JPH02297157A - Method and device for summarizing text - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Industrial applications The present invention relates to searching a large number of documents in a single document search.

Conventional technology Conventional text summarization methods include preparing a certain frame in advance and filling in the slots, and applying text-level rules known as narrative grammar to sentences with a narrative structure. How to extract the core sentences,
Limit sentences to those that develop logically, analyze the connections between sentences, express the structure of the sentence in a binary tree, and scan the binary tree based on the weight given to each connection to summarize. There were ways to get sentences.

In text summaries used for searching documents for inventions or problems to be solved, it is essential to quickly and easily extract the information desired by the user, regardless of the content of individual texts. To achieve this, it is necessary to lighten the burden of sentence analysis and summary sentence generation and achieve high-speed processing.
It is necessary to be able to control the quality and quantity of summaries according to the user's requests.

In the conventional frame-based method, in order to deal with a wide range of texts, it is necessary to prepare a large number of frames according to the written content, which is not practical. In addition, methods based on narrative grammar and inter-sentence connections require a high degree of semantic analysis, which places a heavy burden on sentence analysis and is not suitable for high-speed processing.

Furthermore, in conventional text summarization devices, there was no way to easily obtain summaries online.

The present invention solves the above problems, and aims to provide a text summarization method and device that reduces the burden of sentence analysis and summary sentence generation and enables high-speed online processing.〇 Solving the Problems Means for achieving the above object The technical means of the invention according to claims 1 to 4 is to perform a syntactic and semantic analysis of each sentence of an input sentence, and use the information of the analysis result as pre-prepared word spirit information. Sentences considered to be important are selected by comparing selection criteria such as syntactic information, inter-sentence relationships, and text characteristics. Further, the technical means in the invention according to claims 5 to 7 includes a sentence analysis unit that performs syntactic analysis of an input sentence, a database unit that stores selection criteria for obtaining a summary sentence, and a syntactic analysis result and selection criteria. and a summary sentence selection section that selects sentences that are considered important.

It is equipped with at least a display section for displaying a summary sentence.

Effect The effect in the above configuration is as follows. Various sentences have special idiom syntactic structures depending on the nature of the sentence. Therefore, by setting these special elements as selection criteria in advance, comparing the information obtained by parsing the input text with these selection criteria, and extracting and selecting those that are considered important, each It is possible to obtain a summary sentence by processing a wide range of sentences at high speed, regardless of the specific written content of the sentences.

Example An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a conceptual diagram of a document summarization device according to an embodiment of the present invention. The text to be summarized shall be in English.

A sentence inputted from the input device 1 is subjected to syntactic and semantic analysis by the sentence analysis device 2, and converted into an internal representation representing the content of each sentence of the inputted sentence. The summary sentence selection device 3 learns the nature of the input text and obtains selection criteria corresponding to the nature from the selection criteria database 4. Note that the nature of the input text can also be given by the user using the input device 1.

The selection criteria stored in the selection criteria database 4 are determined as follows. Various types of sentences, such as assertions or problem statements that express a certain attitude of the author, use only special words or phrases, or have special syntactic structures. In addition, the request letter has a sentence structure that states the author's request. These are special boarding information,
It can be expressed as syntactic information or information on the linkage between sentences. The selection criteria database 4 includes information for each characteristic of a sentence: - What type of sentence is important?
It also describes what type of sentence should be selected to obtain a summary sentence.

Judgment criteria for fixing the form of these sentences are described in the form of the above-mentioned collection information, syntax information, and information on the linkage relationships between sentences. A degree of importance is added to the form and judgment information of these sentences.

By calculating the importance of all sentences using the importance added to the selection criteria, it becomes possible to select sentences whose importance exceeds a certain value according to the level value corresponding to the quantity and quality of summary sentences. , the quantity and quality of the summary text can be controlled as needed. Details will be described later.

The summary sentence selection device 3 selects sentences considered to be important by comparing the selection criteria from the selection criteria database 4 with the internal representation of each sentence in the input sentence obtained from the sentence analysis device 2. The quality and quantity levels of summary sentences are given as default values, and the sentences to be selected are determined by taking these level values into consideration. By changing the default level, the quality and quantity of the summary text can be controlled and a summary text that meets the user's purpose can be obtained. Figure 2 shows the internal configuration of the selection criteria database 4 shown in Figure 1. . From a sentence form selection criteria section 7 that describes what type of sentence should be selected for each property of the sentence, and a judgment information section 8 that describes information for determining whether each sentence has that form for each form. It has become. By dividing the database into the sentence form selection criteria section 8, when the sentence forms to be selected overlap depending on the nature of some sentences, the judgment criteria can be shared and duplicate descriptions can be avoided.

FIG. 3 shows the internal configuration of the sentence form selection reference section 7 shown in FIG. For each nature of the text, the format of the text, such as the assertion or problem statement that is considered necessary to obtain a summary from the text, is described. Each of these sentence forms has an importance level a, b, c, etc. that takes into account the nature of the sentence.
... is given. Even if the form of the sentence is the same, the importance value differs depending on the nature of the requested sentence.

For example, for sentences in which the author's argument is important in the sentence format selection criteria section 7, sentences that state an argument or conclusion,
We believe that a summary sentence can be obtained by selecting sentences that state the problem that is the basis of the claim or conclusion, or sentences that state the basis for that claim or conclusion. In this case, sentences that raise the issue or state the basis serve a supplementary role to the sentences that state the argument or conclusion, so the importance value is set low.

FIG. 4 is an internal configuration diagram of the determination information section 8 of FIG. 2.

For each sentence type, comprehensive information, syntactic information, and information on the linkage relationships between sentences are described for determining whether the sentence is in that form. Also, each piece of information has an importance level of s,
t, . . . are added. At least one of each of these pieces of information is used for selection.

For example, in the case of a sentence that expresses an assertion, the syntactic information includes auxiliary verbs that express an assertion such as [5hold J and [mustJ], and occurrence information of expressions such as ``We hope J''.As syntactic information, Information such as emphatic syntax that expresses assertions is used.As for the conjunction relation information between sentences, sentences that are recognized to express assertions based on gathering information and syntactic information are connected with conjunctions such as [and J or "but". Use information such as written sentences. As for the importance level, for example, a sentence with a 1-emphasized structure is set to a high importance value, considering that the sentences that appear also represent a strong argument. The importance of the link relationship information is dynamically assigned a value by the summary sentence selection unit 2 in consideration of the importance of the sentences that have the link relation.

FIG. 5 shows a conceptual diagram of the processing of the summary sentence selection device 3 of FIG. 1. The selection criteria extraction unit 9 extracts information on the sentence form for obtaining a summary sentence according to the nature of the input text from the selection criteria database 40 and the sentence form selection criteria unit 7, and the information on the sentence form from the judgment information unit 8 of the selection criteria database 4. Extract information for determining the form of a sentence. The importance calculation unit 10 compares the selected information with the internal representation of each sentence obtained from the sentence analysis device 2. When a certain sentence matches the judgment information, the importance of the sentence is determined by considering the importance of the form of the sentence and the importance of the information used for the judgment information. The importance of a sentence selected in accordance with the conjunction relation information is calculated from the importance of the original sentence that became the conjunction relation, and for sentences that meet multiple criteria, the importance corresponding to each criterion is calculated. Gives a value higher than the maximum value of degrees. Sentences that do not match any judgment information are given the lowest importance value, and all sentences in the input text are given some degree of importance. The level selection section 11 is
The level value corresponding to the quality and quantity of the summary sentence is known from the input device 1. At this time, if no level value is input, a default value is used. According to this level value, sentences with 1 importance of a certain value or more are selected.

The range of importance corresponding to the level value is determined from the quality and quantity of summary sentences to be selected and the distribution of importance values.

For example, if a level value of 1 corresponds to the request for quantity, ``make the summary sentence half of the whole text'', then the amount of selection will be half of the whole text based on the distribution of importance values. A value of importance is determined, and a sentence that is equal to or higher than that value is selected as a summary sentence. Note that it is also possible to preset the range of importance corresponding to the level value and select 2. However, in this case, be careful as all sentences may be selected or, conversely, not selected at all. is necessary.

Returning to FIG. 1, the summary sentence selected by the summary sentence selection device 3 in this manner is supplied to the output device 6, and is displayed on the input sentence in a light manner such as color coding or bolding. This highlighted display can support so-called diagonal reading.

In addition, in response to a user's request, the summary sentence generation device 5 generates a Japanese translation corresponding to the selected sentence from the internal representation of the sentence by the sentence analysis device 2, and displays it on the same screen by the output device 6. .

Effects of the Invention As described above, the present invention provides a text summarization method and device that selects sentences considered to be important by comparing the results of syntactic and semantic analysis of each sentence of an input text with predetermined selection criteria. As long as the characteristics of the sentences are the same, it does not depend on the written content of each individual sentence, so it can be applied to a wide range of sentences. In addition, since it is a criterion that can be judged even with shallow sentence analysis, there is less burden on sentence analysis, and "summary sentences can also be generated just by selection, so there is less processing burden. Therefore, high-speed processing is possible."

Furthermore, by assigning importance to the entire input text when selecting a summary sentence, selected sentences can be added or deleted at high speed, and the quality and quantity of the summary sentences can be adjusted interactively. Furthermore, if only information about words is used as selection criterion information, even faster processing is possible without performing syntactic analysis.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a Mataryu summary device according to an embodiment of the present invention;
Figure 2 is an internal configuration diagram of the selection criteria database, Figure 3 is an internal configuration diagram of the sentence form selection criteria section of the selection criteria database, Figure 4 is an internal configuration diagram of the judgment information section of the selection criteria database, and Figure 5 is an internal configuration diagram of the selection criteria database. FIG. 3 is a conceptual diagram of processing of the summary sentence selection device. DESCRIPTION OF SYMBOLS 1... Input device, 2... Sentence analysis device, 3... Summary sentence selection device, 4... Selection criteria database, 5...
Summary sentence generation device, 6... Output device, 7... Sentence form selection reference section. 8... Judgment information section, 9... Selection criteria extraction section, 10.
...Importance calculation section, 11...Level selection section. Figure 1 Name of agent Patent attorney Shigetaka Awano #1 or 1 person No. L''j Figure 1.

Claims (7)

[Claims] (1) A text summarization method characterized by obtaining a summary sentence according to selection criteria for selecting sentences considered to be important from the input text based on the results of syntactic and semantic analysis of each sentence of the input text. (2) The text summarization method according to claim 1, characterized in that at least one of lexical information, syntactic information, and information on linkages between sentences obtained from syntactic-semantic analysis is used as the selection criterion. (3) The text summarization method according to claim 1, wherein information that depends only on the properties of the text is used as the selection criterion. (4) The text summarization method according to claim 1, wherein importance is given to the selection criteria. (5) A sentence analysis unit that performs syntactic and semantic analysis of each sentence of the input text, a selection criteria database that stores selection criteria, and a selection criteria database that compares the analysis results of the above sentence analysis unit with the selection criteria of the selection criteria database to identify important ones. 1. A text summarizing device comprising: a summary sentence selection unit that selects a sentence that seems likely; and a summary sentence display unit that displays the selected sentence. (6) The text summarization device according to claim 5, wherein the selected sentence is highlighted on the input text by color coding or the like. (7) The text summarization device according to claim 5, wherein the input text is a foreign language text, and further comprises means for generating and displaying a translated text of the selected text.