JP2007025939A - Multilingual document retrieval device, multilingual document retrieval method and program for retrieving multilingual document - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that a conventional method for retrieving multilingual documents is not suitable for a system which needs the new establishment of a category or the review of a classification axis in an as-needed base because the correspondence relation of the category between different languages must be 1:1. <P>SOLUTION: Translated documents are generated from original documents of each language by translation processing, and the translated documents are classified to a document category of the language. When a retrieval query is inputted, matched candidates matched to the retrieval query are retrieved from original documents of the same language as the query and translated documents thereof. For a document lingually paired with each matched candidate or a document having the relation of an original document-translated document to the matched candidate, the category thereof is determined. Further, among documents paired with other documents belonging to the determined category, documents of the same language as the retrieval query are extracted as relevant candidates, and the relevant candidates and the matched candidates previously retrieved are outputted as a retrieval result. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a technique for retrieving desired document data from a document data group described in various languages such as Japanese and English, and particularly to a document data group in which a unique category is provided for each language. The present invention relates to a technique for executing a search.

  Conventionally, there are two ways to search for multilingual documents using a computer: a search query that is entered as a search keyword is translated into the same language as the search target, and then a document search is performed. After that, it can be roughly divided into two methods of document retrieval. The former method has the advantage that the search target document does not need to be translated in advance compared to the latter method. On the other hand, the accuracy of the search depends on the translation accuracy of the query. It is difficult to obtain.

  Conventionally proposed techniques for searching for multilingual documents include, for example, those described in Patent Documents 1 to 6 described later. In the method described in Patent Document 1, when a document in each language is classified into categories in advance and the categories corresponding to the contents are correlated one-to-one, a matching category is selected from the search result of the source language. Then, the search in the target language is limited to the category corresponding to the selected category.

  The technique described in Patent Document 2 is to obtain a synonym of a search term from a thesaurus before and after translating the search query in a multilingual search for translating the query, and add it to the query. The technique described in Patent Document 3 is a query that uses a query in which the number of words in a query is short and an appropriate translation query cannot be obtained when it is a new word in a translation dictionary. Then, the related words are extracted using the TFIDF score based on the search result of the source language, and the translated words of the related words are added to the translation query.

  In the method described in Patent Document 4, in a multilingual search for translating a query, a searcher selects a matching document corresponding to a search request from a search result of a source language by an initial search request, and the search request in the selected matching document is related to A characteristic word list is generated on the basis of statistical information of appearance words, and this is translated into a target language and searched in the target language. In the multilingual search for translating a query, the technique described in Patent Literature 5 classifies search results using hierarchical Bayesian clustering for the purpose of improving the efficiency of document selection of search results. The search results are presented in a unified manner.

In the technique described in Patent Document 6, in multilingual search for translating a query, metadata such as an issue date, category, and field is added to a document. Let the searcher select By using this condition for narrowing down other languages, documents in foreign languages that are difficult to read are efficiently filtered using screening in the source language.
JP 2002-189745 A Japanese Patent Laid-Open No. 08-305728 JP 2003-208441 A JP 2001-022787 A JP 2003-76710 A Japanese Patent Laid-Open No. 2003-050821

  However, in order to obtain an appropriate search result in the conventional method of narrowing down the search result using the category correspondence, it is necessary that the category correspondence between different languages is one-to-one. Such a condition is not suitable for a system in which, for example, new original documents are added to the search target one after another, and a new category or a review of the classification axis is required accordingly. A typical example of this type of system is, for example, a knowledge base used in the contact center of a service provider, that is, a database in which customer service representatives register and share reports on response cases.

  It is an object of the present invention to provide a method capable of presenting a more appropriate document even when a document to be searched is managed for each language and each document is classified by a unique category for each language. Is to provide.

  The multilingual document search apparatus according to the present invention includes a processor and a storage medium that stores a plurality of original documents assigned with document categories defined for each language for each language. Means for generating a translation document, means for associating the translation document with the original document and storing it in the storage medium according to language, means for obtaining the document category of the translation document from the document category of the same language as the translation document, and Means for searching for a matching candidate that matches the search query from the original document and the translation document in the same language as the search query, recognizes the document category of the translation document or the original document for the matching candidate, and other documents belonging to the document category A document in the same language as the search query is extracted as a related candidate from a translated document or an original document for the document, and the related candidate and the matching candidate are extracted And means for outputting the retrieval result.

  A multilingual document search method according to the present invention includes: a processor connected to a storage medium that stores a plurality of original documents to which a document category defined for each language is assigned for each language, and generates a translated document by translating the original document A step of storing the translated document in association with the original document in the storage medium according to the language, a step of obtaining the document category of the translated document from the document category of the same language as the translated document, and an input means A step of searching for a matching candidate that matches the search query from an original document and a translation document in the same language as the search query, a translation document corresponding to the matching candidate or a document category of the original document, and other documents belonging to the document category Documents in the same language as the search query are extracted as related candidates from translated documents or original documents for And a step of outputting from the output means the fit candidate as a search result.

  The basic concept of the present invention will be described. As a pre-process for document search, a translation document is generated from an original document of each language by translation processing, and the generated translation document is classified into a document category of the language. For example, when the original document is Japanese, the translated English document is classified into an English document category.

  After the pre-processing, when a search query is input in a certain language, matching candidates that match the search query are searched from the original document and translation document in the same language. Further, a category is obtained for a document that is paired in language with the detected matching candidate, that is, a document that has a relationship between the original document and the translated document with respect to the matching candidate.

  Furthermore, a document in the same language as the search query is detected as a related candidate among documents paired with other documents belonging to the obtained category. As a result, matching candidates and related candidates in the same language as the search query are detected, and these candidates are output as search results. By adopting such a configuration and performing multilingual document search, the object of the present invention can be achieved.

  According to the present invention, in a multilingual document search process, in addition to matching candidates that match a search query, related candidates are extracted using a category to which a document paired with the matching candidate belongs. Even when the classification axes and the granularity of are not consistent and the correspondence relationship of categories between different languages is not 1: 1, an appropriate search result can be presented. Thereby, it is possible to improve the search accuracy even in a system in which new original documents are added to the search target one after another.

  The best mode for carrying out the present invention will be described in detail with reference to the drawings. In this embodiment, it is assumed that document data to be searched is in two languages, Japanese and English, and each of these two languages has a unique document category. The state where the original document category is set indicates that the contents of the document category such as literature, politics, and science are not common in each language. In the following description, the Japanese category and the English category refer to document categories that are associated with the original documents in each language in advance.

  Referring to FIG. 1, a computer that is a multilingual document search apparatus according to this embodiment includes an input unit 100 such as a keyboard, a processor 200 that operates under program control, a storage medium 300 such as a hard disk, a display, and a printer. Output means 400.

  The processor 200 includes, as its functional configuration, a document search integration unit 201, a Japanese document search unit 202, and an English document search unit 203 that perform processing related to multilingual document search, and an English-Japanese translation process that performs preprocessing prior to document search Means 204, Japanese-English translation processing means 205, Japanese category discrimination means 206, and English category discrimination means 207. These functional configurations correspond to functions realized by the processor 200 executing a program (not shown) stored in the storage medium 300 or the like.

  The storage medium 300 includes a Japanese document holding unit 301 that stores a Japanese original document, an English document holding unit 302 that stores an English original document, and a document category and an original document that are uniquely set for each language. Japanese category holding means 303 and English category holding means 304 for storing associations, English-Japanese translation document holding means 305 for storing translation documents by English-Japanese translation processing means 204, and translation documents by Japanese-English translation processing means 205 are stored. A Japanese-English translation document holding means 306, an English-Japanese bilingual dictionary holding means 307 for storing a Japanese-English translation dictionary, and a Japanese-English bilingual dictionary holding means 308.

  The overall operation of this embodiment will be described in detail with reference to the flowchart shown in FIG. First, as a pre-stage for executing the search, it is checked whether or not the data necessary for the document search is registered in the storage medium 300 (step A0). If not, the preprocessing described below is executed.

  The Japanese-English translation processing unit 205 inputs a Japanese document stored in the Japanese document holding unit 301 and performs a Japanese-English translation process. Are recorded in the Japanese-English translation document holding means 306 (step A1). Further, the Japanese-English translation document and the translation-source Japanese document are output to the English category discrimination means 207 together with the word boundary information in each language obtained in the course of the translation process.

  The English category discriminating means 207 classifies the input Japanese-English translation document into the English category (step A2). This corresponds to a process in which a currently unclassified Japanese-English translation document is regarded as a new English document and it is determined in which of the English categories it is classified. Specifically, the English category discriminating unit 207 determines which of the English categories stored in the English category holding unit 304 the target Japanese-English translation document is classified into. Note that one Japanese-English translation document may be classified into three or more categories, or may not be classified into any category.

  There are two ways to classify Japanese-English translation documents into English categories. One is a method of classifying Japanese-English translation documents as they are, and the other is a method of classifying Japanese-English translation documents into English categories using a Japanese document as a translation source. The former is a method in which a target Japanese-English translation document is compared with an existing English document, and an English category of the Japanese-English translation document is determined based on the degree of coincidence of words between the two.

  Hereinafter, the latter method of classification using Japanese documents will be described in detail. The English category discriminating unit 207 acquires the correspondence between the English category and the English document belonging to each category from the English category holding unit 304, and acquires the English document belonging to each category from the English document holding unit 302. Then, the degree of similarity between the Japanese document as the translation source and each English document is calculated based on the ratio of the words in both documents.

  In calculating the similarity between a Japanese document and an English document, a method conventionally known in this field is used. For example, each word appearing in a document is taken as a dimension, and the element has an appearance frequency or a value obtained by multiplying the appearance frequency by the word weight (tf * idf (tf: Term Frequency, idf: Inverse Document Frequency)). A method of obtaining such a document vector and using the inner product of the document vectors as the similarity of two documents can be used.

  In addition, the English category discriminating means 207 recognizes the words constituting the Japanese document and the English document using the word boundary information when collating the word in the Japanese document with the word of the English document. Then, translation candidates for words in the Japanese document are acquired from the Japanese-English bilingual dictionary holding means 308, and word candidates that match these translation word candidates are obtained from the English document. If the translation candidate of a word in a Japanese document is different from the translation of the corresponding word in a Japanese-English translation document, it can be inserted as a second translation candidate in parentheses or the like at the corresponding location in the Japanese-English translation document. It can be expected that the search accuracy is reduced due to a translation error and the difficulty in reading due to a translation error is improved.

  The English category discriminating means 207 is based on the calculation result of the similarity between the Japanese document and the English document, the English category in which there is an English document very similar to the Japanese document, or most English documents and Japanese documents. Select an English category that shows a certain degree of similarity. Then, the selected English category is determined as the category of the Japanese-English translation document that has not been classified, and this Japanese-English translation document is stored in the Japanese-English bilingual dictionary holding means 308 in association with the English category.

  The above-described method using Japanese translation source documents for categorizing Japanese-English translation documents has the following advantages. In general, in the translation process, the translated word is uniquely determined with respect to the original word, so even if the translation is done correctly, the English words with the same meaning in Japanese do not match literally. It may not be detected. On the other hand, in the above method of collating the word of the Japanese document with the English document while translating it in the bilingual dictionary, if there is a match with the word of the collation partner among a plurality of translation word candidates, Flexibility can be given to document collation between different languages.

  The English-Japanese translation processing means 204 translates the English document stored in the English document holding means 302 into Japanese by the same procedure as the processing procedure performed by the English category discrimination means 207 (step A3). The Japanese category discriminating means 206 performs the same processing in the English-Japanese direction as the processing performed by the English category discriminating means 207 in the Japanese-English direction (step A4). In the procedure shown in FIG. 2, the English-Japanese translation is performed after the Japanese-English translation, but this order is not limited and may be the reverse order shown in the figure.

  When the user performs a search keyword input operation after the above pre-processing is completed and a search query is input from the input means 100 (step A5), the search processing for this query is performed according to the following procedure.

  First, the document search integration unit 201 determines the language of the input search query (step A5-1). As a result, if the language of the search query is Japanese, the document search integration unit 201 cooperates with the Japanese document search unit 202. The search is made for Japanese matching candidates and related candidates to be described later (step A6). If the language of the search query is English, the document search integration unit 201 searches for English matching candidates and related candidates in cooperation with the English document search unit 203 (step A7).

  Here, with reference to the flowchart shown in FIG. 3, a procedure for searching for a match candidate and a related candidate will be described. Hereinafter, a case where the search query is Japanese will be described as an example.

  The document search integration unit 201 outputs a Japanese search query to the Japanese document search unit 202. The Japanese document search unit 202 searches the input search query for all Japanese documents in the Japanese document holding unit 301 and all English-Japanese translation documents in the English-Japanese translation document holding unit 305. Documents that match the query conditions are searched, and the document set obtained as a result is output to the document search integration unit 201 as matching candidates (step S1). The matching candidates are not limited to either Japanese documents or English-Japanese translation documents, and both may be mixed.

  When the document search integration unit 201 acquires the match candidate from the Japanese document search unit 202, the document search integration unit 201 acquires the Japanese category to which the match candidate Japanese document and the English-Japanese translation document belong from the Japanese category holding unit 303 (step S2). ). If the category acquisition is successful, that is, if the matching candidate belongs to any Japanese category (step S3: Yes), other documents included in the category are extracted as related candidates (step S4).

  At this time, when a plurality of categories are obtained for one document, related candidates are extracted from each category. If no other document exists in the target category, or if the matching candidate does not belong to any category (step S3: No), the process proceeds to the next step.

  Next, the document search integration unit 201 searches for a document paired with the matching candidate, that is, a Japanese-English translation document corresponding to the translation result of the Japanese document, and an English document corresponding to the translation source of the English-Japanese translation document, and their categories. Is obtained from the English category holding means 304 (step S5). As a result, when the document paired with the matching candidate belongs to any English category (step S6: Yes), pay attention to the other document belonging to the category, and the document paired with each focused document, that is, Japanese A document or an English-Japanese translation document is extracted as a related candidate (step S7).

  Finally, the document search integration unit 201 outputs the Japanese matching candidates and related candidates obtained by the above procedure together to the output unit 400 (step S8).

  The above description is a case where the search query is in Japanese. However, in the case of English, the document search integration unit 201 outputs the query to the English document search unit 203, so that the procedure shown in FIG. A similar search process is performed along. Thereby, an English matching candidate and a related candidate are obtained for an English search query.

  Through the above processing, the matching candidate and the related candidate are presented to the user by the output unit 400 as a search result for the previously input search query (FIG. 2: Step A8).

"Concrete example"
In order to deepen the understanding of the present embodiment, a detailed example will be described. Here, as shown in FIG. 4, the Japanese documents d _J1 , d _J2 , d _J3 in the Japanese document holding unit 301 are classified into the category Ja, and the English documents d _E1 , d in the English document holding unit 302. _{E2, d E3} is assumed to be classified in the category Eb. FIG. 5 shows an example of these original documents. Hereinafter, an example in which a multilingual document is searched using the Japanese search query q _J shown in FIG. 5 will be described.

The Japanese-English translation processing unit 205 generates Japanese-English translation documents d _J1 ′, d _J2 ′, d _J3 ′ from the Japanese documents d _J1 , d _J2 , d _J3 and records them in the Japanese-English translation document holding unit 306 ( FIG. 2: Step A1). The English-Japanese translation processing means 204 generates English-Japanese translation documents d _E1 ′, d _E2 ′, d _E3 ′ from the English documents d _E1 , d _E2 , d _E3 and records them in the English-Japanese translation document holding means 305 (FIG. 2: Step A3). FIG. 6 schematically shows the relationship between the translation document generated by the translation process and the original document.

The English category discriminating means 207 classifies the Japanese-English translation documents d _J1 ′, d _J2 ′, d _J3 ′ into English categories (FIG. 2: Step A2). Here, as shown in FIG. 7, the Japanese-English translation documents d _J1 ′ are classified into the same English category Eb as the English documents d _E1 , d _E2 , d _E3, and the remaining Japanese-English translation documents d _J2 ′ and d _J3 ′. Are classified into different English categories Ec and Ed.

Further, the Japanese category discriminating means 206 classifies the English-Japanese translation documents d _E1 ′, d _E2 ′, and d _E3 ′ into Japanese categories (FIG. 2: step A4). Here, as shown in FIG. 7, it is assumed that all the English-Japanese translation documents d _E1 ′, d _E2 ′, and d _E3 ′ generated this time do not correspond to any Japanese category.

After the preprocessing is completed in the state shown in FIG. 7, when a Japanese search query q _J is input (FIG. 2: step A5), the Japanese document search means 202 determines that the Japanese search query q _J Relevant candidates are searched for Japanese documents d _J1 , d _J2 , d _J3 and English-Japanese translation documents d _E1 ′, d _E2 ′, d _E3 ′, which are the same language. As a result, as shown in (1) of FIG. 8, it is assumed that English-Japanese translation documents d _E1 ′ and d _E2 ′ are searched as matching candidates for the search query q _J.

Next, the document search integration unit 201 obtains the categories of the English-Japanese translation documents d _E1 ′ and d _E2 ′ as matching candidates and the categories of the English documents d _E1 and d _E2 that are the translation sources, and from each category. Extract related candidates. Here, since the English-Japanese translation documents d _E1 ′ and d _E2 ′ do not belong to any Japanese category, no related candidate is extracted at this time. If the current candidate for matching is the Japanese document d _J1 , the Japanese documents d _J2 and d _J3 , which are other documents belonging to the category Ja, are extracted as related candidates.

On the other hand, English-Japanese translation document _d E1 _{', d E2'} and will be verified English document _{d E1, d E2} is a translation source of, that category is the English category Eb as shown in (2) of FIG. 8. Therefore, in the category Eb, focusing on the English document d _E1, d _E2 other than the document in which English document d _E3 and Japanese-to-English translation document d _{J1 ',} associated candidate a document to be paired on each document and the language you have these attention Extract as That is, as shown in FIG. 8 (3), the English-Japanese translation document d _E3 ′, which is the translation result of the focused English document d _E3 , and the Japanese translation source of the focused Japanese-English translation document d _J1 ′. The document d _J1 is a related candidate.

Finally, as shown in (4) of FIG. 8, the document search integration unit 201 performs two English-Japanese translation documents d _E1 ′ and d _E2 ′ as matching candidates and two English-Japanese translation documents as related candidates. d _E3 ′ and the Japanese document d _J1 are output to the output unit 400 as a search result for the search query q _J.

  According to the embodiment described above, in the multilingual document search process, in addition to the match candidate matching the search query, a related candidate is extracted using the category to which the document paired with the match candidate belongs, Since related candidates are output together, the classification axis and granularity of categories in each language do not match, and appropriate search results are presented even in a system where the correspondence between categories in different languages is not 1: 1. can do.

  In the above embodiment, there are two types of language to be searched. However, the present invention can also be applied to a multilingual document search having three or more language types. At this time, the document as the related candidate is controlled to extract a document in the same language as the search query. As an example, multilingual search in which the language type is Japanese / English / German is described below.

  In performing the document search in the above three languages, the configuration of the processor 200 includes, as shown in FIG. 1, a German document search means, a German language category determination means, a German-Japanese translation processing means, a Japanese-German translation processing means, and a German-English translation. Add processing means and English-German translation processing means. In addition, the storage medium 300 has a configuration related to German, similar to the configuration shown in FIG. As pre-processing by the processor 200, the original document in German / Japanese is translated into Japanese / English, the original document in Japanese / English is translated into German, and the obtained translation document is classified into the category of the language. Add processing to classify.

Next, the search processing in the three languages will be described using a specific example. Here, as shown in FIG. 9, a Japanese document d _J1 / d _J2 / d _J3 , an English document d _E1 / d _E2 / d _E3 , and a German document d _G1 / d _G2 , which are original documents in each language, respectively. Assume that the categories are previously classified into the corresponding Japanese category Ja, English category Eb, and German language category Gc.

Further, it is assumed that the category of the translation document for the original document in each language is determined as shown in FIG. The sign (J) shown in the figure indicates that the document is translated into Japanese. For example, d _E1 ^(J) indicates an English-Japanese translation document obtained by translating the English document d _E1 into Japanese. Similarly, (E) indicates an English translation document, and (G) indicates a German translation document.

After the preprocessing is completed in the state shown in FIG. 10, when a Japanese query q _J is input, the Japanese document search means causes the Japanese document d _J1 / d having the same language as the Japanese search query. _J2 / d _J3 , English-Japanese translation documents d _E1 ^(J) / d _E2 ^(J) / d _E3 ^(J) , and German-Japanese translation documents d _G1 ^(J) / d _G2 ^(J) Perform a search. As a result, as shown in (1) in FIG. 11, the search query _{q J} EJ to translate document ^{_{d E1 (J) / d E2}} (J) is retrieved as adapted candidates.

Next, the document search integration unit obtains the category of the English-Japanese translation document d _E1 ^(J) / d _E2 ^(J) as the candidate for matching and the category of the English document d _E1 / d _E2 that is the translation source, Extract related candidates from each category. Here, the English-Japanese translation document d _E1 ^(J) / d _E2 ^(J) does not belong to any Japanese category, so no related candidate is extracted at this time.

On the other hand, when the English document d _E1 / d _E2 which is the translation source of the English-Japanese translation document d _E1 ^(J) / d _E2 ^(J) is verified, the category is the English category Eb as shown in (2) of FIG. It is. Therefore, in the category Eb, the English document d _E3 which is a document other than the English document d _E1 / d _E2 , the Japanese-English translation document d _J1 ^(E), and the German-English translation document d _G1 ^(E) / d _G2 ^(E) Then, the document that is paired with each of the focused documents is extracted as a related candidate.

As shown in (3) of FIG. 11, as related candidates, first, an English-Japanese translation document d _E3 ^(J) , which is a translation result of the focused English document d _E3 , and a focused Japanese-English translation document d _J1 ^{( The} Japanese document d _J1 which is the translation source of ^E) is extracted.

Also, in the case of the German-English translation document d _G1 ^(E) / d _G2 ^(E) , which is also focused on, the translation source is the German document d _G1 / d _G2 , and the language is the same as the current search query (Japanese) do not do. In this case, the German-English translation document d _G1 ^(E) / d _G2 ^(E) may be excluded from the related candidate extraction process, but by adding a procedure as shown in FIG. Many related candidates can be obtained. That is, the focused German-English translation document ^{_{d G1 (E) / d G2}} (E) of the translation will originate German Japanese translation document for document _d G1 _{/ d G2,} i.e. DE-Japanese translation document _d ^{G1 (J)} / _d Add _G2 ^(J) to the related candidates.

Finally, as shown in FIG. 11 (4), the document search integration means performs two English-Japanese translation documents d _E1 ^(J) / d _E2 ^(J) as matching candidates and a Japanese document as a related candidate. d _J1 , English-Japanese translation document d _E3 ^(J) and German-Japanese translation document d _G1 ^(J) / d _G2 ^(J) are output to the output means as search results for search query q _J.

In a document search of three or more languages, a document such as a German-English translation document d _G1 ^(E) / d _G2 ^(E) that is not related to the language of the search query is included in the category (Eb) of the pair of matching candidates. However, by adding the search shown in (3) ′ using these documents as a medium, the search range of the documents is expanded, and as a result, more search results can be presented to the user.

  The present invention is suitable for searching for multilingual documents classified into categories such as patents, papers, product and service FAQs, contact center response records, office documents, and the like. The present invention is also applicable to a document sharing system that performs classification and retrieval of such multilingual documents.

It is a block diagram which shows the structure of embodiment of this invention. It is a flowchart which shows the operation | movement procedure of embodiment. It is a flowchart which shows the procedure regarding the document search in embodiment. It is explanatory drawing (the 1) for demonstrating the specific example of embodiment. It is explanatory drawing (the 2) for demonstrating the specific example of embodiment. It is explanatory drawing (the 3) for demonstrating the specific example of embodiment. It is explanatory drawing (the 4) for demonstrating the specific example of embodiment. It is explanatory drawing (the 5) for demonstrating the specific example of embodiment. It is explanatory drawing (the 1) for demonstrating the specific example of other embodiment. It is explanatory drawing (the 2) for demonstrating the specific example of other embodiment. It is explanatory drawing (the 3) for demonstrating the specific example of other embodiment.

Explanation of symbols

100 input means 200 processor 300 storage medium 400 output means 201: document search integration means 202: Japanese document search means 203: English document search means 204: English-Japanese translation processing means 205: Japanese-English translation processing means 206 : Japanese category discriminating means, 207: English category discriminating means 301: Japanese document holding means, 302: English document holding means, 303: Japanese category holding means, 304: English category holding means, 305: English-Japanese translation document holding Means 306: Japanese-English translation document holding means 307: English-Japanese bilingual dictionary holding means 308: Japanese-English bilingual dictionary holding means

Claims (9)

A processor and a storage medium for storing a plurality of original documents assigned with a document category defined for each language for each language;
The processor is
Means for generating a translation document by translating the original document;
Means for associating the translated document with the original document and storing it in the storage medium by language;
Means for obtaining a document category of the translated document from a document category of the same language as the translated document;
Means for searching for matching candidates that match the input search query from an original document and a translation document in the same language as the search query;
Recognizing a document category of a translation document or an original document for the matching candidate, extracting a document in the same language as the search query from a translation document or an original document for another document belonging to the document category as the related candidate And a multilingual document search apparatus characterized by comprising: means for outputting the matching candidates as search results.   When the processor extracts related candidates, if there is a translated document in which the language of the corresponding original document is different from the language of the search query in another document belonging to the recognized document category, the translated document for the original document The multilingual document search apparatus according to claim 1, wherein a translation document in the same language as the search query is added to the related candidates.   When obtaining the document category of the translation document, the processor collates the original document belonging to the document category of the language and the original document of the translation source of the translation document by word-by-word parallel translation, and the similarity between both documents by the collation 3. The multilingual document retrieval apparatus according to claim 1, wherein a document category of the translated document is determined based on the document.   When obtaining the document category of the translated document, the processor collates the original document belonging to the document category of the language and the translated document for each word, and based on the similarity between the two documents by the collation, the document category of the translated document The multilingual document search apparatus according to claim 1, wherein the multilingual document search apparatus determines the number of words.   A program for causing a computer to function as the multilingual document search apparatus according to any one of claims 1 to 4. A processor connected to a storage medium for storing a plurality of original documents assigned with document categories defined for each language for each language,
Generating a translation document by translating the original document;
Storing a translated document in the storage medium in association with the original document in a language;
Obtaining a document category of the translated document from a document category in the same language as the translated document;
Searching for matching candidates that match the search query input by the input means from the original document and the translation document in the same language as the search query;
Recognizing a document category of a translation document or an original document for the matching candidate, extracting a document in the same language as the search query from a translation document or an original document for another document belonging to the document category as the related candidate And a step of outputting the matching candidate as a search result from an output means. In the step of outputting the search result by the processor,
When there is a translated document in which the language of the corresponding original document is different from the language of the search query in another document belonging to the recognized document category, the translated document of the same language as the search query among the translated documents for the original document The multilingual document search method according to claim 6, wherein: is added to the related candidates. In the step, wherein the processor determines a document category of the translation document,
Collating the original document belonging to the document category of the language and the original document of translation of the translated document by word-by-word parallel translation, and determining the document category of the translated document based on the similarity between the two documents by the collation The multilingual document search method according to claim 6 or 7, wherein the multilingual document search method is characterized. In the step, wherein the processor determines a document category of the translation document,
8. The original document belonging to the document category of the language and the translated document are collated for each word, and the document category of the translated document is determined based on the similarity between the two documents by the collation. The described multilingual document search method.

Images