JP2606056B2 - Knowledge base construction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge base construction method for constructing a knowledge base of a target area in a system using a computer application such as a database search system in a natural language.

[0002]

2. Description of the Related Art A system using a computer application such as a database search system using a natural language requires a target region knowledge base in which a target region semantic model is associated with a natural language.

[0003] For example, in a database relating to the status of establishment of a company, in order to answer a query "Which company belongs to the electric industry?", Natural languages such as the verb "belonging" and the particle phrase "to the electric industry" are targeted. It is necessary to define which component of the domain semantic model corresponds to which component.

In a conventional system using a computer application in a natural language, a user who has a knowledge description format of a target area knowledge base and natural language processing expertise is required to use components and natural language in a target area semantic model. The target area knowledge base was constructed by associating with the.

However, the above-described method has a problem that a user who does not have knowledge of the knowledge description format and the natural language processing in the target area knowledge base cannot construct a knowledge base.

[0006] In order to solve this problem, there is a patent application in which a user who does not have specialized knowledge can construct a target area knowledge base by receiving knowledge expressed in a natural language.

For example, in the invention described in Japanese Patent Application Laid-Open No. Hei 1-180046, "Knowledge Base System and Natural Language Understanding System", an analysis dictionary used for natural language analysis is incorporated in a target area knowledge base. A method is used in which knowledge in a natural language input by reference is analyzed, and further converted into a knowledge description format of a knowledge base and registered.

In the invention described in Japanese Patent Application Laid-Open No. Hei 2-112027, "Natural Language Type Knowledge Input Method", the analysis dictionary and the target area knowledge base are separately provided, and are input in the natural language using the analysis dictionary. The knowledge is analyzed, the structure is converted, and the knowledge is stored in the target area knowledge base.

[0009]

However, in the above "knowledge base system and natural language understanding method",
Since the knowledge base and the analysis dictionary are used together, the analysis dictionary must be recreated for each target area.

[0010] In the "input method of natural language type knowledge", there is no means for registering words depending on the target area in the analysis dictionary. Therefore, if a word that depends on the target area is not in the analysis dictionary, every time an expression containing that word is input,
It is necessary to fetch syntactic and semantic information about the word from other sources or to obtain it by inference, which is problematic in terms of efficiency.

As described above, in a conventional system using a computer application in a natural language, there is a problem to be solved regarding the construction of a knowledge base.

An object of the present invention is to solve the above-mentioned problems,
When a natural language knowledge description is input, a natural language analysis is performed, and by referring to the target area semantic model,
An object of the present invention is to provide a knowledge base construction method for efficiently and automatically constructing a target area knowledge base without having to construct an analysis dictionary for each target area.

[0013]

According to the present invention, there is provided a knowledge base construction method comprising: an analysis dictionary used for analyzing a natural language; a target region semantic model relating to words of a target region; and a deep case determination rule for determining a deep case. Natural language analysis means for receiving a natural language knowledge description input by a user, performing a natural language analysis with reference to the analysis dictionary and outputting an analysis result, and receiving the analysis result, Syntax information determining means for determining syntax information by referring to a case determination rule; dictionary registration means for registering the syntax information in the analysis dictionary; and syntax information for adding the syntax information to an analysis result and outputting an intermediate result An adding unit for converting the analysis result into a knowledge description format of a target region knowledge base with reference to the target region semantic model;

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of one embodiment of the present invention. FIG. 2 is an explanatory diagram showing an example of the analysis dictionary. FIG. 3 is an explanatory diagram showing an example of the analysis result. FIG. 4 is an explanatory diagram illustrating an example of a target area semantic model. FIG. 5 is an explanatory diagram showing an example of the deep case determination rule.

In FIG. 1, knowledge description 1 in natural language
Is a representation of an expression or concept unique to the target area by a user in a natural language such as Japanese.

The analysis dictionary 2 is used for analyzing a natural language, and includes a surface layer, a part of speech, and syntax information of the natural language. Here, the syntactic information mainly indicates information related to a case frame such as a deep case of a verb or a semantic category of a noun corresponding to the deep case.

The natural language analysis means 3 analyzes the knowledge description 1 in natural language with reference to the analysis dictionary 2 and outputs an analysis result. This natural language analysis means 3 uses an analysis method described in, for example, Chapters 1 and 2 of "Basic Technology of Natural Language Processing" (Hirosato Nomura, published by the Institute of Electronics, Information and Communication Engineers, 1988). Perform morphological and syntactic analysis using

The target area semantic model 4 describes the target area words constituting the target area semantic model, and the corresponding natural language notation and semantic category.

The deep case determination rule 5 is a rule describing a determination condition for determining a deep case from a case particle.

The syntax information determining means 6 determines syntax information such as a case frame of the analysis dictionary 2 with reference to the target area semantic model 4 and the deep case determination rule 5. First, the syntax information determining means 6 acquires the meaning category of the noun included in the analysis result from the natural language analyzing means 3 while referring to the target area semantic model 4. Further, for the case particles connected to those nouns, it is verified whether or not the determination condition of the deep case determination rule 5 is satisfied, and the deep case is determined.

The dictionary registration means 7 registers the syntax information determined by the syntax information determination means 6 in the analysis dictionary 2.

The syntax information adding means 8 adds the syntax information determined by the syntax information determining means 6 to the analysis result output by the natural language analysis means 3 and outputs an intermediate result.

The structure conversion means 9 converts the structure of the intermediate result and stores it in the target area knowledge base 10.

The target area knowledge base 10 associates a target area semantic model with a natural language.

First, the structure conversion means 9 refers to the target area semantic model 4 and searches for a word that can be replaced with the target area word in the intermediate result, and replaces it with the target area word.
Further, the structure conversion of the intermediate result is performed according to the knowledge description format of the target area knowledge base 10, and the result is stored in the target area knowledge base 10.

Here, an example will be described. For the sake of explanation, consider the case where a database search system is used in a natural language, and assume that the target area is a "database on the status of establishment of a company".

First, it is assumed that "company name belongs to industry" is input to knowledge description 1 in natural language. The natural language analysis means 3 performs a natural language analysis on the knowledge description 1 in a natural language. First, the natural language analysis means 3 performs morphological analysis with reference to the analysis dictionary 2 of FIG.

At this time, it is assumed that the verb "belongs" is a verb of a concept depending on the target area, and that the syntactic information such as the deep case or the semantic category of the noun corresponding to the deep case is not described in the analysis dictionary 2. Natural language analysis means 3
Morphological analysis is performed on knowledge description 1 in natural language, such as (noun) (case particle) (noun) (case particle) (verb), which belongs to / company / industry / /. Furthermore, for the verb "belonging", the noun "company name" is replaced by the case particle "ga"
3 (1)
Outputs the analysis result of.

Next, the syntax information determining means 6 receives the analysis result of FIG. 3A, and refers to the deep case determination rule 5 and the target area semantic model 4 to obtain the syntax information of the verb “belonging”, that is, the deep case. And the semantic category of the noun that corresponds to the deep case. First, the syntactic information determining means 6 refers to the target area semantic model 4 in FIG. 4 for the noun “company name” connected to the case particle “ga” and the noun “industry” connected to the case particle “ni”. And the semantic category "abstract, organization, location, attribute" of "company name" and the semantic category "object,
Abstraction, attributes ". Further, the syntax information determining means 6
With reference to the deep case determination rule 5 in FIG. 4 for the case particles “ga” and “ni”, the respective deep cases are determined. Regarding “ga”, the analysis result satisfies the determination condition of “there is no case particle [を]”, so that it is understood that the subject case is an object case (obj). In addition, the deep case determination rule 5 for the case particle “ni”
Inspection shows that the deep case changes in the semantic category. Since the semantic category of “industry” connected to “ni” is “object, abstract, attribute”, the deep case of “ni” is a goal case (tar).

Next, the dictionary registration means 7 is a syntactic information determination means 6
Is registered in the analysis dictionary 2. FIG. 2B shows the analysis dictionary 2 in which the syntax information has been registered by the dictionary registration unit 7.

The syntax information adding means 8 is a syntax information determining means 6
Is added to the analysis result of FIG. 3A, and an intermediate result is output. In this case, syntax information is added by replacing the case particle “GA” in FIG. 3 (1) with a symbol indicating a deep case called the target case “obj” and the case particle “NI” as a goal case “tar”. . FIG. 3B is an explanatory diagram illustrating an example of the intermediate result output by the syntactic information dictionary adding unit.

Finally, the structure conversion means 9 receives the intermediate result shown in FIG. 3 (2) and performs structure conversion. First, the structure conversion means 9 replaces the noun “company name” and the noun “industry” with the target area words “kname” and “category” with reference to the target area semantic model 4 of FIG. Next, the structure conversion means 9 performs structure conversion according to the knowledge description format of the target area knowledge base 10. Here, it is assumed that knowledge is stored in the target area knowledge base 10 in a network structure including nodes and links as shown in FIG. Therefore, the structure conversion means 9 converts the intermediate result in FIG. 3B into the network structure in FIG. 3C and stores the result in the target area knowledge base 10.

Although the present invention has been described with reference to the case where the knowledge description 1 is input in Japanese, it can be applied to other languages in principle. Further, the knowledge description format of the target area knowledge base 10 can be extended not only to a network structure but also to another structure such as a frame expression.

[0035]

As described above, according to the present invention, a user simply inputs a knowledge description in a natural language, and can obtain a target area knowledge base without special knowledge such as natural language processing or knowledge engineering. It is possible to construct

According to the present invention, since the target area knowledge base and the analysis dictionary are separately provided, it is not necessary to construct an analysis dictionary for each target area, and it is possible to flexibly cope with different target areas. it can.

Further, according to the present invention, if syntactic information of a word depending on a target area is insufficient in the analysis dictionary, the syntactic information is automatically acquired when an expression including the word is first inputted, Means for registering in the analysis dictionary are provided. Therefore, even if such an expression is input a second time, it is not necessary to register the same syntax information, which is efficient.

[Brief description of the drawings]

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

FIG. 2 is an explanatory diagram showing an example of a natural language analysis dictionary.

FIG. 3 is an explanatory diagram showing an example of an analysis result.

FIG. 4 is an explanatory diagram illustrating an example of a target area semantic model.

FIG. 5 is an explanatory diagram showing an example of a deep case determination rule.

[Description of Signs] 1 Knowledge description in natural language 2 Analysis dictionary 3 Natural language analysis means 4 Target area semantic model 5 Deep case determination rule 6 Syntax information determination means 7 Dictionary registration means 8 Syntax information addition means 9 Structure conversion means 10 Target area Knowledge base

Claims (1)

(57) [Claims] 1. An analysis dictionary used for analysis of a natural language, a target region semantic model for words of a target region, a deep case determination rule for determining a deep case, and a knowledge description in a natural language input by a user. A natural language analysis means for receiving and analyzing the natural language by referring to the analysis dictionary and outputting the analysis result; and a syntax for receiving the analysis result and determining syntax information by referring to the target area semantic model and the deep case determination rule Information determination means, dictionary registration means for registering the syntax information in the analysis dictionary, syntax information addition means for adding the syntax information to the analysis result and outputting an intermediate result, and referring to the target area semantic model. A structure conversion means for converting the analysis result into a knowledge description format of a target area knowledge base.