JPH07262189A - Extracting device for sentential form pattern - Google Patents

Abstract

PURPOSE:To easily correct a syntax analysis rule by preparing a large number of texts in a targeted range, integrating all of them by sampling the sentential patterns of them, automatically collecting all the sentential patterns and receiving the support of those sentential patterns. CONSTITUTION:This device is constituted of a morpheme analysis part 2 which applies morpheme analysis to a large number of texts, a sentential pattern sampling part 3 which samples the sentential pattern by applying a sentential pattern sampling rule 4 generated in advance to a result of application of the morpheme analysis by the morpheme analysis part 2 and a sentential pattern integrating part 6 which integrates a set of sentential patterns sampled by the sentential pattern sampling part 3.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is a sentence pattern extracting device for automatically extracting sentence patterns from text, and is a basic material for creating syntax rules and keyword extraction rules in natural language processing. The present invention relates to a device for automatically extracting sentence pattern patterns, which is particularly relevant to a sentence pattern extraction device for automatically extracting and creating sentence patterns by giving a large amount of input text.

[0002]

2. Description of the Related Art In natural language processing, syntactic analysis for analyzing the sentence structure of an input text is one of the most basic processing. For this purpose, the syntax analysis rule prepared in advance is used, and the structure of the entire sentence is analyzed by repeating the rewriting of the sentence element in accordance with the syntax analysis rule. The parsing rules used here are the grammars theoretically proposed by many linguists, rewritten into a form that can be processed by a computer.However, there are exceptions in the sentences in the actual text distributed in the world. In many cases, it will actually be necessary to modify such parsing rules to a large extent.

Further, as a part of information extraction from a large amount of text, rules regarding sentence pattern are defined for the purpose of extracting keywords, relationships between a plurality of keywords, and knowledge such as text summarization. There is a method of preparing and extracting information of a portion depending on each pattern from a sentence matching this sentence pattern.

As described above, the parsing rules and sentence pattern are essential for a wide range of natural language processing.
In the past, the method has been used in which the linguistic insights of the human being created are emphasized and created in a theoretical way.

[0005]

In the above-mentioned method, since the created rule is applied to the target text and evaluated, and the correction is repeated based on the result, generally, a very delicate and difficult correction work is performed. The problem is that it often lasts for a long time. Furthermore, there is no definite guarantee that the rules thus created will be correctly applied to the entire target text, or to new target texts as well. In addition, there was a problem that the correction work would continue endlessly.

The present invention solves these problems.
Prepare a large amount of text in the target range, extract and integrate sentence pattern patterns over the entire text, collect all sentence patterns automatically, and correct parsing rules with the support of this sentence pattern. The purpose is to enable easily.

[0007]

FIG. 1 shows a block diagram of the principle of the present invention. In FIG. 1, the text base portion 1
Is for accumulating a large amount of text. Text that matches the selection condition can be selected and extracted from a large amount of text. Here, as the selection condition, there are a condition for designating a text field and a condition for designating a surface character string. Texts (sentences) that match these selection conditions are selected, and sentence pattern patterns to be described later are automatically extracted.

The morphological analysis unit 2 performs morphological analysis on text. The sentence pattern extracting unit 3 applies the sentence pattern extraction rule 4 to the result of the morphological analysis to extract sentence patterns.

The sentence pattern extraction rule 4 is a rule for extracting sentence patterns from the result of morphological analysis. The sentence pattern 5 is a set of sentence patterns extracted from the sentence by the sentence pattern extraction unit 3.

The sentence pattern integration unit 6 integrates a set of sentence patterns extracted by the sentence pattern extraction unit 3. The syntactic analysis unit 7 converts the syntactic analysis rule 8 into a natural language sentence.
Is applied for parsing.

The result display / editing unit 9 displays the syntactic analysis result syntactically analyzed by the syntactic analyzing unit 7 and the syntactic analysis rule 8 used, and, in response to the modification of the syntactic analysis rule 8, the modified syntactic analysis. The result is displayed.

[0012]

In the present invention, as shown in FIG.
Morphological analysis is performed one sentence at a time from a large amount of input text, and the sentence pattern extraction unit 3 applies sentence pattern pattern extraction rules 4 to the morphological analysis result to extract sentence patterns 5, and the sentence pattern integration unit 6 extracts these sentence patterns. A set of pattern 5 is collected and integrated to generate a sentence pattern.

Also, for the natural language sentence, the syntax analysis unit 7
Applies the previously created syntactic analysis rule 8 to perform syntactic analysis processing, displays the syntactic analysis result, and applies the syntactic analysis rule 8 used to obtain the syntactic analysis result to the relevant part of the syntactic analysis result. Correspondingly displayed, the result display / editing unit 8 extracts and displays the integrated sentence pattern regarding the natural language sentence together, and responds to the correction instruction of the syntax analysis rule 8 used for the displayed syntax analysis result. , Passes the corrected syntactic analysis rule 8 to the syntactic analysis unit 7 when it is adapted to the integrated sentence pattern pattern, causes the syntactic analysis process to be performed, displays the syntactic analysis result, and is used to obtain the syntactic analysis result. The syntax analysis rule 8 is edited and displayed repeatedly to edit the syntax analysis rule 8.

Therefore, a large amount of text in the target range is prepared, the sentence pattern is extracted and integrated over the entire text, or the syntactic analysis rule 8 is modified by referring to the sentence pattern to obtain a desired syntactic analysis result. It is possible to automatically collect all sentence patterns from a large number of sentences by editing so as to generate, and to easily revise and create parsing rules with the support of sentence patterns. Is possible.

[0015]

EXAMPLE Next, the operation of the configuration of FIG. 1 when automatically extracting sentence pattern from a large number of sentences will be described in detail with reference to FIG. 2 (a) and FIGS. 3 to 6.

FIG. 2A shows a flowchart for automatically extracting sentence pattern. In FIG. 2A, S1 is
Take in one sentence. This is to fetch one sentence from a large amount of texts (sentences) from the text base unit 1 of FIG. 1, for example, one sentence from a large amount of texts (sentences) in a specific field designated by the selection condition.

In S2, morphological analysis is performed. This is S1
Perform morphological analysis of the captured sentence. For example, one sentence "Parent
The slot of Russ is called the parent slot. Morphological analysis
If it is divided further, it will not make sense.
It Here, as shown in FIG.Parent class of slot To Parent slot When Call
 . Split as underlined.

In step S3, a sentence pattern is extracted. this is,
The sentence pattern extraction rule 4 is applied to the result of the morphological analysis in S2 to extract the sentence pattern. For example, the sentence pattern extraction rule 4 in FIG. 4 is repeatedly applied to the result of the morphological analysis in FIG.
As shown in (c), (d), and (e), the sentence pattern pattern of (e) in FIG. 3 that can no longer be applied is extracted. For example, sentence pattern pattern extraction rule 4 is repeatedly applied to the result of the morphological analysis of FIG.
The replacement is performed as shown in (c), (d), and (e). This will be specifically described below.

(1) Since the parent class is "ordinary noun", it is replaced with "noun class", and the ordinary noun is expressed by an arbitrary surface expression (*), resulting in (noun class, *) replace.

[0020] (2) is given, because it is "particle" replaced with "particle", and further, to represent the particle as it is "no", as a result (particle,
"No").

(3) Since the slot is a "common noun", it is replaced with "noun class", and the common noun is expressed with an arbitrary surface expression (*), and as a result, replaced with (noun class, *). .

(4) By these (1) to (3), as shown in (b) of FIG. 3, it is replaced with (noun class, *) (particle, "no") (noun class, *).

(5) In contrast to (4), (nouns, *) (particles, in the sentence pattern pattern extraction rule of FIG. 4A)
[No]) (nouns, *) → (nouns, N)
(Nouns, N) (see (c) and (d) of FIG. 3).

[0024] (6) Similarly, replacing the was morphological analysis (the particle, "Wo").

(7) Similarly, the morphologically analyzed parent slot is replaced with (noun class, N).

[0026] (8) Similarly, replacing the capital that was morphological analysis (the particle, "and").

(9) Similarly, the morphologically analyzed call is replaced with (verb "call").

(10) By replacing the above (1) to (9), as shown in (d) of FIG.
N) (particles, [wa]) (nouns, N) (particles, "to")
(Verb, "call")

(11) Since (10) is a definition sentence,
N is called N. And the sentence pattern that you want is automatically extracted.

In step S4, a sentence pattern corresponding to one sentence is stored. This is stored as the sentence pattern 5 in FIG.
In S5, it is determined whether it is the end. This determines whether or not the extraction of the sentence pattern is completed for a large number of sentences in each field, for example. YE
If S, go to S6. In the case of NO, the process returns to S1, the next one sentence is fetched, and S2 and subsequent steps are repeated.

In step S6, the sentence types are summarized. this is,
The sentence type patterns accumulated in S4 are summarized for each sentence type (for example, definition sentence, method sentence, etc.). S7 integrates the summarized results. This is a set of sentence type patterns collected for each sentence type, for example, four of the set of definition sentences in FIG.
4 are integrated and integrated into “,” in FIG. 6 (see the description of FIG. 6).

As described above, morphological analysis is performed on a large number of sentences in each field one by one, the sentence pattern pattern extraction rule of FIG. 4 is applied to extract the sentence pattern pattern of FIG. 5, and the set of sentence pattern patterns of FIG. The integrated sentence pattern of FIG. 6 is automatically generated.
This makes it possible to automatically extract sentence patterns from a large number of sentences.

FIG. 3 is a diagram illustrating a specific example of the present invention. FIG. 3A shows the result of morphological analysis. The underlined portion indicates the character string that has been divided by morphological analysis (a unit that makes no sense if further divided).

FIG. 3 (b) shows that among the results of the morphological analysis, the slots of the parent class are based on the common noun "parent class", the particle "no", and the common noun "slot" (nouns, * )(Particle,
It shows a state in which "no") (nouns, *) is replaced.

FIG. 3C is different from FIG. 3B in that
The result (noun class, N) which replaced by applying the sentence pattern extraction rule of FIG. 4 is shown.

FIG. 3D is similar to FIG. 3C,
The results (nouns, N) (particles, [wa]) (nouns, N) (particles, “to”) (verbs, “call”) of the sentence pattern extraction rules of FIG. 4 replaced are shown.

In (e) of FIG. 3, (d) of FIG. 3 is the definition sentence, and hence N is referred to as N. It is shown that it is converted to and replaced with a sentence pattern.

FIG. 4 shows an example of the sentence pattern extraction rule of the present invention. Here, * represents anything. For example, * in (nouns, *) indicates that any noun is acceptable.

N is a symbol representing a noun. Is a sentence pattern pattern extraction rule when applied to (b) of FIG. 3 and replaced with (c) of FIG.

FIG. 5 shows an example of the extracted sentence pattern according to the present invention. This is a set of sentence pattern patterns automatically extracted by applying the sentence pattern extraction rules of FIG. 4 by morphologically analyzing the sentence as shown in FIG. Here, when the four definition sentences of and are integrated, they are summarized into the definition sentences of'and 'in FIG.

FIG. 6 shows an example of the integrated sentence pattern according to the present invention. As described above, this is a sentence pattern in which the four sentence pattern patterns of FIG. 5 are integrated and integrated. [] In the integrated sentence pattern is a symbol indicating that it may or may not be present.

Is a symbol representing or (any one). Therefore, in the integrated sentence pattern of'of FIG.
N, excluding those enclosed in [], which may or may not be present
Is referred to as N. Since | is or, this is: ・ N is called N.

* N is called N.・ N is called N. Will be represented.

Next, FIG. 2B and FIGS. 7 to 9
The operation of the configuration shown in FIG. 1 for displaying the parsing result and editing the parsing rule will be described in detail using. FIG. 2B shows an editing flowchart of the syntax analysis rule.

In FIG. 2B, S11 fetches one sentence. This takes in one sentence from the text (sentence) group to be parsed from the text base unit 1 in FIG.

In S12, the syntax is analyzed. This is S11
After morphological analysis of one sentence captured by
Is applied for parsing. S13 displays the syntax analysis result. For example, as shown in FIG. 8 described later,
The syntax analysis result and the syntax analysis rule 8 applied at that time are displayed in association with each other on the screen.

In S14, the user determines whether the syntactic analysis result displayed in S13 is good. In the case of YES, the syntactic analysis result of one sentence is good, so that the process returns to S11 to fetch the next one sentence. On the other hand, in the case of NO, since the syntactic analysis result of one sentence is not good, the user corrects the syntactic analysis rule 8 in S15, that is, the syntactic analysis result displayed on the screen in FIG. The syntactic analysis rule 8 used for the above and the sentence pattern corresponding to the sentence at this time are displayed together, and the user compares the two so that the syntactic analysis result is good, and the syntax used at this time is displayed. Parsing rule 8
To fix.

In S16, the syntactic analysis rule 8 corrected in S15 is collated with the sentence pattern 5, that is, collated with the sentence pattern of the sentence to be syntactically analyzed to determine whether or not the sentence is correct. In the case of YES, since it is suitable, S12
The syntactic analysis is performed again using the corrected syntactic analysis rule 8, and the syntactic analysis result and the syntactic analysis rule used at that time are associated and displayed in S13 on the screen.
Then, if the result of S14 is good, the corrected parsing rule 8 is stored. Then, the process returns to S11 to proceed to fetch the next one sentence. On the other hand, if NO in S17, a message prompting the user to redo the correction of the syntax analysis rule is displayed again, and the process returns to S15.

As described above, the natural language sentences are captured one by one, the syntactic analysis rule 8 is applied, the syntactic analysis result and the syntactic analysis rule 8 used at that time are associated and displayed on the screen, and the syntactic analysis is performed by the user. When it judges that the result is good, it proceeds to fetch the next one sentence. On the other hand, when it judges that the result of parsing is not good, it compares the sentence pattern pattern corresponding to the sentence with the parsing rule 8 used, and The user modifies the parsing rules to obtain the desired parsing results. Then, parsing is performed using the corrected parsing rule, the parsing result and the used parsing rule are displayed in association with each other. If the parsing rule is good, the parsing rule has been corrected, and this is saved. . The details will be sequentially described below.

FIG. 7 shows a modification flowchart of the parsing rule of the present invention. In FIG. 7, S21 displays the analysis result. This graphically displays the syntactic analysis result generated by applying the syntactic analysis rule 8 created in advance, as shown by the syntactic analysis result of FIG.

In S22, the related sentence pattern is extracted from the accumulated sentence patterns using the type of the example sentence as a key and displayed. This extracts and displays the integrated sentence pattern including the sentence pattern extracted by applying the sentence pattern extraction rule 4 as described above using the type of the example sentence (eg, definition sentence) of FIG. 8 as a key.

In step S23, the parsing rules related to the parsing rule 8 displayed in the parsing result are used as keys to extract the related parsing rules and display them side by side. This is the same as the syntax analysis rule used in FIG. 8 obtained by extracting the syntax analysis rule used for the syntax analysis result displayed in S21.

In step S24, the user (sentence pattern and syntactic analysis rule 8) is compared and examined to create a better syntactic analysis rule, and the syntactic analysis is performed based on this to correct the syntactic analysis result.

In step S25, the corrected syntactic analysis result is displayed. These S24 and S25 are used, for example, in the syntactic analysis result of FIG. 8 to correct the syntactic analysis rule 8, perform syntactic analysis again, and correct and display the syntactic analysis result of FIG.

In S26, it is determined whether it is OK. This is S2
It is determined whether the syntax analysis result corrected and displayed in step 5 is OK.
In the case of YES, the parsing rule 8 is saved because it has been corrected, and the process ends. On the other hand, if NO, S23
Return to and repeat.

As described above, the syntactic analysis rule 8 is applied to the natural language sentence, the syntactic analysis is performed, the syntactic analysis result is displayed, and the syntactic analysis rule used at that time is displayed in association with the natural language sentence. The sentence pattern pattern of is also displayed, and the user compares and examines both the sentence pattern pattern and the parsing rule used, and corrects the parsing rule. And
Repeat the parsing again using the modified parsing rule and display the parsing result and the used parsing rule, and modify the parsing rule to obtain the desired parsing result. , Save the parsing rule when you can modify it. As a result, it became possible to easily modify and create the parsing rules necessary for parsing by referring to the sentence pattern.

FIG. 8 shows an example of the syntax analysis result of the sentence of the present invention. Here, the upper part of the column of the syntactic analysis result shows the input natural language sentence, the underline shows the result of the morphological analysis, and the third and subsequent lines show the tree structure of the syntactic analysis result and the syntactic analysis rule used at that time. No. of

The used parsing rule column indicates the parsing rule corresponding to the parsing rule number used in the parsing result column. The sentence pattern indicates a sentence pattern of a natural language sentence.

FIG. 9 shows an example of the corrected syntactic analysis result of the sentence of the present invention. This shows the generated syntactic analysis result in response to the fact that the syntactic analysis result of FIG. 8 is undesired and the user has modified the syntactic analysis rule as shown by referring to the sentence pattern. In the figure, (10), (1
1) corresponds to the same number as the corrected parsing rule. Since the parsing result of FIG. 9 is OK, the parsing rule used at this time is saved.

[0060]

As described above, according to the present invention,
Edit so that a large amount of text in the target range is prepared, sentence pattern patterns are extracted and integrated over the entire text, or parsing rule 8 is modified by referring to the sentence pattern to generate a desired syntactic analysis result. Because it adopts a configuration that allows you to automatically collect all sentence pattern from a large amount of natural language sentences, you can easily modify and create parsing rules with the support of sentence pattern. You can

[Brief description of drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 is a flowchart for explaining the overall operation of the present invention.

FIG. 3 is a diagram illustrating a specific example of the present invention.

FIG. 4 is an example of a sentence pattern pattern extraction rule of the present invention.

FIG. 5 is an example of an extracted sentence pattern according to the present invention.

FIG. 6 is an example of an integrated sentence pattern according to the present invention.

FIG. 7 is a flowchart for modifying the parsing rule of the present invention.

FIG. 8 is an example of a syntactic analysis result of a sentence according to the present invention.

FIG. 9 is an example of a corrected syntactic analysis result of the sentence of the present invention.

[Explanation of symbols]

 1: Text base part 2: Morphological analysis part 3: Sentence pattern pattern extraction part 4: Sentence pattern pattern extraction rule 5: Sentence pattern pattern 6: Sentence pattern pattern integration part 7: Syntax analysis part 8: Syntax analysis rule 9: Result display / editing part

Claims (2)

[Claims] 1. A morphological analysis unit (2) for morphologically analyzing a large amount of text in a sentence pattern extracting apparatus for automatically extracting sentence patterns from text.
And a sentence pattern extracting unit (3) for extracting sentence patterns by applying a sentence pattern extraction rule (4) created in advance to the result of morphological analysis by the morphological analyzing unit (2).
And a sentence pattern integrating unit (6) for integrating a set of sentence patterns extracted by the sentence pattern extracting unit (3). 2. A natural language sentence is subjected to a syntactic analysis process by applying a syntactic analysis rule (8) created in advance, and the syntactic analysis result is displayed and the syntactic analysis used to obtain the syntactic analysis result. The syntactic analysis unit (7) for displaying the rule (8) in association with the corresponding portion of the syntactic analysis result and the integrated sentence pattern related to the natural language sentence are extracted and displayed together, and used. In response to the modification instruction of the syntax analysis rule (8), it is determined whether or not the integrated sentence pattern is matched, and when matched, the modified syntax analysis rule (8) is used as the syntax analysis unit (7). ), The parsing process is performed, the parsing result is displayed, and the parsing rule (8) used to obtain the parsing result is displayed in association with the parsing rule (8). ) The sentence pattern extracting device according to claim 1, further comprising a result display / editing unit (9) for collecting and editing the syntax analysis rule (8).