JPH0416826B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to the recognition and management of topics in communication between a computer and its users.

[Prior Art] In general, currently used communication between computers and their users does not use the concept or method of recognizing and managing topics, and therefore the man-machine interface of computers is It lacks flexibility and is difficult to use.

In contrast, we have developed a method that limits the target to database access using restricted Japanese, treats user input as one topic, and manages it, and implements and utilizes it as a program. A computer system has been proposed. FIG. 9 shows the configuration of this computer system, and FIG. 10 shows the processing flow of the controller. Further, FIG. 11 shows the short term memory (abbreviated as STM) format used by the controller.

In FIG. 9, 73 is a controller that controls the entire computer system, and the processing order of this controller corresponds to that in FIG. 10.
In this computer system, a sentence analyzer 80 analyzes a question 76 input by a user 75, and the STM7
4 in the format shown in FIG. One unit 86 of this format corresponds to one question. In the problem solving process 81, depending on the contents of the STM 74,
A user inquiry 82, confirmation 83, and answer 84 are performed.

In the computer system described above, questions input by users are topics of discussion. If an answer is given to the question, the topic is considered to have ended. The key 85 indicates the processing status of such a question, and includes processing/processed. Problem solving process 81 determines whether the question has sufficient information for database access.
If there is missing information, a user inquiry 82 occurs. At this time, the user can ask other questions instead of inputting an answer to the query. If another question is input, since the key for the previous question has not been processed, it is assumed that a change in topic has occurred, and the controller 73 is called recursively to proceed with the processing of the new question. Once the new question has been processed, the controller automatically returns to its original state. Here, the scene returns to the previous user inquiry.

[Problems to be solved by the invention] The following problems arose in the topic recognition and management method described above.

(1) Since "topic = user's question" and topic change is treated as another question input during question processing, the controller must be called recursively for processing, resulting in several steps of topic change. If this happens (a situation in which a topic change occurs while the topic is being changed), efficiency will drop significantly.

(2) In response to a user inquiry, it is not possible to re-enter the original question (revise), which limits its usage.

(3) Since the scope is limited in this way, there is no versatility.

(4) There is no limit to the size of STM, and it is possible to change the topic in any number of steps.
In communication with actual users, even if you return to a topic that was changed several steps ago one by one, it will impair the flexibility of the man-machine interface.

The present invention was made to solve these problems, and aims to realize a flexible man-machine interface that is easy to use for users.

[Means for Solving the Problems] The topic management method according to the present invention uses a dialogue pair dictionary containing dialogue pairs in which question sentence types and response sentence types are defined in advance, and a dialogue pair dictionary between a computer and a user. Dialogue pair management means that determines the establishment of a dialogue pair using a dialogue pair stack that maintains the process of communication as a history of dialogue pairs; It is equipped with topic management means for managing topics by manipulating the stack.

[Operation] In this invention, when a dialogue pair is not established, it is assumed that a topic change has occurred, and the dialogue pair stack is switched to a new, different dialogue pair stack.
Next, if a dialogue pair is established, it is assumed that the topic change has ended, and it is confirmed whether or not to return to the user. When returning, switch back to the original dialogue pair stack. When the dialogue stack that changes the topic after many steps is no longer available,
Instead, use the dialogue stack that was used previously. Depending on how many dialogue stacks you prepare will determine how many steps of topic change are allowed.

[Embodiment of the Invention] FIG. 1 is a block diagram of an embodiment of a topic management system according to the present invention. As is clear from FIG.
, a dialogue pair dictionary 5, and a dialogue pair stack group 6. Each of blocks 1 to 6 is a program that implements the function indicated by each name. Further, the arrows in FIG. 1 indicate the existence of data/requests exchanged between these blocks, and the direction of the exchange if they exist. 7 is a user input analysis result, 8 is an access result from the dialogue management means 3, 9 is an access request to the dialogue management means, 10
is an access request to the input/output processing means, 11 is a user input execution format, 12 is a user input execution result, 1
3 is a dialogue pair or message, 14 is a request for access to the dialogue pair stack, 15 is an access result from the dialogue pair stack, and 16 is a request for access to the dialogue pair dictionary.

Next, the operation of the above embodiment will be explained with reference to FIGS. 2 to 8. FIG. 2 is a flowchart showing the operation of the topic management means 2, FIG. 3 is a flowchart showing the format and example of dialogue vs. dictionary, and dialogue vs. dictionary access, and FIG. 4 is a flowchart showing the operation of the dialogue vs. management means. , FIG. 5 is a flowchart showing the operation of the dialog pair stack group, FIG. 6 is a flowchart showing the operation of input/output processing, FIG. 7 is a flowchart showing the operation of the user input executable, and FIG. 8 is a flowchart showing the operation of the user input executable. FIG. 4 is a schematic diagram of an example of a user's interaction with the embodiment, and a change in the state of the interaction versus the stack at that time;

First, the input/output processing means 1 analyzes the input from the user, and the user input analysis result 7 is sent to the topic management means 2.
[Step 66 in Figure 6]. The input/output processing means 1 waits for the next state designation [70].

The topic management means 2 accesses the dialogue pair dictionary 5 based on the user input analysis result 7 [FIG. 1 16],
Dialogue pair 13 is obtained from dialogue pair dictionary 5 [Fig.
8, 29, 31]. At this time, as dialogue pair 13,
If there is no corresponding user input analysis result 7, the dialogue pair dictionary 5 returns an error message 13 to the topic management means 2 [30].

When the topic management means 2 receives the error message 13 [“NO” side in step 17 in FIG. 2],
The input/output processing means 1 sends it to the input/output processing means 1 [see 67], and after displaying the message [Fig. 6 67], the input/output processing means 1 receives the user's input again [see 6.
6].

When the dialogue pair 13 is obtained from the dialogue pair dictionary 5 [“YES” side in step 17 in FIG. ]. If the dialogue pair stack is empty at step 20 in FIG.
to accept user input again [same 2
5], continue the dialogue with the user. If the dialogue pair stack is not empty in step 20, it is checked whether the previous user input was reworded [22].

This is determined to be a rewording if the previous user input analysis result 7 and the current user input analysis result 7 are compared and the central object and the action content for that object match. In the case of rewording, a user input execution form 11 is created from the user input analysis result 7 and passed to the user input execution body 4 in step 24 .

As shown in FIG. 7, the user input executable 4 executes the user input execution format 11 (71) and sends the execution result to the input/output processing means 1 (72). In the input/output processing means 1, as shown in FIG. 6, after formatting and displaying the user input execution result 12 (68), the execution result 12 is sent to the topic management means 2 (69).

The topic management means 2 performs the same procedure for the execution result 12 as it does for the user input analysis result 7. If it is determined that it is not a restatement,
The topic management means 2 is the dialogue management means 3 and the dialogue management means 3 is the dialogue management means 3.
[Figure 2, 23] and confirm whether or not a dialogue pair is established [Figure 2, 26]. If so, the input/output processing means 1 is put into a state of accepting input again [same 25], and the dialogue continues. If this has not been established, it is assumed that a change of topic has occurred, and the dialogue pair 13 is sent to the dialogue management means 3, and a stack change request and a registration request are sent [ibid. 27]. after that,
The input/output processing 1 is set to accept new input [25], and the dialogue continues again.

The dialogue pair management means 3 processes requests sent from the topic management means 2 using the dialogue pair stack group 6 [FIG. 4, 32 to 47]. The registration request requests the dialogue pair stack group 6 to register the dialogue pair 14 [45] and obtains the result 15 [46]. If the result is successful, the process returns to the topic management means 2 [44]. If it is not successful, the input/output processing 1 displays a message informing that the dialogue has been interrupted [47].

The success or failure of the registration request is determined by the dialog pair stack group 6. Attempts to register deeper than the preset stack depth (max) will fail [FIG. 5, 61-65]. The depth of the stack is, for example, 5 to 7, corresponding to the short-term memory (temporary memory, which is also imitated by the above-mentioned STM) of a normal human being. Checking whether the stack is empty is also performed by checking whether the depth of the stack is 0 in the dialog pair stack group 6 [FIG. 5, 55, 56].

To change the stack, use dialogue to stack group 6 to switch the current stack ID to the next stack ID,
This is achieved by returning that ID [Figure 5 57~
60]. For example, there are three stacks for switching. This means that topic changes may occur up to three levels deep at once. This also depends on the relationship with short-term memory.

Confirmation of the establishment of a dialogue pair is performed by extracting the latest dialogue pair from the dialogue pair stack group 6 [FIG. 4, 33] and comparing it with dialogue pair 13 [FIG. 4, 34].
When the response classification ID [Fig. 3] included in the latest dialogue pair extracted from dialogue pair stack group 6 matches the sentence pattern sentence type [Fig. 3] of dialogue pair 13 corresponding to the user's input sentence, the dialogue pair is is determined to have been established. If it is not established, that fact is returned to the topic management means 2 [FIG. 4 41]. If it is true, pop up one stack in dialogue pair stack group 6 [Double 3]
5].

When it pops up and becomes empty [Ibid. 36], it means returning from the topic change, so I send a stack recovery request to dialogue stack group 6 [Ibid. 37], and change the previously used stack to the current one. Stack [FIG. 5 49-52], and then return success to the topic management means 2 [FIG. 4 38]. If it pops up and is not empty, it is considered that the series of dialogues is still continuing, so just return success to the topic management means 2.

In the manner described above, the dialogue progresses between the user and the embodiment. FIG. 8 schematically shows an example of dialogue in the embodiment and the accompanying change in the state of the dialogue pair stack.

In the above embodiment, the topic management means 2, the dialogue pair management means 3, and the dialogue pair stack group 6 are each realized like the former subroutine program, but
This division is arbitrary depending on the description language, etc., as long as the topic management method described above is realized. Further, the depth of the dialogue pair stack and the number of dialogue pair stacks for switching can be changed depending on the function of the user input executable 4 of the embodiment. Of course, the user input executable 4 can be in any field. Furthermore, various techniques can be applied to the input/output processing 1 as long as the operation shown in FIG. 6 is guaranteed.

[Effects of the Invention] As described above, according to the present invention, the conversation pair dictionary is used to recognize the change in topic, and the topics are managed by switching the conversation pair stack, which improves processing efficiency. It is possible to provide a man-machine interface that has versatility and is more natural to the user without degrading the performance, is not limited in usage, and is versatile.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram of an embodiment of the topic management method according to the present invention, FIG. 2 is a flowchart showing the operation of the topic management means, and FIG. 3 is a format and example of the dialogue pair dictionary, and a dialogue pair dictionary. Flowchart of dictionary access; FIG. 4 is a flowchart showing the interaction and operation of the management means; FIG. 5 is a flowchart showing the interaction and operation of the stack group; FIG. 6 is a flowchart showing the operation of the input/output processing means. FIG. 7 is a flowchart showing the operation of the user input executable, and FIG. 8 is a schematic diagram of an example of interaction between the user and the embodiment, and a change in the state of the interaction versus stack at that time. Figure 9 is a configuration diagram of a computer system that implements the conventional topic management method.
Figure 10 shows the processing flow of the controller, the first
Figure 1 shows the format of short-term memory used by the controller. In the figure, 2 is a topic management means, 3 is a dialogue pair management means, 5 is a dialogue pair dictionary, and 6 is a dialogue pair stack group. Note that the same reference numerals in each figure indicate the same or corresponding parts.

Claims (1)

[Claims] 1. In a computer system that executes processing through communication between a computer and a user, there is a dialogue pair dictionary that stores dialogue pairs that define the correspondence between question sentence types and response sentence types, and a dialogue pair dictionary that stores dialogue pairs that define the correspondence between question sentence types and response sentence types, and the analysis results of question sentences. A plurality of dialogue pair stacks store dialogue pairs retrieved from a dialogue pair dictionary using the obtained question sentence type as a key, and a dialogue pair stack that stores dialogue pairs retrieved from a dialogue pair dictionary using the response sentence type obtained as a result of analysis of response sentences as a key. and the dialogue pairs stored in the dialogue pair stack, determine whether the correspondence between the question sentence and the response sentence is established, and if the judgment results match, remove the dialogue pair from the dialogue pair stack. 1. A topic management method in a computer system, comprising a dialog pair management means and a topic management means for switching a dialog pair stack to another dialog pair stack when the determination results do not match.