JPS6146530A - Conversation system - Google Patents

Abstract

PURPOSE:To obtain an easy-to-use, efficient conversation system wherein an inference is made by displaying the progress of the proper inference in the middle of the inference like a soliloquy. CONSTITUTION:A request input signal 8 from a user is inputted to a buffer 10 after being converted by an inference part 1 to usable expression form. The inference part 1 advances an inference making process according to various knowledge data in internal, short-period, external temporary knowledge memories 3, 4, and 5 and the inference rule in an inference rule memory 2. The inference making process is advanced while the progress of the inference making process and halfway information on the inference are recorded in an intermediate state memory 7. An output data generation part 11 monitors the memory 7 to set a goal and makes an inference for a specific group of items and then edits the contents in document form. Then, an output signal composition part 12 converts the result into an output signal like a soliloquy and outputs a signal 13.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a conversation system, and particularly to an improvement of a system that responds while making inferences.

(Background of the Invention) In conventional systems, such as the well-known MMCIN, the system requests the user to input necessary information, makes inferences based on this information, presents the results to the user, and responds to the request. If so, it would be revolutionary in that it would be possible to explain the reasons for that inference.

However, as is clear from human conversations,
Even in cases where there is a difference in position between the two parties, it is rare for the purpose, way of thinking, and necessary information to be clear in advance, and it is common for both parties to arrive at a conclusion through repeated thinking and reasoning. , it is not uncommon for the purpose to change during the process.

In addition, in advanced systems with a huge number of inference rules, the processing time becomes longer, which means more time for the user.

This creates a problem in that the conversation becomes long and the conversation is effectively interrupted.

[Purpose of the invention]

The purpose of the present invention is to eliminate such drawbacks, provide easy-to-use
and to provide an efficient conversation system.

[Summary of the invention]

That is, in the system according to the present invention.

He is characterized by the fact that he sometimes expresses the progress of his reasoning as if he were talking to himself. As a result, if a user interrupts the system in the middle of an inference process that is clearly incorrect, or if a new inference that the user did not anticipate is expected, the system can accelerate the development in that direction. , it is possible to stop the expansion to others. Furthermore, when the inference is completed, the progress of the inference is known without requiring a new explanation.

[Embodiments of the invention]

In FIG. 1, an input signal 8 indicating a user's request is analyzed by an input signal processing unit 9, and converted into an expression format that can be used by an inference unit 1.
It is stored in the buffer 10.

The input signal takes the form of natural sentences such as keyboard input or voice signals, and the input signal processing unit detects important keywords from among them and performs processing to interpret the user's request from the mutual relationship between them. However, since the specific manner in which these processes are carried out has nothing to do with the essence of the present invention, detailed explanation thereof will be omitted here.

The inference unit '1 uses the data in the input data buffer 10 and the internal knowledge memory 3. Short-term knowledge memory 4. Interprets the user's request based on the various knowledge in the temporary external knowledge memory 5 and the inference rules in the inference rule memory 2, and performs various inference processes in response to the request. The internal knowledge memory 3 stores in advance the specialized knowledge of the system.

In addition, the short-term knowledge memory 4 stores temporary knowledge obtained through the conversation between the system and the user, and what pronouns such as "that" refer to. From information such as what the user is talking about, to information such as what the subject of the conversation is, information that the system needs to judge the user's request, such as information obtained by requesting the user. Stored.

The temporary external knowledge memory 5 is information necessary for processing in order to respond to user requests, and is temporary knowledge obtained from outside the system. External knowledge includes information such as various databases. sell.

The inference unit 1 proceeds with inference using the above-mentioned various kinds of knowledge in accordance with inference rules, and proceeds with the inference while recording the inference process and information during the inference in the intermediate state memory 7. In the intermediate state memory 7, hypothetical goals and the like of goal-based reasoning are also recorded.

The output data creation unit 11 monitors the intermediate state memory 7,
When a goal is set and when inferences are made for a certain set of items, the content is edited into a text format and output signal synthesis unit 12 as if it were a “hitorigoto”.
is converted into an output signal 13 and output. Similar to the input section, the editing method for text and the format of the output signal (character display on a display, synthesized speech), etc. are not essential parts of the present invention, so a detailed description of the method will be omitted. ``The output data creation section 11 outputs information about which data in the intermediate state memory the output content corresponds to and records it in the synchronization processing section 14 at the same time as outputting.

When the user's response to the output (input 8) is detected, the synchronization processing unit 14 correlates which output the input was detected after, so that the inference unit 1 can determine. The user's response (input 8) is processed by the input signal processing unit 9,
The information is input to the input data buffer 1o as information such as confirmation or denial. , the input information may be ambiguous confirmations, denials, or additional information provided by the user.

When the user's response is negative, the reasoning unit 1 first stops reasoning in that direction and moves on to reasoning in another direction. If it is abject, we will proceed with more detailed reasoning in that direction and will not make inferences based on other rules or situations.

Next, an example of a conversation according to the present invention will be given. This is an example related to a database.

User: ゛′I want data about my physique. ″Device =
(Talk to yourself while sequentially inferring the content of the word physique using knowledge information) “Physique includes height, weight, chest circumference,...P'' User = (
(Listen to the device's monologue and talk to the device) ``Height and weight are all you need.'' Device: Based on the user's words, chest circumference or below, sitting height.

The inference processing for the items . As a result, wasted processing time for various inferences, data processing, etc. required for processing each item below the chest circumference is eliminated, and subsequent processing can be carried out efficiently.

〔Effect of the invention〕

As explained above, according to the present invention, in a conversation system that involves inference, inference tasks are sequentially output during inference, and the user's reaction to the questions is received and received. By limiting the possibility of reasoning, or conversely, by using the reasoning process, it becomes possible to direct decisions that the user himself has not clearly given, and by limiting the range of possibilities of reasoning, the system This makes it possible to create a system that efficiently achieves its goals.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an embodiment of the present invention.

Claims (1)

[Claims] 1. An input section, an inference section, a knowledge storage section, an inference rule storage section, a section that records the progress of the inference, a recording section that records intermediate results during the progress of the inference, and a section that outputs information on the intermediate results. , a synchronization processing unit that associates the timing of the output from the output unit and the input from the input unit with information in a recording unit that records the intermediate results, and a synchronization processing unit that matches the timing of the output from the output unit and the input from the input unit; A conversation system characterized by outputting from the output unit and performing subsequent inference processing based on information of a reaction input to the output.