JP2007257644A - Program, method and device for acquiring translation word based on translation word candidate character string prediction - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a program, a method and a device for acquiring translation words based on translation word candidate character string prediction. <P>SOLUTION: This translation word acquisition program functions a computer having a word element parallel translation dictionary which stores correspondence relation between a word element of an original language and a translation word element which is a word element of a target language as a word element division part which divides query of the input original language into word elements, a translation word element extracting part which extracts the corresponding translation word element by reading and referring to the word element parallel translation dictionary to the divided word elements, a logic operation expression construction part which constructs a logic operation expression for retrieval by combining the extracted translation word element and the query, a retrieval part which retrieves a document matched to a query 2 from a network or a data base by using the constructed logic operation expression as the query 2 and a translation word candidate acquisition part which extracts the translation word candidates from the retrieved document. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to computer information processing, and more particularly to processing that uses web mining to discover knowledge. In particular, the present invention estimates a translation of a target language (referred to as a translation target language) for a specific phrase (entity item) from a normal document in a network or database, and the translation in the translation The translation acquisition efficiency is improved by guessing the word element. Here, the entity item is, for example, a word, a proper noun, an idiomatic phrase, or a phrase. A word element is a minimum unit having a finer meaning than a normal word.

（自動車のナンバープレート：英語ではLicense plate number）、
［文字２］

（三国志：英語ではThe Romance of Three Kingdoms）等がある。機械翻訳システムは通常、（機械翻訳辞書として保有していない）未知な表現に出会うとシステム全体の翻訳の正確性が突然低下する。また、クロスリンガル情報検索では、クエリ語が辞書に収録されておらず当該クエリ語に対応する訳語が存在しないと、これがボトルネックとなって情報検索システム全体の性能の向上が妨げられてしまう。 When writing in a foreign language, we often encounter technical terms, proper nouns, idiomatic phrases and phrases that are not listed in general dictionaries. Finding an accurate translation of such an expression can take a long time, even if you search a dictionary or search a document. As words that are difficult to find a translation,
[Character 1]

(Automobile license plate: License plate number in English),
[Character 2]

(Sangokushi: The Romance of Three Kingdoms in English). A machine translation system usually encounters an unknown expression (not held as a machine translation dictionary) and suddenly reduces the accuracy of translation of the entire system. Further, in the cross-lingual information search, if a query word is not recorded in the dictionary and there is no translation corresponding to the query word, this becomes a bottleneck and prevents the performance of the entire information search system from being improved.

  Therefore, it is extremely important to obtain translations of important entry words (unregistered words) that are not recorded in the dictionary. In order to solve such problems, there are those skilled in the art who have considered using a network search engine, but a large amount of irrelevant web pages and redundant information are searched to find useful translations and related knowledge that users need. Was extremely difficult. One of the objects of the present invention is to estimate a large amount of information on the Internet by predicting a character string (translation word word) that is likely to be included in a translation word in order to eliminate unrelated web pages in the translation search. Using resources, it is possible to efficiently obtain bilingual translations that are not in the dictionary using a search engine.

There are roughly the following types of techniques for acquiring corresponding pairs of translated words in two languages or techniques for translating between two languages.
(1) A translation is acquired from a bilingual corpus. In this method, a large-scale bilingual corpus must be constructed. Some researchers have proposed building an automatically aligned corpus from a corpus obtained from the net to reduce the effort of manually building a bilingual corpus. However, it is quite difficult to construct a corpus that is compatible with unstructured data on the net.

  (2) Acquire translations from a non-translated corpus. This method obtains a translation using the fact that when a large-scale balanced corpus is used, the context between two languages is the same or similar for a certain word. According to this method, the corpus existing at that time can be fully utilized while avoiding the problem of a shortage of parallel corpus. However, it is inferior to the above-described method in terms of the effect of acquiring translated words. Furthermore, as can be seen from the research results to date, this method mainly focuses on the translation of individual words.

  (3) A phrase translation is acquired from a combination of translation candidates for each phrase. The candidate translations of each word constituting the phrase are combined, and the appropriateness of each combination is evaluated. This method is used for acquiring translated words such as noun collocations, but often fails.

  (4) Acquire translations by transliteration or transliteration. This method is used to obtain translations between two similar languages. For example, in order to obtain a Chinese translation from a Japanese Kanji word, it may be only necessary to convert between characters (transliteration). To translate English into Japanese, the pronunciation of English can be expressed in katakana (transliteration).

  (5) A translation is acquired based on anchor text information. This method uses the hypothesis that anchor texts linked to the same web page are likely to have the same meaning. When anchor texts in multiple languages are extracted, they represent the same concept. It is regarded as a word (related to a translated word). It works relatively well when you get translations for company names that have websites written in multiple languages.

  The above is the background technology related to the translation extraction from the network that is the basis of the present invention, and the technology that is the basis of the present invention is described next.

  (6) Acquisition of translations from partially bilingual texts on the Internet. In scientific and technical papers and specialized articles, especially in Asian languages such as Chinese, Japanese, and Korean, technical terms and well-known names such as “Acquisition of translations from partially bilingual texts” are well known. Often no proper nouns are given a corresponding English translation. As documents in specialized fields such as scientific and technical papers are released on the Internet, web mining techniques for acquiring translated words such as technical terms from contents on the Internet have become effective means.

  The system described in Non-Patent Document 1 describes means for obtaining a parallel translation of a target language e of a word j in the source language. From all documents including the word j, a document including at least one target language word Thus, for every word e of the target language that appears in the document, the likelihood of the translation between the word j and the word e is calculated, and the word e that seems to be the most translated word is obtained. In reality, instead of all documents, use a search engine to narrow down to a specific number of documents, and use the distance between j and e to determine the likelihood of translation between j and e in each document. It has been shown.

  In the system described in Non-Patent Document 2, in a cross-language information search in which a Chinese query is input and an English document is searched, there may be an unregistered word in the query. This is to solve this problem (maintaining the performance of cross-language search) by using a document including a partial translation. This can be said to show one application of translation search using a network.

  Non-Patent Document 3 also deals with the problem of maintaining cross-language search performance using network search for the problem of the presence of unregistered words in a query in cross-language information search.

M.Nagata, T.Saito, and K.Suzuki, Using the Web as a Bilingual Dictionary, Proc.ACL 2001 Workshop Data-Driven Methods in Machine Translation, 2001 P.J.Cheng, et al. Translating unknown queries with web corpora for cross-language information retrieval.SIGIR 2004 Y.Zhang, P.Vines, RMIT Chinese-English CLIR at NTCIR-4, In Working Notes of the Fourth NTCIR Workshop Meeting

  As described above, partial translations on the network are used to obtain translations of technical terms and proper nouns, and a network using source language words as queries to obtain documents containing these partial translations. Search is being used. However, there are a large number of documents including source language words (search result documents obtained by searching source language words as queries), and it is not realistic to download all of them. Therefore, in the above-mentioned document, only a realistic number of documents is used by the cut-off using the ranking of the search engine.

  However, search engine rankings are ranked according to the frequency of words given as a query, the prominence of the site of the document that contains the query word (eg Google's PageRank algorithm), and so on. However, it does not always include partial translations. Rather, in general, it can be said that there are more documents that do not contain partial translations.

  In order to maintain the accuracy of translation extraction, it is necessary to guarantee the number of documents including partial parallel translations. To do so, a large number of documents are extracted (downloaded) from the search results and checked for partial translations. This requires more time for processing. It is necessary to devise in order to maintain the accuracy of translated word extraction and not increase the processing time.

  The present invention has been made to solve the above-described problems of the prior art, and in order to efficiently search for documents including partial parallel translations, translation candidate candidates are estimated and only documents including these are searched. An object of the present invention is to provide a translated word acquisition method and a translated word acquisition apparatus for efficiently retrieving a document including a partial parallel translation by giving a query to a search engine.

  The source language query (word or phrase) is divided into word elements (the smallest unit of meaning that is finer than ordinary words), and the target language corresponding to each word element of the query using the word parallel translation dictionary A conventional network is constructed by acquiring a set of word candidates and constructing a logical expression for search so that a document including not only the original query but also at least one of the word candidates is searched from here. Compared with the translated word search method using, it is possible to improve the search efficiency of a document having a partial translation for a query (the rate of having a partial translation of a query in a retrieved document). As a result, if the same cut-off as in the prior art is performed, a significant improvement in translation accuracy can be expected, or if the same number of valid documents are taken out, a significant reduction in processing time is realized.

  Exemplary embodiments of a translated word acquisition program, a translated word acquisition method, and a translated word acquisition apparatus according to the present invention will be described below in detail with reference to the accompanying drawings.

  FIG. 2 is a block diagram of the translated word acquisition apparatus based on the translated word candidate character string prediction according to the embodiment of the present invention. As shown in FIG. 2, the translation acquisition device according to the embodiment of the present invention includes a query reception unit 10, a word element division unit 12, a word element dictionary 14, a translation word element set extraction unit (translation word element extraction unit) 16, A word parallel translation dictionary 18, a logical operation expression construction unit 20, a search unit 22, and a translation word candidate acquisition unit 24 are provided. Further, the translation candidate acquisition unit 24 includes a document download unit 30, a translation candidate extraction unit 32, and a translation candidate evaluation unit 34.

  The query receiving unit 10 receives a word for which a translated word is to be checked as a query. The character string input as the query is displayed on the display unit as shown in FIG. 3, for example. The word element dividing unit 12 divides an input word or phrase (query) into word elements. The word element bilingual dictionary 18 holds a target word element set corresponding to each word element in the source language. The translated word element set extraction unit 16 refers to the word element bilingual dictionary 18 for each of the word elements divided and output by the word element dividing unit 12, thereby obtaining the corresponding word word translated word set. Take out. The logical operation expression constructing unit 20 combines a set of translated word element sets corresponding to each word element of the source language extracted by the translated word element set extracting unit 16 and a query inputted first, for search. Build logical expressions. The retrieval unit 22 retrieves a document that matches the query from the network or database using the constructed logical operation expression as a query. The translation candidate acquisition unit 24 downloads the document searched / acquired by the search unit 22, finds and extracts a partial parallel translation for the query, creates a parallel translation candidate together with its evaluation value, and extracts from the parallel translation candidates extracted from a plurality of documents The likelihood of the translation candidate is calculated using various heuristics, and the translation candidate of the query is extracted based on the calculation result.

  The input to the query receiving unit 10 may be, for example, inputting a word or phrase using a keyboard or the like. It is also possible to input words and phrases through a network such as a LAN or the Internet. When inputting words or phrases via a network, the input unit may be configured in the form of a network interface. In addition, words and phrases can be input from a storage device such as a hard disk drive or a scanner. In this case, the input unit may be connected to a scanner, a storage device, or the like to perform data communication. The input unit can be connected by, for example, a wired connection such as USB (Universal Serial Bus) or a wireless connection such as Bluetooth. It is also possible to input phrase items stored in a storage medium such as a flash memory, a floppy (registered trademark) disk, a CD (compact disk), and a DVD (digital versatile disk, digital video disk). In this case, the input unit may be configured as a device that reads data from a storage medium. Examples of the data reader include a flash memory reader, a floppy (registered trademark) disk drive, a CD drive, and a DVD drive.

  Furthermore, the input unit may be configured to be applicable to a plurality of cases.

  The translated word information output by the translated word candidate acquisition unit can be output through a network. In this case, the output unit is configured to include a network interface. Also, the translated word information can be output to other information processing devices such as personal computers and storage devices. In this case, the output unit is configured to perform data communication with another information processing apparatus such as a personal computer and a storage device. Also, the translated word information can be output / written on the storage medium. In this case, the output unit is configured as a writing device that writes data to a storage device or a storage medium. Examples of the writing device include a flash memory recorder, a floppy (registered trademark) disk drive, a CD-R drive, and a DVD-R drive.

  Further, the output unit may be configured as an interface for performing data communication with the display device or connected to the display device in order to output and display the translated word information on a display device such as a monitor so that the data can be used. The interface may be configured to transmit data to the built-in information processing apparatus.

  Furthermore, the output unit may be configured to be applicable to a plurality of cases.

  FIG. 4 is a flowchart showing a processing procedure of the word element division processing in the word element dividing unit 12. In the word segmentation process, the word segmentation unit 12 receives the input query as the target of the word segmentation, and divides the word or phrase that is the target of the word segmentation into the word units that are the smallest meaningful units To do. The word segmentation method may be the same as a method generally called morphological analysis. The difference from the morpheme analysis is that the morpheme analysis uses the morpheme dictionary, whereas the word element division uses the word element dictionary. For example, “Sangokushi” is considered as one word (morpheme), but the word element can be divided into “three”, “country”, and “shi”. Most kanji characters are considered to be the smallest unit that has meaning in a single character, so a word dictionary may not be prepared, and it may be simply divided into one kanji unit. In this case, only a single character string that has no minimum meaning in the dictionary is stored in the dictionary, and only when they match, those character strings are used as word elements, and the remaining kanji characters are divided one by one. It may be a thing. Specifically, when receiving the character string to be divided, the word element dividing unit 12 sets the received character string to the target character string s (step S401), and sets the division list L to an empty list (step S402). If the target character string s is empty (step S403, Yes), the process is terminated. If the target character string s is not empty (step S403, No), the word dictionary is used from the beginning of the target character string s. A character string (word element) that matches is extracted (step S404). The extracted character string is added to the divided list L (step S405), while the matched partial character string is removed from the target character string s (step S406). And the process after step S403 is repeated again.

  The word-element bilingual dictionary 18 holds a set of word elements and their parallel translations. The retained sets are, for example, “three” and “three, third, triple”, “country” and “nation, country, land, state, world”, and the like.

  The translation word element set extraction unit 16 creates a list of translation word candidates by referring to the word element bilingual dictionary for each word element extracted from the word element division unit 12. FIG. 5 shows a processing procedure of a translation word candidate list creation process by the translation word element set extraction unit 16. Specifically, when the translated word element set extraction unit 16 receives the word element list, it sets the received word element list in the list L (step S501), and further emptyes the translated word list T (step S502). If the list L is empty (step S503, Yes), the translated word list T is returned and the process is terminated. If the list L is not empty (step S503, No), the word element w is extracted from the list L (step S503). Step S504). Next, referring to the word element bilingual dictionary, the translated words t1 to tn of the word element w are added to the translated word list T (step S505). Then, the word element w is removed from the list L, and the processes after step S503 are repeated again.

という入力に対して、訳語リスト｛three,third,triple,nation,country,land,state,world,...｝が出力された場合には、論理演算式、「“
［文字２］

”＆（"three"|"third"|"triple"|"nation"|"country"|"land"|"state"|"world"|...）」が構築され出力される。 The logical operation expression construction unit 20 constructs a logical operation expression for search from the source language query and the translation word list extracted by the translation word element set extraction unit 16. The logical operation expression is output in a format that can be interpreted by the search engine of the search unit 22 in the subsequent stage. For example,
[Character 2]

When the translated word list {three, third, triple, nation, country, land, state, world, ...} is output, the logical operation expression ““
[Character 2]

"&(" Three "|" third "|" triple "|" nation "|" country "|" land "|" state "|" world "| ...)" is constructed and output.

The retrieval unit 22 retrieves a document that matches the query from the network or database using the logical operation expression constructed as the logical operation expression as a query (query 2). Specifically, for example, the search is realized by giving url as follows in a format according to the API of the search engine. Japanese, Chinese, etc. are encoded according to url encoding.
http://www.google.com/search?hl=en&q=%E4%B8%89%E5%9B%BD+%28three+%7C+third%29&btnG=Google+%E6%A4%9C%E7%B4 % A2 & lr =

  The document download unit 30 downloads the search result. Usually, an appropriate number of documents are downloaded from the higher ranking document according to an appropriate threshold (number of documents or time limit).

  The translation candidate extraction unit 32 extracts translation candidate words for the query. FIG. 6 shows a processing procedure of translated word candidate extraction processing by the translated word candidate extracting unit 32. Arbitrary character strings in the target language are candidates, but in practice, appropriate narrowing is performed using heuristics. As the heuristic, as shown in Non-Patent Document 1, the distance from the query, the difference in the character string length between the query and the target language, the presence or absence of a specific symbol or expression, and the like are used. Specifically, the translated word candidate extraction unit 32 searches for a query character string in the document (step S601), extracts a character string within the preceding and following threshold values (step S602), and extracts only the target language in the extracted character string. A character string is taken out in units in which no other language is sandwiched (step S603). Further, a partial word string is extracted as a candidate for each extracted character string (step S604). Finally, unnecessary candidates are discarded based on the heuristic (step S605).

に対して、検索文書として、
［文字３］

が検索文書中にあれば、"Kingdoms"、"Three Kingdoms"、"of Three Kingdoms"、"Romance of Kingdoms"、"Book"、"online"、"online read"、"online read book"等が候補として抽出される。 For example, query [character 2]

As a search document,
[Character 3]

"Kingdoms", "Three Kingdoms", "of Three Kingdoms", "Romance of Kingdoms", "Book", "online", "online read", "online read book", etc. Extracted as

  The heuristic example described above is merely an example, and the query translation word candidates can be narrowed down by other methods. For example, it is possible to introduce a heuristic that discards a character string that starts with a preposition from among candidate character strings, and in that case, “of Three Kingdoms” is removed from the translation candidates.

  The translated word candidate evaluation unit 34 evaluates the likelihood of the translated words of each translated word candidate. For example, the simplest evaluation is to calculate the frequency of each candidate. In addition, when a candidate character string is extracted, an individual evaluation value is given, and the weighted frequency can be obtained.

」（stockholder）の省略とも考えられる。そこで、原語候補抽出部では、省略復元辞書を用いて、「股」という語素に対して、「股票」、「
［文字４］

」も語素に加える。ここでは、原語候補を抽出する実施例を示したが、語素対訳辞書１８および訳語語素集合取り出し部１６の処理を混合させ、「股」に対して直接、「section」「thigh」という訳語候補を取り出す際に「stock」と「stockholder」を付与するという方法であっても構わない。 The translated word acquisition apparatus according to the embodiment of the present invention further includes a source word candidate extraction unit, and the source word candidate extraction unit refers to the abbreviated restoration dictionary for the given word element and assigns the original word to the word element. To do. The abbreviated restoration dictionary is a dictionary in which a word before abbreviation, for example, “university” with respect to “dai” is associated. For example, the word “distribution crotch” is divided into two word elements “distribution” and “crotch”. Here, the translation of the word element “thigh” is “section” and “thigh”, but this word element “crotch” is “crotch vote” (stock) or “
[Character 4]

"(Stockholder) may be omitted. Therefore, the original word candidate extraction unit uses the abbreviated restoration dictionary to perform “crotch vote”, “
[Character 4]

Is also added to the word element. In this example, the source word candidates are extracted. However, the word pair translation dictionary 18 and the translation word set extraction unit 16 are mixed to directly translate “section” and “thigh” into “translation”. A method of adding “stock” and “stockholder” when extracting candidates may be used.

」を構成するアイテムそれぞれの対応訳語は、「三」については「three」、「国」については、「country、nation」、「演」については「act、practice」、「
［文字５］

」については「meaning、justice」となり、人は、「三つの国に関すること」なのだろうと大まかなイメージを持つことができる。また、「
［文字１］

」を構成する語素それぞれの対応訳語は、「
［文字６］

」については、「vehicle、car」、「牌」については、「brand、plate、board」、「号」については「number、size、data」、となり、人は、「車の番号」なのだろうと大まかなイメージを持つことができる。 Many Chinese terminology and proper nouns are composed of many words or phrases. It is extremely difficult to get its exact meaning directly from each word or phrase, but guessing a partial or related meaning and word from the word or phrase that is a component of that word It is possible to do. For example, "
[Character 2]

"Three" for "three", "country, nation" for "country", "act, practice", "
[Character 5]

"Meaning, justice", and people can have a rough image that it is about "three countries". Also,"
[Character 1]

The corresponding translations of the word elements that make up "
[Character 6]

”For“ vehicle, car ”,“ 牌 ”for“ brand, plate, board ”,“ No. ”for“ number, size, data ”, and for people to be“ car number ” You can have a rough image.

”＋（three(country(nation(act(practice(justice(meaning）」、「“
［文字１］

”＋（vehicle(car(brand(plate(board(number(size(data）」となる。 In this translation candidate prediction method, a query is expanded using a word or translation predicted by the above-mentioned word-pair bilingual dictionary as a keyword, and a translation candidate is selected from documents searched using the expanded query. Extraction and statistical processing are performed to obtain words closely related to the original query word in these documents to obtain a document including a partial translation.
The generated logical operation expression is as follows. ““
[Character 2]

"+ (Three (country (nation (act (practice (justice (meaning)"",""
[Character 1]

"+ (Vehicle (car (brand (plate (board (number (size (data))".

  Further, in each embodiment, each operation is two connected by one or more processors or specific circuits or connections (eg, individual logic gates interconnected to perform a specific function). It can be executed by a combination of processors. Accordingly, each aspect of the invention can be implemented in different embodiments and all these embodiments are considered to be within the scope of the invention disclosed herein. Examples of such embodiments, for each aspect of the present invention, mean “logic configured to perform the operation” or “logic that performs or can execute the operation”.

  Furthermore, according to the embodiment of the present invention, the object of the present invention can also be realized by a computer-readable medium that executes the program. A computer-readable medium is a computer readable medium that contains or stores a program such that it can be combined with a system, device, or apparatus that executes instructions, or can be used to cause instructions to be executed by the system, device, or apparatus. Means any machine that can communicate, transmit or send. For example, a computer-readable medium can be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus or distribution medium. More specific examples of computer readable media include (but are not a comprehensive list) electrical connections with one or more lead wires, portable computer disks, random access memory (RAM), read only memory (ROM). ), Erasable programmable read only memory (EPROM), optical fiber, and compact disc read only memory (CDROM).

  It should be noted that the above description of the embodiments of the present invention is for illustration and description only. The above description is not intended to be exhaustive or to limit the invention to the particular embodiments disclosed. Various modifications and variations of this invention will be apparent to those skilled in the art. This embodiment best describes the principles of the invention and its practical applications, and allows one skilled in the art to understand various embodiments and variations of the invention, thereby providing specific and anticipated applications. It has been selected and explained to be able to do so. It should be understood that the scope of the invention is defined by the appended claims and their equivalents.

  As described above, the translation acquisition program, translation acquisition method, and translation acquisition device based on the translation candidate character string prediction according to the present invention are suitable for processing using web mining to find knowledge, The present invention is useful for a translation acquisition method and a translation acquisition apparatus for estimating a translation of a target language from a normal document in a network or database for a phrase (entity item).

FIG. 1 is a diagram showing the principle of a translated word acquisition apparatus according to the present invention. FIG. 2 is a block diagram of the translated word acquisition apparatus based on the translated word candidate character string prediction according to the embodiment of the present invention. FIG. 3 is a diagram illustrating an input position of a character string accepted as a query. FIG. 4 is a flowchart showing a processing procedure of word element division processing by the word element dividing unit. FIG. 5 is a flowchart showing a processing procedure of a translation word candidate list creation process by the translation word element extraction unit. FIG. 6 is a flowchart showing a processing procedure of translated word candidate extraction processing by the translated word candidate extraction unit.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Query reception part 12 Word element division part 14 Word element dictionary 16 Translation word element set extraction part 18 Word element parallel translation dictionary 20 Logical operation expression construction part 22 Search part 24 Translation word candidate acquisition part 30 Document download part 32 Translation word candidate extraction part 34 Translation word Candidate evaluation department

Claims (6)

A computer having a word-element bilingual dictionary for storing a correspondence relationship between a word element of a source language and a target word element.
A word segmentation unit that divides an input source language query into word elements,
A translated word element extraction unit that extracts a corresponding translated word element by reading and referring to the word element bilingual dictionary for the divided word elements,
A logical operation expression constructing unit that constructs a logical operation expression for search by combining the extracted translation word element and the query;
A retrieval unit that retrieves a document that matches the query 2 from the network or database using the constructed logical operation expression as the query 2,
A translation candidate acquisition unit that extracts translation candidates from the retrieved document;
Translation program to function as A computer further having an abbreviated restoration dictionary for storing a correspondence between an abbreviated word element and an original word candidate that is an abbreviated previous word element;
Reading the abbreviated restoration dictionary, taking out the original word candidate for the word element divided by the word element dividing unit, and as the original word candidate extracting unit using the original word candidate as a new word element,
The translated word acquisition program according to claim 1, further functioning. A word-to-word translation dictionary that stores correspondences between source language word elements and target language word elements,
A word segmentation unit that divides an input source language query into word segments;
A translated word element extraction unit that retrieves a corresponding translated word element by reading and referring to the word element bilingual dictionary for the divided word elements;
A logical operation expression construction unit that constructs a logical operation expression for search by combining the extracted translation word element and the query;
A retrieval unit that retrieves a document that matches the query 2 from the network or database using the constructed logical operation expression as the query 2,
A translation candidate acquisition unit that extracts translation candidates from the retrieved document;
Translated word acquisition device. An abbreviated restoration dictionary that stores the correspondence between the omitted word elements and the original word candidates that are the previous word elements that are omitted;
Reading the abbreviated restoration dictionary, extracting a source word candidate for a word element divided by the word element dividing unit, and a source word candidate extracting unit that uses the original word candidate as a new word element;
The translated word candidate device according to claim 3, further comprising: A computer having a word-element bilingual dictionary storing a correspondence relationship between a source language word element and a target language word element translated word element,
Divide the input source language query into word elements (referred to as a word element division process)
With respect to the divided word elements, the corresponding word word elements are extracted by reading out and referring to the word element bilingual dictionary,
A logical expression for search is constructed by combining the extracted translation word element and the query,
The constructed logical operation expression is used as a query 2, and a document matching the query 2 is retrieved from a network or a database.
Extract translation candidates from the retrieved document,
The translation acquisition method characterized by this. A computer further comprising an abbreviated restoration dictionary for storing a correspondence between an abbreviated word element and an original word candidate that is an abbreviated previous word element,
After the word element division step, the abbreviated restoration dictionary is read out, a source word candidate for the word element divided by the word element division unit is extracted, and the original word candidate is set as a new word element.
The translated word acquisition method according to claim 5.

Images