JP2005258637A - Translation word extraction device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a translation word extraction device and program for extracting, with high precision and on the basis of phonetic similarities, translation words in a second language that correspond to expressions written in a first language, from documents in the second language. <P>SOLUTION: The translation word extraction device, for extracting from documents in the second language translation words in the second language that correspond to expressions in the first language, includes an extraction means for extracting from previously stored documents in both the first and second languages the documents in the first language including expressions inputted in the first language and the documents in the second language corresponding to the documents in the first language; a first conversion means for converting the documents in the first language into phonetic expressions; a second conversion means for converting the documents in the second language into phonetic expressions; a similarity calculation means for calculating similarities between the phonetic expressions obtained by the first conversion means and the phonetic expressions obtained by the second conversion means; an occurrence frequency obtaining means for obtaining the occurrence frequency of the documents in the first language that are included in the documents in the second language; and an output means for outputting translation words on the basis of the similarities obtained by the similarity calculation means and the occurrence frequency obtained by the occurrence frequency obtaining means. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a bilingual word extraction device and a bilingual word extraction program, and in particular, to extract a bilingual word corresponding to an expression made in the first language with high accuracy from a document in the second language. The present invention relates to an apparatus and a parallel word extraction program.

  Conventionally, proper nouns and the like are often translated into foreign words representing the pronunciation of the original word among languages with different characters such as English and Japanese. For this reason, there has been proposed a method of extracting parallel translation words based on pronunciation information of words between two languages.

  For example, Japanese and English are converted into phonetic symbols to extract matching parts (for example, see Non-Patent Document 1), or Japanese and Korean are converted into phonetic symbols to extract similar parts. There is a method (for example, see Non-Patent Document 2).

Here, when expressed in another language as a foreign language, the pronunciation is similar, but the pronunciation system such as the type and number of vowels and consonants is often different depending on the language, and it is not a phonetic character In many cases, pronunciation estimation is not easy. Therefore, even if the pronunciation of a word expressed in two languages is converted into an expression of the same pronunciation system, it does not always match completely. Therefore, it is important to accurately compare the similarity of pronunciation.
Matsuo Yoshihiro, Shirai Satoshi, 1996, “Automatic Extraction of Word Pairs Using Pronunciation Information”, Report of IPSJ Study Group, NL-116-15, pp. 101-106. Kintamanishiki, Satoshi Fujii, Tetsuya Ishikawa, 2003, “Automatic extraction of foreign words from Korean corpus and application to linguistic analysis”, 9th Annual Meeting of the Association for Natural Language Processing, pp. 258-261.

  However, the method described in Non-Patent Document 1 extracts only the phonetic expression that matches all or the consonant strings that match all. In this method, extraction cannot be performed when some of the data do not match. Therefore, it is not possible to realize a highly accurate parallel translation. Further, the method described in Non-Patent Document 2 calculates a similarity between pronunciation expressions and extracts a translated word. At this time, in calculating the similarity between the strings of two phonetic expressions, the minimum number of differences between the strings due to insertion, replacement, and deletion with the importance of the consonant and the vowel differing as the consonant is an important element. Although the normalized value is used, the method for efficiently extracting the upper solution has not been clarified.

  The present invention has been made in view of the above-described problems, and is based on the similarity of pronunciation of a second language translation corresponding to an expression written in the first language based on the similarity of pronunciation in the second language document. It is an object of the present invention to provide a bilingual word extraction device and a bilingual word extraction program for extraction into a word.

  In order to solve the above problems, the present invention employs means for solving the problems having the following characteristics.

  According to a first aspect of the present invention, there is provided a bilingual word extraction apparatus that extracts a translation of a second language corresponding to an expression of the first language from a document of the second language. A first language document including an expression inputted in the first language, a second language document that is a bilingual translation of the document, and a pronunciation of the first language document. Obtained by the first conversion means for converting into a representation, the second conversion means for converting the document in the second language into the pronunciation expression, the phonetic expression obtained by the first conversion means and the second conversion means. A similarity calculation unit that calculates a similarity with a pronunciation expression, an appearance frequency acquisition unit that acquires an appearance frequency of a document in the first language included in the document in the second language, and the similarity calculation unit Obtained by the similarity and the appearance frequency acquisition means Based on the occurrence frequency, and having an output means for outputting the translation.

  According to the first aspect of the present invention, the translation of the second language corresponding to the expression written in the first language can be extracted from the document of the second language with high accuracy based on the similarity of pronunciation. .

  In the invention described in claim 2, when the similarity calculation unit calculates the similarity between the pronunciation expression obtained by the first conversion unit and the pronunciation expression obtained by the second conversion unit, The editing cost is calculated based on the expression length of the pronunciation expression.

  According to the second aspect of the present invention, the similarity can be calculated with high accuracy based on the editing cost. Thereby, a bilingual word can be extracted with high precision.

  In the invention described in claim 3, when the similarity calculation unit calculates the similarity between the pronunciation expression obtained by the first conversion unit and the pronunciation expression obtained by the second conversion unit, The similarity is calculated based on the ratio of the editing costs of insertion and deletion learned from the bilingual document of the first language and the second language.

  According to the third aspect of the present invention, the similarity can be calculated with high accuracy based on the ratio of the editing costs of insertion and deletion. Thereby, a bilingual word can be extracted with high precision.

  The invention described in claim 4 is characterized in that, when the editing cost is calculated by replacement, the similarity calculation means changes and sets the editing cost by a combination of characters before and after replacement.

  According to the fourth aspect of the present invention, the similarity can be calculated with high accuracy by changing the cost for replacement in accordance with the combination of characters before and after the replacement. Thereby, a bilingual word can be extracted with high precision.

  The invention described in claim 5 is stored in advance in a parallel word extraction program for causing a computer to execute a process of extracting a translation of a second language corresponding to an expression of the first language from a document of the second language. An extraction process for extracting a document in a first language including an expression inputted in the first language, and a document in a second language in a parallel translation of the document, from a parallel translation document in the first language and the second language; A first conversion process for converting a document in the first language into a phonetic expression; a second conversion process for converting the document in the second language into a phonetic expression; and a phonetic expression obtained by the first conversion process; Similarity calculation processing for calculating the similarity to the pronunciation expression obtained by the second conversion processing, and appearance frequency acquisition processing for acquiring the appearance frequency of the first language document included in the second language document And obtained by the similarity calculation process Based on the appearance frequency obtained by the frequency acquisition process similarity, to perform an output process for outputting the translation to the computer.

  According to the fifth aspect of the present invention, the translation of the second language corresponding to the expression written in the first language can be extracted from the document of the second language with high accuracy based on the similarity of pronunciation. . Also, by installing the program, it is possible to easily realize the bilingual word extraction process.

  According to the present invention, a translation of a second language corresponding to an expression written in the first language can be extracted with high accuracy from a second language document based on the similarity of pronunciation.

<Outline of the present invention>
The present invention extracts translations of the second language corresponding to expressions written in the first language from the second language document using pronunciation similarity. Here, the extraction method first converts expressions in both languages into pronunciation expressions using a conversion table in units of characters or partial character strings. Then, a second language expression having a high similarity between pronunciation expressions is extracted as a candidate for a translation word. If the similarities are equal, the documents are ranked according to the frequency of the number of appearing documents in the second language document and output.

  As a result, it is possible to extract the translation word of the second language translation word corresponding to the expression written in the first language with high accuracy from the second language document based on the similarity of pronunciation.

  In the present invention, “the expression in the first language and the translation in the second language corresponding to the expression” are “both words generated by transliteration or transliteration”, “the first language. The expression is a word in the source language (not a word generated by translation, but a word specific to that language), and a translation in the second language is a word generated by transliteration or transliteration. " "The expression in one language is a word generated by transliteration or transliteration, and the translation in the second language is a word in the source language."

<Embodiment>
Next, embodiments of the present invention will be described with reference to the drawings. In the embodiment described below, as an example, Japanese is used as the first language and Chinese is used as the second language, and an example of outputting a bilingual Chinese word corresponding to Japanese katakana input is used. In other words, in this embodiment, the expression for which the translated word is desired is written in Japanese, and the translated word is in Chinese. Also, consider the case where Roman letters are used as pronunciation expressions. The present invention is not limited to combinations of these languages and pronunciation expressions.

  FIG. 1 is a diagram illustrating a configuration example of a parallel word extraction apparatus according to the present invention. The bilingual word extraction device 10 shown in FIG. 1 includes a Chinese document acquisition unit 11, a first pronunciation expression conversion unit 12, a second pronunciation expression conversion unit 13, a pronunciation similarity calculation unit 14, a Chinese expression The restoration unit 15, the appearance frequency acquisition unit 16, and the translated word rank selection unit 17 are configured.

  First, the Chinese document acquisition unit 11 receives a Japanese document including a katakana expression inputted from a bilingual bilingual corpus (document-to-corpus) 21 stored in advance for katakana inputted to extract a translated word. Get the Chinese translation of the bilingual. In addition, the Chinese document acquisition unit 11 outputs the acquired Chinese document as a Chinese corpus 22.

  Next, the first phonetic expression conversion unit 12 converts the acquired Chinese document into phonetic symbols (Roman characters) in units of characters using the Kanji / Romaji conversion table 23 stored in advance. Here, the kanji / romaji conversion table 23 is a conversion table in which romaji corresponding to one kanji is registered.

  If there are a plurality of pronunciation expression candidates to be converted, all of them are output. For example, if the pronunciation expression (Roman character) for the Chinese character “A” is “a” and “e”, both are output. Further, the pronunciation expression (Roman character) is not limited to one character. For example, the pronunciation expressions (Roman characters) for the Chinese character “kai” are “hui”, “kuai”, and “kuo”. The first phonetic expression conversion unit 12 outputs the Chinese corpus 24 of the phonetic expression converted by the conversion table as described above.

  On the other hand, the second phonetic expression conversion unit 13 to which the input katakana is input to extract the translated word uses the katakana / romaji conversion table 25 stored in advance for the input katakana to generate a character or a partial character. Convert to phonetic expression (romaji) in units of columns. In addition, the second pronunciation expression conversion unit 13 outputs the converted pronunciation expression to the pronunciation similarity calculation unit 14.

  The pronunciation similarity calculation unit 14, based on the Chinese corpus 24 of the pronunciation expression, edit distance indicating the similarity and information of the expression with similar pronunciation expression (the character position of the corresponding original expression and the cost of the edit distance) Is calculated. A specific example of calculating the similarity based on the editing distance will be described later. Further, the pronunciation similarity calculation unit 14 outputs the calculated similarity value to the Chinese expression restoration unit 15.

  The Chinese expression restoration unit 15 acquires the corresponding Chinese expression based on the Chinese corpus 22 from the corresponding expression having similar extracted pronunciation expressions and the character position information of the expression. In addition, the Chinese expression restoration unit 15 outputs the restored Chinese expression to the appearance frequency acquisition unit 16.

  The appearance frequency acquisition unit 16 acquires the frequency of the document in which the acquired Chinese expression appears. Here, for example, the “Chinese document of the bilingual translation of the Japanese document including katakana” stored in the Chinese corpus 22 in order to prioritize the parallel translation words having the same similarity calculated by the pronunciation similarity calculation unit 14. The frequency of the appearing document is acquired. In addition, the appearance frequency acquisition unit 16 outputs the acquired appearance frequency information and pronunciation similarity information to the translated word rank selection unit 17.

  The translated word rank selecting unit 17 determines the translated word rank and outputs the translated Chinese word. The translated word rank is short among the plurality of parallel translated words obtained from the pronunciation similarity calculating unit 14. Further, if the editing distance is the same, the document frequency obtained by the appearance frequency acquisition unit 16 is set in descending order. Based on this condition, the mistranslation rank selection unit 17 outputs a Chinese translation.

  As shown in FIG. 1, the parallel word extraction device 10 converts the Chinese (second language) translation corresponding to the expression written in Japanese (first language) from the Chinese document to the similarity of pronunciation. Based on this, it can be extracted with high accuracy.

<Specific example of similarity calculation based on edit distance>
Next, the procedure for calculating the similarity in the pronunciation similarity calculator 14 will be specifically described. In the present invention, the similarity is defined by the value of the edit distance. That is, it is determined that the similarity is higher as the cost of the edit distance is smaller. Here, the edit distance is defined as a cost that minimizes the cost when editing the Chinese pronunciation expression to be the same as the Japanese pronunciation expression. Further, the following three types ((A) to (C)) are used as the cost corresponding to the type of editing used in this method.
(A) Character insertion (cost = CostInsert)
(B) Delete character (Cost = CostDelete)
(C) Character replacement (Cost = CostReplace (char1, char2)
Here, the above char1 and char2 indicate characters before and after replacement.

  Further, in this embodiment, as shown in Non-Patent Document 2, the similarity is calculated by correcting the edit distance (editing cost) using a method other than the method of distinguishing the importance between the consonant and the vowel. Accuracy can be improved. The method will be described below.

<Cost correction corresponding to expression length>
First, editing cost correction corresponding to expression length will be described. In general, editing of one character in a long character string and editing of one character in a short character string have different effects on pronunciation expressed by the entire character string. For example, it can be considered that the effect of editing one character in 10 characters is 1/10 of the whole, and the editing of one character in 5 characters has the influence of 1/5 of the whole. Therefore, the difference in the magnitude of this influence is reflected in the editing cost.

Further, the length of the character string of the Japanese phonetic expression after editing is a character, and the length of the character string of the Chinese phonetic expression before editing is b character. Here, after calculating the edit distance C which is the total minimum cost by each operation of insertion / deletion / replacement, it is converted into a cost in consideration of the length of the character string by the following equation (1).
Edit distance according to expression length = C × a / b (1)

<Cost correction by ratio of insertion and deletion costs learned from corpus>
Also, even if the purpose is to express the same pronunciation, if there is a tendency for a short expression in Japanese pronunciation expression and a long expression in Chinese pronunciation expression, the insertion and deletion costs are reduced. The ratio is adjusted so that deletion is more likely to occur than insertion.

  This ratio is obtained from a bilingual corpus of foreign words during the day. Here, the ratio setting method will be described. In this embodiment, since the editing direction is “edit the Chinese pronunciation expression to be the same as the Japanese pronunciation expression”, first, all the Chinese pronunciation expressions are deleted, and then An operation for generating a Japanese phonetic expression from a state where the phonetic expression is 0 characters is performed.

  Here, both the cost of deleting all Chinese phonetic expressions and the cost of generating a Japanese phonetic expression by inserting (adding) characters from the zero-character state are both 1 It can be thought of as the editing cost of one pronunciation expression as a whole. The ratio of insertion and deletion is adjusted by making the editing costs of these expressions equal.

Therefore, the following expression (2) is used as the ratio of insertion and deletion.
CostDelete × b ′ = CostInsert × a ′ (2)
Here, a ′ represents the average number of Japanese pronunciation expressions in the bilingual corpus and b ′ represents the average number of Chinese pronunciation expressions in the bilingual corpus.

Further, the cost of inserting characters (CostInsert) is set to 1. In this case, the deletion cost is expressed by the following equation (3).
CostDelete = a ′ / b ′ (3)
Note that the value of the above equation (3) is calculated using, for example, a Japanese / Chinese foreign name dictionary. Thereby, the cost in insertion and deletion can be corrected.

<Correction of replacement cost corresponding to the character to be replaced>
Based on the closeness of pronunciation between the replaced character (char1) and the replaced character (char2), the similarity of pronunciation can be calculated more finely by using the replacement cost. Here, a table in which the replacement cost is defined empirically is created and used. Here, a table example will be described with reference to the drawings. FIG. 2 is an example of a table used for the replacement cost. As shown in FIG. 2, the cost corresponding to the combination of the characters before and after replacement and the characters before and after the replacement is set. Note that there is no distinction between characters before and after replacement. By using such a table, the editing cost can be calculated with high accuracy.

  Next, the processing for calculating the editing distance using the above-described editing types (the above-mentioned (A) to (C)) and extracting the upper similar expression having a short editing distance from the long expression will be described with reference to the drawings. explain.

<Edit distance calculation method>
FIG. 3 is an example for explaining an example of calculating the edit distance in the present invention. Note that the example of FIG. 3 is an example of calculating the edit distance for extracting bilingual Chinese corresponding to “Arroyo” in Japanese katakana as shown in [External 1] shown below.

図３に示すように、編集距離の算出は、予め抽出する最大文字数Ｎを指定することで、その最大文字数Ｎ以下の文字数の任意の文字列の編集距離を全て得ることができる。ここで、Ｎは、元になるカタカナ（本実施形態では、日本語カタカナ）の文字数に対応して設定することにより、文字数の不足を防ぐことができる。また、図３における最大文字数Ｎを６とし、”＃”記号はダミーのコストを示している。 [Outside 1]

As shown in FIG. 3, the edit distance is calculated by designating a maximum number N of characters to be extracted in advance, thereby obtaining all edit distances of arbitrary character strings having the number of characters equal to or less than the maximum number N. Here, N is set corresponding to the number of characters of the original katakana (Japanese katakana in the present embodiment), thereby preventing a shortage of the number of characters. Also, the maximum number of characters N in FIG. 3 is 6, and the “#” symbol indicates a dummy cost.

  In FIG. 3, the costs for deletion, insertion, and replacement are all set to 1. However, the present invention is not limited to this, and bilingual words can be extracted with higher accuracy by correcting the costs described above. can do.

  Here, an example of calculating the edit distance based on FIG. 3 will be specifically described with reference to a flowchart. FIG. 4 is a flowchart illustrating an example of calculating the edit distance. In FIG. 4, first, a plurality of tables (“Main Table”) in which Japanese Roman characters are associated with a maximum of N Chinese characters are created (S01). These tables move the Chinese characters in the table one character at a time for the Chinese expressions included in the Chinese corpus 24 of pronunciation expressions shown in FIG. 1 (for example, [External 2] shown below). A table with the maximum number of characters 6 is created.

なお、各テーブルの先頭行の前にダミーの行“＃”を挿入する。また、ダミーの行の先頭のみコストを「０」とする。また、＃行は、右に移動すると、コストが１増加させる。ここで、アロヨはローマ字表記により「ａｒｏｙｏ」となるため、＃行は、“０，１，２，３，４，５”となる。 [Outside 2]

A dummy row “#” is inserted before the first row of each table. Further, the cost is set to “0” only at the top of the dummy row. If the # row moves to the right, the cost is increased by one. Here, since Arroyo is “aroyo” in Roman notation, the # line is “0, 1, 2, 3, 4, 5”.

  In FIG. 3, since the editing cost is 1, this value is obtained. However, if the editing cost is 2, for example, “0, 2, 4, 6, 8, 10” is obtained.

  Therefore, in FIG. 3, “Main Table” stores six Chinese characters in the vertical direction and Roman character representations of each Chinese character, and Katakana Roman characters in the horizontal direction. In FIG. 3, for example, the Roman character expression “a / e” corresponding to the Chinese expression “A” of “Main Table” is “a / e”, but this is expressed as “a” as the pronunciation expression of the Chinese expression “A”. “E” is present.

  Next, a table (“Sub Table”) is created for each created “Main Table” table in which a Roman character expression corresponding to one Chinese character is associated with a Japanese Roman character expression (S02). ). Note that the processing of S02 is performed in order from the top of each row for each “Main Table” row. Further, the cost edited in the previous row on the “Main Table” is copied to a dummy row of the “Sub Table” (S03).

  Next, the cost of the edit distance at each position of the “Sub Table” is calculated by dynamic programming or the like (S04). Further, the cost of the last row of “Sub Table” is copied to “Main Table” (S05). When there are a plurality of “Sub Tables” for one Chinese character, the value with the lowest cost is copied to “Main Table”.

  Here, an example of calculating the cost in the above-mentioned “Sub Table” will be specifically described with reference to FIG. As shown in FIG. 3, the state of insertion, replacement, and deletion is indicated by arrows. First, for the Chinese expression “A”, two “Sub Tables” of “a” and “e” are created. Here, the cost is calculated for each of the katakana romanizations “aroyo” and “a” and “e”.

  First, in the comparison between “#aroyo” and “a”, the characters “#” and “a” at the beginning of “#aroyo” are compared and the characters are different. And becomes 1. Next, “a” and “a” of “#aroyo” are compared, and the characters are the same, so the cost is zero. Next, since “a” and “r” are compared and the characters are different and the characters are inserted, the cost becomes 2 by adding 1 to the total cost 0 so far. Similarly, since “o” and “a” are compared and the characters are different, the cost becomes 2 by adding 1 to the total cost 1 so far. Further, since the characters are different compared to “y” and “a”, the cost becomes 3 by adding 1 to the total cost 2 so far. Further, since “o” and “a” are compared and the characters are different, the cost becomes 4 by adding 1 to the total cost 3 so far. Therefore, the final editing costs for “#aroyo” and “a” are “1, 0, 1, 2, 3, 4”.

  Similarly, comparing the Roman character notation “e” and “#aroyo” of the Chinese character “A”, the cost is “1, 1, 2, 3, 4, 5”. Here, the smaller cost is selected from the two costs. As a result, the cost of the final Chinese expression “A” is “1, 0, 1, 2, 3, 4”.

  In the same manner as described above, the cost is also calculated for the Chinese expression in the next line of “Main Table”. The cost “1, 0, 1, 2, 3, 4” of the row before “Main Table” is input to the dummy row.

  By calculating the cost by such a method, the cost is calculated for all the rows of “Main Table”.

  Next, it is determined whether the cost calculation as described above has been performed for all the rows of “Main Table” (S06). If not performed for all the rows (NO in S06), the next of “Main Table” is determined. (S07), and the process returns to S03. Further, in S06, when the cost is calculated for all the rows of “Main Table” (YES in S06), it is determined whether the processing has been performed for all “Main Table” (S08).

  If not performed for all Main Tables (NO in S08), the process moves to the next “Main Table” (S09), and returns to S02 for processing.

  In S08, when the cost calculation is performed for all the lines of “Main Table” (YES in S08), the start position of “Main Table” becomes the start position of Chinese characters, and the end of “Main Table” The position and value of the column becomes the cost of the edit distance of the Romaji expression corresponding to each end position of Chinese Kanji. Specifically, in "Main Table" starting with the Chinese character "A" shown in FIG. 3, the final cost is "4, 2, 2, 6, 9, 12" in the last column.

  In this way, the cost of the final editing distance of all “Main Tables” is calculated. Here, the cost of the edit distance described above can be corrected to a value of “edit cost according to expression length”, for example.

  Next, this cost is sorted in ascending order to extract the upper start position and end position (S10). Here, the low cost indicates that the number of insertion / deletion / replacement by editing is small, and at the same time, the similarity between languages is high. Next, a Chinese expression is extracted from the start position and end position of the acquired Chinese kanji (S11), and the Chinese expression is output together with the cost (S12).

  In the present embodiment, after the Chinese expression is output together with the cost by the above calculation process, the Chinese expression having the same cost includes the katakana expression input to extract the translated word from the bilingual corpus. Sorted and output in descending order of the number (frequency) of articles containing Chinese expressions contained in the Chinese document (Chinese corpus) that is a parallel translation of the Japanese document.

  As described above, by performing the bilingual word extraction process, the Chinese (second language) translation corresponding to the expression written in Japanese (first language) is converted into the similarity of pronunciation from the Chinese document. Based on this, it is possible to extract with high accuracy.

<Hardware configuration>
Here, as described above, the bilingual word extraction device 10 can also perform bilingual word extraction in the present invention with a dedicated device configuration, but can cause a computer to execute bilingual word extraction processing in each configuration described later. By generating an execution program and installing the program in, for example, a general-purpose personal computer or workstation, it is possible to realize the bilingual word extraction in the present invention.

  Here, an example of a hardware configuration of an executable computer in the present invention will be described with reference to the drawings. FIG. 5 is a diagram illustrating an example of a hardware configuration capable of realizing the bilingual word extraction processing according to the present invention.

  5 includes an input device 51, an output device 52, a drive device 53, an auxiliary storage device 54, a memory device 55, a CPU (Central Processing Unit) 56 for performing various controls, and a network connection device. 57 are connected to each other by a system bus B.

  The input device 51 has a pointing device such as a keyboard and a mouse operated by the user, and inputs various operation signals such as execution of a program from the user. The output device 52 has a display for displaying various windows and data necessary for operating the computer main body for performing the bilingual word extraction processing according to the present invention. Etc. can be displayed.

  Here, in the present invention, the execution program installed in the computer main body is provided by, for example, a recording medium 58 such as a CD-ROM. The recording medium 58 on which the program is recorded can be set in the drive device 53, and the execution program included in the recording medium 58 is installed in the auxiliary storage device 54 from the recording medium 58 via the drive device 33.

  The auxiliary storage device 54 is a storage means such as a hard disk, and can store an execution program according to the present invention, a control program provided in a computer, etc., and perform input / output as necessary.

  The CPU 56 performs processing of the entire computer, such as various operations and input / output of data with each hardware component, based on a control program such as an OS (Operating System) and an execution program read and stored by the memory device 55. The above-described bilingual word extraction processing can be realized by controlling the above. Various information necessary during the execution of the program can be acquired from the auxiliary storage device 54 and can also be stored.

  The network connection device 57 obtains an execution program from another terminal connected to the communication network by connecting to a communication network or the like, or an execution result obtained by executing the program or an execution in the present invention The program itself can be provided to other terminals.

  With the hardware configuration described above, it is possible to provide a highly accurate content at low cost without requiring a special device configuration. Also, by installing the program, it is possible to easily realize the bilingual word extraction process.

<Processing procedure>
Next, the parallel word processing procedure in the present invention will be described with reference to a flowchart. In the present invention, when extracting parallel translation words, a corpus of pronunciation expressions is generated in the first language and the second language. Accordingly, the processing procedure will be described with respect to a procedure for generating a corpus and a procedure for extracting a translation.

  Further, in the processing procedure described later, as described above, the first language is Japanese and the second language is Chinese. That is, an example is shown in which an expression for which a translation is desired is written in Japanese and the translation is in Chinese.

  FIG. 6 is a flowchart illustrating an example of a corpus generation processing procedure. In FIG. 6, first, a katakana for extracting a translated word is input (S61), and a bilingual Chinese document including a representation of the input katakana is acquired (S62). Next, the phonetic expression is converted as described above (S63), and the Chinese corpus of the phonetic expression is output (S64).

  Next, the bilingual word extraction processing procedure will be described with reference to a flowchart. FIG. 7 is a flowchart of an example of a parallel word extraction processing procedure.

  In FIG. 7, first, katakana for extracting translated words is input (S71), and converted into pronunciation expression (S72). Next, pronunciation similarity is calculated by calculating the editing cost as described above (S73). Further, the Chinese expression is restored by referring to the Chinese corpus generated by the corpus generation processing procedure (S74). Similarly, the appearance frequency is acquired based on the Chinese expression calculated by the editing cost with reference to the Chinese corpus (S75).

  Next, the translation word rank is selected based on the similarity and the appearance frequency (S76), and the Chinese translation corresponding to the inputted katakana is output (S77).

  Through the processing procedure as described above, the translation of Chinese (second language) corresponding to the expression written in Japanese (first language) can be made with high accuracy based on the similarity of pronunciation from the Chinese document. Can be extracted. Further, it is possible to provide highly accurate content at low cost without requiring a special device configuration. Further, by installing the program, it is possible to easily realize the bilingual word extraction process.

  As described above, according to the present invention, the translation of the second language corresponding to the expression written in the first language can be obtained with high accuracy based on the similarity and appearance frequency of pronunciation from the document in the second language. Can be extracted.

  Thereby, for example, translations of foreign expressions such as proper nouns can be automatically acquired from a bilingual document such as the latest news article. Moreover, automatic construction of a bilingual dictionary can be realized.

  Further, by applying the present invention to the similarity calculation in the case of translation extraction, translation words can be extracted with high accuracy. Furthermore, it is possible to efficiently acquire a translation word candidate of a higher-order solution, and it is possible to realize highly accurate parallel translation extraction.

  Although the preferred embodiment of the present invention has been described in detail above, the present invention is not limited to the specific embodiment, and various modifications, within the scope of the gist of the present invention described in the claims, It can be changed.

It is a figure which shows one structural example of the bilingual word extraction apparatus in this invention. It is an example of the table used for replacement cost. It is a figure of an example for demonstrating the edit distance in this invention. It is a flowchart which shows an example of the example of calculation of edit distance. It is a figure which shows an example of the hardware constitutions which can implement | achieve the bilingual word extraction process in this invention. It is a flowchart which shows an example of a corpus production | generation process procedure. It is a flowchart of an example of a bilingual word extraction processing procedure.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Parallel word extraction apparatus 11 Chinese document acquisition part 12 1st pronunciation expression conversion part 13 2nd pronunciation expression conversion part 14 pronunciation similarity calculation part 15 Chinese expression restoration part 16 appearance frequency acquisition part 17 translation word rank selection part 21 Japanese-Chinese Bilingual Corpus 22 Chinese Corpus 23 Chinese Character / Romaji Conversion Table 24 Chinese Corpus for Pronunciation Expression 25 Katakana / Roman Character Conversion Table 51 Input Device 52 Output Device 53 Drive Device 54 Auxiliary Storage Device 55 Memory Device 56 CPU
57 Network connection device 58 Recording medium

Claims (5)

In a bilingual word extraction device for extracting a translation of a second language corresponding to an expression of the first language from a document of the second language,
Extraction that extracts a document in a first language including an expression input in the first language and a document in a second language that is a parallel translation of the document from previously translated documents in the first language and the second language Means,
First conversion means for converting the document in the first language into a phonetic expression;
Second conversion means for converting the second language document into a phonetic expression;
Similarity calculation means for calculating the similarity between the pronunciation expression obtained by the first conversion means and the pronunciation expression obtained by the second conversion means;
Appearance frequency acquisition means for acquiring an appearance frequency of the first language document included in the second language document;
A bilingual word extraction apparatus comprising: output means for outputting a translation based on the similarity obtained by the similarity calculation means and the appearance frequency obtained by the appearance frequency acquisition means. The similarity calculation means includes:
When calculating the similarity between the pronunciation expression obtained by the first conversion means and the pronunciation expression obtained by the second conversion means, the editing cost is calculated based on the expression length of the pronunciation expression. The bilingual word extraction device according to claim 1. The similarity calculation means includes:
When calculating the similarity between the pronunciation expression obtained by the first conversion means and the pronunciation expression obtained by the second conversion means, the insertion learned from the bilingual document of the first language and the second language, and The bilingual word extraction device according to claim 1, wherein the similarity is calculated based on a ratio of deletion editing costs. The similarity calculation means includes:
The bilingual word extraction device according to claim 2 or 3, wherein when the editing cost is calculated by replacement, the editing cost is changed and set by a combination of characters before and after replacement. In a bilingual word extraction program for causing a computer to execute a process of extracting a translation of a second language corresponding to an expression of the first language from a document of the second language,
Extraction that extracts a document in a first language including an expression input in the first language and a document in a second language that is a parallel translation of the document from previously translated documents in the first language and the second language Processing,
A first conversion process for converting the document in the first language into a phonetic expression;
A second conversion process for converting the second language document into a phonetic expression;
A similarity calculation process for calculating a similarity between the pronunciation expression obtained by the first conversion process and the pronunciation expression obtained by the second conversion process;
An appearance frequency acquisition process for acquiring an appearance frequency of the document in the first language included in the document in the second language;
A parallel word extraction program for causing a computer to execute an output process for outputting a translation based on the similarity obtained by the similarity calculation process and the appearance frequency obtained by the appearance frequency acquisition process.

Images