JPS62232076A - Learning system - Google Patents

Abstract

PURPOSE:To register a new knowledge in a knowledge base by analyzing a sentence by using syntactical information and knowledge, and verifying the ambiguity of the obtained fact. CONSTITUTION:An ambiguity decision means 2 to verify the ambiguity of the fact obtained by a sentence analysis means 1, is provided. The means 1 retrieves the content of the syntactical information 3 and that of the knowledge base 4 respectively through transmission lines 31 and 41, and analyzes a sentence by using these contents. The result of the analysis is transferred to the means 2 through a transmission line 12, so that a fact which is not ambiguous and can be determined as unitary is found. And the result is transferred through a transmission line 24 and stored in the knowledge base 4.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a learning method for automatically acquiring linguistic information necessary for machine translation and natural language processing.

[Prior art and its problems]

Traditionally, when collecting detailed linguistic information and constructing dictionaries,
They were manually registered in the electronic dictionary by referring to existing dictionaries and following the established sixth-order specifications. Various methods have already been proposed for automatic learning, but no practical method exists.

Most of the methods are useful only in very limited fields.

The present invention provides a method for constructing a system that automatically learns linguistic information by analyzing existing text, extracting information to be learned from the analysis results, and evaluating that information. is the purpose.

In order to achieve this purpose, first read the text, use a means to analyze the text, extract new information to be learned from the result of the text analysis, and put all the retrieved information into a knowledge base. means for storing and evaluating the information;
After one review, a new way to rewrite the F knowledge base based on the results must be provided.

There have already been many reports regarding sentence analysis, but since the purpose here is to acquire linguistic information (knowledge), we will focus on analysis of dependencies based on dependent grammars and introduce nine analysis means. The problem we need to solve here is to use our current knowledge to determine the facts (1;'act) contained in the sentence.
) is an ambiguity in the analysis that arises when fixed. Learning ambiguous facts is extremely dangerous and must be avoided. In other words, it is essential to have a mechanism that can identify the meaning of a sentence that includes the new B5 disaster as an unambiguous fact using existing linguistic information.

[Means to settle the issue]

The learning method of the invention has syntactic information and a knowledge base as information for analyzing a sentence, a means for analyzing the sentence using the syntactic information and the knowledge base, and a fact obtained by using the means for analyzing the sentence. The present invention is characterized by comprising: ambiguity determining means for verifying ambiguity; and storing means for storing facts determined to be unambiguous in the knowledge base.

[For production]

Linguistic information is divided into linguistic information such as syntax and knowledge information regarding meaning.Here, it is assumed that sufficient linguistic information has been provided, and learning is based on knowledge information. , consider a simple example. Suppose the following sentence is given fL9.

``A red rose has bloomed.'' This sentence shows the facts that ``The color of a rose flower is red'' and ``A rose is something that blooms.'' 0These two facts can be explained using syntactic rules in the above sentence. You will be able to hang it. 9. If it's a simple sentence.

Even if you do not use knowledge of the facts shown there, ``red'' is an adjective and relates to the noun, and ``baraga'' is a conjunctive form.
The dependency can be identified by the two rules of ``relating to predicates,'' and it is in principle possible to remove the fact from the identified n-friend relationship. For example, in the following sentence, when given a transparent vase of red roses, ``red'' refers to ``rose.''

Regardless of whether it is “transparent” or “rose”, it is related to “flower bottle”.
[Embodiment] Figure 1 is a functional block diagram showing an embodiment of the present invention. be. The text analysis means 1 retrieves the content of the syntactic information 3 and the content of the knowledge base 4 via transfer lines 31 and 41t, respectively, and analyzes the text using these contents. The analyzed results are sent to the ambiguity determination means 2 via the transfer line 12, which discovers facts that are unambiguous and can be determined at will, and sent to the knowledge base 4 via transfer #1j124t.
Store that fact in . Note that the storage means is not illustrated.

〔Effect of the invention〕

In the present invention, Ln analyzes sentences using syntactic information and knowledge,
By verifying the ambiguity of the obtained facts, new knowledge can be registered in the knowledge base.

The invention further enables automatic knowledge acquisition from text.

[Brief explanation of drawings]

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

Claims (1)

[Claims] A text analysis means that has syntactic information and a knowledge base as information for analyzing a text, uses the syntactic information and knowledge base to analyze the text, and an ambiguity judgment that verifies the ambiguity of the facts obtained by the text analysis means. and storage means for storing, in the knowledge base, facts determined to have no ambiguity by the ambiguity determining means.