JPH0713967A - Abstract sentence generator - Google Patents

Abstract

PURPOSE:To prevent the omission occurrence of semantic information by extracting a sentence including a predicate which is previously set at each scenario item expressing a subject to express the item as an important sentence. CONSTITUTION:A scenario setting part 11 sets the scenario item corresponding to a field to which original sentences belong with respect to the original sentences which are pre-processed already within an inputted document file. A field-classified scenario table 12 gives the previously prepared field-classified scenario item to the scenario setting part 11. Then, an important sentence extracting part 13 extracts the important sentence from the original sentences by a scenario item unit and a scenario item unit predicate table 14 gives the predicate of the scenario item unit which is previously prepared by the scenario item unit to the important sentence extracting part 13. Moreover, a sentence shaping part 15 deletes unnecessary words and phrases and a sentence from the important sentence which is extracted by the important sentence extracting part 13 or shapes the sentences by replacing the words and phrases and a shaping instruction table 16 gives an instruction for deleting words and phrases for shaping the sentence to the sentence shaping part 15.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an abstract sentence creating device for creating abstract sentences of documents such as papers by an information processing device. In particular, the present invention relates to an abstract sentence creating apparatus for automatically creating an abstract sentence by extracting an important sentence including semantic information necessary for the abstract sentence from the whole sentence of an abstract target document in a database that performs a full-text search.

In the present specification, the term "sentence" refers to a string of characters that represents a single set of meanings that are separated from the beginning of a sentence by "." Or "."

[0003]

2. Description of the Related Art In recent years, a full-text database system that supports full-text information has been constructed in an information service industry or in-company database system. In such a full-text information database system, the amount of information is enormous. Therefore, in an actual system, bibliographic information obtained by secondary-processing the full-text information is required, and the search is performed based on this bibliographic information.

As the bibliographic information, a system is required which extracts necessary semantic information from the full-text information and creates an abstract sentence which is an abstract of the full-text from the extracted information. As a technique for automating the creation of such an abstract sentence and automatically creating the abstract sentence from the full-text information, the technique disclosed in Japanese Patent Laid-Open No. 61-117658, which extracts one important sentence for each paragraph, or the present invention This is a technology in which one of the persons involved was involved as an inventor, and the technology selects the most important word according to the frequency of appearance of the important words that express the subject of the sentence, and extracts the sentence containing this most important word as the important sentence. JP-A-2-
There is technology of 257266.

[0005]

However, if one important sentence is extracted for each paragraph, the whole abstract sentence will be unnatural, and more than two sentences will express the most important subject in one paragraph. If included, there was a problem that important semantic information was lost.

[0006] Further, the technique of extracting an important sentence based on the frequency of appearance of an important word can also extract along the flow of the sentence,
There were cases in which the abstract flow was not the unified flow necessary for abstracting, because it was affected by the flow of sentences that differed among the papers recorded in the database system.

The present invention is extracted by paying attention to the fact that, for each full-text information recorded in the database system, there is a scenario as the flow of text required for the full-text, and in general, it is expressed according to this scenario. An object of the present invention is to provide an abstract sentence creating device that automatically creates an abstract sentence in which the abstract sentence is natural and no missing of semantic information occurs.

[0008]

The present invention is based on an important sentence extracting means for extracting an important sentence constituting the subject of the sentence from an inputted sentence, and the extracted important sentence based on a predetermined agreement. In the abstract sentence creating device provided with a sentence shaping means for shaping and abstracting the abstract sentence, the important sentence extracting means is set for each field of the inputted sentence, and is preliminarily set for each scenario item expressing the subject. It is characterized by including a means for extracting a sentence including a predicate expressing this set item as an important sentence.

The sentence shaping means extracts an immediately preceding sentence as an important sentence by a word in the extracted important sentence and adds it, a means for replacing a specific phrase with another phrase, and a specific phrase. It is preferable to provide a means for deleting a phrase from an important sentence and a means for deleting a previously designated sentence.

[0010]

The present invention projects a typical scenario in the relevant field for the text of a paper or the like for which an abstract is to be created, and by projecting this scenario, only the plots and ideas that the writer originally intended are extracted. This is to automatically create an abstract sentence.

Generally, in a thesis or an essay, the text is described in accordance with a scenario represented by a succession and conclusion.
For example, in a science and technology paper, the paper is based on assumptions, objectives,
It is described in terms of methods, results, considerations and conclusions, and it is required to state the subject of the paper in that order. Therefore, even if the abstract text is extracted from the whole sentence by extracting the important sentences for each item corresponding to the plot that the author is talking about according to this scenario, even if the target paper does not follow the above-mentioned scenario Even if it is, a natural abstract sentence is created as an abstract of a paper without missing the necessary semantic information as an abstract of a science and technology paper.

The present invention is designed to
When a scenario item is created and, for example, an abstract of a scientific paper is created for each item, prepare a predicate table that expresses that item for each scenario item such as the above-mentioned assumptions, objectives, methods, and results. Then, by extracting the sentence containing this predicate as an important sentence and performing necessary sentence shaping,
The abstract text is a natural sentence and does not lack the semantic information necessary for the abstract text such as a paper.

Here, the predicate is an element that composes a sentence into a "supplemental word", a "modifier", and a "subject".
A word that is located at the end of a sentence and determines the outline of the sentence depending on its content. The present invention pays attention to this predicate, prepares a predicate expressing the subject of the paper for each scenario item, and extracts a sentence including this predicate as an important sentence. Is being extracted.

[0014]

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

FIG. 1 shows the structure of an abstract sentence creating apparatus according to an embodiment of the present invention.

The abstract sentence creating apparatus 1 of the present embodiment sets a scenario item corresponding to a field to which a source text belongs to a source text which is already preprocessed from an input document file. 11, an area-specific scenario table 12 that provides the scenario setting section with scenario items prepared in advance, and an important-sentence extracting section 13 that extracts an important sentence from an original sentence in units of this scenario item.
And a scenario item unit predicate table 14 for giving a predicate prepared in advance for each scenario item to the important sentence extraction unit 13, and unnecessary words and sentences from the important sentence extracted by the important sentence extraction unit 13. The sentence shaping unit 15 that reforms a sentence by deleting or replacing a phrase
And a shaping instruction table 16 for giving the sentence shaping section 15 instructions such as deletion phrases for sentence shaping.

FIG. 2 shows the operation of the scenario setting section 11. A scenario is selected (S21) according to the field of the sentence for which the abstract sentence is created, the scenario table for each field is read (S22), and the scenario item is set. Yes (S23).

An example of scenario items prepared in the field-specific scenario table 12 is shown below. Scientific and technical papers include: 1. Assumption explanation, 2. Purpose / theme sentence, 3. Method, 4.
Result, 5. Scenario items for consideration and conclusion are set up, and important sentences are extracted for each scenario item.

As another example, in the case of a database system for newspaper industry articles, as scenario item examples: 1.
Main text (who did what, what did you do) 2. Explanation (purpose, characteristics, performance, etc.), 3. Future developments (plans, actions, etc.),
4. Set the price.

FIG. 3 is a diagram for explaining an important sentence extraction process in scenario item units in the important sentence extraction unit 13. That is, the original sentence is read (S31), and the scenario item unit predicate table 14 is read (S32). A sentence including a predicate in the scenario item unit predicate table 14 is extracted from the original sentence as an important sentence (S33). Then, the extracted scenario item unit important sentence is passed to the next sentence shaping unit 15.

FIG. 4 is a diagram for explaining more detailed important sentence extraction processing in scenario item units in the important sentence extraction unit, and is a diagram for explaining important sentence extraction processing in science and technology papers.

In the important sentence extraction unit 13, first, all the target sentences are read, the predicate explanation predicate table in the scenario item is referred to, and the sentence including the predicate in the premise explanation predicate table corresponds to the premise explanation. It is extracted as a sentence and this is used as an important sentence corresponding to the premise explanation.

Next, the whole sentence is read again, this time, by referring to the objective / subject sentence predicate table, a sentence including the predicate in the objective / subject sentence predicate table is extracted as an important sentence corresponding to the objective / subject sentence, and this is extracted. Purpose / subject sentence Applicable important sentence.

Next, regarding the method, similarly, the important sentence corresponding to the method is extracted by referring to the method predicate table and set as the important sentence corresponding to the method.

Next, similarly for the resulting scenario items,
The important sentence corresponding to the result is extracted by referring to the result predicate table and is set as the relevant sentence.

Finally, regarding the scenario item of the consideration / result, the consideration / conclusion predicate table is referred to, and the relevant important sentence is extracted to be the consideration / conclusion important sentence.

Here, the scenario item unit predicate table 1
The contents of 4 are shown. As described above, for science and technology papers, five predicate tables are prepared for each scenario item: a predicate explanation predicate table, a purpose / subject sentence predicate table, a method predicate table, a result predicate table, and a consideration / conclusion predicate table. It

An example of specific predicates prepared in this predicate table will be given.

Predicates The predicates set in the predicate table are "was done", "has been done", "has been done", "has been done", and "so that they can be done". "I need it", "I need it", "I need it", "I need it", "I need it", "I need it", "I need it", "Necessary", "Under development", "Install", "Expected",
These are “desired”, “effective”, “emerging”, etc.

The predicates set in the purpose / subject sentence predicate table are “state”, “report”, “reporting”, “introduce”, “suggest”, and “propose”. It is "consider.", "Explain.", "Aiming.", "Possible.", "Aiming.", "Clarify."

Next, the predicates set in the method predicate table are "selected", "shown", "shows",
"It is.", "Indicates that", "Indicates in combination.", "Indicates also.", "I was able to indicate.", "It is a method.", "I have adopted.", "Created.", "Used.", "Characterized.", "Evaluated.", "Evaluated.", "Composed.",
"Build.", "Prototype.", "Guide.",
"Evaluation experiments were performed."

The predicate set in the result predicate table is
"I mentioned.", "I got it.", "I got it.", "I confirmed.", "It was confirmed.", "I confirmed.", "I could improve.", ". It was realized. ”,“ It has been realized. ”,“ Provided. ”,“ Dissolved. ”, Etc.

Consideration / Conclusion The predicates set in the predicate table are “thinking”, “aiming”, “being a subject”, “required”, “cleared”,
“Conclusion is given”, “Conclusion is obtained”, “Point out”, “is shown”, etc.

These predicates are examples extracted from sentences corresponding to scenario items with reference to actual science and technology papers. In reality, a larger number of predicates are set in each scenario item unit predicate table. .

Next, an example of a predicate table in the fields of newspapers and industrial articles other than the system for creating abstract sentences for scientific and technical papers will be given.

As described above, the scenario items in the newspaper / industrial article field are 1. Main text, 2. Explanation, 3. the next deployment,
4. It consists of price, and for each scenario item,
A main sentence predicate table, an explanation predicate table, a future expansion predicate table, and a price predicate table have been created.

In the subject sentence predicate table, "Announced.",
“Sell”, “Sell.”,
"Established", "Started", "Embarked", "Completed", "Ship",
A predicate such as “expand” is set.

In the explanation predicate table corresponding to the explanation of the purpose, characteristics, performance, etc., "possible (is possible)",
“Achieved”, “Apparatus”, “Able to do”, “Industry”, “Use (Can) (Yes)”, “Incorporated”, “Comprehensive”, “Comprehensive” Yes, ”“ There is a reason. ”“ I can improve. ”,
Predicates such as “what you did.”, “Compatible model”, “first in Japan”, “overview”, “specification”, “made”, “take office”, “work” are set.

In the future expansion predicate table corresponding to future schedules, actions, etc., “planned”, “expected (expected).”, “Expanded”, “progress”.
"Aim.", "Planning" (is)
(No). , "Policy.", "Invested.", "Impact momentum.", "Start (make).", "Sales target is".
Predicates such as are set.

In the price predicate table, "price is",
Predicates such as “annual contract fee (usage fee)” and “selling price” are set.

As described above, the sentences expressing scenario items are extracted by referring to the predicate table according to the scenario of the paper or the like, which is the target of the abstract sentence, according to the scenario of the paper or the like. Extract as important sentences for each item.

A sentence shaping process is performed on the extracted important sentence for each scenario item to reform the sentence into a sentence having a natural flow as an abstract sentence. This is because even if an important sentence is actually extracted by referring to the predicate table, unnecessary sentences or identical sentences may be extracted, or necessary sentences may need to be added from the flow of the abstract sentence. is there.

FIG. 5 shows a more detailed structure of the sentence shaping unit 15 and the shaping instruction table 16 for performing this sentence shaping process.
As shown in FIG. 5, the sentence shaping unit 15 that performs sentence shaping is
Contents of an immediately preceding sentence assigning / shaping unit 51 for imparting and shaping the immediately preceding sentence based on the contents of the immediately preceding sentence assigning instruction table 52, and a phrase replacement instruction table, a phrase deleting instruction table, and a sentence deleting instruction table 54. A phrase replacement / sentence deleting unit 53 for performing phrase replacement and deletion processing and sentence deletion processing based on the above.

Specifically, FIG. 6 shows a flowchart showing the operation of the immediately preceding sentence addition and shaping processing, and the processing will be described. It should be noted that here, description will be made assuming that the sentence shaping process is performed on the important sentence extracted from the scientific and technical paper.

First, important sentences are sequentially read from the beginning (S6).
1). It is determined whether all the sentences have been read (S62), and the immediately preceding sentence addition instruction table 52 is sequentially read from the beginning (S6).
3).

The contents of the immediately preceding sentence addition instruction table 52 are "here (wa)", "where", "this (like) (like) (when) (for) (out of) (koto)" , "That (like) (for) (for) (for)", "again", "this", "this (ha) (for)
(By) (By) (By) (By)), "It (By)"
The words such as “()”, “but (while)”, “that is”, “katsu”, “so”, “naka”, “especially”, “extra”, “next”, etc. are set. Has been done.

This immediately preceding sentence addition instruction table 5 is included in the important sentence.
It is determined whether or not there is a word in 2 (S65), and if there is, the original sentence is read and the sentence immediately before the important sentence is added to the important sentence (S66). This is because the words and phrases in the immediately preceding sentence addition instruction table 52 are indicators, as described above, and if such words and phrases exist, this important sentence cannot be understood unless there is a sentence that appears before. Even if this condition is satisfied, if the preceding sentence of the important sentence is a non-sentence with no punctuation, the preceding sentence is not added.

With respect to the immediately preceding sentence that has been added, it is determined whether or not there is a word or phrase in the immediately preceding sentence addition instruction table, as with the important sentence, and the same processing is performed (S67). The sentence just before the addition process becomes an important sentence and is adopted as a constituent sentence of the abstract (S68).

When all the words and phrases in the immediately preceding sentence addition instruction table have been processed, and when all the important sentences have been processed, the next word / phrase is replaced, deleted, and the sentence deletion process is performed.

FIG. 7 is a flow chart for explaining the operations of word replacement, deletion, and sentence deletion processing. This operation will be described.

First, important sentences are sequentially read from the beginning (S7).
1). After determining whether all processing has been completed (S7
2), the phrase replacement instruction table, the phrase deletion instruction table, and the sentence deletion instruction table are sequentially read from the beginning (S
73).

The contents of the phrase replacement instruction table, the phrase deletion instruction table, and the sentence deletion instruction table will be described.

In the phrase replacement instruction table, for example, an instruction to replace the phrase "generally (problem) as follows" with "some (problem)" is set. This means "some" because the previous phrase means enumerating the questions, which is not appropriate for the abstract.
It is replaced with the phrase of and converted into an appropriate sentence as an abstract sentence.

In the word / phrase deletion instruction table, for example, an instruction to delete the number at the beginning of the sentence and an instruction to delete the words "and" and "nao" are set. The number is deleted because the number at the beginning of the sentence when the itemized sentence is adopted as an important sentence is deleted, and “and” and “na” are not appropriate as abstract sentences and unnecessary words and phrases. This is for deleting.

In the sentence deletion instruction table, for example,
Delete the important sentence that explains the table, delete the important sentence that contains the following words and phrases, and in the scenario item unit predicate table, the longest when multiple important sentences are extracted by the same predicate. Instructions are set to adopt only important sentences and delete other important sentences, and to adopt only the important sentence that appears first when there are multiple same sentences or the same sentence that contains the same sentence. . The appearance of the same sentence is because the same important sentence may be extracted by performing the preceding sentence addition shaping process, for example.

In this way, it is judged whether or not the word / phrase in the instruction table is in the important sentence (S75). If there is a word / phrase, if it is a word / phrase replacement instruction, the replacement instruction is executed to obtain the abstract sentence. If it is an instruction to delete a word (S76), if it is an instruction to delete a word, delete it to assemble words in the sentence and use it as a constituent sentence of the abstract (S77). Skip reading (S78). When all processing is completed, an abstract sentence is created.

Here, specifically, an abstract sentence obtained by performing the above-mentioned processing for one paper will be shown. The target papers are Koichi Shibata, which is disclosed in Japanese Patent Laid-Open No. 2-257266.
Kiichi Miyanaga, Koji Tochiuchi "Automatic extraction of technical term information from science and technology literature" is the 35th national conference of the Information Processing Society of Japan.

Abstract sentence created by this embodiment [abstract sentence] 1. Assumptions We are conducting research to automatically extract sentences that provide definitions and explanations of technical terms from scientific and technical literature.

2. Purpose / Subject The automatic extraction method proposed in this research automatically extracts information on technical terms using the characteristics of the text display, and supports the creation of a question-answer system for technical terms and a technical glossary. We are aiming for application to.

This paper describes the outline and the experimental results obtained so far.

3. Method Sentences with a certain purpose are considered to have common characteristics for each purpose. Such features are called extraction rules, and the target sentence is selected by matching with the target sentence.

The accuracy of the extracted sentence is insufficient only by applying the extraction rule. Therefore, the sentence selected by the extraction rule is used as a candidate sentence, and only one valid sentence is selected from those sentences. The clues used at this time are the order in which the target terms appear in the document (term order), the order of sentences extracted by the extraction rules (extraction order), the accuracy of the extraction rules used (priority order), the extraction Whether the target term is a headline before the sentence obtained by the rule (headline rule)
Etc., and was obtained by analyzing the experimental results according to the extraction rule.

The materials used for the experiment were about one half each of the information retrieval and database, the textbook of wired communication engineering, and one each.

4. Results 5. Discussion / Conclusion In the future, it is necessary to conduct experiments on more documents to confirm whether this method is generally effective.

The above-mentioned abstract and Japanese Patent Application Laid-Open No. 2-25
Comparing with the abstract sentences shown in Japanese Patent No. 7266, almost the same sentences are extracted as important sentences, and particularly, the same sentences are extracted for the premise explanation, consideration and conclusion in the scenario item. The first half sentence of the purpose / subject and the first half sentence of the method are sentences that could not be extracted by the above-mentioned prior application.

According to the method of the present application, an abstract sentence in which necessary information is described according to the scenario can be obtained, and a sufficiently practical abstract sentence can be obtained.

[0067]

As described above, according to the present invention,
Without missing the necessary semantic information for the abstract
Since the necessary information is described along the lines of the sentence, it is possible to automatically create a practical abstract sentence.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an abstract sentence creating apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram showing an operation of a scenario item setting unit.

FIG. 3 is a diagram showing an operation of an important sentence extraction unit.

FIG. 4 is a diagram showing an example of important sentence extraction processing.

FIG. 5 is a diagram showing a configuration of a sentence shaping processing unit.

FIG. 6 is a diagram illustrating an operation of a preceding sentence addition shaping process.

FIG. 7 is a diagram for explaining the operation of word replacement / deletion / sentence deletion processing.

[Explanation of symbols]

1 Abstract sentence creation device 11 Scenario setting unit 12 Scenario-specific scenario table 13 Important sentence extraction unit 14 Scenario item unit predicate table 15 Text shaping unit 16 Shaping instruction table 51 Preceding sentence addition shaping unit 52 Preceding sentence addition instruction table 53 Term substitution deletion / Sentence deletion unit 54 Term replacement instruction table / phrase deletion instruction table / Sentence deletion instruction table

Claims (2)

[Claims] 1. An important sentence extracting means for extracting an important sentence constituting the subject of this sentence from the inputted sentence, and shaping the extracted important sentence based on a predetermined agreement to create an abstract sentence. In the abstract sentence creating device including sentence shaping means, the important sentence extracting means expresses this item that is set for each field of the input sentence and is preset for each scenario item that expresses the subject. An abstract sentence creating apparatus characterized by including a means for extracting a sentence including a predicate being performed as an important sentence. 2. The sentence shaping means includes an immediately preceding sentence adding means for extracting and adding the immediately preceding sentence as an important sentence according to a phrase in the extracted important sentence, a means for replacing a specific phrase with another phrase, and a specific sentence The abstract sentence creating apparatus according to claim 1, further comprising: a unit that deletes the phrase from the important sentence and a unit that deletes a sentence designated in advance.