JP2638187B2 - Text summarization device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text summarization apparatus for obtaining a summary text of a desired text as a means for quickly grasping the content of a desired text from a large number of documents in a document search or the like.

2. Description of the Related Art Conventional text summarization methods include a method of preparing a certain frame in advance and filling its slots, and a method of story-level grammar by applying a story-level grammar to a story-like sentence. The method of extracting the core sentence and limiting the sentence to those that can be logically expanded, analyzing the connection between sentences, expressing the structure of the sentence as a binary tree, and assigning it to each connection There was a method of scanning a binary tree from weights to obtain a summary sentence.

In the text summarization used for document search or the like, it is essential that information desired by the user be quickly and easily extracted regardless of the content of each text. For this purpose, it is necessary to realize high-speed processing by reducing the burden of sentence analysis and summary sentence generation, and at the same time, to control the quality and quantity of the summary sentence according to the user's request.

In the conventional method using frames, it is not practical to deal with a wide range of sentences because it is necessary to prepare a large number of frames according to the description contents. Moreover, the method based on the story grammar and the connection between sentences requires a high level of semantic analysis, so that the burden of sentence analysis is large and not suitable for high-speed processing.

SUMMARY OF THE INVENTION The present invention is to solve the above-mentioned problem. First, it provides a text summarization apparatus which has a small burden of sentence analysis and generation of a summary sentence and can be applied to a wide range of texts regardless of the description content of each text. The purpose is to:

Another object of the present invention is to provide a text summarization apparatus capable of controlling the quality and quantity of a summary sentence according to a user's request.

Means for Solving the Problems In order to achieve the above object, a text summarizing apparatus according to the present invention firstly comprises an input section for inputting a text and a type of a text classified according to the intention / purpose of writing the text. A sentence analysis unit that performs a syntax analysis of each sentence of the sentence input by the input unit, a sentence form selection criterion that specifies a sentence form that constitutes a summary sentence for each sentence type, vocabulary information, syntax information, A selection criterion database storing a selection criterion including a criterion for determining a sentence form of a sentence described by using at least one of the information on the concatenation relation, and a sentence constituting a summary from a sentence input from the input unit And a summary sentence display unit for displaying the selected sentence, wherein the summary sentence selection unit responds to the syntax analysis result of the sentence analysis unit by inputting from the input unit. Corresponding to the type of sentence The sentence form selection criterion and the criterion for judging the sentence form specified by the sentence form selection criterion are referred to, and it is determined whether or not the corresponding input sentence is a sentence constituting the summary. It is configured to select a constituent sentence.

Secondly, in the first means, the selection criterion stored in the selection criterion database has a degree of importance, and the summary sentence selecting unit selects a sentence constituting a summary based on the importance.

Operation The operation in the above configuration is as follows. Various texts have a special circling or syntactic structure depending on the purpose, purpose, etc. of the text, regardless of the nature of the text, that is, the specific content of the text. Therefore,
These special elements are set as selection criteria in advance, and the information obtained by parsing the input sentence is compared with the selection criteria to extract and select those considered important, so that individual sentences can be selected. Irrespective of the specific contents of the description, a summary sentence can be obtained by processing a wide range of sentences at high speed.

In the second means, the summary sentence selection unit selects a sentence that achieves the summary based on the importance of the selection criterion, so that the quality and quantity of the summary sentence can be adjusted according to the user's request. .

Embodiments Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a conceptual diagram of a document summarizing apparatus according to an embodiment of the present invention. The text to be summarized should be English text. Input device 1
The sentence input from the input sentence is subjected to syntactic and semantic analysis by the sentence analysis device 2 and is converted into an internal expression representing the content of each sentence of the input sentence. The summary sentence selection device 3 receives a sentence type indicating the nature of the sentence specified by the user together with the input sentence, and obtains a selection criterion corresponding to the type from the selection criterion database 4.

The selection criterion stored in the selection criterion database 4 is determined as follows. A variety of sentences, such as those that describe the author's certain attitudes, such as assertions or questions,
It has special words or phrases, or has a special syntactic structure. The request has a sentence structure that describes the author's request. These are special vocabulary information,
It can be represented as syntax information or information on connection between sentences. In the selection criterion database 4, the nature of the sentence such as "claim", "question", "request", etc., that is, regardless of the specific content of the sentence, For each type of sentence that is classified, what form of sentence is important, and what other form of sentence is selected to obtain a summary sentence are described. A criterion for judging the form of the sentence is described in the form of the vocabulary information, the syntax information, the information of the connection relationship between the sentences, and the like. The importance is added to the sentence form and the determination information.

By calculating the importance for all sentences using the importance added to the selection criterion, it is possible to select sentences with a value greater than or equal to the importance according to the level value corresponding to the quantity and quality of the summary sentence The amount and quality of the summary can be controlled as needed. Details will be described later.

The summary sentence selection device 3 selects a sentence considered to be important by comparing the selection criterion from the selection criterion database 4 with the internal expression of each sentence of the input sentence obtained from the sentence analysis device 2. The level of the quality and quantity of the summary sentence is given as a default value, and the sentence to be selected is determined in consideration of this level value. By changing the default level, it is possible to control the quality and quantity of the summary sentence and obtain a summary sentence that suits the user's purpose.

FIG. 2 is an internal configuration of the selection criterion database 4 of FIG. A sentence form selection reference unit 7 describing what form a sentence is to be selected for each property of the sentence, and a determination information unit 8 describing information for determining the form for each sentence form Has become. The database is divided into a sentence form selection criterion section 7 and a judgment information section 8, so that when the form of a sentence to be selected is duplicated for some sentence properties, the judgment criterion is shared and duplicate descriptions are avoided. Can be.

FIG. 3 shows the internal configuration of the sentence form selection reference section 7 of FIG. For each character of the sentence, the form of the sentence, such as a claim or a problem, which is deemed necessary to obtain a summary from the sentence is described. Each of these sentence forms is given a degree of importance a, b, c,... Taking into account the nature of the sentence.
Even in the same sentence form, the value of importance differs depending on the nature of the requested sentence.

For example, sentences in which it is important to capture the author's claim, such as editorials and editorials, are referred to as "claims" as a document type, and sentences in which it is important to capture the requirements of the author's reader are referred to as "sentence type". Request ”.

And the type of sentence, that is, the nature of the sentence is "claim"
If so, select a statement that states the claim or conclusion, a statement that raises the problem that caused the claim or conclusion, or a statement that states the basis of the claim or conclusion. Would yield a summary sentence. At this time, the value of the importance statement is set low because the statement that raises the problem or the grounds serves as a supplementary role to the statement that states the assertion or conclusion.

FIG. 4 is an internal configuration diagram of the determination information unit 8 of FIG. For each sentence form, lexical information, syntactic information, and information on connection between sentences for determining the sentence form are described. The importance of each piece of information
s, t,... are added. At least one of these pieces of information is used for selection.

For example, in the case of a sentence form expressing an assertion, as lexical information, use is made of an auxiliary verb expressing an assertion such as "should" or "must" or appearance information of a phrase such as "We hope". As the syntactic information, information such as an emphasis syntax expressing an assertion is used. Concatenation relation information between sentences includes `` and '' and `` bu
Information such as sentences connected by a conjunction such as "t" is used. As importance, for example, a sentence that is an emphasized syntax is "shoul
The value of importance is set high, considering that the statement expresses a stronger assertion than the sentence where "d" appears. The importance of the connection relation information is dynamically assigned a value by the summary sentence selection unit 2 in consideration of the importance of the sentence in the connection relation.

FIG. 5 shows a conceptual diagram of the processing of the summary sentence selection device 3 of FIG. The selection criterion extracting unit 9 obtains information on the sentence form for obtaining a summary sentence corresponding to the nature of the input sentence from the sentence form selection criterion unit 7 of the selection criterion database 4 and the judgment information unit 8 of the selection criterion database 4. The information for determining the form of the sentence is extracted. The importance calculation unit 10 compares the selected information with the internal expression of each sentence obtained from the sentence analysis device 2. When a sentence matches the judgment information, the importance of the sentence is determined in consideration of the importance of the form of the sentence and the importance of the information used for the judgment information. The importance of the sentence selected in accordance with the concatenation relationship information is calculated from the importance of the original sentence that has formed the concatenation relationship. Give a value higher than the maximum value. A sentence that does not match any of the pieces of judgment information has the lowest importance, and assigns some importance to all sentences in the input sentence. The level-based selection unit 11 knows a level value corresponding to the quality and quantity of the summary sentence from the input device 1.
At this time, if there is no input of a level value, a default value is used. According to this level value, a sentence whose importance is equal to or higher than a certain value is selected. The range of the importance corresponding to the level value is obtained from the quality and quantity of the summary sentence to be selected and the distribution of the value of the importance. For example, assuming that the level value 1 corresponds to the request for the amount “the summary sentence is reduced to half of the whole sentence”, the selected amount is reduced to half of the whole sentence from the distribution of the importance value. A value of such importance is obtained, and a sentence higher than that value is selected as a summary sentence. Note that the range of importance corresponding to the level value can be set in advance, but in this case, it is possible to select all sentences or conversely not to select them at all. It is.

Returning to FIG. 1, the summary sentence selected by the summary sentence selection device 3 in this way is supplied to the output device 6, and is displayed on the input sentence by highlighting it by color coding or bolding. This highlight display can support so-called oblique reading.

In addition, in response to the request of the user, the summary sentence generation device 5
From the internal representation of the sentence by the sentence analyzer 2, a Japanese translation corresponding to the selected sentence is generated and displayed on the same screen by the output device 6.

Effect of the Invention As described above, the present invention provides a sentence summarizing method and apparatus for selecting a sentence considered to be important by comparing the result of syntactic and semantic analysis of each sentence of an input sentence with a predetermined selection criterion. If the texts have the same properties, they can be applied to a wide range of texts because they do not depend on the description of each text. In addition, since it is a criterion that can be determined even in shallow sentence analysis, the burden of sentence analysis is small, and the processing load for generating a summary sentence only by selection is small. Therefore, high-speed processing is possible.

Furthermore, adding or deleting selected sentences can be performed at high speed by giving importance to the entire input sentence when selecting a summary sentence.
You can interactively adjust the quality and quantity of the summary sentence. Further, if only the information about the vocabulary is used as the information of the selection criterion, a higher-speed processing can be performed without performing the parsing processing.

[Brief description of the drawings]

FIG. 1 is a block diagram of a text summarizing apparatus according to an embodiment of the present invention;
FIG. 2 is an internal configuration diagram of a selection criterion database, FIG. 3 is an internal configuration diagram of a sentence form selection criterion portion of the selection criterion database, FIG. 4 is an internal configuration diagram of a judgment information portion of the selection criterion database, and FIG. It is a conceptual diagram of a process of a summary sentence selection device. DESCRIPTION OF SYMBOLS 1 ... Input device, 2 ... Sentence analysis device, 3 ... Summary sentence selection device, 4 ... Selection criterion database, 5 ... Summary sentence generation device, 6 ... Output device, 7 ... Sentence form selection reference portion , 8 ...
... Judgment information section, 9 ... selection criterion extraction section, 10 ... importance calculation section, 11 ... level selection section.

 ──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Takao Fukushige 1006 Kazuma Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. (56) References JP-A-61-100861 (JP, A) JP-A-63- 298563 (JP, A) Information Processing Society of Japan Research Report, Vol. 87, No. 68, p. 63.6.1-63.6.8, 1987 Information Processing Society of Japan Research Report, Vol. 87, No. 84, p. 85-91, 1987

Claims (3)

(57) [Claims] An input unit for inputting a sentence and a type of the sentence classified according to the intention and purpose in which the sentence is written, and a sentence analyzing unit for performing a syntax analysis of each sentence of the sentence input by the input unit And a sentence form selection criterion that specifies a sentence form constituting a summary sentence for each sentence type, and a sentence form of the sentence described using at least one of vocabulary information, syntax information, and information on connection relations is determined. A selection criterion database storing a selection criterion including a criterion to be determined, a summary sentence selection unit for selecting a sentence constituting a summary from sentences input from the input unit, and a summary sentence display unit for displaying the selected sentence The summary sentence selection unit is configured to specify a sentence form selection criterion corresponding to the type of the sentence input from the input unit with respect to the syntax analysis result of the sentence analysis unit, and the sentence form selection criterion. Judge the sentence form Article Summary apparatus characterized by criterion a reference to the input sentence corresponding to select the sentence that constitutes the summary by determining whether the sentence constituting an abstract that will. 2. The method according to claim 1, wherein the selection criterion stored in the selection criterion database has a degree of importance, and the summary sentence selection unit selects a sentence constituting a summary based on the importance.
Document summarization device as described. 3. The system according to claim 1, further comprising: means for translating the sentence selected by the summary sentence selection unit into another language, wherein the summary sentence display unit displays the translated sentence. The document summarizing apparatus according to claim 1.