JP2006053907A - Information extraction method, information extraction device, information extraction program, and recording medium recording information extraction program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information extraction method accurately simply realizing management of document or a document group, browsing, analysis, reuse, and the like, and to provide an information extraction device, an information extraction program, or a recording medium recording the information extraction program. <P>SOLUTION: The information extraction method extracting information of document has a language information obtaining step (S2) obtaining modification relation of content words, a depth information obtaining step (S3) obtaining depth information representing strength of modification relation of one content word obtained from the document with the language information obtaining step and other content word, and a partial structure data forming step (S4) forming partial structure data comprising content words having depth information of the prescribed value or more of one content word based on the depth information of each content word obtained with the depth information obtaining step. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an information extraction method, an information extraction apparatus, an information extraction program, and a recording medium for extracting document information for document summarization, document search, and the like.

  A large amount of electronic document data has been distributed, and a simple method for managing and reusing documents and document groups is required. As one of the measures for activating the reuse of digitized document data, a technique for processing text by extracting document topics such as document search, document classification, and inter-document comparison has been proposed. In addition, a topic extraction technique in a document is required for document summarization and digest generation for improving the viewability of the document. In document analysis and document browsing support, what kind of processing is performed for document topic extraction greatly affects processing in a later process.

Conventionally, keywords, co-occurrence keywords, information on two phrases in which dependency is established, and the like have been proposed as processing units for document processing. However, in order to express the contents of a document, the information obtained from them contains too little information to express a meaningful topic.
On the other hand, a method using a syntactic function unit as a document search unit has been proposed (see, for example, Patent Document 1). Specifically, the syntactic functional unit includes a noun phrase composed of a noun sequence, a verb phrase composed of a case component and a predicate, and a noun connected with the particle “no”.
However, in the method described in Patent Document 1, there is no mention of a method for expanding these units. When actually processing a document, it is desirable that processing in a further expanded unit is possible in addition to processing in a prescribed unit. This is because even if the effect cannot be obtained with the prescribed unit, the effect may be obtained with the expansion unit. For example, in a document search system, it is known that a method of performing a search using a “related word” that is an extension of a keyword as a query is effective as a measure for improving the search accuracy. . Even when comparing the contents of news articles and generating a digest, the basic topic unit in the news document is extracted, and then the words indicating the development of the topic are acquired and added to the basic topic unit. By generating the extended topic unit, it is possible to find out the development history of the topic in addition to the similarities and differences between the topic units. In addition, when a document having a low similarity between basic topic units is handled in an extended topic unit, a synergistic effect in which the similarity is increased by the added information can be expected.
On the other hand, a method has been proposed in which the target document is subjected to dependency analysis, the result is stored in the form of a syntax tree or a linear list, and the frequent pattern of the word and positional relationship is automatically extracted using constraints and parameters. (For example, refer to Patent Document 2). In the method described in Patent Document 2, when there are a large number of words in a frequent pattern, abundant information may be acquired to express the content of the document.
Japanese Patent No. 3040945 Japanese Patent No. 3353829

However, with the method described in Patent Document 2, frequent patterns may not be extracted depending on the type of target document. For example, in document data with little fluctuation in the text registration method accumulated at a call center etc., there is a possibility that frequent patterns can be acquired effectively using words and positional relationships, but they are handled like documents such as news and questionnaires. In document data whose contents and description manner are arbitrary, the frequent pattern cannot be obtained even if the frequency of the word and the positional relationship is measured.

  In addition, since a construction step of a syntax tree and a linear list is included, it is assumed that a large amount of storage capacity is required for processing, which is not practical. Furthermore, when extracting a frequent pattern as an analysis unit from a syntax tree or a linear list, a constraint condition that specifies the pattern is necessary, but it is difficult to register this constraint condition unless you have expertise in the language processing field. It is.

  In view of the above problems, the present invention provides an information extraction method, an information extraction apparatus, and an information extraction program that can accurately and easily realize management, browsing, analysis, reuse, and the like of a document or a document group without depending on a target document. Alternatively, an object is to provide a recording medium on which an information extraction program is recorded.

  In view of the above problems, the present invention provides an information extraction method for extracting document information, a language information acquisition step for acquiring dependency relations of independent words, and one independent word acquired from the document by the language information acquisition step. Based on the depth information acquisition step for acquiring depth information indicating the strength of the dependency relationship with other independent words, and the depth information of each independent word acquired in the depth information acquisition step. And a partial structure data generation step of generating partial structure data composed of independent words having depth information equal to or greater than a predetermined value of the word.

  According to the present invention, it is possible to provide an information extraction method that can accurately and easily realize management, browsing, analysis, reuse, and the like of a document or a document group without depending on a target document.

  Further, in one aspect of the present invention, the language information acquisition step acquires the independent word of the dependency for each independent word of the dependency source, and the depth information acquisition step uses the predetermined word of the independent word as an extraction key. An extraction key selection step to select; an independent word that is related to the selected extraction key; and an independent word acquisition step that gives depth information of an initial value to the independent word; and Depth that obtains a self-supporting independent word whose word is a dependency, and adds a value obtained by adding a predetermined value to the depth information of the acquired independent word as depth information of the related self-supporting word And an information adding step. The “acquired independent word” is an independent word that modifies the extraction key, and the dependency independent word is an independent word that modifies “the independent word that modifies the extraction key”. Therefore, in the depth information addition step, the independent words having the modification relationship are traced one after another, and the depth information is added by a predetermined value. When all the independent words having the modification relationship are traced, the depth information adding step for the extracted key is finished. The “initial value” is 0 (zero), for example, and the depth information is incremented from 0 by a predetermined value. The predetermined value is preferably 1, but the magnitude of the predetermined value may be changed instead of being constant according to the modification relationship. For example, the depth information is greatly added when the prescription is modified, and the added value of the depth information is decreased when the body is modified.

  According to the present invention, by using a predicate as an extraction key, and extracting an independent word that directly modifies the extracted key and further an independent word that modifies the extracted independent word from the sentence, The extraction range of structure elements can be automatically defined, and information-rich document information can be extracted. It is also possible to extract a self-supporting word that is directly modified by the extraction key, and to extract as many self-supporting words that have modified the extracted self-supporting word from the sentence.

  Further, in one aspect of the present invention, the language information acquisition step acquires the independent word of the dependency for each independent word of the dependency source, and the depth information acquisition step selects a predetermined body language from the independent words as an extraction key. An extraction key selection step, an independent word acquisition step of acquiring an independent word on which the selected extraction key is related, and giving depth information of an initial value to the independent word, and the acquired independent word Depth information that obtains a self-supporting independent word that is a dependency and adds a value obtained by adding a predetermined value to the depth information of the acquired independent word as depth information of the self-supporting word And an adding step.

  According to the present invention, by using a description as an extraction key, and extracting an independent word that directly modifies the extracted key and further an independent word that modifies the extracted independent word from the sentence, the partial structure The extraction range of elements can be automatically defined, and document information rich in information can be extracted.

  In one aspect of the present invention, the depth information acquisition step includes a depth information designation step for designating a value of depth information, and the value of the depth information is designated by the depth information designation step. Independent words of the following depth information are acquired.

  According to the present invention, the analysis process is stopped in the modification relation analysis up to the depth specified by the user, and the entire analysis time is shortened by moving to the modification relation analysis using the next predicate or body as the extraction key. And efficient processing can be realized.

  Further, in one aspect of the present invention, a connection particle or a connection particle having a connection particle or a connection particle is described, and the depth information acquisition step includes a connection particle or a connection particle included in a phrase including the acquired independent word. Based on the conjunction / junction particle dictionary, the connection meaning of the conjunction or the connection particle is extracted, and then the extracted connection meaning is added to the extension element of the acquired independent word. To do.

  According to the present invention, it is possible to add the meaning of connection to the extracted independent words, and it is possible to find out the development process of the document during document processing.

  Further, in one aspect of the present invention, the information processing apparatus includes an intention expression dictionary that describes an expression that represents an intention, and the depth information acquisition step is based on an intention expression included in a phrase that includes the acquired independent word. The meaning of the intention expression is extracted, and then the meaning of the extracted intention expression is added to the extension element of the acquired independent word.

  According to the present invention, an intention expression representing an intention can be added to the extracted independent word, and processing using the intention information effectively is possible during document processing.

  Further, in one aspect of the present invention, the depth information acquisition step, when the acquired independent word is included in the last phrase in one sentence, the end of the sentence is indicated in the extended element of the independent word It is characterized by adding.

  According to the present invention, when the extracted independent word is included in the last phrase in one sentence, the end of the sentence can be added to the extended element of the independent word, and at the time of document processing , It is possible to effectively use the sentence end phrase.

  Further, in one aspect of the present invention, the partial structure data generation step is added to a modification relationship indicating a relationship in which the independent word acquired by the depth information acquisition step modifies the extraction key and an extension element of the acquired independent word. The partial structure data is generated by associating the meaning of the connected connection, the meaning of the intention expression, or the notation at the end of the sentence with the acquired independent word.

  According to the present invention, when extracting and generating document information from the extracted extended partial structure element, the depth information is designated, and the document information is extracted by arbitrarily increasing / decreasing the content of the document information. Can do.

  Further, the present invention provides an information extraction apparatus for extracting information on a document, a language information acquisition unit that acquires a dependency relationship of independent words, a single independent word acquired from the document by the language information acquisition unit, and another Based on the depth information acquisition means for acquiring depth information indicating the strength of the dependency relationship with the independent word, and the depth information of each independent word acquired by the depth information acquisition means, And partial structure data generation means for generating partial structure data composed of independent words having depth information equal to or greater than a predetermined value.

  Moreover, in one aspect of the present invention, the language information acquisition unit acquires the independent word of the dependency for each independent word of the dependency source, and the depth information acquisition unit uses the predetermined word of the independent word as an extraction key. An extraction key selection means for selecting, an independent word acquisition means for acquiring an independent word that is related to the selected extraction key, and giving the initial value depth information to the independent word; and Depth that obtains a self-supporting independent word whose word is a dependency, and adds a value obtained by adding a predetermined value to the depth information of the acquired independent word as depth information of the related self-supporting word And an information adding means.

  Further, in one aspect of the present invention, the language information acquisition unit acquires the independent word of the dependency destination for each independent word of the dependency source, and the depth information acquisition unit selects a predetermined body language from the independent words as an extraction key. An extraction key selection unit that acquires the independent word that is the destination of the selected extraction key and assigns initial depth information to the independent word; and A depth at which a self-supporting word whose dependency word is a dependency is acquired, and a value obtained by adding a predetermined value to the value of the acquired self-supporting word depth information is obtained as depth information of the dependency self-supporting word. And an information adding means.

  In one aspect of the present invention, the depth information acquisition means includes depth information specifying means for specifying a value of depth information, and the value of the depth information specified by the depth information specifying means. Independent words of the following depth information are acquired.

  Further, in one aspect of the present invention, a connection particle or a connection particle is described as a connection particle or a connection particle, and the depth information acquisition unit includes a connection particle or connection particle included in a phrase including the acquired independent word. Based on the conjunction / junction particle dictionary, the connection meaning of the conjunction or connection particle is extracted, and then the extracted connection meaning is added to the acquired independent word extension element. And

  Further, in one aspect of the present invention, there is an intention expression dictionary in which expressions representing intentions are described, and the depth information acquisition unit generates the intention expression dictionary based on the intention expressions included in the phrase including the acquired independent words. The meaning of the intention expression is extracted with reference, and then the meaning of the extracted intention expression is added to the obtained extension element of the independent word.

  Further, in one aspect of the present invention, the depth information acquisition means, when the acquired independent word is included in the last phrase in one sentence, the end of the sentence is indicated in the extended element of the independent word. It is characterized by adding.

  Further, the partial structure data generation means includes a modification relation indicating a relation in which the independent word acquired by the depth information acquisition means modifies the extraction key, and the meaning and intention of the connection added to the acquired extension element of the independent word. The partial structure data is generated by associating the meaning of the expression or the notation at the end of the sentence with the acquired independent word.

  Moreover, this invention provides the information extraction program which makes a computer perform said information extraction method. The present invention also provides a recording medium on which an information extraction program is recorded.

  Further, the present invention provides an information extraction method for extracting document information, a language information acquisition step of acquiring a dependency relation of independent words, a phrase part of speech indicating a part of speech information of a phrase, or a phrase type indicating phrase classification information, Referring to the extraction key selection rule that defines the phrase type in association with the phrase part of speech, the extraction key selection step for selecting the extraction key, and the extended rule defined based on the phrase part of speech and the dependency relationship are extracted keys And an extended structural element generation step for generating an extended structural element by extracting a clause that matches the extended rule from the sentence.

  According to the present invention, an extended structural element extraction key selection rule is provided, and an extended rule for selecting a clause to be an extended structural element is provided, so that an extended structural element with comprehensive and low redundancy can be obtained. Can be generated.

  Further, in one embodiment of the present invention, the phrase part of speech is any one of a predicative clause, a verbal clause, a sa modification clause, a copula clause, an adjective clause, an adverbial clause, a combined clause, a connectivity clause, or an indefinite clause. The phrase type is any one of sentence ending, punctuation, engagement, cut or indefinite.

  Further, in one embodiment of the present invention, the extended structural element includes a predicative clause, a verbal clause, a sacrificial clause, a copula clause, an adjective clause, an adverbial clause, a combined clause, a connectivity clause, or an indefinite clause. It is characterized by being comprised by the basic unit in any one of these, or the combination of this basic unit.

  In one embodiment of the present invention, each clause of the basic unit has any one of the following types of clauses: reading, sentence ending, no relation, relation or indefinite.

  Further, in one aspect of the present invention, the extraction key selection step includes any of the attributes of a predicative phrase that does not have an indefinite attribute, a modified phrase that does not have an indefinite attribute, a copula phrase, a sentence end, a punctuation mark, and a disconnection attribute. Any one of an adjective phrase having an adverbial phrase having an end-of-sentence or disconnection attribute, or an episodic phrase having an end-of-sentence or disconnection attribute is selected as the extraction key.

  In one embodiment of the present invention, the extended structural element generation step uses the extraction key selected in the extraction key selection step as an expansion target clause for extracting a phrase from a sentence. In the case of a sex clause, a samurai clause, a copula clause, an adjective clause, or an adverbial clause, a prescriptive clause, a sacrificial clause, an adjective clause, or a somatic clause that has an attribute related to the extension target clause Or a forward extension processing step of adding an adverbial clause to the extraction key to generate an extended structure element, and the clause added by the forward extension processing step is a prescriptive phrase, a sacrificial phrase, an adjective In the case of a clause or a body phrase, the forward extension processing step is further performed with the clause as an extension target clause.

  In one embodiment of the present invention, the partial structure data generation step uses the extraction key selected in the extraction key selection step as an expansion target clause for extracting a phrase from a sentence, and the expansion target clause is an episodic phrase. A forward extension processing step of adding an adjective clause, a verbal clause or an adverbial clause having an attribute related to the extension target clause to generate an extended structural element by adding the extracted key, When the clause added by the processing step is an adjective clause or a body clause, the forward extension processing step is further performed with the clause as an extension target clause.

  In one form of the present invention, the partial structure data generation step uses the extraction key selected in the extraction key selection step as an expansion target clause for extracting a phrase from the sentence, and sets the attribute related to the expansion target clause. If the clause has a predicative clause, a modified clause, an adjective clause, or an adverbial clause, and if the destination clause of the target clause is an episodic clause, add the destination clause to the extraction key A backward extension processing step for generating an extended structural element, and the verbal clause added by the backward extension processing step has a dependency attribute, and the destination clause of the added verbal phrase is In the case of a sex clause, the backward extension processing step is further performed with the clause as the extension target clause.

  In one embodiment of the present invention, the partial structure data generation step uses the extraction key selected in the extraction key selection step as an expansion target clause for extracting a phrase from the sentence, and the expansion target clause has an associated attribute. A backward extension processing step for generating an extended structural element by adding a related clause to the extraction key if the related clause of the extension target clause is a semantic clause or a copula clause And the semantic clause added by the backward extension processing step has a dependency attribute, and if the destination clause of the added semantic clause is a semantic clause, that clause is used as the extension target clause. Further, a backward extension processing step is performed.

  Further, the present invention provides an information extraction device for extracting document information, language information acquisition means for acquiring a dependency relationship of independent words, a phrase part of speech indicating a part of speech information of a phrase, or a phrase type indicating phrase classification information, With reference to the extraction key selection rule that defines the phrase type in association with the phrase part of speech, the extraction key selection means for selecting the extraction key, and the expansion rule defined based on the phrase part of speech and the dependency relationship are extracted keys And an extended structural element generating means for generating an extended structural element by extracting a clause that matches the extended rule from the sentence.

  Further, in one embodiment of the present invention, the phrase part of speech is any one of a predicative clause, a verbal clause, a sa modification clause, a copula clause, an adjective clause, an adverbial clause, a combined clause, a connectivity clause, or an indefinite clause. The phrase type is any one of sentence ending, punctuation, engagement, cut or indefinite.

  Further, in one embodiment of the present invention, the extended structural element includes a predicative clause, a verbal clause, a sacrificial clause, a copula clause, an adjective clause, an adverbial clause, a combined clause, a connectivity clause, or an indefinite clause. It is characterized by being comprised by the basic unit in any one of these, or the combination of this basic unit.

  In one embodiment of the present invention, each clause of the basic unit has any one of the following types of clauses: reading, sentence ending, no relation, relation or indefinite.

  Further, in one aspect of the present invention, the extraction key selection means includes any of the attributes of a predicative phrase having no indefinite attribute, a sacrificial phrase having no indefinite attribute, a copula phrase, a sentence end, a punctuation mark, and a disconnection attribute. Any one of an adjective phrase having an adverbial phrase having an end-of-sentence or disconnection attribute, or an episodic phrase having an end-of-sentence or disconnection attribute is selected as the extraction key.

  In one embodiment of the present invention, the extended structural element generation means uses the extraction key selected by the extraction key selection means as an extension target phrase for extracting a phrase from a sentence, and the extension target phrase has a descriptive property. In the case of a clause, sa modification clause, copula clause, adjective clause, or adverbial clause, a phrasal clause, a sacrificial clause, an adjective clause, a somatic clause, Alternatively, it has forward extension processing means for adding an adverbial clause to the extraction key to generate an extended structure element, and the clause added by the forward extension processing means is a prescriptive phrase, a sacrificial phrase, an adjective phrase Alternatively, in the case of an episodic clause, the forward extension processing means is further performed with the clause as the extension target clause.

  In one embodiment of the present invention, the partial structure data generation means uses the extraction key selected by the extraction key selection means as an extension target phrase for extracting a phrase from a sentence, and the extension target phrase is an episodic phrase. Forward expansion processing means for generating an extended structural element by adding an adjective clause, a verbal clause or an adverbial clause having an attribute related to the expansion target clause to the extraction key, When the clause added by the means is an adjective clause or an episodic clause, the forward extension processing means is further performed with the clause as an extension target clause.

  In one embodiment of the present invention, the partial structure data generation means uses the extraction key selected by the extraction key selection means as the extension target phrase for extracting a phrase from the sentence, and sets the attribute related to the extension target phrase. If the clause has a predicative clause, a modified clause, an adjective clause, or an adverbial clause, and if the destination clause of the target clause is an episodic clause, add the destination clause to the extraction key And a backward extension processing means for generating an extended structural element, and the verbal phrase added by the backward extension processing means has a dependency attribute, and the destination clause of the added verbal phrase is In the case of a sex clause, the backward extension processing means is further performed with the clause as the extension target clause.

  Further, the partial structure data generation means is a syntactic phrase in which the extension target clause has an attribute related to the extraction key selected by the extraction key selection means as an extension target clause for extracting a phrase from the sentence. And the extension target clause includes a backward extension processing means for adding an extension clause to the extracted key and generating an extension structure element when the dependency clause of the clause to be expanded is a body phrase clause or a copula clause, If the verbal clause added by the processing means has a dependency attribute, and the dependency clause of the added verbal clause is a verbal clause, the clause is set as an extension target clause and further backward extension processing means It is characterized by performing.

  The present invention also provides an information extraction program for causing a computer to execute the information extraction method according to claims 19 to 27. The present invention also provides a recording medium on which an information extraction program is recorded.

  An information extraction method, information extraction device, information extraction program, or information extraction program that can accurately and easily realize management, browsing, analysis, reuse, etc. of a document or a document group without depending on the target document A medium can be provided.

  The best mode for carrying out the present invention will be described below with reference to the accompanying drawings.

  FIG. 1A is a flowchart showing the overall flow of information extraction processing in this embodiment. A solid line indicates the flow of processing, and a broken line indicates the flow of processing target data. Each step is executed by the document registration unit 31, the language information acquisition unit 32, the partial structure element extension extraction unit 33, the partial structure data generation unit 34, and the output unit 35 described in each step.

  FIG. 1B shows a functional configuration diagram of the information extraction apparatus. The information extraction apparatus is configured to include a document registration unit 31, a language information acquisition unit 32, a partial structure element extension extraction unit 33, a partial structure data generation unit 34, and an output unit 35. Processing executed by these means will be described. First, an outline of the overall flow of the information extraction process will be described.

(Step S1)
When the information extraction is started, the document is registered by the document registration process. The document registration process is executed by the document registration unit 31.

(Step S2)
The registered documents are sequentially stored in the document storage unit 11. When all the registration target documents are registered, language information acquisition processing is performed from the stored documents. The language information acquisition process is executed by a language information acquisition unit.

  The language information is acquired by using an existing method such as a method of acquiring language information by performing a language analysis process or a method of acquiring automatically or semi-automatically while receiving a user instruction. Language information is at least information on the words that make up the document, clauses (one independent word or one unit in Japanese grammar with one or more attached words attached to one independent word), information between phrases Refers to dependency relationship information.

  The dependency relationship between clauses is a grammatical relationship that exists between clauses. For example, in the sentence “I ate apples”, the verbal phrase “I am” “I ate” is grammatically modified, and “I” ate “I ate”, or it ’s a dependency. Because the dependency relationship name is accompanied by the case particle “Ga”, it is called “Ga case continuous modification relationship”. Concurrent modification refers to a phenomenon in which a phrase modifies a idiom phrase. In addition, the prescriptive phrase “I ate” grammatically modifies the verbal phrase “Ringo.”, And its dependency relation name is “multiple modification relation”. Linkage modification refers to the phenomenon in which a certain clause modifies an episodic clause. Of the dependency relationship names, “parallel” is processed as one variation of the linkage modification. In the embodiment, “parallel”, which is a dependency relationship of the phrase pair “from the system” to “from the new system”, is processed as a combination modification. Here, a predicate is a word that can be used among independent words that can become a phrase alone, for example, a verb or an adjective. A prescriptive phrase is a phrase that includes a predicate and forms a predicate. In the embodiment, the verbal clauses other than the prescriptive clauses are set as the verbal clauses. A syntactic phrase is composed of independent words that are not used, and the last independent word is defined as a noun phrase. Self-independent words of sa variable nouns generate both prescriptive and somatic verbal clauses.For example, if there is a particle or asserted auxiliary verb immediately followed by a verbal phrase, the verbal verb "do" or "can" is added immediately after. If this is the case, it can be determined at the time of dependency analysis processing, depending on how to connect the idiom clause and the attached word. Documents to which language information is added are sequentially stored in the language information storage unit 12.

(Step S3)
When the linguistic information acquisition process for all the registered documents is completed, the substructure element extension extraction process is executed on the document to which the stored linguistic information is added. The partial structure element extension extraction process is executed by the partial structure element extension extraction means 33.

  As shown in FIG. 1B, the partial structural element extension extraction unit 33 includes depth information acquisition unit 330. The depth information acquisition unit 330 includes extraction key selection unit 331, independent word acquisition unit 332, depth Depth information adding means 333 and depth information specifying means 334.

  The depth information acquisition means 330 is a depth indicating the strength of the dependency relationship between one independent word (hereinafter referred to as an extraction key) acquired from the document by the language information acquisition process (S2) and another independent word. Get information. By the depth information acquisition means 330, the strength of the dependency relationship between the extraction key and each other independent word is represented by the depth information.

  The extraction key selection means 331 extracts the words or the independent words constituting the document as extraction keys in order from the end of the document. The independent word acquisition unit 332 acquires all the independent words that are related to the selected extraction key. Then, initial depth information, for example, 0 is assigned to each independent word. Therefore, the depth information of the independent word that directly modifies the extraction key is zero. The depth information addition means 333 acquires the independent word in which the acquired independent word is a dependency (hereinafter referred to as a dependency independent word), and acquires the acquired independent as the depth information of the dependency independent word. A value obtained by adding a predetermined value to the value of the word depth information is assigned. Therefore, the depth information of the dependency independent word becomes the depth information of 1 by adding a predetermined value (for example, 1) to 0 if the dependency independent word modifies the independent word of depth information 0. . Similarly, if the dependency independent word modifies the independent word of the depth information 1, the depth information becomes 2. The depth information specifying unit 334 specifies the value of depth information. If depth information is specified by the depth information specifying means 334, an independent word of depth information equal to or less than the specified value is acquired, and the information extraction process is terminated.

  For each sentence of one document, partial structure element extension extraction is performed using language information. When the partial structure element expansion extraction of one sentence is completed, the extraction result is stored in the partial structure element storage means 13, and the partial structure element expansion extraction process and storage are similarly executed from the document to which the language information of the next sentence is added. To do. When the partial structure element extension extraction process and storage are completed for the entire sentence of one document, the same extended partial structure element extraction process and storage are executed from the first sentence of the next document.

(Step S4)
When the partial structure element extension extraction process and storage are completed for all registered documents, the partial structure data generation process for generating the partial structure data in order from the head of the extended partial structure element stored in the partial structure element storage unit 13 is performed. Execute. The partial structure data generation process is executed by the partial structure data generation means 34.

  The generated partial structure data is sequentially stored in the partial structure data storage means 14, displayed on a predetermined display means, and when the partial structure data generation processing, storage, and display are completed for all partial structure elements, information extraction processing is performed. Exit.

  Next, the language information acquisition process (S2), the partial structure element extraction process (S3), and the partial structure data generation process (S4) will be described in detail.

[Language information acquisition processing]
S21) First, one sentence is extracted from the head of the document, language analysis processing is performed, and language information is acquired.
S22) The language analysis process uses existing methods to acquire word information, phrase information, and dependency relation information in a sentence. The language information acquisition result of the example sentence is as shown in FIG.
S23) When the language analysis process for one sentence is completed, the sentence whose language information has been acquired is stored in the language information storage means 12.
S24) If there is a next sentence in the registered document, the process returns to S21) to perform language analysis on the next sentence. This operation is performed until there are no more sentences in the document. When the language analysis process for the entire sentence is completed, the process proceeds to the partial structure element extension extraction process.

  FIG. 2 is an example of a language information acquisition result obtained as a result of language analysis. The linguistic information acquisition result in FIG. 2 is “But, if the difference in performance between the current system and the new system opens, consumers and companies will want to purchase such a new system.” did.

  The document with linguistic information obtained as a result of the language analysis is expressed by two lines of one phrase, and the first line starts from the beginning with "Section number of the corresponding phrase", "Section number of the destination of the corresponding phrase", It consists of a dependency relation name for a phrase, an arrow, a description of the corresponding phrase, and a phrase attribute. The phrase attribute refers to whether the phrase is an episodic or a idiom.

  The second line is constituent word information of the corresponding clause, and from the first word of the relevant clause, word notation and part-of-speech pairs are described for the number of constituent words of the relevant clause. In the example of the first clause “E even” in the document of FIG. 2, the clause number is [0], the dependency clause is the clause [14], and the dependency relation name to the dependency clause is It is a modification, and the notation of this clause is [But], and the clause attribute is episodic. The constituent words indicate the conjunction “even” and the punctuation mark “,”.

[Substructure element extension extraction process]
The partial structure element extension extraction process will be described using the language information acquisition result of FIG. First, a case where extended partial structural elements of all depth hierarchies are extracted will be described.

  First, the result of the partial structure element extension extraction process will be briefly described. 3 and 4 are extraction examples in which extended partial structural elements are extracted from the language information acquisition result of FIG. For one extracted key, there are zero or more related independent words attached, and the extracted key line extracts the part of speech as the type of extracted key (prescriptive or syntactic), the extracted key notation, and the extracted key attribute. ing. The relation independent word line includes the depth of the relation (an integer value obtained by adding 0 to 1), the notation of the extracted word, the part of speech as the attribute, the qualified relation name in the dependency with the extraction key, and the extension element (0 or more) Variable number). As will be described later, intention expressions and connection meanings are added to the extension elements. The extension element indicates the development history of the topic and adds to the similarity and difference between the topic units. In addition, it can be expected that a document having a low similarity in word notation or part-of-speech name has a high similarity due to the information of the extension element.

  The substructure element extension extraction process extracts the substructure element by extracting each element of the relationship depth, word notation, part of speech as an attribute, modification relation name in the dependency with the extraction key, and extension element from one sentence. Perform extraction. Partial structure data is generated based on the extracted extended partial structure element shown in FIG.

(Extraction of substructure elements using the word "say" as an extraction key)
FIG. 5A is a flowchart showing a processing flow in the case of extracting an extended partial structure element using a prescription as an extraction key. The flowchart of FIG. 5A shows the policy of the processing 1) to 101) below. First, referring to the language information acquisition result of FIG. 3, the predicates are extracted one by one as an extraction key in order from the end of the sentence (S101). When the extraction key is extracted, a dictionary which will be described later is referred to whether or not the extraction key has an intention expression. If the extraction key has an intention expression, the extraction key is extracted (S102). Next, independent words that use the extracted key in combination are extracted and their depth information is set to an initial value of 0 (S103). Extract (S104). Next, 0 is input to the count constant k for counting the depth information.

  Note that, by setting the processing of step S103 to “extract an independent word that directly modifies the extraction key“ ga ”as an independent word of depth information 0”, it is also possible to extract a dependency relationship from which the extraction key is a source. In the following 1) to 101), the dependency relationship from which the extraction key is the dependency source is also extracted (processing 8) and 163)).

  Next, an independent word that modifies the independent word of the depth information k (= 0) is extracted, and depth information of k + 1 (= 1) is given to the extracted independent word. Further, if there is an intention expression in the phrase including the independent word of the depth information k + 1, it is extracted (S107).

  Next, if there is another independent word whose depth information is k (= 0) (an independent word that uses the extracted key for continuous modification), steps S106 to S107 are repeated (S108).

  Next, it is determined whether or not there is an independent word that modifies the independent word of the depth information k + 1 (= 1) (N in Step S109), and if it exists, k + 1 is input to k (S110). Up to this point, independent words with depth information 0 and 1 have been extracted, and the count constant k has become 1. By repeating steps S106 to S107, the count constant k is incremented until there is no dependency relationship for one extracted key, and the value of the count constant k is given as depth information to the independent word that has become a dependency destination.

  If there is no self-supporting word that modifies the self-supporting of the depth information k + 1 (Y in step S109), the same processing is repeated using the next premise as an extraction key (N in step S111). When the processing is completed for all the predicates (Y in step S111), the processing in FIG.

  Next, the partial structural element extraction process of FIG. 5A will be described in detail based on example sentences. In the partial structure element extraction process, the dictionary shown in FIG. 6 is used. FIG. 6A is an example of a dictionary in which correspondence between connection particles, conjunctions, and connection meanings is described. Connection dictionary or conjunction notation and part of speech and corresponding connection meanings are described. FIG. 6B is an example of a dictionary in which the correspondence between the expression indicating the intention and the intention is described. The notation of the word expressing the intention and the intention corresponding to the part of speech are described.

1) First, the sentence of FIG. 2 with one sentence of language information from the head is extracted from the language information storage means 12, and a predicate is searched from the last sentence. In FIG. 2, the last clause is the clause of clause number [16].
2) Since clause number [16] is an episodic clause, clause number [15] is examined. Since the phrase number [15] is a predicate phrase, the predicate “say” is extracted from the phrase number [15].
3) As an extraction key, the notation “say” and the part of speech attribute “verb” are extracted and stored. Further, it is checked whether or not there is an intention expression in the same phrase with reference to the dictionary of FIG. 5B, and if there is, the corresponding intention is registered in the extension element (hereinafter referred to as intention expression check). Since there is no intention expression in this clause, it is not registered as an extension element.
4) Next, from the phrase number [15], search for independent words in the phrase in which the predicate “say” is directly modified in order from the phrase number [15].
5) Since the destination clause number of clause number [14] is [15], it can be seen that the clause of clause number [14] directly modifies “say”, and “is” is found. Since “Naru” is an independent word that directly modifies the extracted key, the depth 0, the notation “Naru”, the part-of-speech attribute “verb”, and the dependency relation name “Glyph combination modification” are stored. Also, an intention expression check is performed to determine whether there is an intention expression in the same phrase. Referring to the dictionary of FIG. 6, there is no intention expression in the phrase.
6) Further, from the phrase number [14], search for independent words in the phrase in which the predicate “say” is directly modified in order from the phrase number [14]. In this search, if the dependency relationship name of the immediately preceding clause is “Parallel” and the dependency clause number is the clause number of the corresponding clause, the independent word of the immediately preceding clause is also extracted, The reception relationship is stored as the dependency relationship name of the phrase to be processed. This is called a parallel check. When the parallel check is performed, it can be seen that there is no clause that is directly qualified with “parallel” in clause number [14], so the search for the first clause is advanced.
7) Search from phrase number [13] to head phrase number [0]. Since there is no clause whose dependency clause number is [15], the next independent word directly modified by “say” is searched from the clause number [16] toward the end clause.
8) Since the destination clause number of clause number [15] is [16], it can be seen that the clause of clause number [15] directly modifies “that” of clause number [16]. , "Koto" is discovered. Since “Koto” is an independent word directly modified by the extraction key, the depth 0, the notation of “Koto”, the part-of-speech attribute “formal noun”, and the dependency relationship “combined modification” are stored. At this time, a flag indicating a difference from the independent word of the previous consecutive modification relationship is set. In FIG. 3, flags are represented by asterisks. Furthermore, the intention expression in the same clause is checked, but there is no intention expression in the same clause.
9) Further, from the phrase number [16], the self-supporting words of the phrases in which the predicate “say” is directly modified in the order from the phrase number [16] are searched. In this search, if the dependency relationship name of the corresponding clause is “parallel” and the dependency clause number is the clause number of the immediately following clause, the independent word of the immediately following clause is also extracted, The dependency relationship is stored as the dependency relationship name of the phrase to be processed. This is also called a parallel check. A parallel check is performed on clause number [16]. It can be seen that there is no further independent word to be extracted because the dependency relation name is not “parallel” but a sentence end clause.
10) Due to the universal restriction of Japanese grammar “there is only one clause destination”, there is already an independent word that is directly modified by “say”. The extraction of the partial structural element having the depth 0 is finished.

11) Next, the independent words of the phrase having the dependency relationship are further extracted from the independent words having the dependency relationship with the extraction key. Independent words that have a dependency relationship with the extraction key “say” are “become” and “ko”. First, a phrase that directly modifies the phrase number [14] including “Naru” is searched from the phrase number [13] toward the first phrase.
12) Since the destination clause of the clause number [13] is the clause number [14], the independent word “purchase” is found from the clause number [13]. “Purchase” is an independent word that directly modifies the independent word “Naru” with a depth of 0. Therefore, by adding the depth, the depth 1 and the notation of “purchasing”, the part of speech attribute “sa kin”, The dependency relationship name “continuous modification” is stored. When the intention expression check in the same clause is performed, the intention “request” is registered in the extension element because it matches the expression “I want” shown in FIG. 5B.

13) Furthermore, a phrase that directly modifies clause number [14] is searched from clause number [13] toward the first clause, and through parallel checking, the independent word “company” is found from clause number [10]. . Since the independent word “Company” is an independent word that directly modifies the independent word “Naru” with a depth of 0, the depth is added to the depth 1 with the expression “Company” and the part-of-speech attribute “Noun”. , Store the dependency relation name “continuous modification”. When the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
14) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [9] toward the first phrase. At this time, if the parallel check is performed, the dependency relationship name of the clause number [9] is “parallel”, and the dependency clause number is the clause number of the corresponding clause [10]. The independent word “consumer” in [9] is also subject to extraction. Therefore, the notation of depth 1 and “consumer”, the part-of-speech attribute “noun”, and the dependency relation name are stored as the dependency relation name “continuous modification” of the clause to be processed. When the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
15) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the independent word “Den” is found from the phrase number [0]. Since the independent word “but” is an independent word that directly modifies the independent word “being” with a depth of 0, the depth is added, the depth 1 is expressed with “even”, and the part-of-speech attribute “connective” And the dependency relation name “continuous modification” is stored. Even if the part-of-speech is a conjunctive, the depth of the corresponding independent word is not 0. Therefore, the dictionary in which the correspondence between the connecting particle, the conjunctive, and the meaning of the connection is described cannot be referenced. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.

  16) Further, a clause that directly modifies clause number [14] is searched from the clause number [0] toward the first clause through a parallel check. When searching up to the first clause, there is no clause that directly qualifies the clause number [14], so the search for the independent word that directly qualifies the independent word “Naru” ends.

17) Next, a phrase that directly modifies the phrase number [16] including the independent word “koto” is searched from the phrase number [15] toward the first phrase. At this time, the phrase number [15] including the extraction key “say” is not searched.
18) Since the first phrase is searched and there is no phrase that directly modifies the phrase number [16], the retrieval of the independent word that directly modifies the independent word “Koto” is terminated.

19) Next, the phrase directly modified by the phrase number [16] including the independent word “Koto” is searched for toward the sentence end phrase.
20) Since clause number [16] is the end-of-sentence clause, there is no clause directly modified by clause number [16] containing the independent word “koto”, so the search for independent words directly modified by the independent word “koto” is terminated. To do.

  21) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 1 is extracted. The extracted independent words of depth 1 are “Purchase”, “Company”, “Consumer”, and “Even”. First, the phrase directly modifying the phrase number [13] including “purchase” is searched from the phrase number [12] toward the first phrase.

22) Since the dependency clause of the clause number [12] is the clause number [13], the independent word “system” is found from the clause number [12]. Since the independent word “system” is an independent word that directly modifies the independent word “purchase” at depth 1, add the depth to the depth 2, the notation of “system”, and the part of speech attribute “noun” And the dependency relation name “wo case qualification” is stored. When the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
23) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [11] toward the first phrase, and through the parallel check, the independent word “open” is found from the phrase number [8]. Since the independent word “open” is an independent word that directly modifies the independent word “purchase” at depth 1, add the depth, the depth 2, the expression “open”, and the part of speech attribute “verb” And the dependency relation name “continuous modification” is stored. Further, in the dictionary of FIG. 6A, when the presence of a connection particle in a phrase including “open” is checked, the connection particle “ba” matches the dictionary, and therefore the connection meaning “forward assumption” is added to the extension element. sign up. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
24) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [7] toward the first phrase, and through the parallel check, the search is performed up to the first phrase, but the phrase number [13] is directly qualified. Since there is no clause to do, the search for the independent word that directly modifies the independent word “purchase” is terminated.

  25) Next, a phrase that directly modifies the phrase number [10] including “company” is searched from the phrase number [9] toward the first phrase. Since the destination clause of clause number [9] is clause number [10], the independent word “consumer” is found from clause number [9]. The independent word “consumer” is an independent word that directly modifies the independent word “enterprise” at depth 1. Therefore, add the depth to the depth 2, “consumer”, and the part of speech attribute “noun”. , The dependency relation name “parallel” is stored. When the intention expression check in the same phrase is performed, there is no target expression in the same phrase.

  25) Furthermore, a clause that directly qualifies clause number [10] is searched from clause number [8] toward the first clause, and through the parallel check, the first clause is searched, but clause number [10] is directly modified. Since there are no clauses, the search for independent words that directly qualify the independent word “company” is terminated.

  26) Next, a phrase that directly qualifies the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  27) Next, the phrase directly modifying the phrase number [0] including “Even” is searched toward the first phrase. Since clause number [0] is the first clause, there is no clause that directly qualifies clause number [0], so the search for independent words that directly qualify the independent word “even” is terminated.

28) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 2 is extracted. The independent words of depth 2 are “system”, “open”, and “consumer”. First, a clause that directly modifies the clause number [12] including “system” is searched from the clause number [11] toward the first clause.
29) Since the dependency clause of clause number [11] is clause number [12], the independent word “Such” was found from clause number [11], and the independent word “system” of depth 2 was directly modified. Since it is an independent word, the depth is added, and a depth of 3, “notation of such”, a part-of-speech attribute “linkage”, and a dependency relation name “linkage modification” are stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
30) Further, a clause that directly modifies clause number [12] is searched from the clause number [10] toward the first clause through a parallel check. Since there is no clause that directly qualifies clause number [12] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “system” is terminated.

31) Next, a clause that directly modifies clause number [8] including “open” is searched from clause number [7] toward the first clause.
32) Since the destination clause of the clause number [7] is the clause number [8], the independent word “difference” is found from the clause number [7]. The independent word “difference” is an independent word that directly modifies the independent word “open” with depth 2. Therefore, the depth 3 is added by adding the depth, the notation “difference”, the part-of-speech attribute “noun” , Store the dependency relationship name “qualification for continuous use”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
33) Further, a clause that directly modifies clause number [8] is searched from clause number [6] toward the first clause, and after parallel check, the destination clause of clause number [1] is the clause number [8]. Therefore, the self-supporting word “Any” is found from the phrase number [1]. Since the independent word “Any” is an independent word that directly modifies the independent word “Open” with a depth of 2, add the depth to the depth of 3, the notation of “Any”, and the part of speech attribute “adverb”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
34) Further, a clause that directly qualifies clause number [8] is searched from clause number [0] toward the first clause, and through the parallel check, the first clause is searched to directly qualify clause number [8]. Since there is no clause, the search for the independent word that directly modifies the independent word “open” is terminated.

  35) Next, the phrase directly including the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  36) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 3 is extracted. The extracted independent words of depth 3 are “such”, “difference”, and “any”. First, a phrase that directly modifies the phrase number [11] including “these” is searched from the phrase number [10] toward the first phrase. Since there is no clause that directly qualifies the clause number [11] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “these” is terminated.

37) Next, a phrase that directly modifies the phrase number [7] including “difference” is searched from the phrase number [6] toward the first phrase.
38) Since the dependency clause of clause number [6] is clause number [7], the independent word “performance” is found from clause number [6]. Since the independent word “performance” is an independent word that directly modifies the independent word “difference” at depth 3, add the depth to the depth 4, the notation “performance”, and the part of speech attribute “noun”. , Store the dependency relationship name "no-complex modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
39) Further, a clause that directly modifies clause number [7] is searched from clause number [5] toward the first clause, and after parallel check, the destination clause of clause number [5] is the clause number [7]. Therefore, the self-supporting word “system” is found from the phrase number [5]. Since the independent word “system” is an independent word that directly modifies the independent word “difference” at the depth 3, the depth is added to the depth 4, the expression “system”, and the part of speech attribute “noun”. And, the dependency relation name “no-complex modification” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
40) If a clause that directly modifies clause number [7] is searched from clause number [5] toward the first clause and parallel check is performed, the dependency relation name of clause number [4] is “parallel”. Since the related clause number is the clause number of the corresponding clause [5], the independent word “system” of the immediately preceding clause [4] is also extracted. Therefore, the notation of depth 4 and “system”, the part-of-speech attribute “noun”, and the dependency relationship name store the dependency relationship name “no-complex modification” of the clause to be processed. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
41) Furthermore, the phrase directly modifying the phrase number [7] is searched from the phrase number [4] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies clause number [7], the search for the independent word that directly modifies the independent word “difference” is terminated.

  42) Next, the phrase directly modifying the phrase number [1] including “any” is searched from the phrase number [0] toward the first phrase, and the first phrase is searched. Since there is no clause that directly qualifies clause number [1], the search for the independent word that directly qualifies the independent word “any” is terminated.

  43) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 4 is extracted. The extracted independent words of depth 4 are “performance”, “system”, and “system”. First, a clause that directly modifies clause number [6] including “performance” is searched from clause number [5] toward the first clause. Since there is no clause that directly qualifies the clause number [6] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “performance” is terminated.

44) Next, a phrase that directly modifies the phrase number [5] including “system” is searched from the phrase number [4] toward the first phrase. Since the destination clause of clause number [4] is clause number [5], the self-supporting word “system” is found from clause number [4]. Since the independent word “system” is an independent word that directly modifies the independent word “system” with a depth of 4, add the depth, the depth of “5”, the notation of “system”, and the part of speech attribute “noun” And the dependency relationship name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
45) Further, a clause that directly modifies clause number [5] is searched from clause number [3] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [5], the search for the independent word that directly qualifies the independent word “system” is terminated.

46) Next, a phrase that directly modifies the phrase number [4] including “system” is searched from the phrase number [3] toward the first phrase. Since the destination clause of clause number [3] is clause number [4], the independent word “owned” is found from clause number [3]. The independent word “owned” is an independent word that directly modifies the independent word “system” with a depth of 4. Therefore, the depth is added, the depth of “5”, the notation of “owned”, and the part of speech attribute “sa change” Then, the dependency relation name “modification modification” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
47) Further, a clause that directly modifies clause number [4] is searched from clause number [2] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [4], the search for independent words that directly qualify the independent word “system” is terminated.

48) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 5 is extracted. The extracted independent words of depth 5 are “system” and “owned”. First, a clause that directly modifies clause number [4] including “system” is searched from clause number [3] toward the first clause. Since the destination clause of clause number [3] is clause number [4], the independent word “owned” is found from clause number [3]. The freestanding word “owned” is a freestanding word that directly modifies the freestanding word “system” with depth 5; , Store the dependency relationship name "modification modification". Moreover, when the intention expression check in the same phrase was performed, there was no target expression in the same phrase.
49) Further, a clause that directly modifies clause number [4] is searched from clause number [2] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [4], the search for independent words that directly qualify the independent word “system” is terminated.

50) Next, a phrase that directly modifies the phrase number [3] including “owned” is searched from the phrase number [2] toward the first phrase. Since the destination clause of clause number [2] is clause number [3], the independent word “present” is found from clause number [2]. Since the independent word “present” is an independent word that directly modifies the independent word “owned” at a depth of 5, the depth is added to the depth of 6, the expression “current”, and the part of speech attribute “adverb”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
51) Furthermore, a clause that directly qualifies clause number [3] is searched from clause number [1] toward the first clause, and through the parallel check, the first clause is searched, and clause number [3] is directly modified. Since there are no clauses, the search for independent words that directly qualify the independent word “owned” is terminated.

52) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 6 is extracted. The extracted independent words of depth 6 are “owned” and “present”. First, a clause that directly modifies clause number [3] including “owned” is searched from clause number [2] toward the first clause. Since the destination clause of clause number [2] is clause number [3], the independent word “present” is found from clause number [2]. Since the independent word “present” is an independent word that directly modifies the independent word “owned” at a depth of 6, the depth is added to the depth of 7, the notation of “current”, and the part of speech attribute “adverb”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
53) In addition, a clause that directly modifies clause number [3] is searched from clause number [1] toward the first clause, and through the parallel check, the first clause is searched, and clause number [3] is directly modified. Since there are no clauses, the search for independent words that directly qualify the independent word “owned” is terminated.

  54) Next, the phrase directly modifying the phrase number [2] including “present” is searched from the phrase number [1] toward the first phrase. Since there is no clause that directly qualifies clause number [2] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “present” is terminated.

  55) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 7 is extracted. The extracted free standing word of depth 7 is “present”. A phrase that directly modifies the phrase number [2] including “current” is searched from the phrase number [1] toward the first phrase. Since there is no clause that directly qualifies clause number [2] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “present” is terminated. The extraction of the partial structure element using the extraction key “say” as the extraction key is completed by the above extraction process. The result is as shown in FIG.

(Extraction of substructure elements using the predicate “Naru” as an extraction key)
56) Next, when a predicate is searched from the phrase number [14] toward the first phrase, the phrase number [14] is a predicative phrase, and therefore the predicate “is” is found from the phrase number [14]. .
57) As an extraction key, the notation “Naru” and the part of speech attribute “verb” are extracted and stored. In addition, an intention expression check is performed, but there is no intention expression in the same clause.

58) Next, from the phrase number [13], the independent words of the phrases in which the predicate “Naru” is directly modified are sequentially searched from the first phrase in order.
59) Since the dependency clause of the clause number [13] is the clause number [14], the independent word “purchase” is found from the clause number [13]. Since “Purchase” is an independent word that directly modifies the extraction key “Naru”, the depth 0, the notation of “Purchase”, the part-of-speech attribute “sa change”, and the dependency relation name “Continuous modification” are stored. . In addition, when the intention expression check in the same clause is performed, the intention “request” is registered in the extension element because the correspondence between the expression indicating the intention and the intention shown in FIG. To do.

60) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [13] toward the first phrase, and through a parallel check, the independent word “company” is found from the phrase number [10]. Since the independent word “company” is an independent word that directly modifies the extraction key “Naru”, the depth 0, the expression “company”, the part of speech attribute “noun”, and the dependency relation name “continuous modification” Remember. In addition, although the intention expression check in the same clause is performed, there is no intention expression in the same clause.
61) Furthermore, when a clause that directly qualifies clause number [14] is searched from clause number [10] toward the first clause and parallel check is performed, the dependency relation name of clause number [9] is “parallel”. It is. For this reason, when the dependency clause number is the clause number of the corresponding clause [10], the independent word “consumer” of the immediately preceding clause [9] is also extracted, and the depth 0 and “consumer” notation, The part-of-speech attribute “noun” and the dependency relation name are stored. The dependency relationship name is stored as the dependency relationship name “continuous modification” of the clause to be processed. In addition, the intention expression check in the same phrase is performed, but there is no intention expression in the same phrase.
62) Further, a phrase that directly modifies the phrase number [14] is searched from the phrase number [9] toward the first phrase, and through the parallel check, the independent word “Den” is found from the phrase number [0]. Since the independent word “Dai” is an independent word that directly modifies the extracted key “Naru”, the depth 0, the expression “Dai”, the part-of-speech attribute “connective”, and the dependency relation name “Continuous modification” Remember. Since the part of speech is a conjunction, and the depth of the corresponding independent word is 0, the reference to the dictionary in which the correspondence between the connection particle, the conjunction, and the meaning of the connection is described, and there is a match. Register the connection meaning “reverse connection” in the element. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
63) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [0] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [14], the search for the independent word that directly qualifies the independent word “becomes” ends.

  64) Next, extract the independent words of the phrase that are related to the extraction key and directly related to the extracted independent words of depth 0. Independent words regarding the extraction key “Naru” are “Purchase”, “Company”, “Consumer”, and “Even”. First, the phrase directly modifying the phrase number [13] including “purchase” is searched from the phrase number [12] toward the first phrase.

65) Since the dependency clause of the clause number [12] is the clause number [13], the independent word “system” is found from the clause number [12]. Since “system” is an independent word that directly modifies the independent word “purchase” with a depth of 0, the depth is added to the depth 1, the expression “system”, the part of speech attribute “noun”, Receiving relation name “wo case qualification” is stored. In addition, the intention expression check in the same phrase is performed, but there is no intention expression in the same phrase.
66) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [12] toward the first phrase, and through the parallel check, the independent word “open” is found from the phrase number [8]. Since it is a self-supporting word that directly modifies the self-supporting word “purchase” at depth 0, the depth is added to the depth 1, the expression “open”, the part-of-speech attribute “verb”, and the dependency relation name “ “Continuous modification” is stored. Further, when the presence or absence of a connection particle in a phrase including “open” is checked, since the connection particle “BA” matches the dictionary, the connection meaning “forward assumption” is registered in the extension element. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
67) Furthermore, the phrase directly modifying clause number [13] is searched from clause number [8] toward the first clause, and after parallel checking, the first clause is searched and clause number [13] is directly qualified. Since there is no clause to do, the search for free words that directly qualify the free word “purchase” is terminated.

68) Next, a phrase that directly modifies the phrase number [10] including “company” is searched from the phrase number [9] toward the first phrase. Since the destination clause of clause number [9] is clause number [10], the independent word “consumer” is found from clause number [9]. The independent word “consumer” is an independent word that directly modifies the independent word “enterprise” with a depth of 0. , The dependency relationship name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
69) Further, the phrase directly modifying the phrase number [10] is searched from the phrase number [9] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [10], the search for the independent word that directly qualifies the independent word “company” is terminated.

  70) Next, a phrase that directly modifies the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  71) Next, the phrase directly modifying the phrase number [0] including “Even” is searched toward the first phrase. Since clause number [0] is the first clause, there is no clause that directly qualifies clause number [0], so the search for independent words that directly qualify the independent word “even” is terminated.

72) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 1 is extracted. The extracted independent words of depth 1 are “system”, “open”, and “consumer”. First, a clause that directly modifies the clause number [12] including “system” is searched from the clause number [11] toward the first clause.
73) Since the destination clause of the clause number [11] is the clause number [12], the self-supporting word “such” is found from the clause number [11]. “Such” is a self-supporting word that directly modifies the self-supporting word “system” at depth 1, so add depth to depth 2, “notice” and the part-of-speech attribute “communicative” Receiving relationship name “modification modification” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
74) Further, the phrase directly modifying the phrase number [12] is searched from the phrase number [11] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [12], the search for independent words that directly qualify the independent word “system” is terminated.

75) Next, the phrase directly modifying the phrase number [8] including “open” is searched from the phrase number [7] toward the first phrase.
76) Since the destination clause of the clause number [7] is the clause number [8], the independent word “difference” is found from the clause number [7]. Since the independent word “difference” is an independent word that directly modifies the independent word “open” with depth 1, the depth is added by adding depth 2, the notation of “difference”, and the part of speech attribute “noun”. , Store the dependency relationship name “qualification for continuous use”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
77) Further, a clause that directly modifies clause number [8] is searched from clause number [7] toward the first clause, and through the parallel check, the first clause is searched. Since the destination clause of clause number [1] is clause number [8], the independent word “any” is found from clause number [1]. Since the independent word “Any” is an independent word that directly modifies the independent word “open” at the depth 1, the depth is added to the depth 2, the expression “any”, and the part of speech attribute “adverb”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
78) Further, a clause that directly modifies clause number [8] is searched from clause number [1] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [8], the search for the independent word that directly qualifies the independent word “open” is terminated.

  79) Next, the phrase directly modifying the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  80) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 2 is extracted. The extracted independent words of depth 2 are “such”, “difference”, and “any”. First, a phrase that directly modifies the phrase number [11] including “these” is searched from the phrase number [10] toward the first phrase. Since there is no clause that directly qualifies the clause number [11] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “these” is terminated.

81) Next, the phrase directly modifying the phrase number [7] including “difference” is searched from the phrase number [6] toward the first phrase.
82) Since the destination clause of clause number [6] is clause number [7], the independent word “performance” is found from clause number [6]. The independent word “Performance” is an independent word that directly modifies the independent word “Difference” at depth 2. Therefore, add the depth to the depth 3, “Performance”, and the part of speech attribute “Noun”. , Store the dependency relation name "no-complex modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
83) Further, the phrase directly modifying the phrase number [7] is searched from the phrase number [6] toward the first phrase, and through the parallel check, the first phrase is searched. Since the destination clause of clause number [5] is clause number [7], the independent word “system” is found from clause number [5]. Since the independent word “system” is an independent word that directly modifies the independent word “difference” of depth 2, the depth is added to the depth 3, the expression “system”, the part of speech attribute “noun” , Store the dependency relation name "no-complex modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
84) Further, the phrase directly modifying the phrase number [7] is searched from the phrase number [5] toward the first phrase, and the parallel check is performed. Since the dependency relation name of clause number [4] is “Parallel” and the destination clause number is the clause number of the corresponding clause [5], the independent word “system” of the immediately preceding clause [4] is also Extraction target. The notation of depth 3 and “system”, the part-of-speech attribute “noun”, and the dependency relation name are stored as the dependency relation name “no-complex modification” of the clause to be processed. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
85) Further, the phrase directly modifying the phrase number [7] is searched from the phrase number [4] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies clause number [7], the search for the independent word that directly modifies the independent word “difference” is terminated.

  86) Next, the phrase directly modifying the phrase number [1] including “any” is searched from the phrase number [0] toward the first phrase, and the first phrase is searched. Since there is no clause that directly qualifies clause number [1], the search for the independent word that directly qualifies the independent word “any” is terminated.

  87) Next, the independent words of the phrase having a dependency relationship with the extracted independent words of depth 3 are extracted. The extracted independent words of depth 3 are “performance”, “system”, and “system”. First, a clause that directly modifies clause number [6] including “performance” is searched from clause number [5] toward the first clause. Since there is no clause that directly qualifies the clause number [6] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “performance” is terminated.

88) Next, a phrase that directly modifies the phrase number [5] including “system” is searched from the phrase number [4] toward the first phrase. Since the destination clause of clause number [4] is clause number [5], the self-supporting word “system” is found from clause number [4]. Since the independent word “system” is an independent word that directly modifies the independent word “system” at depth 3, the depth is added to the depth 4, the expression “system”, and the part of speech attribute “noun”. , The dependency relationship name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
89) Further, the phrase directly modifying the phrase number [5] is searched from the phrase number [3] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [5], the search for the independent word that directly qualifies the independent word “system” is terminated.

90) Next, a phrase that directly modifies the phrase number [4] including “system” is searched from the phrase number [3] toward the first phrase. Since the destination clause of clause number [3] is clause number [4], the independent word “owned” is found from clause number [3]. The freestanding word “owned” is a freestanding word that directly modifies the freestanding word “system” at depth 3. Therefore, the depth is added to the depth 4 and the notation of “owned” and the part of speech attribute “sa change” , Store the dependency relation name “linkage modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
91) Further, a clause that directly modifies clause number [4] is searched from clause number [2] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [4], the search for independent words that directly qualify the independent word “system” is terminated.

92) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 4 is extracted. The extracted independent words of depth 4 are “system” and “owned”. First, a clause that directly modifies clause number [4] including “system” is searched from clause number [3] toward the first clause. Since the destination clause of clause number [3] is clause number [4], the independent word “owned” is found from clause number [3]. Since the independent word “owned” is an independent word that directly modifies the independent word “system” with a depth of 4, add the depth to the depth of 5 and the notation of “owned” and the part-of-speech attribute “sa change” , Store the dependency relation name “linkage modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
93) Further, a clause that directly modifies clause number [4] is searched from clause number [2] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [4], the search for independent words that directly qualify the independent word “system” is terminated.

94) Next, a clause that directly modifies clause number [3] including “owned” is searched from clause number [2] toward the first clause, and the destination clause of clause number [2] is clause number [3]. ], The self-supporting word “present” is found from the phrase number [2]. Since the independent word “present” is an independent word that directly modifies the independent word “owned” at a depth of 4, the depth is added to the depth of 5, the notation of “current”, and the part of speech attribute “adverb”. , Store the dependency relation name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
95) Further, a clause that directly modifies clause number [3] is searched from clause number [1] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [3], the search for the independent word that directly qualifies the independent word “owned” is terminated.

96) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 5 is extracted. The extracted independent words of depth 5 are “owned” and “present”. First, a clause that directly modifies clause number [3] including “owned” is searched from clause number [2] toward the first clause. Since the destination clause of clause number [2] is clause number [3], the independent word “present” is found from clause number [2]. Since the independent word “present” is an independent word that directly modifies the independent word “owned” at a depth of 5, the depth is added to the depth of 6, the expression “current”, and the part of speech attribute “adverb”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
97) Furthermore, the phrase directly modifying the phrase number [3] is searched from the phrase number [1] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [3], the search for the independent word that directly qualifies the independent word “owned” is terminated.

  98) Next, the phrase directly modifying the phrase number [2] including “current” is searched from the phrase number [1] toward the first phrase. Since there is no clause that directly qualifies clause number [2] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “present” is terminated.

  99) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 6 is extracted. The extracted free standing word of depth 6 is “present”. A phrase that directly modifies the phrase number [2] including “current” is searched from the phrase number [1] toward the first phrase. Since there is no clause that directly qualifies clause number [2] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “present” is terminated. With the above extraction process, the partial structure element extension extraction process using the extraction key “Naru” as the extraction key is completed. The result is as shown in FIG.

100) Next, the next extraction key is selected to perform the partial structure element extension extraction process. When the phrase is searched from the phrase number [13] toward the first phrase, the phrase “Purchase” is found from the phrase number [13] because the phrase number [13] is a phrase phrase.
101) As the extraction key, the notation of “purchase” and the part-of-speech attribute “sa change” are extracted and stored, and when the intention expression check in the same phrase is performed, the correspondence between the expression indicating the intention in FIG. 5B and the intention is described. The intention “request” is registered in the extension element because it matches the auxiliary verb “tai” in the dictionary.

  In accordance with the above procedure, search for a predicate from the last sentence of the linguistic information acquisition result of the example sentence to the first sentence, and the independent word that is in a dependency relationship with the predicate key, and further, the independent word and the dependency Independent words in are extracted with a hierarchy of depth, and extension elements are also registered. When the search for the predicate key and the extraction of the related independent words and the extended elements are completed up to the first phrase, the extended extraction of the partial structure element with the prescriptive key in the example sentence is ended. The subsequent processing is omitted.

(Extraction of substructure elements using the word “Koto” as an extraction key)
FIG. 5B is a flowchart showing the flow of processing in the case of extracting an extended partial structural element using the body language as an extraction key. The flowchart of FIG. 5B shows the processing policy of the following 102) to 154). In the flowchart of FIG. 5B, the description of the same processing as that of FIG. The process of FIG. 5B is different from that of FIG. 5A in that a description is extracted as an extraction key and whether or not a phrase having the extraction key is the end of a sentence. In the case of the end of a sentence, “end of sentence” is stored in the extension element, thereby enabling processing using the end of sentence syntactic vocabulary effectively.

102) First, a sentence clause is retrieved from the last sentence of the language analysis processing result for one sentence from the same head.
103) “Koto” is found from the phrase number [16].
104) The notation of “Koto” and the part-of-speech attribute “formal noun” are stored, and since the phrase is the sentence end sentence, “end of sentence” is registered in the extension element. Furthermore, when the intention expression check is performed in the same phrase, there is no target expression in the same phrase.

105) Since the part-of-speech attribute of “Koto” is a noun class and can be regarded as a dialect positioned at the end of the dialect phrase, then the phrase “Koto” is directly linked and modified in order from the phrase number [15] toward the first phrase. Search for free-standing words in the current phrase.
106) Since the destination clause number of the clause number [15] is [16], the phrase of the clause number [15] finds “say” that directly modifies “that”. Since “Take” directly employs the extraction key “Koto”, it stores the notation of depth 0, “Take”, the part-of-speech attribute “verb”, and the dependency relationship “multiple modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
107) Further, from the clause number [15], the independent words of the clauses in which the body phrase “Koto” is directly modified in order from the clause number [15] are searched through the parallel check. In this embodiment, there is no clause that directly modifies clause [15] with “parallel”.
108) Search up to the first phrase, and since there is no independent word of the phrase that directly modifies “koto”, the extraction of the partial structural element with the synonym “koto” as the extraction key is finished.

  109) Next, the independent words of the phrase having a dependency relationship directly with the independent words of depth 0 extracted by directly modifying the extraction key are extracted. An independent word for the extraction key “Koto” is “say”. First, the phrase directly modifying the phrase number [15] including “say” is searched from the phrase number [14] toward the first phrase.

110) Since the destination clause of the clause number [14] is the clause number [15], the independent word “becomes” is found from the clause number [14]. The independent word “Naru” is an independent word that directly modifies the independent word “say” with a depth of 0. Therefore, the depth is added to the depth 1, the expression “Naru”, and the part of speech attribute “verb”. , Store the dependency relationship name “Gift qualification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
111) Further, the phrase directly modifying the phrase number [15] is searched from the phrase number [14] toward the first phrase. At that time, 107) the same parallel check is performed. There is no clause that directly modifies clause [14] with “parallel”.
112) Search up to the first phrase, and since there is no independent word of the phrase that directly modifies “say”, the extraction of the partial structural element using the independent word “say” as the extraction key is finished.

113) Next, an independent word of a phrase having a dependency relationship directly with an independent word having a depth of 1 is extracted. A self-supporting word of depth 1 is “become”. First, a phrase that directly modifies the phrase number [14] including “Naru” is searched from the phrase number [13] toward the first phrase.
114) Since the destination clause of the clause number [13] is the clause number [14], the independent word “purchase” is found from the clause number [13]. The independent word “Purchase” is an independent word that directly modifies the independent word “Naru” with a depth of 1. Therefore, add the depth to the depth 2, “Purchase”, and the part-of-speech attribute “sa change”. , Store the dependency relation name “continuous modification”. In addition, when the intention expression check in the same clause is performed, the intention “request” is added to the extension element because it matches the auxiliary word “tai” in the dictionary in which the correspondence between the expression indicating the intention and the intention shown in FIG. 5B is described. sign up.

115) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [13] toward the first phrase, and through the parallel check, the first phrase is searched. Discover the self-supporting word “company” from clause number [10]. The independent word “company” is an independent word that directly modifies the independent word “become” with depth 1. Therefore, the depth 2 is added by adding the depth, the expression “company”, and the part-of-speech attribute “noun”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
116) Furthermore, when a clause that directly modifies clause number [14] is searched from clause number [9] toward the first clause and parallel check is performed, the dependency relation name of clause number [9] is “parallel”. It is. Since the related clause number is the clause number of the corresponding clause [10], the independent word “consumer” in the immediately preceding clause [9] is also extracted, the depth 2 and “consumer” notation, and the part of speech attribute “ "Noun" and the dependency relationship are memorized. The dependency relationship name is stored as the dependency relationship name “continuous modification” of the clause to be processed. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
117) Further, a phrase that directly modifies the phrase number [14] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the independent word “even” is found from the phrase number [0]. Since the independent word “but” is an independent word that directly modifies the independent word “becomes” with a depth of 1, add the depth to the depth 2, the notation of “even”, and the part-of-speech attribute “connective”. , Store the dependency relationship name “continuous modification”. Even if the part-of-speech is a conjunctive, the depth of the corresponding independent word is not 0. Therefore, the dictionary in which the correspondence between the connecting particle, the conjunctive, and the meaning of the connection is described cannot be referenced. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
118) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [0] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [14], the search for the independent word that directly qualifies the independent word “becomes” ends.

119) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 2 is extracted. The independent words of depth 2 are “Purchase”, “Company”, “Consumer”, and “Even”. First, the phrase directly modifying the phrase number [13] including “purchase” is searched from the phrase number [12] toward the first phrase.
120) Since the relation clause of the clause number [12] is the clause number [13], the independent word “system” is found from the clause number [12]. Since the independent word “system” is an independent word that directly modifies the independent word “purchase” at depth 2, add depth to depth 3, “system”, and part-of-speech attribute “noun” , Store the dependency relation name "wo case qualification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
121) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [11] toward the first phrase, and through the parallel check, the independent word “open” is found from the phrase number [8]. Since the independent word “open” is an independent word that directly modifies the independent word “purchase” at depth 2, add the depth to the depth 3, the expression “open”, and the part of speech attribute “verb” , Store the dependency relationship name “continuous modification”. Further, when the presence or absence of a connection particle in a phrase including “open” is checked, since the connection particle “BA” matches the dictionary, the connection meaning “forward assumption” is registered in the extension element. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
122) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [7] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [13], the search for the independent word that directly modifies the independent word “purchase” is terminated.

123) Next, a phrase that directly modifies the phrase number [10] including “company” is searched from the phrase number [9] toward the first phrase. Since the destination clause of clause number [9] is clause number [10], the independent word “consumer” is found from clause number [9]. The independent word “consumer” is an independent word that directly modifies the independent word “enterprise” of depth 2. Therefore, add the depth to the depth 3, “consumer”, and the part-of-speech attribute “noun”. , The dependency relationship name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
124) Further, the phrase directly modifying the phrase number [10] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [10], the search for the independent word that directly qualifies the independent word “company” is terminated.

  125) Next, a phrase that directly qualifies the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  126) Next, the phrase directly modifying the phrase number [0] including “Even” is searched toward the first phrase. Since clause number [0] is the first clause, there is no clause that directly qualifies clause number [0], so the search for independent words that directly qualify the independent word “even” is terminated.

127) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 3 is extracted. The extracted free words of depth 3 are “system”, “open”, and “consumer”. First, a clause that directly modifies the clause number [12] including “system” is searched from the clause number [11] toward the first clause.
128) Since the destination clause of the clause number [11] is the clause number [12], the self-supporting word “such” is found from the clause number [11]. The independent word “Such” is a self-supporting word that directly modifies the independent word “system” with a depth of 3. , Store the dependency relation name “linkage modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.

  129) Further, the phrase directly modifying the phrase number [12] is searched from the phrase number [10] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [12], the search for the independent word that directly qualifies the independent word “system” is terminated.

130) Next, the phrase directly modifying the phrase number [8] including “open” is searched from the phrase number [7] toward the first phrase.
131) Since the destination clause of the clause number [7] is the clause number [8], the independent word “difference” is found from the clause number [7]. The independent word “difference” is an independent word that directly modifies the independent word “open” with a depth of 3. Therefore, the depth is added by adding the depth of 4, the notation of “difference”, and the part of speech attribute “noun”. , Store the dependency relationship name “qualification for continuous use”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
132) Further, the phrase directly modifying the phrase number [8] is searched from the phrase number [6] toward the first phrase, and through the parallel check, the first phrase is searched. Since the destination clause of clause number [1] is clause number [8], the independent word “any” is found from clause number [1]. Since the independent word “any” is an independent word that directly modifies the independent word “open” at the depth 3, the depth is added to the depth 4, the expression “any”, and the part of speech attribute “adverb”. , Store the dependency relation name “continuous modification”. Also. When the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
133) Further, the phrase directly modifying the phrase number [8] is searched from the phrase number [0] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [8], the search for the independent word that directly qualifies the independent word “open” is terminated.

  134) Next, the phrase directly modifying the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  135) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 4 is extracted. The extracted independent words of depth 4 are “such”, “difference”, and “any”. First, a phrase that directly modifies the phrase number [11] including “these” is searched from the phrase number [10] toward the first phrase. Since there is no clause that directly qualifies the clause number [11] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “these” is terminated.

136) Next, the phrase directly modifying the phrase number [7] including “difference” is searched from the phrase number [6] toward the first phrase.
137) Since the destination clause of clause number [6] is clause number [7], the independent word “performance” is found from clause number [6]. The independent word “Performance” is an independent word that directly modifies the independent word “Difference” with a depth of 4. Therefore, the depth is added to the depth of 5. , Store the dependency relationship name "no-complex modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
138) Further, the phrase directly modifying the phrase number [7] is searched from the phrase number [5] toward the first phrase, and through the parallel check, the first phrase is searched. Since the destination clause of clause number [5] is clause number [7], the independent word “system” is found from clause number [5]. Since the independent word “system” is an independent word that directly modifies the independent word “difference” with a depth of 4, add the depth to the depth of 5 and the notation of “system” and the part of speech attribute “noun”. , Store the dependency relationship name "no-complex modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
139) Further, the phrase directly modifying the phrase number [7] is searched from the phrase number [5] toward the first phrase, and the parallel check is performed. Since the dependency relation name of clause number [4] is “Parallel” and the destination clause number is the clause number of the corresponding clause [5], the independent word “system” of the immediately preceding clause [4] is also Extraction target. The depth 5, the description of “system”, the part-of-speech attribute “noun”, and the dependency relation name are stored as the dependency relation name “no-complex modification” of the clause to be processed. Moreover, when the intention expression check in the clause is performed, there is no target expression in the clause.
140) Further, the phrase directly modifying the phrase number [7] is searched from the phrase number [4] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies clause number [7], the search for the independent word that directly modifies the independent word “difference” is terminated.

  141) Next, the phrase directly modifying the phrase number [1] including “any” is searched from the phrase number [0] toward the first phrase, and the first phrase is searched. Since there is no clause that directly qualifies clause number [1], the search for the independent word that directly qualifies the independent word “any” is terminated.

  142) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 5 is extracted. The extracted independent words of depth 5 are “performance”, “system”, and “system”. First, a clause that directly modifies clause number [6] including “performance” is searched from clause number [5] toward the first clause. Since there is no clause that directly qualifies the clause number [6] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “performance” is terminated.

143) Next, the phrase directly modifying the phrase number [5] including “system” is searched from the phrase number [4] toward the first phrase. Since the destination clause of clause number [4] is clause number [5], the self-supporting word “system” is found from clause number [4]. Since the independent word “system” is an independent word that directly modifies the independent word “system” with a depth of 5, the depth is added to a depth of 6, a description of “system”, and a part-of-speech attribute “noun”. , The dependency relation name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
144) Further, the phrase directly modifying the phrase number [5] is searched from the phrase number [3] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [5], the search for the independent word that directly qualifies the independent word “system” is terminated.

145) Next, a phrase that directly modifies the phrase number [4] including “system” is searched from the phrase number [3] toward the first phrase. Since the destination clause of clause number [3] is clause number [4], the independent word “owned” is found from clause number [3]. The freestanding word “owned” is a freestanding word that directly modifies the freestanding word “system” with depth 5; , Store the dependency relationship name "modification modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
146) Further, the phrase directly modifying the phrase number [4] is searched from the phrase number [2] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [4], the search for independent words that directly qualify the independent word “system” is terminated.

147) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 6 is extracted. The extracted independent words of depth 6 are “system” and “owned”. First, a clause that directly modifies clause number [4] including “system” is searched from clause number [3] toward the first clause. Since the destination clause of clause number [3] is clause number [4], the independent word “owned” is found from clause number [3]. Since the independent word “owned” is an independent word that directly modifies the independent word “system” with a depth of 6, the depth is added to the depth of 7, the notation of “owned”, the part-of-speech attribute “sa change” , Store the dependency relation name “linkage modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
148) Further, the phrase directly modifying the phrase number [4] is searched from the phrase number [2] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [4], the search for independent words that directly qualify the independent word “system” is terminated.

149) Next, the phrase directly modifying the phrase number [3] including “owned” is searched from the phrase number [2] toward the first phrase. Since the destination clause of clause number [2] is clause number [3], the independent word “present” is found from clause number [2]. Since the independent word “present” is an independent word that directly modifies the independent word “owned” at a depth of 6, the depth is added to the depth of 7, the notation of “current”, and the part of speech attribute “adverb”. , Store the dependency relationship name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
150) Further, the phrase directly modifying the phrase number [3] is searched from the phrase number [1] toward the first phrase, and after the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [3], the search for the independent word that directly qualifies the independent word “owned” is terminated.

151) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 7 is extracted. The extracted independent words of depth 7 are “owned” and “present”. First, a clause that directly modifies clause number [3] including “owned” is searched from clause number [2] toward the first clause. Since the destination clause of clause number [2] is clause number [3], the independent word “present” is found from clause number [2]. Since the independent word “present” is an independent word that directly modifies the independent word “owned” at a depth of 7, the depth is added to the depth of 8, the notation of “current”, and the part of speech attribute “adverb”. , Store the dependency relation name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
152) Further, the phrase directly modifying the phrase number [3] is searched from the phrase number [1] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [3], the search for the independent word that directly qualifies the independent word “owned” is terminated.

  153) Next, the phrase directly modifying the phrase number [2] including “current” is searched from the phrase number [1] toward the first phrase, and the first phrase is searched. Since there is no clause that directly qualifies clause number [2], the search for the independent word that directly qualifies the independent word “present” is terminated.

  154) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 8 is extracted. The extracted free standing word of depth 8 is “present”. A phrase that directly modifies the phrase number [2] including “current” is searched from the phrase number [1] toward the first phrase. Since there is no clause that directly qualifies clause number [2] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “present” is terminated. With the above extraction process, the extraction of the partial structure element using the extraction key “ko” as the extraction key is completed. The result is as shown in FIG.

  In accordance with the above procedure, we search for the syntactic phrase from the last phrase of the linguistic information acquisition result of the example sentence toward the first phrase, the part of speech attribute of the independent word of that phrase is a noun class, If it can be regarded as the last term, it is extracted as an extraction key, and the independent words that are in the dependency relationship, and the independent words that are in the dependency relationship with the independent words are extracted with a depth hierarchy, and the extended elements are also registered. I will do it. When the extraction key search and related independent words and extension elements are extracted up to the first phrase, the extension extraction of the partial structure element with the body key in the example sentence is ended.

  155) If there is a language information acquisition result of the next sentence in the stored language information acquisition result, return to 1) until there is no stored language information acquisition result for the language information acquisition result of the next sentence , Partial structure element extension extraction is performed for each sentence. If there is no language information acquisition result for the next sentence, the substructure element extension extraction process is terminated, and the process proceeds to the substructure data generation process.

[Substructure element extension extraction process when user specifies depth information (precaution)]
Next, a description will be given of a process in the case where the substructure element extension extraction process when the depth n = 2 is specified by the user up to the input depth n.
FIG. 7 is a flowchart showing a flow of processing for extracting an extended partial structural element when a depth is designated by the user. In FIG. 7, the word is used as an extraction key.

  The flowchart of FIG. 7 shows the processing policy of the following 156) to 211). In the flowchart of FIG. 7, the description of the same processing as in FIG. The process of FIG. 7 has n input setting steps (S301) for designating the depth, and when the count constant k becomes n or less, the substructure element expansion extraction process with a depth greater than that is not performed. This is different from FIG. 6A in the point (S302). That is, when the depth specified by the user is reached, the processing using the next predicate as the extraction key is performed without detecting a dependency relationship of a depth greater than that. Therefore, the entire analysis time can be shortened and efficient processing can be realized. The result of the partial structure element extension extraction process in the case of n = 2 is as shown in FIG.

  156) A sentence with language information for one sentence from the beginning as shown in FIG. 2 is extracted from the language information storage means, and a predicate is searched from the last sentence. The last clause is the clause with clause number [16].

  157) Since the phrase number [16] is an episodic phrase, the next examination of the phrase number [15] finds the phrase “say” from the phrase number [15] because it is a phrase phrase .

158) As the extraction key, the notation “say” and the part-of-speech attribute “verb” are extracted and stored. (There is no intention expression in the same clause).
159) Next, from the phrase number [15], the independent words of the phrase in which the phrase “say” is directly modified are sequentially searched from the first phrase toward the first phrase.
160) Since the destination clause number of the clause number [14] is [15], the phrase of the clause number [14] finds “Naru” in which “say” is directly modified. Since “Naru” is an independent word that directly modifies the extracted key, the depth 0, the notation “Naru”, the part-of-speech attribute “verb”, and the dependency relation name “Glyph combination modification” are stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
161) Further, from the phrase number [14], the independent words of the phrases in which the predicate “say” is directly modified are sequentially searched in order from the phrase number [14], and a parallel check is performed. When the parallel check is performed, it is found that there is no clause that is directly qualified with “parallel” in clause number [14], and the search to the first clause is advanced.

162) Search from phrase number [13] to head phrase number [0]. Since there is no clause whose dependency clause number is [15], the next step is to search for independent words directly modified by “say” from clause number [16] toward the end clause.
163) Since the destination clause number of the clause number [15] is [16], it can be seen that the clause of the clause number [15] directly modifies the “number” of the clause number [16]. . Since “Koto” is an independent word directly modified by the extraction key, the depth 0, the notation of “Koto”, the part-of-speech attribute “formal noun”, and the dependency relationship “combined modification” are stored. At this time, a flag indicating a difference from the independent word of the previous consecutive modification relationship is set. In FIG. 8, flags are represented by asterisks. Furthermore, the intention expression in the same phrase is checked, and it can be seen that there is no intention expression in the same phrase.

  164) Further, from the phrase number [16], the independent words of the phrases in which the predicate “say” is directly linked and modified from the phrase number [16] are searched in order. A parallel check is performed on clause number [16], and it can be seen that there is no further independent word to be extracted because the dependency relation name is not “parallel” but a sentence end clause.

  165) Due to the universal restriction of Japanese grammar “There is only one clause destination”, there is already a self-supporting word that is directly modified with “say”. The extraction of the partial structural element having the depth 0 is finished.

166) Next, an independent word of a phrase having a dependency relationship with the extracted independent word that is related to the extraction key is extracted. Independent words for the extraction key “say” are “become” and “ko”. First, a phrase that directly modifies the phrase number [14] including “Naru” is searched from the phrase number [13] toward the first phrase.
167) Since the dependency clause of the clause number [13] is the clause number [14], the independent word “purchase” is found from the clause number [13]. “Purchase” is an independent word that directly modifies the independent word “Naru” with a depth of 0. Receiving relationship name “continuous modification” is stored. When the intention expression check in the same phrase is performed, the intention “request” is registered in the extension element because the correspondence between the expression indicating the intention and the intention shown in FIG. .
168) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [13] toward the first phrase, and through the parallel check, the first phrase is searched. Discover the self-supporting word “company” from clause number [10]. Because it is an independent word that directly modifies the independent word “Naru” with a depth of 0, the depth is added, and the depth 1, the notation of “company”, the part of speech attribute “noun”, and the dependency relation name “ “Continuous modification” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
169) Further, when a clause that directly modifies clause number [14] is searched from clause number [9] toward the first clause and parallel check is performed, the dependency relation name of clause number [9] is “parallel”. It is. Since the related clause number is the clause number of the corresponding clause [10], the independent word “consumer” of the immediately preceding clause [9] is also extracted, the depth 1 and “consumer” notation, and the part of speech attribute “ "Noun" and the dependency relationship are memorized. The dependency relationship name is stored as the dependency relationship name “continuous modification” of the clause to be processed. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.

170) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the self-supporting word “even” is found from the phrase number [0]. “But” is a free-standing word that directly modifies the free-standing word “Naru” with a depth of 0, so the depth is added by adding the depth of 1 and the notation of “but” and the part-of-speech attribute “connective”. Receiving relationship name “continuous modification” is stored. Even if the part-of-speech is a conjunctive, the depth of the corresponding independent word is not 0. Therefore, the dictionary in which the correspondence between the connecting particle, the conjunctive, and the meaning of the connection is described cannot be referenced. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
171) Further, a clause that directly modifies clause number [14] is searched from clause number [0] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly modifies the clause number [14], the search for the independent word that directly qualifies the independent word “becomes” ends.

172) Next, the phrase directly modifying the phrase number [16] including the independent word “koto” is searched from the phrase number [15] toward the first phrase. At this time, the phrase number [15] including the extraction key “say” is not searched.
173) Since the first phrase is searched and there is no phrase that directly modifies the phrase number [16], the retrieval of the independent word that directly modifies the independent word “Koto” is terminated.

174) Next, the phrase directly modified by the phrase number [16] including the independent word “Koto” is searched for toward the sentence end phrase.
175) Since clause number [16] is the end-of-sentence clause, there is no clause directly modified by clause number [16] containing the independent word “koto”. To do.

176) Next, an independent word of a phrase having a dependency relationship with the extracted independent word of depth 1 is extracted. The extracted independent words of depth 1 are “Purchase”, “Company”, “Consumer”, and “Even”. First, the phrase directly modifying the phrase number [13] including “purchase” is searched from the phrase number [12] toward the first phrase.
177) Since the dependency clause of the clause number [12] is the clause number [13], the independent word “system” is found from the clause number [12]. Since the independent word “system” is an independent word that directly modifies the independent word “purchase” at depth 1, add the depth to the depth 2, the notation of “system”, the part-of-speech attribute “noun” , Store the dependency relation name "wo case qualification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
178) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [11] toward the first phrase, and through the parallel check, the independent word “open” is found from the phrase number [8]. Since the independent word “open” is an independent word that directly modifies the independent word “purchase” at depth 1, add the depth to the depth 2, the expression “open”, and the part of speech attribute “verb” , Store the dependency relation name “continuous modification”. Further, when the presence or absence of a connection particle in a phrase including “open” is checked, since the connection particle “BA” matches the dictionary, the connection meaning “forward assumption” is registered in the extension element. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
179) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [7] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [13], the search for the independent word that directly modifies the independent word “purchase” is terminated.

180) Next, a phrase that directly modifies the phrase number [10] including “company” is searched from the phrase number [9] toward the first phrase. Since the destination clause of clause number [9] is clause number [10], the independent word “consumer” is found from clause number [9]. The independent word “consumer” is an independent word that directly modifies the independent word “enterprise” of depth 1. Therefore, add the depth to the depth 2, “consumer”, and the part of speech attribute “noun”. , The dependency relation name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
181) Further, the phrase directly modifying the phrase number [10] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [10], the search for the independent word that directly qualifies the independent word “company” is terminated.

  182) Next, a phrase that directly modifies the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  183) Next, the phrase directly modifying the phrase number [0] including “Even” is searched toward the first phrase. Since clause number [0] is the first clause, there is no clause that directly qualifies clause number [0], so the search for independent words that directly qualify the independent word “even” is terminated.

  184) Here, since the substructure element expansion extraction process is stopped up to the input depth 2, the extraction of independent words that modify the extracted independent words “system”, “open”, and “consumer” at the depth 2 is not performed. In the above extraction process, the partial structure element extension extraction process using “say” as the extraction key is completed. The result is as shown in FIG. In FIG. 8A, independent words up to depth information 2 are extracted.

185) Next, when a predicate is searched from the phrase number [14] toward the first phrase, the phrase number [14] is a predicative phrase, and therefore the predicate “is” is found from the phrase number [14]. .
186) “Naru” notation and part-of-speech attribute “verb” are extracted and stored as an extraction key, and an intention expression check is performed, but the phrase does not have an intention expression.

187) Next, from the phrase number [13], the independent words of the phrase in which the predicate “Naru” is directly modified are sequentially searched from the first phrase toward the first phrase.
188) Since the dependency clause of the clause number [13] is the clause number [14], the independent word “purchase” is found from the clause number [13]. Since the independent word “purchase” is an independent word that directly modifies the extraction key “Naru”, the depth 0, the “purchase” notation, the part-of-speech attribute “sa change”, and the dependency relationship name “continuous modification” Remember. In addition, when the intention expression check in the same clause is performed, the intention “request” is registered in the extension element because the correspondence between the expression indicating the intention and the intention shown in FIG. To do.
189) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [13] toward the first phrase, and through the parallel check, the independent word “company” is found from the phrase number [10]. Since the independent word “company” is an independent word that directly modifies the extraction key “Naru”, the depth 0, the expression “company”, the part of speech attribute “noun”, and the dependency relation name “continuous modification” Remember. In addition, although the intention expression check in the same clause is performed, there is no intention expression in the same clause.
190) Further, a clause that directly modifies clause number [14] is searched from clause number [10] toward the first clause, and a parallel check is performed. Since the dependency relation name of clause number [9] is “Parallel” and the destination clause number is the clause number of the corresponding clause [10], the independent word “consumer” of the immediately preceding clause [9] is also As an extraction target, the notation of depth 0, “consumer”, the part of speech attribute “noun”, and the dependency relationship are stored. The dependency relationship name is stored as the dependency relationship name “continuous modification” of the clause to be processed. In addition, the intention expression check in the same phrase is performed, but there is no intention expression in the same phrase.
191) Further, a phrase that directly modifies the phrase number [14] is searched from the phrase number [9] toward the first phrase, and through the parallel check, the independent word “Den” is found from the phrase number [0]. Since the independent word “Dai” is an independent word that directly modifies the extracted key “Naru”, the depth 0, the expression “Dai”, the part-of-speech attribute “connective”, and the dependency relation name “Continuous modification” Remember. Since the part of speech is a conjunction, and the depth of the corresponding independent word is 0, the reference to the dictionary in which the correspondence between the connection particle, the conjunction, and the meaning of the connection is described, and there is a match. Register the connection meaning “reverse connection” in the element. Moreover, when the intention expression check in the same phrase was performed, there was no target expression in the same phrase.
192) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [0] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [14], the search for the independent word that directly qualifies the independent word “becomes” ends.

193) Next, the independent words of the phrase that are directly related to the extracted independent words of depth 0 and related to the extraction key are extracted. Independent words regarding the extraction key “Naru” are “Purchase”, “Company”, “Consumer”, and “Even”. First, the phrase directly modifying the phrase number [13] including “purchase” is searched from the phrase number [12] toward the first phrase.
194) Since the dependency clause of the clause number [12] is the clause number [13], the independent word “system” is found from the clause number [12]. Since the independent word “system” is an independent word that directly modifies the independent word “purchase” with a depth of 0, the depth is added to the depth 1, the notation of “system”, the part-of-speech attribute “noun” , Store the dependency relation name "wo case qualification". In addition, the intention expression check in the same phrase is performed, but there is no intention expression in the same phrase.
195) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [12] toward the first phrase, and through the parallel check, the independent word “open” is found from the phrase number [8]. Since the independent word “open” is an independent word that directly modifies the independent word “purchase” with a depth of 0, the depth is added to the depth 1, the expression “open”, and the part of speech attribute “verb”. , Store the dependency relationship name “continuous modification”. Further, when the presence or absence of a connection particle in a phrase including “open” is checked, since the connection particle “BA” matches the dictionary, the connection meaning “forward assumption” is registered in the extension element. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
196) Further, the phrase directly modifying the phrase number [13] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [13], the search for the independent word that directly modifies the independent word “purchase” is terminated.

197) Next, a phrase that directly modifies the phrase number [10] including “company” is searched from the phrase number [9] toward the first phrase. Since the destination clause of clause number [9] is clause number [10], the independent word “consumer” is found from clause number [9]. The independent word “consumer” is an independent word that directly modifies the independent word “enterprise” with a depth of 0. , The dependency relationship name “parallel” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
198) Further, the phrase directly modifying the phrase number [10] is searched from the phrase number [9] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [10], the search for the independent word that directly qualifies the independent word “company” is terminated.

  199) Next, a phrase that directly qualifies the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  200) Next, the phrase directly modifying the phrase number [0] including “Even” is searched toward the first phrase. Since clause number [0] is the first clause, there is no clause that directly qualifies clause number [0], so the search for independent words that directly qualify the independent word “even” is terminated.

201) Next, an independent word of a phrase having a dependency relationship with the extracted independent word having a depth of 1 is extracted. The extracted independent words of depth 1 are “system”, “open”, and “consumer”. First, a clause that directly modifies the clause number [12] including “system” is searched from the clause number [11] toward the first clause.
202) Since the relation clause of the clause number [11] is the clause number [12], the self-supporting word “such” is found from the clause number [11]. The freestanding word “Such” is a freestanding word that directly modifies the freestanding word “system” with a depth of 1. Therefore, the depth is added by adding the depth of 2. , Store the dependency relationship name "modification modification". Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
203) The phrase directly modifying the phrase number [12] is searched from the phrase number [11] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly qualifies clause number [12], the search for independent words that directly qualify the independent word “system” is terminated.

204) Next, the phrase directly modifying the phrase number [8] including “open” is searched from the phrase number [7] toward the first phrase.
205) Since the destination clause of the clause number [7] is the clause number [8], the independent word “difference” is found from the clause number [7]. Since the independent word “difference” is an independent word that directly modifies the independent word “open” with depth 1, add the depth, the depth 2, the difference, and the part-of-speech attribute “noun” And the dependency relation name “qualification for continuous use” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
206) Further, a clause directly modifying clause number [8] is searched from clause number [7] toward the first clause, and after parallel check, the destination clause of clause number [1] is the clause number [8]. Therefore, the self-supporting word “Any” is found from the phrase number [1]. Since the independent word “any” is an independent word that directly modifies the independent word “open” at the depth 1, the depth is added to the depth 2, the expression “any”, and the part of speech attribute “adverb”. , Record the dependency relation name “continuous modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
207) Further, a clause that directly modifies clause number [8] is searched from clause number [1] toward the first clause, and through the parallel check, the first clause is searched. Since there is no clause that directly qualifies clause number [8], the search for the independent word that directly qualifies the independent word “open” is terminated.

  208) Next, the phrase directly modifying the phrase number [9] including “consumer” is searched from the phrase number [8] toward the first phrase. Since there is no clause that directly qualifies clause number [9] after searching up to the first clause, the search for the independent word that directly qualifies the independent word “consumer” is terminated.

  209) Here, since the substructure element expansion extraction process is stopped up to the input depth 2, the extraction independent words of “depth 2” “these” “difference” “any” are not extracted. Thus, the extraction of the partial structure element using “Naru” as the extraction key is completed by the above extraction process. The result is as shown in FIG.

210) Further, when the phrase is searched from the phrase number [13] toward the first phrase, the phrase “Purchase” is found from the phrase number [13] because the phrase number [13] is a phrase phrase.
211) When the notation of “purchase” and the part-of-speech attribute “sa change” are extracted and stored as the extraction key, and the intention expression check in the same phrase is performed, the correspondence between the expression indicating the intention in FIG. 5B and the intention is described. The intention “request” is registered in the extension element because it matches the auxiliary verb “tai” in the dictionary.

  In accordance with the above procedure, search for a predicate from the last sentence of the linguistic information acquisition result of the example sentence to the first sentence, and use the independent words that are in a dependency relationship with the extraction key, and also the independent words and dependencies. A self-supporting word is extracted up to a user-specified depth, and an extension element is registered. When the search for the predicate key and the extraction of the related independent words and the extension elements are completed up to the first clause, the partial structure element extension extraction process using the prey as the extraction key is ended.

[Extraction extraction process for substructure element when depth is specified by the user (syntax)]
212) Search for the phrase of the semantic sentence from the last sentence of the language analysis processing result for one sentence from the same head.
213) Find “Koto” from clause number [16].
214) The notation of “Koto” and the part-of-speech attribute “formal noun” are stored, and since the phrase is a sentence end phrase, “end of sentence” is registered in the extension element. Furthermore, when the intention expression check is performed in the same phrase, there is no target expression in the same phrase.

215) Since the part-of-speech attribute of “Koto” is a noun class and can be regarded as a dialect positioned at the end of the dialect phrase, next, the phrase “Koto” is directly linked and modified in order from the phrase number [15] toward the first phrase. Search for free-standing words in the current phrase.
216) Since the destination clause number of the clause number [15] is [16], it can be seen that the clause of the clause number [15] directly modifies “ko”, and “say” is found. Since “Take” directly employs the extraction key “Koto”, it stores the notation of depth 0, “Take”, the part-of-speech attribute “verb”, and the dependency relationship “multiple modification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
217) Further, from the phrase number [15], the independent words of the phrase in which the body phrase “Koto” is directly modified are sequentially searched in order from the phrase number [15], and a parallel check is performed. There is no clause that directly modifies clause [15] with “parallel”.
218) Search up to the first phrase, and since there is no independent word of the phrase that directly modifies “Koto”, the extraction of the substructure element with the word “Koto” as the extraction key is finished.

  219) Next, the independent words of the phrase that are directly related to the independent words of depth 0 extracted by directly modifying the extraction key are extracted. An independent word for the extraction key “Koto” is “say”. First, the phrase directly modifying the phrase number [15] including “say” is searched from the phrase number [14] toward the first phrase.

220) Since the destination clause of the clause number [14] is the clause number [15], the independent word “becomes” is found from the clause number [14]. The independent word “Naru” is an independent word that directly modifies the independent word “say” with a depth of 0. Therefore, the depth is added to the depth 1, the expression “Naru”, and the part of speech attribute “verb”. , Store the dependency relationship name “Gift qualification”. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
221) Further, the phrase directly modifying the phrase number [15] is searched from the phrase number [14] toward the first phrase. At that time, 107) the same parallel check is performed. There is no clause that directly modifies clause [14] with “parallel”.
222) Search up to the first phrase, and since there is no independent word of the phrase that directly modifies “say”, the extraction of the partial structural element with the independent word “say” as the extraction key is finished.

223) Next, the phrase independent words directly related to the independent word of depth 1 are extracted. A self-supporting word of depth 1 is “become”. First, a phrase that directly modifies the phrase number [14] including “Naru” is searched from the phrase number [13] toward the first phrase.
224) Since the relation clause of the clause number [13] is the clause number [14], the independent word “purchase” is found from the clause number [13]. The independent word “Purchase” is an independent word that directly modifies the independent word “Naru” with a depth of 1, so add the depth, the depth 2, “Purchase”, and the part-of-speech attribute “sa change”. And the dependency relation name “continuous modification” is stored. In addition, when the intention expression check in the same clause is performed, the intention “request” is registered in the extension element because the correspondence between the expression indicating the intention and the intention shown in FIG. To do.
225) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [13] toward the first phrase, and through the parallel check, the independent word “company” is found from the phrase number [10]. Since the independent word “company” is an independent word that directly modifies the independent word “become” with depth 1, add depth, depth 2, “notice”, and part-of-speech attribute “noun” And the dependency relation name “continuous modification” is stored. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
226) Further, when a clause that directly modifies clause number [14] is searched from clause number [9] toward the first clause and parallel check is performed, the dependency relation name of clause number [9] is “parallel”. It is. Since the related clause number is the clause number of the corresponding clause [10], the independent word “consumer” in the immediately preceding clause [9] is also extracted, the depth 2 and “consumer” notation, and the part of speech attribute “ "Noun" and the dependency relationship are memorized. The dependency relationship name is stored as the dependency relationship name “continuous modification” of the clause to be processed. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
227) Further, a phrase that directly modifies the phrase number [14] is searched from the phrase number [8] toward the first phrase, and through the parallel check, the independent word “Den” is found from the phrase number [0]. Since the independent word “but” is an independent word that directly modifies the independent word “becomes” with a depth of 1, add the depth to the depth 2, the notation of “even”, and the part-of-speech attribute “connective”. , Store the dependency relationship name “continuous modification”. Even if the part-of-speech is a conjunctive, the depth of the corresponding independent word is not 0. Therefore, the dictionary in which the correspondence between the connecting particle, the conjunctive, and the meaning of the connection is described cannot be referenced. Moreover, when the intention expression check in the same phrase is performed, there is no target expression in the same phrase.
228) Further, the phrase directly modifying the phrase number [14] is searched from the phrase number [0] toward the first phrase, and through the parallel check, the first phrase is searched. Since there is no clause that directly modifies the clause number [14], the search for the independent word that directly qualifies the independent word “becomes” ends.

  229) Here, in order to stop the sub-structure element expansion extraction process up to the input depth 2, the extraction of independent words that qualify the extracted independent words “purchase”, “enterprise”, “consumer”, and “even” at depth 2 Is not performed, and the extraction of the partial structure element using “to” as the extraction key is completed by the above extraction process. The result is as shown in FIG.

  In accordance with the above procedure, we search for the syntactic phrase from the last phrase of the linguistic information acquisition result of the example sentence toward the first phrase, the part of speech attribute of the independent word of that phrase is a noun class, If it can be regarded as the last body phrase, it is extracted as an extraction key, and the independent words that are in the dependency relationship, and the independent words that are in the dependency relationship with the independent words, are extracted up to the user-specified depth and expanded. Elements are also registered. When the extraction key search and related independent words and extension elements are extracted up to the first phrase, the extension extraction of the partial structure element with the body key in the example sentence is ended.

  230) If there is a language information acquisition result of the next sentence in the stored language information acquisition result, return to 156) until there is no stored language information acquisition result for the language information acquisition result of the next sentence. , Partial structure element extension extraction is performed for each sentence. If there is no language information acquisition result for the next sentence, the substructure element extension extraction process is terminated, and the process proceeds to the substructure data generation process. Above, description of the process "partial element expansion extraction step" of step S3 of FIG. 1 is complete | finished.

[Partial structure data generation processing]
When the partial structure element extension extraction is performed, the partial structure data generation process of step S4 in FIG. 1 is executed. In the partial structure data generation process, the partial structure data of the document is generated using an independent word having depth information equal to or less than a specified value for each extraction key.

  For example, when there is an input m for designating a depth value from the user, an extraction key and an extended partial structure element having a depth of m or less are selected from all the extended partial structure elements shown in FIGS. Generate structure data. If there is no input of a depth value from the user, the partial structure data is generated by selecting the extended partial structure elements of all depth layers.

  Also, as shown in FIG. 8, when the partial structure element extension extraction process is stopped up to the input depth n, if n = <m, all the extracted extended partial structure elements are selected and the partial structure data is selected. Generate (if n> m, partial structure data is generated with an independent word having a depth specified by m).

  FIG. 9 shows a generation result example in which partial structure data up to a depth m = 2 is generated from the partial structure element extension extraction results of FIGS. 3 and 4 stored in the partial structure element storage unit 13. An example of the generation result in FIG. 9 will be described.

  a) When the information extraction apparatus of the present invention is used for a document summary that requires high processing accuracy or an application that performs similarity determination of a document with high accuracy, for each extraction key during partial structure data generation processing, The extraction key and its word attributes, extension elements, and the depth and notation of independent words that have a dependency relationship with the extraction key and their word attributes, dependency relationships, and expansion elements are registered as the elements. Since there is an input of depth m = 2, as a result of selecting an independent word element having a depth of 2 or less, contents such as the partial structure data example of FIG. 8 are generated. That is, independent word elements having a depth of 3 or more are not extracted. The partial structure data is data that distinguishes the extracted key from other independent words and retains information such as word attributes, dependency relationships, and extended elements.

  b) On the other hand, when the information extraction apparatus of the present invention is used for a document search that requires a high processing speed or an application that makes a document similarity determination at high speed, For each extraction key, the depth and notation of independent words that have a dependency relationship with the extraction key and the extraction key are registered.

  Since there is an input of depth m = 2, as a result of selecting an independent word element having a depth of 2 or less, contents such as the partial structure data example of FIG. 10 are generated. FIG. 10 shows partial structure data that retains word notation and extension elements without distinguishing the extracted key from other independent words.

  c) After the partial structure data is generated for each sentence with respect to the extended partial structure element, the partial structure data generated as shown in FIG. 9 or 10 is stored in the partial structure data storage means 14.

  d) When the partial structure data generation process is completed for all the extended partial structure elements, the partial structure data generation process is ended.

[Partial structure data generation processing with analysis stop processing]
Next, the partial structure data generation process when the analysis is stopped at the depth 2 because the user inputs the depth n = 2 will be described.

  5) When the information extraction apparatus of the present invention is used for a document summary that requires high processing accuracy or an application that performs similarity determination of a document with high accuracy, for each extraction key during partial structure data generation processing, The extraction key and its word attributes, extension elements, and the depth and notation of independent words that have a dependency relationship with the extraction key and their word attributes, dependency relationships, and expansion elements are registered as the elements. Since there is an input of depth m = 2, as a result of selecting all the elements in FIG. 8, contents such as the partial structure data example in FIG. 11 are generated. FIG. 11 shows data in which the extracted key and other independent words are distinguished and information such as each word attribute, dependency relationship, and extended element is retained.

  6) On the other hand, when the information extraction apparatus of the present invention is used for a document search that requires a high processing speed or an application that makes a similarity determination of documents at high speed, For each extraction key, the depth and notation of independent words that have a dependency relationship with the extraction key and the extraction key are registered. Since there is an input of depth m = 2, the extraction key and the depth and notation of independent words that are in a dependency relationship with the extraction key are selected for each extraction key from all the elements in FIG. Contents such as the partial structure data example are generated. FIG. 12 shows partial structure data that retains word notation and extension elements without distinguishing the extracted key from other independent words.

  7) After the partial structure data is generated for each sentence for the extended partial structure element, the partial structure data generated as shown in FIG. 11 or 12 is stored in the partial structure data storage unit 14.

  8) When the partial structure data generation process is completed for all the extended partial structure elements, the partial structure data generation process is ended.

[Output processing]
The extraction information stored in the partial structure data storage means is output by an output method arbitrarily set by the user (step S5). The output may be printed on paper or displayed on a computer display. Moreover, you may output to another program. When the output process ends, the process of FIG. 1 ends.

  As described above, the information extraction method of the present embodiment has no effect even if the field of the target document changes, and does not require a dictionary for restrictions requiring maintenance. In addition, the user does not need specialized knowledge in the language processing field, can find out the development process of the document topic, and can extract document information suitable for practical use with high accuracy.

[Information extraction program and its recording medium]
A program for realizing the information extraction method described so far (hereinafter referred to as an information extraction program) and its recording medium will be described. FIG. 13 shows an example of a hardware configuration diagram of a computer that executes an information extraction program.

  The hardware configuration diagram of FIG. 13 includes a CPU 21, an input / output device 22, a drive device 23, a communication device 24, a main storage device 25, a storage device 26, and a display device 27 that are mutually connected by a bus. Is done.

  The CPU 21 comprehensively controls processing performed by the computer. The input / output device 22 includes a keyboard and a mouse, and is used to input various operation instructions from the user. The communication device 24 is an interface for connecting to a network such as the Internet or a LAN, and includes, for example, a modem, a router, and the like. The main storage device 25 is a storage area for temporarily storing an operating system, programs, and data. A program that realizes the information extraction method is installed in the storage device 26. The drive device 23 can be inserted into a DVD-ROM or CD-ROM, can read programs and data from the recording medium 20, and can write programs and data to the recording medium 20. The display device 27 forms a GUI (Graphical User Interface) screen and displays various windows and data necessary for operation.

  The information extraction program causes the computer to execute each of the document registration unit, language information acquisition unit, partial structure element extension extraction unit, partial structure data generation unit, and output unit shown in FIG. As an example, the flowcharts described in FIGS. 6 and 7 are encoded by a predetermined program language, compiled, and the like.

  The information extraction program is provided by a recording medium 20 such as a DVD-ROM. The recording medium 20 on which the information extraction program is recorded is set in the drive device 23 and installed in the storage device 26.

  The storage medium 20 on which the information extraction program is recorded is a recording medium that records information optically, electrically, or magnetically, such as a CD-ROM, DVD-ROM, flexible disk, magneto-optical disk (MO), or the like. Various types of recording media such as a semiconductor memory that electrically records information, such as a ROM and a flash memory, can be used. Further, the information extraction program may be downloaded via a network.

  In the first embodiment, by using the dependency relationship of independent words, partial structural elements with grammatical limitations are regularly extracted, and information for identifying combinations of these elements as basic topic units for text processing In the extraction method, the information extraction method for further extending the partial structural elements has been described.

  By the way, when the extracted partial structural elements are used as materials in high-level document processing methods such as document retrieval and document summarization (hereinafter referred to as feature quantities), the feature quantities are comprehensive and redundant with respect to the entire target document. Is desired to be low. Redundancy increases as substructure elements are extracted redundantly.

For example, the sentence “I think there are dogs but no cats”
(1: dogs: verbality: engagement: 2)
(2: Yes, but: Usefulness: Dependency / Reading marks: 4),
(3: cat is: body language: relationship: 4),
(4: If not: Usefulness: Dependency: 5),
(5: Think: Usefulness: End of sentence: -1),
Can be broken down into phrases such as

  Here, a sentence is expressed as a five-element element using “:” in parentheses as a separator, and a phrase number, a phrase notation, a phrase part of speech, a phrase type, and a related phrase number are shown in this order. When there is no dependency clause, the destination clause number is set to “−1”. The definition of phrase parts of speech etc. will be described later.

According to the extraction procedure described in Example 1, this sentence is extracted as an extended substructure element.
An extended substructure element consisting of clause 1 and clause 2 with the extraction key being clause 2;
An extended substructure element consisting of clause 1, clause 2, clause 3 and clause 4 with the extraction key being clause 4;
An extended substructure element consisting of clause 1, clause 2, clause 3, clause 4 and clause 5 with the extraction key as clause 5;
The three extended substructure elements can be extracted.

  However, although this extended partial structural element is exhaustive with respect to the target sentence, it becomes highly redundant, for example, the phrase 1 and the phrase 2 are extracted three times.

  The inconvenience of Example 1 that the extracted extended substructure elements are exhaustive but the redundancy is high is that there are few rules for selecting the extraction key of the extended substructure elements, and the clauses that are the extended substructure element members This is thought to be due to the fact that there are few restrictions on the rules for selecting the.

  Therefore, in this embodiment, predetermined extended substructure elements with grammatical limitations are regularly extracted using language information in the document, and a combination of these elements is a basic topic unit for text processing. An information extraction method to be identified will be described. Hereinafter, the extended partial structural element extracted in the present embodiment is referred to as an extended structural element in order to distinguish it from the first embodiment.

  In this embodiment, any one of the phrase types of reading, sentence end, no relation, relation, or indefinite, prescriptiveness, verbality, sa modification, copula, adjective, adverbiality, connectivity, connectivity , Or indefinite, using a clause with one of the clause parts of speech as the basic unit of the extension structure element, the extraction rule of the extension structure element key and the extension structure based on the clause information consisting of the clause type and the clause part of speech By describing element configuration rules, information extraction information is provided that can extract an extended structural element that is exhaustive and has low redundancy for a document or a document set from which an extended structural element is to be extracted. Although copula will be described later, it is a name as a “sentence component” that connects a subject and a predicate, such as a be verb in an English sentence, and is not a part of speech name.

  FIG. 14A shows an example of the procedure of the information extraction process of this embodiment, and FIG. 14B shows a functional block diagram of the information extraction apparatus. In FIG. 14, the same parts as those in FIG. Each step is executed by the document registration unit 31, the language information acquisition unit 32, the extended structural element generation unit 530, and the output unit 35 described in each step.

(Step S1)
When the information extraction is started, the document is registered by the document registration process. The document registration process is executed by the document registration unit 31.

(Step S2)
The registered documents are sequentially stored in the document storage unit 11. When all the registration target documents are registered, language information acquisition processing is performed from the stored documents. The language information acquisition process is executed by a language information acquisition unit.

(Step S30)
When the linguistic information acquisition processing for all the registered documents is completed, the extended structural element generation processing is executed for the document to which the stored linguistic information is added. The extension structure element generation process is executed by the extension structure element generation means 530.

  The extension structure element generation unit 530 includes an extraction key selection unit 531, a front extension processing unit 532, and a rear extension processing unit 533. The extended structural element generation unit 530 generates an extended structural element by extracting a phrase from a sentence with reference to the extended rule 132 defined based on the dependency part information acquired in the phrase part of speech and language information acquisition step. The extraction key selection unit 531 selects an extraction key with reference to an extraction key selection rule 131 described later. The forward extension processing means 532 adds a clause preceding the clause of interest (an extension target clause described later) to the extension structure element. The backward extension processing means 533 adds a clause after the clause of interest (an extension target clause described later) to the extension structure element. Details of each unit included in the extended structural element generation unit 530 will be described later.

(Step S5)
When the extended structural element is generated, the extended structural element is output from the output device by an output method arbitrarily set by the user (step S5). The output may be printed on paper or displayed on a computer display.

  Of steps S1 to S4, the document registration process (S1) and the output process (S5) are the same as those in the first embodiment, and thus description thereof is omitted. Next, the language information acquisition process (S2) and the extended structure element generation process (S30) will be described in detail.

[Language information acquisition processing]
The processing in the language information acquisition processing is the same as that in the first embodiment, but the expression format of the acquired language information is changed. The language information is acquired by using an existing method such as a method of acquiring language information by performing language analysis processing or a method of acquiring automatically or semi-automatically while receiving a user instruction.

  The linguistic information acquisition means 32 acquires from the sentence the dependency relationship of the independent words, the phrase part of speech indicating the part of speech information of the phrase, or the phrase type indicating the classification information of the phrase.

FIG. 15 shows an example of language information obtained as a result of sentence language analysis. The text of this sentence is
“A big bear is walking slowly, but there is no small bear nearby. So, I thought I would try to approach quietly.”… Sentence
It is.

  In FIG. 15, “:” is used as a separator and five elements are shown, and the phrase number, the phrase notation, the phrase part of speech, the phrase type, and the destination phrase number are sequentially shown. If there is no destination clause, the destination clause number is set to “−1”.

  For example, clause 4 in FIG. 15 indicates that the clause number is 4, the clause part of speech is continuity, the clause type is a punctuation mark, and there is no dependency clause. In this embodiment, word information is not managed separately for simplicity, but phrase information and word information may be managed independently for efficiency.

[Extended structure element generation processing]
Next, the extended structural element generation process will be described. First, words and phrases in this embodiment are defined.
a) A word / word has at least information about a real expression, a representative expression, and a part of speech.
・ Words are classified into independent words and functional words according to the part of speech they have.
・ Words with adjunct parts of speech shall be adjunct words.
・ Words with parts of speech other than the adjunct part of speech are independent words.

b) Word parts of speech / nouns, symbols, numbers, and parts of speech similar to unregistered words are referred to as body part of speech.
・ Parts of speech similar to verbs, sa variable nouns, adjectives, and adjective verbs are called prescriptive parts of speech.
・ Parts of speech similar to adverbs are called adverb parts of speech.
・ Parts of speech similar to conjunctions are called connectivity parts of speech.
・ Parts of speech similar to conjunctions are referred to as conjunction parts of speech.
・ Parts of speech similar to particles, auxiliary verbs, auxiliary verbs, auxiliary adjectives, prefixes, and suffixes are referred to as adjunct parts of speech.
・ Nouns, symbols, numerals, unregistered words, verbs, sa-variant nouns, adjectives, adjective verbs, adverbs, conjunctions, conjunctions, particles, auxiliary verbs, auxiliary verbs, auxiliary adjectives, prefixes, and suffixes Part of speech is called indefinite part of speech.

c) A phrase / sentence is one unit of a sentence component composed of words in which one or more independent words are accompanied by zero or more attached words. However, when there are a plurality of independent words, the independent words other than the first independent word are treated as appendages.
A phrase has information on at least constituent words, phrase parts of speech, phrase types, and related phrases.
The phrase part-of-speech is information according to the word information constituting the phrase.
The phrase type is classification information that classifies the phrase according to the modification relationship between the word information constituting the phrase and the related phrase.

d) The phrase part-of-speech / "copula" is a phrase part-of-speech that a self-supporting word is a body part-of-speech part, and a phrase that forms a sentence-end sentence with assertive auxiliary verbs, body stop, and the like.
“Sa modification” is a phrase part-of-speech possessed by a phrase containing a sa-variant noun among the verbal parts of speech.
“Adjective” is a phrase part-of-speech possessed by a phrase including an adjective and an adjective verb among self-supporting part-of-speech parts of speech.
“Verbalism” is a phrase part of speech that a self-supporting word has a body part of speech and a phrase other than a copula phrase corresponds to.
“Adverbiality” is a phrase part of speech that a clause whose independent word is an adverb has.
“Peculiarity” is a phrase part of speech that a self-supporting word has in a phrase containing a verb among the part of speech idiom.
“Community” is a phrase part of speech that a phrase whose independence word is a combination word has.
“Connectivity” is a phrase part-of-speech possessed by a phrase whose independent word is a conjunction.
“Undetermined” is a phrase part of speech that has a phrase that does not satisfy the above-mentioned conditions for the phrase part of speech.

e) Phrase types / phrase types have descending priorities in the order of sentence ending, punctuation, engagement, disconnection, and indefinite.
“End of sentence” is an attribute of a phrase in which a phrase or an alternative symbol exists in the phrase.
“Reading mark” is an attribute of a phrase that has a punctuation mark or a substitute symbol in the phrase, and that has a phrase part of speech that is idiom or degenerate.
“Relationship” is an attribute of a clause that forms a compound word with a dependency clause, whether the modification of the dependency clause is a combination modification or a combination modification.
“Disconnect” is an attribute of a clause that has no relationship with another clause.
“Undetermined” is a phrase type possessed by a phrase that does not satisfy the above-mentioned conditions for the phrase type.

  In the extension structure element generation process, the extension structure element generation means 530 extracts the extension structure element using the language information extracted or stored by the language information acquisition process.

Extraction of extended structural elements
(I) First, an extraction key is selected using an extraction key selection rule 131 composed of a phrase part of speech and a phrase type.
(Ii) For each selected extraction key, the clause is extracted from the sentence using the extension rule 132 composed of the phrase part of speech, and an extended structural element is generated.

(I) Selection of extraction key The extraction key selection rule 131 is a rule for selecting an extraction key. FIG. 16 shows an example of the extraction key selection rule 131. In FIG. 16, one extraction key selection rule 131 is described in one pair of parentheses, and is expressed as a double element having “:” as a separator.

  Each of the two elements is a phrase part of speech and a phrase type, and the phrase type is set to include one or more “|” having a selection meaning as a separator. The extraction key selection rule 131 in FIG. 16 means a rule that uses a phrase having the phrase part of speech of the first element and any of the phrase types of the second element as the extraction key.

  For example, the rule on the first line in FIG. 16 represents a rule in which a phrase whose part of speech is a predicate and a phrase whose phrase type is a sentence end, a punctuation mark, a relation, or a cut is an extraction key. When the extraction key selection unit 531 selects an extraction key for the language information shown in FIG. 15 using the extraction key selection rule shown in the first line of FIG. 16, the phrase 4, the phrase 7, the phrase 10, and the phrase 11 are extracted. Elect as key.

  The extracted key itself becomes one extended structural element. In this embodiment, the clause extracted based on the extension rule 132 is added to the extension structure element (here, the extraction key. After the clause is added to the extraction key, the extension structure element is generated). To do.

(Ii) Generation of Extended Structure Element for Extracting Phrase from Sentence In this embodiment, in the extended structure element generation process (S30), generation processing is performed for a phrase whose position in the document is before the phrase that is the extraction key. Processing is called “forward extension”, and conversely, processing that performs generation processing for a clause whose position in the document is behind the extraction key is called “backward extension”. Use things. The extracted extension structure element is stored in the extension structure element storage unit 130 in an appropriate format.

  FIG. 17 shows an example of an extension rule 132 for forward extension (hereinafter referred to as a forward extension process rule) in the extension structure element generation process. In the forward extension processing rule of FIG. 17, one forward extension processing rule is described in one parenthesis pair, and is expressed as a quadruple element having “:” as a separator. Each element is a phrase part of speech, a dependency depth from the extraction key, and two phrase parts of speech, respectively. One or more of the phrase parts of speech are set with “|” having a selection meaning as a separator.

In the forward extension processing rule or the backward extension process, the clause to be processed is processed as an extended structure clause. A process in which the forward extension processing means 532 generates the extension structure element based on the forward extension processing rule will be described with reference to the flowchart of FIG.
First, using the extracted key as an extended structure clause, a forward extension process in which the extension target clause is the phrase part of speech indicated in the first element is extracted (S11).
It is determined whether or not the extracted second element of the forward extension process is the engagement depth from the extraction key (S12).
Next, it is determined whether or not the phrase having the phrase type “involved” as the extraction key is any of the phrase parts of speech indicated in the third element (S13).
If the third element has a phrase having the phrase type “Relationship” in the extraction key, the phrase is added to the extended structure element (S14).
-Also, it is determined whether or not the phrase part of speech of the clause is any of the clause parts of speech shown in the fourth element (S15). If it is any of the clause parts of speech of the fourth element, the clause is expanded As a phrase, the same processing is performed according to the forward extension processing rule (S16).

  However, in the second element of the forward extension processing rule shown in FIG. 17, when the engagement depth from the extraction key is “0”, all the engagement depths that are not specified (that is, specified in the first and second lines). All other than “0” and “1”).

  For example, the rule on the first line shown in FIG. 17 is the case where the phrase part of speech of the extension target phrase other than depth 1 from the extraction key or the extraction key is a predicate, and the phrase type of “relation” is assigned to the phrase. If the clause has a idiom, submorphism, adjective, adverbial, or verbal, add that clause to the extension structure element, and if the clause has a idiom, submodification, or adjective, add the clause Indicates a rule that is an expansion target clause. Also, the rule shown on the second line is that if the clause part of speech of the extension target clause with a depth of 1 from the extraction key is a predicate, the clause with the clause type of “related” in the clause is a predicate, If it is adjective, adverbial, or episodic, add the clause to the extended structure element, and if it is a prescriptive, subdenaturing, adjective, or syntactic, make the clause a clause to be expanded Show.

  A process of generating an extended structural element based on the forward extension rule shown in FIG. 17 using the language information shown in FIG. 15 and the phrase 4 as the extraction key selected by the extraction key selection rule 131 of FIG. 16 will be described. First, it can be seen from FIG. 15 that the phrase part-of-speech of the phrase 4 is a predicate, and since the phrase 4 is an extraction key, the rule on the first line in FIG. 17 is selected as the forward extension rule. Next, it can be seen from FIG. 15 that the clauses having the relationship “related” to the clause 4 are the clause 3 and the clause 2, and the phrase part of speech of the clause 2 and the clause 3 are the verbal and adverbial properties, respectively. Looking at the third element of the selected extension rule, both the verbality and adverbity are present in the element, so both the phrase 2 and the phrase 3 are added as extension structure elements. Furthermore, since the fourth element of the forward extension process rule has an episodic character, clause 2 of the episodic clause is an extension target clause.

  Since the phrase 2 has a verbal part of speech and the depth of engagement from the extracted key is 1, the forward extension processing rule is selected as the 11th line rule in FIG. From FIG. 15, it can be seen that the phrase having the relationship “related” to the phrase 2 is the phrase 1, and the phrase part of speech of the phrase 1 is adjective. Therefore, the adjective is added to the third element of the selected forward extension processing rule. Therefore, clause 1 is also added to the extended structure element. Moreover, since the fourth element also has a description, clause 1 becomes an extension target clause.

  Since phrase 1 has an adjective phrase part-of-speech and the depth of engagement from the extraction key is 2, the forward extension processing rule is selected as the rule on the seventh line in FIG. 17, but from FIG. Since there is no clause having a relationship of “relation”, the processing is ended, and as a result, an extended structural element including the clause 1, the clause 2, the clause 3, and the clause 4 is extracted.

  Similarly, FIG. 19 shows the result of generating an extended structural element based on the forward extension processing rule shown in FIG. 17 using the language information shown in FIG. 15 and the extracted key selected based on the extracted key selection rule in FIG. Show. However, in FIG. 19, one extended structural element is described in one pair of parentheses, and consists of one or more pieces of clause information with “:” as a separator. Each phrase information is a double element with “,” as a separator. The first element indicates the phrase number, and the second element indicates the depth of engagement from the extraction key. In FIG. 19, clauses 4, 7, 11 and 13 are extracted keys, and a clause ahead of each extraction key (with a younger clause number) is added as an extended structural element. In FIG. 19, the phrase part of speech and the phrase type are not included in the extended structural element. However, referring to the language information in FIG. 15, the extended structural element can include the phrase part of speech and the phrase type (FIG. 21). The same).

  Subsequently, the backward expansion process will be described. FIG. 20 shows an example of an extension rule (hereinafter referred to as a backward extension process rule) 132 in the extension structure element generation process. In FIG. 20, one backward extension processing rule is described in one parenthesis pair. When the phrase part-of-speech of the third and fourth elements is the character string “none”, it indicates that there is no candidate part-of-speech, that is, no addition is made to the extended structural element.

The backward extension process performed by the backward extension processing unit 533 will be described with reference to the flowchart of FIG.
First, using the extraction key as an extension target clause, a backward extension processing rule in which the extension target clause is the phrase part of speech indicated in the first element is extracted (S21).
It is determined whether or not the extracted second element of the backward extension processing rule is the engagement depth from the extraction key (S22).
Next, it is determined whether or not the expansion target clause has a clause type of “relation” (S23).
Next, it is determined whether or not the destination clause of the clause to be expanded is any of the clause parts of speech indicated in the third element (S24).
When the dependency clause of the extension target clause is any of the phrase parts of speech of the third element, the dependency clause is added to the extension structure element (S25).
It is also determined whether or not the phrase part of speech of the dependency clause is any of the phrase parts of speech indicated in the fourth element (S26), and the phrase part of speech of the dependency clause is any of the phrase parts of speech of the fourth element In this case, the backward extension process is repeated with the dependency clause as the extension target clause (S27).

  For example, according to the rule on the second line shown in FIG. 20, when the phrase type of the extraction key is “relation”, the phrase part of speech is a predicate, and the dependency clause of the phrase is a discourse The clause is added to the extension structure element, and the added clause is set as the extension target clause. Further, the rule shown in the first line indicates a rule that does not perform the extension process when the phrase type of the expansion target phrase that is not the extraction key is “relation” and the phrase part of speech is a predicate.

  A process for adding an extension structure element to the extension structure element generated as shown in FIG. 19 based on the backward extension process rule shown in FIG. 20 will be described. Note that the extension structure element to be generated is not affected by the execution order of the extension extension element forward extension process and the backward extension process. Therefore, contrary to the present embodiment, the backward extension process may be executed first, and the forward extension process may be executed on the result, or only one of the forward extension process and the backward extension process may be performed depending on the application. May be executed.

  In FIG. 19, the extraction key of the extended structural element on the first line is clause 4, and clause 4 is not related to clause information. Also, the extraction key of the extended structural element on the second line is the phrase 7, and the phrase 7 is not related to the phrase information. In addition, the extraction key of the extended structural element on the third line is the phrase 10, and the phrase information of the phrase 11 that is the relation destination of the phrase 10 is not syntactic. Further, the extraction key of the extended structure element on the fifth line is the phrase 13, and the phrase information of the phrase 13 is not relevant. Accordingly, the first, second, third, and fifth lines in FIG. 19 are not added to the expansion structure element by the backward expansion process.

  The extraction key of the extended structural element on the fourth line is the phrase 11, and from FIG. 15, it can be seen that the phrase type is “related” and the phrase part of speech is a predicate, so the backward extension processing rule is the second line in FIG. 20. Rules are selected. Next, it can be seen from FIG. 15 that the dependency clause of the clause 11 is the clause 12 and the clause part of speech of the clause 12 is episodic, so when looking at the third element of the selected backward extension processing rule, Therefore, the clause 12 is added as an extended structural element. Furthermore, the phrase 12 which is a body phrase according to the fourth element of the extension rule becomes the extension target phrase.

  Next, the phrase 12 has a phrase part-of-speech, so the backward expansion process rule is selected from the rule on the ninth line in FIG. From FIG. 15, it can be seen that the phrase type of the phrase 12 is “relation” and the relation destination phrase is the phrase 13. Further, from FIG. 15, the phrase part of speech of the phrase 13 is adjective, but since the adjective is not present in the third element of the selected backward extension processing rule, the processing is ended, and as a result, the phrase 12 is included in the extended structure element. Added.

  FIG. 22 shows an extension structure element generated based on the backward extension processing rule with respect to the extension structure element shown in FIG. The expression format is the same as the extended structural element in FIG.

  FIG. 23 shows an extended structural element in which the extended structural element shown in FIG. According to FIG. 23, it can be seen that the extracted extended structural elements are exhaustive and can be extracted so as to reduce redundancy.

  Therefore, by providing an extraction key selection rule for an extension structure element and providing an extension rule 132 for selecting a clause to be an extension structure element, an extension structure element with comprehensive and low redundancy can be generated.

[Information extraction program and its recording medium]
A program for realizing the information extraction method described so far (hereinafter referred to as an information extraction program) and its recording medium will be described. The hardware configuration diagram of the computer that executes the information extraction program is the same as that in FIG.

  The information extraction program causes the computer to execute each unit of the document registration unit 31, the language information acquisition unit 32, the extended structural element generation unit 530, and the output unit 35 of FIG. As an example, the processes in the flowcharts described with reference to FIGS. 18 and 21 are coded in a predetermined program language and compiled.

  The information extraction program is provided via a recording medium 20 such as a DVD-ROM or a network. The recording medium 20 on which the information extraction program is recorded is set in the drive device 23 and installed in the storage device 26.

  The storage medium 20 on which the information extraction program is recorded is a recording medium that records information optically, electrically, or magnetically, such as a CD-ROM, DVD-ROM, flexible disk, magneto-optical disk (MO), or the like. Various types of recording media such as a semiconductor memory that electrically records information, such as a ROM and a flash memory, can be used.

  As described above, the information extraction method of the present embodiment has no effect even if the field of the target document changes, and does not require a dictionary for restrictions requiring maintenance. In addition, the user does not need specialized knowledge in the language processing field, can find out the development process of the document topic, and can extract document information suitable for practical use with high accuracy. Also, an extended structural element extraction key selection rule is provided, and an extended rule for selecting a clause to be the extended structural element is provided, so that an extended structural element with comprehensive and low redundancy can be generated.

It is a flowchart figure which shows the whole flow of an information extraction process. It is a figure which shows an example of a linguistic information acquisition result. It is the extraction example which extracted the extended partial structure element of the precaution. It is the extraction example which extracted the extended partial structure element of the word. It is an example of the flowchart figure of a partial structural element extraction process. It is a dictionary in which connection meanings and intention expressions are described for connection particles or conjunctions. It is an example of the flowchart figure of the partial structural element extraction process with designation | designated depth information. It is the extraction example which extracted the extended substructure element in case there is specification of depth information. FIG. 5 is a generation result example in which partial structure data from a partial structure element extension extraction result of FIG. 3 or 4 to a depth n = 2 is generated. This is partial structure data that retains word notation and extension elements without distinguishing the extracted key from other independent words. FIG. 9 is a generation result example in which partial structure data from a partial structure element extension extraction result of FIG. 8 to a depth n = 2 is generated. This is partial structure data that retains word notation and extension elements without distinguishing the extracted key from other independent words. It is an example of the hardware block diagram of the computer which performs an information extraction program. FIG. 10 is a flowchart illustrating an example of a procedure of information extraction processing according to the second embodiment. It is an example of the linguistic information obtained as a result of the linguistic analysis of a sentence. It is an example of an extraction key selection rule. It is an example of an extension rule (forward extension process rule). It is an example of the flowchart figure which shows the procedure of the process which a front expansion process means produces | generates an expansion structure element. It is an example of the extension structure element produced | generated based on the forward extension process rule. It is an example of an extension rule (backward extension process rule). It is an example of the flowchart figure which shows the procedure of the process which a back extension process means produces | generates an expansion structure element. It is an example of the extension structural element produced | generated based on the back extension process rule. FIG. 22 is an extended structural element in which the extended structural element in FIG. 21 is displayed with text characters.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Document storage means 12 Language information storage means 13 Partial structure element storage means 14 Partial structure data storage means 31 Document registration means 32 Language information acquisition result 33 Partial structure element expansion extraction means 34 Partial structure data generation means 35 Output means 131 Extraction key selection Rule 132 Extended rule 330 Depth information acquisition means 331 Extraction key selection means 332 Autonomous word acquisition means 333 Depth information addition means 334 Depth information designation means 530 Extended structure element generation means 531 Extraction key selection means 532 Forward extension processing means 533 Back Extended processing means

Claims (38)

In an information extraction method for extracting document information,
A language information acquisition step for acquiring a dependency relationship of independent words;
A depth information acquisition step of acquiring depth information indicating the strength of the dependency relationship between one independent word acquired from the document and the other independent words by the language information acquisition step;
Partial structure data generation for generating partial structure data composed of independent words having depth information greater than or equal to a predetermined value of the one independent word based on the depth information of each independent word acquired in the depth information acquisition step Steps,
An information extraction method characterized by comprising: The linguistic information acquisition step acquires the independent word of the destination for each independent word of the source,
The depth information acquisition step includes:
An extraction key selection step for selecting a predetermined word out of the independent words as an extraction key and an independent word on which the selected extraction key is related are acquired, and initial value depth information is acquired in the independent word. An independent word acquisition step to be given;
A value obtained by acquiring a dependency independent word whose dependency is the acquired independent word and adding a predetermined value to the value of the acquired independent word depth information as depth information of the dependency independent word A depth information adding step for providing
The information extracting method according to claim 1, further comprising: The linguistic information acquisition step acquires the independent word of the destination for each independent word of the source,
The depth information acquisition step includes:
An extraction key selection step for selecting a predetermined word from the independent words as an extraction key, and obtaining an independent word on which the selected extraction key is related, and giving initial value depth information to the independent word An independent word acquisition step,
A value obtained by acquiring a dependency independent word whose dependency is the acquired independent word and adding a predetermined value to the value of the acquired independent word depth information as depth information of the dependency independent word A depth information adding step for providing
The information extracting method according to claim 1, further comprising: The depth information acquisition step includes a depth information designation step for designating a value of depth information,
Obtaining an independent word of depth information whose depth information value is less than or equal to the value specified by the depth information specifying step;
The information extraction method according to claim 2 or 3, wherein Conjunctive particle or Conjunctive particle / conjunctive particle dictionary with connection particle described
The depth information acquisition step includes:
Based on the conjunction or connective particle that the clause containing the acquired independent word has, refer to the conjunction / connective particle dictionary, and extract the connection meaning of the conjunction or connective particle,
Then, the meaning of the extracted connection is added to the obtained independent word extension element.
The information extracting method according to claim 2, wherein the information extracting method is an information extracting method. It has an intention expression dictionary describing expressions expressing intentions,
The depth information acquisition step includes:
Based on the intention expression that the clause containing the acquired independent word has, refer to the intention expression dictionary, extract the meaning of the intention expression,
Then, the meaning of the extracted intention expression is added to the obtained independent word extension element.
The information extracting method according to claim 2, wherein the information extracting method is an information extracting method. The depth information acquisition step includes:
When the acquired independent word is included in the last clause in one sentence, a notation that the sentence is at the end is added to the extended element of the independent word.
The information extraction method according to claim 3, wherein: The partial structure data generation step includes:
The modification relationship indicating the relationship in which the independent word acquired by the depth information acquisition step modifies the extraction key, the meaning of the connection added to the extended element of the acquired independent word, the meaning of the intention expression, or the end of the sentence Notation and
Is generated as partial structure data in association with the acquired independent word,
The information extraction method according to claim 1, wherein: In an information extraction device for extracting document information,
Language information acquisition means for acquiring dependency relations of independent words;
Depth information acquisition means for acquiring depth information indicating the strength of the dependency relationship between one independent word acquired from the document by the language information acquisition means and another independent word;
Partial structure data generation for generating partial structure data composed of independent words having depth information of a predetermined value or more of the one independent word based on the depth information of each independent word acquired by the depth information acquisition means Means,
An information extraction apparatus comprising: The linguistic information acquisition means acquires the independent word of the destination for each independent word of the source,
The depth information acquisition means includes
An extraction key selection means for selecting a predetermined word out of the independent words as an extraction key and an independent word on which the selected extraction key is related are acquired, and initial depth information is stored in the independent word. An independent word acquisition means to be given;
A value obtained by acquiring a dependency independent word whose dependency is the acquired independent word and adding a predetermined value to the value of the acquired independent word depth information as depth information of the dependency independent word Depth information adding means for providing
10. The information extraction device according to claim 9, further comprising: The linguistic information acquisition means acquires the independent word of the destination for each independent word of the source,
The depth information acquisition means includes
An extraction key selection means for selecting a predetermined body language as an extraction key from the independent words and a dependency independent word having the selected extraction key as a dependency, and obtaining depth information of an initial value in the independent word An independent word acquisition means for providing
A value obtained by acquiring a dependency independent word whose dependency is the acquired independent word and adding a predetermined value to the value of the acquired independent word depth information as depth information of the dependency independent word Depth information adding means for providing
10. The information extraction device according to claim 9, further comprising: The depth information acquisition means includes depth information designation means for designating a value of depth information,
Obtaining an independent word of depth information whose depth information value is equal to or less than the value specified by the depth information specifying means;
12. The information extraction apparatus according to claim 10 or 11, wherein Conjunctive particle or Conjunctive particle / conjunctive particle dictionary with connection particle described
The depth information acquisition means includes
Based on the conjunction or connective particle that the clause containing the acquired independent word has, refer to the conjunction / connective particle dictionary, and extract the connection meaning of the conjunction or connective particle,
Then, the meaning of the extracted connection is added to the obtained independent word extension element.
14. The information extraction device according to claim 10, wherein the information extraction device is a device. It has an intention expression dictionary describing expressions expressing intentions,
The depth information acquisition means includes
Based on the intention expression that the clause containing the acquired independent word has, refer to the intention expression dictionary, extract the meaning of the intention expression,
Then, the meaning of the extracted intention expression is added to the obtained independent word extension element.
14. The information extraction device according to claim 10, wherein the information extraction device is a device. The depth information acquisition means includes
If the acquired independent word is included in the last sentence in a sentence, a notation that the sentence ends is added to the extended element of the independent word.
The information extraction apparatus according to claim 11, wherein The partial structure data generating means
A modification relationship indicating a relationship in which the independent word acquired by the depth information acquisition unit modifies the extraction key, a meaning of connection added to an extension element of the acquired independent word, a meaning of intention expression, or a sentence end Notation and
Is generated as partial structure data in association with the acquired independent word,
The information extraction device according to claim 9.   An information extraction program for causing a computer to execute the information extraction method according to claim 1.   A recording medium on which the information extraction program according to claim 17 is recorded. In an information extraction method for extracting document information,
A language information acquisition step of acquiring a dependency relation of independent words, a phrase part of speech indicating a part of speech information of a phrase or a phrase type indicating classification information of the phrase;
An extraction key selection step of selecting an extraction key with reference to an extraction key selection rule that defines the phrase type in association with the phrase part of speech;
An extended structural element that generates an extended structural element by referring to the extended rule defined based on the phrase part of speech and the dependency relationship based on the extraction key and extracting a clause that matches the extended rule from the sentence. Generation step;
An information extraction method characterized by comprising: The phrase part-of-speech is one of a predicative clause, a verbal clause, a sacrificial clause, a copula clause, an adjective clause, an adverbial clause, a conjunctive clause, a connectivity clause, or an indefinite clause,
The phrase type is one of the following:
The information extraction method according to claim 19, wherein: The extended structural element is
The basic unit of any of the predicative clauses, the verbal clauses, the samurai clauses, the copula clauses, the adjective clauses, the adverbial clauses, the conjunctive clauses, the connectivity clauses, or the indefinite clauses, or Composed of combinations,
20. The information extraction method according to claim 19, wherein: Each clause of the basic unit has any one of the following phrase types: reading, end of sentence, unrelated, unrelated or undefined
The information extraction method according to claim 21, wherein: The extraction key selection step includes:
Predicate clauses that do not have indefinite attributes,
A modified clause that does not have indefinite attributes,
Copula clause,
An adjective clause with one of the following attributes:
Adverbial clauses with end-of-sentence or truncation attributes,
Selecting either the end of sentence or the narrative phrase having a cutting attribute as the extraction key,
23. The information extracting method according to claim 19, wherein the information extracting method is performed. The extended structural element generation step includes:
Using the extracted key selected in the extraction key selection step as an extension target clause for extracting a clause from the sentence, the extension target clause is a prescriptive phrase, a samurai phrase, a copula phrase, an adjective phrase, or In the case of adverbial clauses,
Forward extension that generates an extended structure element by adding a predicative clause, a sacrificial clause, an adjective clause, an episodic clause, or an adverbial clause having an attribute related to the extension target clause to the extraction key Has processing steps,
If the clause added by the forward extension processing step is a prescriptive clause, a sacrificial clause, an adjective clause, or a verbal clause, the clause is set as the extension target clause, and the forward extension processing step is further performed. ,
24. The information extracting method according to claim 19, wherein the information extracting method is performed. The extended structural element generation step includes:
When the extraction key selected in the extraction key selection step is an expansion target clause for extracting a clause from the sentence, and when the expansion target clause is a body phrase,
A forward extension processing step of adding an adjective clause, a verbal clause or an adverbial clause having an attribute related to the extension target clause to generate an extended structural element by adding to the extraction key;
When the clause added by the forward extension processing step is an adjective clause or a verbal clause, the forward extension processing step is further performed with the clause as the extension target clause.
24. The information extracting method according to claim 19, wherein the information extracting method is performed. The extended structural element generation step includes:
The extraction key selected in the extraction key selection step is used as an expansion target clause for extracting a clause from the sentence, and the expansion target clause has a related attribute, a prescriptive phrase, a modified phrase, and an adjective phrase Or adverbial clauses,
If the destination clause of the extension target clause is a body clause, it has a backward extension processing step of generating an extension structure element by adding the destination clause to the extraction key,
In the case where the verbal clause added by the backward extension processing step has a dependency attribute, and the dependency clause of the added verbal clause is a verbal clause, the clause is further set as the extension target clause. Performing the backward extension processing step;
24. The information extracting method according to claim 19, wherein the information extracting method is performed. The extended structural element generation step includes:
When the extracted key selected by the extraction key selection step is an expansion target clause for extracting a clause from the sentence, the expansion target clause has a related attribute,
If the destination clause of the extension target clause is a body clause or a copula clause, a backward extension processing step of generating an extension structure element by adding the clause of the destination to the extraction key;
In the case where the verbal clause added by the backward extension processing step has a dependency attribute, and the dependency clause of the added verbal clause is a verbal clause, the clause is further set as the extension target clause. Performing the backward extension processing step;
24. The information extracting method according to claim 19, wherein the information extracting method is performed. In an information extraction device for extracting document information,
Language information acquisition means for acquiring dependency relations of independent words, phrase part of speech indicating the part of speech information of the phrase or phrase type indicating the classification information of the phrase;
An extraction key selection means for selecting an extraction key with reference to an extraction key selection rule that defines the phrase type in association with the phrase part of speech;
An extended structural element that generates an extended structural element by referring to the extended rule defined based on the phrase part of speech and the dependency relation based on the extraction key and extracting a clause that matches the extended rule from the sentence. Generating means;
An information extraction apparatus comprising: The phrase part-of-speech is one of a predicative clause, a verbal clause, a sacrificial clause, a copula clause, an adjective clause, an adverbial clause, a conjunctive clause, a connectivity clause, or an indefinite clause,
The phrase type is one of the following:
29. The information extraction device according to claim 28, wherein: The extended structural element is
The basic unit of any of the predicative clauses, the verbal clauses, the samurai clauses, the copula clauses, the adjective clauses, the adverbial clauses, the conjunctive clauses, the connectivity clauses, or the indefinite clauses, or Composed of combinations,
29. The information extraction device according to claim 28. Each clause of the basic unit has any one of the following phrase types: reading, end of sentence, unrelated, unrelated or undefined
30. The information extracting apparatus according to claim 29. The extraction key selection means includes
Predicate clauses that do not have indefinite attributes,
A modified clause that does not have indefinite attributes,
Copula clause,
An adjective clause with one of the following attributes:
Adverbial clauses with end-of-sentence or truncation attributes,
Selecting either the end of sentence or the narrative phrase having a cutting attribute as the extraction key,
32. The information extraction device according to claim 28, wherein the information extraction device is a device. The extended structural element generation means includes
Using the extracted key selected by the extraction key selection means as an extension target clause for extracting a clause from the sentence, the extension target clause is a prescriptive phrase, a sacrificial phrase, a copula phrase, an adjective phrase, or In the case of adverbial clauses,
Forward extension that generates an extended structure element by adding a predicative clause, a sacrificial clause, an adjective clause, an episodic clause, or an adverbial clause having an attribute related to the extension target clause to the extraction key Having processing means,
When the clause added by the forward extension processing means is a predicative clause, a modified clause, an adjective clause, or a semantic clause, the forward extension processing means is further performed with the clause as the extension target clause. ,
33. The information extraction device according to claim 28, wherein the information extraction device is a device. The extended structural element generation means includes
When the extraction target selected by the extraction key selection means is an expansion target clause for extracting a clause from the sentence, and the expansion target clause is a body phrase,
Forward extension processing means for generating an extended structural element by adding an adjective clause, a verbal clause or an adverbial clause having an attribute related to the extension target clause to the extraction key;
If the clause added by the forward extension processing means is an adjective clause or a verbal clause, the forward extension processing means is further performed with the clause as the extension target clause;
33. The information extraction device according to claim 28, wherein the information extraction device is a device. The extended structural element generation means includes
Using the extracted key selected by the extraction key selecting means as an expansion target clause for extracting a clause from the sentence, a predicative phrase, a sacrificial phrase, an adjective phrase having an attribute related to the expansion target phrase Or adverbial clauses,
If the destination clause of the extension target clause is a body clause, it has a backward extension processing means for generating an extension structure element by adding the clause of the destination to the extraction key,
In the case where the verbal clause added by the backward extension processing means has a dependency attribute, and the dependency clause of the added verbal clause is a verbal clause, the clause is further set as the extension target clause. Performing the backward extension processing means;
33. The information extraction device according to claim 28, wherein the information extraction device is a device. The extended structural element generation means includes
The extraction key selected by the extraction key selection means is an extension target clause for extracting a clause from the sentence, and the extension target clause has a related attribute,
If the destination clause of the extension target clause is a body language clause or a copula clause, it has a backward extension processing means for generating an extension structure element by adding the clause of the destination to the extraction key,
In the case where the verbal clause added by the backward extension processing means has a dependency attribute, and the dependency clause of the added verbal clause is a verbal clause, the clause is further set as the extension target clause. Performing the backward extension processing means;
33. The information extraction method according to any one of claims 28 to 32, wherein:   An information extraction program for causing a computer to execute the information extraction method according to claim 19 to 27.   38. A recording medium on which the information extraction program according to claim 37 is recorded.

Images