JPH0454670A - Natural language analyzing method - Google Patents

Abstract

PURPOSE:To realize to efficiently execute the processing by providing a step for changing the acknowledgement of a word, based on information of a word and a syntax for constituting a syntax tree, and correcting the syntax tree in accordance with this change. CONSTITUTION:A natural language sentence inputted from an input device 1 is divided into words by a morpheme analyzing device 2. In a syntax analyzing device 3, syntax analysis is executed by using a syntax analysis rule 7, based on an output from the device 2, and a syntax tree is generated. Subsequently, in a work/syntax tree reacknowledging device 4, from word information obtained by the device 2 and the syntax tree obtained by the device 3, reacknowledgment of the word and a correction of the syntax tree are executed in accordance with a control table 8. Its result is outputted from an output device 5. In such a way, it is possible to cope easily with such an expression as a method for the word division being different in accordance with a structure of a sentence.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a natural language analysis method, that is, a morphological analysis method for analyzing words included in an input sentence, and a syntactic analysis method for analyzing the structure of an input sentence. This invention relates to a natural language analysis method that can handle expressions in which words are divided in different ways depending on the structure of the sentence.

[Conventional technology]

In conventional general natural language analysis methods, expressions that require word division depending on the structure of the sentence are basically excluded from the target, and expressions that require word division depending on the structure of the sentence are basically excluded. It had become.

For example, if we take the expression "around" as an example, "
In the case of the expression "controversy over authenticity", it is better to treat this as a particle equivalent, but in the case of the expression "tourist boat sailing around the bay", it is better to treat it as a particle equivalent. Unfortunately, the problem is something like ``I need to extract the verb ``Meguru J.''

In addition, in order to avoid being fixed to one of the above patterns, an improved method has been proposed in which parsing is performed partially at the word segmentation stage and the method of word segmentation is determined. .

In this regard, for example, the description in ``Japanese Information Processing'' edited by Makoto Nagao (IEICE series, March 1985) is helpful. In addition, Masaaki Miyazaki's doctoral thesis at Tokyo Institute of Technology, ``Research on language processing for Japanese speech output J
(1986) is also helpful.

[Problem to be solved by the invention]

In the conventional general natural language analysis method mentioned above,
Since word segmentation can only be divided into one of the word segmentation patterns, there is a problem in that the results of word segmentation must be monitored by some means, and if the segmentation pattern is not the desired result, corrections must be made manually. Ta.

Furthermore, in the case of the latter of the above-mentioned conventional techniques, that is, the improved method, it is necessary to incorporate a large number of exception handlings on top of the basic method of morphological analysis, which has the problem of poor processing visibility. Furthermore, at the stage of morphological analysis, there is a problem in that parsing performed partially and parsing the entire sentence overlap, resulting in an unavoidable drop in efficiency.

Note that among the above-mentioned conventional techniques, the last-mentioned paper describes a method of performing processing by applying various types of exception handling, but this has the downside of limiting the scope of application.

The present invention has been made in view of the above-mentioned circumstances, and its purpose is to solve the above-mentioned problems in the conventional technology, and to be able to cope with expressions in which the method of word division differs depending on the structure of the sentence. The objective is to provide a natural language analysis method. Another object of the present invention is to provide a natural language analysis method that eliminates the need to incorporate various types of exception handling into morphological analysis and syntactic analysis, thereby making it possible to improve processing efficiency.

[Means to solve the problem]

The above object of the present invention is to provide a natural language analysis system that analyzes the structure of an input sentence by identifying words included in an input sentence expressed in natural language and converting the input sentence into a syntax tree based on this result. This is achieved by a natural language analysis method characterized by the steps of: changing the recognition of the words based on information on the words and syntax constituting the syntax tree; and modifying the syntax tree in accordance with this change.

[Effect]

In the natural language analysis method according to the present invention, after first performing morphological analysis and then performing syntactic analysis, these results are used to review the word segmentation method and the syntactic analysis results to adjust the structure of the sentence. It is designed to obtain new word information and syntax trees.

In other words, at the stage of morphological analysis, division into basic words is performed regardless of syntactic issues, and at the stage of syntactic analysis, it is sufficient to analyze basic dependencies. By reviewing the word segmentation using the word information and syntax information obtained above, the words are re-qualified and the syntax tree is changed as necessary.

For example, for the above-mentioned ``wo,'' we first split the words into ``wo (particle)'' and ``wo (verb, adnominal form)''.
'', and the syntactic analysis yields the dependencies of [authenticity J - ``around'', ``inside the bay'' → ``around'' 0
In the next step, by reconsidering the word segmentation using syntactic information and word information such as ``authenticity'' and ``controversy,'' ``inside the bay,'' and ``tour boat J,'' in the former case, ``around'' is equivalent to the particle of one word. In addition to re-certifying it as a word, the syntax tree will be changed so that ``over authenticity'' is related to ``controversy'' due to the adnominal modification, but the latter will not be changed.

〔Example〕

Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a diagram showing the basic block configuration of a natural language analysis system that is an embodiment of the present invention.

In the figure, 1 is an input device for inputting natural language sentences, 2 is a morphological analyzer that divides the input sentence into words using a word dictionary 6, and 3 is a syntactic tree that performs syntactic analysis using syntactic analysis rules 7. 4 is a word/syntax analyzer that requalifies words and corrects the syntax tree according to a control table 8 from the word information obtained by the morphological analyzer 2 and the syntax tree obtained by the syntax analyzer 3. A tree re-certification device 5 indicates an output device that outputs the re-certification results obtained by the word/syntax tree re-certification device 4.

An overview of the operation of the natural language analysis system shown in this example is as follows.
It is as follows.

A natural language sentence input from the input device 1 is divided into words by the morphological analysis device 2. The syntactic analysis device 3 performs syntactic analysis using the syntactic analysis rules 7 based on the output from the morphological analysis device 2, and creates a syntactic tree. Next, the word/syntax tree recognition device 4 requalifies the words and corrects the syntactic tree according to the control table 8 from the word information obtained by the morphological analysis device 2 and the syntactic tree obtained by the syntactic analysis device 3. be exposed. This result is output from the output device 5.

FIG. 2 is a diagram showing the processing contents of the word/syntactic tree ear recognition device 4 described above, and shows the key word detection unit 41. Pattern matching section 42. It is composed of a word/syntax tree generation section 43. Further, FIG. 3 is a diagram showing an example of the configuration of the control table 8 that controls the operation of the word/syntax tree identification device 4.

The operation of the natural language analysis system of this embodiment configured as described above will be explained in detail below using the input sentence examples shown in FIGS. 4(a) and 4(b).

First, in the case of an input sentence a shown in FIG. It is parsed as follows. Although the syntax analysis result is generally represented in a tree structure as shown in FIG. 4(C), here, for convenience, it will be shown as the above-mentioned syntax analysis result a.

Word/syntax tree identification device 4 that receives the above analysis results
First, the key word detection unit 41 examines the control table 8 and obtains the rule 51 for "meguru". next,
The pattern matching unit 42 matches the pattern condition specification of the above-mentioned rule 51 and the syntax analysis result. That is, first, X2 of the rule 51 that includes "meguru" is associated with clause 4, and it is determined whether the structures match. In this case, they match, so next, XI and clause 2 are associated as clauses received by x2 and clause 3, and x3 and clause 4 are associated as corresponding clauses, and it is determined whether the structures of each correspond. In this case, since they all match, the condition is specified as "○K" and the matching is completed.

Next, the word/syntax tree generation unit 43 generates the word (particle) "around" according to the word/syntax tree generation of rule 51, changes the word structure of clause 2, and deletes clause 3. By doing this, the word segmentation results and syntactic analysis results are
Word segmentation result al shown in Figure (a).

Modify the syntax analysis result as shown in a'.

In the case of the input sentence shown in FIG. 4(b), as in the case of the input sentence a shown in FIG. The division results and syntax analysis results are analyzed as shown below. Furthermore, in the word/syntax tree identification device 4, the key word detection unit 4
1, the rule 51 for [meguru] is obtained, and the process proceeds to the pattern matching unit 42, which matches X2 and clause 3, but xl
When matching phrase 2 with ``[abstract object] 10'' according to the rule, the syntax analysis result does not match as ``[location] +'', so the word/syntax according to this rule The tree is not requalified, so in this case no corrections are made as in the case of FIG. 4(a).

By performing the above-described processing, in the case of the input sentence a in FIG. 4(a), the particle equivalent word "meguru" is recognized, and in the case of the input sentence i in FIG. It becomes possible to leave it as a verb.

On the other hand, even if the morphological analysis device 2 shown in FIG.
By registering the rule 61 shown in FIG. 3 in the control table 8, as shown in FIG. 5, it is also possible to control an operation opposite to the example described first.

According to the above embodiment, it is possible to easily deal with expressions in which words are divided in different ways depending on the structure of the sentence.

In addition, since there is no need to incorporate various types of exception handling into morphological analysis and syntactic analysis, you can maintain a clear outlook for processing.
Another effect is that it becomes possible to expand the scope of application of analysis.

It should be noted that the above-mentioned embodiment is shown as an example of the present invention, and it goes without saying that the present invention should not be limited thereto.

〔Effect of the invention〕

As described in detail above, according to the present invention, the words contained in an input sentence expressed in natural language are identified, and the input sentence is converted into a syntax tree based on this result, thereby analyzing the structure of the input sentence. In a natural language analysis system that performs We are developing a natural language analysis method that can deal with expressions in which word segmentation methods differ depending on the situation, and that can improve processing efficiency by eliminating the need to incorporate various types of exception handling into morphological analysis and syntactic analysis. This has the remarkable effect of being achievable.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing the basic block configuration of a natural language analysis system that is an embodiment of the present invention, Fig. 2 is a diagram showing the processing contents of the word/syntax tree fragrance recognition device, and Fig. 3 is a diagram showing the processing contents of the word/syntax tree scent recognition device. FIGS. 4 and 5 are diagrams showing an example of the configuration of a control table for controlling the operation of the word/syntax tree recognition device, and are diagrams showing input sentences and their analysis results for explaining the operation of the word/syntax tree recognition device. . an input device, 2: a morphological analyzer, 3: a syntactic analyzer,
4: Word/syntax recognition device, 5: Output device, 6: Word dictionary, 7: Syntax analysis rules, 8: Control table. Figure Natural language sentence Figure Word information/syntax tree Word information/syntax tree analysis results (word information/syntax tree) ＾ Mi diagram (Part 3) (c) Occurs (embedding) Circulating (case relation) Authenticity/falsity

Claims (1)

[Claims] (1) In a natural language analysis system that analyzes the structure of an input sentence by identifying words contained in an input sentence expressed in natural language and converting the input sentence into a syntactic tree based on the results, 1. A natural language analysis method comprising the steps of changing the recognition of said words based on information on words and syntax constituting said words, and modifying said syntax tree in accordance with this change.