JPH09237278A - Precise dialogue processing system - Google Patents

Abstract

(57) [Abstract] [Problem] In a system for explaining a message to the other party through voice or a keyboard and conducting a question-and-answer dialogue with the other party, the intention of the other party is promptly extracted, and this is analyzed instantly. To provide a precise dialogue processing system that reflects the will of the other person, who can deal with it properly. SOLUTION: The present system, for example, allows the other party to interrupt a question or confirmation even while talking, and the characteristics of reflexive utterance such that the other person stagnate or hits a partner, and, for example, By obtaining together with the characteristics of the deliberate utterance that considers the state of the other party and the situation of the dialogue that are necessary to achieve it, it is possible to actively generate utterances that direct the task execution in response to the other party's reaction. As a result, it is possible to efficiently obtain the same effect as when humans directly interact with each other.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a detailed dialogue processing system for a dialogue system that allows a partner to perform a task while answering a question or confirmation from a user via a voice or a keyboard.

[0002]

2. Description of the Related Art In the conventional dialogue processing technique, in order to generate an utterance for causing the other party to perform a task,
First, the task is divided into all the procedures in advance and the utterance content is planned.After the task is divided into the procedures that can be executed in detail, the actual utterance is executed.If the utterance execution fails, the plan is created from the beginning. The method of re-doing was adopted.

In addition, as one of the plans that is responsive to the situation, the activation energy is obtained from the goal or the state and the Strips is obtained.
Which is an active propagation mechanism in which active energy propagates between nodes in the plan operator's flow and which selects the node with the highest active energy in accordance with the current state.
Network Architecture (ANA)
(P. Maes. How To DoTheRight
Things. Connection Science
e, Vol. 1, No. 3,1989. ) Has been proposed.

[0004]

In the conventional dialogue system, utterance planning and execution are separated, so that there is a time lag between utterance planning and execution. Therefore, the situation of the dialogue changes between the time of utterance planning and the time of execution,
When making a utterance, the utterance is no longer appropriate, and there arises a problem that the utterance cannot be generated according to the situation of the dialogue. Also, under the cycle of utterance planning and execution, it is not possible to accept interrupts such as questions or confirmations from the other party during the utterance execution, and the other party is planned in order to plan an appropriate response after the utterance ends. There is a problem that it is impossible to reflexively answer the interruption of.

In order to deal with such a problem,
When NA is used, in ANA, a single network operates with a single parameter, which causes the problem that the reflexivity and contemplation of dialogue cannot be dealt with. Furthermore, all states are expressed as true or false. Therefore, there is a problem that it is not possible to represent a gradual state change such as the confidence of the other party.

Therefore, an object of the present invention is to provide a precise dialogue processing system capable of interrupting such an opponent and capable of performing prompt decision processing in response to gradual state changes. .

[0007]

In order to solve such a problem, the precise dialogue processing system of the present invention carries out dialogue for answering questions or confirmations from the other party through voice or keyboard while performing a task. In an unmanned dialogue system,
By presenting the utterance information to the other party and analyzing the linguistic content of the response from the other party to the utterance information, the proposition to be solved is extracted, and the physics of the reaction of the other party to the utterance information such as approval or disapproval or understanding situation Parameter values representing the psychological state of the other party regarding the proposition to be solved are calculated by analyzing the social characteristics, and the dialogue scenario with the other party is rapidly changed according to the parameter value, The content of the process is to rapidly correct the utterance information content and the utterance information presentation method.

More specifically, the present interactive processing system is composed of only utterance nodes having a reflexive property, and stores a set of propositions having a continuous partner confidence of 0 to 1 and storing them. From the result of referring to all the propositions in the storage unit, an optimal utterance node is selected from a plurality of utterance nodes that direct communication by the active propagation mechanism of the ANA that is more strongly directed to the current state, and the selected utterance node is selected. Concerning the proposition obtained from the contribution proposition derivation unit, which is transmitted to the output unit and derives which proposition the selected utterance node contributes to the achievement of the proposition, the certainty factor change unit changes the certainty factor of the corresponding proposition in the partner state storage unit. And try to maintain the dialogue by notifying the change of the threshold size in relation to the proposition obtained from the contribution proposition derivation unit by the threshold change unit. From the result of referring to all the propositions in the target state storage part and the partner state storage part, which consist only of the emissive utterance generation part and the utterance node having a deliberate property, the more target state Select the most suitable utterance node from the plurality of utterance nodes that aim to achieve the task by the active propagation mechanism that directs the task, and transmit the selected utterance node to the output unit. Degree is defined by the number of sub-goals that contribute to achieving that goal, and the difficulty level of the goal is

[0009]

[Equation 1] (However, Δβ j is the magnitude of the change in the certainty factor of the relevant proposition j,
A (j) is the number of contributing sub-targets) and is obtained from the contribution proposition derivation unit that derives which proposition the selected utterance node contributes to the achievement of the calculated difficulty level of the target. With regard to the given proposition, the utterance of the user and the two utterance generation units, which are a thoughtful utterance generation unit that aims to accomplish the task by changing the certainty factor of the proposition in the partner state storage unit, are used as the propositions. It has a transfer knowledge storage unit for converting and storing the knowledge for achieving the goal, and further refers to the transfer knowledge storage unit to judge whether or not the utterance of the partner matches the knowledge to be transferred, and the partner status storage unit The utterance analysis unit that changes the certainty factor of the relevant proposition of, and the certainty factor thresholds of the held proposition and the certainty factors of all the propositions in the state storage unit are compared, and the certainty factor exceeds the threshold value. Is true, and those that do not exceed are false An authenticity determination unit added to the pair of the proposition and the certainty factor of the proposition, an output unit that obtains the utterance node selected by the utterance node selection unit of the reflexive utterance generation unit and the thoughtful utterance generation unit, and outputs the utterance node by voice, Always monitor the other person,
When the utterance of the other party is input by the keyboard or voice, the utterance analysis unit is called, the truth determination unit is inquired about whether the goal is achieved, and the reflexive utterance generation unit and thoughtful utterance until the goal is achieved. A control unit that repeats the operation of calling the generation unit is provided, and means for repeating the interaction of the two utterance generation units via the state storage unit described above is added while accepting the utterance from the other party.

[0010]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of a detailed dialogue processing system of the present invention, FIG. 2 is a main flowchart of this example, and FIG. 3A is a diagram of the reflexive utterance generator 2 of this example. A flow chart, (b) is also a flow chart of the reflective utterance generation unit 3, FIG. 4 is a block diagram of the partner state storage unit 11 of this example, and FIG. 5 is an utterance node selection unit of the reflective utterance generation unit 3 of this example. 32 is a block diagram of the contribution proposition derivation unit 21 of the reflexive utterance generation unit 2 of this example, and FIG. 6B is a contribution proposition derivation unit 31 of the reflection utterance generation unit 3 of the same example. Diagram,
FIG. 7 is a diagram showing a change in the certainty factor of a proposition in this example. FIG.
(B) is a case where there is a wrong interruption of the other party.

As shown in FIG. 1, the state storage unit 1 includes a partner state storage unit 11 composed of a set of pairs of each proposition and a certainty factor for the proposition, and a target state storage unit 12 for storing a goal of a service. Composed of.

The reflexive utterance generating unit 2 has a utterance node selecting unit 22 for selecting an utterance by an active propagation mechanism with reference to the authenticity judging unit 5, and a utterance node name and a list having prerequisites and effects of the utterance node. Then, using the selected utterance node name as a key, the contribution proposition derivation unit 21 that derives the proposition that the utterance node contributes to the achievement, which is the effect of the utterance node, and the proposition obtained from the contribution proposition derivation unit 21 The partner state storage unit 11 includes a certainty factor changing unit 23 that changes the certainty factor of the relevant proposition, and a threshold value changing unit 24 that changes the threshold value in relation to the proposition obtained from the contribution proposition deriving unit 21.

The deliberative utterance generation unit 3 includes a partner state storage unit 11
And a target state storage unit 12 and a utterance node selecting unit 32 for selecting an utterance by an active propagation mechanism by referring to the true / false determination unit 5, and a list having a utterance node name and preconditions and effects of the utterance node. , A contribution proposition derivation unit 31 that derives a proposition that contributes to the achievement of the utterance node, which is the effect of the utterance node, by using the selected utterance node name as a key, and a partner about the proposition obtained from the contribution proposition derivation unit 31 The state storage unit 11 includes a certainty factor changing unit 33 that calculates and changes the certainty factor of the relevant proposition.

The utterance analysis unit 4 has a transfer knowledge storage unit 41 for converting the utterance of the other party using a voice or a keyboard into a proposition and storing the knowledge to be transferred to achieve the goal, and the transfer knowledge storage unit 41. Referring to, it is determined whether the utterance of the other party matches the knowledge to be transmitted, and the other party state storage unit 11 is changed.

The true / false determination unit 5 holds a threshold value of the certainty factor of the proposition that is variable according to the state of the other party, compares all the propositions in the state storage unit 1 with the threshold value, and if the certainty factor exceeds the threshold value. A judgment result of true and false if not exceeded is added to each proposition and the pair of certainty of the proposition.

The output unit 6 obtains the utterance node generated by the reflexive utterance generation unit 2 and the thoughtful utterance generation unit 3 and outputs it as a voice.

The control unit 7 inquires of the authenticity determination unit whether or not the target has been achieved, and calls the reflexive utterance generation unit 2 and the thoughtful utterance generation unit 3 until the target is achieved, monitors the other party, and monitors the other party. If there is an utterance, a series of operations of calling the utterance analysis unit 4 is repeated.

Next, the operation of this example will be described.

As shown in FIG. 2, the control unit 7 inquires of the true / false determination unit 5 whether the target is achieved, and repeats the following until the proposition of the target state is regarded as true.

When there is no utterance from the other party, FIG.
As shown in (a), in the reflexive utterance generation unit 2, the utterance node selection unit 22 similarly selects the utterance node by the active propagation mechanism using the active energy obtained by referring to the true / false determination unit 5. , Proposition that the utterance node contributes to achievement by the contribution proposition deriving unit 21 (effect of the utterance node)
Since the reliance is emphasized, the certainty factor of the corresponding proposition in the opponent state storage unit 11 is changed by the certainty factor changing unit 23 so that the changed certainty factor is always 1.0. Then, the selected speech node is sent to the output unit 6, and the speech fragment is output.

Further, as shown in FIG. 3B, in the thoughtful utterance generator 3, the utterance node selector 32 uses the active propagation mechanism using the active energy obtained by referring to the true / false determination unit 5. The utterance node is selected, the contribution proposition derivation unit 31 calculates the degree of certainty to be changed by the certainty factor changing unit 33 with respect to the proposition that the utterance node contributes to achievement (effect of the utterance node), and the partner state storage unit 11 Change the certainty of the relevant proposition of. Then, the selected speech node is sent to the output unit 6, and the speech fragment is output.

When the other party makes an interruption by using a voice or a keyboard, the utterance analysis section 4 converts the other party's utterance into a proposition and refers to the transfer knowledge storage section 41 to check whether the other party's interruption utterance is correct. It is judged whether it is wrong, and the certainty factor of the relevant proposition in the partner state storage unit 11 is changed.

The reflexive utterance generation unit 2 refers to the authenticity determination unit 5 for determining the authenticity of the changed partner state storage unit 11, and the utterance node selection unit 22 conveys whether or not the interrupt of the partner is correct. Is selected, the certainty factor changing unit 23 changes the certainty factor of the corresponding proposition in the partner state storage unit 11 to 1.0 for the proposition that the utterance node obtained from the contributing proposition derivation unit 21 contributes to achievement, and the threshold changing unit 24 If the utterance from the other party conforms to the content to be transmitted, the threshold value is lowered, and if not, the threshold value is increased.

Further, in the deliberative utterance generating unit 3, the utterance node selecting unit 32 reactivates the active propagation using the active energy obtained from the true / false determining unit 5 which refers to the changed partner state storing unit 11. As a result, the utterance node suitable for the current state is selected.

For example, in a dialogue system having a task of explaining an answering machine, explanation of the position and characteristics of a button necessary for recording operation, request of an act of pushing the button, and effects of what happens when the button is pushed will be explained. Must.

The target state storage unit 12 stores the target proposition e
xplane-how-to-record-mess
The knowledge about the telephone to be transmitted is stored in the transmission knowledge storage unit 41, and the conjecture utterance generation unit 3 stores the proposition explain-how-t as shown in FIG.
In order to achieve the o-record-message,
There are utterance nodes that convey the location, color, and shape of the button that must be pressed, and utterance nodes that request the action of pressing the button and utterance nodes that explain the result of the action.

As shown in FIG. 5, in the reflective utterance generating unit 3, the utterance node selecting unit 32 selects the utterance node "at the upper left of the main body" from the utterance nodes, which explains the location of the button, by the active propagation algorithm. select.

As shown in FIG. 6B, the contribution proposition derivation unit 31 determines from the utterance node “at the upper left of the main body” that the utterance node is the effect of the utterance node, that is, the proposition i that contributes to achievement.
Derify-button is derived. For the derived proposition identity-button, the certainty factor changing unit 33 of the deliberative utterance generation unit 3 calculates the certainty factor to be changed based on the mathematical formula that defines the difficulty level of the target, and the utterance node related to the button is the place. , Shape, color, name, nearby features, and display.

[0030]

[Equation 2] Only increase the certainty of the corresponding proposition identify-button.

As shown in FIG. 7, the initial value of this proposition is 0.
Therefore, the current certainty factor is 0.378. Since the threshold value is initially set to 0.5, it is still below the threshold value at this point, so the authenticity determination unit 5 does not determine that the button has been understood yet.

FIG. 4 shows the state of the partner state storage unit 11,
It consists of a set of propositions and certainty factors of the propositions. The true / false determination unit 5 considers that the proposition is understood if the certainty factor of the proposition exceeds the threshold value, and considers that the proposition is not understood if it does not exceed the threshold value.

However, when a proactive interrupt from the other party, such as "Is it a response 1 button?", Comes in, the utterance analysis unit 4 refers to the transmitted knowledge storage unit 41 to make a utterance proposition user.
-Identify-name-of-button and convert the certainty factor of the corresponding proposition in the partner state storage unit 11 to 1.
Raise to 0. Utterance node selection unit 2 of reflexive utterance generation unit 2
As a result of referring to the authenticity determination unit 5, 2 obtains activation energy, and selects the utterance node "Yes, that is the button with that name" by the activation propagation mechanism.

As shown in FIG. 6 (a), the contribution proposition derivation unit 21 is a proposition that the utterance node "yes, it is a button with that name", which is the effect of the utterance node, contributes to the achievement.
Derive the configure-button. Confidence degree changing unit 23
Is the identity-butto of the partner state storage unit 11.
The certainty factor of n is increased to 1.0, and in this case, it is determined that the other party has correctly identified the button, and the threshold changing unit 24
Lowers the threshold to 0.3. The threshold affects the choice of speech node. When the threshold is lowered, all propositions are likely to exceed the threshold when the next utterance node is selected, that is, the true / false determination unit 5 is likely to determine that the other party understands, and the thoughtful utterance generation unit 3 is redundant. No need to generate utterances.

The utterance node selection unit 3 of the deliberative utterance generation unit 3
2 refers to the authenticity determination unit 5 that refers to the changed partner state storage unit 11, and the certainty factor of the proposition identify-button is 1.0 and exceeds the current threshold value of 3.0. Decides that the other person understands,
The next proposition explain-a without the other five utterances
Select the utterance for the action.

On the contrary, the other party asks, "Is this dial button 1?"
When the erroneous interrupt is input, the utterance analysis unit 4 refers to the transfer knowledge storage unit 41 and utters the utterance as a proposition user-not-
It is converted into the utterance "identify-name-of-button". The utterance node selection unit 22 of the reflexive utterance generation unit 2 refers to the true / false determination unit 5 and determines that it indicates that the other party's interrupt is not understood, and the active propagation mechanism displays “No, button with that name”. It is not ". Contribution proposition derivation part 2
Since there is no proposition that is the effect of the utterance node "No, it is not a button with that name" by 1, and it does not contribute to the achievement of any proposition, the confidence factor changing unit 23 does not change the confidence factor of any proposition. In this case, since the other party has not correctly identified the button, the threshold changing unit 24 sets the threshold to 0.7.
Raise to When the threshold is raised, it becomes difficult for all the propositions to exceed the threshold when the next utterance node is selected, that is,
It becomes difficult for the authenticity determination unit 5 to determine that the other party understands,
The deliberative utterance generation unit 3 generates more utterances and can explain them in detail.

The utterance node selection unit 3 of the thoughtful utterance generation unit 3
2 refers to the authenticity determination unit 5 that refers to the changed partner state, and the certainty factor of the propositional identity-button is 0.378, which does not exceed the current threshold value of 0.7.
Without determining that the proposition is understood, the utterance node selection unit 32 further activates the active propagation mechanism so as to select the utterance node for explaining the proposition, and selects the utterance node "flashing red". . The contribution proposition derivation unit 31 is a proposition that the utterance node "flashes in red", which is an effect of the utterance node, contributes to achievement.
Derive the define-button. Confidence level change section 3
3 is an identity-button certainty factor of 0.3
It is calculated as 78, and the certainty factor of the relevant proposition in the opponent state storage unit 11 is increased to 0.756. Proposition identity-but
Since the certainty factor of ton exceeds the current threshold value of 0.7, the utterance node for explaining another proposition is selected in the next utterance selection.

Although the present embodiment has been described above, the main advantages of the system of the present invention can be summarized as follows.

In this system, since the control unit 7 constantly monitors the other party, the other party's utterance can be accepted even while the dialogue system is speaking, and the analysis result of the interrupt utterance is always stored in the other party's state. As reflected in the unit 11, the reflexive utterance generation unit 2 immediately determines whether or not the proposition that the dialogue system is trying to explain matches the utterance of the other party, and utters the result or generates a thoughtful utterance. The section 3 can select an appropriate utterance suitable for the understanding situation of the other party at that time. This solves the problems caused by the time lag between the planning and execution of speech.

For example, if the opponent makes a proactive interrupt, the reflexive utterance generator 2 immediately selects the utterance indicating that the interrupt is correct, and the contemplation utterance generator 3 further details the relevant proposition. You can proceed without explaining. On the other hand, when the other party makes an interrupt indicating a mistake or incomprehension, the reflexive utterance generator 2 immediately selects an utterance indicating that the interrupt is different from the proposition described or an utterance that stutters, and the reflective utterance generator 3 Can be further elaborated by selecting another utterance that describes the relevant proposition.

[0041]

As described above, according to the present invention, when the other party interacts with the dialogue system by voice or keyboard, even if the dialogue system is in the middle of uttering the other party to perform the task, the other party can ask a question or confirm. The system's dialogue strategy dynamically appears according to the other party's utterance content, so the other party does not have to wait until the system completes the utterance and starts the other party's utterance content. It is possible to maintain the communication with the other party and to dynamically generate the utterance for performing the task while always watching the current state of the other party without having to reconsider. The appearance timing of each utterance is automatically determined by the contents of interruption of the other party, and there is an effect that it is possible to provide a precise dialogue processing system that enables natural dialogue with the system like direct dialogue between humans.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of a precise dialogue processing system of the present invention.

FIG. 2 is a main flowchart of this example.

FIG. 3A is a flowchart of a reflexive utterance generation unit 2 of the present example, and FIG. 3B is a flowchart of a reflective utterance generation unit 3 of the same example.

FIG. 4 is a configuration diagram of a partner state storage unit 11 of this example.

FIG. 5 is a configuration diagram of an utterance node selection unit 32 of the reflective utterance generation unit 3 of this example.

6A is a block diagram of a contribution proposition derivation unit 21 of the reflexive utterance generation unit 2 of the present example, and FIG. 6B is a block diagram of a contribution proposition derivation unit 31 of the reflection utterance generation unit 3 of the same example. is there.

FIG. 7 is a diagram showing a change in certainty of a proposition in this example, where (a) is a case where there is a preceding interruption of the other party,
(B) is a case where there is a wrong interruption of the other party.

[Explanation of symbols]

 1 state storage unit 2 reflexive utterance generation unit 3 contemplation utterance generation unit 4 utterance analysis unit 5 truth determination unit 6 output unit 7 control unit 10 partner 11 partner state storage unit 12 target state storage unit 21 2 contribution proposition derivation unit 22 2 utterance node selecting unit 23 2 certainty factor changing unit 24 2 threshold changing unit 31 3 contribution proposition deriving unit 32 3 utterance node selecting unit 33 3 certainty factor changing unit 41 transfer knowledge storage unit

Claims (6)

[Claims] 1. An unmanned dialogue system for carrying out dialogue while answering a question or confirmation from the other party through voice or a keyboard, to present speech information to the other party, and a language of a response from the other party to the speech information. By extracting the proposition to be solved by analyzing the content, the physical characteristics of the reaction such as the pros and cons of the other party or the understanding situation to the utterance information are analyzed, and the psychological situation of the other person about the proposition to be solved is analyzed. Calculating a parameter value representing a different state, quickly changing the dialogue scenario with the other party according to the parameter value, and quickly correcting the utterance information content and the utterance information presentation method to the other party. A precise dialogue processing system characterized by 2. The dense dialogue processing system according to claim 1, wherein the parameter value indicating the psychological state uses a certainty factor indicating how much the opponent understands the proposition. 3. In the step, the state storage unit (1) represents a user state or a target state of the dialogue system, and a true / false determination unit (5) is stored in the state storage unit (1). The authenticity of each proposition is judged, the reflexive utterance generator (2) maintains communication with the other party, and the deliberative utterance generator (3) aims to make the other person perform the task. The analysis unit (4) analyzes the utterance from the other party, the output unit (6) outputs the utterance of the dialogue system, the control unit (7) controls these, and the reflexive utterance generation unit ( 2) and the thoughtful utterance generation unit (3) interact via the state storage unit (1), allowing interruptions such as questions and confirmations of the other party even during the utterance of the dialogue system, and performing tasks according to the other party's reaction. 3. The denseness according to claim 1 or 2, which comprises dynamically generating the utterance of Story processing system. 4. The partner status storage unit (11) of the status storage unit (1) stores a continuous value from 0 to 1 representing a proposition and how much the partner understands the proposition. The set of pairs with the confidence of the opponent to be taken is stored, the goal state storage unit (12) stores the goal of the service, and the authenticity determination unit (5) is a confidence factor that is variable according to the state of the opponent. Of the probabilities of all the propositions in the state storage unit (1) are compared with the thresholds of the probabilities stored in the state storage unit (1). Is added to each proposition and a certainty factor of the proposition, and in the reflexive utterance generation unit (2), the utterance node selection unit (22) obtains active energy from the true / false determination unit (5). , A situation-oriented, active-propagation mechanism that directs communication Select the best speech node from the speech node, contribute proposition deriving unit (21)
Has a list of utterance node names, preconditions and effects of the utterance nodes, and derives the effect of the utterance node, that is, the proposition that the utterance node contributes to achievement, by using the selected utterance node name as a key. , Confidence factor changing unit (2
3) changes the certainty factor of the relevant proposition in the opponent state storage unit (11) according to the proposition obtained from the contribution proposition derivation unit (21), and the threshold value changing unit (24) derives the contribution proposition. The threshold value of the certainty factor is changed according to the proposition obtained by the unit (21), the communication with the other party is maintained by these, and the utterance node selection unit (32) of the thoughtful utterance generation unit (3) is The active energy is obtained from the authenticity determination unit (5), an optimal utterance node is selected from a plurality of utterance nodes that aim to achieve a task by a goal-oriented active propagation mechanism, and a contribution proposition deriving unit (31). Has a list of utterance node names, preconditions and effects of the utterance nodes, and with the selected utterance node name as a key, the effect of the utterance node, that is, a proposition that the utterance node contributes to achievement. And the confidence factor changing unit (33) determines the difficulty of targeting the change in the confidence factor of the relevant proposition in the opponent state storage unit (11) according to the proposition obtained from the contributing proposition derivation unit (31). The utterance analysis unit (4) converts the utterance of the other party into a proposition, by changing the confidence level of the relevant proposition calculated by
It has a transfer knowledge storage unit (41) for storing knowledge to be transferred to achieve the goal, judges whether the utterance of the other party matches the content to be transferred from the transfer knowledge storage unit (41), and Based on the result, the certainty factor of the proposition in the partner state storage unit (11) is changed, and the output unit (6) outputs the reflexive utterance generation unit (2) and the thoughtful utterance generation unit (3). The utterance node selection unit (22, 32) obtains the utterance node selected and outputs it by voice, and the control unit (7) confirms with the true / false determination unit (5) whether or not the target has been achieved, Until the above is achieved, the reflexive utterance generation unit (2) and the contemplation utterance generation unit (3) are called, the utterance from the other party is monitored, and if there is an utterance from the other party, the utterance analysis unit (4) is called. The precision according to claim 3, wherein a series of operations of calling is repeated. Interactive processing system. 5. The partner state storage unit (11) of the state storage unit (1) is a set of pairs of propositions and certainty factors of the propositions, and the target state storage unit (12) is a target. The state is stored, the authenticity determination unit (5) holds a threshold value of the certainty factor of the proposition that is variable according to the state of the opponent, and the threshold value that is held for all the propositions of the state storage unit (1). Comparing with the above, a determination result that is true if the threshold is exceeded and is false if the threshold is not exceeded is added to each proposition and the confidence level of the proposition, and these always represent the current opponent's state. The detailed dialogue processing system described in 3. 6. The deliberative utterance generation unit (3) generates the utterance on the assumption that the degree of difficulty of the goal is defined by the number of sub-goals contributing to the achievement of the goal, and then the deliberative utterance is generated. Confidence degree changing unit (33) of the dynamic utterance generation unit (3)
The precise dialogue processing system according to claim 3, wherein the degree of change in the certainty factor of the proposition in the opponent state storage unit (11) is calculated according to the difficulty level of the target.