JPS6033636A - Inferential device - Google Patents

Abstract

PURPOSE:To obtain a high-speed inferential device for minimizing a rule storing capacity by using thesourus knowledge (associated knowledge regarding a similar conception), and generalizing a rule. CONSTITUTION:When conception names A, B and a relation name C are inputted to an inference operating mechanism 1 through data lines 011, 012 and 013, a rule storage mechanism 2 receives a rule number (n) serving as a command signal sent from a data line 121, sends out values of the first field and the second field in the rule number (n) to data lines 231, 232, respectively, and sends out a value of the second field from a data line 122. A thesourus storage mechanism 3 receives signals 131, 132 from the inference operating mechanism 1, and sends out a bit address 0/1 in an address of a signal 131, a signal from the data line 231, a signal 132 and a signal from the data line 232 in a bit array, from data lines 133, 134 respectively.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a sieve speed inference device in an intelligent information processing device.

Conventionally, the method for realizing logical reasoning mechanisms is the decomposition proof algorithm (Reference Chan & Lee: 1972, Acad
emic press″8y+nbolicLogic
and Mechanical the Or-e'r
However, as the number of rules increases, the inference efficiency decreases significantly. One of the drawbacks of this is that it uses a very general inference method, so there is no way to avoid the fact that in the worst case, even if you use a non-deterministic machine, you may not be able to run in polynomial time. Second, there is no 71-ware version.

Usually, in language understanding, consultation systems (diagnosis systems), decision support systems (decision support systems), etc. that are included in the scope of intelligent information processing, the semantic relationships between concepts that occur in individual processing are discovered. This is very important. For example, in natural language understanding, it is an inference about what kind of relationship X is possible between word concepts A and B, or in a consultation system, what kind of situation X is it that causes event B to be derived from event A? It is essential to make inferences in the decision support system, such as under what economic conditions the investor will receive profit B.

For such simple questions, it is possible to quickly reason and obtain answers without using general decomposition proof methods. However, if we describe the above A, B, and X individually, the basic concepts A, B.

As the number of X increases, the storage capacity for storing it increases dramatically (lAl=n,, IBI=n
If b, l Xl = nx, then n, Xnb
Xnx = n': n = Max (ns, n
b, nx)). However, if each concept has a similar meaning to each other (for example, in natural language understanding, semantically similar words can be identified relatively easily), then all individual a( Logical rule A that holds true
By describing rules using concepts that represent similar concepts without writing -B-X, an explosion in the number of rules can be prevented. This associative knowledge about similar concepts is called a thesaurus, and by using a thesaurus, it is also possible to generalize the inferences based on the rules ζ. In the following examples, the thesaurus is used as a superordinate/subordinate relationship between concepts (for example, a baby is a subordinate concept for humans,
Biology is considered a superordinate concept).

According to the present invention, a rule storage mechanism that stores inference rules;
A high-speed memory necessary for storing inference rules, consisting of a thesaurus storage mechanism that stores associative knowledge, and an inference calculation mechanism that applies the inference rules via the associative knowledge that describes the superior/subordinate relationships of concepts to perform inference. A reasoning device is obtained that minimizes capacity.

Next, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing one embodiment of the present invention,
FIG. 2 is an explanatory diagram of the inference calculation mechanism, FIG. 3 is an explanatory diagram showing the data format in the thesaurus storage mechanism, and FIG. 4 is an explanatory diagram showing the format of rules in the rule storage mechanism.

In Figure 1, data line 01+, Olt -013
The concept master, concept name B, and relation name C are connected to the inference calculation mechanism 1 through
When entered, the rule search command 12. is the rule storage mechanism 2
and the relation code in the selected rule is 12. The information is sent to the inference mechanism 1 via. Further, the concept code is sent to the thesaurus storage mechanism 3 via data lines 23+ and 232.

The rule storage mechanism 2 receives a rule number n, which is a command signal sent from the data line 12+, and stores the values of the first and second fields in the rule code n on the data line 23.
+, 232, respectively, and the value of the second field is sent to data line 12. Send from If the rule number n exceeds the rule number K, a control signal is sent from the data line 123.

The thesaurus storage mechanism 3 receives signals 13s and 13z from the inference calculation mechanism, and in the bit array in the third Nl (described later), (signal 133, signal from data line 23°), (signal 13□, data line 232) bit value (0/l) at the address of each data &! 133 and 134.

FIG. 2 is a diagram showing the configuration of the inference calculation mechanism 1. In the figure, 101-10t, 104 are coating mechanisms that map symbols to integers, and 10. is a decoding mechanism that maps integers to symbols; 10. is an integer comparison mechanism, 106 is A
The ND circuit 107 is a counter. The data line 0
11.012 The concept name manually entered is converted into an integer by the cope ink mechanism 101.102, and triggered by the input of the signal line 114, the data a], 3+,
13z to the thesaurus storage mechanism 2 I7. At this time, the coded integer is the data line 10+
.

The first signal is retained within the coding mechanism unless a newer signal than 102 is input. The data line 01. The relationship name sent from the Co-op Ink mechanism 104 is converted into an integer and the ratio is +9. Mechanism 10. The contents of the data line 122 are compared with the contents of the data line 122, and if they do not match, the counter 10
A control signal 1,1 is sent to increase 7 by 1.

If they do not match, a signal sending command signal is output to the coding mechanisms 101 and 102 via data lines 1 and 4.

The data line 01. When data is input from 7, the counter 107 sets itself to 'l' and the data line 12
, the contents of the same counter 10□ are stored in the rule storage mechanism 2.
Send to. The counter 10 is connected to the data line 11.
When it receives a signal from 1*115, it increases its value by 1, and when it receives a signal from the data line 3.2s, it stops.

The AND circuit 106 receives signals from the thesaurus storage mechanism 3 via the data lines 13g and 134, calculates the product, and sends the value to the decoding mechanism 103 as data II Lt.

Front 81: The decoding mechanism 103 is connected to the data line 11.
2J-11" is received, the integer value output from the data line 122 is decoded from the integer value l"j to the data line O14, and the decoded value is output from the data line O14.

In Figure 43, the thesaurus storage mechanism has N lines.

Memorize the data array of N columns (the 7 concept names correspond to the row numbers and 9 column numbers without any excess or deficiency due to coding.(i,j)1<i
, if j<N is 1, there exists a subordinate-superordinate relationship between the concept disease coded i and the concept warmth coded j (i is J's sub41!V, Nen).

In FIG. 4, the rule storage mechanism records l rules, each consisting of three fields. Same Jll.

In the rain, integers that code each concept disease are stored in the first field and second field, and integers that code the person concerned are stored in the third field.

According to the present invention, it is possible to obtain a high-speed inference device that generalizes rules 7 by using thesaurus knowledge, thereby minimizing the rule storage capacity.

In this embodiment, the related parties are not generalized due to thesaurus knowledge, but if necessary, Seki Atsushi himself can be generalized easily. Furthermore, although superior/subordinate relationships are used as thesaurus knowledge, other logical transition relationships can be used instead of superior/subordinate relationships.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a first embodiment of the present invention, FIG. 2 is an explanatory diagram of the inference calculation mechanism 1, FIG. 3 is an explanatory diagram of the data storage format in the thesaurus storage mechanism 3, and FIG. 4 is an explanatory diagram of the data storage format in the thesaurus storage mechanism 3. 2 is an explanatory diagram of a data storage format in the search mechanism 2. FIG. In the figure, 1 is an inference calculation mechanism, 2 is a rule storage mechanism, 3 is a thesaurus storage mechanism, and 10. ．． 10□, 104 is a coating mechanism, 10Y is a decoding mechanism, 10. is a counter, 1
05- is a comparison mechanism, 101. (ANI) circuits respectively. 123N

Claims (1)

[Claims] An inference device in an information processing device performs inference using a rule storage mechanism that stores inference rules, a thesaurus storage mechanism that stores associative knowledge, and rules on the rule storage mechanism and associative knowledge on the thesaurus storage mechanism. and an inference calculation mechanism that executes the rules using the associative knowledge.
An inference device characterized by being more general, minimizing the same rule storage equivalence, and speeding up inference.