JPS63263542A - Expert system - Google Patents

Abstract

PURPOSE:To enable even an expert unaccustomed to computers to easily develop an expert system by describing both of forward and backward rules together in a natural form of Japanese-language sentences in one file. CONSTITUTION:A Japanese language input and editing means 1, a means 2 in which an inputted character string is stored, an expression limiting means (syntax) 3 which is so devised that knowledges for both of forward and backward inference can be described together, a means 4 which segments required terms from the inputted character string based on a prescribed syntax, a means 5 which syntactically analyzes the character string based on the prescribed syntax and segmented terms, an inference executing means 10 which has the internal format of knowledge based on the Japanese-language sentence expression, a means 6 which converts the syntactical analysis result to a prescribed internal format, a means 7 on which syntactic and semantic errors are displayed if they exist in the character string, and a means 9 in which the converted prescribed knowledge internal format is stored are provided. Input knowledges for both of forward and backward inference are made into a knowledge base in each inference format in accordance with the syntax prepared on the system side.

Description

[Detailed Description of the Invention] <Industrial Application Field> The present invention is a computer system that performs the work of experts, a so-called expert system, in which knowledge for forward reasoning and knowledge for backward reasoning are mixed in a natural manner. It relates to an expert system with a knowledge input method that can be described as follows.

<Conventional Technology> When converting the knowledge held by experts into a knowledge base, it is necessary to describe the input knowledge in a form that the system can understand, according to a certain syntax provided by the expert system used.

However, in conventional expert systems, it is difficult to express knowledge, and the syntax of many of them is similar to programming languages, making it difficult for experts who are not familiar with computers to input knowledge directly into expert systems. It has become a difficult task.

As one method to solve these problems, attempts have been made to equip the knowledge acquisition section of an expert system with a flexible man-machine interface such as natural language input.

However, even in such systems, those based on the prior art (1) cannot be described in the form of natural sentences; (2) the input ports for forward inference and backward inference are different systems.

The current situation is that we have such problems.

<Problem that the invention seeks to solve> - Difficult to describe knowledge in natural sentences.

・The input format for forward reasoning and backward reasoning is inconsistent. ・It is not possible to write a mixture of forward reasoning and backward reasoning.

The present invention attempts to solve this problem.

By solving the first problem, you can avoid writing that has a syntax that is biased towards a programming language style.
Since it is possible to get rid of the level of notation that is difficult to read, the development of expert systems becomes more accessible to experts who are not familiar with computers.

By solving the second and third points, it becomes easier to think and write when using forward reasoning for this part and backward reasoning for this part in the flow of problem-solving programming.

Methods for resolving problems> A means for inputting and editing Japanese, a means for memorizing the input character strings, and knowledge for both forward and backward reasoning are devised so that it can be written in a mixed manner. means (syntax) for restricting notation (syntax); means for cutting out necessary words from the input character string based on a predetermined syntax; means for parsing a string; a means for performing inference having an internal format of knowledge based on Japanese sentence notation; a means for converting the parsed result into a predetermined internal format; It consists of means for displaying syntactic or semantic errors, if any, and means for storing the internal format of the converted predetermined knowledge.

<Operation> It functions to convert forward and backward mixed input knowledge created by the user using Japanese sentences into a knowledge base for each inference format according to the syntax prepared by the system.

<Example> An example of the present invention will be described below with reference to the drawings.

FIG. 1 shows the basic configuration of the mixed input method of the present invention.

In the figure, the input/editing means 1 is a tool for actually creating mixed input sentences, such as a kanji editor.

Said input/editing means! The input sentence storage means 2 shown in the figure is a tool for storing the input sentence created in the storage device in a storage device, and this specifically refers to a storage device such as a file system of a computer or a disk.

A control means 8 is a programming system called a compiler that controls the processing from reading an input text to converting it into an internal format (knowledge base data) of the expert system.

The control means 8 manages and manages the following components.
It's in control.

- Lexical analysis means 4 Cuts out the smallest unit of syntactically meaningful character strings while referring to a table that defines the types of tokens.

- Syntactic analysis means 5 An intermediate data structure (surface structure) for the internal format is created based on the character string and grammar rules extracted by the lexical analysis means 4.

During analysis, if a string of tokens that does not conform to the grammar rules is input, the horizontal sentence analysis process will fail, so if the result is passed to the control means, error handling can be performed at the same time.

・Conversion means 6 to internal format Interpret the results of parsing and generate an internal format.

Similar to the syntax analysis process, if an intermediate structure that is determined to be structurally incorrect to be written into the internal format is passed as a result of (5), semantic error processing can be performed at the same time.

The above three are general programming systems.

・Display means 7 Displays errors and results.

A display device and software for screen display.

- Syntax storage means 3 A means for storing and referencing data (tokens, grammar rules) for syntactically interpreting mixed input sentences.

This is the main memory of a computer.

The data converted into the internal format is stored as a knowledge base on the memory by the knowledge base storage means IO. This also refers to normal disks, file systems, etc. The inference mechanism side 10 of the expert system reads this knowledge base data, executes inference 1, and solves the problem.

FIG. 2 shows an example of conversion of an input text into an internal format using this method.

In the case of this method, it is desirable that the internal format is determined as follows.

[1] A data structure (internal format of the simple sentence) that is the basis of matching is prepared for each simple sentence in the fact, condition, and conclusion parts.

[2] The internal forms of each simple sentence are jf , And , O
A data structure is prepared that represents functional connection using r, etc.

[3] The internal form of each simple sentence is structured such that predicates, nominals, and case particles are the basic elements of matching.

In the example of the internal format in the same figure, the predicate slot (prd:)
and some internal slots (pos
:, top:, 5ubj:, etc.), and can contain character strings representing predicates and nominals.

During inference, matching is performed using these slots and their values corresponding to the syntactic elements of the input sentence as keys. In this way, the internal tank′ that reflects the surface structure of the input text is created.
If you give it some structure, it will be possible to input Japanese sentences.

In both forward and backward reasoning, it is possible to express the internal form in a form that reflects the surface structure as in the example.

That is, if the notation of a simple sentence, which is the basic unit of backward matching, is the same as the notation of a fact in working memory, which is the basic unit of forward matching, at least the notation of the conditional part can be unified.

However, in forward reasoning, it is generally necessary to describe operations on working memory in the rules, so sentences with special syntax for working memory operations (
action statement) is required.

To deal with this, reserved words such as ``create'' and ``delete'' are prepared as operation names (actions) in action sentences, and the syntax of each action sentence as a Japanese sentence is prepared as a grammar rule. If you do this, you will be able to create natural Japanese sentences.

In addition, according to normal parsing methods, each node in the parse tree can have a specific data structure, or a program can be called in a grammar rule, so the parsed sentence can be forward-facing rules that include action sentences. Whether it is a backward rule or a backward rule can be easily determined from the data in the parse tree.

FIG. 3 shows an example of an input sentence.

FIG. 4 shows an example of the syntax of mixed input.

Figure 5 shows the meanings of action names.

FIG. 6 shows an example of the forward/backward determination procedure.

In this example, ・Determine the type of case particle as a reserved word as shown in 1 in Figure 4, ・Determine the action and action sentence as shown in Figure 5 and 2 in Figure 4, and the result will be as shown in Figure 3. It is possible to input text.

FIG. 6 illustrates the flow of processing.

The input text is scanned character by character in the lexical analysis unit, and is decomposed into a string of words having token names defined in the token table, with reference to a symbol table for specific characters, a reserved word table for specific words, etc.

In the syntax analysis processing unit, the token string passed from the lexical analysis processing unit is treated as an input part-of-speech sequence, and the main structure representing the syntactic structure as shown in Figure 2 is created by applying the grammar rules described in the grammar table. seek.

Using normal parsing techniques, each node (category) in the tree structure can have a predetermined data structure, and each grammar rule can contain procedures for manipulating that data. Here, the data necessary for the internal format is extracted and structured at the syntax level.

・Also, if you create unambiguous grammar rules, the analysis results and analysis paths will be uniquely determined.

Therefore, if a token string that is not in the analysis path is entered manually, the judgment can be made based on the presence or absence of the application rule, so error handling can be easily performed.

The internal format conversion section creates an internal format from the syntactic data extracted by the syntactic analysis process. The processing of this part depends on the internal format, so the algorithm will differ depending on the system.

For example, if the internal format is similar to the structure of Table 1 as shown in Figure 3, simple procedures such as data surge, data replacement with the table, and writing to the internal format buffer may be sufficient.

Effects> As explained above, according to the input method of the present invention, knowledge can be input using natural Japanese sentences.

This makes it possible to get rid of the difficult-to-write, hard-to-read notation where the syntax is biased toward programming language style, and makes the development of expert systems more accessible to experts who are not familiar with computers.

- Rules for forward inference and backward inference can be written in a unified format in a single file.

This improves the ease of thinking and writing when forward inference and backward inference are processed in parallel or while calling each other during programming.
This is particularly effective when performing fusion inference later.

As described above, this method is useful as a natural language interface when developing an expert system that incorporates a two-way inference mechanism for descendants and backwards.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the basic configuration for realizing this method, FIG. 2 is a diagram showing an example of input/output of this method, and FIG. 3 is a diagram showing an example of an input text according to this method. Figure 4 is a diagram showing an example of the syntax that defines the person-based method, Figure 5 is a diagram showing an example of a forward-looking action name, and Figure 6 is a diagram showing an example of the procedure for determining whether it is a forward-looking rule or a backward-looking rule. , 7th
The figure is a schematic diagram showing the flow of processing according to this method. ■...Input/editing means, 2...Input sentence storage means, 3
... syntax storage means, 4... lexical analysis means,
5... Syntactic analysis means, 6... Conversion means into internal format, 7... Display means, 8... Control means, 9... Knowledge base storage means, 10... Forward inference and backward reasoning. Reasoning mechanism. Agent Patent Attorney Takeshi Sugiyama (and 1 other person) Figure 1 (b) Figure 2 Figure 6 Determine whether facing forward or backward Figure 7 (a) - Figure 7 (b) Figure 7 (c) )

Claims (1)

[Claims] A means for inputting and editing Japanese, a means for storing the input character string, a means for restricting notation (syntax),
means for cutting out necessary words from the input character string based on a predetermined syntax; means for parsing the input character string based on the predetermined syntax and the cut words; An inference execution means having an internal form of knowledge based on text notation, a means for converting the parsed result into a predetermined internal form, and a syntactic or semantic error in the input character string, if any. In the knowledge acquisition processing unit of an expert system that is capable of forward and backward reasoning and has means for displaying the internal format of the converted predetermined knowledge and means for storing the internal format of the converted predetermined knowledge, both forward and backward rules are stored in one file. This expert system is characterized by being able to write mixed Japanese sentences in a natural manner.