JPH07210392A - User supporting method and user supporting device - Google Patents

Abstract

PURPOSE:To reduce the burdens of rebuilding a user's plan and performing request to a system again by a user himself when the plan is unexecutable. CONSTITUTION:When a request from the user is present, a plan inference part 2 infers the plan of the user. However, when the plan can not be executed, the values of the variables of respective sub goals included in the plan are changed and the change is performed in a predetermined priority order. Thus, alternate plans are generated from the one closer to the request of the user. When the executable alternate plan among them is presented, the user does not need to rebuild the plan by himself and the prompt execution of the plan by the alternate plan becomes possible.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention applies a plan inference method to a method of controlling a goal-oriented collaborative dialogue, and is used for a natural language response of an information processing apparatus and the like. Regarding the device.

[0002]

2. Description of the Related Art In order to improve the operability of an information processing device, a method of operating the information processing device in an interactive form using natural language has been introduced (Japanese Patent Laid-Open No. 1-26382).
3 gazette). In such processing, the input natural language is analyzed, the intended processing is inferred and an appropriate response is made, or a certain processing is specifically executed by the information processing device according to the inference result. As a method of controlling a goal-oriented cooperative dialogue with a system using such a natural language, the plan inference method shown in the following documents is widely adopted. Reference: James Allen: “Recognizing in
tentions from natural language utterances ”, Comp
utational Models of Discourse, The MIT Press, pp10
7-166, (1983)

In this method, a user's plan is inferred from a user's utterance, and if the user's plan is feasible, a system plan for achieving a goal is generated, executed, and responded in natural language.

[0004]

The conventional user support method as described above has the following problems to be solved. For example, when the user support method based on the above plan inference method cannot execute the user's plan, the system tells the user why the execution is impossible. The user receives this and re-establishes an executable plan.

Therefore, for example, when the user inquires in natural language, "Is it possible to use the first conference room from 1 am to 3 pm today?", The system side says that the user's plan is from 1 am to 3 pm today. It is inferred to reserve the first conference room until time. On the other hand, if the first conference room is already reserved from 1 am to 3 pm today, the plan is inexecutable, and a response indicating that it has already been reserved and cannot be reserved is made.

The user receives this and plans again
Change from 1 am to 3 pm today to reserve the second conference room. However, in such a system, we would like to reduce the burden on the user side, such as having the user make a plan again and make a request again. Further, it is preferable that the system establishes some rule for plan re-establishment and services the user. The present invention has been made to solve the above object.

[0007]

According to the user support method of the present invention, a plan inference method is applied to a goal-oriented collaborative dialogue control method for a plan requested by a user, and a response for plan execution is obtained. If so, perform processing according to the following procedure. First, a priority order is set for each variable of all subgoals included in the plan, focusing on the possibility of changing the value of the variable. That is, the variables are ranked such that the higher the possibility of change, the higher the priority.

After that, when the user's utterance is input, it is determined whether the plan inferred from the utterance is feasible. If any of the subgoals included in the plan is not satisfied by the value of that variable and the plan cannot be executed, the value of the variable is changed from the variable with the higher priority to generate the alternative plan. Select a viable option from these alternative plans and present it to the user.

[0009]

[Function] For variables of all subgoals included in the plan, if the priorities are set in order from those that can be changed,
When the requested plan cannot be executed as it is, alternative plans can be generated in order from the one closest to the request. If an alternative feasible plan is selected and presented to the user, the burden of rethinking the plan by the user is lightened as compared with the response that the user's plan cannot be executed.

[0010]

The present invention will be described in detail below with reference to the embodiments shown in the drawings. [Embodiment 1] FIG. 1 shows a block diagram of a user support device for carrying out a user support method of the present invention. As shown in this figure, this method is suitable for a system in which when a utterance suggesting a user's plan is input, the content is analyzed and inferred, and a response is returned to the user by an utterance by the device. It should be noted that the user support method of the present invention itself can be implemented by data input from a keyboard such as a general information processing apparatus, command input via a communication line, or response thereto without using natural language. .

In the figure, in this apparatus, a natural language analysis unit 1, a plan inference unit 2, a plan generation unit 3, a plan execution unit 4, a natural language generation unit 5, and an alternative plan generation unit 6 are provided.
Is provided. In this method, for example, when the user inputs an utterance "Today, the first conference room is free from 1 to 3 o'clock" using the device of such a configuration, the user's plan is set. If the plan is inferred and it is an executable plan, processing such as confirmation of the reservation of the conference room is performed, and the user is answered "Yes, it is vacant."

On the other hand, if it is an infeasible plan, rather than just replying that it is not free,
Generate an alternative plan. Subgoals that compose a plan in which values such as "today", "first meeting room", "1 to 3 o'clock", "is it free" output from the natural language analysis unit 1 are configured? The variables are assigned a priority in the order in which their values can be changed. Here, as shown in the priority order table 6A, the priority order of the conference room, time, and day is set. Therefore, the conference room is set to the one that is most likely to be changed.

For example, if the first conference room is already reserved, the alternative plan generator 6 changes the first conference room to the second conference room. Then, an alternative plan is generated by this, and a response is made to the user via the natural language generation unit 5. Of course, in this case it is confirmed in advance that the alternative plan is feasible. By receiving such a response, the user does not need to assemble a new plan and request the request again.

In order to perform the above-mentioned processing, the natural language analysis unit 1 includes, for example, a device which receives the voice of the user as input, performs voice recognition, and converts this into natural language data.
A microphone and a well-known voice recognition device are used for this. The natural language data obtained in this way is parsed by a well-known language processing program and separated into clauses. Then, it is determined what role each clause plays in the sentence, and the value of the variable of the subgoal included in the plan requested by the user, that is, "today", "first conference room" ... The value of the variable is extracted. The plan inference unit 2 is a unit that inputs a value of such a variable and infers a plan for achieving a goal which is a purpose of user interaction.

If this user's plan is executable, the plan generating unit 3 generates a system plan.
Actions and plan inference rules and plan generation rules as described below are used for such plan inference and generation. FIG. 2 shows a content explanatory view of such an action. This action includes a header (HEADER) slot that indicates the content and name of the action, a precondition (PRECONDITION) slot that must be satisfied before the action is executed, and an effect (EFFECT) when the action is executed. It is described by the slot and the relation (CONSTRAINT) slot of the term inside the action. As a result, when a certain condition is satisfied, an effect produced by performing a certain operation is expressed.

Incidentally, FIG. 3 shows a content explanatory view of each item in the action shown in FIG. The items used in the above actions represent the contents shown in this figure. For example, the first action shown in FIG. 2 has the content “the speaker tells the listener that the proposition P has been established.” (Inform) ). In this case, the precondition is that the speaker knows that the proposition P holds. Then, as the effect, the listener is informed that the proposition P is established (know).

The second action is to inform the listener whether the proposition P holds or not (informif). In this case, the precondition is that the speaker knows whether the proposition P holds (know), and as a result, the listener knows whether the proposition holds. By logically connecting these actions one by one and executing them,
By inferring a user's plan, the system can decide what the system should do, what to do, etc.

FIG. 4 shows a specific user request and a solution example of the system. This is an example of the dialogue in the conference room reservation system, and when the user inputs the utterance "Today, the first conference room is open from 1 am to 3 pm." After performing processing and inference to determine whether the user's plan is viable, he answers "yes".

FIG. 5 is an explanatory diagram of the inference path in the plan inference unit 2 when such a plan is actually input. Further, FIG. 6 shows an explanatory diagram of the plan inference rule. As shown in FIG. 5, the inference path is composed of the above-described action and the plan inference rule established between the slots of each action. For example, the ACTION shown in FIG.
The rule -EFFECT is that if an action is executed, the effect of that action will occur.

The lowest action shown in FIG. 5 is a process in which the user has requested a certain process to the system. This fixed process is a process in which the system notifies the user whether or not the first conference room is reserved from 1 am to 3 pm today. We deduce that the effect of such actions is that the system wants to perform the actions shown below the informif.

The inference rule WANT-ACTION is a rule that if the actor of the action wants to execute the action, the actor of the action executes the action. Therefore, the inference is then made that the machine tells the user if the first meeting room is not booked (informif). In addition, the user knows if the first conference room is not reserved by anyone (knowif).
As a matter of proposition, it is inferred that the purpose is to know that the first conference room is not reserved for anyone from 1 to 3 today. According to this inference,
The system reserves the first conference room for the user (res
erve) and finally concludes that the first conference room is reserved for the user.

From the result of such inference, the system is first
Recognizing that the conference room should be reserved, the plan generation unit 3 shown in FIG. 1 generates a plan for executing a specific process. FIG. 7 shows an explanatory diagram of plan generation according to such inference. Further, FIG. 8 is a diagram for explaining an inference rule when generating such a plan. In this figure, from top to bottom, reasoning proceeds and in order for the user to know that the first room is not reserved for anyone, the system tells the user that the first room is for everyone. It is concluded that it has to be said that it is not reserved and in order to perform that action the system needs to know as a prerequisite that the first conference room is not reserved to anyone.

As described above, a user's plan inferred by the system is a chain of actions linked by plan inference rules. The system plan generated by the system is a chain of actions linked by the plan generation rule. Here, it is preferable that the plan requested by the user can be executed as described above, but the feature of the present invention is the correspondence when the plan cannot be executed.

FIG. 9 is an explanatory view of a response example according to the present invention. In the present invention, as shown in this figure, when the utterance of the user is, for example, "Today, the first conference room is vacant from 1:00 to 3:00." While answering "No", an alternative plan is further generated and provided to the user with the utterance "I can make a reservation in the second conference room".

In order to execute such an operation, the following processing is performed in the present invention. FIG. 10 shows an explanatory diagram of processing by the method of the present invention. In the plan reasoning method, which is a goal-oriented collaborative dialogue control method, the user's plan is a fixed goal, and subgoals (s that must be established in order to achieve that goal).
ubgoal) and are associated with each other. (A) is a user's plan, (b) is an alternative plan.

In the user's plan shown in (a), one subgoal at the left end is a subgoal that does not hold, for example, because it has already been reserved by another reservation person. Other subgoals are established or may be established. These subgoals have one or more variables. The values of these variables are, for example, “today”, “first conference room”, or “1 to 3 o'clock”. In the present invention, the values of these variables are changed. Variables can change their values according to the rules described below. For example, the variable "meeting room" is set so that its value can be changed to "first meeting room", "second meeting room", and "third meeting room".

However, if an arbitrary plan included in the user's request is freely changed to create an alternative plan, the plan may not be desired by the user at all. Therefore, in the present invention, first, for each variable, a priority order is set by focusing on the possibility of changing the value of the variable. FIG. 11 shows a variable priority order explanatory diagram in the above example, for example. When reserving a conference room, each variable is prioritized in the order of conference room, end time, start time, and finally day. That is, the higher the priority, the higher the possibility that the change will not disobey the user's request. For example, in the conference room, there is often no great problem as a user even if the first conference room is changed to the second conference room. On the contrary, it is not easy to change the date and time due to the schedule of the user. From such a relationship, the priority order shown in this figure is set.

Further, a change rule for each variable is defined. That is,
It is desirable that the change is a gradual change that does not overrule the user's plan as much as possible. Therefore, rules for changing the conference room, end time, start time, date, etc. are set in advance. As shown in FIG. 12, in the case of a conference room, for example, if there is a request for the first conference room, it is changed to the second conference room. If there is a request for the second conference room, it is changed to the first conference room. Furthermore, if the conference room A is changed to the conference room B, a rule is set such that the conference room is relatively similar in size and quality. Regarding the end time, the start time, the day, etc., the rules are further tightened, and are changed so that the difference between the values before and after the change is reduced as much as possible. That is,
Make changes in a way that avoids major schedule changes.

A priority table 6A is connected to the alternative plan generator 6 shown in FIG. This is composed of a storage device or the like in which information regarding the priority order and the like, which has already been described with reference to FIG. 11, is stored. The plan generation unit 3 determines the feasibility of the user's plan, generates a system plan according to the user's request, and the plan execution unit 4 starts execution of the system plan, reservation processing of the conference room, etc. I do. Then, the result is transmitted to the user through the natural language generation unit 5. The natural language generation unit 5 is composed of a well-known voice conversion device or the like that generates a natural language from the utterance action of the system, displays it on the display, or further converts it into voice and outputs it using a speaker or the like. It

When the user's request is made by a handwritten document, the natural language analysis unit 1 reads the sentence as image data, performs morphological analysis, syntactic analysis,
This is the part that performs semantic analysis, context processing, etc. and performs analysis.

FIG. 13 shows a flow chart of the alternative plan generation operation. In generating the alternative plan, the values of the variables are sequentially changed and processed in this procedure. First, in step S1, the first rule of the variable value change rules described with reference to FIG. 12 is set. The [loop 1] formed by step S1 and step S3 executes the loop while changing a certain parameter (here, the rule for changing the value of the variable), and when all the switching of the parameter is completed, the loop is executed. This represents a process such as exiting. The same applies to the [loop 2] formed between step S5 and step S8 of this flowchart.

That is, first, in step S1, a rule for changing the value of a variable is set, and in step S2,
Determine if the user's plan has a value assigned to the variable. If the user does not specify the value of the variable, the rule for changing the value of the variable is switched to the next rule and the loop is executed. On the other hand, if the value of any variable is substituted, the process proceeds to step S4, and step S1
Change the value of the variable according to the rule defined in. Next, between steps S5 and S8, the subgoal is changed and the repetitive processing is executed.

If the plan requested by the user includes a plurality of subgoals, step S6 for each subgoal,
The determination of step S7 is executed. First, step S6
Determines whether the subgoal has been established. Further, in step S7, it is determined whether or not a system plan that may fulfill the subgoal can be generated. That is, by the loop 2 of steps S5 to S8, all subgoals may be established or established after the value of the variable is changed, and it is determined whether the plan is executable. If it cannot be executed for some reason, the process is skipped, and steps S1 and S3 are performed again.
Return to the loop sandwiched between.

Here, the rule for changing the value of the variable is changed, and the process of changing the value of the variable is performed again in step S4. When it is determined by the loop of steps S5 and S8 that a system plan in which all subgoals are satisfied or can be satisfied can be generated, the loop is exited and the process proceeds to step S9. Then, the speech act of the system is generated. That is, the semantic expression of the utterance for communicating the alternative plan to the user is generated.

In the example shown in FIG. 9, in response to the request "Is the first meeting room free today from 1:00 to 3:00?", The user's plan is from 1:00 to 3:00 today. We infer that it is to reserve a room, and also know that the first conference room is already reserved from 1 to 3 today,
Tell the user that the plan is not viable. Then, according to the change rule shown in FIG. 12, the conference room is changed from the first conference room to the second conference room, and the alternative plan is generated. When it is determined that the plan is feasible, the user is notified that the second conference room is available. In this way, the system side can voluntarily provide an alternative plan.

[Embodiment 2] Next, an example of appropriately selecting the method of changing the value of the variable as described above will be described. The value of the variable is changed according to the priority order that focuses on the possibility of change. This priority may be determined in advance by the system designer, or may have the following change history as data and be determined according to this.

FIG. 14 is a diagram for explaining a variable value change history by the user. For example, when a user makes a reservation for a conference room by a conventional device and the plan cannot be executed, the user changes the plan and requests again. At this time, for example, the conference room is changed from the first conference room to the second conference room.
When a change is made to the conference room or the third conference room, the number of times such a change is made is counted. Then, the priority order is determined according to the count value. In this example, since the number of times the conference room is changed from the first conference room to the second conference room is the largest, the rule is given the highest priority. Since the number of changes from the first conference room to the A conference room is small, the rule has the lowest priority. Also, the date of the meeting can be changed from today to tomorrow,
The rules for changing from tomorrow to the day after tomorrow will be adopted for the changes.

This makes it possible to set an appropriate change rule that satisfies the user's request as much as possible. In this case, only the case of using which variable and how the variable is changed as described above as a change rule and the number of times the value of which variable is changed are simply memorized. Alternatively, a method of preferentially changing to another value can be adopted.

[Third Embodiment] Next, a method for extracting the weighting of each variable value required by the user by analyzing the natural language input by the user will be described. That is, when the user inputs an utterance in natural language, each variable value is provided with an appropriate modifier. The attached word that modifies the value of each variable is effective as information indicating the weighting of the value of each variable. Therefore, the user's intention based on such an adjunct word is determined, and the non-deterministic expression of the adjunct word gives a higher priority to each variable.

FIG. 15 shows an explanatory diagram of an example of a user's utterance in such a case. Utterance example 1 in the figure is "Is the first conference room vacant tomorrow from 1 am to 3 pm?" That is, the adjunct word "rank" following "tomorrow" does not always specify tomorrow, but is an indeterminate expression that it may be tomorrow or the day after tomorrow. When there is such a nondeterministic subordinate word, the value of the variable for the condition of tomorrow is changed most preferentially. Therefore, when an utterance having such a non-deterministic expression portion is input, after the syntax analysis, the priority order table 6A shown in FIG. 1 is changed. In this way an appropriate alternative plan can be determined.

Further, the utterance example 2 shown in FIG.
Is the first conference room vacant from 1 am to 3 pm? It is. In this case, the adjunct word "around" from "around 1 o'clock to 3 o'clock" is also a non-deterministic expression part. From this, it can be seen that the reserved time portion can be changed most preferentially.
It should be noted that an example of adding such means for determining the order of strength of the user's intention to the natural language analysis unit 1 is, for example, Japanese Patent Application Laid-Open No.
It is also shown in Japanese Patent Publication No.-263823. Therefore, a specific analysis process may follow this method.

The present invention is not limited to the above embodiments. The natural language input / output processing method as described above can be carried out either by characters, by voice, or by other communication data input / output by the information processing apparatus.
Further, the configuration of the apparatus itself for carrying out the method of the present invention is not limited to the above-mentioned embodiment, and blocks having the same function may be added, changed, and divided and combined.

[0043]

According to the user support method and the user support apparatus of the present invention described above, a plan inference method is applied to a plan requested by a user as a purpose-oriented collaborative dialog control method and executed. If you want a response for
When the user's plan cannot be executed, a priority is set that focuses on the possibility of changing the value of each sub-goal included in the plan. Since the response is generated and presented, it is possible to improve the operability of the user by eliminating the process of making a plan that the user can execute again and requesting the system again.

[Brief description of drawings]

FIG. 1 is a block diagram of a specific example of a support device for carrying out the method of the present invention.

FIG. 2 is an explanatory diagram of contents of actions for inference processing.

FIG. 3 is an explanatory diagram of the contents of each item in the action.

FIG. 4 is a diagram illustrating an example of a dialogue in the conference room reservation system.

FIG. 5 is an explanatory diagram of a plan inference path.

FIG. 6 is an explanatory diagram of a plan inference rule.

FIG. 7 is an explanatory diagram of a plan generation path.

FIG. 8 is an explanatory diagram of a plan generation rule.

FIG. 9 is a diagram illustrating a response example according to the present invention.

FIG. 10 is an explanatory diagram of processing according to the method of the present invention.

FIG. 11 is an explanatory diagram of variable priorities.

FIG. 12 is an explanatory diagram of a rule for changing a variable value.

FIG. 13 is a flowchart of an alternative plan generation operation.

FIG. 14 is an explanatory diagram of a variable value change history by a user.

FIG. 15 is an explanatory diagram of an example of a plan requested by a user.

[Explanation of symbols]

 1 Natural Language Analysis Section 2 Plan Inference Section 3 Plan Generation Section 4 Plan Execution Section 5 Natural Language Generation Section 6 Alternative Plan Generation Section

Claims (5)

[Claims] 1. When a plan inference method is applied to a plan requested by a user and a plan inference method is applied to a method for controlling a collaborative dialogue oriented to a goal, each of the plans included in the plan is included. For each variable of the sub-goal, a priority order focusing on the possibility of changing the value of the variable is set, and when the plan of the user is inferred, any sub-goal included in the plan is not satisfied with the value of the variable Is not executable, the value of the variable is changed from the one with the highest priority to generate a viable alternative plan, and a response is presented to present it. 2. A natural language analysis unit that analyzes a utterance input by a user using a natural language and extracts a condition included in the utterance, and a plan reasoning method as a purpose-oriented collaborative control method. , A plan inference unit that infers a user's plan, a plan generation unit that generates a system plan according to the inference result, and for each variable of all subgoals included in the plan, the value of that variable is When the priority is determined by focusing on the possibility of change, and any subgoal included in the plan is not satisfied by the value of the variable and the plan cannot be executed, the value of the variable is determined from the one with the highest priority. To generate an alternative plan, a natural language that is a response generated by executing the plan, and a natural language that expresses an executable alternative plan. User assistance apparatus characterized by comprising a natural language generation unit tell over. 3. When a user requests an arbitrary plan in the past, and when the user changes the value of any subgoal variable by the time the plan is executed, which variable was changed and how the variable was changed. 2. The user support method according to claim 1, wherein the change history shown is recorded, and the change method of the value of each variable is determined by learning the change history. 4. When a user requests an arbitrary plan in the past, and when the user changes the value of a variable of any subgoal by the execution of the plan, it indicates which variable and how many times the variable was changed. 2. The user support method according to claim 1, wherein a change history is recorded, and a variable having a larger number of changes is given a higher priority in accordance with the change history. 5. When a user's utterance is input in natural language, an adjunct word that modifies a plurality of conditions included in the utterance is extracted, and the adjunct word determines the user's intention for the necessity of each condition. The user support method according to claim 1, wherein the priority of changing the value of each variable is set to be higher as the adjunct word has a nondeterministic expression.