JP2005038074A - Device and method for process of creating bilingual dictionary data, process program thereof, and recording medium for recording this program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a process device for automatically creating bilingual dictionary data in a certain language and another language via bilingual data with regard to a third language. <P>SOLUTION: A translation candidate obtaining means 2 obtains an English translation of a keyword (Japanese) in bilingual dictionary data 6 from a Japanese-English dictionary 41 and obtains translation candidates (Chinese) for the obtained English translation from an English-Chinese dictionary 51. A translation evaluating/selecting means 3 uses a word class correspondence evaluating part 31 to evaluate the correspondence between the word class of each translation candidate and the word class of the keyword according to the correspondence between Japanese and Chinese word classes; the translation candidate which is evaluated to be equal to or higher than a predetermined value is extracted as a translation of the keyword in the bilingual dictionary data 6. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

【００５７】
訳語評価・選択手段３は、獲得したすべての訳語候補Ｃについて上記のペナルティ値Ｐｅｎ（Ｊ，Ｃ）を算出する。
【００５８】
品詞対応関係評価部３１により、上記のＦ１（Ｊ，Ｃ）が算出される。例えば、獲得された訳語候補のうち、名詞の訳語候補は適切な訳語であるが、形容詞の訳語候補は適切ではない。形容詞の訳語候補は、英語の訳語「ｓｔｏｕｔ」の形容詞の意味にもとづいて獲得されたものだからである。日本語「アーミン」の品詞（名詞）と対応関係の評価が高い名詞の訳語候補に対する値は低くなる。一方、対応関係の評価が低い形容詞の訳語候補に対する値は高くなる。
【００５９】
さらに、共通語数評価部３２により、上記のＦ３（Ｊ，Ｃ）が算出される。見出し語および訳語候補の英訳の集合中に最も多く現れる英訳「ｅｒｍｉｎｅ」と対応する訳語候補の値が高くなる。
【００６０】
さらに、訳語長評価部３３により、Ｆ２（Ｃ）が算出される。獲得された訳語候補が説明文であるような訳語候補の値が低くなる。
【００６１】
そして、訳語評価・選択手段３は、Ｆ１、Ｆ２、Ｆ３のそれぞれの値からペナルティ値Ｐｅｎ（Ｊ，Ｃ）を算出する。そして、ペナルティ値Ｐｅｎ（Ｊ，Ｃ）が一定の値より小さい訳語候補を良い訳語であると評価し、この訳語候補を対訳辞書データ６における見出し語「アーミン」の訳語として抽出する。
【００６２】
次に、日本語の単語「重大だ」を見出し語とする場合の処理の具体例にもとづき本発明の効果を説明する。
【００６３】
ここでは、日英辞書４１としてＥＤＲ辞書データ（日本電子化辞書研究所）を用い、英中辞書５１としてＬＤＣ英中・中英単語対応表データ（ｈｔｔｐ：／／ｗｗｗ．ｌｄｃ．ｕｐｅｎｎ．ｅｄｕ／Ｐｒｏｊｅｃｔｓ／Ｃｈｉｎｅｓｅ／）を用いて、日本語の各見出し語について中国語の訳語候補を獲得した。なお、ＬＤＣ英中・中英単語対応表データには、品詞情報が付与されていないため、中国語単語分割ツールにより、各単語に品詞情報を付与して用いた。
【００６４】
処理の結果、中国語の訳語候補を２０個以上獲得し、かつ、獲得した訳語候補中に正しい訳語が含まれていることが検証できた見出し語は１７２語あった。そのうち、最も訳語候補数が多い見出し語は、形容詞「重大だ」であった。
【００６５】
図７（Ａ）に示すように、本発明の対訳辞書データ作成処理装置１は、訳語候補獲得手段２において、見出し語「重大だ」に対して、英訳「ｍａｊｏｒ、ｉｍｐｏｒｔａｎｔなど」を介して１４５個の中国語の訳語候補を獲得した。
【００６６】
しかし、１つの見出し語に対して１４５個の訳語候補が存在するという状態は、適切な訳語も獲得されているが、不適切もしくは誤った訳語も多数獲得されているという状態を意味し、これらの訳語候補にもとづいて作成した対訳辞書データ６を機械翻訳処理に利用することはできない。
【００６７】
本発明の対訳辞書データ作成処理装置１は、訳語評価・選択手段３において、獲得された１４５個の訳語候補全てについて、その訳語候補が適切な訳語であるかどうかの評価を行い、さらに、一定の評価を得た訳語候補を対訳辞書データ６における訳語として抽出した。
【００６８】
ここで、品詞対応関係情報による評価と、共通する中継言語の訳語による評価との効果を比較するために、２つの異なる処理を行ってみた。正しい訳語候補「重大」を指標とした。
【００６９】
（１）処理例Ｉ
まず、処理例Ｉとして、訳語評価・選択手段３は、訳語候補の品詞対応関係（Ｆ１）および訳語候補の文字列長（Ｆ２）を評価手法として用いて、以下のようなペナルティ値Ｐｅｎ（Ｊ，Ｃ）を算出し、算出したペナルティ値の順位で訳語候補を並べた。
【００７０】
Ｐｅｎ（Ｊ，Ｃ）＝Ｆ１（Ｊ，Ｃ）＋Ｆ２（Ｃ）
図７（Ｂ）に、所定の評価を得た訳語候補を含むまでの順位と、含まれる訳語候補の一部とを示す。所定の評価を得た訳語候補が含まれる候補数は、１位において２７個、７位において１６個、８位において３３個であった。この処理例Ｉでは、正しい訳語候補「重大」を獲得するために２７個の訳語候補が獲得されることになる。そして、正しい訳語「重大」は１位に含まれ、以下のような動詞、名詞の訳語は、低い順位となった。
【００７１】
【外２】

【００７２】
（２）処理例ＩＩ
また、別の処理例ＩＩとして、訳語候補の品詞対応関係（Ｆ１）および訳語候補の文字列長（Ｆ２）および共通する英訳数（Ｆ３）を評価手法として用いて、以下のようなペナルティ値Ｐｅｎ（Ｊ，Ｃ）を算出し、算出したペナルティ値の順位で訳語候補を並べた。
【００７３】
Ｐｅｎ（Ｊ，Ｃ）＝Ｆ１（Ｊ，Ｃ）＋Ｆ２（Ｃ）−Ｆ３（Ｊ，Ｃ）
図７（Ｃ）に、所定の評価を得た訳語候補を含むまでの順位と、含まれる訳語候補の一部とを示す。
【００７４】
処理例ＩＩでは、所定の評価を得た訳語候補が含まれる候補数は、１位において２個、２位において３個であった。また、１位および２位の訳語候補は、すべて正しいものであり、正しい訳語「重大」は１位に含まれた。
【００７５】
【発明の効果】
以上の説明のように、本発明によれば、第二の中継言語を介して、基言語および目的言語について対訳辞書データを自動作成する場合に、自動的に獲得された多数の訳語候補を、所定の評価式を用いて絞り込むことにより、不適切な訳語候補や誤った訳語候補を排除して、適切な訳語候補のみを対訳辞書データの訳語として抽出する。
【００７６】
これにより、あまり国際的でない基言語および目的言語についても精度のよい対訳辞書データを自動的に作成する処理装置、処理方法、または処理プログラムなどを提供でき、対訳辞書データ作成処理における種々のコスト軽減を図ることができる。
【図面の簡単な説明】
【図１】本発明にかかる対訳辞書データ作成処理装置の構成例を示す図である。
【図２】第一辞書データ（日英辞書）の例を示す図である。
【図３】第二辞書データ（英中辞書）の例を示す図である。
【図４】品詞対応関係情報の例を示す図である。
【図５】共通する英訳の数にもとづく評価処理を説明するための図である。
【図６】本発明の処理の概略を説明するための図である。
【図７】獲得した訳語候補の例を示す図である。
【符号の説明】
１ 対訳辞書データ作成処理装置
２ 訳語候補獲得手段
３ 訳語評価・選択手段
３１ 品詞対応関係評価部
３２ 共通語数評価部
３３ 訳語長評価部
４ 第１辞書記憶部
４１ 第一辞書データ（日英辞書）
５ 第２辞書記憶部
５１ 第二辞書データ（英中辞書）
６ 対訳辞書データ（日中辞書）[0001]
BACKGROUND OF THE INVENTION
The present invention relates to bilingual dictionary data creation processing technology for creating bilingual dictionary data that describes translations from one natural language to another, and particularly describes bilingual data between a first language and a second language. From the first language to the third language using the digitized first dictionary data and the digitized second dictionary data describing the parallel translation data of the second language and the third language The present invention relates to a bilingual dictionary data creation processing apparatus for creating bilingual dictionary data, a processing method thereof, a bilingual dictionary data creation processing program, and a recording medium on which the bilingual dictionary data creation processing program is recorded.
[0002]
[Prior art]
In the process of acquiring bilingual data for a word in a certain language, the language of the word / phrase that the user wants to obtain the bilingual is called a base language, and the language of the acquired word is called a target language.
[0003]
I want to look up a translation of a phrase in one language into another language, but depending on the language, there is a situation where there is no bilingual dictionary from the base language to the target language. In such a situation, a more international language having a bilingual dictionary for the base language and the target language is used as a relay language. First, using the bilingual dictionary (first dictionary data) of the base language and the relay language, the translation language of the relay language is acquired from the base language. Thereafter, using the bilingual dictionary (second dictionary data) of the relay language and the target language, the target language translation is acquired from the already acquired relay language translation.
[0004]
The first dictionary data and second dictionary data that have been digitized are the processes for automatically creating bilingual dictionary data from the base language to the target language by relaying the bilingual data with a more international language. There has been proposed a method of automatically using a computer and using a computer (see Non-Patent Document 1).
[0005]
[Non-Patent Document 1]
By Tanaka (Ishii) Kumiko, Umemura Koji, Iwasaki Hideya,
"Creating a bilingual dictionary via a third language," IPSJ Journal,
In June 1998, Vol. 39, no. 6, pp. 1915-1924
[0006]
[Problems to be solved by the invention]
It is well known that a phrase in a natural language has multiple meanings, and that the conceptual scope of the meaning of a phrase varies from language to language.
[0007]
When creating bilingual dictionary data of the base language and the target language via the relay language, in the process of acquiring the translation of the relay language, the translation for the multiple meanings of the words in the base language is acquired, and the acquired translation is Based on the target language translation.
[0008]
For this reason, there may be a shift in the conceptual range of significance of these words / phrases for the words / phrases in the base language, the translated words in the relay language, and the translated words in the target language. Due to the shift in the conceptual range of this meaning, there is a problem that the finally acquired translation of the target language includes inappropriate words and the accuracy of the bilingual dictionary data is lowered.
[0009]
In the conventional processing method described in Non-Patent Document 1 described above, inappropriate translations are eliminated by reversely retrieving the base language words from the acquired translations of the target language. However, the process of performing reverse lookup on all acquired translations of the target language and determining the appropriateness of the parallel translation requires a large amount of calculation, and thus the load is large.
[0010]
The present invention has been made in view of the above points, and uses bilingual dictionary data of a base language, a target language, and a relay language, and uses bilingual dictionary data of the base language and the target language using a computer. A bilingual dictionary data creation processing device, a processing method thereof, a bilingual dictionary data creation processing program, and a processing program thereof that can create accurate bilingual dictionary data without increasing the processing amount when automatically creating It is an object to provide a recorded recording medium.
[0011]
[Means for Solving the Problems]
In order to achieve the above-mentioned object, the present invention describes the first electronic dictionary data describing the parallel translation data of the first language and the second language, and the parallel translation data of the second language and the third language. A bilingual dictionary data creation processing apparatus that creates bilingual dictionary data from a first language to a third language using electronic second dictionary data, and has the following configuration.
[0012]
The present invention stores means for storing first dictionary data describing parallel translation data between a first language and a second language, and means for storing second dictionary data describing parallel translation data between a second language and a third language. A translation of the second language for the headword that is a word in the first language is obtained from the first dictionary data, and a translation of the third language is obtained for the translation of the second language from the second dictionary data. Correspondence between the part-of-speech of the candidate word and the part-of-speech of the entry word based on the part-of-speech correspondence information indicating the ease of correspondence of the part-of-speech between the first language and the third language A translation selection unit that evaluates a relationship and extracts a translation word candidate as a translation of the entry word in the bilingual dictionary data when the evaluation of the correspondence is equal to or greater than a predetermined value;
[0013]
Alternatively, the present invention provides a means for storing the first dictionary data, a means for storing the second dictionary data, and a translation of a second language for a headword that is a word of the first language from the first dictionary data. A translation candidate acquisition means for acquiring a translation candidate of the third language from the second dictionary data and obtaining a translation candidate of the third language from the second dictionary data, and a second translation of the entry word and a second candidate of the translation candidate A translation evaluation / selection unit that obtains the number of translations common in the bilingual translations and extracts translation candidates with the translations that are common in a predetermined number or more as translations of the entry words in the bilingual dictionary data;
[0014]
Alternatively, the present invention provides a means for storing the first dictionary data, a means for storing the second dictionary data, and a translation of a second language for a headword that is a word of the first language from the first dictionary data. A translation candidate acquisition means for acquiring a translation candidate in a third language for a translation in the second language from the second dictionary data, and determining the number of character strings of the translation candidate, Translation word evaluation / selection means is provided for extracting less than a predetermined number of translation candidates as translations of the entry word in the bilingual dictionary data.
[0015]
Furthermore, the present invention is a processing method executed in the bilingual dictionary data creation processing apparatus, a processing program for causing a computer to execute processing in the processing method, or a recording medium on which the processing program is recorded.
[0016]
The present invention provides, for example, bilingual dictionary data between Japanese and Chinese, where the first language as the base language is Japanese, the third language as the target language is Chinese, and the second language as the relay language is English. When creating, it works as follows.
[0017]
In the present invention, Japanese-English dictionary data as first dictionary data and English-Chinese dictionary data as second dictionary data are stored. The English translation is acquired from the Japanese-English dictionary data for the Japanese word that is the headword of the bilingual dictionary data. Further, a Chinese translation is acquired for the English translation from English-Chinese dictionary data, and a translation candidate is obtained.
[0018]
Then, the acquired translation word candidate is evaluated as follows, and the translation word candidate that has obtained a certain evaluation is extracted as a translation of the headword in the bilingual dictionary data.
[0019]
Based on the part-of-speech correspondence information indicating the ease of correspondence between parts of speech between Japanese and Chinese, the correspondence between the part-of-speech of the candidate word and the part-of-speech of the headword is evaluated, and the evaluation of the correspondence is predetermined. If it is greater than or equal to the value, the translation candidate is extracted as the translation of the entry word in the bilingual dictionary data.
[0020]
In general, parts of speech between different languages have a certain restrictive correspondence. In the present invention, by using the restriction of the correspondence relationship between the parts of speech, for example, by giving a low evaluation to a part of speech that has a close correspondence, that is, a part of speech that is considered to have no correspondence, It is something to exclude.
[0021]
Alternatively, the present invention obtains the number of translation words common in the English translation of the Japanese translation and the Chinese translation candidate, and extracts a translation candidate with the number of translations common in a predetermined number or more as the Japanese translation in the bilingual dictionary data.
[0022]
If the English translation of the entry word and the English translation of the acquired translation candidate are in common, it is considered that the semantic connection between the entry word and the translation candidate is strong. In the present invention, a common second language translation (English translation) that is considered to indicate the strength of the semantic relationship between the headword and the translation word candidate is used, and the English translation common to the English translation of the headword has a predetermined number or more. The candidate words are highly evaluated, and candidate words having good evaluation are extracted.
[0023]
Alternatively, the present invention obtains the number of character strings of the acquired Chinese translation candidates and extracts the translation candidates having the number of character strings less than a predetermined number as the translation of the entry word in the bilingual dictionary data.
[0024]
The dictionary data may include a description of a headword in addition to translated words such as words and phrases. Generally, a word or phrase is considered preferable as a translated word rather than an explanatory sentence. Therefore, based on the acquired candidate word string, the evaluation of the candidate word that seems to be an explanatory sentence is lowered, and the word or phrase is Extract appropriate word candidates as translated words.
[0025]
In the present invention, the suitability of acquired translation word candidates is evaluated using the above evaluation criteria. As a result, inappropriate or incorrect translations can be eliminated from the acquired translation candidates, and appropriate translation candidates can be extracted as translations of bilingual dictionary data, and accurate bilingual dictionary data can be automatically created. Is possible.
[0026]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, as one embodiment of the present invention, when creating bilingual dictionary data from Japanese as a base language to Chinese as a target language, a more international as a relay language between the base language and the target language. A bilingual dictionary data creation process when English as a language is used will be described.
[0027]
FIG. 1 shows a configuration example of a bilingual dictionary data creation processing apparatus according to the present invention. The bilingual dictionary data creation processing device 1 includes a translated word candidate acquisition unit 2 and a translated word evaluation / selection unit 3. The bilingual dictionary data creation processing device 1 can access the first dictionary storage unit 4 and the second dictionary storage unit 5 to acquire data.
[0028]
The first dictionary storage unit 4 is means for storing first dictionary data (Japanese-English dictionary) 41 describing parallel translation data from the base language to the relay language of the parallel translation dictionary data 6 to be created.
[0029]
FIG. 2 shows an example of the Japanese-English dictionary 41. In the Japanese-English dictionary 41, at least the part of speech of the head word, the translation in English (relay language), and the part of speech of the translation are described for each Japanese (base language) head word.
[0030]
In the Japanese-English dictionary 41 shown in FIG. 2, information such as record numbers, parts of speech, semantic codes, semantic concept information, and translated words (English) are associated with headwords. For example, the following information is stored for the Japanese headword “Armin”.
[0031]
“Part of speech: noun,
Semantic concept information: ermine, okojo, an animal called ermine,
Translation: stout (ad) / ermine (n)
The second dictionary storage unit 5 stores second dictionary data (English-Chinese dictionary) 51 describing parallel translation data from the relay language to the target language adopted in the first dictionary data 41 of the first dictionary storage unit 4. Means.
[0032]
FIG. 3 shows an example of the English-Chinese dictionary 51. The English-Chinese dictionary 51 describes at least the part-of-speech of the headword, the translation in Chinese (target language), and the part-of-speech of the translation for each headword in English (relay language).
[0033]
For example, the following information is stored for the English headword “ermine”.
“Part of speech: noun,
Translation: 貂 / 貂 的 White fur (noun) ”
The translation candidate acquisition means 2 takes out the translation word for each entry word in the Japanese-English dictionary 41, searches for the entry word in the English-Chinese dictionary 51 using the extracted translation word (translation in the second language) as a key, and further searches for the Chinese word. This is a means for acquiring translations and making them translation candidates.
[0034]
The translation evaluation / selection means 3 evaluates the translation candidate acquired by the translation candidate acquisition means 2 based on a predetermined evaluation rule, and selects a translation candidate that has obtained a certain evaluation as a translation of the bilingual dictionary data 6 It is.
[0035]
The translation evaluation / selection means 3 includes any one, two, or all of the processing units of the part of speech correspondence evaluation unit 31, the common word number evaluation unit 32, or the translation word length evaluation unit 33.
[0036]
The part-of-speech correspondence evaluation unit 31 refers to the part-of-speech correspondence information describing the closeness of the correspondence between parts-of-speech between two natural languages, and the part-of-speech of a Japanese headword and the part-of-speech of a candidate for a Chinese translation. This is a processing unit for evaluating candidate translations based on the correspondence relationship.
[0037]
Here, it is the part-of-speech correspondence information in Japanese and Chinese, and the part of speech of Japanese (base language) is classified into 37 parts of speech, and the part of speech of Chinese (target language) is classified into 39 parts of speech. In such a case, the combination of all parts of speech between the two languages, which defines the degree of ease of correspondence, is used.
[0038]
FIG. 4 shows an example of part of speech correspondence information. The degree of ease of correspondence of part of speech is classified into four categories: “corresponding, semi-corresponding, non-corresponding, undecided”. For example, in the translation from Japanese to Chinese, the level of Japanese common nouns corresponding to Chinese nouns is very high. A thing with a high correspondence between parts of speech in this way is defined as “correspondence”. In addition, the degree to which Japanese common nouns correspond to Chinese particles is even lower. Such correspondence is referred to as “non-correspondence”. Japanese common nouns may also correspond to “noun word elements”, which are word elements that generate nouns in Chinese. Such correspondence is referred to as “semi-correspondence”. Assume that relationships other than these three correspondence relationships are “undecided”.
[0039]
A predetermined condition can also be set for the correspondence. For example, “Correspondence” is only applied to the correspondence relationship that satisfies the condition of “adjective verb” in Japanese and the part of speech of the last word of the Chinese candidate word is a particle and “the last character is“ target ”” May be set.
[0040]
Then, the part-of-speech correspondence evaluation unit 31 evaluates Chinese translation word candidates for Japanese headwords based on the degree of the four correspondences defined in such part-of-speech correspondence information. When the translation candidate is composed of a plurality of words (morphemes), the part of speech of the last word (morpheme) is set as an evaluation target.
[0041]
The common word number evaluation unit 32 is a processing unit that evaluates the translation word candidates acquired based on the number of translation words common to the English translation of the entry word and the English translation of the acquired translation word candidate.
[0042]
This is because if the English translations of Japanese headwords and the English translations of the translation candidates are more common, the translation candidates are closer to the Japanese headwords and are considered to be more appropriate translations. .
[0043]
FIG. 5 shows an example of commonality between English translations of Japanese headwords and Chinese translation candidates. Here, there is “ermine, stout” as an English translation for the Japanese headword “Armin”, and “ermine, marten, stable”, “ermine”, “strong” "powerful", "firm, solid", and "ermine" is common to two candidate words.
[0044]
In this case, it is considered that the headwords with common English translation “ermine” and the translation word candidates have a close semantic relationship. Since the translation candidate having such a relationship is highly likely to be an appropriate translation, the translation candidate having the English translation “ermine” is highly evaluated.
[0045]
The translated word length evaluation unit 33 is a processing unit that evaluates the acquired translated word candidate by a predetermined calculation formula F3 (C) based on the length of the acquired character string of the translated word candidate.
[0046]
Like English-Chinese dictionary 51 that acquires translation word candidates, the dictionary data may include a description text of the head word in addition to the translation for the head word, so the translation word candidate based on the length of the character string of the translation word candidate So that appropriate translations such as words and phrases can be extracted.
[0047]
The translation evaluation / selection means 3 evaluates each of the translation word candidates acquired by the translation word candidate acquisition means 2 using, for example, the following formula for calculating a penalty value (Pen (J, C)).
[0048]
Pen (J, C) = F1 (J, C) + F2 (C) -F3 (J, C)
[J = Japanese headword, C = Chinese translation candidate]
Among the above Pen (J, C), an evaluation value based on the part-of-speech correspondence between the entry word J and the translation word candidate C is acquired by F1 (J, C).
[0049]
For example, for four correspondence relationships [corresponding, semi-corresponding, non-corresponding, undecided], predetermined values having the following relationships are determined, and the values are acquired.
[0050]
“Correspondence”: smallest value “semi-correspondence”: second smallest value “non-correspondence”: largest value “undecided”: second largest value Therefore, the value of F1 becomes smaller as the part of speech corresponds. Therefore, the penalty value Pen (J, C) is reduced, and the evaluation of the translation candidate is improved.
[0051]
F2 (C) is determined so as to be directly proportional to the length of the translation word candidate C. Therefore, the shorter the character string of the translated word candidate, the smaller the value of F2, the smaller the penalty value Pen (J, C), and the better the evaluation of the translated word candidate.
[0052]
Further, F3 (J, C) is determined so that the English translations of the entry word J and the translation word candidate C are in direct proportion to the number of common words. Therefore, as the number of common words increases, the value of F3 increases, the penalty value Pen (J, C) decreases, and the evaluation of the candidate translation is improved.
[0053]
The outline of the processing of the present invention will be described with reference to FIG.
[0054]
First, the translation candidate acquisition unit 2 refers to the Japanese-English dictionary 41 in the first dictionary storage unit 4 and acquires all English translations of the word (eg, “(noun) armin”) that is the headword J.
[0055]
Further, the translation candidate acquisition means 2 refers to the English / Chinese dictionary 51 of the second dictionary storage unit 5, acquires all translations (Chinese) for each acquired English translation, and sets it as a translation candidate C. For example, the following “noun” translation and “adjective” translation are acquired.
[0056]
[Outside 1]

[0057]
The translation evaluation / selection means 3 calculates the penalty value Pen (J, C) for all acquired translation word candidates C.
[0058]
The part-of-speech correspondence evaluation unit 31 calculates F1 (J, C). For example, among the acquired translation candidates, the noun translation candidate is an appropriate translation, but the adjective translation candidate is not appropriate. This is because the adjective translation candidates are acquired based on the meaning of the adjective of the English translation “stout”. The value for a candidate noun that has a high evaluation of the correspondence between the part of speech (noun) of Japanese “Armin” is low. On the other hand, the value for an adjective candidate with a low evaluation of the correspondence relationship is high.
[0059]
Further, the common word number evaluation unit 32 calculates the above F3 (J, C). The value of the translation candidate corresponding to the English translation “ermine” that appears most frequently in the set of English translations of the headwords and translation candidates becomes high.
[0060]
Further, the translated word length evaluation unit 33 calculates F2 (C). The value of the candidate word for which the acquired candidate word is an explanatory sentence becomes low.
[0061]
Then, the translated word evaluation / selection means 3 calculates a penalty value Pen (J, C) from the respective values of F1, F2, and F3. Then, a translation candidate whose penalty value Pen (J, C) is smaller than a certain value is evaluated as a good translation, and this translation candidate is extracted as a translation of the entry word “Armin” in the parallel translation dictionary data 6.
[0062]
Next, the effect of the present invention will be described based on a specific example of processing when the Japanese word “serious” is used as an entry word.
[0063]
Here, EDR dictionary data (Japan Electronic Dictionary Laboratory) is used as the Japanese-English dictionary 41, and LDC English-Chinese / English dictionary correspondence data (http://www.ldc.upenn.edu/ Using Projects / Chinese /), Chinese translation candidates were obtained for each Japanese headword. Note that the part-of-speech information is not given to the LDC English / Chinese English-language correspondence table data, so the part-of-speech information is given to each word using a Chinese word segmentation tool.
[0064]
As a result of the processing, there were 172 headwords that acquired 20 or more Chinese translation candidates and verified that the acquired translation candidates contained the correct translation. Among them, the headword with the largest number of candidate translations was the adjective “serious”.
[0065]
As shown in FIG. 7A, the bilingual dictionary data creation processing device 1 of the present invention uses the English translation “major, important, etc.” 145 for the headword “serious” in the translated word candidate acquisition means 2. Acquired Chinese candidate words.
[0066]
However, the state that there are 145 candidate translations for one headword means that a proper translation has been acquired but many inappropriate or incorrect translations have been acquired. The bilingual dictionary data 6 created based on the translated word candidates cannot be used for the machine translation process.
[0067]
The bilingual dictionary data creation processing device 1 of the present invention evaluates whether or not the translation candidate is an appropriate translation for all 145 translation candidates obtained by the translation evaluation / selection means 3, and further The translated word candidates obtained with the above evaluation were extracted as translated words in the bilingual dictionary data 6.
[0068]
Here, in order to compare the effect of the evaluation based on the part-of-speech correspondence information and the evaluation based on the translation of the common relay language, two different processes were performed. The correct translation candidate “serious” was used as an index.
[0069]
(1) Processing example I
First, as processing example I, the translation evaluation / selection means 3 uses the part-of-speech correspondence (F1) of the translation word candidate and the character string length (F2) of the translation word candidate as evaluation methods, and the following penalty value Pen (J , C), and translated word candidates are arranged in the order of the calculated penalty value.
[0070]
Pen (J, C) = F1 (J, C) + F2 (C)
FIG. 7B shows the ranks until the translation candidate candidates that have obtained the predetermined evaluation are included, and some of the translation word candidates that are included. The number of candidates including translated word candidates having a predetermined evaluation was 27 in the 1st place, 16 in the 7th place, and 33 in the 8th place. In this processing example I, 27 candidate words are acquired in order to acquire a correct candidate word “serious”. The correct translation “serious” was included in the first place, and the following verb and noun translations were ranked low.
[0071]
[Outside 2]

[0072]
(2) Processing example II
Also, as another processing example II, the following penalty value Pen is used by using the part-of-speech correspondence (F1) of the translation candidate, the character string length (F2) of the translation candidate and the number of common English translations (F3) as evaluation methods. (J, C) was calculated, and translation word candidates were arranged in the order of the calculated penalty value.
[0073]
Pen (J, C) = F1 (J, C) + F2 (C) -F3 (J, C)
FIG. 7C shows the ranks until the translation candidate candidates obtained with a predetermined evaluation are included, and some of the translation candidate candidates included.
[0074]
In the processing example II, the number of candidates including the translated word candidates obtained with the predetermined evaluation was 2 in the first place and 3 in the second place. In addition, the first and second candidate translations were all correct, and the correct translation “serious” was included in the first place.
[0075]
【The invention's effect】
As described above, according to the present invention, when automatically creating bilingual dictionary data for the base language and the target language via the second relay language, a large number of translation word candidates that are automatically acquired, By narrowing down using a predetermined evaluation formula, inappropriate translation candidates and incorrect translation candidates are excluded, and only appropriate translation candidates are extracted as translation words in the bilingual dictionary data.
[0076]
As a result, it is possible to provide a processing device, a processing method, a processing program, or the like that automatically creates accurate bilingual dictionary data even in a base language and target language that are not very international, thereby reducing various costs in bilingual dictionary data creation processing. Can be achieved.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration example of a bilingual dictionary data creation processing apparatus according to the present invention.
FIG. 2 is a diagram showing an example of first dictionary data (Japanese-English dictionary).
FIG. 3 is a diagram showing an example of second dictionary data (English-Chinese dictionary).
FIG. 4 is a diagram illustrating an example of part of speech correspondence information.
FIG. 5 is a diagram for explaining an evaluation process based on the number of common English translations.
FIG. 6 is a diagram for explaining the outline of processing of the present invention.
FIG. 7 is a diagram illustrating an example of acquired translation word candidates.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Bilingual dictionary data creation processing apparatus 2 Translation word candidate acquisition means 3 Translation word evaluation / selection means 31 Part-of-speech correspondence evaluation part 32 Common word number evaluation part 33 Translation word length evaluation part 4 1st dictionary memory | storage part 41 1st dictionary data (Japanese-English dictionary)
5 Second dictionary storage 51 Second dictionary data (English-Chinese dictionary)
6 Bilingual dictionary data (Japanese-Chinese dictionary)

Claims (12)

Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language In the bilingual dictionary data creation processing device for creating bilingual dictionary data from the first language to the third language,
Means for storing first dictionary data describing parallel translation data of the first language and the second language;
Means for storing second dictionary data describing parallel translation data of the second language and the third language;
A translation of a second language is obtained from the first dictionary data for a headword that is a word of the first language, and a translation of a third language is obtained from the second dictionary data for a translation of the second language. Means for acquiring candidate translations,
Based on the part of speech correspondence information indicating the ease of correspondence of the part of speech between the first language and the third language, the correspondence between the part of speech of the candidate word and the part of speech of the headword is evaluated, and the correspondence A bilingual dictionary data creation processing device comprising: a translation selection unit that extracts a translation candidate as a translation of the entry word in the bilingual dictionary data when the evaluation is equal to or higher than a predetermined value. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language In the bilingual dictionary data creation processing device for creating bilingual dictionary data from the first language to the third language,
Means for storing first dictionary data describing parallel translation data of the first language and the second language;
Means for storing second dictionary data describing parallel translation data of the second language and the third language;
A translation of a second language is obtained from the first dictionary data for a headword that is a word of the first language, and a translation of a third language is obtained from the second dictionary data for a translation of the second language. Means for acquiring candidate translations,
The number of translation words common in the second language translation of the entry word and the second language translation of the translation candidate is obtained, and the translation word candidate having the number of translation words common in a predetermined number or more is extracted as the translation of the entry word in the bilingual dictionary data A bilingual dictionary data creation processing device comprising: a translation word evaluation / selection means for performing translation. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language In the bilingual dictionary data creation processing device for creating bilingual dictionary data from the first language to the third language,
Means for storing first dictionary data describing parallel translation data of the first language and the second language;
Means for storing second dictionary data describing parallel translation data of the second language and the third language;
A translation of a second language is obtained from the first dictionary data for a headword that is a word of the first language, and a translation of a third language is obtained from the second dictionary data for a translation of the second language. Means for acquiring candidate translations,
Bilingual dictionary data, comprising: a translation word evaluation / selection unit that obtains the number of character strings of the translation word candidates and extracts translation word candidates whose number of character strings is less than a predetermined number as translations of the entry words in the translation dictionary data Creation processing device. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language In the bilingual dictionary data creation processing method for creating bilingual dictionary data from the first language to the third language,
The translation of the second language is acquired from the first dictionary data describing the parallel translation data of the first language and the second language for the headword that is the word of the first language,
From the second dictionary data describing the bilingual data of the second language and the third language, the translation of the third language is obtained as the translation candidate for the translation of the second language,
Based on the part of speech correspondence information indicating the ease of correspondence of the part of speech between the first language and the third language, the correspondence between the part of speech of the candidate word and the part of speech of the headword is evaluated, and the correspondence A bilingual dictionary data creation processing method, wherein, when the evaluation is equal to or higher than a predetermined value, the translated word candidate is extracted as a translated word of the headword in bilingual dictionary data. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language In the bilingual dictionary data creation processing method for creating bilingual dictionary data from the first language to the third language,
The translation of the second language is acquired from the first dictionary data describing the parallel translation data of the first language and the second language for the headword that is the word of the first language,
From the second dictionary data describing the bilingual data of the second language and the third language, the translation of the third language is obtained as the translation candidate for the translation of the second language,
Find the number of translations common in the second language translation of the headword and the second language translation of the translation candidate,
A bilingual dictionary data creation processing method, wherein a number of translation candidates with a common number of translations equal to or greater than a predetermined number are extracted as translations of the headword in bilingual dictionary data. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language In the bilingual dictionary data creation processing method for creating bilingual dictionary data from the first language to the third language,
The translation of the second language is acquired from the first dictionary data describing the parallel translation data of the first language and the second language for the headword that is the word of the first language,
From the second dictionary data describing the bilingual data of the second language and the third language, the translation of the third language is obtained as the translation candidate for the translation of the second language,
Obtain the number of character strings of the translation candidates,
6. A bilingual dictionary data creation processing method, wherein translated word candidates having a number of character strings smaller than a predetermined number are extracted as translated words of the headword in bilingual dictionary data. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language , A program for causing a computer to execute bilingual dictionary data creation processing for creating bilingual dictionary data from a first language to a third language,
A process of obtaining a translation of the second language for a headword that is a word of the first language from the first dictionary data describing the parallel translation data of the first language and the second language;
Processing to acquire a translation of the third language from the second dictionary data describing the parallel translation data of the second language and the third language, and to obtain a translation candidate of the third language;
A process of evaluating the correspondence between the part of speech of the candidate word and the part of speech of the headword based on the part of speech correspondence information indicating the ease of correspondence of the part of speech between the first language and the third language;
A bilingual dictionary data creation processing program for causing a computer to execute a process of extracting a translation word candidate as a translation of the entry word in bilingual dictionary data when the evaluation of the correspondence relationship is a predetermined value or more. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language , A program for causing a computer to execute bilingual dictionary data creation processing for creating bilingual dictionary data from a first language to a third language,
A process of obtaining a translation of the second language for a headword that is a word of the first language from the first dictionary data describing the parallel translation data of the first language and the second language;
Processing to acquire a translation of the third language from the second dictionary data describing the parallel translation data of the second language and the third language, and to obtain a translation candidate of the third language;
A process of calculating the number of translations common in the second language translation of the headword and the second language translation of the translation candidate;
A process of extracting translation candidates of the number of translations common in a predetermined number or more as translations of the entry word in the parallel translation dictionary data,
A bilingual dictionary data creation processing program to be executed by a computer. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language , A program for causing a computer to execute bilingual dictionary data creation processing for creating bilingual dictionary data from a first language to a third language,
A process of obtaining a translation of the second language for a headword that is a word of the first language from the first dictionary data describing the parallel translation data of the first language and the second language;
Processing to acquire a translation of the third language from the second dictionary data describing the parallel translation data of the second language and the third language, and to obtain a translation candidate of the third language;
Processing for obtaining the number of character strings of the translation candidates;
A bilingual dictionary data creation processing program for causing a computer to execute a process of extracting candidate words whose number of character strings is smaller than a predetermined number as translated words of the headword in bilingual dictionary data. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language , A recording medium recording a program for causing a computer to execute parallel dictionary data creation processing for creating parallel dictionary data from a first language to a third language,
A process of obtaining a translation of the second language for a headword that is a word of the first language from the first dictionary data describing the parallel translation data of the first language and the second language;
Processing to acquire a translation of the third language from the second dictionary data describing the parallel translation data of the second language and the third language, and to obtain a translation candidate of the third language;
A process of evaluating the correspondence between the part of speech of the candidate word and the part of speech of the headword based on the part of speech correspondence information indicating the ease of correspondence of the part of speech between the first language and the third language;
When the evaluation of the correspondence relationship is equal to or higher than a predetermined value, a bilingual dictionary data creation processing program for causing a computer to execute a process of extracting the translation word candidate as a translation of the headword in bilingual dictionary data is recorded recoding media. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language , A recording medium recording a program for causing a computer to execute parallel dictionary data creation processing for creating parallel dictionary data from a first language to a third language,
A process of obtaining a translation of the second language for a headword that is a word of the first language from the first dictionary data describing the parallel translation data of the first language and the second language;
Processing to acquire a translation of the third language from the second dictionary data describing the parallel translation data of the second language and the third language, and to obtain a translation candidate of the third language;
A process of calculating the number of translations common in the second language translation of the headword and the second language translation of the translation candidate;
A recording medium on which a bilingual dictionary data creation processing program for causing a computer to execute a process of extracting as many translated word candidates as a translation of the entry word in bilingual dictionary data at a predetermined number or more is used. Using the computerized first dictionary data describing the bilingual data of the first language and the second language, and the computerized second dictionary data describing the bilingual data of the second language and the third language , A recording medium recording a program for causing a computer to execute parallel dictionary data creation processing for creating parallel dictionary data from a first language to a third language,
A process of obtaining a translation of the second language for a headword that is a word of the first language from the first dictionary data describing the parallel translation data of the first language and the second language;
Processing to acquire a translation of the third language from the second dictionary data describing the parallel translation data of the second language and the third language, and to obtain a translation candidate of the third language;
A process for obtaining the number of character strings of the translation candidates;
A recording medium recording a bilingual dictionary data creation processing program for causing a computer to execute a process of extracting translated word candidates having a number of character strings smaller than a predetermined number as translated words of the headword in bilingual dictionary data.

Images