JP2002334091A - Method, device and program for acquiring knowledge and recording medium recording the program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract a part explaining a thing from an electronic text. SOLUTION: A spot information extraction part 1 inputs a pair of a spot name and an explanation text and adds a pair of the spot name and a relation sort with a spot value of the same vocabulary as a spot value in a spot vocabulary DB 6 to each sentence divided from the explanation text to constitute explanation text information data. A clause information extraction part 2 outputs a clause string constituting a sentence in the explanation text information data. A modification information extraction part 3 inputs the clause string and outputs the modification relation of clauses. An explanation phrase constitution part 4 extracts a part writing about a spot from the sentence on the basis of the relation sort corresponding to the sentence in the explanation text information data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge acquisition apparatus for extracting a part (hereinafter referred to as an explanation phrase) describing an object from an electronic text describing an object such as a travel destination or a product.

[0002]

2. Description of the Related Art In recent years, with the spread of the Internet, many texts have been provided electronically.
In view of such circumstances, research has been conducted to automatically construct a knowledge base used for intelligent systems and machine translation from texts on the Internet.

As an example, an intelligent system for selecting a travel destination will be described. As a system for presenting travel destinations, there has conventionally been a travel search site that inputs “place”, “price”, etc., and recommends a travel destination that meets a request. However, with such a search system, it is difficult to present an appropriate travel destination to “users who do not have a specific plan for travel” such as location and price. Further, in the conventional search system, when a search is performed using a keyword such as "enjoy in a family" or "a thrill," there is no knowledge base that meets such a user's request. Therefore, there is a problem that a search based on such ambiguous desires is difficult.

As described above, as a method of recommending an appropriate travel destination to a user who does not yet have a specific plan, a phrase that can be a user's desire is presented from the system side,
There is a way to make it select. In this way, by presenting a phrase that can be a user's desire, the user's potential desire can be extracted, and an appropriate travel destination can be searched based on the potential desire. To achieve this,
It is necessary to collect a large number of phrases that can be a user's desire, and a technique for automatically collecting such phrases from an electronic text is desired.

As a technique for extracting a phrase from an electronic text, there has been an automatic text summarization, and there is an important sentence extraction method for extracting the phrase based on the frequency of appearance of a keyword in the text.

[0006]

However, with such a method, there is a problem that words that are not adjacent to important words, such as "enjoy in the family" in the above example, and keywords with a low frequency of appearance cannot be extracted. As described above, in the related art, a method that focuses on words that appear frequently is used to extract a phrase that can be a user's desire from the text. There was a point.

[0007] An object of the present invention is to provide a knowledge acquisition method and apparatus for extracting a part describing an object such as a tourist spot from an electronic text such as a tourist guide.

[0008]

In order to achieve the above object, a knowledge acquiring apparatus according to the present invention comprises a spot vocabulary database, a text information initial database, an explanatory text information database, a spot information extracting means, a phrase information extracting means, It has a dependency information extracting unit and an explanation phrase forming unit.

The spot information extracting means includes: a spot name;
A set of explanatory texts describing a spot name object is input, and the explanatory text is divided for each sentence to form partial text information including a set of a text number, a spot name, and a partial text. And the same vocabulary as the spot values in the spot vocabulary database that stores a plurality of spot values, which are vocabulary, and a relationship type (relation type) between these spot values for each spot name for each partial text information Appears, a set of the input spot name and the sentence is assigned a set of the spot value and a relation type between the input spot name and the spot value, and the explanatory text information data is generated. Configure and store in the explanatory text information database.

The phrase information extracting means inputs a sentence of explanatory text information data from an explanatory text information database,
A sequence of phrases (phrase sequence) constituting the sentence is output.

The dependency information extracting means inputs a phrase string,
The dependency relation of the phrase is extracted and output.

The explanation phrase constructing means inputs explanation text information from the explanation text information database. (1) When the relation type corresponding to a sentence in the explanation text information is the same as the spot name in the explanation text information, Extract the set of clauses whose values are in a case relation and the clauses that are sequentially related from the clauses. (2) When the relation type corresponding to the sentence in the explanation text information is a word representing the type of the spot name, the phrase having the adjunct form of the predicate in the spot value and the phrase (3) When the relation type corresponding to the sentence in the explanation text information is a word that represents a part of the spot name, A set of clauses whose values are in a case relation and the clauses that are sequentially affected from the clauses are extracted, and the clauses that are adjuncts of the predicates in the spot value and the clauses that are affected by the clause are extracted. Perform a process to retrieve the set.

When the spot information extracting means forms the explanation text information data, it is possible to know the relation type between the spot value and the spot name in each sentence. The phrase information extraction unit divides the sentence into phrase units, so that the phrase can be handled in phrase units. The dependency information extracting means outputs the dependency relationship between the phrases,
It is possible to know the dependency relation of the phrase, and based on this, it is possible to specify the portion written about the spot name. The explanatory phrase constructing means extracts, from the sentence based on the relation type corresponding to the sentence in the explanation text information, a portion written about the spot name.

As described above, according to the present invention, it is possible to obtain a portion written about a spot name from a text.

[0015]

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

Referring to FIG. 1, a knowledge acquisition apparatus according to an embodiment of the present invention includes a spot information extraction unit 1, a phrase information extraction unit 2, a dependency information extraction unit 3, an explanatory phrase construction unit 4, a content DB 5, a spot vocabulary, and the like. It consists of DB6, text information initial DB7, explanation text information DB8, and explanation phrase DB9.

As shown in FIG. 2, the content DB 5 stores a set of pairs of names (spot names) indicating spots and texts (explanatory texts) describing the spots. In addition, the spot vocabulary DB6, as shown in FIG.
For a plurality of vocabularies (spot values), data is stored in which the spot values are associated with the respective spot names and in what relation (relation type). Note that the relation type in the spot vocabulary data is “sp” associated with the spot name.
ot "," part "having a relationship of spot name and part of (lower)," part * "having a relationship of spot name and part of (higher), and" kind "having a relationship of spot name and kind of ".

The spot information extracting unit 1 has two processes, a "text dividing process" for dividing the explanation text for each sentence and a "text information extracting process" for extracting text information from the sentences divided by the text dividing process. I do.

In the text division process, a set of a spot name and a description text is extracted from the content DB 5, the description text is divided into sentence units, each of which is a partial text.
A list of data (hereinafter, referred to as partial text information) composed of three sets of | text number, spot name, and partial text | The division into sentence units can be realized by, for example, examining the character string of the explanation text one character at a time, and dividing the text at the place where the period "."
In the text division processing, partial text information is created for all pairs of spot names and explanation texts in the content DB 5 and is constructed as the text information initial DB 7. FIG. 4 shows an example of a list of partial text information.

In the text information extraction process, partial text information is sequentially extracted from the text information initial DB 7, and spot vocabulary data including a spot name matching a spot name in the partial text information is extracted from the spot vocabulary DB 6. Here, the spot vocabulary data is data in which a plurality of spot values are associated with a type of relationship between the spot values and each spot, as shown in FIG. Next, the text information extraction process selects, for each spot value in the spot vocabulary data extracted from the spot vocabulary DB 6, only those whose partial text in the partial text information includes a character string that matches the spot value. . Then, the spot value selected in this way and the “relation type corresponding to the spot value” are combined to form explanatory text information data together with the partial text information. In the text information extraction processing, explanatory text information data is created for all the partial text information in the initial text information DB 7, and this is constructed as the explanatory text information DB 8. Explanation text information DB
8 shows an example of the configuration of FIG.

The phrase information extraction unit 2 includes an explanation text information DB
The description text information data is input from step 8 and a sequence of phrases (phrase sequence) constituting the sentence is output. To get the phrase sequence,
The conventional morphological analysis technology is used. In particular,
The explanation text information data is sequentially read from the explanation text information DB 8, and the partial text in the explanation text information data is used as an input character string, and the character string of the input sentence starting with the character string at the left end of the input character string and the headword of the word dictionary And the longest headword among all corresponding headwords is set as the first word. Next, a character string obtained by removing the word from the input character string is used as an input character string, and collation with the word dictionary is performed. The next collation is performed in the same manner as the first collation, and words are extracted until there is no more input character string. The word sequence (hereinafter referred to as a word sequence) and the part-of-speech sequence (hereinafter referred to as a part-of-speech sequence) thus collected are obtained by arranging part-of-speech names corresponding to headwords in a word dictionary. Are grouped into a unitary phrase string consisting of one or more independent words (nouns, verbs, etc.) + 0 or more attached words (particles, suffixes, etc.). The phrase sequence thus collected is a phrase sequence.

The dependency information extraction unit 3 inputs the phrase string acquired by the phrase information extraction unit 2 and outputs dependency information of the phrase. In order to obtain phrase dependency information, a conventional dependency analysis technique is used. Specifically, an attribute indicating a function is added to a phrase from a phrase sequence, and a parallel structure and a dependency structure are analyzed based on the attribute. The attributes to be added to the clause include “nominal”, “decim”, “ga”, “、”, “key of parallel”, “parallel key”, and the like. The expression “parallel key” refers to an expression indicating the existence of a parallel structure, such as “book and pencil” and “book”. In the analysis of the parallel structure, for example, if there is a clause of “Nominal parallel key”, the end of “Nominal”
The most similar phrase sequence is found in the phrase, and "the range of the parallel structure" is determined. The analysis of the dependency is performed according to a rule such as, for example, “a“ ga-ga ”clause pertains to“ declarations ””. The dependency information between the phrases thus obtained is configured as a dependency information list. Here, the dependency information list is
This is a list of data consisting of three sets of {dependency source clause, dependency destination clause, dependency attribute}. The dependency attribute indicates a dependency relationship between a higher-level clause and a lower-level clause.
For example, there is a case relation "case", a case particle "no", an adjunct case particle, an "adjective" relation "body", and a predicate adnominal form "embed". .
For example, in the example sentence of the three phrases {"Longest in Japan", "Proud", and "Suspension Bridge"}, the dependency of this spot value "Suspension Bridge" is "" Proud "→" Suspension Bridge "}. As for the dependency relationship from the phrase “proud” to the “suspension bridge”, the dependency attribute “embedded” is given by “relation by the union form of the predicate”. Thus, "embed", "body",
An attribute such as “case” is assigned to each combination of clauses in which dependency occurs.

The explanatory phrase constructing unit 4 performs two processes: a "pattern determining process" for determining the text pattern of the partial text and an "explanatory phrase generating process" for generating an explanatory phrase from the dependency information of the partial text. .

The pattern determination process is performed in the explanation text information DB
8, the description text information data is sequentially extracted, and all the relation types existing in the explanation text information data are extracted. Next, it is checked whether a plurality of relation types exist. If there are a plurality of them, first, it is checked whether or not they match the relation type of the column whose combination flag in the pattern determination table is “1”. Here, the pattern determination table is a table that defines a combination of a dependency relationship and a relationship type for determining a text pattern from a dependency relationship of a clause including a spot value in a partial text. As shown in FIG. 6, the pattern determination table includes a “combination flag” for determining whether the relation type is composed of one or two sets, a “pattern” representing a pattern value, and a spot corresponding to the relation type. A “higher-order dependency flag 1” for determining whether a dependency exists from a higher-ranking clause in the value, and a spot value corresponding to the relation type for determining whether a dependency exists in a lower-ranking clause It is composed of a set of “lower-order dependency flag 1” and “relation type 1” of the relation type corresponding to the spot value. When the set flag is “1”, the set further includes a set of “upper-order dependency flag 2”, “lower-order dependency flag 2”, and “relation type 2”. Next, it is checked whether the relation type matches the first relation type in the pattern determination table and the next relation type matches the next relation type in the pattern determination table. If both relationship types match,
A spot value corresponding to the relation type is extracted from the explanation text information data, and a character string that matches the spot value is
The dependency attribute included in the dependency source clause of the dependency information is “case” and the dependency attribute corresponding to the dependency source clause is also included in the dependency destination clause. Check whether the corresponding dependency attribute is “embedded” or “body”. If a character string that matches the spot value is included in both, and both dependency attributes match, the upper dependency flag “other than 0” and the lower dependency flag “1” in the pattern determination table are set. Is determined. When a character string that matches the spot value is included only in the dependency source clause of the dependency information, it is checked whether the dependency attribute corresponding to the dependency source clause is “case”. If they match, it is determined whether the upper-order dependency flag is “0” and the lower-order dependency flag is “1” in the pattern determination table. Next, the spot value corresponding to the next relation type in the explanation text information data is extracted,
If a character string that matches the spot value is included in the dependency clause of the dependency information, it is determined whether the upper dependency flag is “1” and the lower dependency flag is “0” in the pattern determination table. I do. The pattern value that matches the combination of the above patterns becomes the text pattern of the partial text in the explanation text information data.

When the patterns do not match and when the number of relation types in the explanatory text information data is one, it is checked whether they individually match the relation types of the set flag “0” in the pattern judgment table. In the case of matching with the relation type, the spot value corresponding to the relation type is extracted, and the character string matching with the spot value is included in the dependency source clause of the dependency information and corresponds to the dependency source clause. Dependency attribute is "case" and is also included in the dependency clause,
It is checked whether the dependency attribute corresponding to the dependency destination clause is “embedded” or “body”. Included in both,
If both dependency attributes match, it is determined whether the upper dependency flag is "other than 0" and the lower dependency flag is "1" in the pattern determination table. Further, it is checked whether a character string that matches the spot value is included only in the source clause of the dependency information, and whether the dependency attribute corresponding to the dependency source clause is “case”. If they match, it is determined whether the upper-order dependency flag is “0” and the lower-order dependency flag is “1” in the pattern determination table. When the character string that matches the spot value is included only in the dependency clause of the dependency information, the upper dependency flag “1” and the lower dependency flag “0” of the pattern determination table are set. Is determined. The pattern value that matches the combination of the above patterns becomes the text pattern of the partial text in the explanation text information data.

In the explanatory phrase generation processing, a higher-order dependency flag of the pattern determination table corresponding to the relation type is determined from the pattern value obtained in the pattern determination processing. When the higher-order dependency flag is “2”, the spot value corresponding to the relation type in the explanation text information is extracted, and it is determined whether the spot value is included in the character string in the dependency destination clause of the dependency information. I do. If there is a matching character string, the phrase of the source clause corresponding to the dependency clause is extracted and set as the explanatory phrase 2. Then, using the extracted character string of the dependency source clause as a key, the dependency destination clause of the dependency information is searched again, and if a matching character string exists, the dependency source clause corresponding to the dependency destination clause is found. Is extracted and concatenated before description phrase 2 to form description phrase 2. This operation is repeated until the character strings of the destination clause no longer match. When the higher-order dependency flag of the pattern determination table corresponding to the relation type is “1”, a spot value corresponding to the relation type in the explanatory text information data is extracted, and the spot value is used as the dependency information. It is determined whether it is included in the character string in the destination clause. If there is a matching character string, the clause of the dependency source clause corresponding to the dependency destination clause is extracted and set as the explanatory phrase 1. Then, using the extracted character string of the dependency source clause as a key, a search is again made for the dependency clause of the dependency information. If a matching character string exists, the dependency source clause corresponding to the dependency source clause is searched. The clause is taken out and concatenated before the explanatory phrase 1 to obtain the explanatory phrase 1. This operation is repeated until the character strings of the destination clause no longer match. When the generation of the description phrase corresponding to the relation type is completed, the spot name of the column of the description text information data corresponding to the relation type, description phrase 1, description phrase 2
To form an explanatory phrase DB9. Here, explanation phrase data is created by three sets of {spot name, pattern value, explanation phrase 1, explanation phrase 2}, and is constructed as an explanation phrase DB9. However, if multiple explanation phrases 1 or 2 are created,
The explanation phrase data is created in accordance with the larger one.

As described above, it is possible to output the explanation phrase DB9 necessary for explaining the spot name from the set of the spot name and the explanation text in the content DB5.

Next, as shown in FIG. 7, an example of a set of a spot name {Kotohikihama} read from the content DB 5 and a description text {Kotohikihama located in the northeastern part of Amino-cho,.
The processing of each section will be described in detail based on the example of the spot vocabulary DB 6 for the spot names shown in FIG.

The spot information extraction unit 1 inputs the spot name {Kotohikihama} in the example shown in FIG. 7 and the explanation text {Kotohikihama located in the northeastern part of Amino-cho,. Is divided (text division processing). The explanatory text is a partial text に Northeast part of Amino-cho ...｝,
A part of the beach is divided into five parts as in..., And a text number and a spot name are assigned to each part, and a list of partial text information is configured as a text information initial DB 7. FIG. 9 shows a storage example of the text information initial DB 7.

In the text information extraction processing, the partial text information is sequentially extracted from the text information initial DB 7, and the vocabulary data of the spot vocabulary DB 6 corresponding to the spot name {Kotohikihama} is extracted. The character strings in the partial text of the text number “1” that match the spot value of the spot vocabulary DB 6 are “Kotobiki Beach” and “Coast”. The text number “1” of the partial text information, the spot name “Kotohikihama”, the partial text “in the northeastern part of Amino-cho ...” and the set of two relation types and spot values included in the partial text It creates explanation text information data consisting of {spot: Kotohikihama} and {kind: coast} and composes explanation text information DB8. The same processing is performed on the remaining partial text. For the partial text information of text number "5", the spot vocabulary
Since there is no character string that matches the spot value in DB6, no explanation text information data is output. FIG. 10 shows an output example of the explanatory text information DB8.

Hereinafter, an example in which the text number "1" is processed will be described.

The phrase information extraction unit 2 is provided with an explanation text information DB
8 are extracted in order and phrase information is output. For the explanatory text information data of text number "1", "｛" in Amino-cho, "in the northeast", "located", "Kotobikihama", "total length: 1.8km",
The phrase string of “white sand blue pine”, “very”, “beautiful”, “shore” is output.

The dependency information extraction unit 3 receives the phrase string output from the spot information extraction unit 1 and outputs dependency information for each phrase. For the description text information of text information number “1”, as shown in FIG. 11, {Amino town: in the northeast: body}, {in the northeast: located: case in charge}, {position To: Kotobiki beach: embedded｝, Kotobiki beach: coast. : Body｝, {Length: 1.8km: Body}, 1.8km: White sand and blue pine: Body, ｛White sand and blue pine: Beach. : Body｝, ｛Very: Beautiful: Adverb｝,
｛Beautiful: The coast. : Embed 埋,｝ beach. : NIL: N
Outputs dependency information of IL｝｝.

The explanation phrase constructing section 4 first extracts the explanation text information data from the explanation text information DB 8 in order as the pattern judgment processing. “Spot” and “kind” are extracted as the relation types corresponding to the text number “1”. Since there are a plurality of relationship types, it is determined whether the relationship type of the column with the set flag “1” in the pattern determination table of FIG. Since the relationship type of the explanation text information data is a combination of “spot” and “kind”, the pattern values “pt1” and “pt2” can be determined as candidates. Next, a character string that matches the spot value “Kotohikihama” corresponding to the first relation type “spot” is used as the dependency source clause and the dependency destination clause of the dependency information output from the dependency information extraction unit 3. Check whether it is included and determine whether the dependency attribute corresponding to the dependency source clause is "case" and the dependency attribute corresponding to the dependency clause is "embedded" or "body" I do. The third line of the dependency information is located at: Kotobikihama: The embedding destination clause and
The fourth is Kotohiki Beach: The coast. : Exists in both of the dependency source clauses of case｝, and the dependency attribute matches “embed” and “case”. The pattern value corresponding to the text number 1 is “pt2” because the conditions of the upper dependency flag “other than 0” and the lower dependency flag “1” of the pattern determination table are matched.

The explanation phrase generation processing is based on the acquired pattern value “pt2” and the pattern value pt2 in the pattern determination table.
Since the upper-level dependency flag in the column of “2” is “2”, the spot value “Kotohikihama” corresponding to the relation type “spot” in the explanation text information data is extracted. This spot value is
It is determined whether the character string matches the character string in the dependency destination clause of the dependency information shown in FIG. Since the spot value “Kotobikihama” is included in the third dependency clause of the dependency information, the dependency source clause information “located” corresponding to this dependency clause is extracted as the explanatory phrase 2. Then, the extracted character string of the dependency source phrase “located” is used as a key, and the dependency destination phrase of the dependency information is searched again. Since the matching character string is included in the second dependency clause, the dependency source clause "northeast" corresponding to this dependency clause is connected before description phrase 2, and the description phrase 2 I do. By repeating this operation, the explanatory phrase 2 becomes “located in the northeastern part of Amino-cho”. Next, a higher order dependency flag next to the pattern value pt2 in the pattern determination table is determined. Since the higher-order dependency flag is “1”, the spot value “coast” corresponding to the next relation type in the explanation text information data is extracted. It is determined whether this spot value matches the character string in the dependency destination clause of the dependency information in FIG. Since the spot value “coast” is included in the seventh and ninth clauses of the dependency information, the dependency phrases “white sand and blue pine” and “beautiful” corresponding to the two dependency clauses are explanatory phrases 1 Take out as. 2 taken out here
The dependency destination phrase of the dependency information is searched again using the character strings of the two dependency source phrases "white sand blue pine" and "beautiful" as keys. Since “Shirasago Aomatsu no” is included in the sixth dependency clause, the dependency source clause corresponding to this dependency clause is connected before description phrase 1 to form description phrase 1. The dependency destination clause extracted from the dependency information is searched again using the extracted dependency source clause as a key.
Repeat until no phrase with a matching character string is found, and one of the explanatory phrases 1 is “1.8 km of white sand and blue pine”. Then, the same determination is made for another explanatory phrase 1 “beautiful”, and the explanatory phrase 1 becomes “very beautiful”. At the time when the description phrase for text number "1" has been generated, this time, two description phrases 1 have been created, so this is located in Kotohikihama; 1.8km in length of white sand and blue pine; northeast of Amino-cho説明 and ｛Kotohikihama; very beautiful; located in the northeastern part of Amino Town｝ are composed in the explanatory phrase DB9.
The procedure of the description phrase extracted here is shown in FIGS.
FIG. 15 shows the description DB 9 configured.

The processing including the spot information extracting unit 1, the phrase information extracting unit 2, the dependency information extracting unit 3, and the explanatory phrase constructing unit 4 shown in FIG. 1 is a knowledge acquisition program such as a floppy disk, a CD-ROM, It can be recorded on a recording medium such as a magneto-optical disk or a DVD and executed on a computer such as a personal computer.

[0037]

As described above, according to the present invention, a portion describing an object such as a sightseeing spot is extracted from an electronic text such as a tourist guide document, and a knowledge base is constructed. On the other hand, it is possible to present a phrase that can be a user's desire from the system side, derive a potential desire of the user, and construct a search system that matches the user's desire.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a knowledge acquisition device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a relationship between a spot name in a content DB 5 and an explanation text.

FIG. 3 is a diagram showing a configuration of a spot vocabulary DB6.

FIG. 4 is a diagram showing a configuration of partial text information in a text information initial DB 7;

FIG. 5 is a diagram showing a configuration of explanation text information data in an explanation text information DB8.

FIG. 6 is a diagram showing a configuration of a pattern determination table.

FIG. 7 is a diagram showing spot names and explanation texts.

FIG. 8 is a diagram showing an example of a spot vocabulary DB 6.

FIG. 9 is a diagram showing an example of a spot information initial DB 7;

FIG. 10 is a diagram showing an example of an explanation text information DB8.

FIG. 11 is a diagram illustrating an example of dependency information.

FIG. 12 is a diagram illustrating an example of a configuration procedure of an explanatory phrase.

FIG. 13 is a diagram illustrating an example of a configuration procedure of an explanatory phrase.

FIG. 14 is a diagram illustrating an example of a configuration procedure of an explanatory phrase.

FIG. 15 is a diagram showing an example of an explanatory phrase DB9.

[Description of Signs] 1 Spot Information Extraction Unit 2 Clause Information Extraction Unit 3 Dependency Information Extraction Unit 4 Explanation Phrase Composition Unit 5 Content DB 6 Spot Vocabulary DB 7 Text Information Initial DB 8 Explanation Text Information DB 9 Explanation Phrase DB

Claims (4)

[Claims] 1. A set of a spot name, which is a name of a certain thing, and a description text, which is a text describing the thing with the spot name, is inputted, and the description text is divided into sentence units, and each is divided into partial texts. A text number, a spot name, and partial text information composed of a set of partial texts are configured and stored in a text information initial database, and partial text information is sequentially extracted from the text information initial database. If a vocabulary that matches a spot value in a spot vocabulary database that holds a spot type and a relationship type that is a type of relationship between the spot value and the spot value is included in the partial text information, the spot value and the partial The description text information data is added by adding a pair of the spot name in the text information and the relation type of the spot value. A spot information extracting step of creating a description data and storing the spot text information in the explanatory text information database; and a phrase information outputting step of inputting explanatory text information data from the explanatory text information database and outputting a sequence of phrases constituting the explanatory text information data. Inputting a sequence of the phrases, extracting a dependency relationship of the phrases and outputting the dependency information, and inputting the description text information from the description text information database; When the relation type corresponding to the sentence is the same as the spot name in the explanation text information, a set of clauses whose spot values are related to each other and the clauses that are sequentially affected from that clause are extracted, and the spot (2) performing a process of retrieving a set of clauses that are adjunct forms of the predicates in the value and a set of clauses related to the clauses; When the relation type corresponding to the sentence in the text information is a word representing the type of the spot name, the process of extracting all the clauses that are connected to the predicate in the spot value and the set of the clauses related to the phrase are performed. Do
(3) When the relation type corresponding to a sentence in the explanation text information is a word representing a part of a spot name, a phrase whose spot value has a relationship with a case, and a phrase that is sequentially affected from the phrase. And a step of taking out a set of clauses that are adjuncts of predicates in spot values and a set of clauses related to the clauses. 2. A spot vocabulary database that holds a plurality of spot values that are vocabulary and a relationship type that is a type of a relationship between these spot values for each spot name that is a name of a certain thing; a text information initial database; A description text information database, a spot name, and a set of description texts, which are texts describing the thing of the spot name, are input, and the description text is divided into sentence units and each is divided into partial texts.
A text number, a spot name, and a partial text information composed of a set of partial texts are configured, stored in the text information initial database, and the partial text information is sequentially extracted from the text information initial database. If a matching vocabulary is included, the partial text information is added with the spot value and a set of a spot name in the partial text information and a relation type of the spot value to create explanatory text information data, Spot information extraction means for storing in the explanation text information database; description information information means for inputting explanation text information data from the explanation text information database and outputting a sequence of phrases constituting the explanation text information data; , And extract and output the dependency relation of the phrase. Receiving information extracting means, and inputting the explanation text information from the explanation text information database; (1) when the relation type corresponding to the sentence in the explanation text information is the same as the spot name in the explanation text information, A set of clauses whose values are in a case relation and the clauses that are sequentially affected from the clauses are extracted, and the clauses that are adjuncts of the predicates in the spot value and the clauses that are affected by the clause are extracted. (2) When the relation type corresponding to the sentence in the explanatory text information is a word representing the type of the spot name, the spot value is associated with the phrase which is a continuous form of the predicate and the phrase. Perform the process of extracting all the set of clauses received,
(3) When the relation type corresponding to a sentence in the explanation text information is a word representing a part of a spot name, a phrase whose spot value has a relationship with a case, and a phrase that is sequentially affected from the phrase. A knowledge acquisition apparatus having an explanatory phrase construction step of performing a process of extracting a set of phrases having a continuous form of a predicate in a spot value and a set of phrases related to the phrase. 3. A set of a spot name, which is the name of a certain thing, and a description text, which is a text describing the thing with the spot name, is input, and the description text is divided into sentence units to form partial texts. Partial text information consisting of a set of a number, a spot name, and a partial text is configured and stored in a text information initial database, and partial text information is sequentially extracted from the text information initial database. If the vocabulary that matches the spot value in the spot vocabulary database that holds the relationship type that is the relationship type between the value and the spot value is included in the partial text information, the spot value and the partial text The description text information data is added by adding a set of the spot name in the information and the relation type of the spot value. A spot information extraction process for creating and storing in the explanation text information database; a clause information output process for inputting explanation text information data from the explanation text information database and outputting a sequence of phrases constituting the explanation text information data; Inputting a column of the phrase, extracting a dependency relationship of the phrase and outputting the same, and inputting the description text information from the description text information database, and (1) inputting the description text information from the description text information database; When the relation type corresponding to the sentence is the same as the spot name in the explanation text information, a set of clauses whose spot values are related to each other and the clauses which are sequentially affected from the clause are taken out, and the spot value is extracted. Performs a process of extracting a set of clauses that are adjunct forms of the predicate and the clauses related to the clause, and (2) When the relation type corresponding to the sentence in the text information is a word representing the type of the spot name, the process of extracting all the clauses that are connected to the predicate in the spot value and the set of the clauses related to the phrase are performed. Do
(3) When the relation type corresponding to a sentence in the explanation text information is a word representing a part of a spot name, a phrase whose spot value has a relationship with a case, and a phrase that is sequentially affected from the phrase. Is a knowledge acquisition program that causes a computer to execute an explanation phrase construction process of extracting a set of phrases having a continuous form of a predicate in a spot value and a set of phrases related to the phrase. 4. A recording medium on which the knowledge acquisition program according to claim 3 is recorded.