JPH0744566A - Abstract preparation device - Google Patents

Abstract

PURPOSE:To improve the reading comprehensibility and naturalness of abstract sentences by including the corresponding destination of a correspondence expression in the abstract sentence or omission part display in the abstract sentence. CONSTITUTION:A format analysis part 1 analyzes electronized input sentences and analyzes the break of the sentences, the end of paragraphs and the structure of chapters and passages, etc. A morpheme/syntax analysis part 3 performs morpheme/ syntax analysis by each sentence for the part of a text and the titles of the chapters and the passages. A correspondence solving part 5 refers to the morpheme/syntax analyzed results of the respective sentences and decides the corresponding destination of the correspondence expression (participal adjectives and pronouns) included in the sentences. An important sentence judgement part 9 judges important sentences based on the morpheme/syntax analyzed results of the respective sentences and the titles of the chapters and the passages. In such constitution, when unselected words and phrases, the part of the correspondence expression such as the participal adjective and the pronoun or a non-important sentence for instance, are present between the words and phrases selected from inputted original sentences corresponding to a prescribed reference, that effect is indicated between the words an phrases in the abstract sentence with a prescribed code or the like for instance.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an abstract creating apparatus for creating and displaying an abstract of a sentence (text) in natural language.

[0002]

2. Description of the Related Art Conventionally, a mainstream technology in a system for creating an abstract is to select important sentences from a document and connect the selected important sentences to create an abstract. The problem with such a conventional abstract creation system is that the sentence that has been selected as an important sentence is "that"
In this case, anaphoric expressions such as “these,” “this,” and “the same person” are included. If the reference (reference) of those terms happens to be an important sentence, there is no word or sentence in the abstract that points to those words, so there is no problem in understanding the abstract, but it is not an important sentence. If you have a problem. In such a case, not only the word or phrase of the reference destination is not known, but if it points to a different word or phrase in the abstract sentence that happens to come, there is a possibility that it will be misread and a large misreading will occur. In the past, such anaphoric expressions were deleted, but this method cannot be denied the drawback of impairing the readability and naturalness of the abstract.

Ideally, it is desirable to determine / identify the sentence or phrase to which the adjunct or pronoun is anaphorated and replace the adjunct or pronoun in the abstract sentence with the phrase or sentence. At this time,
If the sentence or phrase of the reference is originally included in the abstract sentence,
It is preferable to leave the pronouns and adnominals as they are.

As the anaphora identification processing technology (anaphora problem solving technology), the Japanese objects include "Japanese sentence anaphora problem solving system", Takeshi Yamamura, Noboru Onishi, Noboru Sugie (The Institute of Electronics, Information and Communication Engineers) Magazine Vol.J73-D-II, No.6, p
p.887-896, 1990.6) and "Analysis of Japanese sentences using meaning and context information-processing in consideration of context" (Makoto Nagao, Junichi Tsujii, Kazutoshi Tanaka: IPSJ Journal Vol.17, N.
o.1, Jan.1976) has been proposed.

However, these techniques are premised on that technical terms and knowledge in the field of texts to be processed, as well as daily knowledge (common sense), are prepared in advance in a machine-processable form. Therefore, these techniques cannot be used for a general natural language processing system in which the domain knowledge cannot be prepared in advance because the domain of the text to be handled is unknown. Also'common sense '
The preparation issue of is an unsolved technical issue.
Moreover, the application of these anaphoric solution techniques to the abstract preparation system has not been studied so far. Moreover, in the conventional document processing / display system including the abstract creating system, the usage of the anaphora analysis result was not sufficient.

On the other hand, with the recent digitization of documents,
New developments can be seen in document utilization technology. For example, a text database may be centrally managed or a specific symbol may be inserted into an electronic document, such as SGML, to facilitate various processes. There is also a method such as hypertext in which a document or various parts in the document are electronically associated with each other and stored to expand the way of using the document. There are multiple ways to search and display documents, such as searching and displaying only the relevant parts instead of all of one document.
It is becoming more complex.

[0007] Under such circumstances, a technique of supplementing the word or topic expression of the anaphoric expression is important so that only a part of the sentence can be read to understand the tentative content. I think it will be.

Along with this, not only the processing result is unilaterally displayed, but also the additional information such as the change of the display portion and the presentation of the phrase referred to in the sentence is interactively displayed according to the operation of the user. The function of such a system, that is, the man-machine interface of the system, is becoming more important. In such a situation, there are various possibilities of displaying the abstract and using the anaphora analysis processing result accompanying it, but it cannot be said that it has been fully realized yet.

[0009]

However, in the conventional abstract generation system, the anaphora analysis is not performed, or the anaphora analysis result is not sufficiently used, and the abstract text is poor in readability. It also responds to user requests for displaying abstract sentences and original texts (such as changing the length of abstracts, requesting references to adjuncts and pronouns in abstracts, and displaying original texts not included in abstracts). It was difficult.

The present invention has been made in view of the above-mentioned problems, and makes it possible to improve the comprehension and naturalness of an abstract sentence by including the reference destination of the anaphoric expression in the abstract sentence or the abbreviation display. It is an object to provide an abstract preparation device that improves.

[0011]

In order to achieve the above object, the present invention is an abstract creating device for creating an abstract from an input original sentence, and selects an abstract sentence by selecting a phrase from the original sentence according to a predetermined criterion. The abstract sentence creating means to be created, and the words and phrases not selected between the words and phrases selected by this abstract sentence creating means, such as insignificant sentences or anaphoric expressions such as adnominals, pronouns, etc. Between the words and phrases in the abstract sentence, it is necessary to provide a presentation means for presenting with, for example, a predetermined code and a word or phrase corresponding to the location presented by this presentation means, for example, an anaphoric expression part such as an insignificant sentence or a noun or a pronoun. The gist of the present invention is to have display means for displaying the information accordingly.

[0012] Preferably, in an abstract creating system for selecting an important sentence from an original sentence to create an abstract sentence, if the selected important sentences are not continuous in the original sentence, a specific symbol such as "..." is inserted between them. To do.

Preferably, when a particular symbol portion such as "..." in the display is designated by clicking with a mouse, the original text of that portion is displayed.

Preferably, in an abstract creating system for selecting an important sentence from an original sentence to create an abstract sentence, an anaphora analysis unit is provided, and an abstract sentence is created and displayed by using an analysis result of the anaphora analysis unit. .

Preferably, when the anaphoric expression portion such as a noun or a pronoun being displayed is designated by clicking with a mouse, the word or sentence of the anaphoric destination is displayed. Alternatively, if they are already displayed, the relevant part being displayed is highlighted.

[0016] Preferably, in the abstract creating system for creating an abstract sentence by selecting an important sentence from an original sentence, an anaphoric analysis unit is provided, and the selected important sentence includes a noun and a pronoun, and its destination is an unimportant sentence. If, then the non-important sentence is included in the abstract (as an important sentence).

Preferably, in the abstract creating system for creating an abstract sentence by selecting an important sentence from the original sentence, when an anaphoric analysis unit is provided and the selected important sentence includes a conjunction or a pronoun,
It replaces those adnominals and pronouns with the phrase or sentence of the destination.

Preferably, in the abstract creating system for creating an abstract sentence by selecting an important sentence from the original sentence, when an anaphoric analysis unit is provided and the selected important sentence includes a conjunction or a pronoun,
Before and after these adnominals and pronouns, the phrase or sentence of the recipient is inserted together with an appropriate symbol indicating insertion.

Preferably, in the abstract creating system for selecting an important sentence from an original sentence to create an abstract sentence, an anaphoric analysis unit is provided, and the selected important sentence includes a noun / pronoun, and its destination is an unimportant sentence. , The sentence containing the anaphoric expression is not included in the abstract.

Preferably, in the abstract creating system for creating an abstract sentence by selecting an important sentence from the original sentence, an anaphoric analysis unit is provided, and the selected important sentence includes a conjunction, and the destination of the anaphora is a non-important sentence. In this case, the adnominal is deleted.

[0021] Preferably, in an abstract creation system for creating an abstract sentence by deleting non-primary (non-important) parts such as adverbial clauses and subordinate clauses of the original sentence, an anaphoric analysis unit is provided, and a noun / pronoun appearing in the main clause If the reference destination of is included in a clause that is not the main clause, those clauses are not deleted.

[0022]

According to the present invention, when an unselected word, such as an unimportant sentence or an anaphoric expression part such as a noun or a pronoun, exists between the word or phrase selected according to a predetermined criterion from the input original sentence. This is shown between the words and phrases in the abstract text, for example, by using a predetermined code or the like. In addition, words and phrases corresponding to the part that is further presented,
For example, by displaying an anaphoric expression part such as a non-important sentence or a noun sentence or a pronoun, it is possible to improve the readability of the abstract sentence.

Further, according to the present invention, since the abstract is generated on the basis of the anaphora analysis result, the abstract sentence is highly comprehensible, and when the abstract sentence is subjected to processing such as full-text search or information extraction, Better results can be expected because there are fewer leaks than without anaphora treatment. Some recent information retrieval systems use co-occurrence relations or syntactic relations of multiple words in one sentence, but if anaphora processing is appropriate, even if one of the words is a pronoun. It does not reduce the search accuracy of these search systems. Further, it is possible to interactively change the length of the abstract, display an original text part that is not in the abstract, or display an anaphoric destination such as a noun or a pronoun in the abstract in accordance with the user's operation.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an automatic abstract generation / display system which is an embodiment of the abstract creation device according to the present invention.

Referring to FIG. 1, the format analysis unit 1 analyzes the digitized input sentence, and analyzes sentence breaks, paragraph transitions, chapter and section structures, and the like. morpheme·
The syntactic analysis unit 3 morphologically and syntactically analyzes the chapter and section headings and the body part for each sentence. Anaphora Resolution Department 5
Is to determine the anaphoric destination of the anaphoric expression (adjunct / pronoun) included in the sentence by referring to the morpheme / syntax result of each sentence. The important sentence determination unit 9 determines an important sentence based on a chapter or section heading and a morphological / syntactic analysis result of each sentence.

Next, the configuration of the anaphora solving section 5 will be described with reference to FIG. This anaphora resolution unit 5 is a topic analysis unit 51.
And an anaphora identification section 55.

The topic analysis unit 51 extracts a topic expression directly related to the content of the sentence from the morpheme / syntactic analysis result of the original sentence. The rhetorical topic expression storage unit 53a stores rhetorical sentence patterns related to topic expressions. Topic expressions that match this sentence pattern are rhetorical usages and are not directly related to the content of the sentence. The inversion syntax sentence pattern storage unit 53b stores sentence pattern information about a topic of a sentence such as an inversion syntax and a topic presentation expression such as'wa ',' mo 'and'koso' which is not a noun phrase at the end. There is.

The anaphoric identification unit 55 identifies the anaphoric type of the adnominal / pronoun included in the original sentence and the sentence or phrase of the anaphoric destination from the morpheme / syntactic analysis result of the original sentence and the topic analysis result of the topic analysis unit 51. To do. Anaphoric expression sentence pattern storage unit 5
3c stores information about sentence patterns including anaphoric expressions. This information is used to determine the anaphoric type of the sentence. In the propositional anaphoric expression storage section 53d, the'thing ',
Registered are nouns that often refer to the whole sentence including predicates such as'event 'and'situation'.

Referring to FIG. 1, the abstract creation / display unit 7
Creates an abstract sentence based on the important sentence judgment result and anaphora resolution result. Further, it is possible to interactively change the length of the abstract, display an original text part that is not in the abstract, or display an anaphoric destination such as a noun or a pronoun in the abstract in accordance with the user's operation. FIG. 3 is a block diagram of the abstract generation / display unit.

The operation of each processing unit will be described below based on an analysis example. FIG. 18 is a diagram showing an example of an input document (in the figure, a portion where Rv is displayed indicates a so-called inverted display. The same applies to other drawings hereinafter). The format analysis unit 1 analyzes the digitized input sentence,
Analyze sentence breaks, paragraph breaks, chapter and section structures, etc. The analysis result is used to select which part of the sentence is to be subjected to the abstraction process or to determine the unit of the abstraction process. For example, it is used for processing such as abstracting each chapter or section, or abstracting only the chapters or sections having the headings “Conclusion” and “Summary” to obtain the entire abstract.

The morpheme / syntactic analysis section 3 uses chapter and section headings,
And the text part is morpheme / syntactically analyzed for each sentence. Figure 2
0 is a part of the document morpheme / syntactic analysis result for the input document example shown in FIG. In this embodiment, a morpheme break is indicated by +, a bunsetsu break is indicated by /, and a morpheme name is indicated by <...>. In addition, a break of a phrase or a clause (hereinafter referred to as a main term) directly related to the main artistic part of the sentence is indicated by "[...]". What is described between these two symbols is the syntactic information of the clause obtained as a result of the syntactic analysis, such as the subject ", a complement, an adverb phrase.

The important sentence judgment unit 9 judges an important sentence based on the chapter or section heading and the morphological / syntactic analysis result of each sentence. For this processing, various methods have been already proposed as a conventional technique. For example, Japanese Patent Application No. 2-203865 extracts the context structure of a sentence based on the sentence connection method and the connection relation,
According to the method, the important sentence is judged. The important sentence determination unit 9 normally determines an important sentence within the number of designated important sentences or the designated number of characters.

The important sentence determination unit of the character number designating method is also usually the sentence number designating method internally. Calculate the sum of the number of characters of the sentence judged as important sentence, and if it exceeds the specified number of characters, decrement the specified number of sentences internally,
The only difference is that the post-processing of determining the important sentence again is repeated until the number of specified characters is reached.

Therefore, in the following description, the case of the sentence number designating method will be continued as an example. However, even when the important sentence determining unit of the character number designating method is used, the above-mentioned post-processing is added, and analogy is easy. Is.

As a conventional method, there is also an important sentence determination method of a compression ratio (number of characters of abstract sentence / number of characters of original sentence, or number of sentences of abstract sentence / number of sentences of original sentence) specification method. This is also
If the number of characters or the number of sentences in the original sentence is obtained and the preprocessing for calculating the number of characters or sentences allowed in the abstract sentence is added, it is equivalent to the above method. Omit it.

The topic analysis unit 51 shown in FIG. 2 extracts a topic expression from the morpheme / syntactic analysis result of the original sentence. The algorithm of the topic analysis unit is shown in FIG. First, the topic analysis section
A noun phrase that includes a particle or rhetorical expression that presents a topic such as'wa ',' mo ',' koso 'at the end of the phrase or clause.
Extract clauses or adverb phrases / clauses from the original sentence (step S1)
7). Some phrases and clauses including these topic presentation expressions are rhetorical and have nothing to do with the content of the sentence. The rhetorical topic expression storage unit stores information about such styles. FIG. 16 shows an example of registration. The style is written in regular expressions, pattern matching is performed with the morpheme sequence of the original sentence, and if there is a match, the topic expression is determined to be rhetorical usage and rejected as a topic (step S19, 21).

The regular expression is a method of designating a character string used in a computer based on unixOS. '. 'Is any one character,' a * 'is 0 or more consecutive characters a,' ^ 'is the beginning of a sentence,' $ 'is the end of a sentence, and' [^ a] 'is a character other than the letter a. , '(X | y)' is the character string x or the character string y, and '(x)? 'Indicates an empty character string or a character string x, respectively. "..." is "." * 'Represents. Further, in the standard regular expression, “+” indicates one or more times of the preceding character, but in this example, it is treated as a mere character (a symbol that separates morphemes).

In the inverted syntax, the topic of the sentence is not a noun phrase including a topic presentation expression such as'wa 'or'mo' at the end. Information about such sentence patterns is registered in the inverted syntax sentence pattern storage unit. An example of registration is shown in FIG. This sentence pattern is written by a regular expression described later, but the original sentence part that matches the '~' part in the pattern indicates the topic of the sentence. In addition, "-" is processed as a character string equivalent to "..." That is, ". *" At the time of pattern matching.

FIG. 21 shows the result of topic analysis for the input document example shown in FIG. In the second sentence, the 16th sentence, etc., the topic of the previous sentence is inherited. Also, the fifth sentence,
The information in the inverted syntax sentence pattern storage unit is used to extract the topic expression from the seventh sentence. The topic extracted from one sentence is not limited to one, and two topics are extracted in the ninth sentence and the eleventh sentence.

The anaphora resolution unit 5 refers to the topic analysis result and the morpheme / syntax result of each sentence to determine the anaphoric destination of the anaphoric expression such as adnominal / pronoun included in the sentence. here,
In addition to common sense things such as'this' it 'these' these ', expressions like'the same company',
I am thinking of including personal nouns such as'Mr. Doctor 'in the anaphoric expression. 5 and 6 show the recipient identification algorithm.

The anaphora resolution unit 5 first extracts pronouns and adnominals in the original sentence (steps S47, S49, S51).
This extraction and determination whether the phrase of the anaphoric destination of the extracted adnominal / pronoun is in the own sentence or in the preceding sentence (step S45) determines the style pattern registered in the anaphoric expression storage unit as the morpheme of the original sentence. -It is performed at the same time by matching the parsing result. When multiple templates match, the template closer to the beginning of the sentence is given priority.
If the matched parts overlap, the template that matches longer in the sentence is given priority. FIG. 15 shows an example of registration in the anaphoric expression sentence pattern storage unit. Rhetorical topic expression storage unit 53a
Like the inverted syntax sentence pattern storage unit 53b, it is described by a regular expression.

In FIG. 16, is a template that matches a sentence starting with "that", "that", or "this". Since the morpheme sequence is specified, it does not match a conjunction such as "So". The adjunct / pronoun existing at the beginning of a sentence can be regarded as the destination of the preceding sentence.

Matches a sentence such as "A is B, but it is ...". Let A and B be proper noun phrases. In such a sentence pattern, the reference destination of "the" can be regarded as A or B, or the entire part "A is B".

"C where A is B is
…. Matches the sentence "." In this case, the destination of "the" is C.

Matches a sentence such as "A is B, but C is ...". In this case, it can be considered that the "destination" is A or B, or "A is B".

When the destination is forward, the destination is determined. When the anaphoric expression is a pronoun such as "that" or "kore" (step S51), the topic of the preceding sentence is set as the anaphoric destination. When the anaphoric expression is an adjunct such as "that" or "this" and the modified word appears in the preceding sentence, the noun phrase of the preceding sentence including the modified word is set as the destination (step S7).
7).

If there is no such phrase in the preceding sentence and the modified word is a phrase such as “event”, “method”, “situation” that makes it easy to refer to an event instead of an object (step S63), The whole is set as the reference destination (step S79). Information about what kind of phrase it is is registered in the propositional anaphora expression storage unit 53d. If not, the topic in the previous sentence is set as the reference destination (step S).
65).

FIG. 22 shows the result of anaphora resolution for the sentence shown in FIG. It is determined that the reference destinations of the anaphoric expressions in sentences 2, 12, and 15 are within the own sentence. In sentences 4 and 16, since the modified word exists in the preceding sentence, the noun phrase of the preceding sentence including it is determined to be the destination. Sentence 8 is an example in which the topic of the previous sentence is determined to be the reference destination. In sentence 17, the word "policy" is a propositional anaphoric expression.
The entire preamble (16 sentences) is the reference destination.

The abstract generation / display unit 7 creates an abstract sentence to be displayed based on the anaphora resolution result, the morpheme / syntax analysis result of each sentence, and the user-specified abstract sentence number specification. Change the display, display the phrase of the reference destination, and redo the abstract according to the instructions. 6 to 15 and FIGS. 24 to 33 describe the operation of the abstract generation / display unit 7.

In the following description, the number of sentences specified by the user is 5, and the abstract generation / display unit 7 activates the important sentence determination unit 9 with that number as an argument, and the sentence determined as an important sentence is It is assumed that there are four sentences of 1, 8, 12, and 17.

FIG. 6 shows an algorithm for displaying. "..." is inserted between the sentences that are not continuous in the original sentence, and the connection expression of the sentence is deleted. FIG. 23 shows a display example of the abstract thus created.

In Japanese Patent Application No. 04-100168, when the selected important sentences are discontinuous in the original sentence, a method of referring to the context structure and replacing the connection expression in the sentence with an appropriate one is proposed. Has been done. It is easy to use the technique in this embodiment.

FIG. 7 shows an algorithm for performing the operation described in claim 2.

When the abstract sentence as shown in FIG. 23 is being displayed, for example, it is assumed that the symbol “...” In front of the second sentence being displayed is clicked with a mouse or the like. Then,
In this case, the sentences from the second sentence to the seventh sentence are displayed. Figure 2
4 is an example of the display. In this example, the second sentence to the seventh sentence are inserted in the display, and the inserted portion is shown by inversion (indicated by Rv in the drawing) or the like so that it is easy to understand.

FIG. 25 shows another display example. Here, the second sentence to the seventh sentence are displayed in another window, and the clicked place and the window are connected by a line.

FIG. 8 shows an algorithm for displaying. Now, suppose that the abstract sentence shown in FIG. 23 is being displayed. Here, FIG. 26 shows a display example when the anaphoric expression “here” of the second sentence being displayed is clicked. A sentence including the phrase of the reference destination is inserted and highlighted by reverse display or the like.

Similarly, the anaphoric expression "Kono" of the fourth sentence being displayed
FIG. 27 shows a display example when the button is clicked. In this case, the sentence containing the phrase of the reference destination is already displayed, so
The relevant part is indicated by reverse display or the like.

FIG. 9 shows an algorithm for displaying. Important sentences that include anaphoric expressions for non-important sentences are important sentence 2 (original sentence number 8) and important sentence 4 (original sentence number 17). If both are included, the number of important sentences becomes 6, which exceeds the number of sentences specified by the user. In this case, the important sentence determination unit is restarted with argument 4. Then, it is assumed that the sentences determined as the important sentences at that time are three sentences of 1, 8, and 16. Just as before, there are two sentences that include anaphoric expressions for non-important sentences. This time, since the number of sentences including the reference destination is less than the number of sentences specified by the user, the abstract sentence is generated without restarting the important sentence determination unit. FIG. 28 is an example of the display. The display method described above is also used.

FIG. 10 shows an algorithm for displaying. This means that if the referent of an anaphoric expression in an important sentence is the whole sentence (in a non-important sentence), it is included in the abstract as an important sentence, and if the referent is a phrase (in the non-important sentence), it is in the important sentence. The anaphoric expression of is replaced with the phrase.

The basic operation is almost the same as that shown in FIG. 9, but in this case, the reference destination of the important sentence 2 (original sentence number 8) is a word and phrase, and when replacing it, the number of important sentences is The reference destination of the important sentence 4 (original sentence number 17) that does not increase is the entire sentence, and this is added to the abstract sentence by insertion instead of replacement. The number of important sentences is incremented. After all, the number of sentences after anaphoric processing is 5, which is equal to or less than the number of sentences designated by the user, so that the important sentence determination unit is displayed as it is as an abstract sentence without being restarted. FIG. 29 is a display example thereof.
The display method described above is also used.

FIG. 11 shows an algorithm for displaying. The difference from the example shown in FIG. 10 is that replacement is not performed, and the symbols '[', ']' indicating the inserted portion are added at the time of insertion. FIG. 30 is a display example thereof.
The display method described above is also used.

FIG. 12 shows an algorithm for displaying. This is to delete an important sentence from the abstract that includes an anaphoric expression to an unimportant sentence. In this example, important sentence 2 (original sentence number 8) and important sentence 4 (original sentence number 17) are deleted. FIG. 31 is a display example. The display method described above is also used.

FIG. 13 shows an algorithm for displaying. This removes adnominals that include anaphora from non-important sentences from important sentences. If the anaphoric expression is a pronoun, deleting the whole sentence usually removes the meaning of the sentence. In this example, important sentence 2
“Original” of (original number 8) is deleted.

However, in important sentence 4 (original sentence number 17), even if "this" is deleted, only "policy"
Difficult to grasp sentences. Generally, the modified word of an adnominal is a type of propositional anaphoric expression such as “event” or “situation”, and if there is no other modifier clause associated with the word, deleting the adjunct will make it impossible to grasp the sentence. In such cases,
Delete the entire sentence as well as the pronoun.

FIG. 32 is a display example. The display system according to claim 1 is also used.

As a conventional method, there is a technique for reducing a sentence by deleting unimportant parts such as adverb phrases and subordinate clauses in the sentence. An example of applying anaphora resolution processing to the technique will be described below.

A problem when deleting an adverb phrase or a subordinate phrase is when there is an anaphoric expression in the main clause and the destination of the anaphora is a phrase in the adverb phrase or subordinate phrase. In such cases, the adverbial subordinate clause should not be deleted.

FIG. 14 shows an algorithm for displaying. In the example of this sentence, "kore" at the beginning of the main clause of sentence number 12 is inserted before that, "although it is possible to solve anaphora to some extent only with rhetorical and practical information such as style and sentence pattern". This clause remains undeleted because it refers to a phrase. FIG. 33 is a display example.

The operation of the abstract creation / display unit has been described above in detail, but these processes may be freely combined. Further, these display processes may be dynamically switched to each other in accordance with the designation of the user. The process of the example shown in FIG.
It is also useful to combine it with an important sentence selection type automatic abstraction system when creating and displaying abstract sentences using the analysis results of the anaphora analysis part.

Although the actual structure of the data processed by each processing unit and storage unit is not mentioned in this embodiment, all the data are character string type and one processing unit (one registration) is one. You can think of it as the text that makes up a line.
When storing the correspondence of a plurality of data, it is divided by an appropriate field separator and the information is arranged and registered in one line in a predetermined order.

When referring to and retrieving such information, the number of field separators from the beginning of the sentence may be counted and the required field (column) data may be extracted. Registration and search for these data may be performed by adding or searching the text file, and any widely used method may be used. Further, although this example mainly describes the display of the abstract, the compression rate is 100%, that is, it can be used as a high-performance text browser in the case of not abstracting.

As described above, in this embodiment, since the abstract is generated based on the anaphora analysis result, the readability of the abstract text is high. In addition, when performing processing such as full-text search or information extraction on the abstract sentence, the number of keywords is less than that in the case without anaphora processing, and therefore better results can be expected. Some recent information retrieval systems use co-occurrence relations and syntactic relations of multiple words in one sentence.
Proper anaphoric processing does not reduce the search accuracy of such a search system even if one of the words is a pronoun. Further, it is possible to interactively change the length of the abstract, display an original text part that is not in the abstract, or display an anaphoric destination such as a noun or a pronoun in the abstract in accordance with the user's operation.

[0073]

As described above, according to the present invention, since the non-important sentence in the abstract sentence, for example, the reference destination of the anaphoric expression or the abbreviation display, is included in the abstract sentence, the readability of the abstract sentence, The naturalness can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an example of an abstract creation / display system according to the present invention.

FIG. 2 is a block diagram showing a schematic configuration of an anaphora resolution processing unit shown in FIG.

FIG. 3 is a block diagram showing a schematic configuration of an abstract generation / display unit shown in FIG.

FIG. 4 is a flowchart showing an operation algorithm of a topic analysis unit of the anaphora analysis processing unit shown in FIG.

5 is a flowchart showing an operation algorithm of an anaphora fixing unit of the anaphora resolution processing unit shown in FIG. 1; Is.

FIG. 6 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG. 1.

FIG. 7 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG. 1.

FIG. 8 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG.

9 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG.

10 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG.

11 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG.

12 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG.

FIG. 13 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG. 1.

FIG. 14 is a flowchart showing an algorithm of the abstract generation / display unit shown in FIG. 1.

FIG. 15 is a diagram showing an example of registration in the anaphoric expression sentence pattern storage unit shown in FIG. 2;

16 is a diagram showing an example of registration in the rhetorical topic expression storage unit shown in FIG.

17 is a diagram showing an example of registration in the inverted syntax sentence pattern storage unit shown in FIG.

FIG. 18 is a diagram illustrating a processing sentence example for explaining each processing.

FIG. 19 is a diagram showing the sentence numbers of sentences and the positions of anaphoric expressions.

FIG. 20 is a diagram showing a display example of a part of a morpheme / syntactic analysis result of a sentence.

FIG. 21 is a diagram showing a topic analysis result of a sentence.

FIG. 22 is a diagram showing a result of anaphora resolution of a sentence.

FIG. 23 is a diagram showing a display example.

FIG. 24 is a diagram showing a display example.

FIG. 25 is a diagram showing a display example.

FIG. 26 is a diagram showing a display example.

FIG. 27 is a diagram showing a display example.

FIG. 28 is a diagram showing a display example.

FIG. 29 is a diagram showing a display example.

FIG. 30 is a diagram showing a display example.

FIG. 31 is a diagram showing a display example.

FIG. 32 is a diagram showing a display example.

FIG. 33 is a diagram showing a display example.

[Explanation of symbols]

 1 Format Analysis Section 3 Morphological / Syntax Analysis Section 5 Anaphoric Resolution Section 7 Abstract Generation / Display Section 9 Important Sentence Determination Section 11 Keyboard 13 Display / Printer

Claims (1)

[Claims] 1. An abstract creating device for creating an abstract from an inputted original sentence, comprising: an abstract sentence creating means for creating an abstract sentence by selecting a word or phrase from the original sentence according to a predetermined standard; and the abstract sentence creating means. When there is a word that is not selected between the selected words and phrases, etc., a presentation means for presenting that fact between the words and phrases in the abstract sentence, and a word or phrase corresponding to the part presented by this presentation means are provided as necessary. An abstract creating device, comprising: a display unit for displaying the abstract.