JP3666066B2 - Multilingual document registration and retrieval device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a multilingual document registration / retrieval apparatus for registering a search index for a document including sentences described in a plurality of types of languages, and more specifically, a plurality of types of documents. Multilingual document registration and retrieval device that extracts keywords from a text database of documents written in multiple languages and registers them as an index and searches the multilingual text database using the registered index It is about.
[0002]
[Prior art]
Conventionally, as a document search apparatus for a multilingual document including sentences described in a plurality of types of languages, for example, a “document search apparatus” described in JP-A-4-21180 is known. This "document search device" is a database that has keywords created in its own language and a document search device that searches the database by inputting the keywords, and has a dictionary of the native language and other languages, and is input in other languages. For the keyword, the keyword is converted into the native language, and the database is searched with the keyword of the native language. Then, the document name of the retrieved document and the entire text of the selected document are converted into another language and displayed on the screen.
[0003]
[Problems to be solved by the invention]
By the way, when performing a document search using an index for document search processing, the document to be searched is not only a single language (for example, Japanese) as in the native language, but is also cited as a cited document. Languages from other countries (for example, English) are often included. For this reason, if the index in the document search device does not correspond to multiple languages (multiple languages), sufficient documents are available. You cannot search for.
[0004]
2. Description of the Related Art Conventionally, in a document search apparatus that uses an index for document search processing, an index generation method used for document search performs morphological analysis, cuts out words, and creates an index using the words as keywords. In that case, since the language for performing morphological analysis is usually a language of one country (for example, only Japanese), words of languages of other countries can only be extracted as proper noun words. Therefore, in order to register the extracted word as a keyword in the index, it cannot be restored to the standard notation or original form. Therefore, in this case, in the creation of the index for search, the unregistered word for the already registered word is left as it is. I could only extract it and register it in the index.
[0005]
For this reason, when the target document contains many languages from different countries, the number of keywords (proprietary nouns) of words to be registered as a search index increases and the index size becomes unnecessarily large. there were.
[0006]
In addition, in such a document search apparatus, when searching for a document, it is possible to search only with a character string having the same pattern as, for example, a proper noun index registered as a keyword, and an appropriate range can be sufficiently searched. There was no problem. That is, when searching for a document, there is a problem in that a character string having exactly the same pattern as a word (proprietary noun) cut out by morphological analysis will not be hit, and a desired search document will not be hit even if searched.
[0007]
When a keyword in a certain language is registered and a search request is made in a language other than the language of the country as in the “document search device” described in Japanese Patent Application Laid-Open No. Hei 4-21180, for example, It is possible to search for documents including languages of other countries by translating into words of already registered languages that have the same meaning. There is a problem in that it is composed of keywords only in Japanese, and no consideration is given to index registration for documents composed of multiple languages.
[0008]
The present invention has been made to solve these problems, and a first object of the present invention is to provide a document including sentences written in a plurality of languages, as much as possible to each of those different languages. The morpheme analysis is performed in response to the above, the words are extracted, the keywords are extracted, and the keywords can be registered, so that unregistered words can be minimized even for documents containing sentences written in multiple languages. An object of the present invention is to provide a multilingual document registration / retrieval apparatus.
[0009]
A second object of the present invention is to extract a word by morphological analysis corresponding to a plurality of languages, extract a keyword, and register the keyword for a document including sentences written in a plurality of languages. Another object of the present invention is to provide a multilingual document registration / retrieval device that avoids duplication of the range to be analyzed, minimizes the index size, and improves retrieval accuracy when performing retrieval.
[0010]
In addition, the third object of the present invention is to cope with a change in the form of words in a language other than the native language and fluctuation of the notation for a document including sentences described in a plurality of languages, and when searching. An object of the present invention is to provide a multilingual document registration / retrieval device capable of improving search accuracy, minimizing the extraction of unnecessary unregistered words and minimizing the index size.
[0011]
Furthermore, a fourth object of the present invention is to extract a word by morphological analysis corresponding to a plurality of languages for a document including sentences described in a plurality of languages, and to perform a morphological analysis corresponding to each of the plurality of languages. Multi-language that avoids duplicate analysis of target texts by combination and performs morphological analysis optimally as efficiently as possible, and minimizes unregistered words for documents containing sentences written in multiple languages The object is to provide a document registration / retrieval apparatus.
[0012]
[Means for Solving the Problems]
In order to achieve the above object, a multilingual document registration / retrieval device as a first feature of the present invention creates and registers an index used for retrieval for a document including sentences in a plurality of languages, In a multilingual document registration / retrieval apparatus that searches documents by index, multilingual document storage means (3) for storing documents including sentences in a plurality of languages, and corresponding to sentences in different languages with respect to the documents Controlling multiple word segmentation means A keyword extraction unit (2) for performing morphological analysis and extracting a keyword of the document, an index registration unit (4) for registering the keyword extracted by the keyword extraction unit together with an identifier of the corresponding document, and a search condition A search condition input means (11) for inputting, an index matching means (12) for cutting out a word from the search conditions input by the search condition input means, and matching the cut-out word with an index keyword, It is characterized by comprising reading means (14) for reading out a document that meets the search condition based on the collation result.
[0013]
Also, a multilingual document registration / retrieval device as a second feature of the present invention In the keyword extraction means Process priority is given to a plurality of word segmentation means (21a, 22a, 23a) having different segmentation target languages, a setting means (27) for setting processing priority of the plurality of word segmentation means, and the plurality of word segmentation means. Index that controls according to the degree, extracts a keyword from a document, and extracts a keyword, and an index that associates the extracted keyword with the identifier of the document from which the word of the keyword is extracted and registers it in the index Registration means (28).
[0014]
In the multilingual document registration / retrieval apparatus according to the third aspect of the present invention, the keyword extraction control unit may identify a word that cannot be identified by a word extraction unit with a certain processing priority, with a next processing priority. The extracted word is processed by the word extracting means, the identifier of the word is used as a keyword, and the word that cannot be identified to the end by a plurality of word extracting means is used as the keyword. .
[0015]
A multilingual document registration / retrieval device as a fourth feature of the present invention is: Furthermore, Input means (1) for inputting a document to be registered and instructing keyword extraction; , It is characterized by comprising holding means (3, 5) for holding a document to be registered, an index, and a word not registered in the dictionary file.
[0016]
In the multilingual document registration / retrieval apparatus according to the fifth feature of the present invention, the keyword extracting means extracts a word from a document composed of a plurality of language sentences by morphological analysis for each language sentence. A plurality of word extraction means (21a, 22a, 23a), a plurality of dictionary files (21b, 22b, 23b) for storing dictionaries corresponding to languages referred to by the plurality of word extraction means; An order setting means (27) for setting the order in which the word extracting means is applied, and a plurality of word extracting means are controlled in the order set by the order setting means to extract corresponding multilingual sentence words from the document. And control means (26) for performing control.
[0017]
In the multilingual document registration / retrieval apparatus as the sixth feature of the present invention, Furthermore, An unregistered keyword candidate holding means (25) that temporarily holds the word determined as an unregistered word by the word extracting means, and a keyword that is temporarily extracted from other dictionaries. Keyword candidate holding means (24) for holding as candidates, and the control means (26, 27) controls the first word extraction means to input a document including sentences in a plurality of languages. The word is extracted by morphological analysis, and the word determined to be an unregistered word is temporarily stored in the unregistered keyword candidate holding means as an unregistered keyword candidate, and the keyword candidate is extracted from the dictionary The keyword candidate storage means performs processing to control each word extraction means in order, The unregistered word candidate held in the unregistered keyword candidate holding unit by the output unit is input, the word is cut out by morphological analysis, and the word that is determined to be an unregistered word is directly used as the unregistered keyword candidate holding unit. The word extracted from the dictionary is deleted from the unregistered keyword candidate holding unit, and added and held in the keyword candidate holding unit. Finally, the keyword candidates held in the keyword candidate holding unit are A keyword is registered, and the unregistered keyword held in the unregistered keyword candidate holding means is registered in the index together with the identifier of the corresponding document as an unregistered keyword.
[0018]
A multilingual document registration / retrieval device as a seventh feature of the present invention creates and registers an index used for searching for a document including sentences in a plurality of languages, and searches the document using the index. In the language document registration and retrieval device, Keyword extraction for extracting text keywords by controlling morphological analysis by controlling a text database storing documents containing sentences in multiple languages and a plurality of word segmenting means corresponding to sentences in different languages. Means, and an index registration means for registering the keyword extracted by the keyword extraction means together with an identifier of the corresponding document as an index, Search condition input means (11) for inputting a search condition, index collation means (12) for multilingual correspondence that cuts out a word from the search condition input by the search condition input means and collates with an index, and the index collation means And extracting means (14) for extracting a corresponding document from a text database based on the collation result.
[0019]
In the multilingual document registration / retrieval apparatus according to the eighth feature of the present invention, the index matching means performs morphological analysis on a sentence in a corresponding language from a document composed of a plurality of languages, A plurality of word extracting means (131a, 132a, 133a) for extracting Multiple A word setting means (137) for combining the word cutting means to set the order in which the word cutting means is applied, and a control for cutting out the search condition words input by the search condition input means in the order set by the order setting means And control means (136) for performing the above.
[0020]
In the multilingual document registration / retrieval apparatus according to the ninth feature of the present invention, unregistered search word candidates that are temporarily stored as unregistered search word candidates for words determined by the word extracting means as unregistered words. A storage means (134) and a search word candidate storage means (135) for temporarily storing words extracted from other dictionaries as search word candidates, the control means (13) 6 , 137) controls the word extraction means in the first stage, receives a document including sentences in a plurality of languages, extracts words by morphological analysis, and temporarily determines words that are determined as unregistered words. The unregistered search word candidate is held in the unregistered search word candidate holding unit, and the word extracted from the dictionary is stored as the search word candidate in the search word candidate holding unit. The word extraction means is controlled, the unregistered word candidate held in the unregistered search word candidate holding means is input by the word extraction means in the previous stage, the word is extracted by morphological analysis, and it is determined as an unregistered word The words are left as they are in the unregistered search word candidate holding means, and the words extracted from the dictionary are deleted from the unregistered search word candidate holding means, and the search word candidate holding means Performs additional holding processing, finally sets the search word candidate held in the search word candidate holding means as a search word, and sets the unregistered search word candidate held in the unregistered search word candidate holding means as an unregistered search word As a feature, index matching is performed, a corresponding document is extracted by a text database unit, and result information is output.
[0021]
According to the multilingual document registration / retrieval apparatus of the present invention having such various features, in the multilingual document registration / retrieval apparatus having the first feature, the multilingual document storage means (3) includes a plurality of A document including a sentence in a language is stored, and the keyword extracting means (2) corresponds to a sentence in a different language for the document. Controlling multiple word segmentation means Perform morphological analysis and extract keyword of document. The index registration means (4) registers the keyword extracted by the keyword extraction means as an index together with the identifier of the corresponding document.
[0022]
When searching for a document, when the search condition is input by the search condition input means (11), the index matching means (12) cuts out a word from the search condition input by the search condition input means, Match the index keyword. Then, the reading means (14) reads a document that meets the search condition based on the keyword / word matching result. In this way, an index used for search is created and registered for a document including sentences in a plurality of languages, and the document is searched using the index.
[0023]
In the multilingual document registration / retrieval apparatus as the second feature of the present invention, an index is created. Therefore, keyword extraction means include A plurality of word segmentation means (21a, 22a, 23a) having different segmentation target languages are provided. When the setting means (27) sets the processing priority of the plurality of word segmentation means, the keyword extraction control means (26 ) Controls the plurality of word segmenting means according to the processing priority, segments a word from the document, and extracts a keyword. The index registration means (28) registers the extracted keyword and the identifier of the document from which the word of the keyword is extracted in association with the index. Thus, for a document including sentences in a plurality of languages, a keyword word can be cut out by performing morphological analysis by each of a plurality of word cutout means corresponding to each target language. For this reason, it is possible to create an index to be used at the time of searching while minimizing unregistered words for a document described in multiple languages.
[0024]
In the multilingual document registration / retrieval apparatus according to the third feature of the present invention, a morphological analysis is performed on a document containing sentences in a plurality of languages by a plurality of word segmenting means corresponding to each target language. When the keyword extraction unit extracts a keyword word, the keyword extraction control unit processes the word that cannot be identified by the word extraction unit with a certain processing priority by the word extraction unit with the next processing priority, and is extracted. For a word, the identifier of the word is a keyword. In addition, regarding a word that cannot be identified to the end by a plurality of word segmenting means, the word is used as a keyword. As a result, a word described in a plurality of languages can be extracted by morphological analysis corresponding to each language, and keywords can be extracted while avoiding overlapping analysis ranges. In this way, the index size when registering keywords can be minimized.
[0025]
According to the multilingual document registration / retrieval apparatus as the fourth feature of the present invention, document registration In addition, to Input means (1) and , Holding means (3, 5) Is provided. When the input means (1) inputs a document to be registered and instructs keyword extraction, the keyword extraction means (2) includes a dictionary used for word segmentation, extracts a keyword of the document by morphological analysis, and registers (4 ) Registers the keyword extracted by the keyword extracting means in the index together with the identifier of the corresponding document. As a result, the holding means (3, 5) holds the document to be registered, the index, and the word not registered in the dictionary file. As a result, an index used for search can be created and registered for a document including sentences in a plurality of languages, and the document can be searched using the index.
[0026]
According to the multilingual document registration / retrieval apparatus as the fifth feature of the present invention, in the keyword extracting means, the plurality of word extracting means (21a, 22a, 23a) is based on a document composed of sentences in a plurality of languages. When words are extracted by morphological analysis for sentences in each language, a plurality of dictionary files (21b, 22b, 23b) correspond to languages referred to by the plurality of word extraction means (21a, 22a, 23a), respectively. When the order setting unit (27) sets the order in which the plurality of word extraction units are applied, the control unit (26) sets a plurality of dictionaries in the order set by the order setting unit. Is controlled to cut out words of a corresponding multilingual sentence from the document.
[0027]
Further, in the multilingual document registration / retrieval apparatus as the sixth feature of the present invention, the unregistered keyword candidate holding means (25) temporarily unregisters words determined as unregistered words by the word extracting means. The words stored as keyword candidates and the words extracted from other dictionaries are temporarily stored as keyword candidates by the keyword candidate holding means (24). When extracting keywords for multilingual correspondence, the control means (26, 27) controls the word extraction means in the first stage to input a document including sentences in a plurality of languages and input a word by morphological analysis. Cut out. Thereby, the word determined to be an unregistered word is temporarily stored in the unregistered keyword candidate holding unit as an unregistered keyword candidate, and the keyword candidate holding unit is used as a keyword candidate for a word extracted from the dictionary. Perform the process to hold.
[0028]
Subsequently, each word extraction unit is sequentially controlled, and the unregistered word candidate held in the unregistered keyword candidate holding unit is input by the preceding word extraction unit, and the word is extracted by morphological analysis, A word determined to be an unregistered word is left as it is in the unregistered keyword candidate holding unit, and a word extracted from the dictionary is deleted from the unregistered keyword candidate holding unit and additionally held in the keyword candidate holding unit. Process.
[0029]
Finally, the keyword candidates held in the keyword candidate holding unit are set as keywords, and the unregistered keywords held in the unregistered keyword candidate holding unit are registered in the index together with the identifiers of the corresponding documents as unregistered keywords. In this way, the words are cut out sequentially by morphological analysis, so that the morpheme is optimized as efficiently as possible by avoiding duplicate analysis of the target text by the combination of multiple morphological analyzes corresponding to multiple languages. Analysis can be performed, and unregistered words can be suppressed to a minimum with respect to a document including sentences written in multiple languages.
[0030]
According to the multilingual document registration / retrieval apparatus as the seventh feature of the present invention, In order to A search condition input means (11), a multilingual index matching means (12), and an extraction means (14) are provided. When the search condition input means (11) inputs a search condition, multilingual support is provided. Index collating means (12) extracts words from the search condition input by the search condition input means and collates them with the index. The extraction means (14) extracts the corresponding document from the text database based on the collation result of the index collation means. Accordingly, when an index used for searching is created and registered for a document including sentences in a plurality of languages, the document can be searched using the index.
[0031]
Also, when performing a document search, according to the multilingual document registration / retrieval apparatus as the eighth feature of the present invention, in the index collating means, a plurality of word extracting means (131a, 132a, 133a) are provided in a plurality of languages. Morphological analysis is performed on a sentence in a corresponding language from a document composed of sentences of Multiple Control for cutting out words from the search condition input by the search condition input means in the order set by the order setting means by the control means (136). Then, the document is searched by the word of the search condition.
[0032]
Further, in the multilingual document registration / retrieval apparatus according to the ninth feature of the present invention, the unregistered search word candidate holding means (134) temporarily does not store a word determined as an unregistered word by the word extracting means. Words stored as registered search word candidates and words extracted from other dictionaries are temporarily stored as search word candidates by the search word candidate holding means (135).
[0033]
The control means (136, 137) controls the word extraction means at the first stage, receives a document including sentences in a plurality of languages, extracts words by morphological analysis, and is determined to be an unregistered word The word is temporarily stored in the unregistered search word candidate holding unit as an unregistered search word candidate, and the word extracted from the dictionary is stored in the search word candidate holding unit as a search word candidate. .
[0034]
Subsequently, each word extracting unit is sequentially controlled, and unregistered word candidates held in the unregistered search word candidate holding unit are input by the previous word extracting unit, and words are cut out by morphological analysis, The words determined to be unregistered words are left as they are in the unregistered search word candidate holding means, and the words extracted from the dictionary are deleted from the unregistered search word candidate holding means, and the search word candidate holding means Perform the process to add and hold.
[0035]
Finally, the search word candidate held in the search word candidate holding unit is set as a search word, the unregistered search word held in the unregistered search word candidate holding unit is set as an unregistered search word, and index matching is performed. The corresponding document is extracted by the text database unit and the result information is output.
[0036]
In this way, an index used for search is created and registered for a document including sentences in a plurality of languages, and the document is searched using the index, so that a document described in a plurality of languages can be automatically searched. It can cope with changes in the form of words other than Japanese and fluctuations in notation, improving search accuracy. In addition, unnecessary unregistered word extraction is minimized, and the index size can be minimized. In addition, for documents written in multiple languages, optimal morphological analysis is performed as efficiently as possible by avoiding duplicate analysis of the target text by combining multiple morphological analyzes corresponding to multiple languages. It is possible to keep unregistered words to a minimum for documents written in multiple languages.
[0037]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments for carrying out the present invention will be specifically described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a multilingual document registration / retrieval apparatus according to an embodiment of the present invention. In FIG. 1, 1 is an input processing unit, 2 is a multilingual keyword extraction unit, 3 is a text database unit, 4 is an index registration unit, 5 is an index file unit, 11 is a search condition input unit, and 12 is a multilingual index registration unit. , 13 is a display unit, and 14 is a text extraction unit.
[0038]
The text database unit 3 stores a document (110: FIG. 11) including sentences written in a plurality of languages, such as sentences written in English and sentences written in Japanese. When the input processing unit 1 receives an input instruction for document registration from the user, the multilingual keyword extraction unit 2 performs processing for extracting the keyword for the document instructed by the user. Here, the document from which the keyword is extracted is, for example, a document input from the input processing unit 1 and registered in the text database unit 3 or a document already registered in the text database unit 3. A document from which a keyword is extracted is designated for the document, and a process for extracting the keyword is performed on the document.
[0039]
As will be described later, the multilingual keyword extraction unit 2 includes a plurality of morpheme analysis units (21a to 23a: FIG. 2) that perform morphological analysis corresponding to the language of each country. Corresponding morpheme analyzers are sequentially controlled to efficiently perform morpheme analysis corresponding to sentences in different languages in the document, extract words, and extract keywords. A plurality of morpheme analyzers are provided with language-corresponding dictionary files corresponding to the languages to be analyzed, and when morphological analysis is performed and words are extracted, the words corresponding to each language are It is compared with the corresponding dictionary corresponding to the language.
[0040]
Words registered in a dictionary file in a language are temporarily held as keyword candidates in the keyword candidate holding unit, and words not registered in a dictionary in a language are temporarily stored as unregistered keyword candidates. It is stored in the unregistered keyword candidate storage unit. The unregistered keyword candidate word is subjected to morphological analysis by the morphological analysis unit of the next language. In this way, when comparing with the dictionary corresponding to the language of the next country, the morphological analysis is performed including the word of the previous unregistered keyword candidate, the word is extracted, and the keyword is extracted.
[0041]
The keyword extracted in this way is indexed by the index registration unit 4 as an index together with the identifier of the document corresponding to the keyword. The Registered in the table.
[0042]
When a user searches for a desired document, the user inputs a search condition from the search condition input unit 11. When the search condition is input, the multilingual index matching unit 12 cuts out the word from the input search condition, and extracts the extracted word and the index. The Matches against index keyword in table. Then, the text extraction unit 14 reads out a document that matches the search condition based on the matching result of the keyword and the word based on the collation result, and displays the read document on the display unit 13.
[0043]
FIG. 2 is a block diagram illustrating a configuration of a main part of the multilingual keyword extraction unit. FIG. 2 shows a data flow together with a block of each element in the multilingual keyword extraction unit. In FIG. 2, 1 is an input processing unit, 21a is a first morpheme analysis unit, 21b is a first dictionary file unit, 22a is a second morpheme analysis unit, 22b is a second dictionary file unit, 23a is the Nth morpheme analyzer. , 23b is the Nth dictionary file part It is. Reference numeral 24 is a keyword candidate holding unit, 25 is an unregistered keyword candidate holding unit, 26 is a keyword / unregistered keyword determining unit, 27 is an order setting unit, 28 is an index registration unit, and 29 is an index file unit.
[0044]
As shown in FIG. 2, the multilingual keyword extraction unit includes a plurality of morpheme analysis units (21 a to 23 a) corresponding to each language for performing morphological analysis of sentences in each country language, and each morpheme analysis. Are provided with a plurality of dictionary file sections (21b to 23b) corresponding to national languages for supplying dictionary data of each language to the section, and the plurality of morpheme analysis sections (21a to 23a) are controlled to improve efficiency. In order to perform multilingual sentence morphological analysis, the order setting unit 27 for setting the order of performing morphological analysis of the multilingual document, and temporarily registering the analyzed words as keyword candidates as its working memory The word candidate holding unit 24 and the words that have not been analyzed by the morpheme analysis unit corresponding to one language, Unregistered keyword candidate holding unit 25 is provided to keep to register. When the morphological analysis is completed, the keyword / unregistered keyword determination unit 26 determines a word to be registered and a word to be unregistered keyword, and the index registration unit 28 extracts the keyword. Are registered in the index file unit 29 in association with the identifier.
[0045]
FIG. 3 is a diagram for explaining the contents of a control table used when performing keyword extraction processing. 3A shows the keyword extraction management table 30, FIG. 3B shows the morphological analysis management table 36, and FIG. 3C shows the analysis target character string type setting table 37. Show.
[0046]
As shown in FIG. 3A, the keyword extraction management table 30 is a table for managing the usage state of each morpheme analysis unit provided in the multilingual keyword extraction unit, and includes a number field 31 and a corresponding language type field. 32, an order field 33, a use flag field 34, and an analysis target character string type field 35. The condition data of the morphological analyzer used for each field is set. For example, in the second entry from the top, the condition of the morpheme analysis unit that performs morphological analysis in Japanese is set, “2” is set in the number field 31, and “2” is set in the corresponding language type field 32. “Japanese” is set, “1” is set in the order field 33, and “ON” is set in the use flag field 34. In addition, “text-ALL” is set in the analysis target character string type field 35, and the condition data here is set. In other words, from this condition data, “The second morpheme analysis unit is used for Japanese, performs morpheme analysis of multilingual documents first, and performs the range of the analysis target character string as the entire text. "Means.
[0047]
The morpheme analysis unit used in the keyword extraction process is managed by the morpheme analysis management table 36. In the morpheme analysis management table 36, as shown in FIG. 3B, data on the number n of usable morpheme analysis units corresponding to each language and the number i of the currently used morpheme analysis unit is managed. ing. Further, an analysis target character string type setting table 37 is provided for setting the range of the analysis target character string, as shown in FIG. The analysis target character string type setting table 37 includes a range of the analysis target character string according to the analysis target character string type. But It is prescribed. For example, when the analysis target character string type is “unregistered word group”, the range of the analysis target character string is set to “all keyword candidates in the unregistered keyword candidate holding unit or the unregistered search word holding unit”. In addition, when the analysis target character string type is “text-ALL”, the range of the analysis target character string is set to “all text of registered document or search expression”.
[0048]
Next, multilingual document registration processing in the keyword extraction processing unit as shown in FIG. 2 will be described using these control tables. FIG. 4 is a flowchart showing an overall processing flow of the multilingual document registration processing. The flowchart shown in FIG. 4 is an overall flow of processing for extracting a word as a keyword and registering it in an index by performing a morphological analysis corresponding to each language of a document described in one or more languages. Is shown. FIG. 5 is a flowchart showing the processing flow of the keyword setting management table condition setting process in the multilingual document registration process. FIG. 6 shows the analysis target character string range in the multilingual document registration process. FIG. 7 is a flowchart showing a processing flow of the keyword extraction processing in the multilingual document registration processing. FIG. 8 is a flowchart showing a processing flow of unregistered keyword candidate processing in multilingual document registration processing.
[0049]
First, the overall process of the multilingual document registration process will be described with reference to the flowchart of FIG. When the processing is started, first, in step 41, the input processing unit 1 issues a document registration instruction. Next, in step 42, keyword extraction condition setting processing (FIG. 5) is performed, and subsequently, in step 43, character string set setting processing (FIG. 6) to be analyzed in morphological analysis is performed. That is, the condition of the morpheme analysis unit used in the keyword extraction process is set, and then the target character string set to be analyzed is set using the morpheme analysis unit in which the condition is set. In the character string set setting process, for example, it is set whether all the text of the registered document or the character string set of the unregistered keyword candidate group, and the setting of the character string set to be analyzed that specifically performs the morphological analysis is set. Do.
[0050]
When these setting processes are completed, the variable i indicating the number (order) of the currently used morpheme analysis unit is set to “1” in step 44 to control the keyword extraction process. Set the morphological analyzer used at the beginning of the registration process. Next, in step 45, it is determined whether or not the use flag of the morphological analyzer of the language corresponding to the variable i is on. If the use flag is not turned on, the morpheme analysis process by the morpheme analysis unit having the number of the variable i is not performed.
[0051]
If the use flag of the morphological analysis unit of the language corresponding to the variable i is turned on in the determination in step 45, that is, the language corresponding to the variable i indicating the order in the keyword extraction management table 30 ("analysis order"). When the “usage flag” of the morphological analysis entry of the language in which i is i) is ON, the process proceeds to the next step 46, and the character set is sequentially processed by the i-th corresponding morphological analysis unit. The keyword extraction process (FIG. 7) is performed. In this keyword extraction processing, as will be described later, the first character set is set for the character string set set by the type of the analysis target character string of the entry of the ith morphological analysis unit in accordance with the condition data of the keyword extraction management table 30. A keyword is extracted by the i-th corresponding morphological analyzer.
[0052]
Next, in step 47, it is determined whether or not the extracted keyword is registered in the (i-th) morpheme analysis dictionary. If it is not registered, the process proceeds to step 48, where it is handled as an unregistered keyword candidate. Together with the document ID to be stored in the unregistered keyword candidate holding unit, the process proceeds to step 51. If it is determined in step 47 that the extracted keyword is registered in the morphological analysis dictionary, the process proceeds to step 49 and is stored in the keyword candidate holding unit together with the corresponding document ID as a keyword candidate. Then, in step 50, unregistered keyword candidate processing (FIG. 8) is performed. In this unregistered keyword candidate process, as will be described later, character string matching is performed on words that have been unregistered keyword candidates in the process of the previous morpheme analysis unit, but have been determined as keyword candidates in the process of the subsequent morpheme analysis unit. The collated word is removed from the unregistered keyword. When this unregistered keyword candidate process ends, the process proceeds to step 51.
[0053]
In step 51, it is determined whether or not the keyword extraction by the i-th morphological analysis unit of the registered document is completed. If keyword extraction has not been completed, the process returns to step 46 and the processing from step 46 is repeated. Further, if the completion of the keyword extraction process by the i-th corresponding morpheme analysis unit can be confirmed by the determination process of step 51, the keyword extraction process by the next morpheme analysis unit is performed. The variable i indicating the order of the morpheme analyzers is incremented, that is, the variable i is counted up (i = i + 1), and in the next step 53, the number n of morpheme analyzers corresponding to each national language that can be used and the next Is compared with a variable i indicating the order of the morpheme analysis units used for
[0054]
As a result of this comparison, if n ≧ i, the keyword extraction process by the i-th morpheme analyzer has not been completed, so the process returns to step 45 and the processes from step 45 are repeated. If n <i, the analysis by the morpheme analysis unit that is in a usable state according to the conditions set in the keyword extraction management table has been completed. An index is created from an unregistered keyword group. This completes the document registration process. Therefore, in step 55, it is determined whether or not the document registration is to be ended. If the other document registration process is to be performed, step 41 is performed. The processing is returned to step 4 and the processing from step 41 is repeated. Further, when the registration of the document is finished, the series of processes here is finished.
[0055]
Next, the process for setting the conditions of the keyword extraction management table in the multilingual document registration process will be described. This process is a process executed in step 42 of the overall process flow of the multilingual document registration process of FIG. Reference is made to the flowchart of FIG. When the process is started, the process proceeds to step 61, where the number of morpheme analyzers corresponding to each language stored in the keyword extraction management table is counted and set to a variable n indicating the number of morpheme analyzers. . That is, if the number of morphological analysis units registered in the keyword extraction management table 30 is m, n = m is set. Next, step 62 In The order of analysis is set in the order field 33 of the keyword extraction management table 30 using the morphological analysis unit corresponding to each language. Then, in the next step 63, in order to set the morphological analysis corresponding to the national language used at the time of document registration, whether to use the morphological analysis unit corresponding to the language of each country is used. In the field 34, the ON / OFF flag is set. Thereby, the setting process of the condition of the keyword extraction management table ends.
[0056]
Next, in the multilingual document registration process Solution of The analysis target character string range setting process will be described. As described above, this process is a process executed in step 43 of the overall process flow of the multilingual document registration process of FIG. Refer to the flowchart of FIG. When this process is started, first, in step 71, a variable j indicating the entry number of the morphological analysis unit corresponding to each national language stored in the keyword extraction management table 30 is set to “1”. Next, in step 72, the type of the analysis target character string for the morphological analysis unit corresponding to each language in the jth entry of the keyword extraction management table is selected from the analysis target character string type setting table. As described above, the analysis target character string type setting table 37 sets a range of character strings to be subjected to morphological analysis in accordance with the type of analysis target character string type. The type of analysis target character string is selected from the types of analysis target character string types in the column type setting table 37 and set in the analysis target character string type field 35 of the keyword extraction management table 30.
[0057]
Next, the process proceeds to step 73, where it is determined whether or not the “order of analysis” of the j-th morpheme analyzer is first. If the order of the j-th morpheme analyzer is first, the process proceeds to step 74 to forcibly set “text-ALL” ignoring the previously set analysis target character string type. Then, the process proceeds to step 75. As a result, the range of the solution target character string analyzed by the first morpheme analyzer is always the entire text of the registered document. If it is determined in step 73 that the order of the morphological analysis units is not first, the process proceeds to step 75 as it is.
[0058]
Subsequently, in order to set the analysis target character string type in the entry of the next morphological analysis unit, next, in step 75, the variable j is incremented, and in the next step 76, whether or not j ≦ n is satisfied. Determine. If j ≦ n, there is an entry corresponding to a morpheme analyzer that has not been set, so the process returns to step 72 and the processing from step 72 is repeated. If j ≦ n is not satisfied, the processing here ends. That is, the variable j is compared with the number n of morphological analysis units stored in the keyword extraction management table 30. If j is equal to or smaller than n, the process returns to step 72. If not, the process ends.
[0059]
Thereby, in the keyword extraction management table 30, the setting processing of the character string set to be analyzed is completed for each morphological analysis unit. As the analysis target character string set, for example, the process of setting all of the text of the registered document or the set of character strings of the unregistered keyword candidate group is completed. Perform keyword extraction processing according to the settings.
[0060]
Next, a keyword extraction process in the multilingual document registration process will be described. Reference is made to the flowchart of FIG. As described above, this process is a process executed in step 46 of the entire process flow of the multilingual document registration process of FIG. When the keyword extraction process is started and the process proceeds to step 81, a word is cut out in step 81 by performing morphological analysis on the character string set set by the character string type to be analyzed in the keyword extraction management table. That is, a word is cut out by morphological analysis from a position where the morphological analysis of the character string set is not completed.
[0061]
Next, in step 82, it is determined whether or not an unnecessary word is included in the extracted word. If an unnecessary word is not included, the process proceeds to step 88 as it is, and a word other than the unnecessary word is immediately extracted as a keyword, and the processing here ends. If the extracted word includes an unnecessary word, the process proceeds to step 83 to determine whether or not the variable i is “1”. When the variable i is “1”, the currently used morpheme analysis unit is the first morpheme analysis unit, so there is no processing for the unregistered keyword candidate. The processing here ends as a keyword obtained by extracting words other than.
[0062]
If it is determined in step 83 that the variable i is not “1”, the process proceeds to step 84, where the analysis target character string of the i-th morpheme analysis is an unregistered word group. Determine whether or not. If the character string to be analyzed is not an unregistered word group, the process proceeds to step 85, and character string matching is performed on the unregistered keyword candidate using the extracted word. Then, in the next step 86, the result of character string matching is determined. If the collation is successful, the process proceeds to step 87, where the words extracted from the unregistered keyword candidates or the words subjected to the character string collation are removed, and then in step 88, the words other than unnecessary words are extracted as keywords. Terminate the process.
[0063]
If it is not determined in step 86 that the character string has been collated, the process in step 87 is performed. And Then, the process proceeds to step 88, and the process here ends as a keyword obtained by extracting words other than unnecessary words.
[0064]
Next, the setting process of the unregistered keyword candidate process in the multilingual document registration process will be described. Reference is made to the flowchart of FIG. As described above, this process is a process executed in step 50 of the entire process flow of the multilingual document registration process of FIG. When the unregistered keyword candidate process is started and the process proceeds to step 91, first, it is determined whether or not the variable i indicating the currently used morphological analysis order is “1”. If the variable i is “1”, as described above, the morpheme analysis unit currently used is the first morpheme analysis unit, so there is no processing for the unregistered keyword candidate. The candidate process is terminated.
[0065]
If it is confirmed in step 91 that the variable i is not “1”, the process proceeds to step 92 to determine whether or not the character string to be analyzed in the i-th morpheme analysis is an unregistered word group. To do. That is, in the keyword extraction management table 30, it is determined whether or not the setting of the analysis target character string type field 35 of the entry corresponding to the i-th morpheme analysis unit in the analysis order is “unregistered word group”. If the result of this determination is that the character string type to be analyzed is “unregistered word group”, in step 95, the keywords extracted from the unregistered keyword candidates are removed, and this process ends.
[0066]
If it is determined in step 92 that the character string type to be analyzed is not “unregistered word group”, the process proceeds to step 93, where the extracted keyword is subjected to character string matching against unregistered keyword candidates. The result of this character string matching is determined. As a result of the determination, if the character string can be collated, the process proceeds to step 95, the keyword extracted from the unregistered keyword candidates is removed, and this process is terminated. If the character string cannot be collated, this process is terminated as it is.
[0067]
In this way, the processing for the unregistered keyword candidate is performed. As a result, the word which has been determined as the unregistered keyword candidate in the processing of the previous morpheme analysis unit, but is determined as the keyword candidate in the processing of the subsequent morpheme analysis unit. Character string matching is performed, and the matched word is removed from unregistered keywords.
[0068]
Next, in the case of registering a document including a sentence described in a plurality of languages, an example of the operation will be described by specifically illustrating a document including a sentence described in a plurality of languages. FIG. 9 is a diagram illustrating an example of a multilingual document. As shown in FIG. 9, the multilingual document 99 here is a document in which Japanese and English sentences exist, and a case where the multilingual document 99 is newly registered will be described. In this case, the entire document is first subjected to morphological analysis in Japanese, then the unanalyzed part is subjected to morphological analysis in English, keyword extraction is performed, and the document (document Registration number).
[0069]
This multilingual document registration / retrieval apparatus is provided with a morpheme analysis unit capable of performing morphological analysis corresponding to each of four languages corresponding to “English”, “Japanese”, “Chinese”, and “Arabic”. In this case, in the keyword extraction management table 30 that defines the keyword extraction conditions, the control conditions of the respective morphological analysis units are set as shown in FIG. Therefore, in this case, the number of morphological analysis units registered in the keyword extraction management table 30 is counted as “4” (n = 4) (step 61: FIG. 5), and as shown in FIG. It is temporarily stored in the morphological analysis management table 36 in a record format (or as a variable).
[0070]
Further, in setting the conditions of the keyword extraction management table 30, the order in which each morphological analysis unit analyzes the document is, for example, “Japanese”, “English”, “Arabic”, “Chinese”. Therefore, in the order field 33 of the keyword extraction management table 30, the order is “2”, “1”, “4”, “3” in order from the top, corresponding to each language-compatible morphological analysis unit. Setting is made (step 62: FIG. 5).
[0071]
In addition, since the language type of the morphological analysis unit used in the keyword extraction process in the document registration process is “Japanese”, “English”, and “Chinese”, the keyword extraction management table 30 is used. , The use flag field 34 of the entry corresponding to “Japanese”, “English” and “Chinese” is set to “ON”, and the use flag field 34 of the entry corresponding to “Arabic” is set to “OFF”.
[0072]
Further, in order to efficiently execute the keyword extraction process by specifying the range of the character string to be analyzed by each morphological analysis unit that performs morphological analysis in the keyword extraction process in the document registration process, In the morpheme analysis process set for the second and subsequent morpheme analysis units corresponding to the language, the entire document in which the character string group to be analyzed is not always registered is necessarily included in the scope, and is subject to analysis. Specifies a range of text or a string or a set of them.
[0073]
Therefore, as shown in FIG. 3C, in the analysis target character string type setting table 37, an analysis target character string type corresponding to a predefined character string range to be subjected to morphological analysis is displayed as a keyword extraction management table. It is set in 30 analysis target character string type fields 35. In this example, the analysis target character string type field 35 of the keyword extraction management table 30 includes, in order from the top, “unregistered word group”, “text-ALL”, “unregistered word group”, “text-range designation”. In the Japanese morphological analysis unit, the entire document is subject to analysis, but the English morphological analysis unit and the Chinese morphological analysis unit set the analysis target as an unregistered word group. . In this case, the first morpheme analysis unit that performs morpheme analysis is forcibly reset to default to “text-ALL” by default so that the entire document of the registered document is first analyzed (step). 73-step 74: FIG. 6).
[0074]
Thus, when conditions such as the order of each morpheme analysis unit used in the keyword extraction management table 30 and the character string range to be analyzed are set, each morpheme analysis unit is controlled according to the set conditions, Keyword extraction processing is executed. When the keyword extraction process is started, first, morpheme analysis is performed using the morpheme analysis unit set in the first order. In this example, after confirming that the “use flag” is “ON” by the first “Japanese correspondence” morpheme analysis unit (step 45), this morpheme analysis unit is supported. Keyword extraction is performed on the set of character strings. That is, in this case, since “text-ALL” is set, keyword extraction is performed on all texts of the registered document (step 46).
[0075]
In the keyword extraction process (FIG. 7), keywords other than those that are determined as unnecessary words among the extracted words are used as keywords. Since the “Japanese” morpheme analysis process is the first analysis process, unnecessary word processing for unregistered keyword candidates is not performed. Then, it is determined whether or not the next extracted keyword is registered in the Japanese morphological analysis dictionary (step 47), and the registered keyword is stored as a keyword candidate together with the document ID (step 48). .
[0076]
FIG. 10A and FIG. 10B are diagrams showing the contents of the keyword candidate holding unit and the unregistered keyword candidate holding unit at the stage where the keyword extraction processing by the Japanese-compatible morpheme analysis unit is completed. . For example, if the identification number (document ID) of the document to be registered (FIG. 9) is “20204”, the registration process of all keyword candidates by morphological analysis of “Japanese” is completed, and FIG. As shown, the keyword candidates 102 for the document (fileID) 101 of the keyword candidate holding unit 100 include the words “Iraq”, “Kuwait”, “country” of words extracted by morphological analysis as keyword candidates for the document with document ID = 20204. Border "...""Bomber" is memorized. On the other hand, those that are not registered in the dictionary for morpheme analysis are the unregistered keywords as shown in FIG. 10B at the stage where all the unregistered keyword candidates are registered by the “Japanese” morpheme analysis. For the unregistered keyword candidate 105 for the document (fileID) 104 in the candidate holding unit 103, the words “patriot”, “The”, “ "Ministry", "of", ..., "recently" are stored.
[0077]
In this way, when the process based on the “Japanese” morpheme analysis is completed, the process based on the “English” morpheme analysis of the next corresponding language is started. In this case, as in the case described above, according to the conditions of the keyword extraction management table 30, the second “English” morpheme analyzer confirms that the “usage flag” is also “ON”. Then (step 45), keyword extraction is performed on the character string set set corresponding to this morphological analysis unit. That is, in this case, “unregistered word group” is set as the analysis target character string type in the analysis target character string type field 35 of the keyword extraction management table 30, and therefore, as shown in FIG. In this manner, keyword extraction processing is performed on the character string stored in the unregistered keyword candidate 105 for the document (fileID) 104 in the unregistered keyword candidate holding unit 103. That is, a keyword with respect to a character string previously extracted as an unregistered keyword candidate without performing an English-supported morphological analysis on all documents for a document with document ID = 20204, which is the current registration target document. An extraction process is performed (step 46).
[0078]
Also in the keyword extraction processing (FIG. 7) in this case, as in the case described above, the extracted words other than those determined as unnecessary words are used as keywords. That is, by this processing, for example, “The”, “of”, etc., which are determined as unnecessary words, are excluded from the unregistered keyword candidates as processing of unnecessary words for the unregistered word keyword candidates. Then, it is determined whether or not the extracted keyword is registered in the English morphological analysis dictionary (step 47), and the registered keyword is stored as a keyword candidate together with the document ID (step 48).
[0079]
FIG. 11A and FIG. 11B are diagrams showing the contents of the keyword candidate holding unit and the unregistered keyword candidate holding unit at the stage where the keyword extraction processing by the next English-compatible morpheme analysis unit is finished, in comparison. is there. As described above, if the identification number (document ID) of the document to be registered (FIG. 9) is “20204”, all registration processing by “English” morphological analysis is completed for the unregistered keyword candidate. In this stage, the character strings “Ministry”, “Education”,... Stored in the unregistered keyword candidate 105 corresponding to the document (fileID) 104 of the unregistered keyword candidate holding unit 103 shown in FIG. For “said” and “recently”, morphological analysis is performed by the morphological analysis unit for English, and as a result, the words registered in the English morphological analysis dictionary are extracted. As shown in FIG. 11A, the keyword candidate 112 is additionally stored in the keyword candidate 112 for the document (fileID) 111 in the keyword candidate holding unit 110. That is, as keyword candidates for the document with document ID = 20204, “ministry”, “education”,..., “Say”, “recent” are additionally stored in the corresponding entries.
[0080]
Although not particularly mentioned in the processing of morphological analysis in this description, when a word is cut out by morphological analysis, a process of unifying several table words of the word into a standard form is also performed at the same time. In other words, processes such as “Ministry” → “ministry”, “Education” → “education”, and processing to unify uppercase letters into lowercase, and processes to unify the original form such as “said” → “say” are performed. . In this way, a word that has been an unregistered keyword candidate in one language is extracted as a keyword candidate by performing a morphological analysis in another language, and the keyword extracted as the keyword candidate is Process to remove from registered keyword candidates.
[0081]
In this way, those not registered in the English morphological analysis dictionary, in this example, “Monbushou” remain, and are stored as unregistered word keyword candidates. At the stage when all the keyword extraction processes by the “English” morphological analysis are finished, as shown in FIG. 11B, the unregistered keyword candidates 115 corresponding to the document (fileID) 114 of the unregistered keyword candidate holding unit 113 are displayed. As a non-registered keyword candidate for the document with document ID = 20204, “patriot” and “Monbushou” are stored in the corresponding entries.
[0082]
In this way, when the processing by the “English” morpheme analysis is completed, the process by the morpheme analysis unit corresponding to the language in the third order is subsequently entered. That is, the processing by the morpheme analyzer corresponding to “Arabic” of the next corresponding language is entered, but the “usage flag” of the “Arabic” morpheme analyzer in the keyword extraction management table 30 is “OFF”. In this case, as in the case described above, “ON” of the “usage flag” of the third “Arabic” morpheme analyzer corresponding to the condition of the keyword extraction management table 30 is used. "Cannot be confirmed (step 45), and the processing by the morpheme analyzer corresponding to" Arabic "is skipped.
[0083]
Thus, if the process by the morpheme analysis unit corresponding to the third order language is skipped, the process by the morpheme analysis unit corresponding to the fourth order language is entered. Also in this case, the processing proceeds according to the conditions of the keyword extraction management table 30 as in the case described above. In this case, it can be confirmed that the “usage flag” of the fourth “Chinese” corresponding morpheme analysis unit is “ON” (step 45). The analysis unit performs keyword extraction on the character string set set for the correspondence. In this case, since the “unregistered word group” is set as the analysis target character string type in the analysis target character string type field 35 of the keyword extraction management table 30, the unregistered keyword candidates shown in FIG. Subsequently, the morphological analysis is performed on the character string stored in the unregistered keyword candidate 115 for the document (fileID) 114 in the holding unit 113, and the keyword extraction process is performed. That is, the keyword extraction process is continuously performed on the character string previously extracted as the current unregistered keyword candidate for the document with the document ID = 20204 that is the current registration target document (step 46). ).
[0084]
In the keyword extraction process (FIG. 7) in this case, as in the case described above, the extracted words other than those judged as unnecessary words are used as keywords. Since no character string stored as the unregistered keyword candidate 115 in the unregistered keyword candidate holding unit 113 corresponds to “Chinese”, keyword extraction processing by morphological analysis of “Chinese” is performed. Even if the processing is completed, the contents of the keyword candidate storage unit 110 and the unregistered keyword candidate storage unit 113 do not change as shown in FIGS.
[0085]
In this way, when the keyword extraction process by morphological analysis for all languages is completed, an index table (120: FIG. 12) is created in accordance with the contents of the keyword extracted by the processes so far. As shown in FIG. 12, the index table 120 includes a document (file ID) 122 and an unregistered word flag 123 of the document corresponding to each extracted keyword 121 as an index for multilingual document search. It is a registered table. The index registered in the index table 120 is based on the case where the document ID of the corresponding document (fileID) 122 is sorted based on the word of the keyword 121, and there is an unregistered keyword for the document with the document ID. , An unregistered flag to that effect is set. When such an index table 120 is created, the document registration work for multilingual documents ends here.
[0086]
Then the index created like this The A case where a multilingual document search is performed using the index of the table 120 will be described.
[0087]
When searching for a document desired by the user, as described above, the user inputs a search condition using the search condition input unit (11: FIG. 1). When the search condition is input, the multilingual index matching unit (12: FIG. 1) cuts out a word from the search condition expression of the input search condition, and the extracted word and index The Matches against index keyword in table. Based on the collation result, the text extraction unit (14: FIG. 1) reads a document that matches the search condition based on the collation result between the index and the word, and the display unit (13: FIG. 1) displays the read document. .
[0088]
FIG. 13 is a block diagram illustrating a configuration of a main part of the multilingual index matching unit. FIG. 13 shows a data flow when a search expression is determined by performing a morphological analysis from a search condition together with a block of each element in the multilingual index matching unit. In FIG. 13, 3 is a text database part, 11 is a search condition input part, 14 is a text extraction part, 131a is a first morpheme analysis part, 131b is a first dictionary file part, and 132a is a second morpheme. An analysis unit, 132b is a second dictionary file unit, and 133a is an Nth morpheme analysis unit. 133b is the Nth dictionary file part It is. 134 is a search word candidate holding unit, 135 is an unregistered search word candidate holding unit, 136 is a search formula determining unit, and 137 is an order setting unit.
[0089]
As shown in FIG. 13, the multilingual index matching unit includes a plurality of morpheme analysis units corresponding to each language for performing a morphological analysis of a sentence in each language of a search condition expression input as a search condition. (131a to 133a) and a plurality of dictionary file units (131b to 133b) corresponding to each language that supply dictionary data of each language to each morpheme analysis unit, and the plurality of morpheme analysis units ( 131a to 133a) to control the morphological analysis of the multilingual search condition in an efficient manner, the order setting unit 137 for setting the order in which the morphological analysis of the multilingual search condition is performed, and its working memory The search term candidate holding unit 134 for temporarily registering the words of the analyzed search conditions as search term candidates and the search terms not analyzed by the morpheme analysis unit corresponding to one language For word, in order to perform the morphological analysis by the morphological analysis unit corresponding to another language, unregistered search word candidate holding unit 135 that temporarily registers are provided. Then, when the morphological analysis of the search condition expression of the search condition is completed, the search expression determination unit 136 determines a word to be the search expression, and the text extraction unit 14 determines from the text database unit 3 according to the index document ID. Extract documents that meet the search criteria.
[0090]
FIG. 14 is a flowchart showing the overall processing flow of the multilingual document search processing. The flowchart shown in FIG. 14 cuts out a word as a search word by performing a morphological analysis corresponding to each search condition expression in a search condition described in one or more languages, and performs a search. An overall flow of processing for creating an expression and searching for a document is shown. FIG. 15 is a flowchart showing a processing flow of search word extraction processing in multilingual document search processing, and FIG. 16 shows a processing flow of unregistered search word candidate processing in multilingual document search processing. It is a flowchart.
[0091]
First, the overall processing of multilingual document search processing will be described with reference to the flowchart of FIG. When the process is started, first, in step 141, the search expression input unit 11 inputs a multilingual search condition expression for document search. Next, in step 142, search word extraction condition setting processing is performed, and then in step 143, setting processing of a character string set to be analyzed in morphological analysis is performed. That is, in this case, similarly to the keyword extraction management table in the keyword extraction process described above, the search word extraction management table (not shown) is used to set the conditions of the morphological analysis unit for extracting the search word, In addition, a set of character strings to be analyzed is set using a morphological analysis unit in which conditions are set. In the character string set setting process, for example, the entire text of the search condition expression or the character string set of the unregistered search word candidate group (unregistered word group) is set, and the analysis is performed specifically for morphological analysis. Set the target character string set. It should be noted that the keyword extraction management table 30 described above may be used as the search word extraction management table without providing the search word extraction management table.
[0092]
When these setting processes are completed, the variable i indicating the order of the currently used morpheme analysis unit is set to “1” in step 144 in order to control the extraction process of the search word for creating the search expression. Set and set the morpheme analyzer used at the beginning of the morphological analysis process of the search condition formula. Next, in step 145, it is determined whether or not the use flag of the morphological analyzer of the language corresponding to the variable i is on. If the use flag is not on, the morpheme analysis process by the morpheme analysis unit with the number of the variable i is not performed.
[0093]
If the use flag of the morphological analysis unit of the language corresponding to the variable i is turned on in the determination in step 145, that is, the language corresponding to the variable i indicating the order in the search word extraction management table ("analysis order"). When the “usage flag” of the morphological analysis entry of the language in which i is i) is ON, the process proceeds to the next step 146, where the i-th corresponding morphological analysis unit sequentially performs the character string set. The search word extraction process (FIG. 15) is performed. In this search word extraction process, as will be described later, for a set of character strings set by the type of analysis target character string of the entry of the i-th order morpheme analysis unit according to the condition data of the search word extraction management table Then, a search word (keyword) is extracted by the i-th corresponding morphological analysis unit.
[0094]
Next, in step 147, it is determined whether or not the extracted search term is registered in the (i-th) morphological analysis dictionary. If it is determined in step 147 that it is not registered, the process proceeds to step 148 where the search condition expression corresponding to the unregistered search word candidate is stored in the unregistered search word candidate holding unit, and the process proceeds to step 151. If it is determined in step 147 that the extracted search word is registered in the morphological analysis dictionary, the process proceeds to step 149 and the search word candidate holding unit together with the search condition expression corresponding to the search word candidate. To remember. Then, in step 150, unregistered search word candidate processing (FIG. 16) is performed. In this unregistered search word candidate process, as will be described later, a word that has been determined as an unregistered search word candidate in the process of the previous morpheme analysis unit, Column matching is performed, and the matched word is removed from unregistered search terms. When the unregistered search word candidate process is finished, the process proceeds to step 151.
[0095]
In step 151, it is determined whether or not the search word extraction by the i-th morpheme analyzer has been completed. If the search term extraction is not completed, the process returns to step 146 and the process from step 146 is repeated. If the end of the search word extraction process by the i-th corresponding morpheme analysis unit can be confirmed by the determination process of step 151, the search word extraction process by the next morpheme analysis unit is performed. The variable i indicating the order of the morpheme analysis units to be used is incremented, that is, the variable i is counted up (i = i + 1), and in the next step 153, the number n of morpheme analysis units corresponding to each national language that can be used is n. And a variable i indicating the order of the morpheme analyzer to be used next.
[0096]
As a result of this comparison, if n ≧ i, the search word extraction process by the i-th morpheme analyzer has not been completed, so the process returns to step 145 and the processes from step 145 are repeated. If n <i, the analysis by the morpheme analysis unit that is in a usable state according to the conditions set in the search word extraction management table has been completed, so the process proceeds to step 154 to search word candidates. A process for creating a search expression from the group and the unregistered search word group is performed. As a result, a search expression for multilingual document search is created. Next, in step 155, the created search expression is used to match the index keyword, and the corresponding document is extracted. The process ends.
[0097]
Next, a search word extraction process in the multilingual document search process will be described. Reference is made to the flowchart of FIG. As described above, this process is a process executed in step 146 of the overall process flow of the multilingual document search process of FIG. When the search word extraction process is started and the process proceeds to step 161, morphological analysis is performed on the character string set of the search condition formula set in the search target character string type of the search word extraction management table in step 161. Cut out a word. That is, the word is cut out by the morphological analysis process from the position where the previous morphological analysis of the character string set is not completed.
[0098]
Next, in step 162, it is determined whether or not an unnecessary word is included in the extracted word. If an unnecessary word is not included, the process proceeds to step 168 as it is, and immediately, a search word other than the unnecessary word is extracted as a search word, and the process ends here. If the extracted word includes an unnecessary word, the process proceeds to step 163 to determine whether or not the variable i is “1”. When the variable i is “1”, the currently used morpheme analyzer is the first morpheme analyzer, so there is no processing for unregistered search word candidates. In this case as well, the process proceeds to step 168 and is unnecessary. The processing here ends as a search term obtained by extracting a word other than the word.
[0099]
If it is determined in step 163 that the variable i is not “1”, the process proceeds to step 164, and whether or not the analysis target character string of the i-th morphological analysis in the analysis order is an unregistered word group. Determine. If the analysis target character string is not an unregistered word group, the process proceeds to step 165, and character string matching is performed on the unregistered search word candidate using the extracted word. Then, in the next step 166, the result of character string matching is determined. If collation is successful, the process proceeds to step 167 to remove words or character strings collated from unregistered search word candidates, and then in step 168, as search words from which words other than unnecessary words are extracted, This process ends.
[0100]
If it is not determined in step 166 that the character string has been matched, the process proceeds to step 168 without performing the process in step 167, and a search term extracted from words other than unnecessary words is used here. The process in is terminated.
[0101]
Next, an unregistered search word candidate process in the multilingual document search process will be described. Reference is made to the flowchart of FIG. As described above, this process is a process executed in step 150 of the entire process flow of the multilingual document search process of FIG. When the unregistered search word candidate process is started and the process proceeds to step 171, first, it is determined whether or not the variable i indicating the order of the currently used morpheme analyzer is “1”. If the variable i is “1”, as described above, the morpheme analysis unit currently used is the first morpheme analysis unit, so there is no processing for the unregistered search word candidate, and this unregistered is immediately performed. The search term candidate process is terminated.
[0102]
If it is confirmed in step 171 that the variable i is not “1”, the process proceeds to step 172 to determine whether or not the analysis target character string of the i-th morphological analysis is an unregistered word group. To do. That is, it is determined whether or not the setting of the analysis target character string type field of the entry corresponding to the i-th morpheme analysis unit in the search word extraction condition control table is “unregistered word group”. . If the result of this determination is that the character string type to be analyzed is “unregistered word group”, in step 175, the search word extracted from the unregistered search word candidates is removed, and this process ends.
[0103]
If it is determined in step 172 that the character string type to be analyzed is not “unregistered word group”, the process proceeds to step 173, the extracted search word is subjected to character string matching with the unregistered search word candidate, and the next step In 174, the result of this character string matching is determined. As a result of the determination, if the character string can be collated, the process proceeds to step 175, the search word extracted from the unregistered search word candidates is removed, and this process is terminated. If the character string cannot be collated, this process is terminated as it is.
[0104]
In this way, the processing for the unregistered search word candidate is performed, and as a result, it is determined as the unregistered search word candidate in the process of the previous morpheme analysis unit, but is determined as the search word candidate in the process of the later morpheme analysis unit. Character strings are collated with respect to the detected words, and processing for removing the collated words from unregistered search words is performed.
[0105]
Next, when searching a multilingual document using a search condition expression of search conditions including sentences described in a plurality of languages, a specific example of the search condition expression will be described and an operation example will be described. FIG. 17 is a diagram illustrating an example of a search condition expression of a search condition including a sentence described in multiple languages. As shown in FIG. 17, the search condition 179 here is a search condition expression for a sentence in which Japanese and English sentences exist, and a document is searched by extracting a search word from the sentence of the search condition 179. Will be described. In this case, the entire search condition expression described in a plurality of languages is first subjected to morphological analysis in Japanese, and then the portion that has not been analyzed is subjected to morphological analysis in English. A search term is extracted, a search expression is generated from the extracted search term, and a corresponding document is searched.
[0106]
This multilingual document registration / retrieval apparatus is provided with a morpheme analysis unit that can perform morpheme analysis corresponding to each of four languages “English”, “Japanese”, “Chinese”, and “Arabic”. In the search term extraction management table that defines the search term extraction conditions, the control conditions of the respective morphological analysis units are set with the same contents as the keyword extraction management table 30 shown in FIG. It shall be. Here, in particular, the contents of the search word extraction management table are not shown, but the keyword extraction management table 30 shown in FIG. 3A is referred to as similar to the search word extraction management table as necessary. Therefore, the number of morpheme analyzers registered in the search word extraction management table in this case is counted as “4” (n = 4), and similarly, the morpheme analysis management table (36: FIG. 3 (b)). , Temporarily stored in record format (or as a variable).
[0107]
In setting the search term extraction management table conditions, the order of document analysis by each morphological analysis unit is, for example, “Japanese”, “English”, “Arabic”, and “Chinese”. The “analysis order” in the order field of the search word extraction management table (keyword extraction management table 30) corresponds to each language-compatible morpheme analysis unit, and the order is “2” and “1”, respectively. , “4”, “3”.
[0108]
Also, since the language type of the morphological analysis unit used in the search word extraction processing is “Japanese”, “English”, and “Chinese” here, the search word extraction management table (keyword extraction management table 30). ), The use flag of the entry corresponding to “Japanese”, “English” and “Chinese” is set to “ON”, and the use flag of the entry corresponding to “Arabic” is set to “OFF”.
[0109]
Further, in order to efficiently execute the search word extraction process by specifying the range of the character string to be analyzed by each morpheme analysis unit that performs the morphological analysis in the search word extraction process in the document search process, In the morphological analysis processing set for the second and subsequent morphological analysis units corresponding to each language, the character string group to be analyzed does not always fall within the scope of the entire sentence of the search condition expression. Specify the range of text or character string to be analyzed or a set of them.
[0110]
For this reason, as in the case of keyword extraction, as shown in FIG. 3C, in the analysis target character string type setting table 37, the analysis target character corresponding to the predefined character string range to be subjected to morphological analysis. The column type is the character string type to be analyzed for each entry in the search word extraction management table (keyword extraction management table 30). Set up Determine. In this example, the character string types to be analyzed in the search word extraction management table are “unregistered” corresponding to the morphological analysis units corresponding to “English”, “Japanese”, “Chinese”, and “Arabic”. “Word group”, “text-ALL”, “unregistered word group”, “text-range specification” are set.
[0111]
Therefore, in this case, the Japanese-compatible morphological analysis unit sets the entire document as an analysis target, but the English-compatible morpheme analysis unit and the Chinese-compatible morpheme analysis unit set the analysis target as an unregistered word group. Note that the morpheme analysis unit that performs the first morpheme analysis is always forcibly set as “text-ALL” by default to always analyze the entire document of the registered document.
[0112]
In this way, when conditions such as the order of each morpheme analysis unit to be used and the character string range to be analyzed are set by the search word extraction management table (keyword extraction management table 30), each morpheme is set according to the set conditions. The analysis unit is controlled, and search word extraction processing is executed. When the search word extraction process is started, first, morpheme analysis is performed using the morpheme analyzer that is set in the first order. In this example, after confirming that the “usage flag” is “ON” by the first “Japanese correspondence” morpheme analysis unit (step 145), this morpheme analysis unit is supported. Search term extraction is performed on the set of character strings. That is, in this case, since “text-ALL” is set, search word extraction is performed for all texts in the search condition formula (step 146).
[0113]
In the search word extraction process (FIG. 15), the extracted words other than those that are determined as unnecessary words are used as search words. Since the “Japanese” morphological analysis process is the first analysis process, unnecessary word processing for unregistered search word candidates is not performed. Then, it is determined whether or not the next extracted search word is registered in the Japanese morphological analysis dictionary (step 147), and the registered search word is stored as a search word candidate together with the corresponding search condition expression. (Step 148).
[0114]
FIG. 18A and FIG. 18B show the contents of the search word candidate holding unit and the unregistered search word candidate holding unit at the stage where the search word extraction processing by the Japanese-compatible morpheme analysis unit is completed. FIG. For example, with respect to the multilingual sentence “Retreat of Iraqi unit, Patriot missile and Ministry of Education” in the search condition formula 179 as shown in FIG. 17, registration processing of all search word candidates by morphological analysis of “Japanese” is performed. At the final stage, as shown in FIG. 18 (a), the search word candidate 181 of the search word candidate holding unit includes “Iraq”, “unit”, “withdrawal” of the words extracted by morphological analysis as search word candidates. ”And“ missiles ”are stored, while those not registered in the dictionary for morphological analysis are registered at the stage where all unregistered search word candidate registration processing by“ Japanese ”morphological analysis is completed. As shown in FIG. 18B, unregistered search word candidates 182 in the unregistered search word candidate holding unit are similarly “patriot”, “Ministry”, “of”, "Education Is stored as an unregistered search word candidate.
[0115]
In this way, when the process based on the “Japanese” morpheme analysis is completed, the process based on the “English” morpheme analysis of the next corresponding language is started. In this case, as in the case described above, the processing by the second “English” morpheme analyzer is performed according to the conditions of the search word extraction management table (keyword extraction management table 30). After confirming that the “usage flag” is also “ON” (step 145), search word extraction is performed on the character string set set corresponding to this morpheme analyzer. That is, in this case, “unregistered word group” is set as the character string type to be analyzed in the search word extraction management table (keyword extraction management table 30), so as shown in FIG. Search word extraction processing is performed on the character string of the unregistered search word candidate 182 in the registered search word candidate holding unit (step 146).
[0116]
In the search word extraction process (FIG. 15) in this case, as in the case described above, the extracted words other than those determined as unnecessary words are used as search words. That is, by this process, “of” determined as an unnecessary word is excluded from the unregistered search word candidates as an unnecessary word process for the unregistered word search word candidates. Then, it is determined whether or not the extracted search word is registered in the dictionary for English morphological analysis (step 147), and the registered search word is stored as a search word candidate together with the corresponding search condition formula ( Step 148).
[0117]
19 (a) and 19 (b) compare the contents of the search word candidate holding unit and the unregistered search word candidate holding unit at the stage where the search word extraction processing by the next English-compatible morpheme analysis unit is completed. FIG. In the stage where all the registration processes by the morphological analysis of “English” are completed for the unregistered search word candidate 182 in FIG. 18B, the character strings “Ministry” and “Education” are supported in English. As a result of the morpheme analysis in the morpheme analysis unit, among the extracted words, the words registered in the English morpheme analysis dictionary are searched word candidates as shown in FIG. “Ministry” and “education” are additionally stored as search word candidates in the search word candidate 191 of the holding unit. As in the case described above, the morphological analysis process described in this explanation is not particularly mentioned, but when extracting words by morphological analysis, the process of unifying uppercase letters into lowercase letters is standardized. Are performed at the same time. In this way, in the morphological analysis of one language, a word that is an unregistered search word candidate is extracted as a search word candidate by performing a morphological analysis in another language, and is extracted as the search word candidate A process for removing a search term from unregistered search term candidates is performed.
[0118]
As a result, those that are not registered in the English-compatible morphological analysis dictionary, in this example, “patriot” remain, and are stored as unregistered word search word candidates. At the stage where the processing of extracting all the search words by the “English” morpheme analysis is finished, as shown in FIG. 19B, “patriot” is stored as the unregistered search word candidate 192 of the unregistered search word candidate holding unit. It will be in the state.
[0119]
In this way, when the processing by the “English” morpheme analysis is completed, the process by the morpheme analysis unit corresponding to the language in the third order is subsequently entered. That is, the processing by the morpheme analyzer corresponding to “Arabic” of the next corresponding language is entered, but in the search word extraction management table (keyword extraction management table 30), the morpheme analyzer of “Arabic” is used. Since the “use flag” is “OFF”, in this case, as in the case described above, the third “Arabic” in the order according to the conditions of the search word extraction management table (keyword extraction management table 30). "ON" of the "usage flag" of the morpheme analysis unit corresponding to "" cannot be confirmed (step 145), and the process by the morpheme analysis unit corresponding to "Arabic" is skipped.
[0120]
Thus, if the process by the morpheme analysis unit corresponding to the third order language is skipped, the process by the morpheme analysis unit corresponding to the fourth order language is entered. In this case as well, the process proceeds according to the conditions of the search word extraction management table (keyword extraction management table 30) as in the case described above. In this case, it can be confirmed that the “usage flag” of the fourth “Chinese” corresponding morpheme analysis unit is “ON” (step 45). The analysis unit extracts a search term from the character string set set for the correspondence. In this case, since “unregistered word group” is set as the analysis target character string type in the analysis target character string type of the search word extraction management table (keyword extraction management table 30), FIG. Subsequently, a morphological analysis is performed on the character string stored in the unregistered search word candidate 192 of the unregistered search word candidate holding section shown, and the search word extraction process is performed (step 146).
[0121]
In the search word extraction process (FIG. 15) in this case, as in the case described above, a process other than a word that is determined as an unnecessary word among the extracted words is performed as a search word. There is nothing, and no character string stored as the unregistered search word candidate 192 of the unregistered search word candidate holding unit corresponds to “Chinese”, so the morphological analysis of “Chinese” Even if the search word extraction process by is completed, as shown in FIGS. 19A and 19B, there is no change in the contents of the search word candidate storage unit and the unregistered search word candidate storage unit.
[0122]
In this way, when the search word extraction process by morphological analysis for all languages is completed, a search expression 200 is created as shown in FIG. 20 according to the contents of the search word extracted by the process so far. The search formula 200 is an array of a search key 201 and an unregistered word flag 202 for searching a multilingual document, and the search key 200 and the unregistered word flag 202 are, for example, shown in FIG. Are collated with the index (keyword, document ID, unregistered word flag) of the index table 120 as shown in FIG. 9 and a multilingual document (FIG. 9) is read from the corresponding document ID.
[0123]
【The invention's effect】
As described above, according to the multilingual document registration / retrieval apparatus of the present invention, there are documents including texts or search condition expressions described in one or more languages, and those documents are registered and searched. When performing a search using conditional expressions, Meet By combining the processing of the morpheme analyzer, the word can be extracted with as much accuracy as possible. For this reason, the size of the index created at the time of registration can be made compact. In addition, the accuracy (reproducibility) of collation with the index can be improved during document search. Also, by embedding the control information of the keyword extraction conditions such as “analysis order”, “use flag”, and “analysis target character string type” in the keyword extraction management table, the target text is simply duplicated in each morphological analysis. Analysis can be performed optimally and efficiently according to the situation. By customizing these conditions by the user, it becomes possible to arbitrarily register and search multilingual documents corresponding to the user.
[Brief description of the drawings]
FIG. 1 is a block diagram showing the configuration of a multilingual document registration / retrieval apparatus according to an embodiment of the present invention;
FIG. 2 is a block diagram showing a configuration of a main part of a multilingual keyword extraction unit;
FIG. 3 is a diagram for explaining the contents of a control table used when performing keyword extraction processing;
FIG. 4 is a flowchart showing an overall processing flow of multilingual document registration processing;
FIG. 5 is a flowchart showing a processing flow of a keyword setting management table condition setting process in a multilingual document registration process;
FIG. 6 is a flowchart showing a processing flow of analysis target character string range setting processing in multilingual document registration processing;
FIG. 7 is a flowchart showing a processing flow of keyword extraction processing in multilingual document registration processing;
FIG. 8 is a flowchart showing a processing flow of unregistered keyword candidate processing in multilingual document registration processing;
FIG. 9 is a diagram showing an example of a multilingual document;
FIGS. 10A and 10B show the contents of the keyword candidate holding unit and the unregistered keyword candidate holding unit at the stage where the keyword extraction processing by the Japanese-compatible morpheme analysis unit has been completed, in comparison with each other. Figure,
FIGS. 11 (a) and 11 (b) compare the contents of the keyword candidate holding unit and the unregistered keyword candidate holding unit at the stage where the keyword extraction processing by the next English-compatible morpheme analysis unit is completed. Figure showing
FIG. 12 is a view showing an example of a created multilingual index table;
FIG. 13 is a block diagram showing a configuration of a main part of a multilingual index matching unit;
FIG. 14 is a flowchart showing an overall processing flow of multilingual document search processing;
FIG. 15 is a flowchart showing a processing flow of search word extraction processing in multilingual document search processing;
FIG. 16 is a flowchart showing a processing flow of unregistered search word candidate processing in multilingual document search processing;
FIG. 17 is a diagram illustrating an example of a search condition expression for a search condition including sentences written in multiple languages;
FIGS. 18A and 18B show the contents of the search word candidate holding unit and the unregistered search word candidate holding unit at the stage where the search word extraction processing by the Japanese-compatible morpheme analysis unit is finished. Figure showing in comparison,
FIGS. 19A and 19B show the contents of the search word candidate holding unit and the unregistered search word candidate holding unit at the stage where the search word extraction processing by the next English-compatible morpheme analysis unit is completed. Figure showing the comparison,
FIG. 20 is a diagram showing an example of a search expression generated by a multilingual search condition expression.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Input processing part, 2 ... Multilingual keyword extraction part, 3 ... Text database part, 4 ... Index registration part, 5 ... Index file part, 11 ... Search condition input part, 12 ... Multilingual index registration part, 13 ... Display 14, text extraction unit, 21 a, first morpheme analysis unit, 21 b, first dictionary file unit, 22 a, second morpheme analysis unit, 22 b, second dictionary file unit, 23 a,. Nth morphological analyzer , 23b ... Nth dictionary file part , 24 ... keyword candidate holding unit, 25 ... unregistered keyword candidate holding unit, 26 ... keyword / unregistered keyword determining unit, 27 ... order setting unit, 28 ... index registration unit, 29 ... index file unit, 30 ... keyword extraction management Table 31, number field 32, corresponding language type field 33, order field 34, use flag field 35, analysis target character string type field, 36 morphological analysis management table 37, analysis target character string type setting table 99 ... Multilingual document, 100 ... Keyword candidate storage unit, 101 ... Document (fileID), 102 ... Keyword candidate, 103 ... Unregistered keyword candidate storage holding unit, 104 ... Document (fileID), 105 ... Unregistered keyword candidate, 110 ... keyword candidate storage unit, 111 ... document (fileID , 112 ... keyword candidates, 113 ... unregistered keyword candidate storage unit, 114 ... document (fileID), 115 ... unregistered keyword candidates, 120 ... index table, 121 ... extracted keywords, 122 ... document (fileID), 123 ... Unregistered flag, 131a ... first morpheme analyzer, 131b ... first dictionary file part, 132a ... second morpheme analyzer, 132b ... second dictionary file part, 133a ... N Th morpheme analyzer 133b ... Nth dictionary file part 134 ... Search word candidate holding unit, 135 ... Unregistered search word candidate holding unit, 136 ... Search formula determining unit, 137 ... Order setting unit, 179: Search condition, 181 ... Search word candidate, 182 ... Unregistered search word candidate 191 ... Search word candidate, 192 ... Unregistered search word candidate, 200 ... Search formula, 201 ... Search key, 202 ... Unregistered word flag.

Claims (9)

In a multilingual document registration / retrieval device that creates and registers an index used for searching for a document including sentences in a plurality of languages, and searches for a document using the index,
Multilingual document storage means for storing a document including sentences in a plurality of languages;
A keyword extracting unit that controls a plurality of word segmenting units corresponding to sentences in different languages for the document, performs morphological analysis, and extracts a keyword of the document;
Index registration means for registering the keyword extracted by the keyword extraction means as an index together with the identifier of the corresponding document;
Search condition input means for inputting search conditions;
An index matching unit that cuts out a word from the search condition input by the search condition input unit and matches the extracted word with an index keyword;
A multilingual document registration / retrieval apparatus, comprising: a reading unit that reads out a document that meets a search condition based on a keyword and word collation result. The multilingual document registration / retrieval apparatus according to claim 1 ,
The keyword extracting means includes
A plurality of word segmentation means having different segmentation target languages;
Setting means for setting processing priorities of the plurality of word extracting means;
Keyword extraction control means for controlling the plurality of word extracting means according to processing priority, extracting words from a document, and extracting keywords;
A multilingual document registration / retrieval device comprising: index registration means for registering an extracted keyword and an identifier of a document from which the word of the keyword is extracted in association with each other. The multilingual document registration / retrieval apparatus according to claim 2,
The keyword extraction control unit processes a word that cannot be identified by a word extraction unit with a certain processing priority by a word extraction unit with a next processing priority. A multilingual document registration / retrieval apparatus characterized in that, for a word that cannot be identified to the end by a plurality of word segmenting means, the word is used as a keyword. The multilingual document registration / retrieval apparatus according to claim 1 , further comprising:
An input means for inputting a document to be registered and instructing keyword extraction ;
A multilingual document registration / retrieval apparatus comprising: a document to be registered; an index; and a holding unit that holds a word that is not registered in the dictionary file. The multilingual document registration / retrieval apparatus according to claim 4,
The keyword extracting means includes
A plurality of word cutout means for cutting out a word by the morphological analysis and corresponds to the statement of the respective language from the document consisting of text in multiple languages,
A plurality of dictionary files storing dictionaries corresponding to languages referred to by the plurality of word segmenting means;
Order setting means for setting the order of applying the plurality of word cutting means;
A multilingual document registration / retrieval device comprising: a control unit configured to control a plurality of word segmenting units in the order set by the sequence setting unit to perform segmenting of words of a corresponding multilingual sentence from the document. . The multilingual document registration / retrieval apparatus according to claim 5, further comprising:
Unregistered keyword candidate holding means for temporarily holding as an unregistered keyword candidate a word determined as an unregistered word by the word extracting means;
A keyword candidate holding unit that temporarily holds a word extracted from another dictionary as a keyword candidate, and the control unit includes:
The first word extraction means is controlled to input a document including sentences in a plurality of languages, and a word is extracted by morphological analysis. For a word determined to be an unregistered word, an unregistered keyword is temporarily stored. The unregistered keyword candidate holding means is stored as a candidate, and the word extracted from the dictionary is stored as the keyword candidate in the keyword candidate holding means.
Sequentially controlling each word extracting means, the unregistered word candidate held in the unregistered keyword candidate holding means by the previous word extracting means as an input, the word is extracted by morphological analysis, and unregistered words and The determined word is left as it is in the unregistered keyword candidate holding unit, the word extracted from the dictionary is deleted from the unregistered keyword candidate holding unit, and additionally stored in the keyword candidate holding unit.
Finally, the keyword candidate held in the keyword candidate holding unit is set as a keyword, and the unregistered keyword held in the unregistered keyword candidate holding unit is registered as an unregistered keyword together with the identifier of the corresponding document. Multilingual document registration and retrieval device. In a multilingual document registration / retrieval device that creates and registers an index used for searching for a document including sentences in a plurality of languages, and searches for a document using the index,
A text database for storing documents containing sentences in multiple languages;
A keyword extracting unit that controls a plurality of word segmenting units corresponding to sentences in different languages for the document, performs morphological analysis, and extracts a keyword of the document;
Index registration means for registering the keyword extracted by the keyword extraction means as an index together with the identifier of the corresponding document;
Search condition input means for inputting search conditions;
An index matching unit that cuts out a word from the search condition input by the search condition input unit and matches the extracted word with an index keyword;
A multilingual document registration / retrieval apparatus comprising: an extraction unit that extracts a corresponding document from the text database based on a collation result of the index collation unit. The multilingual document registration / retrieval apparatus according to claim 7,
The index matching means includes
A plurality of word extraction means for extracting words by performing morphological analysis on sentences in a corresponding language from documents composed of a plurality of languages;
Order setting means for setting the order of applying the word cutting means by combining a plurality of word cutting means;
A multilingual document registration / retrieval apparatus, comprising: a control unit that performs control for extracting words of the search condition input by the search condition input unit in the order set by the order setting unit. The multilingual document registration / retrieval apparatus according to claim 8, further comprising:
Unregistered search word candidate holding means for temporarily holding the word determined as an unregistered word by the word cutting means as an unregistered search word candidate;
Search word candidate holding means for temporarily holding as a search word candidate for words extracted from other dictionaries, the control means,
The first word extraction unit is controlled to input a document including sentences in a plurality of languages, extract words by morphological analysis, and temporarily search for unregistered words determined to be unregistered words A word candidate is held in the unregistered search word candidate holding means, and a word extracted from the dictionary is processed as a search word candidate and held in the search word candidate holding means,
Sequentially controlling each word segmentation means, using the unregistered word candidate held in the unregistered search word candidate holding means by the previous word segmentation means as input, segmenting words by morphological analysis, and unregistered words For words determined to be left as they are in the unregistered search word candidate holding means, words extracted from the dictionary are deleted from the unregistered search word candidate holding means and additionally held in the search word candidate holding means. Process,
Finally, the search word candidate held in the search word candidate holding means is used as a search word, the unregistered search word held in the unregistered search word candidate holding means is used as an unregistered search word, and index matching is performed. A multilingual document registration / retrieval apparatus for extracting a document by a text database unit and outputting result information.

Images