JPS62169220A - Man-machine interface system - Google Patents

Abstract

PURPOSE:To input information in highly natural conversation by generating and outputting a system response without outputting inquiry contents repeatedly when information of inquiry contents is inputted. CONSTITUTION:A man-machine interface in a natural language consists of an input part 1, an input interpreting part 2, a conversation control part 3, a task processing part 4, a response generating part 5, and an output part 6. The response generating part 5 discriminates whether contents of input information from a user are related to an inquiry generated and outputted as the system response in conversation to the user or not and controls generation and output of the next system response in accordance with the discrimination result. If information related to inquiry contents is inputted, the next inquiry sentence is generated and outputted as the system response without repeating these contents of input information. Only when information different from inquiry contents is inputted, input information contents are outputted repeatedly, and the next inquiry sentence is generated and outputted thereafter.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a man-machine interface method that allows natural conversation between a user and a system.

[Technical background of the invention and its problems]

Recently, various information processing systems have been developed and are widely used in various industrial fields. Recently, instead of inputting information using a keyboard or tablet into this type of information processing system, voice recognition technology is being used to input information directly by voice uttered by the operator (user). This is also being attempted.

Now, in order to effectively operate this type of information processing system, it is important to realize an interface that allows interactive information exchange between the system and the user. For example, it is desired to realize an interface that allows users to input necessary information into the system while making interactive requests and inquiries in a free format, without being bound by the responses from the system. Moreover, in order to expand the user base of information processing systems, it is desired to realize a human-machine interface using natural language.

By the way, in conversations between humans, generally speaking, ■the truth,■
It is done in the minimum necessary amount, ■ while being related to each other, and ■ speaking in easy-to-understand expressions. For this reason, there are many abbreviations.

Although this principle of conversation is not necessarily followed on a daily basis, it is undeniable that conversations that deviate from this principle become extremely unnatural and difficult to understand.

On the other hand, when carrying out a conversation between a human and a transport machine via a man-machine interface, information containing many abbreviations is input interactively, just like in a conversation between humans.
It becomes extremely difficult for the system to interpret its meaning. For this reason, various studies have been made on how to correctly interpret the meaning of input information including abbreviations with high precision.

On the other hand, determining what kind of system response to generate in response to input information is a very important issue in realizing natural conversation. However, the reality is that conventional systems pay little attention to this point.

The system exclusively generates and outputs question texts that prompt information input as system responses. Or each time information is entered.

It generates and outputs a system response that prompts confirmation of the contents of the input information, and when the confirmation is obtained, generates and outputs the next question.

However, when the conversation between the user and the system proceeds in this manner, the following problems occur. For example, when information is input sequentially in response to question sentences that are output one after another from the system side, there may be anxiety as to whether or not the input information is correctly interpreted by the system. Furthermore, if the user is prompted to confirm the input information after inputting information, it is undeniable that the task of inputting information through conversation becomes extremely troublesome. Furthermore, the system response that prompts for confirmation unnecessarily increases the redundancy of the conversation, and the content of the question is likely to be buried in other information, causing problems such as making it difficult to understand the meaning of the system response.

Therefore, it is strongly desired to realize a man-machine interface that enables natural conversation with an information processing system.

[Purpose of the invention]

The present invention was made in consideration of these circumstances, and its purpose is to provide a man-machine that controls the generation and output of system responses to input information and enables information input through highly natural conversation.・The purpose is to provide an interface.

[Summary of the invention]

The present invention provides a man-machine interface that includes an input interpretation section that analyzes input information, and an output response section that generates and outputs a system response to the input information according to the analysis result. This feature is characterized in that a system response is generated that does not repeatedly output the contents of the system response or information that has been output as a system response in the past.

In particular, when information regarding the content of a question is input, a system response that does not repeatedly output the content of the input information is generated and output, and when information different from the content of the question is input, the content of the input information is output. When a user repeatedly outputs and responds, the next system response other than the question prompting the user to input the input information is generated and output.

〔Effect of the invention〕

Thus, according to the present invention, when response information to a question from the system is input, the input information is not repeatedly output and the next question is issued, so the original question content is buried in other information. It becomes easier to hear.

Furthermore, since there is no repeated expression of the input information and no questions prompting the user to input the same information, it can be known that the input information has been correctly interpreted.

Therefore, the feeling of anxiety in dialogue can be eliminated.

Furthermore, if information different from the content of the question is input, the content of the input information is repeatedly responded to, so it becomes possible to know from this point on that the input information has been accepted by the system. Furthermore, if the information content that has been repeatedly responded to differs, it is also possible to immediately correct it.

Therefore, unnecessary repetitive responses in conversation are avoided, redundancy is reduced, the troublesomeness of conversation is reduced, conversation with the system is carried out naturally, and necessary information is efficiently input into the information processing system. This has great practical effects, such as making it possible to

(Embodiment of the Invention) Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic diagram of a man-machine interface using natural language to which the embodiment method is applied.

The input section 1 consists of, for example, a keyboard and a microphone, and supplies input information input as a code string via the keyboard or input information input by voice via the microphone to the input interpretation section 2.

The input interpreter 2 executes syntactic processing and semantic processing on the input information to interpret the contents of the input information. Then, a semantic interpretation representation of the input information content is created and provided to the task processing unit 4' via the dialogue control unit 3. This semantic interpretation, for example, is based on a semantic network (
It is expressed using a case frame, which is a type of frame.

The task processing unit 4 executes task processing for each field of work to be performed by the information processing system, such as seat reservation work and information search.

This task processing requires the system to provide a semantic representation of the information to be presented to the user.

The response generation section 5 generates response information to respond to the user according to the semantic expression obtained from the task processing section 4 via the interaction control section 3, and outputs this via the output section 6. ing.

Here, I will briefly explain the process of interpreting input information.
First, the input character string forming the input information inputted from the input unit 1 is divided into phrase units, and the phrase structure of the input information is determined.

The part-of-speech structure of a clause is given, for example, as (prefix) * + (independent word) + (attached word) *. Syntactic analysis of input information is achieved by analyzing whether the input information (input sentence) is composed of one or more clauses. Note that the * above indicates that the part of speech is repeated zero or more times.

In this syntactic processing, for example, words prepared in advance in a dictionary are matched sequentially from the end of the input string, starting from the longest matching layer, to find a string of clauses (words) that correspond to the entire input string. This is done by In this case, it goes without saying that the matching process is performed while taking into account the usage of the predicate. Furthermore, if the matching process fails during the process, it goes without saying that the matching process can be continued while backtracking as necessary.

Furthermore, if the format of the input information is predicted for the question text etc. output from the system side, it goes without saying that the input information may be interpreted and processed according to the predicted input format. . Specifically, when the system makes a query regarding time, the user can use the shorthand expression
A case may be considered in which only llIN data is input. In such a case, if numerical data is input, it may be interpreted as indicating time information, depending on the prediction information of the input format.

In this way, even if the input information includes an abbreviation, its semantic information is determined as appropriate and the input information is interpreted.

By the way, the response generation unit 5 basically uses, for example, the second
The system response is generated according to the steps shown in the figure.

That is, step a) generates and outputs a question related to the content as a system response, and when a response from the user is given as input information (step b), the content of the input information is first interpreted semantically (step C). ).

Then, it is checked whether the content of the input information is related to the content of the question (step d).

According to this determination result, if the content of the input information is related to the previous question, it is determined whether there are any other questions remaining (step e), and the next question text is generated (step f).

On the other hand, if the content of the input information differs from the content of the question, the content of the input information is first repeatedly output as a response to prompt the user to confirm (step q), and then other questions are left. Step e) and generates the next question sentence (Step f) - In other words, in the response generation unit 5, in the dialogue with the user, the content of the input information from the user is related to the question generated and output as a system response. It determines whether or not it is true, and controls the generation output of the next system response according to the determination result.

Specifically, when information related to the question content is input, the next question text is generated and output as a system response without repeating the input information content, and only when information different from the question content is input. , the input information content is repeatedly output, and a model question text is generated and output.

By controlling the generation and output of system responses in this manner, unnecessary repetitive response outputs of the input information content from the system side are omitted, and the naturalness of the conversation is ensured.

Next, a specific example of the system response generation output will be explained.

Figure 3 shows that in response to the system's question, ``From what time do you want to make a reservation?'', the user asks, ``From 13R to 1.
This figure shows the flow of the system response generation procedure when the information "Until 4 o'clock" is input.

Here, the process indicates a list of information necessary for performing the task, and in the case of [reserved task processing] in this example, it is given as follows, for example.

((from-1111e?) ([0-口a+e?))
.

(place?), (name?)
, -...Also, RL is a semantic expression for generating a system response, and RL is a history list of semantic expressions generated in the past as a system response.

Further, UL is a semantic representation of input information, and tJLL is the past 111fi of the above LJL.

When processing a reservation for a meeting, the month and day information for each reservation is input, and when the next time an inquiry is made about the reservation time, the task list TL is as follows: ■ ((from-time) ) (to-ti
Me? ))■ (place?) ■ (name?) is left. At this time, when "1r13:00 to 14:00" is given as input information, the content is interpreted and, for example, a semantic expression tJL having the following semantic network structure is obtained.

(nil (from-time 13) (to-
tille 14)) In this example, since there is no predicate in the input sentence, (nil) is used in the semantic expression.

Then, task processing is performed on the semantic representation of this input information, and the input information contents are expressed as ((from-time?) (to-time?) in the task list TL.
? ) ) ) is required.

As a result, the corresponding item is deleted from the task list TL,
The next question text is generated using the remaining item information. In this example, since the information regarding the location has been input, and by referring to the information of the conference rooms that are vacant during that vI period from the database, the semantic expression for the system response, for example, (( (aku (*teinei) ) (or(pla)
ce 4A) (place 4B) ) ) ((y
oyaku (*teinei) (*oimon)
(place ? )), and according to this semantic expression TL, ``Conference rooms 4A and 4B are vacant.

Which one would you like to book? ” will be output as a system response.

Therefore, in this case, for example, upon receiving the input information, a system response such as ``The reservation is from 13:00 to 14:00'' is not repeatedly output.

In the course of the conversation, for example, when information about the month and day to reserve a conference room is input, the system may know information about meetings that are open on that month and day. In such a case, for example, the r4A conference room will be available from 10:00 to 16:00 as a system response. ”
You may also output .

In such a case, the above information is recorded as the information BL presented to the user by the system, and for example, the fourth
As shown in the figure showing the system response generation procedure, the next system response may be generated by removing previously presented information content from the task list TL.

On the other hand, due to the nature of the conversation, for example, the question ``From what time to what time do you want to make a reservation?Even though the system responded ``J'', the user asked ``R4A conference room.'''' may be entered. In such a case, the system response generation output is controlled as shown in FIG. 5, for example. In this case, first, the task list TL is checked for (nt+ (place 4 A)'), which is a semantic representation of the input information. and task list T
Since the corresponding item exists in the task list TL, the input information is taken in as necessary information and the corresponding item is deleted from the task list TL.

After that, since the input information is different from the question content,
For example, ((desu (*kakunin) ) (p
lace 4 A) )). Then, referring to the task list TL, in this case, information regarding the time has been input, and the system information shows that the task is free from 11:00 to 16:00. For example, "The reservation is for the 4A meeting.

It is open from 11:00 to 16:00.

From what time to what time do you want to make a reservation? J A system response is generated as follows.

That is, when information different from the content of the question is input, the content of the input information is repeatedly output and confirmed, and the next question is generated accordingly.

In this manner, in the man-machine interface according to the present invention, it is determined whether or not the content of the input information is related to the content of the question, and only when the content of the input information is different from the content of the question, the content of the input information is repeatedly output as a response. be done. If information matching the content of the question is input, the content of the input information is not repeated, thereby generating a system response that omits useless response content. In particular, since information that has already been output is not repeatedly output, it is possible to have a natural conversation with less redundancy.

Since the system response does not have to be repeated as a conversation, the system response can be made easier to listen to, and information can be entered as appropriate without being restricted by the system response, which has many practical benefits. This produces a certain effect.

Note that the present invention is not limited to the embodiments described above. For example, the semantic representation structure of input information and system responses, their interpretation processing, etc. may be determined according to the interface specifications. Furthermore, the generation algorithm of the system response, that is, which item in the task list should be used to generate the question text, can be determined according to the specifications. Furthermore, the form of task processing is not particularly limited. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

Figure 1 is a schematic configuration diagram of a man-machine interface according to an embodiment of the present invention, Figure 2 is a diagram showing the basic control flow for system response generation, and Figures 3 to 5 are respectively FIG. 3 is a diagram showing a specific flow of system response generation. DESCRIPTION OF SYMBOLS 1... Input unit, 2... Input interpretation unit, 3... Dialogue control unit, 4... Task processing unit, 5... Response generation unit, 6
...Output section. Applicant's agent Patent attorney Takehiko Suzue Figure 1 Figure 2

Claims (4)

[Claims] (1) An input interpretation unit that analyzes input information, and an output response unit that generates and outputs a system response to the input information according to the analysis result, and the output response unit repeatedly outputs the contents of the input information. This is a man-machine interface method that is characterized by generating system responses that never occur. (2) The man-machine interface method according to claim 1, wherein the output response section generates and outputs a system response that does not repeatedly output information that has been output as a response in the past. (3) An input interpretation unit that analyzes input information; and an output response unit that generates and outputs a system response to the input information according to the analysis result; When the input information is input, a system response is generated that does not repeatedly output the content of the input information, and when information different from the above question content is input, the content of the input information is repeatedly output and output, and then the next system response is generated. A man-machine interface method that is characterized by generating and outputting. (4) The man-machine interface according to claim 3, wherein, when information different from the question content is input, the output response unit generates a system response other than the question content to prompt the input of the information. method.