JP2015026281A - Bilingual dictionary generation device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a bilingual dictionary intended for a wide range of vocabularies.SOLUTION: A template attribute set extraction section 20 extracts a set of attribute names included in an article template of a first language, and extracts a set of attribute names included in an article template of a second language. A template quotation article set extraction section 22 extracts a set of articles described by using the article template of the first language, and extracts a set of articles described by using the article template of the second language. An association section 3 associates the attribute names of the first language with the attribute names of the second language, and associates an attribute value of the first language with an attribute value of the second language. A bilingual dictionary generation section 50 generates respective pairs of character strings of attribute names of the first language and character strings of attribute names of the second language which are associated, and respective pairs of character strings of attribute values of the first language and character strings of attribute values of the second language which are associated as a bilingual pair.

Description

  The present invention relates to a bilingual dictionary generation device, method, and program, and more particularly, to a bilingual dictionary generation device, method, and program for generating a bilingual dictionary.

  There are known studies that extract bilingual relations and dictionaries from the titles of article pairs linked by interlingual links using Wikipedia (R) as an information source (Non-Patent Documents 1 and 2).

  In addition, there is a known research for extracting a bilingual relationship / dictionary from a corpus data of a different language called a parallel corpus or comparable corpus by a statistical method or the like (Non-Patent Documents 3 and 4).

Arai and three others, “Translation Extraction for Multilingual Blog Search Using Wikipedia”, Proc. 5J-4, 70th Annual Conference of Information Processing Society of Japan, 2008 Eight Sato et al., “Association of related news and blogs via Wikipedia”, IPSJ Research Report Natural Language Processing Study Group 2009-NL-194 (10), 2009 Gamallo, P., `` Extraction of Translation Equivalents from Parallel Corpora Using Sense-sensitive Contexts '', Proc.EAMT2005, 2005, p.97-102 梶 One other person, "Translation selection using comparable corpus. 4th Machine Translation Technology Innovation Symposium", 2010, Internet <http://www.congre.co.jp/imttsympo/2010/program/pdf/ p5_kaji.pdf>

  However, many of the prior arts of Non-Patent Documents 1 and 2 shown above extract a bilingual dictionary from a title pair of an article in which correspondence between languages is specified in advance by an inter-language link or the like. The range of possible translations is limited.

  Further, since the conventional techniques of Non-Patent Documents 3 and 4 extract the parallel translation from the general text description portion, it is difficult to obtain additional information regarding the applicable area of the parallel translation.

  The present invention has been made in view of the above circumstances, and an object thereof is to provide a bilingual dictionary generating apparatus, method, and program capable of generating a bilingual dictionary in which bilingual pairs are extracted from a wide range of parts.

  In order to achieve the above object, a bilingual dictionary generating apparatus according to the present invention includes a bilingual dictionary storing a bilingual pair that is a combination of a character string of a first language and a character string of a second language. A bilingual dictionary generating device for generating an article template in which attribute names related to the entity are listed for describing an article about the entity to be described, and corresponding to the type of the entity to be described Based on the article template in the first language and the article template in the second language, a set of attribute names included in the article template in the first language is extracted and included in the article template in the second language. A template attribute set extraction unit for extracting a set of attribute names and a set of articles described using the article template in the first language are extracted, and the article in the second language is extracted. A template-cited article set extraction unit that extracts a set of articles described using a template, a set of attribute names of the first language extracted by the template attribute set extraction unit, and a second language In the set of attribute names, the attribute name of the first language is associated with the attribute name of the second language, the set of articles in the first language extracted by the template-cited article set extraction unit, and the A set of attribute values for the attribute name of the associated first language extracted from the set of articles in the first language based on a set of articles in the second language; and the second The attribute value of the first language is associated with the attribute value of the second language in the attribute value set for the attribute name of the associated second language extracted from the set of articles in the language versus A pair of an attribute name character string of the first language and a character string of the attribute name of the second language associated by the associating unit, and the first language associated with the first language A bilingual dictionary generating unit that generates the bilingual dictionary storing each pair of the attribute value character string of the second language and the attribute value character string of the second language as a bilingual pair. Yes.

A bilingual dictionary generating method according to the present invention includes a template attribute set extracting unit, a template cited article set extracting unit, an associating unit, and a bilingual dictionary generating unit. A bilingual dictionary generating method in a bilingual dictionary generating device for generating a bilingual dictionary storing bilingual pairs that are combinations with character strings in a second language, wherein the template attribute set extracting unit reports articles about entities to be described. Is an article template in which attribute names related to the entity are listed, and the article template of the first language and the article template of the second language corresponding to the type of the entity to be described are included in the article template. Based on this, a set of attribute names included in the article template of the first language is extracted, and the attributes included in the article template of the second language are extracted. Extracting a set of articles, and extracting a set of articles described using the article template of the first language by the template-cited article set extraction unit, and using the article template of the second language A step of extracting a set of described articles, and a set of attribute names of the first language and a set of attribute names of the second language extracted by the template attribute set extraction unit by the association unit The set of articles in the first language and the second language extracted by the template-cited article set extraction unit in association with the attribute name of the first language and the attribute name of the second language A set of attribute values for the associated attribute name of the first language, extracted from the set of articles in the first language based on the set of articles In the set of attribute values for the attribute name of the associated second language extracted from the set of articles in the language, the attribute value of the first language and the attribute value of the second language are A step of associating with each of a pair of a character string of the attribute name of the first language and a character string of the attribute name of the second language associated by the correspondence unit by the bilingual dictionary generation unit; and Generating the bilingual dictionary storing each pair of the character string of the attribute value of the associated first language and the character string of the attribute value of the second language as a bilingual pair. .

A program according to the present invention is a program for generating a bilingual dictionary storing a bilingual pair that is a combination of a character string in a first language and a character string in a second language, which is a bilingual translation. An article template in which attribute names related to the entity are listed for describing an article about the entity to be described, the article template in the first language corresponding to the type of the entity to be described, and the A template attribute for extracting a set of attribute names included in the first language article template and extracting a set of attribute names included in the second language article template based on the second language article template Set extraction unit,
A template-cited article set extraction unit that extracts a set of articles described using the article template of the first language and extracts a set of articles described using the article template of the second language; In the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit, the attribute names of the first language and the attribute names of the second language Are extracted from the set of articles in the first language based on the set of articles in the first language and the set of articles in the second language extracted by the template-cited article set extraction unit. The attribute name of the associated second language extracted from the set of attribute values for the attribute name of the associated first language and the set of articles of the second language In In the attribute value set, the association unit associating the attribute value of the first language and the attribute value of the second language, and the attribute name of the first language associated by the association unit Each of a pair of a character string and a character string of the attribute name of the second language, and an associated character string of the attribute value of the first language and a character string of the attribute value of the second language It is a program for causing each pair to function as a bilingual dictionary generation unit that generates the bilingual dictionary stored as a bilingual pair.

  The association unit according to the present invention is characterized in that the attribute of the first language in the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit. For each pair of a name and an attribute name of the second language, the similarity of the pair is calculated, and based on the similarity calculated for each pair, the attribute name of the first language and the second A set of attribute values for the associated attribute name of the first language, extracted from the set of articles in the first language, and the attribute name of the second language For each pair of attribute value of the first language and attribute value of the second language in the attribute value set for the associated attribute name of the second language extracted from the set of articles , Calculate the similarity of the pair, The attribute value of the first language and the attribute value of the second language are associated with each other based on the similarity calculated, and the bilingual dictionary generation unit is associated with the association unit Each of the character string of the attribute name of the first language and the character string of the attribute name of the second language, and the character string of the attribute value of the first language and the second language associated with each other Each pair of the attribute value and the character string may be stored in the bilingual dictionary as a bilingual pair together with the similarity calculated for the pair.

  Further, the association unit according to the present invention includes: the first language in the set of attribute names in the first language and the set of attribute names in the second language extracted by the template attribute set extraction unit. For each pair of the attribute name and the attribute name of the second language, the pair similarity is calculated, and based on the similarity calculated for each pair and a predetermined first threshold, An attribute value for the associated attribute name of the first language extracted from the set of articles of the first language by associating the attribute name of the first language with the attribute name of the second language And a set of attribute values for the associated attribute name of the second language extracted from the set of articles in the second language and the attribute value of the first language and the second For each pair with a language attribute value of The pair similarity is calculated, and the attribute value of the first language is associated with the attribute value of the second language based on the similarity calculated for each pair and a predetermined second threshold value. Can be.

  Further, the associating unit according to the present invention is configured so that the attribute names included in the attribute name set of the first language extracted by the template attribute set extracting unit are in the predetermined order in the first order. The attribute name of the first language is the processing target, the attribute name of the first language to be processed is associated with the attribute name of the second language, and the attribute name of the first language to be processed The attribute value of the second language is repeatedly associated with the attribute value of the second language for each attribute name of the first language to be processed, and the attribute name of the first language to be processed is Each time the association is performed, the first threshold value and the second threshold value can be reduced.

  Further, the association unit according to the present invention includes a set of attribute values for the attribute name of the first language to be processed from the set of articles in the first language extracted by the template cited article set extraction unit. A first attribute value instance set extracting unit for extracting the first attribute value instance set, and a set of attribute names in the second language extracted by the template attribute set extracting unit based on the similarity of the attribute names. A corresponding attribute candidate extraction unit that extracts a set of attribute name candidates of the second language corresponding to attribute names of the second language, and the second language included in the set of candidate attribute names of the second language For each attribute name, a set of attribute values for the attribute name of the second language is extracted from the set of articles in the second language extracted by the template-cited article set extraction unit in the second language. A second attribute value instance set extraction unit; and for each attribute name of the second language included in the set of candidate attribute names of the second language, the second attribute value for the attribute name of the second language A set of attribute values for the attribute name of the second language extracted by the instance set extraction unit, and an attribute value for the attribute name of the first language to be processed extracted by the first attribute value instance set extraction unit An attribute value instance set similarity calculation unit that calculates a similarity between attribute value instance sets that is a similarity between the second language attribute set and the second language attribute name candidate set. For each language attribute name, if the similarity between attribute value instance sets calculated by the attribute value instance set similarity calculation unit is greater than or equal to the first threshold, the processing target The attribute name of the first language and the attribute name of the second language are associated with each other, and the associated attribute name of the first language and the attribute name of the second language are associated with each other. For each pair of the attribute value of the first language and the attribute value of the second language in the set of attribute values, if the pair similarity is not less than the second threshold value, the first of the pair An association determining unit that associates the attribute value of the language with the attribute value of the second language can be included.

  As described above, according to the bilingual dictionary generation device, method, and program of the present invention, the bilingual dictionary generating device includes the first language article template based on the first language article template and the second language article template. A set of attribute names and a set of attribute names included in the article template of the second language, a set of articles described using the article template of the first language, and the second language The set of articles described using the article template is extracted, and the attribute name of the first language and the second set of attribute names of the first language and the set of attribute names of the second language are extracted. In the set of attribute values for the attribute name of the associated first language and the set of attribute values for the attribute name of the associated second language, Attribute value and second word Each of the pair of the attribute name character string of the first language and the character string of the attribute name of the second language, and the attribute of the first language associated with each other Generating a bilingual dictionary in which bilingual pairs are extracted from a wide range by generating bilingual dictionaries in which each pair of a value string and a second language attribute value string is stored as a bilingual pair The effect of being able to be obtained.

It is a figure which shows the conceptual diagram of the information structure which embodiment of this invention makes object. It is the schematic which shows the structure of the bilingual dictionary production | generation apparatus concerning embodiment of this invention. It is a flowchart which shows the content of the first half part of the bilingual dictionary production | generation processing routine in the bilingual dictionary production | generation apparatus which concerns on embodiment of this invention. It is a flowchart which shows the content of the second half part of the bilingual dictionary production | generation processing routine in the bilingual dictionary production | generation apparatus which concerns on embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<Overview>
For an entity of a certain type τ, in order to describe attributes and attribute values for entities with that type, an article template prepared in advance for each language is used to create an article for a specific entity ε. Is described in various languages, article template ID: S of source language X and article template ID: T of target language Y prepared for the type τ are input, and from these article templates Correspondence between attribute names is performed, and further, by correlating attribute values from the article group that cites the template specified by these article template IDs, the bilingual dictionary is obtained from the correspondence information. Extract.

  Here, the article template is a framework defined in advance for describing an entity (eg, Mt. Fuji / Mt. Fuji) of a specific type τ (eg, mountain / mountain). In the article template, an attribute name (attribute name) (for example, name, altitude, name, altitude) for describing an entity of type τ in the target language is presented. The source code of the article template specified by the given article template ID can be acquired / analyzed by a separate means, and thus the above attribute name can be easily extracted. However, even for article templates of the same type, the article template is created for each language, and attribute names included in the article template for each language are not necessarily translated.

FIG. 1 is a conceptual diagram illustrating the information structure targeted by this embodiment.
An article that describes a type of entity is described by quoting an article template that reflects that type of entity. The article template lists attributes for describing an entity of the type. Also, each article describes an attribute value that gives information on each attribute for the entity targeted by the article.

  According to the method of the present embodiment, in the example of FIG. 1 above, parallel translations such as “name: name” and “height: elevation” can be extracted from the correspondence of attribute names, and “Mt. Fuji: Mt.Fuji, 3,776m: 12,388ft "can be extracted.

<System configuration>
The bilingual dictionary generation device 100 according to the first exemplary embodiment of the present invention includes a character string (word) in a source language (first language) and a character string (word) in a target language (second language). Create a bilingual dictionary. This bilingual dictionary generation device 100 is constituted by a computer including a CPU, a RAM, and a ROM storing a program for executing a bilingual dictionary generation processing routine to be described later, and functionally configured as shown below. Has been. As shown in FIG. 2, the bilingual dictionary generation device 100 includes an input unit 1, a calculation unit 2, and an output unit 4.

  The input unit 1 accepts an input of a pair of an article template ID: S of the source language X and an article template ID: T of the target language Y corresponding to the type of entity to be described. Note that the article template ID: S of the source language X and the article template ID: T of the target language Y do not necessarily have the same type of entity to be described. For example, the type of article template ID: S in the source language X may include the type of article template ID: T in the target language Y.

  The calculation unit 2 is a character string that is a translation of the source language X and the target language Y based on the article template ID: S of the source language X received by the input unit 1 and the article template ID: T of the target language Y. A bilingual dictionary storing pairs is generated.

  The calculation unit 2 includes an association unit 3, a template attribute set extraction unit 20, a template cited article set extraction unit 22, a character string translation unit 24, a character string similarity calculation unit 26, and an attribute name similarity calculation unit. 28 and a bilingual dictionary generation unit 50.

The template attribute set extraction unit 20 extracts a set of attribute names included in the source language article template based on the source language X article template ID and the target language Y article template ID received by the input unit 1. Extract a set of attribute names included in a language article template.
Specifically, the template attribute set extraction unit 20 acquires / analyzes the source code of the article template with the article template ID received by the input unit 1 via the Internet 5, and sets each attribute name included in the article template. An attribute ID is assigned to the attribute ID, and a set of these attribute IDs is extracted. Here, the attribute ID is an ID for identifying a set of attribute names (eg, {name, name, common name}) representing attributes having the same meaning. That is, for example, the same attribute ID is assigned to a set of attribute names {name, name, common name}. Whether or not the attribute names have the same meaning may be determined by using a conventionally known synonym determination technique, and thus the description thereof is omitted.
Further, in this process, the template attribute set extraction unit 20 generates an attribute ID table that is an internal table that holds the correspondence between the attribute ID and the attribute name character string set, and stores the attribute ID table in an internal table database 30 described later. The template attribute set extraction unit 20 extracts the attribute ID set α for the article template ID: S received by the input unit 1 and the attribute ID set β for the article template ID: T. Here, the attribute ID set α = {α ₁ , α ₂ ,..., Α _M }, and the attribute ID set β = {β ₁ , β ₂ ,.

The template-cited article set extraction unit 22 extracts a set of articles described using the article template with the article template ID of the source language X received by the input unit 1 and the article of the target language Y received by the input unit 1 A set of articles described using an article template with a template ID is extracted.
Specifically, based on the article template ID received by the input unit 1, the template cited article set extraction unit 22 obtains a group of articles quoting the article template corresponding to the article template ID, and the articles An article ID set is extracted. Here, each article has an article ID. It is assumed that the article template ID of the article template that it cites is specified in each article. Therefore, for example, by performing a search using the template ID as a key for the article group to be processed, it is possible to extract the article ID set of the article group that cites the designated article template.

  The character string translation unit 24 translates the designated source language X character string into the designated target language Y character string. The string translation unit 24 uses existing technologies and services (Internet 1 <http://translate.google.co.jp/?hl=en&tab=wT>, Internet 2 <http://langrid.org/tools/ (See toolbox /> etc.), and details thereof will not be described in this embodiment.

  The character string similarity calculation unit 26 calculates the similarity between the designated character string 1 and character string 2. Details of the character string similarity calculation unit 26 will not be described. For example, an edit distance (see, for example, the Internet <http://en.wikipedia.org/wiki/Edit_distance>) or Jaro-Winkler This can be realized by using a known technique such as distance (see, for example, the Internet <http://en.wikipedia.org/wiki/Jaro%E2%80%93Winkler_distance>).

  The attribute name similarity calculation unit 28 calculates the similarity between the attribute name of the source language X and the attribute name of the target language Y. Here, the attribute name similarity is a numerical value representing the cross-language similarity between the following two character strings.

(1) An attribute name character string that is an element of an attribute name set represented by a certain attribute ID in source language X. (2) An attribute name that is an element in an attribute name set represented by a certain attribute ID in target language Y. String

More specifically, the attribute name similarity calculation unit 28 determines the attribute ID set a selected by the corresponding attribute candidate extraction unit 36 (to be described later). Each element β _i ) of β is input, and the similarity between attribute name sets belonging to each attribute ID is calculated. In the present embodiment, the attribute name similarity is calculated as follows.

Each attribute name character string that is an element of the attribute name character string set corresponding to the attribute ID: a of the source language X is translated into a character string of the target language Y by the character string translation unit 24 and translated into the target language Y get the attribute string set a ^T. The attribute ID obtained from the attribute ID table: seek the union of the attribute string set a ^T in which the attribute name string set a 'correspondence has been translated into the target language Y in a. Here, since the union is obtained, it is not an empty set. The reason for obtaining the union is to assume a case where the source language character string is used as it is in the target language. For all combinations of the elements of this union and each element of the attribute character string set b ′ obtained from the attribute ID table for the attribute ID: b in the target language Y, the character string similarity calculating unit 26 The similarity of the character string of the combination is calculated. Then, the maximum similarity among the similarities obtained for all combinations is set as the attribute name similarity for the attribute ID: a and the attribute ID: b.

In the attribute ID set α of the source language X and the attribute ID set β of the target language Y extracted by the template attribute set extracting unit 20, the associating unit 3 determines the attribute name of the source language X and the attribute name of the target language Y. Corresponding source data is extracted from the source language X article set based on the source language X article set and the target language Y article set extracted by the template-cited article set extraction unit 22. In the set of attribute values for the attribute name of the target language Y extracted from the set of attribute values for the attribute name of language X and the set of articles of the target language Y, the attribute value of the source language X and the target language Corresponds to the Y attribute value.
In addition, the associating unit 3 includes an internal table database 30, an attribute ID set sorting unit 32, a first attribute value instance set extracting unit 34, a corresponding attribute candidate extracting unit 36, and a second attribute value instance set extracting unit 38. And an attribute value instance set similarity calculation unit 40, an association determination unit 42, and an iterative determination unit 44.

  The internal table database 30 stores an attribute ID table generated by the template attribute set extraction unit 20. The internal table database 30 stores an attribute name similarity table and an attribute value instance similarity table. Here, the attribute ID table is

<Attribute ID, attribute name string set>

It consists of two sets of information. The attribute name similarity table

<Attribute ID of source language X, attribute ID of target language Y, attribute name character string of source language X, attribute name character string of target language Y, attribute name similarity, translation determination flag>

It consists of six sets of information. The translation determination flag is used to determine whether or not to use the attribute name character string of the attribute ID of the source language X and the attribute name of the attribute ID of the source language Y as a translation pair. “True” is displayed when it is selected, and “False” is displayed when it is determined that it is not a translated word pair.

  In addition, attribute value instance similarity table

<Attribute ID of source language X, attribute ID of target language Y, article ID of source language X, article ID of target language Y, source language X attribute value string, target language Y attribute value string, attribute value similarity Degree, translation determination flag>

It consists of eight sets of information.

  The attribute value similarity is a numerical value representing the cross-language similarity between the following two character strings.

(1) An attribute value character string that appears in a certain article for any attribute name of an attribute name set represented by a certain attribute ID in the source language X. (2) An attribute represented by a certain attribute ID in the target language Y. Attribute value string that appears in an article for any attribute name in the name set

The attribute ID set sorting unit 32 applies the elements (= {α ₁ , α ₂ ,..., Α _M }) of the attribute ID set α to the attribute ID set α extracted by the template attribute set extracting unit 20. The attribute IDs are sorted in descending order of priority, and the resulting sorted attribute ID set α is newly generated. In the present embodiment, details of the criteria regarding the priority of the attribute ID will not be described. However, for example, it is conceivable that the priority is determined according to the number of articles including the attribute corresponding to the attribute ID.

The first attribute value instance set extraction unit 34 extracts a set of attribute values for the attribute name of the source language X to be processed from the set of articles in the source language X extracted by the template cited article set extraction unit 22.
Specifically, the first attribute value instance set extraction unit 34 extracts the article ID and the processing target source language X from each article in the target language X article ID set generated by the template cited article set extraction unit 22. An attribute value instance set V = {v ₁ , v ₂ ,..., Which is a set of pairs of attribute names and attribute value character strings is extracted. More specifically, the first attribute value instance set extraction unit 34 acquires / analyzes the source code of an article via the Internet 5 and when the attribute represented by the specified attribute ID is included in the article, A pair of the attribute value character string given to the attribute and the article ID is extracted.

The corresponding attribute candidate extraction unit 36 selects the object corresponding to the attribute name of the target language X to be processed from the attribute name set of the target language Y extracted by the template attribute set extraction unit 20 based on the similarity of the attribute name. A candidate set of attribute names for language Y is extracted.
Specifically, the corresponding attribute candidate extraction unit 36 extracts the attribute ID: a of the source language X located at the head of the attribute ID set generated by the attribute ID set sorting unit 32 and the template attribute set extraction unit 20. Based on the attribute ID set β of the target language Y, the candidate for the inter-language association of the attribute ID: a is selected from the elements ({β ₁ , β ₂ ,...}) Of the attribute ID set β. Are extracted, and an attribute ID set β ′ (= {β ′ ₁ , β ′ ₂ ,...) Composed of these matching candidates is extracted. For example, the corresponding attribute candidate extraction unit 36 uses the attribute name similarity calculation unit 28 to calculate the attribute name similarity between the attribute ID: a and each element β _i of the attribute ID set β, and the attribute name similarity An element β _i having a value greater than a predetermined threshold θ ₀ is selected as an association candidate. Further, the obtained attribute name similarity is stored in the attribute name similarity table stored in the internal table database 30 together with the attribute name character strings of attribute ID: a, attribute ID: β _i , source language X, and target language Y. Record. Note that the value of the translated word determination flag is set to False.

The second attribute value instance set extraction unit 38 extracts articles of the target language Y extracted by the template-cited article set extraction unit 22 for each attribute name of the target language Y included in the candidate name candidate set of the target language Y. A set of attribute values for the attribute name of the target language Y is extracted from the set.
Specifically, the second attribute value instance set extraction unit 38, like the first attribute value instance set extraction unit 34, each article in the article ID set of the target language Y generated by the template cited article set extraction unit 22. Then, an attribute value instance set W = {w ₁ , w ₂ ,...}, Which is a set of pairs of article IDs and attribute value character strings for attribute names of the target language Y, is extracted.

  The attribute value instance set similarity calculation unit 40 extracts a second attribute value instance set for the attribute name of the target language Y for each attribute name of the target language Y included in the set of candidate attribute names of the target language Y. Between a set of attribute values for the attribute name of the target language Y extracted by the unit 38 and a set of attribute values for the attribute name of the target language X to be processed extracted by the first attribute value instance set extraction unit 34 The similarity between attribute value instance sets, which is a similarity, is calculated.

  Here, specifically, the similarity between attribute value instance sets is a numerical value representing the cross-language similarity between the following two sets.

(1) A set of character strings (hereinafter referred to as attribute value character strings) appearing as attribute values for an attribute represented by a certain attribute ID in the source language X. (2) An attribute represented by a certain attribute ID in the target language Y. Set of attribute value strings for

Specifically, the attribute value instance set similarity calculation unit 40 includes source language X and target language Y attribute IDs: a, β ′ _i , and source language X, target language Y attribute value instance set v, Using w as an input, the similarity between these attribute value instance sets is calculated. In the present embodiment, the attribute value instance set similarity is calculated as follows.

First, the attribute value character string in each element of the attribute value instances set v of the source language X, by the character string translation unit 24 obtains the attribute value character string set v ^T translated into the target language Y. Seek the union of the target language Y in the translation attribute value set of character strings v ^T and the original language X of the attribute value corresponding to the instance set v attribute value set of strings v '. Since the union is obtained, it is not an empty set. The reason for obtaining the union is to assume a case where the source language character string is used as it is in the target language. For every combination of each element of this union and each element of the attribute value character string set w ′ of each element of the attribute value instance set w, the character string similarity calculation unit 26 performs the characters of the combination. Calculate the similarity of columns. In this process, the attribute value instance similarity table, which is an internal table, includes an attribute ID of source language X: a, an attribute ID of target language Y: β′i, an article ID of source language X, an article ID of target language Y, The attribute value character string of the source language X, the attribute value character string of the target language Y, and the obtained character string similarity are recorded as the attribute value similarity. Note that the value of the translated word determination flag is set to False.

  Here, not only all combinations, but only combinations that exist in articles that are referred to by inter-language links with each other, by calculating the string similarity, it imposes stronger restrictions and handles with high accuracy It is also possible to limit to the attachment. Here, whether or not an article pair of the source language X and the target language Y is referred to by an inter-language link can be separately determined based on the article ID of each language.

  The maximum attribute value similarity among the similarities obtained for all combinations is defined as the similarity between attribute value instance sets for the attribute value instance set v and the attribute value instance set w.

The association determination unit 42 calculates the similarity between attribute value instance sets calculated by the attribute value instance set similarity calculation unit 40 for each attribute name of the target language included in the set of candidate attribute names of the target language Y. Is equal to or greater than the threshold θ ₁ , it is determined to associate the attribute name of the target language X to be processed with the attribute name of the target language Y, and the translation determination flag of the corresponding entry in the attribute name similarity table is set to True. Change to In addition, for each pair of the attribute value of the source language and the attribute value of the target language between the attribute value set for each of the attribute name of the target language X to be processed and the attribute name of the target language, if the attribute value similarity of the pair threshold theta ₂ or more, translation entries decides to associate the attribute value of the attribute value and the target language of the source language of the pair, corresponding in the attribute value instances similarity table Change the decision flag to True.

The iterative determination unit 44 sets the sum of the preset threshold value N for the association condition relaxation and the number M of elements in the ID set α of the source language X as the number of repetitions, and the processing of the association unit 3 is repeated N + M times. It is determined whether or not it has been done. If it is determined that the process of the associating unit 3 has not been repeated N + M times, the threshold θ ₁ and the threshold θ ₂ are reduced, and then the process of the associating unit 3 is repeated.

The bilingual dictionary generation unit 50 stores each of a pair of a source language attribute name character string and a target language attribute name character string associated by the associating unit 3 and an associated source language attribute value. A bilingual dictionary storing each pair of a character string and a character string of a target language attribute value as a bilingual pair is generated.
Specifically, the bilingual dictionary generation unit 50 is obtained by each process of the association unit 3 and is stored in the attribute name similarity table and the attribute value instance similarity table stored in the internal table database 30. The bilingual dictionary is generated by converting the entry whose translation word determination flag is True into the bilingual dictionary table format that is a set of the following seven sets and integrating (merging) these.

<Template ID (= S) of source language X, attribute ID of source language X, character string of source language X, template ID of target language Y (= T), attribute ID of target language Y, character string of target language Y , Translation compatibility>

More specifically, the bilingual dictionary generation unit 50 generates a bilingual dictionary by the following processing.
(1) Convert attribute name similarity table into bilingual dictionary format Extract entries whose translation word determination flag is True from the entries in the attribute name similarity table and store the source language X and the target language Y in the entry. The attribute ID is copied to the attribute of the source language X and the target language Y of the bilingual dictionary table. Also, the attribute name character strings of the source language X and the target language Y are copied to the character strings of the languages X and Y of the bilingual dictionary, respectively. Further, the attribute name similarity is copied to the translation correspondence in the bilingual dictionary table.
(2) Converting the attribute value instance similarity table into the bilingual dictionary format The entry whose translation determination flag is True is extracted from the attribute value instance similarity table, and the source language X and the target language Y of the entry are extracted. The attribute ID is copied to the attribute of the source language X and the target language Y of the bilingual dictionary. Further, the attribute value character strings of the source language X and the target language Y are copied to the character strings of the source language X and the target language Y of the bilingual dictionary, respectively. Further, the attribute value similarity is copied to the translation correspondence of the bilingual dictionary.

  The output unit 4 outputs the bilingual dictionary generated by the bilingual dictionary generating unit 50 as a result.

<Operation of the bilingual dictionary generator>
Next, the operation of the bilingual dictionary generation device 100 according to the present embodiment will be described. First, when a pair of an article template ID of the source language X and an article template ID of the target language Y corresponding to the type of entity to be described is input to the bilingual dictionary generating apparatus 100, the bilingual dictionary generating apparatus 100 Thus, the bilingual dictionary generation processing routine shown in FIG. 3 is executed.

  First, in step S100, the input unit 1 receives a pair of an article template ID in the source language X and an article template ID in the target language Y.

  Next, in step S102, based on the article template with the article template ID of the source language X and the article template with the article template ID of the target language Y received by the template attribute set extraction unit 20 in step S100, the source language X The attribute ID set α included in the article template is extracted, and the attribute ID set β included in the article template of the target language Y is extracted.

  In step S104, the attribute ID set sorting unit 32 sorts the attribute ID set α extracted in step S102 in descending order of the priority of the attribute IDs that are elements of the attribute ID set α. A sorted attribute ID set α is newly generated.

  In step S106, 0 is substituted for the number of repetitions loop_count.

  In step S108, the first attribute ID: a is extracted from the sorted attribute ID set α generated in step S104, set as a processing target, and the first attribute ID: a is extracted from the attribute ID set α. The removed items are again referred to as an attribute ID set α.

  In step S110, the template citation article set extraction unit 22 extracts a set of articles described using the article template with the article template ID of the input source language X, and the article template DI of the input target language Y is input. The set of articles described using the article template is extracted.

In step S112, for the attribute ID a to be processed, the attribute of the source language X to be processed from the set of articles in the source language X extracted in step S110 by the first attribute value instance set extraction unit 34. Attribute value instance set V = {v ₁ , v ₂ ,.

In step S113, the attribute name similarity calculation unit 28 calculates the attribute name similarity between the attribute name of the processing target attribute ID: a and each element β _i of the attribute ID set β extracted in step S102. Calculate each.

In step S114, the corresponding attribute candidate extracting unit 36 converts the attribute name set β extracted in step S102 into the attribute name of the processing target source language X based on the attribute name similarity calculated in step S113. Corresponding attribute name candidate set β ′ = {β ′ ₁ , β ′ ₂ ,...} Is extracted.

In step S116, the second attribute value instance set extraction unit 38 extracts elements of the corresponding attribute name candidate set β ′ = {β ′ ₁ , β ′ ₂ ,... For β ′ _i , the attribute value instance set W = {w ₁ , w ₂ ,... of the target language Y for the attribute name of the element β ′ _i from the set of articles in the target language Y extracted in step S110. } Is extracted.

In step S118, the attribute value instance set similarity calculation unit 40 extracts the attribute value instance set W = {w ₁ , w ₂ ,...} For the target language Y extracted in step S116, and in step S112. An attribute value instance set similarity sim _i , which is a similarity between the extracted attribute value instance set V = {v ₁ , v ₂ ,...} For the attribute of the processing target source language X, is calculated.

In step S119, for all elements included in the corresponding attribute name candidate set β ′ = {β ′ ₁ , β ′ ₂ ,... Whether or not is executed. If there is an element β ′ _i that has not been subjected to the processes in steps S116 and S118, the process returns to step S116, and the processes in steps S116 and S118 are performed on the β ′ _i . On the other hand, when the processes in steps S116 and S118 are executed for all elements included in the corresponding attribute name candidate set β ′ = {β ′ ₁ , β ′ ₂ ,...}, The process proceeds to step S120. To do.
In step S120, among the attribute value instance set similarity sim _i calculated in step S118, the element β ′ _i that gives the maximum attribute value instance set similarity sim _i is set to β *.

In step S122, among the attribute value instance set similarity sim _i calculated in step S118, the maximum attribute value instance set similarity sim _i is set to sim *.

In step S124, sim * set at step S122 is, whether greater than the threshold theta ₁ predetermined judges. If sim * is larger than the threshold θ _1, it is determined to associate the attribute name of the processing target attribute ID: a with the attribute name of the attribute ID: β * obtained in step S120. The process proceeds to step S126. On the other hand, sim * is, in the case of the threshold theta ₁ below, the process proceeds to step S132.

  In step S126, in the attribute name similarity table stored in the internal table database 30 by the association determination unit 42, the attribute ID of the source language X matches the attribute ID: a to be processed, and the attribute of the target language Y For the entry whose ID matches the attribute ID: β * obtained in step S120, the translated word determination flag is changed to True.

In step S128, the association determining unit 42 matches the attribute ID of the source language X with the attribute ID: a to be processed in the attribute value instance similarity table stored in the internal table database 30, and the target language Y. attribute ID is attribute ID obtained in step S120: among the entry that matches the beta *, attribute value similarity, for each of the threshold theta ₂ is greater than the entry, changes the translation decision flag to True.

  In step S130, the set β obtained by removing β * obtained in step S120 from the set β extracted in step S102 is referred to as a set β.

  In step S132, the processing target attribute ID: a is added as the last element of the set α.

  In step S134, the iteration determination unit 44 increments the iteration count loop_count.

  In step S136, the repetition determination unit 44 determines whether or not the processing in steps S108 to S134 has been repeated N + M times. If the processes of step S108 to step S134 are repeated N + M times or more, the process proceeds to step S140. On the other hand, when the process of step S108 to step S134 is repeated less than N + M times, the process proceeds to step S138.

In step S138, a threshold theta _1, reduces the threshold theta _2. Specifically, θ ₁ is multiplied by an attenuation coefficient ω ₁ , and θ ₂ is multiplied by an attenuation coefficient ω ₂ . Here, 0 <ω ₁ ≦ 1 and 0 <ω ₂ ≦ 1.

  In step S140, the bilingual dictionary generation unit 50 makes an entry in the attribute name similarity table in which the translated word determination flag is set to “True” in step S126, and an attribute value in which the translated word determination flag is set to “True” in step S128. A bilingual dictionary is generated by merging the entries in the instance similarity table.

  In step S142, the bilingual dictionary generated in step S140 is output as a result, and the bilingual dictionary generation processing routine is terminated.

  As described above, according to the bilingual dictionary generation device according to the present embodiment, based on the source language X article template and the target language Y article template, a set of attribute names included in the source language article template A set of attribute names included in the article template of the target language Y, and a set of articles described using the article template of the source language X and an article template of the target language Y And a set of attribute names in the source language X and an attribute name in the target language Y are associated with each other in association with the attribute name in the source language X and the attribute name in the target language Y. In the set of attribute values for the attribute name of the source language X and the set of attribute values for the attribute name of the associated target language Y, the attribute value of the source language X and the attribute value of the target language Y are associated with each other Bilingual dictionary extracted from a wide range by generating a bilingual dictionary that stores each pair of attribute name character strings and associated attribute value character string pairs as a bilingual pair A bilingual dictionary storing pairs can be generated.

  In addition, since the bilingual dictionary is generated based on the correspondence between the attribute names obtained from the article template and the correspondence between the attribute values obtained from the article quoting the article template, the title of the article associated between the languages. It is possible to extract more bilingual entries (vocabulary) than bilingual extraction from the correspondence of.

  The bilingual dictionary entry obtained in the present embodiment is linked to a template for extracting a bilingual translation, and the template reflects the type of entity to be described (for example, mountain / mountain). Since this entity type imposes restrictions on the application area of the bilingual translation (eg, the translation of the mountain “height” is “elevation”), a bilingual dictionary is created to include additional information about the applicable area of the bilingual translation. be able to.

  In this embodiment, a pair of source language X template and target language Y template is input, but it is repeatedly executed for elements of a set of template pairs that are known to be in a correspondence relationship. By doing so, more bilingual dictionary entries can be obtained.

  In addition, since the source language X and the target language Y in the present embodiment are not limited to a specific language, by applying it to a set of template pairs in other types of language pairs, Multilingual bilingual dictionaries can be obtained in the form of sets.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, the attribute name similarity calculation unit 28 may calculate the attribute name similarity by another similarity calculation method. Further, the attribute value instance set similarity calculation unit 40 may calculate the attribute value instance set similarity by another similarity calculation method.

  The internal table database 30 may be provided outside and connected to the bilingual dictionary generation device via a network.

  The bilingual dictionary generating apparatus described above has a computer system inside, but the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.

DESCRIPTION OF SYMBOLS 1 Input part 2 Operation part 3 Correlation part 4 Output part 5 Internet 20 Template attribute set extraction part 22 Template cited article set extraction part 24 Character string translation part 26 Character string similarity calculation part 28 Attribute name similarity calculation part 30 Internal table Database 32 Attribute ID set sort unit 34 First attribute value instance set extraction unit 36 Corresponding attribute candidate extraction unit 38 Second attribute value instance set extraction unit 40 Attribute value instance set similarity calculation unit 42 Association determination unit 44 Iterative determination unit 50 Bilingual Dictionary Generation Unit 100 Bilingual Dictionary Generation Device

Claims (7)

A bilingual dictionary generating device for generating a bilingual dictionary storing a bilingual pair that is a combination of a character string of a first language to be translated and a character string of a second language,
An article template in which attribute names related to the entity are listed for describing an article about the entity to be described, the article template of the first language corresponding to the type of the entity to be described, and the first A template attribute set for extracting a set of attribute names included in the article template of the first language based on an article template of the second language and extracting a set of attribute names included in the article template of the second language An extractor;
A template-cited article set extraction unit that extracts a set of articles described using the article template of the first language and extracts a set of articles described using the article template of the second language; ,
In the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit, the attribute names of the first language and the attribute names of the second language And
The correspondence extracted from the set of articles in the first language based on the set of articles in the first language and the set of articles in the second language extracted by the template cited article set extraction unit. A set of attribute values for the attribute name of the first language attached and a set of attribute values for the attribute name of the associated second language extracted from the set of articles of the second language An association unit associating the attribute value of the first language with the attribute value of the second language;
Each pair of the character string of the attribute name of the first language and the character string of the attribute name of the second language associated by the association unit, and the attribute of the first language associated with each other A bilingual dictionary generation unit for generating the bilingual dictionary storing each pair of a character string of a value and a character string of an attribute value of the second language as a bilingual pair;
A bilingual dictionary generation device including The association unit
In the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit, the attribute names of the first language and the attribute names of the second language For each pair, and based on the similarity calculated for each pair, associate the attribute name of the first language with the attribute name of the second language,
The correspondence extracted from the set of attribute values for the attribute name of the associated first language and the set of articles of the second language extracted from the set of articles in the first language In the set of attribute values for the attribute name of the second language attached, for each pair of the attribute value of the first language and the attribute value of the second language, the similarity of the pair is calculated; Based on the similarity calculated for each pair, the attribute value of the first language is associated with the attribute value of the second language,
The bilingual dictionary generation unit
Each pair of the character string of the attribute name of the first language and the character string of the attribute name of the second language associated by the association unit, and the attribute of the first language associated with each other 2. The bilingual dictionary according to claim 1, wherein each pair of a character string of a value and a character string of an attribute value of the second language is stored in the bilingual dictionary as a bilingual pair together with the similarity calculated for the pair. Generator. The association unit
In the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit, the attribute names of the first language and the attribute names of the second language For each pair, and based on the similarity calculated for each pair and a predetermined first threshold, the attribute name of the first language and the second language To the attribute name of
The correspondence extracted from the set of attribute values for the attribute name of the associated first language and the set of articles of the second language extracted from the set of articles in the first language In the set of attribute values for the attribute name of the second language attached, for each pair of the attribute value of the first language and the attribute value of the second language, the similarity of the pair is calculated; 3. The bilingual dictionary generation device according to claim 2, wherein the attribute value of the first language and the attribute value of the second language are associated with each other based on the similarity calculated for each pair and a predetermined second threshold value. . The association unit
For each attribute name included in the attribute name set of the first language extracted by the template attribute set extraction unit, the attribute name of the first language is set as a processing target in a predetermined order.
The attribute name of the first language to be processed is associated with the attribute name of the second language, and the attribute value of the attribute name of the first language to be processed is assigned to the attribute name of the second language. The association with the attribute value is repeated for each attribute name of the first language to be processed,
The bilingual dictionary generation device according to claim 3, wherein the first threshold value and the second threshold value are reduced each time the association is performed on the attribute name of the first language to be processed. The association unit
A first attribute value instance set extracting unit for extracting a set of attribute values for the attribute name of the first language to be processed from the set of articles in the first language extracted by the template cited article set extracting unit; ,
The second language corresponding to the attribute name of the first language to be processed based on the attribute name similarity from the attribute name set of the second language extracted by the template attribute set extraction unit A corresponding attribute candidate extraction unit that extracts a set of candidate attribute names of
Articles of the second language extracted by the second language and the template-cited article set extraction unit for each of the attribute names of the second language included in the set of candidate attribute names of the second language A second attribute value instance set extraction unit for extracting a set of attribute values for the attribute name of the second language from the set of:
For each attribute name of the second language included in the set of candidate attribute names of the second language, the attribute name of the second language is extracted by the second attribute value instance set extraction unit. The similarity between the attribute value set for the attribute name of the second language and the attribute value set for the attribute name of the first language to be processed extracted by the first attribute value instance set extraction unit An attribute value instance set similarity calculation unit for calculating the similarity between attribute value instance sets;
The attribute value instance set similarity calculated by the attribute value instance set similarity calculation unit for each attribute name of the second language included in the attribute name candidate set of the second language is: If it is equal to or greater than the first threshold, the attribute name of the first language to be processed is associated with the attribute name of the second language,
The attribute value of the first language and the attribute of the second language in the attribute value set for each of the attribute name of the first language and the attribute name of the second language that are associated with each other For each pair with a value, if the pair similarity is equal to or greater than the second threshold, an association determination unit that associates the attribute value of the first language and the attribute value of the second language of the pair; The bilingual dictionary production | generation apparatus of Claim 4 containing. A combination of a first language character string and a second language character string that include a template attribute set extraction unit, a template-cited article set extraction unit, an association unit, and a bilingual dictionary generation unit. A bilingual dictionary generating method in a bilingual dictionary generating device for generating a bilingual dictionary storing bilingual pairs,
An article template in which attribute names related to the entity are listed for describing an article about the entity to be described by the template attribute set extraction unit, and the type of the entity to be described corresponds to the first template A set of attribute names included in the article template of the first language based on the article template of the second language and the article template of the second language, and the attribute names included in the article template of the second language Extracting a set of
The template cited article set extraction unit extracts a set of articles described using the article template of the first language, and sets a set of articles described using the article template of the second language. Extracting, and
In the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit by the association unit, the attribute names of the first language and the first Associate attribute names in two languages
The correspondence extracted from the set of articles in the first language based on the set of articles in the first language and the set of articles in the second language extracted by the template cited article set extraction unit. A set of attribute values for the attribute name of the first language attached and a set of attribute values for the attribute name of the associated second language extracted from the set of articles of the second language And associating the attribute value of the first language with the attribute value of the second language;
Each of the pair of the attribute name character string of the first language and the character string of the attribute name of the second language associated by the association unit is associated with the bilingual dictionary generation unit. Generating the bilingual dictionary storing each pair of a character string of the attribute value of the first language and a character string of the attribute value of the second language as a bilingual pair;
A bilingual dictionary generation method including A program for generating a bilingual dictionary storing a bilingual pair that is a combination of a character string of a first language to be translated and a character string of a second language,
Computer
An article template in which attribute names related to the entity are listed for describing an article about the entity to be described, the article template of the first language corresponding to the type of the entity to be described, and the first A template attribute set for extracting a set of attribute names included in the article template of the first language based on an article template of the second language and extracting a set of attribute names included in the article template of the second language Extractor,
A template-cited article set extraction unit that extracts a set of articles described using the article template of the first language and extracts a set of articles described using the article template of the second language;
In the set of attribute names of the first language and the set of attribute names of the second language extracted by the template attribute set extraction unit, the attribute names of the first language and the attribute names of the second language And
The correspondence extracted from the set of articles in the first language based on the set of articles in the first language and the set of articles in the second language extracted by the template cited article set extraction unit. A set of attribute values for the attribute name of the first language attached and a set of attribute values for the attribute name of the associated second language extracted from the set of articles of the second language The association unit associating the attribute value of the first language with the attribute value of the second language, and the character string of the attribute name of the first language associated with the association unit and the first language Each of a pair of attribute names of two languages and a pair of attribute values of the first language and character strings of the attribute values of the second language, The bilingual dictionary stored as a bilingual pair Program to function as a parallel translation dictionary generator.