JPH05101120A - Document abstract generating device - Google Patents

Abstract

PURPOSE:To obtain a document abstract generating device which automatically generate abstract sentences while maintaining context as to all the sentences of a Japanese document. CONSTITUTION:The document abstract generating device consists of a morpheme analysis part 1 which divides an input source sentence into constituent words and phrases, a context analysis part 2 which analyzes connection relation between the phrases divided by the morpheme analysis part 1, a document abstract generation part 3 which generates abstract according to the processed result in the context analysis part 2, an abstract display part 4 which generates an abstract according to the processing result of the context analysis part 2. And an abstract display part 4 which displays the processing result of the document abstract generation part 3. Further, the document abstract generation part 3 is provided with means that specifies the type of the connection of a phrase which is not marked, means which specifies a phrase as a header and marks the header, means which marks phrases except the connection type specified by said means in the phrase regarding the header, and means which decides a phrase including declinable word as a header in the marked phrase.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sentence abstract generation apparatus for automatically generating an abstract while maintaining the context of Japanese sentences.

[0002]

2. Description of the Related Art Conventionally, as a technique in such a field,
Reference 1; Ota, Matsuda, Inudo, "Attempt of Automatic Abstraction by Text Structure Model", Proceedings of National Conference of the Japanese Society for Artificial Intelligence, 1989.
Year, p. 411-414, Document 2; JP-A-02-112.
There was one described in Japanese Patent No. 069.

The method described in the above document 1 is a method of generating an abstract of an important portion using surface information of a sentence. That is, a sentence is divided into sentence constituent elements, the main verb of the sentence is focused, and the connection relation between sentences (logical relation between sentences)
Is a method of generating an abstract of an important sentence or a part of a sentence in a sentence.

Further, the method described in the above-mentioned document 2 is a method of generating an abstract by obtaining in advance the subject or the core important word (keyword) of the description content of a sentence and extracting the sentence containing this important word. Is.

[0005]

However, in the above-described abstract generation method, only the portion of the text that is judged to be important is extracted. Therefore, the part to be extracted is one sentence or a part of one sentence, and it is extracted in a scattered manner. Although it is grammatically correct, the context of the original sentence is ignored.

Further, it is difficult to judge whether or not a sentence is really important, and thus an important sentence may be overlooked and a less important sentence may be extracted. Furthermore, there is a problem that the judgment as to what is important in a sentence differs depending on the reader. On the other hand, the method using the important word has a problem that the important word cannot be set unless the content of the sentence is known in advance.

The present invention has been made in view of the above points, and it is an object of the present invention to solve the above problems and to provide a sentence abstract generation apparatus that generates an abstract sentence while maintaining the context for all sentences in a sentence. And

[0008]

In order to solve the above-mentioned problems, the present invention is directed to a sentence abstraction generating apparatus, which comprises a morphological analysis unit for dividing an input original sentence into constituent words and phrases, and a phrase divided by the morphological analysis unit. Consists of a syntactic analysis unit that analyzes the dependency relations between clauses, a sentence abstract generation unit that generates an abstract based on the processing results in the syntactic analysis unit, and an abstract display unit that displays the processing results in the sentence abstract generation unit. is doing. Further, in the sentence abstract generation unit, a means for designating a relation type of a phrase which is not marked, a means for designating a clause in a header and marking the header, and a relation type of the phrase among the clauses related to the header. There is provided a means for marking a clause excluding the designated relation type to the means for designating a clause, and a means for making a clause including a phrase among the marked clauses as a header. In the sentence abstract display section, a means for displaying only the marked phrases in the input original text,
A means for highlighting the marked phrases in the input original sentence is provided.

[0009]

According to the present invention, since the sentence abstraction generation device is configured as described above, it is possible to generate the abstract sentence while maintaining the context for all the sentences in the sentence. Therefore, the above problem can be solved.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a system configuration diagram showing an embodiment for realizing the text abstract generating device of the present invention. 1 is a morphological analysis unit, 2 is a syntax analysis unit, 3 is a sentence abstract generation unit, 4
Is an abstract display section. Hereinafter, description will be given with reference to the drawings.
First, the original sentence written in Japanese is input to the morphological analysis unit 1. The morphological analysis unit 1 divides the input original sentence into constituent words and clauses. Figure 2 shows the processing result when the original sentence "One ship with sails gradually appeared from the far sea and headed for the port." Was input to the morphological analysis unit 1. Here, “/” indicates a break between constituent words, and “//” indicates a break between phrases.

The syntactic analysis unit 2 inputs the processing result of the morphological analysis unit 1 and analyzes the dependency relation between the clauses. That is, it analyzes which bunsetsu the bunsetsu is related to. FIG. 3 shows the processing result in the syntax analysis unit 2. In the figure, arrows indicate which bunsetsu the bunsetsu is related to.

The text abstract generation unit 3 extracts a phrase representing the meaning of a sentence based on the processing result of the syntax analysis unit 2 and marks the phrase. Then, by extracting the marked clauses in order from the beginning of the sentence, a context-preserving abstract is generated. The procedure to mark a phrase is from (a) to
It is (d). (A) The end of the sentence is the head. (B) Mark the head, and also mark the clauses other than the clauses of the relation type specified in advance among the clauses related to the head. (C) If there is a phrase including a declination among the marked phrases, that phrase is newly set as the head. (D) (b) and (c) are repeated until there are no more phrases related to the head.

FIG. 4 shows the flow of processing in the text abstract generation unit 3, and FIG. 5 shows the processing steps. First, the relationship type to be deleted is designated in step S1 of FIG. As the engagement type, as shown in FIG. 5A, “mo”, “f”,
"Wo", "kara", "adverb or adverb phrase", "conjunction",
"Continuous", "connector", etc. The engagement type "Ga" is
This is the case when "I ..." is involved. Similarly,
"F" means "to the port ...", "Wo" means "the book ...", "Kara" means "From Tokyo ...", "Adverb or adverb phrase" means "Dandant ...", "Conjunction""But...","Continuoususe"
Is the case of "representation ...", and "relation" is "far ...". The circle mark in the lower column of the relationship type in FIG. 5A indicates that the relationship type is designated to be deleted. Here, "adverb or adverb phrase" is designated as a non-marking type.

Next, in step S2, the dependency list 100 and the dependency type 101 shown in FIG. 5B are created based on the analysis result of the syntax analysis unit 2. The contact list 100
Indicates which bunsetsu the bunsetsu is related to.
01 indicates the type of the clause. Next, the reverse contact list 102 is created based on the contact list 100. The reverse reference destination list 102 indicates which number of clauses the clause is related to. For example, the twelfth bunsetsu "Go to." Indicates that the eleventh bunsetsu "To Minato" and the tenth bunsetsu "Akira," are involved.

Next, according to the procedure (a) of marking the phrase, the sentence-end phrase is set to the head in step S3. Therefore, "headed" becomes the head. In FIG. 5B, the circle in the head column 103 of the phrase number 12 indicates that the phrase is a head. Next, in step S4, the clauses relating to the elements of the head reverse relation list 102 are checked one by one. Head No. 11 and No. 1 on the opposite list of "Headed."
Since there are two clause numbers 0, we first check 11th. In steps S5 and S6, according to (b) of the procedure for marking the phrase, “it headed” is marked, and the relation type of the 11th phrase “to the port” is “he” and the mark is made. Do not specify it as a clerk type, so mark it. In step S7,
According to the procedure (c) of marking the phrase, it is checked whether the phrase includes a verb. Since the 11th clause does not include a noun, the process returns to step S4 to check the next 10th clause.

The tenth clause "represent," is also marked because it is not designated as a non-marking relation type. Next, since "reveal," is a verb, it is set to the head in step S3. The clause numbers 9, 8, 7, and 3 are described in the reverse reference list 102 of the head "represent,". First of all, the phrase number "look" is examined. Mark "figure" because it is not a special type that specifies deletion. Also, "figure" does not include a verb, so check the following clauses. The next clause number 8 "Dandanto" is an adverb, so it is not marked. In the same manner, by checking the clauses according to the procedure for marking the clauses, the clause number 7 "Funa ga" and the clause number 3 "beyond" are marked. Since all the clauses relating to the head "revealed" and the head "headed" have been examined, the marking process ends.

In the abstract display unit 4, based on the result of the marking process in the sentence abstract generation unit 3, only the clauses marked in the input original sentence are displayed or marked in the input original sentence. The highlighted phrase is displayed.
Display examples are shown in FIGS. 6 (a) and 6 (b). FIG. 6A shows
Only the clauses marked by the sentence abstract generator 3 are displayed. FIG. 6B shows the entire original sentence. Here, in particular, the bunsetsu marked by the text abstract generation unit 3 is emphasized by the shading process. In order to emphasize the phrase, a method such as changing the display color or drawing an underline may be used in addition to the shading processing of FIG.

[0018]

As described above in detail, according to the present invention, since the abstracts are generated for all the sentences in the sentence, it is possible to generate the abstracts while keeping the context, and It is possible to provide a sentence abstract generation device capable of deleting an unnecessary phrase from an abstract by designating a relation type of an unnecessary phrase.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of a text abstract generation device according to an embodiment of the present invention.

FIG. 2 is a diagram showing an input sentence and a processing result in a morphological analysis unit.

FIG. 3 is a diagram showing a processing result in a syntax analysis unit.

FIG. 4 is a diagram showing a flow of processing in a text abstract generation unit.

FIG. 5 is a diagram showing a processing process in a text abstract generation unit.

FIG. 6 is a diagram showing a display example on an abstract display unit.

[Explanation of symbols]

 1 Morphological analysis unit 2 Syntax analysis unit 3 Text abstract generation unit 4 Abstract display unit

Claims (2)

[Claims] 1. A sentence abstraction generation apparatus for inputting an original sentence and automatically generating an abstract of the original sentence, a morphological analysis unit for dividing an input original sentence into constituent words and phrases, and a phrase divided by the morphological analysis unit. The parsing unit that analyzes the dependency relations between clauses, the sentence abstraction generation unit that generates a sentence abstract based on the dependency relations between the clauses, and the processing result of the sentence abstraction generator are displayed. An abstract display unit, wherein the sentence abstract generation unit further specifies means for specifying a relation type of a clause not to be marked, means for specifying a clause in a header, and marking the header, and a clause for the header. And a means for marking a clause excluding the specified clause type in the means for designating the clause type of the clause, and a means for using a clause containing a declination as a header among the marked clauses. A text abstract generation device characterized by comprising: 2. The abstract display unit has means for displaying only the verses marked in the input original sentence, and means for displaying the verses marked in the input original sentence in an emphasized manner. The text abstract generation device according to claim 1.