discussion dialogue generation procedure in which information indicating one of two opposing positions on a certain discussion topic is input and a dialogue scenario is generated so that the discussion will converge to one of the positions, and a discussion dialogue output control procedure in which a dialogue based on the scenario is output. Effect of the Invention

[0009] It is possible to help change the thinking of those who hold a particular position on a certain issue. [Brief description of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating an example of a hardware configuration of a dialogue playback device 10 according to an embodiment of the present invention. [Diagram 2] 1 is a diagram showing an example of a functional configuration of a dialogue playback device 10 according to an embodiment of the present invention. [Diagram 3] A figure showing an example of generating a discussion dialogue scenario using discussion knowledge. [Figure 4] FIG. 2 is a diagram showing an example of a discussion dialogue scenario. [Diagram 5] FIG. 13 is a diagram showing an example of an output form of a dialogue. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] As a premise, a topic of discussion with multiple opposing positions (for example, "for" and "against") is targeted. At this time, as a user who changes his / her opinion, a person who strongly holds a specific position (for example, a person who answers above a certain threshold when asked about the degree of certainty of a specific opinion in a numerical questionnaire that asks for an opinion on the subject) is targeted. When there is a specific position X on the topic of discussion, the discussion generated in this embodiment is made to listen to the user who strongly holds the position X. At this time, a robot that will engage in dialogue may be prepared in advance for each different position. As the content of the discussion, after starting a discussion from different positions, the conclusion of the discussion is made to be the position X (converge) by having the robot other than the position X agree with the robot of the position X through the discussion. By having the user listen to such a discussion, it is possible to increase the degree of certainty of the user's opinion X (significantly increase the score of the degree of certainty in the questionnaire).

[0012] The selection of target users can be realized by specifically asking about the degree of certainty of a particular opinion using a questionnaire. In addition, the generation of discussions on a particular topic can be realized by using the method described in Non-Patent Document 2.

[0013] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing an example of the hardware configuration of a dialogue playback device 10 in an embodiment of the present invention. The dialogue playback device 10 in Fig. 1 has a drive device 100, an auxiliary storage device 102, a memory device 103, a processor 104, an interface device 105, etc., which are all connected to each other via a bus B.

[0014] The program for realizing the processing in the interactive playback device 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed from the recording medium 101 via the drive device 100 into the auxiliary storage device 102. However, the program does not necessarily have to be installed from the recording medium 101, but may be downloaded from another computer via a network. The auxiliary storage device 102 stores the installed program as well as necessary files, data, etc.

[0015] The memory device 103 reads out and stores the program from the auxiliary storage device 102 when an instruction to start the program is received. The processor 104 is a CPU or a GPU (Graphics Processing Unit), or a CPU and a GPU, and executes functions related to the dialogue playback device 10 in accordance with the program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

[0016] 2 is a diagram showing an example of a functional configuration of the dialogue playback device 10 according to an embodiment of the present invention. The dialogue playback device 10 is a device capable of automatically and diversifyingly generating discussion dialogues that increase the confidence of a user's opinion that is strongly confident about a particular position on a particular discussion topic.

[0017] 2, the dialogue playback device 10 has a discussion dialogue generation unit 11 and a discussion dialogue output control unit 12. Each of these units is realized by a process executed by a processor 104 of one or more programs installed in the dialogue playback device 10. Each unit will be described below.

[0018] [Discussion dialogue generation part 11] The discussion dialogue generation unit 11 receives as input information indicating a particular position S (a position on which a user has a strong opinion) on a certain discussion theme T and a graph summarizing knowledge of discussions on theme T, and generates a discussion dialogue scenario in which the discussion on theme T converges to position S. A discussion dialogue scenario refers to data that chronologically indicates the content of each utterance in a dialogue held as a discussion on a certain theme. Position S is, for example, either an opinion "in favor" or an opinion "against" theme T. In this case, the information indicating position S input to the discussion dialogue generation unit 11 may be something like a label indicating "in favor" or "against".

[0019] A graph summarizing the knowledge of discussions on theme T is constructed (created) manually in advance. The graph may be constructed, for example, using the method disclosed in Reference 1. A method for generating a discussion dialogue scenario based on a graph summarizing the knowledge of discussions is described in Reference 2, and the same method is used in this embodiment. The literature information for each reference will be described later.

[0020] Fig. 3 is a diagram showing an example of generating a discussion dialogue scenario using discussion knowledge. The left side of Fig. 3 shows a template of a discussion dialogue scenario prepared manually. The right side of Fig. 3 shows a graph (hereinafter simply referred to as a "graph") summarizing discussion knowledge on a certain topic (in the example of Fig. 3, "Are you for or against autonomous driving?").

[0021] The template is data showing a template for each utterance according to the progress of the discussion (showing the template for each utterance in chronological order), and data in which the parts of the content of each utterance that depend on the discussion knowledge on a specific topic are in a hole state (blank). In the example of FIG. 3, the part that was already in a hole state is shown in a filled state, but the part surrounded by the dashed line in FIG. 3 indicates that it was in a hole state at the beginning. Therefore, the template can be used universally for various topics. Basically, the content of the template is configured so that the discussion progresses with utterances from the pro-side and the anti-side on an unknown topic (to the template) and converges to one of the opinions (so that the other agrees with one of the opinions). Since it is unknown at the time of creating the template whether the position to be input is pro-side or anti-side, two templates are created in advance: a template that converges to the pro-side position (hereinafter referred to as the "affirmative side template") and a template that converges to the anti-side position (hereinafter referred to as the "negative side template"). The template in Figure 3 shows an example of an affirmative side template that is configured so that the discussion converges on A's opinion in a situation where A is in favor and B is in the opposing position.

[0022] The template in FIG. 3 includes the items "Part," "U," and "Utterance" for each utterance in the discussion. "Part" indicates the stage of the discussion. "U" indicates the serial number of each utterance (hereinafter referred to as "utterance number"). "Utterance" is a template for the content of the utterance. FIG. 3 shows that a specific utterance content (a character string indicating it) has already been applied to "Utterance" based on the graph, but "Utterance" is an item that includes a part that is in a hole state (hereinafter referred to as "hole part") as described above. Specifically, "Utterance" is set with a symbol ("A" or "B") indicating whether it is the position of A or B, and information indicating which node in the graph is a candidate for application (hereinafter referred to as "application information"). Note that, while FIG. 3 shows an example in which all the words in the utterance content (the utterance content) are fitted from the graph, for example, only a portion of the words in the utterance content (for example, a part corresponding to a word) may be the holed portion.

[0023] On the other hand, the graph on the right side of FIG. 3 is a graph based on argument knowledge. The root node of the graph is a node in which a character string indicating a specific theme ("for or against autonomous driving") is set. The root node has two child nodes (hereinafter referred to as "position nodes") corresponding to the for and against sides of the theme. The position node connected to the root node by the edge of "Support" is a node in which a wording corresponding to the position of the for side of the theme is set, and the position node connected to the root node by the edge of "Attack" is a node in which a wording corresponding to the position of the opposite side of the theme is set. The nodes of each hierarchy below the position node have multiple child nodes (hereinafter referred to as "support nodes") connected by the edge of "Support" and multiple child nodes (hereinafter referred to as "rebuttal nodes") connected by the edge of "Attack". Each support node is a node in which a wording indicating support for the position or opinion of the parent node is set. Each rebuttal node is a node in which a wording indicating a rebuttal to the position or opinion of the parent node is set. It is preferable that the content of the template is already known to the creator of the graph. In this case, the creator of the graph can decide the wording to be set for each node by being aware of which part of the template (i.e., discussion dialogue scenario) each node of the graph is a candidate for fitting into, thereby reducing the possibility that the final discussion dialogue scenario will be unnatural.

[0024] In the blank portion of the template, a node to be filled in the blank portion is specified with the content of "support node for <certain node>" or "counterargument node for <certain node>". Here, <certain node> indicates identification information of a specific node. For example, for a blank portion in which "support node for <certain position node>" is set as the fitting information, multiple supporting nodes among all child nodes of the position node become candidates for filling. The discussion dialogue generation unit 11 selects one supporting node from the multiple supporting nodes randomly or according to a preset rule, and fills the blank portion with the text set in the selected supporting node. On the other hand, for a blank portion in which "counterargument node for <certain position node>" is set as the fitting information, multiple counterargument nodes among all child nodes of the position node become candidates for filling. The discussion dialogue generation unit 11 selects one counterargument node from the multiple counterargument nodes randomly or according to a preset rule, and fills the blank portion with the text set in the selected counterargument node. The discussion dialogue generation unit 11 executes such processing for all the perforated parts of the template, thereby generating a discussion dialogue scenario.

[0025] The discussion dialogue generation unit 11 uses the affirmative side template when the input position S is "for", and uses the negative side template when the position S is "against". In this way, the discussion dialogue generation unit 11 can generate a discussion dialogue scenario that converges to the position S. Which of the two templates is the affirmative side template and which is the negative side template may be identifiable, for example, by the file names of the files that store the templates.

[0026] In addition, from the viewpoint of the template, "Agree" or "Again" is merely a relative relationship. For example, in the template of FIG. 3, "A" is merely one position, and "B" is merely the other position. Therefore, only either the affirmative side template or the negative side template may be prepared. In this case, the discussion dialogue generation unit 11 may generate a discussion dialogue scenario by changing the way of applying the discussion knowledge to the prepared template depending on whether the position corresponding to the template matches the position S. For example, when only the affirmative side template is created, if the position S is "Agree", the discussion dialogue generation unit 11 may apply the wording set in the supporting node of the graph to the holed portion in which the supporting node is specified as a candidate for application in the template application information. When the position S is "Again", the discussion dialogue generation unit 11 may apply the wording set in the refutation node of the graph (i.e., the node corresponding to the opposite position to the position specified for application) to the holed portion.

[0027] FIG. 4 is a diagram showing an example of a discussion dialogue scenario. In FIG. 4, the numbers in brackets ([]) indicate the utterance number. "A" and "B" indicate the speaker and the position of the discussion (A is in favor, B is against). The character string following ":" after "A" or "B" indicates the content of the utterance. An '*' at the beginning of the utterance content indicates that the utterance content is a fixed (template) character string, and the rest of the utterance content indicates an utterance filled in from discussion knowledge. A dialogue based on a discussion dialogue scenario starts in a situation where each party has a different position, and after each party has stated their opinion, it ends with the opposing party B agreeing with the supporting party A.

[0028] [Discussion dialogue output control section 12] The discussion dialogue output control unit 12 presents the state of the discussion to the user by controlling the output device 20 to output (play) the dialogue based on the discussion dialogue scenario generated by the discussion dialogue generation unit 11. The output device 20 may be prepared for each position in the theme T, for example. In this case, the discussion dialogue output control unit 12 may cause each output device 20 corresponding to each position in the theme T to output utterances from the position to which the output device 20 corresponds in the discussion on the theme T. Any output device 20 can be used, and for example, a display device that displays messenger-type images such as LINE (registered trademark), videos of things that have substance in virtual space such as CG agents, or things that have substance in real space such as robots may be used as the output device 20. Alternatively, a simple speaker may be used as the output device 20. In this case, the discussion dialogue output control unit 12 may, for example, generate the voice of the utterance in the discussion using a voice synthesis technology and output the generated voice to the speaker.

[0029] Fig. 5 is a diagram showing an example of a dialogue output form. Fig. 5 shows a case where a liquid crystal screen and two social dialogue robots, CommU (registered trademark), which is one of the representative robots, are used as the output device 20. The two CommU (registered trademark) operate as speakers with different positions, and the liquid crystal screen displays the contents of the utterances by each speaker. For example, the CommU (registered trademark) on the left corresponds to the pro-side, and the one on the right corresponds to the anti-side.

[0030] The discussion dialogue output control unit 12 reads out the utterances in the discussion dialogue scenario generated by the discussion dialogue generation unit 11 to CommU (registered trademark) corresponding to the position of the utterance, and outputs a character string indicating the utterance on the liquid crystal screen. The discussion dialogue output control unit 12 moves the body of CommU (registered trademark) according to a predetermined rule (for example, action A for a question utterance, action B for a filler utterance, and action C for other utterances), and further updates the dialogue on the liquid crystal screen, making it possible to clearly present the content of the discussion dialogue to the user.

[0031] As described above, according to this embodiment, it is possible to automatically generate and play back a dialogue that converges on the same position (opinion) as a user who has a strong opinion on a particular position among multiple opposing positions on a certain topic. As a result, it is possible to strengthen the certainty of the user's own opinion. Therefore, it is possible to support the transformation of the thinking of a person who holds a particular position on a certain discussion.

[0032] The literature information for the above cited references 1 and 2 is as follows:

[0033] [Reference 1] Kazuki Sakai et al., "Creating large-scale argumentation structures for dialogue systems", In Proc. of LREC, pp3975-3980, 2018 [Reference 2] Kazuki Sakai et al., "Hierarchical argumentation structure for persuasive argumentative dialogue generation", IEICE Transactions on Information and Systems, E103.D(2): 424-434, 2020 Although the embodiment of the present invention has been described in detail above, the present invention is not limited to such specific embodiment, and various modifications and variations are possible within the scope of the gist of the present invention described in the claims. [Explanation of symbols]

[0034] 10 Dialogue playback device 11 Discussion dialogue generation section 12 Discussion dialogue output control section 20 Output Devices 100 Drive device 101 Recording media 102 Auxiliary storage device 103 Memory device 104 processors 105 Interface device B Bus

Claims

1. a discussion dialogue generation procedure for inputting information indicating one of two opposing positions on a certain discussion topic and generating a dialogue scenario so that the discussion will converge to one of the positions; A discussion dialogue output control procedure for outputting a dialogue based on the scenario; The method for reproducing dialogue is characterized in that the method is executed by a computer.

2. the discussion dialogue generation step generates the scenario by applying discussion knowledge of the topic to a template corresponding to one of the positions among dialogue templates prepared for each position in the discussion; 2. The method of claim 1, wherein the dialogue is reproduced by a plurality of voices.

3. the discussion dialogue generation step generates the scenario by changing how the discussion knowledge of the theme is applied to a dialogue template prepared for a certain position in the discussion depending on whether the position matches one of the positions.

2. The method of claim 1, wherein the dialogue is reproduced by a plurality of voices.

4. A discussion dialogue generation unit configured to input information indicating one of two opposing positions on a certain discussion topic and generate a dialogue scenario so that the discussion will converge to the one position; A discussion dialogue output control unit configured to output a dialogue based on the scenario; 1. A dialogue playback device comprising:

5. The discussion dialogue generation unit is configured to generate the scenario by applying discussion knowledge of the theme to a template corresponding to one of the positions among dialogue templates prepared for each position in the discussion.

5. The dialogue playback device according to claim 4.

6. The discussion dialogue generation unit is configured to generate the scenario by changing a way of applying discussion knowledge of the theme to a dialogue template prepared for a certain position in a discussion depending on whether the position coincides with one of the positions.

5. The dialogue playback device according to claim 4.

7. 4. A program for causing a computer to execute the dialogue reproduction method according to claim 1.

Citation Information

Patent Citations

  • Dialog system, method, and program

    JP2018147189A

  • Argumentation structure extension device, argumentation structure extension method, program and data structure

    JP2019144692A

  • Dialogue control apparatus, dialogue control method, and program

    JP2020038397A

  • Speech sentence generation model learning device, speech sentence collection device, speech sentence generation model learning method, speech sentence collection method, and program

    JP2020106905A

  • Automatic conversation device, automatic conversation method, and program

    JP2020190585A