JPH06223109A - Knowledge acquisition system - Google Patents

Abstract

PURPOSE:To gradually acquire an objective area dictionary and objective area knowledge by repeating a cycle to register a Japanese notation on the objective area dictionary and objective area meaning network knowledge on the stage of analyzing the Japanese notation. CONSTITUTION:A Japanese notation imparting part 103 imparto Japanese language to a database schemer extraction part 101. An objective database structure selection part 120 selects an objective database structure, to which natural language analysis is performed, in the order of a theme, column and data defines a Japanese character string, which is not registered on an objective area dictionary 107 and a basic dictionary 106, as a non-registered word. A database structure meaning classification holding part 114 estimates a dictionary description corresponding to the non-registered word according to a syntax analysis rule 109 and gradually increases the objective area dictionary and the objective area knowledge by performing the next morpheme analysis by using the objective area dictionary 107, which vocabulary is increased, while performing registration on the objective area dictionary 107 and an objective area value knowledge holding part 110.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language interface for information retrieval means such as a database,
The present invention relates to a knowledge acquisition method in which the target area dictionary and the target area knowledge are gradually increased.

[0002]

2. Description of the Related Art With the development of database technology and AI (artificial intelligence) technology, there is an increasing demand for an interface that can be easily used by not only specialized operators but also users who are not familiar with computers. One of the interfaces that answers this request has been developed that inquires a computer in natural language. Such a natural language interface includes a semantic analysis unit that performs natural language processing,
Understand the meaning of the input sentence in natural language, create an input string according to the operation means specific to each application, and execute the application.

In order for the semantic analysis unit of the system to understand the meaning of a word included in an input sentence, it is necessary to perform a semantic analysis by collating with a dictionary. However, it is not possible to register all words included in various input sentences in the dictionary in advance, so some unregistered words that cannot be collated occur, and as a result the system In many cases, the results were incomprehensible.

Therefore, in order to acquire the knowledge of the target area that describes the relationship between the vocabulary of natural language and the internal representation on the target application, the concept of natural language is obtained by the tabular input method and the connection of nodes and links. There have been proposed methods such as a method of acquiring a mapping between a prime and a conceptual prime on a target application. However, the conventional knowledge acquisition method is
Each required knowledge about knowledge representation.

[0005]

Conventionally, when constructing such a Japanese interface, a dictionary item for analyzing a Japanese input sentence is described based on the dictionary expression,
It was necessary to describe the target domain knowledge for converting the parsed structure into an application language in the form of a system-dependent knowledge representation. Therefore, knowledge of the dictionary expression and knowledge of the knowledge expression of the target area knowledge are indispensable, and it is difficult for an expert who is familiar with the target area to directly input the knowledge.

The present invention has been made to solve the above problems. An object of the present invention is to gradually acquire the target area dictionary and the target area knowledge by repeating the cycle of registering the target area dictionary and the target area knowledge at the stage of analyzing the Japanese notation of the database, and By using information and Japanese grammar knowledge to the maximum extent, a large part of the knowledge required for registration can be automatically acquired, and even for grammatical / semantic classification information that cannot be estimated, it is possible to select an example sentence. It is possible to obtain the target region dictionary and target region knowledge even by an expert who has no knowledge of dictionary notation and target region knowledge representation.

[0007]

In order to solve the above-mentioned problems, the knowledge acquisition method of the present invention uses a database schema extraction unit for extracting a database schema from a database and a database schema extracted by the database schema extraction unit. A database schema holding unit for holding, a Japanese notation giving unit for giving Japanese notation to the database schema output by the database schema extraction unit, and a schema for holding the database schema given Japanese by the Japanese notation giving unit When defining a database structure, which is a constituent element of the database schema and a Japanese notation holding unit, a database to be targeted from the schema Japanese notation holding unit when defining a database structure Target database for selecting the structure Select the target character string for executing the target character string language analysis registration unit from the Japanese structure and the Japanese synonym notation corresponding to the database structure selected by the target structure selecting unit. The target character string selection unit and the target character string language analysis registration unit for analyzing and registering the target character string selected by the target character string selection unit, and the target character string language analysis registration unit stores the target character string. A morphological analysis unit for performing morphological analysis,
The basic dictionary and the target area dictionary referred to by the morphological analysis unit, the syntactic analysis unit that parses the morphological analysis sequence output by the morphological analysis unit, the syntactic analysis rule referred to by the syntactic analysis unit, and the syntactic analysis unit. Is storing and referencing an intermediate result, an intermediate result storage unit, a dictionary description estimation rule for estimating a dictionary description that is insufficient when the syntax analysis unit performs dictionary registration, and a syntax of the morphological analysis sequence in the syntax analysis unit. A target area knowledge holding unit that converts the syntax analysis result obtained by analysis into a database operation sequence, a dictionary registration unit that registers a vocabulary in the target area dictionary, and a dictionary that registers the syntax analysis result in the target area dictionary. A dictionary structure transformation unit that changes the format to a format, a knowledge structure transformation unit that changes the syntax analysis result to a format that registers the target area knowledge holding unit, and the syntactic analysis result to the pair. A database structure semantic classification holding unit for use in estimating semantic classification information required when registering in a region dictionary, and a target region knowledge registration unit for registering new region knowledge in the target region holding unit And have
When the target database structure selecting unit selects a large database structure and performs morphological analysis by the morphological analysis, Japanese character strings that do not exist in the target area dictionary and the basic dictionary are treated as unregistered words. , The dictionary description for the unregistered word is estimated by using the syntax analysis rule and the database structure semantic classification holding unit, and is registered in the target region dictionary by using the dictionary registration unit. The target area knowledge registration unit is registered in the knowledge holding unit, and the target area dictionary and the target area knowledge are gradually increased by performing the following morphological analysis using the target area dictionary with increased vocabulary. It is characterized by letting you go.

[0008]

The present invention will be described below with reference to the drawings. The gradual increase unit is the target database structure semantic classification storage unit 114, the target area knowledge storage unit 110, and the target area dictionary 107.

FIG. 1 is a basic configuration diagram of a knowledge acquisition system according to an embodiment of the present invention, FIG. 2 is a diagram showing a schema example of the database in FIG. 1, and FIG. 3 is a Japanese notation example of the schema of the database in FIG. FIG. 4 is a diagram showing a gradual increase portion immediately before the target character string language analysis registration processing of the input target area name in FIG. 1, and FIG. 5 is the state in FIG. FIG. 6 is a diagram showing a result of the target character string language analysis registration processing of the target area name, FIG. 6 is a diagram showing a gradual increase portion changed as a result of the target character string language analysis registration processing executed in FIG. 5, and FIG. In the Japanese notation of the database schema in 1, the diagram showing the gradual increase part immediately before the target database structure language analysis registration process when the KAISHA table is the target database structure, and FIG. 8 shows the KAISHA table. Pair when Japanese notation is the target character string The figure which shows the result of having performed the character string language analysis registration process, FIG. 9 shows the progressive increase part immediately after performing the target character string language analysis registration process when the Japanese description of the said KAISHA table is made into a target character string. FIG. 10 and FIG. 10 are diagrams showing the results of subject character string language analysis registration processing for “company”, which is a Japanese synonym notation of the KAISHA table, and here it is assumed that “company” does not exist in the basic dictionary. ing.

FIG. 11 is a diagram showing a gradual increase portion immediately after the target character string language analysis registration processing is performed for "company", which is a Japanese synonym notation of the KAISHA table, and FIG.
"Company", a Japanese synonym for the ISHA table
In the figure which shows the result of having performed the target character string language analysis registration process, it is assumed here that "company" does not exist in the basic dictionary.

FIG. 13 is a diagram showing a gradual increase portion immediately after performing the target character string language analysis registration process for "Company" which is a Japanese synonym notation of the KAISHA table, and FIG. 14 is a diagram of the database schema in FIG. KAIS in Japanese notation
FIG. 15 is a diagram showing a gradual increase portion immediately before the target database structure language analysis registration process when the TELNO column of the HA table is used as the target database structure. FIG. 15 shows the Japanese notation of the TELNO column as the target character string. FIG. 16 is a diagram showing the result of performing the target character string language analysis registration process of FIG.
FIG. 17 is a diagram showing a gradual increase portion immediately after performing the target character string language analysis registration process when the Japanese notation of the LNO column is used as the target character string. FIG. FIG. 18 is a diagram showing the result of performing the target character string language analysis registration processing on "telephone number", and FIG. 18 is a view immediately after the target character string language analysis registration processing on "telephone number of the head office" which is a Japanese synonym notation of the TELNO column. FIG. 19 is a diagram showing a format of a dictionary used in morphological analysis, FIG. 20 is a diagram showing an example of a format of a dictionary used in morphological analysis, and FIG. 21 is a diagram showing a rough processing flow. , FIG. 22 shows
It is a figure showing the flow of target database structure language analysis registration processing and target character string language analysis registration processing.

Referring to FIG. 1, in this embodiment, a database schema extraction unit 101 for extracting a database structure from a database and the database schema extraction unit 10 are provided.
The database schema holding unit 102 for holding the database schema extracted in 1; the Japanese notation giving unit 103 for giving Japanese notation to the database schema output by the database schema extraction unit 101; and the Japanese notation giving unit 103. A schema Japanese notation storage unit 104 that stores a database schema to which Japanese is added;
The morpheme analysis unit 105 for performing a morpheme analysis of the schema Japanese expression output by the Japanese notation giving unit 103, the basic dictionary 106 and the target area dictionary 107 referred to by the morpheme analysis unit 105, and the morpheme analysis unit 105 A syntax analysis unit 108 that analyzes the syntax of the output morphological analysis sequence;
Parsing rule 109 referenced by the parsing unit 108
An intermediate result storage unit 116 for the syntax analysis unit 108 to store and refer to an intermediate result, and a dictionary description estimation rule 117 for estimating a dictionary description that is insufficient when the syntax analysis unit 108 registers a dictionary. A target area knowledge holding unit 110 for converting the syntactic analysis result into a database operation sequence, a dictionary registration unit 111 for registering a vocabulary in the target area dictionary 107, and the syntactic analysis result for the target area dictionary. A dictionary structure modification unit 112 that changes the dictionary format to be registered in 107, a knowledge structure modification unit 113 that changes the syntax analysis result to a format that is registered in the target area knowledge holding unit 110, and the syntax analysis result in the target area. Database structure semantic classification holding unit 11 for use in estimating semantic classification information required when registering in the dictionary
4 and a target area knowledge registration unit 115 for registering new area knowledge in the target area knowledge holding unit 110.

Next, regarding the operation of this embodiment, while following the flow of FIGS. 21 and 22, the intermediate results shown in FIGS.
It will be described using 0.

The database schema extraction unit 101 inputs a database search command for outputting schema information to the database management system 90 having the target database 91, acquires the schema information, and stores it in the database schema holding unit 102 as shown in FIG. Table name 201
And the column name 202 are stored so as to correspond to each other.

A database search command for outputting schema information is executed by the database management system 90, for example.
If you have a QL interface, you can use the method described in "Introduction to SQL Language" by Kazuki Kawamura, published by Nikkan Kogyo Shimbun.

The Japanese notation giving unit 103 gives the Japanese notation 304 and the Japanese synonym notation 305 of FIG. 3 to each database structure of the schema shown in FIG. To do. For the Japanese notation imparting unit 103, for example, the method described in Proceedings 45th National Convention of Information Processing Society of Japan, 2F-10, "Object Field Knowledge Input Support Tool for Database Search by Natural Language" may be used. . The target area name input by the target area name input unit 123 is the target character string language analysis registration unit 122.
Analysis is performed by.

The state of the gradual increase portion immediately before this is as shown in FIG. 4, and the database structure meaning classification 401 held by the database structure meaning classification holding portion 114 is also the target held in the target area holding portion 110. Domain knowledge 402
Also, the target area dictionary content 403 held in the target area dictionary 107 is empty.

As shown in FIG. 5, the target character string 501 "company information" is morphologically analyzed using the target area dictionary 107 and the basic dictionary 106 to obtain a morpheme string 502.

The morphological analysis unit 105 includes, for example, Showa 63
You can use the method described in the annual conference of the IEICE Fall Meeting, "PIVOT JE: Japanese Morphological Decomposition".

The morpheme string 502 is parsed using the syntax analysis rule 109 and the intermediate result storage unit 116, and the syntax analysis result 503 is stored in the syntax analysis result storage unit 124.
To store. For the syntactic analysis unit 108, for example, syntactic analysis means known from Japanese Patent Application No. 61-175034 may be used. For the conceptual structure representing the syntactic analysis result 503, for example, the method described in the Institute of Electronics, Information and Communication Engineers Technical Report NLC91-62 "Natural Language Interface Construction Kit: IF-Kit" may be used.

The dictionary structure transformation unit 112 registers the semantic classification for the target area name in the database structure semantic classification holding unit 114 as shown in 601 in FIG. 6, and creates the target area dictionary registration item 504 from the syntax analysis result, Dictionary registration unit 11
1, the target area dictionary content 60 is added to the target area dictionary 107.
Store as 3.

The knowledge structure transformation unit 113 uses the syntax analysis result 50.
Create 505 as the target area knowledge registration item from 3,
The target area knowledge registration unit 115 stores the target area knowledge holding unit 110 like the target area knowledge 602. The target area knowledge 602 includes a node surrounded by a circle like 604 and 6
It is made up of labeled links like 05. 6
Links with a label 0 as indicated by 06 are synonymous links and connect nodes having the same meaning.

FIG. 6 shows a gradual increase portion immediately after the target character string language analysis registration processing is completed with the target area name as the target character string.

Next, the target character string language analysis registration processing is performed by using each table belonging to the input target area name as a target database structure and using its Japanese notation and Japanese synonym notation as a target character string. The flow when the target database structure is a KAISHA table is shown in Figs.
Description will be made with reference to 10, 11, 12, and 13.

Since the target database structure is the KAISHA table, the Japanese notation “company” of the KAISHA table, the Japanese synonym notation “company”, “company”, and “company” are sequentially targeted as target character strings. Performs character string language analysis registration processing. The gradual increase portion immediately before this is as shown in FIG. 7, which is the same as FIG. 6 at the stage when the target character string language analysis registration processing of the target area name is completed.

First, FIG. 8 shows the result of the target character string language analysis registration process for the Japanese notation "company" in the target database structure KAISHA table. As a result, the gradually increasing portion becomes as shown in FIG.

Next, the target character string language analysis registration processing is performed for the Japanese synonym notation “company” in the target database structure KAISH table. This result is shown in FIG. 10 and shows the case where there is no “company” in the basic dictionary. Since “company” is an unregistered word, the syntactic analysis result 1002 and the dictionary registration item 1003 are obtained from the dictionary description estimation rule 117 and the database structure semantic classification 901, and the progressive increase portion becomes as shown in FIG.

Similarly, the Japanese synonym notation "company"
As for the result, the result is shown in FIG. 12, and as a result, the gradually increasing portion becomes as shown in FIG.

Similarly, the target database structure language analysis registration processing is performed on the unprocessed table.

Next, the same processing is performed for the columns belonging to the target area name. Here, regarding the TELNO column of the KAISHA table, FIG. 14 shows the change in the gradually increasing part when the target character string analysis is performed for the "phone number" which is the Japanese notation and the "phone number of the head office" which is the Japanese synonym notation. ,
The results of the language registration analysis are shown in FIGS.

As described above, the schema Japanese notation holding unit 1
The target area dictionary 106 and the target area knowledge holding unit 110 are gradually increased by performing the target character string language analysis registration processing of Japanese notation and Japanese synonym notation on all the database structures held in 04. Can be made.

Although the present invention has been specifically described based on the embodiments, the present invention is not limited to the above embodiments, and various modifications can be made without departing from the scope of the invention. Needless to say.

[0033]

As described above, according to the present invention, the cycle of registering in the target area dictionary and the target area semantic network knowledge at the stage of analyzing the Japanese notation of the registered database is performed. By repeating, the target area dictionary and the target area knowledge can be gradually acquired. Also, by maximizing the use of schema information and Japanese grammar knowledge, a large part of the knowledge necessary for registration can be acquired automatically, and even for grammatical / semantic classification information that cannot be estimated, Since it can be acquired by a simple inquiry such as selecting an example sentence, even an expert who has no knowledge of dictionary notation and semantic network knowledge representation can construct a target area dictionary and target area knowledge.

[Brief description of drawings]

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

FIG. 2 is a diagram showing a schema example of a database in FIG.

3 is a diagram showing an example of Japanese notation of a schema of the database in FIG.

FIG. 4 is a diagram showing a gradual increase unit immediately before subject character string language analysis registration processing of the input subject area name “company information” in FIG. 1;

5 is a diagram showing a result of subject character string language analysis registration processing of the inputted target area name "company information" in FIG. 1 in the state of FIG. 4;

6 is a diagram showing a gradual increase portion changed as a result of the target character string language analysis registration processing executed in FIG. 5;

FIG. 7 shows a gradual increase portion immediately before the target character string language analysis registration process with the Japanese character “company” in the KAISHA table in the Japanese notation of the database schema in FIG. 1 as the target character string. It is a figure.

FIG. 8 is a diagram showing a result of performing a target character string language analysis registration process when the Japanese notation “company” in the KAISHA table is set as a target character string.

FIG. 9 is a diagram showing a gradual increase unit immediately after performing a target character string language analysis registration process when the Japanese notation of the KAISHA table is used as a target character string.

FIG. 10 is a diagram showing a result of subject character string language analysis registration processing performed on “company”, which is a Japanese synonym notation of the KAISHA table, assuming that “company” does not exist in the basic dictionary. There is.

FIG. 11 is a diagram showing a gradual increase unit immediately after performing a target character string language analysis registration process on “company”, which is a Japanese synonym notation of the KAISHA table.

FIG. 12 is a diagram showing a result of subject character string language analysis registration processing of “Company”, which is a Japanese synonym notation of the KAISHA table, assuming that “Company” does not exist in the basic dictionary. There is.

FIG. 13 is a diagram showing a gradual increase unit immediately after performing a target character string language analysis registration process on “Company” which is a Japanese synonym notation of the KAISHA table.

[FIG. 14] In the Japanese notation of the database schema in FIG. 1, when the Japanese notation “telephone number” of the TELNO column of the KAISHA table is the target character string, the gradual progress immediately before the target character string language analysis registration process It is a figure which shows a static increase part.

[Fig. 15] "TEL number" written in Japanese on the TELNO column
It is a figure which shows the result of having performed the target character string language analysis registration process at the time of making into a target character string.

[Fig. 16] Japanese notation "phone number" of the TELNO column
It is a figure which shows the gradual increase part immediately after performing target character string language analysis registration processing when using as a target character string.

FIG. 17 is a diagram showing a result of performing target character string language analysis registration processing on “telephone number of head office” which is a Japanese synonym notation of the TELNO column.

FIG. 18 is a diagram showing a gradual increase portion immediately after the target character string language analysis registration process is performed for “telephone number of headquarters”, which is a Japanese synonym notation of the TELNO column.

FIG. 19 is a diagram showing a format of a dictionary used in morphological analysis.

FIG. 20 is a diagram showing an example of a format of a dictionary used in morphological analysis.

FIG. 21 is a diagram showing a rough processing flow in the embodiment of FIG.

22 is a diagram showing a flow of a target database structure language analysis registration process and a target character string language analysis registration process in the embodiment of FIG.

[Explanation of symbols]

90 database management system 91 database 101 database schema extraction unit 102 database schema holding unit 103 Japanese notation giving unit 104 schema Japanese notation holding unit 105 morphological analysis unit 106 basic dictionary 107 target area dictionary 108 syntax analysis unit 109 syntax analysis rule 110 target Domain knowledge storage unit 111 Dictionary registration unit 112 Dictionary structure modification unit 113 Knowledge structure modification unit 114 Database structure semantic classification storage unit 115 Target area knowledge registration unit 116 Intermediate result storage unit 117 Dictionary description estimation rule 120 Target database structure selection unit 121 Target character string selection unit 122 Target character string language analysis registration unit 123 Target area name input unit 201 Table name storage area 202 Column name storage area 301 Target area name storage area 302 Table name storage Area 303 Column name storage area 304 Japanese notation storage area 305 Japanese synonym storage area 401 Database structure Semantic classification 402 Target area knowledge 403 Target area dictionary content 501 Target character string 502 Morphological string 503 Syntax analysis result 504 Dictionary registration item 505 Target area knowledge registration item 601 Semantic classification for target area name 602 Target area knowledge 603 Target area dictionary content 604 Node representing company information 605 Link that transitions when a company is entered in the information node 606 Synonym link with label 0 901 Semantic classification for KAISHA table 1001 Morphological sequence for "company" 1002 Parsing result for "company" 1003 Dictionary registration item for "company"

Claims (1)

[Claims] 1. A database schema extraction unit for extracting a database schema from a database, a database schema holding unit for holding the database schema extracted by the database schema extraction unit, and a database schema output by the database schema extraction unit in Japan. A Japanese notation giving unit for giving word notation, a schema Japanese notation holding unit for holding a database schema to which Japanese has been given by the Japanese notation giving unit, an element constituting the database schema, and a target area , A target database structure selecting unit that selects a target database structure from the schema Japanese expression holding unit, and a target database structure selecting unit Selected by A target character string selection unit that selects a target character string for executing the target character string language analysis registration unit from among the Japanese notation and Japanese synonym notation corresponding to the database structure, and the target selected by the target character string selection unit. A target character string language analysis registration unit for analyzing and registering a character string, the target character string language analysis registration unit performs a morpheme analysis on the target character string, and a basic dictionary referred to by the morpheme analysis unit. And a target area dictionary, a syntactic analysis unit that syntactically analyzes the morphological analysis sequence output by the morphological analysis unit, a syntactic analysis rule referred to by the syntactic analysis unit, and an intermediate result in which the syntactic analysis unit stores and refers to an intermediate result. A storage unit; a dictionary description estimation rule for estimating a dictionary description that is insufficient when the syntactic analysis unit registers a dictionary; A target area knowledge holding unit that converts the syntactic analysis result obtained by the syntactic analysis into a database operation sequence, a dictionary registration unit that registers a vocabulary in the target area dictionary, and the syntactic analysis result is registered in the target area dictionary. A dictionary structure transformation unit that changes to a dictionary format, a knowledge structure transformation unit that changes the syntax analysis result to a format that registers the target area knowledge holding unit, and a necessary when registering the syntax analysis result to the target area dictionary The target database structure has a database structure semantic classification holding unit for use in estimating semantic classification information that is assumed to be, and a target region knowledge registration unit that registers new region knowledge in the target region holding unit. When the selecting unit selects a large database structure, and when performing the morphological analysis by the morphological analysis, the target area dictionary and the Using a Japanese character string that does not exist in this dictionary as an unregistered word, a dictionary description for the unregistered word is estimated using the syntax analysis rule and the database structure semantic classification holding unit, and in the target area dictionary, By performing registration using the dictionary registration unit, registration using the target region knowledge registration unit in the target region knowledge holding unit, and performing the next morphological analysis using the target region dictionary with increased vocabulary. A knowledge acquisition method characterized in that the target area dictionary and the target area knowledge are gradually increased.