Conversation control method, device, and program

The conversation control method addresses the challenge of maintaining dialogue flow by switching between scenarios based on user intentions, ensuring continuous and relevant responses.

JP7581651B2Active Publication Date: 2024-11-13FUJITSU LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2020075466
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-04-21
Publication Date
2024-11-13
Estimated Expiration
2040-04-21

AI Technical Summary

Technical Problem

Conventional dialogue systems struggle to maintain a natural flow in conversations, often ending abruptly or providing irrelevant responses when unexpected user utterances occur, leading to user boredom.

Method used

A conversation control method that switches between multiple predefined conversation scenarios based on user intentions, allowing seamless transitions and natural continuation of dialogue by selecting appropriate responses from a scenario database.

Benefits of technology

Enables continuous and natural flow in conversations by adapting to user inputs, ensuring relevant responses even when scenarios change or topics shift.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007581651000001
    Figure 0007581651000001
  • Figure 0007581651000002
    Figure 0007581651000002
  • Figure 0007581651000003
    Figure 0007581651000003
Patent Text Reader

Abstract

To output a reply which enables continuance of a conversation with a user with a natural flow.SOLUTION: In a scenario DB, a conversation scenario indicating each of a plurality of topics corresponding to conversion scenes is defined, and each conversion scenario is defined by connecting unit conversations which are questions and answers based on a set of an intension label and an answer with a sequence given to a conversation flow. When an intension label estimated from a user speech during execution of conversation control based on a current conversation scenario corresponds to an intension label of the unit conversation of one of the current conversation scenarios, the unit conversation is determined as a next conversation. When the estimated intension label does not correspond to the intension label of any of the unit conversations included in the current conversation scenarios and there is an intension label corresponding to another conversation scenario included in the same conversion scene, conversion control is changed from the current conversation scenario to another conversion scenario.SELECTED DRAWING: Figure 8
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The disclosed technology relates to a conversation control method, a conversation control device, and a conversation control program. [Background technology]

[0002] Conventionally, dialogue systems such as task-oriented dialogue systems and question-and-answer chatbots have been used in FAQ (Frequently Asked Questions) systems and web page searches. In recent years, dialogue systems have become known that learn the dialogue relationship between a user's utterance and a response utterance to the user's utterance, and select and utter a response content from the user's utterance using a learned model after learning.

[0003] For example, a data processing device has been proposed that appropriately assists a user by enriching the content of a conversation with the user, etc. This data processing device determines an action for the user by referring to state management data that manages the state of the user. Also, this data processing device updates the user's state management data based on user state data that indicates the state of the user, which is determined by collecting and analyzing activity history data of the user acquired by the user's mobile terminal. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2018-128747 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, when a dialogue system is intended to have a continuous conversation rather than a one-question-one-answer format, it is necessary to prepare a scenario that anticipates user utterances. In this case, if a user utters an utterance that is unexpected in response to the scenario, the system cannot respond and the conversation ends. Furthermore, if there is no scenario, the system may return a response that is out of context and ignores the flow of the conversation before and after. In this way, if a dialogue system does not continue a conversation or if the system returns a response that is out of context, the user is likely to become "bored" with the dialogue system.

[0006] In one aspect, the disclosed technology aims to output a response that allows a conversation with a user to continue in a natural flow. [Means for solving the problem]

[0007] In one aspect, the disclosed technology is a conversation control method for executing conversation control based on one of a plurality of conversation scenarios. When an utterance content received during execution of conversation control based on a first conversation scenario corresponds to one of the conversation contents included in the first conversation scenario, the disclosed technology outputs a response corresponding to the one of the conversation contents. When the utterance content does not correspond to any of the conversation contents included in the first conversation scenario and corresponds to any of the conversation contents included in a second conversation scenario different from the first conversation scenario, the disclosed technology performs the following process. The disclosed technology outputs a response corresponding to any of the conversation contents included in the second conversation scenario, and transitions the conversation control based on the first conversation scenario to the conversation control based on the second conversation scenario. Effect of the Invention

[0008] One aspect is that it has the effect of being able to output a response that allows the conversation with the user to continue in a natural flow. [Brief description of the drawings]

[0009] [Figure 1] FIG. 2 is a functional block diagram of a conversation control device. [Diagram 2] FIG. 13 is a diagram illustrating an example of learning data. [Diagram 3] FIG. 2 is a diagram conceptually illustrating a data structure of a scenario DB. [Figure 4] FIG. 11 is a diagram showing another example of a conversation scenario. [Diagram 5] FIG. 1 is a block diagram showing a schematic configuration of a computer that functions as a conversation control device. [Figure 6] 13 is a flowchart illustrating an example of a conversation control process. [Figure 7] FIG. 13 is a diagram for explaining a case where a conversation continues on the same topic. [Figure 8] FIG. 13 is a diagram for explaining a case where a conversation continues by changing the topic in the same conversation scene. [Figure 9] FIG. 13 is a diagram for explaining a case where a conversation scene is changed, a new topic is started, and a conversation is continued. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, an example of an embodiment according to the disclosed technology will be described with reference to the drawings. In this embodiment, an example in which the disclosed technology is applied to a conversation control device that conducts a voice conversation with a user will be described.

[0011] As shown in Fig. 1, the conversation control device 10 receives voice data indicating a user utterance via a microphone (not shown). The conversation control device 10 also outputs voice data indicating a response (system utterance) corresponding to the user utterance via a speaker (not shown). Functionally, the conversation control device 10 includes an estimation unit 12, a determination unit 14, and an output unit 16. Furthermore, an intention estimation model 22 and a scenario DB (Database) 24 are stored in a predetermined storage area of ​​the conversation control device 10.

[0012] The estimation unit 12 estimates the intention of the user utterance indicated by the received voice data. Specifically, the estimation unit 12 inputs the user utterance, which is obtained by converting the voice data into text data through voice recognition processing, to a pre-trained intention estimation model 22. The estimation unit 12 receives an intention label estimated and output by the intention estimation model 22 in accordance with the input user utterance, and passes it to the determination unit 14.

[0013] Here, the intention estimation model 22 will be described. The intention estimation model 22 is a model such as a neural network that is machine-learned in advance using learning data that is a pair of an intention label and a user utterance. FIG. 2 shows an example of the learning data. Each row in the list shown in FIG. 2 corresponds to one learning data. As shown in FIG. 2, the same intention label is associated with user utterances having a common intention among various expressions of user utterances. In the example of FIG. 2, the intention label "favorite food" is associated with a user utterance asking about a favorite food. Similarly, the intention label "ask again when there is a matter" is associated with a user utterance asking again about a matter. In addition, an expression that unifies the variation in the expression of the user utterance may be used as the intention label, such as the intention label "What can I make for breakfast?" shown in FIG. 2.

[0014] The intent estimation model 22 is generated by learning model parameters using a large amount of learning data as described above. The intent estimation model 22 outputs an intent label that is most highly correlated with an input user utterance among the learned intent labels. Note that the intent label is an example of the utterance content of the disclosed technology.

[0015] The determination unit 14 determines a conversation corresponding to the intention label estimated by the estimation unit 12 from conversations included in a plurality of conversation scenarios stored in the scenario DB 24.

[0016] Fig. 3 conceptually illustrates the data structure of the scenario DB 24. As illustrated in Fig. 3, the scenario DB 24 defines, for example, a conversation scenario indicating each of a plurality of topics corresponding to a plurality of conversation scenes assuming daily life situations. In the example of Fig. 3, conversation scenes assuming each of the following situations are defined: "breakfast" as conversation scene 1, "returning home" as conversation scene 2, and "before going to bed" as conversation scene 3. Also, in the example of Fig. 3, conversation scene 1 (breakfast) includes conversation scenarios indicating each of the topics "breakfast" as topic 1, "weather" as topic 2, "news" as topic 3, and "school" as topic 4.

[0017] Each conversation scenario is defined by connecting a plurality of unit conversations with a certain order in the flow of the story. A unit conversation is a question-and-answer conversation in which an intention label and an answer are paired. In the example of FIG. 3, the j-th unit conversation of a conversation scenario showing a topic i is written as "conversation ij". In the following, the unit conversation of conversation ij is also simply called "conversation ij". The order of the flow of the story means that the unit conversations are connected in an order so that the intention label that is assumed to be a user's utterance in response to a certain unit conversation is included in the next unit conversation. Note that, in the example of FIG. 3, for the sake of simplicity, an example of a conversation scenario in which unit conversations are arranged in a line is shown, but as shown in FIG. 4, the unit conversation may be branched midway. In the example of FIG. 4, the unit conversation following conversation 1-2 is branched into conversations 1-3-1 and 1-3-2.

[0018] The answer included in the unit conversation corresponds to the paired intention label, and the content according to the topic indicated by the conversation scenario including the unit conversation and the conversation scene including the conversation scenario is predefined. The predefined answer may be the system utterance itself output from the conversation control device 10, or may be a template for creating an answer by embedding a predetermined word or word string. The predetermined word or word string may be, for example, a word or word string according to a time period or a user attribute, a word or word string extracted from a user utterance, or a word or word string searched from a search site or the like according to the corresponding intention label.

[0019] As described above, since the content of the answer is defined according to the topic and the conversation scene, different answers may be associated with the same intent label when the topic or conversation scene is different. For example, the answer "Hato Sable" can be associated with the intent label "favorite food" in one topic, and the answer "Will you treat me?" can be associated with the same intent label "favorite food" in another topic.

[0020] Also, for example, suppose that a conversation scenario showing the topic of weather in a breakfast conversation scene includes an intent label of "How's the weather?". A template can be defined in which a word string indicating today's weather retrieved from a weather forecast website or the like is embedded as a response to this intent label. Also, suppose that a conversation scenario showing the topic of weather in a bedtime conversation scene includes the same intent label of "How's the weather?". Unlike the case of the breakfast conversation scene described above, a template can be defined in which a word string indicating tomorrow's weather retrieved from a weather forecast website or the like is embedded as a response to this intent label.

[0021] Specifically, the determination unit 14 determines the conversation scenario at the start of the conversation according to a predetermined rule. For example, the determination unit 14 can determine a conversation scenario indicating an ith (e.g., the first) topic included in a conversation scene according to a time period as the conversation scenario at the start of the conversation. For example, the determination unit 14 may determine the conversation scenario at the start of the conversation based on a user utterance. In this case, the determination unit 14 may receive a user utterance after performing a system utterance via the output unit 16 described later, such as "Is there anything I can do for you?", which prompts the user to make an utterance for determining the conversation scenario at the start of the conversation. The determination unit 14 may determine the conversation scenario in which the first unit conversation includes an intention label estimated from the user utterance, as the conversation scenario at the start of the conversation.

[0022] Furthermore, when an intention label transferred from the estimation unit 12 during execution of conversation control based on the current conversation scenario corresponds to an intention label of any unit conversation included in the current conversation scenario, the determination unit 14 determines the unit conversation as the next conversation. Hereinafter, the intention label transferred from the estimation unit 12 is referred to as an "estimated intention label." Furthermore, the case where the intention labels correspond can be the case where the estimated intention label and the intention label in the conversation scenario match or are similar. The case where the intention labels are similar can be, for example, the case where similar intention labels are grouped in advance, and intention labels belonging to the same group are set as similar intention labels. When the current conversation scenario includes multiple unit conversations including the corresponding intention label, the determination unit 14 can select, from among the unit conversations following the current unit conversation, a unit conversation whose order of appearance in the conversation scenario is closer to the current unit conversation.

[0023] Furthermore, when the estimated intention label does not correspond to the intention label of any unit conversation included in the current conversation scenario, the determination unit 14 searches for another conversation scenario different from the current conversation scenario that is included in the same conversation scene. Specifically, when the estimated intention label corresponds to the intention label of any unit conversation included in another conversation scenario, the determination unit 14 determines that unit conversation as the next conversation. When there are multiple other conversation scenarios including the corresponding intention label, the determination unit 14 selects, for example, a conversation scenario in which the corresponding intention label appears earlier in the conversation scenario. As a result, the determination unit 14 transitions conversation control based on the current conversation scenario to conversation control based on the other conversation scenario.

[0024] Furthermore, when the estimated intention label is not included in any of the conversation scenarios included in the current conversation scene, the determination unit 14 searches for other conversation scenes different from the current conversation scene. Specifically, when the estimated intention label corresponds to the intention label of any unit conversation in any of the other conversation scenarios included in the other conversation scenes, the determination unit 14 determines the unit conversation as the next conversation. When there are multiple other conversation scenes including the corresponding intention label, the determination unit 14 selects, for example, a conversation scene including a conversation scenario in which the corresponding intention label appears earlier in the conversation scenario. As a result, the determination unit 14 transitions conversation control based on the current conversation scene to conversation control based on the other conversation scene.

[0025] The current conversation scenario is an example of a first conversation scenario of the disclosed technology, and the other conversation scenario is an example of a second conversation scenario of the disclosed technology. The current conversation scene is an example of a first conversation scene of the disclosed technology, and the other conversation scene is an example of a second conversation scene of the disclosed technology.

[0026] Furthermore, when none of the conversation scenarios in any of the conversation scenes includes an intention label corresponding to the estimated intention label, the determination unit 14 determines a standard conversation as the next conversation. The standard conversation may be, for example, a predetermined response such as a greeting, a backchannel, or a repeat question associated with an intention label.

[0027] The determination unit 14 transfers information about the unit conversation determined as the next conversation to the output unit 16. The information about the determined unit conversation is identification information of the conversation scene and conversation scenario including the unit conversation, and information indicating the ordinal number of the unit conversation within the conversation scenario.

[0028] The output unit 16 refers to the scenario DB 24 based on the information of the unit conversation passed from the determination unit 14, and generates a system utterance corresponding to the answer included in the corresponding unit conversation. The output unit 16 converts the generated system utterance into voice data and outputs it.

[0029] The conversation control device 10 can be realized, for example, by a computer 40 shown in Fig. 5. The computer 40 includes a CPU (Central Processing Unit) 41, a memory 42 as a temporary storage area, and a non-volatile storage unit 43. The computer 40 also includes an input / output device 44 such as an input unit and a display unit, and an R / W (Read / Write) unit 45 that controls reading and writing of data from and to a storage medium 49. The computer 40 also includes a communication I / F (Interface) 46 that is connected to a network such as the Internet. The CPU 41, the memory 42, the storage unit 43, the input / output device 44, the R / W unit 45, and the communication I / F 46 are connected to one another via a bus 47.

[0030] The storage unit 43 can be realized by a hard disk drive (HDD), a solid state drive (SSD), a flash memory, etc. The storage unit 43 as a storage medium stores a conversation control program 50 for causing the computer 40 to function as the conversation control device 10. The conversation control program 50 has an estimation process 52, a decision process 54, and an output process 56. The storage unit 43 also has an information storage area 60 in which information constituting each of the intention estimation model 22 and the scenario DB 24 is stored.

[0031] The CPU 41 reads out the conversation control program 50 from the storage unit 43, loads it in the memory 42, and sequentially executes the processes of the conversation control program 50. The CPU 41 operates as the estimation unit 12 shown in FIG. 1 by executing the estimation process 52. The CPU 41 operates as the determination unit 14 shown in FIG. 1 by executing the determination process 54. The CPU 41 operates as the output unit 16 shown in FIG. 1 by executing the output process 56. The CPU 41 reads out information from the information storage area 60, and loads the intention estimation model 22 and the scenario DB 24 in the memory 42. As a result, the computer 40 that has executed the conversation control program 50 functions as the conversation control device 10. The CPU 41 that executes the program is hardware.

[0032] The functions realized by the conversation control program 50 can also be realized by, for example, a semiconductor integrated circuit, more specifically, an ASIC (Application Specific Integrated Circuit) or the like.

[0033] Next, the operation of the conversation control device 10 according to the present embodiment will be described. When the start of conversation control is instructed in the conversation control device 10, the conversation control device 10 executes a conversation control process shown in Fig. 6. Note that the conversation control process is an example of a conversation control method of the disclosed technology.

[0034] In step S12, the determination unit 14 determines a conversation scenario at the start of a conversation according to a predetermined rule. The determination unit 14 determines, for example, a system utterance for encouraging a user to speak according to an intention label included in the first unit conversation of the determined conversation scenario, and passes the system utterance to the output unit 16. The output unit 16 converts the passed system utterance into voice data and outputs it.

[0035] Next, in step S14, the estimation unit 12 determines whether or not a user utterance has been received. If a user utterance has been received, the process proceeds to step S16, and if not, the process proceeds to step S30.

[0036] In step S16, the estimation unit 12 performs speech recognition processing on the voice data indicating the received user utterance, converts it into text data, and inputs it to the pre-trained intention estimation model 22. Then, the estimation unit 12 receives the intention label estimated and output by the intention estimation model 22 in accordance with the input user utterance, and passes it to the determination unit 14.

[0037] Next, in step S18, the determination unit 14 determines whether or not an intention label corresponding to the estimated intention label exists in the current conversation scenario. If a corresponding intention label exists, the process proceeds to step S24. If not, the process proceeds to step S20.

[0038] In step S20, the determination unit 14 determines whether or not an intention label corresponding to the estimated intention label exists in another conversation scenario included in the same conversation scene including the current conversation scenario. If a corresponding intention label exists, the process proceeds to step S24. If not, the process proceeds to step S22.

[0039] In step S22, the determination unit 14 determines whether or not an intention label corresponding to the estimated intention label exists in any of the conversation scenarios included in another conversation scene. If a corresponding intention label exists, the process proceeds to step S24. If not, the process proceeds to step S26.

[0040] In step S24, the determination unit 14 determines a unit conversation that includes an intention label corresponding to the estimated intention label as the next conversation. That is, when a positive determination is made in the above step S18 and the process proceeds to this step, the determination unit 14 determines a unit conversation that will be the next conversation from within the current conversation scenario. When there are multiple corresponding intention labels, the determination unit 14 determines, as the next conversation, a unit conversation that is closer in order of appearance in the conversation scenario to the current unit conversation among the unit conversations that follow the current unit conversation.

[0041] Furthermore, if a positive determination is made in step S20 and the process proceeds to this step, the determination unit 14 determines a unit conversation that will become the next conversation from another conversation scenario in the current conversation scene. If there are multiple other conversation scenarios, the determination unit 14 determines a unit conversation that will become the next conversation from the conversation scenario in which the corresponding intention label appears earliest in the conversation scenario.

[0042] Also, if the above step S22 is judged to be positive and the process proceeds to this step, the determination unit 14 determines a unit conversation that will become the next conversation from within any of the conversation scenarios of the other conversation scenes. If there are multiple other conversation scenes, the determination unit 14 determines a unit conversation that will become the next conversation from a conversation scene including a conversation scenario that has the earliest appearance order of the corresponding intention label in the conversation scenario. Once the next conversation is determined, the process proceeds to step S28.

[0043] On the other hand, in step S26, the standard conversation corresponding to the estimated intention label is determined as the next conversation, and the process proceeds to step S28.

[0044] In step S28, the determination unit 14 passes information on the unit conversation determined as the next conversation to the output unit 16. Then, the output unit 16 refers to the scenario DB 24 based on the information on the unit conversation passed from the determination unit 14, and generates a system utterance corresponding to the answer included in the corresponding unit conversation. The output unit 16 converts the generated system utterance into voice data and outputs it.

[0045] Next, in step S30, the estimation unit 12 determines whether or not to end the conversation control. For example, it can determine that the conversation control is to be ended when the user instructs the end of the conversation, or when the next user utterance is not accepted even after a predetermined time has passed since the previous answer was output. If the conversation control is not to be ended, the process returns to step S14, and if the conversation control is to be ended, the conversation control process is ended.

[0046] For example, assume that a conversation scenario showing topic 1 in conversation scene 1 shown in FIG. 7 is determined as a conversation scenario at the start of a conversation. In this case, when an intention label corresponding to the intention label included in conversation 1-1 in conversation scene 1 is estimated from a user utterance, a system utterance based on a reply included in conversation 1-1 is output. When an intention label estimated from a user utterance in response to this corresponds to an intention label included in the following conversation 1-2, the conversation transitions from conversation 1-1 to conversation 1-2 according to the conversation scenario showing topic 1. Similarly, when the estimated intention label corresponds to the intention label of conversation 1-3 and then the intention label of conversation 1-4, a conversation according to the conversation scenario is conducted as shown by the row of unit conversations shaded in FIG. 7. As a result, in a state where a topic is specified, that is, a conversation that is not a question and answer is continued about the same topic, and a natural conversation can be continued.

[0047] Also, as shown in FIG. 8, it is assumed that an intention label corresponding to the intention label estimated from the user utterance in response to the output of the system utterance based on the answer included in conversation 1-2 does not exist in the same conversation scenario. In this case, the conversation control transitions to a conversation scenario having an intention label corresponding to the estimated intention label among other conversation scenarios of the same conversation scene. The example of FIG. 8 shows that the conversation transitions from conversation 1-2 in the conversation scenario showing topic 1 to conversation 2-3 in the conversation scenario showing topic 2 as the next conversation, and the conversation continues from there according to the conversation scenario showing topic 2. Also, after conversation 2-4, the conversation transitions to conversation 4-2 in the conversation scenario showing topic 4, and the conversation continues according to that conversation scenario, and after conversation 4-4, the conversation transitions to conversation 3-5 in the conversation scenario showing topic 3.

[0048] In this way, if there is no intention label corresponding to the estimated intention label in the continuing unit conversation in the same conversation scenario, a transition is made to another conversation scenario of the same conversation scene. This allows the conversation to continue while changing the topic with the conversation scene specified. That is, since the topic may change even in a normal conversation between people, by transitioning to a conversation scenario indicating a different topic under the constraint of the same conversation scene, the conversation can be continued without any awkwardness.

[0049] Also, as shown in FIG. 9, assume that the conversation transitions from conversation 1-1, conversation 1-2, conversation 2-3, and conversation 2-4, and an intention label corresponding to an intention label estimated from a user utterance to a system utterance based on a response included in conversation 2-4 does not exist in the same conversation scene. In this case, conversation control transitions to another conversation scene. The example of FIG. 9 shows that after conversation 2-4 in conversation scene 1, the conversation transitions to conversation 1-1 in conversation scene 2. This makes it possible to avoid a sudden end of the conversation, and to start a conversation in a new conversation scene to continue the conversation.

[0050] As described above, when an intention label estimated during execution of conversation control based on the current conversation scenario corresponds to any of the intention labels included in the current conversation scenario, the conversation control device according to this embodiment outputs an answer included in a unit conversation including the intention label. On the other hand, when the estimated intention label does not correspond to any of the intention labels included in the current conversation scenario, the conversation control device searches for an intention label corresponding to the estimated intention label from other conversation scenarios included in the same conversation scene. When an intention label corresponding to the estimated intention label is included in the other conversation scenario, the conversation control device transitions from conversation control based on the current conversation scenario to conversation control based on the other conversation scenario including the searched intention label. This makes it possible to output an answer that allows the conversation with the user to continue in a natural flow.

[0051] In the above embodiment, the example of the conversation by voice has been described, but the present invention is not limited to this. For example, the content of the user's conversation may be received as text data, and the conversation may be controlled based on the same scenario DB as above to output a response as text data.

[0052] Furthermore, the conversation scenes, topics, responses, and the like given in the above embodiment are merely examples, and the contents of each may be defined according to the application scene of the conversation control device.

[0053] In the above embodiment, when searching for an intention label corresponding to an estimated intention label from a conversation scenario, the intention label paired with a previously output answer in a series of conversation controls may be excluded from the search target, thereby making it possible to avoid unnatural repetition or backtracking of a conversation.

[0054] In the above embodiment, the conversation control program is stored (installed) in the storage unit in advance, but the present invention is not limited to this. The program according to the disclosed technology can also be provided in a form stored in a storage medium such as a CD-ROM, a DVD-ROM, or a USB memory.

[0055] The following supplementary notes are further disclosed regarding the above embodiment.

[0056] (Appendix 1) A conversation control method for executing conversation control based on one of a plurality of conversation scenarios, comprising: When a conversation content received during execution of a conversation control based on a first conversation scenario corresponds to any of the conversation contents included in the first conversation scenario, outputting a response corresponding to any of the conversation contents; When the utterance content does not correspond to any of the conversation contents included in the first conversation scenario and corresponds to any of the conversation contents included in a second conversation scenario different from the first conversation scenario, a response corresponding to any of the conversation contents included in the second conversation scenario is output, and conversation control based on the first conversation scenario is transitioned to conversation control based on the second conversation scenario. The method includes the steps of:

[0057] (Appendix 2) 2. The conversation control method according to claim 1, wherein each of the plurality of conversation scenarios corresponds to each of a plurality of topics according to a conversation scene.

[0058] (Appendix 3) A conversation control method as described in Appendix 2, wherein each of the plurality of conversation scenarios is defined by concatenating a plurality of unit conversations, each of which pairs the utterance content with a response according to a topic corresponding to the conversation scenario in which the utterance content is included.

[0059] (Appendix 4) A plurality of conversation scenarios are defined for each of the plurality of conversation scenes; When the utterance content does not correspond to any conversation content of any conversation scenario included in a first conversation scene, and corresponds to any conversation content of any conversation scenario included in a second conversation scene different from the first conversation scene, a response according to any of the conversation content of any of the conversation scenarios included in the second conversation scene is output, and conversation control based on the first conversation scene is transitioned to conversation control based on the second conversation scene. 4. A conversation control method according to claim 2 or 3.

[0060] (Appendix 5) The conversation control method according to any one of Supplementary Note 1 to Supplementary Note 4, further comprising accepting, as the content of the utterance, the user's intention estimated based on the user's utterance and a model that estimates the user's intention according to the utterance.

[0061] (Appendix 6) A conversation control method as described in Appendix 5, wherein, when the utterance content indicating the estimated intention of the user does not correspond to any of the conversation contents included in the multiple conversation scenarios, a predetermined standard response is output in accordance with the estimated intention of the user.

[0062] (Appendix 7) A conversation control device that executes conversation control based on one of a plurality of conversation scenarios, a determination unit that, when an utterance content received during execution of conversation control based on a first conversation scenario corresponds to any one of the conversation contents included in the first conversation scenario, determines the any one of the conversation contents as a transition destination, and, when the utterance content does not correspond to any one of the conversation contents included in the first conversation scenario and corresponds to any one of the conversation contents included in a second conversation scenario different from the first conversation scenario, determines the any one of the conversation contents included in the second conversation scenario as a transition destination; an output unit that outputs a response according to the conversation content of the transition destination determined by the determination unit, and when any of the conversation contents included in the second conversation scenario is determined as the transition destination, transitions conversation control based on the first conversation scenario to conversation control based on the second conversation scenario. A conversation control device including:

[0063] (Appendix 8) 8. The conversation control device according to claim 7, wherein each of the plurality of conversation scenarios corresponds to each of a plurality of topics according to a conversation scene.

[0064] (Appendix 9) The conversation control device according to claim 8, wherein each of the plurality of conversation scenarios is defined by linking together a plurality of unit conversations, each of which is a pair of the utterance content and a response according to a topic corresponding to the conversation scenario in which the utterance content and the utterance content are included.

[0065] (Appendix 10) A plurality of conversation scenarios are defined for each of the plurality of conversation scenes; When the utterance content does not correspond to any conversation content of any conversation scenario included in a first conversation scene and corresponds to any conversation content of any conversation scenario included in a second conversation scene different from the first conversation scene, the determination unit determines any of the conversation content of any of the conversation scenarios included in the second conversation scene as a transition destination. 10. The conversation control device according to claim 8 or 9.

[0066] (Appendix 11) The conversation control device according to any one of Supplementary Note 7 to Supplementary Note 10, further comprising an estimation unit that receives, as the content of the utterance, an intention of the user estimated based on the user's utterance and a model that estimates the intention of the user according to the utterance.

[0067] (Appendix 12) when the utterance content indicating the estimated intention of the user does not correspond to any of the conversation contents included in the plurality of conversation scenarios, the determination unit determines a predetermined standard conversation as a transition destination in accordance with the estimated intention of the user; The output unit outputs a response according to the standard conversation. 12. The conversation control device of claim 11.

[0068] (Appendix 13) A conversation control program for executing conversation control based on one of a plurality of conversation scenarios, When a conversation content received during execution of a conversation control based on a first conversation scenario corresponds to any of the conversation contents included in the first conversation scenario, outputting a response corresponding to any of the conversation contents; When the utterance content does not correspond to any of the conversation contents included in the first conversation scenario and corresponds to any of the conversation contents included in a second conversation scenario different from the first conversation scenario, a response corresponding to any of the conversation contents included in the second conversation scenario is output, and conversation control based on the first conversation scenario is transitioned to conversation control based on the second conversation scenario. A conversation control program for causing a computer to execute processes including the above.

[0069] (Appendix 14) The conversation control program according to claim 13, wherein each of the plurality of conversation scenarios corresponds to each of a plurality of topics according to a conversation scene.

[0070] (Appendix 15) A conversation control program as described in Appendix 14, wherein each of the plurality of conversation scenarios is defined by linking together a plurality of unit conversations, each of which pairs the utterance content with a response according to a topic corresponding to the conversation scenario in which the utterance content and the utterance content are included.

[0071] (Appendix 16) A plurality of conversation scenarios are defined for each of the plurality of conversation scenes; When the utterance content does not correspond to any conversation content of any conversation scenario included in a first conversation scene, and corresponds to any conversation content of any conversation scenario included in a second conversation scene different from the first conversation scene, a response according to any of the conversation content of any of the conversation scenarios included in the second conversation scene is output, and conversation control based on the first conversation scene is transitioned to conversation control based on the second conversation scene. 16. A conversation control program according to claim 14 or 15.

[0072] (Appendix 17) The conversation control program according to any one of appendices 13 to 16, wherein the conversation control program accepts, as the content of the utterance, an intention of the user estimated based on the user's utterance and a model that estimates the user's intention according to the utterance.

[0073] (Appendix 18) 18. The conversation control program according to claim 17, wherein, when the utterance content indicating the estimated intention of the user does not correspond to any of the conversation contents included in the plurality of conversation scenarios, a predetermined standard response is output in accordance with the estimated intention of the user.

[0074] (Appendix 19) A conversation control program for executing conversation control based on one of a plurality of conversation scenarios, When a conversation content received during execution of a conversation control based on a first conversation scenario corresponds to any of the conversation contents included in the first conversation scenario, outputting a response corresponding to any of the conversation contents; When the utterance content does not correspond to any of the conversation contents included in the first conversation scenario and corresponds to any of the conversation contents included in a second conversation scenario different from the first conversation scenario, a response corresponding to any of the conversation contents included in the second conversation scenario is output, and conversation control based on the first conversation scenario is transitioned to conversation control based on the second conversation scenario. A storage medium storing a conversation control program for causing a computer to execute a process including the above. [Explanation of symbols]

[0075] 10 Conversation control device 12 Estimation part 14 Decision Section 16 Output section 22 Intention Estimation Model 24 Scenario DB 40 Computer 41 CPU 42 Memory 43 Storage section 49 Storage medium 50 Conversation Control Program

Claims

1. A conversation control method for executing conversation control based on one of a plurality of conversation scenarios in which unit conversations, each of which is a pair of a conversation content and a response, are linked in a sequential manner to form a story flow, comprising: accepting a user's utterance while executing conversation control based on a first conversation scenario, accepting as an utterance content a user's intention corresponding to the user's utterance that is estimated based on the user's utterance and a model that has been generated in advance by machine learning so as to estimate the user's intention corresponding to the utterance, and if the accepted utterance content corresponds to a conversation content in any one of the unit conversations included in the first conversation scenario, outputting an answer in any one of the unit conversations; When the utterance content does not correspond to the conversation content in any of the unit conversations included in the first conversation scenario, and corresponds to the conversation content in any of the unit conversations included in a second conversation scenario different from the first conversation scenario, a response in any of the unit conversations included in the second conversation scenario is output, and conversation control based on the first conversation scenario is transitioned to conversation control from any of the unit conversations included in the second conversation scenario. The method includes the steps of:

2. 2. The conversation control method according to claim 1, wherein each of the plurality of conversation scenarios corresponds to a plurality of topics according to a conversation scene.

3. The conversation control method according to claim 2, wherein each of the plurality of conversation scenarios is defined by linking together a plurality of unit conversations each of which pairs the conversation content with a response according to a topic corresponding to the conversation scenario in which the conversation content is included.

4. A plurality of conversation scenarios are defined for each of a plurality of conversation scenes; When the utterance content does not correspond to any conversation content of any conversation scenario included in a first conversation scene and corresponds to any conversation content of any conversation scenario included in a second conversation scene different from the first conversation scene, a response according to any of the conversation content of any of the conversation scenarios included in the second conversation scene is output, and conversation control based on the first conversation scene is transitioned to conversation control based on the second conversation scene. The conversation control method according to claim 2 or 3.

5. The conversation control method according to any one of claims 1 to 4, further comprising the step of outputting a predetermined standard response in accordance with the estimated user's intention when the utterance content indicating the estimated user's intention does not correspond to any of the conversation contents included in the plurality of conversation scenarios.

6. A conversation control device that executes conversation control based on one of a plurality of conversation scenarios in which unit conversations, each of which is a pair of a conversation content and a response, are linked in a sequential manner to form a story flow, a determination unit that receives a user's utterance while executing conversation control based on a first conversation scenario, receives as utterance content a user's intention corresponding to the user's utterance that is estimated based on the user's utterance and a model previously generated by machine learning so as to estimate the user's intention corresponding to the utterance, and when the received utterance content corresponds to the conversation content in any one of the unit conversations included in the first conversation scenario, determines the any one of the unit conversations as a transition destination, and when the utterance content does not correspond to the conversation content in any of the unit conversations included in the first conversation scenario and corresponds to the conversation content in any one of the unit conversations included in a second conversation scenario different from the first conversation scenario, determines the any one of the unit conversations included in the second conversation scenario as a transition destination; an output unit that outputs a response in the unit conversation of the transition destination determined by said determination unit, and when any one of the unit conversations included in said second conversation scenario is determined as the transition destination, transitions conversation control based on said first conversation scenario to conversation control based on said second conversation scenario; A conversation control device including:

7. A conversation control program for executing conversation control based on one of a plurality of conversation scenarios in which unit conversations, each of which is a pair of a conversation content and a response, are linked in a sequential manner to form a story flow, comprising: accepting a user's utterance while executing conversation control based on a first conversation scenario, accepting as an utterance content a user's intention corresponding to the user's utterance that is estimated based on the user's utterance and a model that has been generated in advance by machine learning so as to estimate the user's intention corresponding to the utterance, and if the accepted utterance content corresponds to a conversation content in any one of the unit conversations included in the first conversation scenario, outputting an answer in any one of the unit conversations; When the utterance content does not correspond to the conversation content in any of the unit conversations included in the first conversation scenario, and corresponds to the conversation content in any of the unit conversations included in a second conversation scenario different from the first conversation scenario, a response in any of the unit conversations included in the second conversation scenario is output, and conversation control based on the first conversation scenario is transitioned to conversation control from any of the unit conversations included in the second conversation scenario. A conversation control program for causing a computer to execute processes including the above.

Citation Information

Patent Citations

  • Interactive method, interactive device, interactive program, and recording medium

    JP2007047488A

  • Interaction method, interaction device, interaction program, and recording medium

    JP2007079397A

  • Interaction control device, interaction control method and interaction control program

    JP2008170817A

  • Data processing device, data processing method and program

    JP2018128747A

  • Growth type interactive device

    JP2019101291A