JP3124389B2 - Hypothetical reasoning device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a hypothesis inference apparatus based on a production system.

[0002]

2. Description of the Related Art Conventionally, in a hypothesis inference apparatus that generates a hypothesis world by an inference function based on a production system, a parent-child relationship is maintained between a hypothesis world generated by applying a production rule, If there is more than one hypothetical world that successfully matches the antecedent part of the rule, select the parent's hypothetical world closest to the root and use the data existing in that hypothetical world to determine the consequent part of the rule. Had to run.

[0003]

However, such a conventional hypothesis inference apparatus has the following problems.
For example, in a production system in which data is frequently deleted, or in a production system-based inference device in which there is a successful match due to the absence of certain data, the inference that the hypothesis world of the parent is always selected and executed. In the hypothesis world merge function that generates a single hypothesis world by merging multiple hypothesis worlds, data is especially deleted if there is a parent-child relationship between the hypothesis worlds. At times, there was a problem that ambiguity occurred in the existence of data.

In other words, the hypothesis world generated by hypothesis inference does not always require a parent-child relationship. In an actual application, the hypothesis generation is performed in order to avoid a case where the existence of data becomes ambiguous. In some cases, it is appropriate that the hypothesized worlds are subject to inference independently of each other, and that the existence of data is clear for each hypothetical world.

The present invention has been made in view of such a conventional problem, and there is no parent-child relationship between each hypothesis world generated by a hypothesis generation command in the consequent part of the production rule, and the inference engine collation. , Conflict resolution and operation phases are performed independently in each hypothetical world where data exists, and when matching is not successful due to the absence of a common hypothetical world of data to be matched, multiple hypothetical worlds It is an object of the present invention to provide a hypothesis inference apparatus which can generate a new hypothesis world in which such matching is successful by automatically merging, and push inference.

According to the present invention, when a hypothetical world is automatically generated, when a hypothetical world corresponding to a contradiction rule is generated, the hypothetical world is automatically eliminated, and only a valid hypothetical world can be efficiently generated. It is an object to provide an inference device.

[0007]

A hypothesis inference apparatus according to the present invention comprises: a rule storage unit for storing a production rule having an antecedent part and a consequent part; and a hypothesis described in a consequent part of the production rule. A hypothesis world storage unit that stores a hypothesis world generated by a hypothesis generation command for standing up, a premise part of a production rule given from the production rules stored in the rule storage unit, and a hypothesis world storage unit An inference engine that performs an inference operation of independently collating each hypothesis world, resolving conflicts, and generating a hypothesis world having no parent-child relationship by a hypothesis generation command described in the consequent part of the production rule; and When a certain production rule is compared in the hypothesis world stored in the hypothesis world storage unit, Is not successful, generates a new hypothesis world that matches by taking the union of the hypothesis worlds by merging a plurality of hypothesis worlds, and stores the hypothesis world storage unit in the automatic merge function unit. It is provided with.

[0008] The hypothesis inference apparatus according to the present invention further includes a contradiction rule storage unit for storing contradiction rules of the hypothesis world and a contradiction rule of the contradiction rule storage unit. And an inconsistency hypothesis elimination function section for eliminating the above-mentioned item.

[0009]

In the hypothesis inference apparatus according to the present invention, the inference engine collates the antecedent part of the production rule stored in the rule storage with the hypothesis world stored in the hypothesis world storage, resolves conflicts, and creates a new hypothesis world. Is performed, each hypothesis world stored in the hypothesis world storage unit is treated independently of each other regardless of parent and child, and an inference operation is performed. Therefore, deletion of data in one hypothetical world does not spread to other hypothetical worlds, and the existence of data in each hypothetical world can be clarified.

At the time of inference engine collation, the automatic merging function unit automatically merges appropriate hypothetical worlds among the hypothetical worlds stored in the hypothetical world storage unit to generate a new hypothetical world. By making inferences about the world, the efficiency of hypothetical inference can be improved.

In the hypothesis inference apparatus of the present invention, the efficiency of the hypothesis inference operation is improved by excluding a generated hypothesis world to which an inconsistency rule is applied so as not to be an inference object. can do.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a functional block diagram of an embodiment of the present invention. A hypothesis inference apparatus of this embodiment includes a rule database 3 storing a production rule 1 and a contradiction rule 2, and a production rule 1 of the rule database 3. Is provided with a hypothesis world management function unit 5 for storing and managing the hypothesis world 4 generated by the application of.

The above-mentioned hypothesis inference apparatus includes a collation, a conflict resolution,
The inference engine 6 includes a collation processing unit 6a, a conflict resolution processing unit 6b, and an operation processing unit 6c for performing inference by repeating three phases of operation. The antecedent part of the inconsistency rule is compared with the hypothesis world 4 stored in the hypothesis world management function unit 5, and the conflict of the corresponding hypothesis world is resolved to select one hypothesis world. A new hypothesis world is generated by applying the hypothesis generation command of the consequent part of the production rule, or a contradictory hypothesis world is generated by applying the consequent part of the contradiction rule.
Is executed.

Further, the above-mentioned hypothesis inference apparatus generates a new hypothesis world by the automatic merging function, which does not correspond to the antecedent of the production rule in the hypothesis world 4 of the working memory when the inference engine 6 checks. An automatic merge function unit 7 is provided.

Next, the operation of the hypothesis inference device having the above configuration will be described.

First, as shown in FIG. 1, the inference engine 6 executes inference based on the inference engine cycle of the matching phase, the conflict resolution phase, and the operation phase, and the consequent part of the production rule selected in the conflict resolution phase is Executed in the operation phase. Then, when a command that makes a hypothesis among the commands described in the consequent part is executed,
Another working memory composed of a new working memory element in which the data of the established hypothesis is added to the content of the working memory whose collation has succeeded is generated, and the hypothesis world 4 is constructed. In other words, the hypothesis world is generated by the hypothesis command, so that the hypothesis world consisting of all elements of the working memory whose rule matching including the hypothesis command succeeds is further added to the working memory element of the hypothesis world whose matching has succeeded. A new hypothetical world consisting of the working memory elements with the added data is created.

The basic operation will be described with reference to FIGS. 2 to 4. In FIG. 2, the collation is performed on the working memory element of W0, then rule 1 is selected by conflict resolution, and 5 shows a procedure in which the unit is executed. That is, in the working memory element W0,
The element "x is present and y is present" is applied, but by applying Rule 1, "assume" is a command to make a hypothesis in the consequent part, and this is executed. As a result, a new hypothesis world W1, in which the hypothesis data A and B are respectively added to the consequent part,
W2 is generated. Each of these hypothesis worlds W1 and W2 is composed of a working memory element in which respective hypothesis data A and B are added to a working memory element before being set. Therefore, from the condition of "there is x and there is y" in the hypothesis world W0, "if there is x and there is y, is it not A?" (W1), or "if there is x and there is y"Isn't it B? "(W
The hypothesis world of 2) is generated.

For convenience of expression, the illustrated hypothetical world W
, W2,..., Above the dotted line is the hypothesis data generated by the hypothesis command, and below the dotted line is the hypothesis data generated by the normal data registration command to the working memory. Shall be shown. Therefore, in this hypothesis inference method, it is assumed that the hypothesis data derived by the hypothesis command and the normal data other than the hypothesis data are separately managed in the hypothesis world 4.

The hypothesis inference operation performed by the inference engine 6 of this embodiment is performed independently for each hypothesis world in matching, conflict resolution,
The feature is that the rule of the operation inference engine cycle is executed, and as shown in FIG.
If there is and hypothesis A does not exist, W0 and W2 in the hypothesis worlds W0, W1, and W2 are compared.
Matching succeeds in 2. Therefore, these hypothesis worlds W0,
For W2, delete the consequent "y" of rule 2 (de
Let)), the data y is deleted in these hypothetical worlds, and these are made new hypothetical worlds W0 and W2.

The reason for this inference operation is that, if the data y of the parent W0 is deleted between W0 and W1 in the parent-child relationship, the y of the child W1 is also deleted in the prior art. Also, the hypothesis worlds W1 and W2 generated from the hypothesis world W0 are not dependent on parent and child, but are inferred completely independently.

Next, the automatic merge function will be described.
In the hypothesis inference apparatus of this embodiment, even when the matching is not successful, the matching of the pattern matching is successful.
The automatic merging function unit 7 operates in order to prevent a case where the verification does not succeed as a whole because there is no common hypothesis world. In other words, a hypothetical world in which data to be collated exists is created by performing an automatic merge of taking the union of a plurality of hypothetical worlds, and an operation is performed to make the overall collation succeed in the merged hypothetical world. It is.

As an example of this operation, as shown in FIG. 4, when the antecedent part of the production rule "if there is a hypothesis A and there is a hypothesis B" is compared with the hypothesis world 4 of the hypothesis world management function unit 5, Although the data A exists in the hypothesis world W1 and the data B exists in the hypothesis world W2, there is no hypothesis world in which the data A and B exist in common, so that the pattern matching succeeds. Does not succeed.

Therefore, the automatic merging function unit 7 operates at the time of collation to generate a hypothetical world W3 in which the data A and B are present in common, and succeeds the collation there to advance the inference. Here, the collation succeeds in the hypothesis world W3, the consequent part of Rule 3 is executed, and z is further registered in the data x and y of W3 (a
The hypothesis world W3 is newly generated and registered in the hypothesis world management function unit 5.

The detailed operation of the hypothesis inference apparatus of this embodiment will be described with reference to FIGS.

The inference engine 6 repeats a three-phase inference engine cycle of matching, conflict resolution, and operation. Hypothesis inference is first started from one hypothesis world, that is, a state in which no hypothesis has been generated. In other words, the working memory in a normal production system is compared with the antecedent of the rule, and when the execution command “assume” for establishing a hypothesis of the consequent of the rule that has been successfully matched is executed, the hypothesis command is executed. A hypothetical world including the data generated by the above is newly created, and subsequently, the data existing in the hypothetical world in which the rule is executed is copied. As an example of this operation, as shown in FIG. 5, ordinary inference is first performed in a single hypothetical world called W0, and the hypothetical world W0 is obtained according to the following rule, if x then assume (operation A10). , The hypothesis data (work A 1)
0) is added to generate a new hypothesis world W1. The meaning of this rule is easy to understand, for example, if it is considered as a kind of planning problem for the worker named Mr. x, "If Mr. x, let work A be assigned to 10 days."

After the hypothesis world W1 is generated in this way, inference is performed simultaneously and independently on the hypothesis worlds W0 and W1, as shown in FIG. In other words, the antecedent part of the rule is matched independently for each hypothesis world with respect to which data the hypothesis world holds, and the subsequent part executes the hypothesis world that has been successfully matched with the successfully matched data. There is no parent-child relationship between the hypothetical worlds as in the past.

The hypothesis world W of the inference result shown in FIG.
Assuming that the following rule, if x then assay (operation B10), is executed for 0 and W1, as shown in FIG. 6, a hypothesis that data x to be collated in the antecedent part of this rule exists. The world is W
0 and W1, the matching is successful, and the hypothesis data (work B
Hypothesis worlds W2 and W3 to which (10) is added are generated.

Here, the application of the contradiction rule 2 registered in the rule database 3, which is another feature of this embodiment, will be described. Now, when a hypothesis world as shown in FIG. 6 is generated, the following inconsistency rule is immediately executed in the hypothesis inference apparatus of this embodiment.
That is, if (work A? Date) & (work B? Da)
te) The then contrast is independently applied to each of the hypothetical worlds W0 to W3, the antecedent is checked, and the consequent “contradiction” is executed for the corresponding hypothetical world W2.

This contradiction rule states that "If the same day? Da
If work A and work B are assigned to te, the plan is contradictory. " In this embodiment, a symbol starting with "?", Such as "? Date", present in the antecedent of a rule represents a variable, and any value can be successfully matched. Assume that the rule is to match values.

The execution of this contradiction rule is controlled in the conflict resolution phase of the inference engine cycle so that it is executed first before any other rules. Therefore, the hypothesis world W2 is invalidated by the application of this contradiction rule, and no inference will be made for this hypothesis world in the future. This is indicated by a broken line x in the drawing, and the hypothesis world W2 is thereafter managed by the hypothesis world management function unit 5 as a hypothesis contradictory world.

In executing the next hypothesis inference, the following rule, if x then assume (operation B12), is executed, and as shown in FIG.
The hypothesis world W0, W1, W3 that the collation of "if there is" succeeds
The hypothesis data (operation B12) of the consequent part is independently added to each, and new hypothesis worlds W4, W5, and W6 are generated.

Following this, the contradiction rule, if (? Operation 10) & (? Operation 12) then contrast is executed. This inconsistency rule may mean, for example, "If the same work? If the work is assigned twice on 10th and 12th, the plan is inconsistent." Therefore, as a result of the execution of this contradiction rule, the hypothesis world W5 to which the same work B is allocated twice is excluded as a hypothesis contradiction world. As a result, when a plan that can be finally made is considered as a hypothesis, the hypothesis world that realizes the assignment plan is W4. "If Mr. x, work A is 10 days and work B is 12 days. On the day
Let's assign. "

Next, a description will be given of a processing operation when deleting data which has conventionally been a problem. In other words, in the conventional inference method in which there is a parent-child relationship between the hypothetical world,
The rule was executed in the hypothetical world of the parent, and the data deleted by the consequent part was to be deleted in the hypothetical world of the child under the same influence, but the inference method by the inference device of this embodiment The feature of is that each hypothesis world is independently subjected to inference, and inference is performed as follows.

From the inference results in FIG. 7, it has been found that the hypothesis world W4 has an appropriate assignment plan so far.
Thereby, the assignment plan of the worker x is completed. In this case, only the hypothetical world W4 is important, and it is not necessary to proceed with inference from the hypothetical worlds W1, W3, and W6. Data on Mr. x is not needed in future plans. Therefore, hypothesis world W
The data x is deleted at 1, W3, and W6, and the allocation plan, for example, the allocation plan for Mr. y can be continued by focusing only on the hypothesis world W4.

At this time, conventionally, if the rule for deleting the data x in the hypothetical world of W1 is executed, the data x of the hypothetical world W4 corresponding to the child of the hypothetical world W1 is deleted, and the object is achieved. Can not do it. However, in the inference method of the hypothesis inference apparatus of this embodiment, after the hypothesis world is generated, they are independent from each other and have no parent-child relationship. Is deleted, the data x is not deleted in the hypothesis world W4, and the inference as intended can be continued thereafter.

In this embodiment, such a mechanism for performing independent inference in each hypothesis world is realized by a method of information on the world to which data belongs. In other words, each data to be collated has the information of the hypothetical world to which it belongs as data with belonging world information, and the collation is also performed using this. That is, at the same time as checking the data elements by pattern matching, the belonging world information of the data is checked. that is,
Since it is a common part between the belonging world information of the data to be collated, AND of the belonging world information is performed in the same manner as performing pattern matching by AND logic of data.
, It is possible to determine a hypothetical world that exists in common.

An example of this method is shown in FIG.
8A shows an example of realizing the extent of the data shown in FIG. 8A, in which the data x belongs to the hypothesis world W0, W1, and the hypothesis data (operation A10) belongs to only the hypothesis world W1. . 6B shows an example of realizing the extent of the data shown in FIG.
0, W1, W3, hypothetical data (work A10)
Belongs to only the hypothesis world W1, and the hypothesis data (work B
10) indicates that it belongs only to the hypothesis world W3. FIG. 11C similarly shows an example of realizing the extent of the data in FIG.

A description will now be given of a collation method using belonging world information. When the belonging world information of the data is, for example, u = {W0, W2, W4} v = {W2, W3, W4}, For the rule if u & v then ***, u & v = {W0, W2, W4} & {W2, W3, W4} = {W2, W4}, so a rule instance as a matching result of this rule Will be {W2, W4}.

When a hypothesis is made, a hypothesis is generated, and apparently, a copy is made from the hypothesis world where the rules are executed to a newly generated hypothesis world. This will be dealt with by additionally rewriting this belonging world information.

The affiliation world information is also used in the conflict resolution processing. That is, when the matching process is completed, the rule instance with the belonging world information is registered in the conflict resolution set.
In the conflict resolution process, one rule instance to be executed is selected from the registered set by using an arbitrary strategy. Then, the consequent part of the rule is executed for one belonging world information-added rule instance selected by the conflict resolution processing. That is, the affiliation world information of the selected rule instance becomes a set of hypothetical worlds that execute the rule. The execution of the rule is executed equally for the hypothetical world of the world information to which the rule instance belongs. For example, the execution of the consequent part of the rule instance whose belonging world information is W8 and W9 is executed equally for W8 and W9. Even if W9 is generated from W8, it will be executed as such.

Next, an automatic merging function at the time of matching, which is a feature of the hypothesis inference apparatus of this embodiment, will be described with reference to FIGS.

Taking the above work planning problem as an example,
From the situation shown in FIG. 9, the hypothetical world shown in FIG.
It is generated by executing two rules sequentially.

Rule 11: if x then Assume (operation x10) First, according to this rule, a hypothetical world of W1 is generated from a hypothetical world W0 in which x data exists as shown in FIG.

Rule 12: if then then (operation y 10) According to this rule, a hypothesis world W0 is generated from the hypothesis world W0 as shown in FIG.

Rule 13: if y & (operation x10) then delete y According to this rule, the hypothesis world W1 in FIG.
Is deleted.

Rule 14: if x & (operation y 10) then delete x According to this rule, the hypothesis world W2
, The data x is deleted, and finally a hypothetical world in the state shown in FIG. 10 is generated.

Next, the following rule is considered in the hypothetical world shown in FIG.

Rule 15: if (operation x10) & (operation y10) then delete (operation y10) This rule is applicable if the operation plan (operation x10) and the operation plan (operation y10) are present together. Since the work plan (work x10) has a higher assignment priority, the work plan (work y10) is deleted.

However, in this rule 15, data (operation x10) and (operation y10) are derived in the antecedent part, but the hypothetical worlds in which these exist are W1 and W2.
Since there is no hypothesis world in which the two data coexist, the matching of the rule 15 does not succeed.

However, in the automatic merging function of this embodiment, since data is derived during inference, such a hypothesis world in which a plurality of data are present in common is generated at the time of rule matching, and the rule matching is performed. Is made to succeed in the merged hypothesis world. As shown in FIG. 11, a hypothesis world W3 is generated by an automatic merge function prior to the application of the rule 15. The data that should exist in the hypothesis world W3 is the union of the data that exists in each of the hypothesis worlds W1 and W2 to be merged. Therefore, the data x not deleted in W1 and the data y not deleted in W2 are merged, and both x and y exist in the hypothesis world W3. This is because in the conventional hypothesis inference method in which there is a parent-child relationship between the hypothetical worlds, the hypothetical world of the other parent to be merged does not exist in the hypothetical world of one parent due to deletion at the time of this merge. Then, the existence makes it ambiguous whether the data should exist in the new hypothesis being merged, but in this example, there is no parent-child relationship and all Since the hypothesis world is handled independently, automatic merging can be performed by the union as described above.

As a result of the automatic merging, the matching of the rule 15 is successful, the conflict is resolved, and when this rule 15 is selected, the consequent part is executed, and the work in the hypothesis world W3 is performed as shown in FIG. The plan (operation y10) is deleted.

By the way, in this embodiment, as the correspondence between the hypothesis data set as the hypothesis and the hypothesis world generated at that time, the deletion of the hypothesis data is a contradiction of the hypothesis world in which the hypothesis data is executed. This is based on the model that, since the hypothesis world was generated by the hypothesis, if the hypothesis was deleted, the hypothesis world could not exist. In that sense, this hypothetical world becomes a contradictory hypothetical world and will not be the subject of inference in the future. FIG. 12 shows this fact, and the hypothetical world W3 is indicated by a broken line with a cross mark, which has become a contradictory hypothetical world.

Thus, in the case of this embodiment, only one rule is described without execution of two rules of generation of a hypothesis and checking by a contradictory rule. The inference can be made efficiently by using. If this automatic merging function is not provided, a rule for making a hypothesis of data (work y10) in the situation of the hypothesis world W1 must be separately described in advance. Inferences can be made.

Other Examples of Automatic Merging FIGS. 13 to 1
5 will be described. In the automatic merging of FIGS. 13 to 15, the hypothesis world of FIG. 13 is first generated by the inference flow substantially similar to the automatic merging operation of FIGS. 9 to 12 described above, and the inference is performed there. I do. At this time, consider the following rules.

Rule 16: if (operation P13)
& (Work Q15) then assume (work S14) This rule 16 is defined as "work plan (work P13)
If the work plan (work Q15) is assigned, then a hypothesis for the work plan of (work S14) is made based on the work plan (work Q15).
Again, the hypothesis world W3 shown in FIG. 14 is generated at the time of matching by the same automatic merging function as described above, and after the matching, the consequent part of the rule operates in the operation phase of the inference engine cycle. "Assum
The "e" command generates a new hypothesis world W4.

As described above, the relation between the hypothesis data and the rule expression as described above is merely used without considering the hypothesis world in which the data exists. Can automatically generate the hypothesis world that you want to infer by hypothesis.

Next, a hypothesis world management function of the hypothesis inference apparatus of this embodiment will be described. In the hypothesis inference apparatus of this embodiment, the hypothesis world management function unit 5 that manages the hypothesis and the hypothesis world is based on the following functions. In other words, the hypothetical world is managed based on hypothetical data when the hypothetical world is generated, and specifically, is managed by a combination of hypothetical data above the broken line of each hypothetical world shown in the figure. . Then, management is performed so that a hypothetical world that is a combination of the same hypothetical data is not generated.

This management function will be described with reference to FIGS. 16 and 17. Assume that the following rule is executed when a hypothetical world is generated as shown in FIG.

Rule 17: if B then Assume A Since the rule 17 succeeds in the hypothesis world W2 in FIG. 16, the hypothesis world W4 should be generated as shown in FIG. Note, however, that the hypothesis world W3 has already been generated based on the hypothesis data A and B. That is, if the hypothesis world W4 is also generated, two hypothesis worlds based on the hypothesis data A and B are generated, which means that a redundant hypothesis world is generated. In this case, it is sufficient to perform inference in the hypothesis world W3 without generating the hypothesis world W4. Therefore, management is performed so that a hypothesis world like W4 is not generated, and efficient hypothesis inference can be performed. You do it.

[0061]

As described above, according to the present invention, the inference engine collates the antecedent part of the production rules stored in the rule storage unit with the hypothesis world stored in the hypothesis world storage unit, and resolves conflicts. When performing an inference operation to generate a new hypothesis world, each hypothesis world stored in the hypothesis world storage unit is treated independently of each other regardless of parent and child, and the inference operation is performed. The hypothesis that the automatic merge function is stored in the hypothetical world storage unit when the inference engine is compared can be merged between arbitrary hypothetical worlds without deleting the data of other hypothetical worlds. A match is not established by itself in the world, but a new hypothesis world that succeeds by taking the union of the hypothesis worlds is generated by merging appropriate hypothesis worlds. Since inference is made to the hypothetical world, parent-child relationships remain between the hypothetical worlds as in the past, and there is no ambiguity regarding deleted data, and the efficiency of hypothetical inference can be improved. .

Further, according to the present invention, for a hypothetical world to be generated to which a contradiction rule is applied,
Thereafter, since it is excluded so as not to be an inference object, it is possible to avoid matching against a given rule with a meaningless hypothesis world, and it is possible to further improve the efficiency of the hypothesis inference operation.

[Brief description of the drawings]

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

FIG. 2 is an explanatory diagram showing an inference operation of the embodiment.

FIG. 3 is an explanatory diagram showing an inference operation of the embodiment.

FIG. 4 is an explanatory diagram showing an inference operation of the embodiment.

FIG. 5 is an explanatory diagram showing an inconsistency rule matching operation of the embodiment.

FIG. 6 is an explanatory diagram showing an inconsistency rule matching operation of the embodiment.

FIG. 7 is an explanatory diagram showing an inconsistency rule matching operation of the embodiment.

FIG. 8 is an explanatory diagram showing a collation operation using belonging world information in the embodiment.

FIG. 9 is an explanatory diagram showing an automatic merge operation of the embodiment.

FIG. 10 is an explanatory diagram showing an automatic merge operation of the embodiment.

FIG. 11 is an explanatory diagram showing an automatic merge operation of the embodiment.

FIG. 12 is an explanatory diagram showing an automatic merge operation of the embodiment.

FIG. 13 is an explanatory diagram showing another example of the automatic merge operation of the embodiment.

FIG. 14 is an explanatory diagram showing another example of the automatic merge operation of the embodiment.

FIG. 15 is an explanatory diagram showing another example of the automatic merge operation of the embodiment.

FIG. 16 is an explanatory diagram showing a hypothetical world management function of the embodiment.

FIG. 17 is an explanatory diagram showing a hypothetical world management function of the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Production rule 2 Contradiction rule 3 Rule database 4 Hypothesis world 5 Hypothesis world management function part 6 Inference engine 6a Collation processing part 6b Conflict resolution processing part 6c Operation processing part 7 Automatic merge function part

Continuation of the front page (56) References JP-A-2-211537 (JP, A) JP-A-3-37725 (JP, A) JP-A-1-140337 (JP, A) JP-A-5-165642 (JP) , A) JP-A-5-2488 (JP, A) Information Processing Society of Japan, Vol. 92, No. 42 (92-AI-82), published by the Information Processing Society of Japan (May 26, 1992). 41-50 (Patent Office CSDB document number: CSN 199800017005) IPSJ 36th (Early 1988) National Conference Lecture Papers (Volume 2), Japan Society for Information Processing (1988), pp. 1367- 1368 Information Processing Society of Japan 37th (late 1988) National Conference Lecture Papers (Volume 2), Japan Information Processing Society of Japan (1988), pp. 1208-1209 IEICE Technical Report, Vol. . 90, no. 122 (AI90-47-58), The Institute of Electronics, Information and Communication Engineers (1990), pp. 61-68 (Patent Office CSDB literature number: CNT199900951008) Information Processing Society of Japan Research Report, Vol. 90, No. 88 (90-AI-73), Information Processing Society of Japan (1990), pp. 21-30 (Patent Office CSDB Document Number: CCNT 199900832002) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 9/44 JICST file (JOIS) CSDB (Japan Patent Office)

Claims (1)

(57) [Claims] 1. A rule storage unit for storing a production rule having an antecedent part and a consequent part; and a hypothesis world generated by a hypothesis generation command for forming a hypothesis described in a consequent part of the production rule. A hypothesis world storage unit that stores the antecedent parts of the production rules given from the production rules stored in the rule storage unit and the respective hypothesis worlds stored in the hypothesis world storage unit, independently checking each other, An inference engine that performs an inference operation of resolving a conflict and generating a hypothesis world having no parent-child relationship by a hypothesis generation command described in a consequent part of the production rule; and, when comparing a certain production rule in the inference engine, storing the hypothesis world memory. In the hypothetical world stored in the section, direct matching does not succeed, but the union of the hypothetical worlds is A hypothesis inference apparatus comprising: an automatic merge function unit that generates a new hypothesis world in which the matching is successful by merging a plurality of hypothesis worlds and stores the generated hypothesis world in the hypothesis world storage unit.