JP2006053867A - Bilingual dictionary creation method and apparatus, and computer program - Google Patents

Abstract

A bilingual dictionary can be automatically created for a combination of arbitrary languages.
The method includes a first dictionary for a first language speaker having a direction from a first language to a third language, and a second language for a second language speaker. Preparing a second dictionary having directionality from the first language to the third language in an electronically readable form, each entry relating to a content word in the first dictionary, and a content word in the second dictionary A step 54 of extracting an entry pair whose similarity value determined as a predetermined function with respect to the set of translations of the third language included in each entry is equal to or greater than a predetermined threshold, among the entry pairs composed of each entry; And step 56 for storing the content word pairs corresponding to the entry pairs extracted in the extracting step in an electronically readable form.
[Selection] Figure 1

Description

  The present invention relates to an automatic construction technique for an electronic dictionary relating to a natural language, and more particularly to a dictionary creation method and apparatus capable of automatically constructing a bilingual dictionary suitable for machine translation regardless of the combination of languages.

  In machine translation, the bilingual dictionary is the most basic and important resource. However, it takes a long time and a great cost to construct a bilingual dictionary. In the future, the demand for automatic translation between various languages is expected to increase, but the number of language combinations increases exponentially with the number of languages, so how to build a bilingual dictionary Is an important issue.

  There are various methods for constructing an electronic bilingual dictionary. Among the conventional methods, the most interesting one is the Papillon Project (Non-Patent Document 1). This document focuses on building multilingual vocabulary data for creating a large-scale, detailed, and compliant dictionary. The primary resource for creating a multilingual dictionary is a monolingual dictionary. These monolingual dictionaries are associated with links between languages and are used as a database for creating a multilingual dictionary. When a multilingual dictionary is created by such a method, a large number of collaborators and experts are required.

As research related to this project, an attempt to create a Japanese-French dictionary from an English-French dictionary and a Japanese-English dictionary is disclosed in Non-Patent Document 2.
Christian Boat et al., “Papillon Project: Collaborative construction of an open source dictionary and multilingual vocabulary database for deriving vocabulary catalogs”, 2nd Workshop, NLPXML-2002, pp. 93-96, 2002 (Christian Boitet et al., The Papillon project: cooperatively building a multilingual lexical data-base to derive open source dictionaries & lexicons ", the 2nd Workshop NLPXML-2002, pp. 93-96) Kumiko Tanaka and one other, “Creation of a bilingual dictionary through a third language”, CORING-94, pp. 297-303, 1994 (Kumiko Tanaka et al., "Construction of a Bilingual Dictionary Intermediated by a Third Language", COLING-94, pp. 297-303.)

  In order to realize machine translation between multiple languages as described above, bilingual dictionaries for many language combinations are required. Moreover, it is necessary to minimize the time and cost for that purpose. In addition, an expert who is familiar with both languages is usually required to create a bilingual dictionary. However, when there are various combinations of languages, it becomes difficult to find such experts. Therefore, there is a technology for automatically creating a bilingual dictionary so that even a worker who is not familiar with the language to be translated can efficiently create a bilingual dictionary for any combination of languages. is needed.

  Although the prior art described above shows the necessity of a technique for creating a bilingual dictionary and an attempt to do so, it is possible to automatically and efficiently create a bilingual dictionary for any combination of languages. No strategy is shown. In addition, it is desirable that as many entries as possible can be extracted from available resources because of the nature of machine translation, but no measures for that purpose are shown in the prior art.

  SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a bilingual dictionary creation method and apparatus capable of automatically creating a bilingual dictionary for any combination of languages.

  Another object of the present invention is to provide a bilingual dictionary creation method and apparatus capable of automatically creating a bilingual dictionary having as many entries as possible for any combination of languages.

  A bilingual dictionary creating method according to a first aspect of the present invention is a bilingual dictionary creating method for automatically creating a bilingual dictionary between a first language and a second language with a third language as an intermediary. A first dictionary with directionality from the first language to the third language for speakers of the first language and a direction from second language to the third language for speakers of the second language A second dictionary having the characteristics in an electronically readable form, and an entry pair comprising each entry relating to the content word of the first dictionary and each entry relating to the content word of the second dictionary Of these, a step of extracting an entry pair whose similarity value defined as a predetermined function for the set of translations of the third language included in each entry is equal to or greater than a predetermined threshold, and an entry extracted in the extracting step The content word pair corresponding to the pair can be read electronically And a first step of storing by the formula.

  By selecting the first and second dictionaries having the above directionality and extracting the content word pairs corresponding to the entry pairs having a similarity value equal to or greater than a threshold value from these dictionaries, Experiments have confirmed that bilingual dictionaries for the first language and the second language can be created with higher accuracy and more headlines than when using dictionaries with different directions. The dictionary creation procedure is automatic, and the user only needs to specify the language combination of the bilingual dictionary. Therefore, a highly accurate bilingual dictionary can be created in a short time for many combinations of languages. Further, according to this method, an expert who is familiar with both the first language and the second language is unnecessary, and a bilingual dictionary can be easily created even for a very large number of language combinations.

  Preferably, the preparing step includes preparing a plurality of electronically readable bilingual dictionaries each having a predetermined direction for a predetermined language speaker, and a first language and a second language. A step of receiving a specification related to a language, a dictionary having directionality from the first language specified in the step of receiving a specification to a different language from a plurality of bilingual dictionaries, and a second language specified in the step of receiving the specification Selecting a dictionary pair consisting of a dictionary having a direction to the other language.

  More preferably, the preparing step comprises: preparing a plurality of electronically readable bilingual dictionaries each having a predetermined direction for a predetermined language speaker; Receiving a designation relating to two languages and a third language; a dictionary having directionality from the first language to the third language designated in the designation receiving step from a plurality of two-language dictionaries; and receiving the designation Selecting a dictionary pair consisting of a dictionary having directionality from the second language to the third language specified in (1).

  More preferably, a plurality of dictionary pairs may be selected in the selecting step, and the preparing step further includes an entry included in each dictionary pair among the plurality of dictionary pairs selected in the selecting step. Selecting the one with the largest sum of numbers.

Preferably, the extracting step includes a step of calculating a similarity S ₁ (x, y) between each entry x in the first dictionary and each entry y in the second dictionary by the following equation.

ただしＺ（ｘ）、Ｚ（ｙ）はそれぞれ、エントリｘ、ｙ内に含まれる第３言語の訳語の集合を表し、記号｜・｜は集合の要素数を表す。抽出するステップはさらに、類似度Ｓ₁（ｘ，ｙ）が予め定めるしきい値以上であるエントリ対（ｘ，ｙ）に対応する内容語対を抽出するステップを含んでもよい。

However, Z (x) and Z (y) represent a set of third language translations included in the entries x and y, respectively, and the symbols | · | represent the number of elements of the set. The step of extracting may further include a step of extracting a content word pair corresponding to the entry pair (x, y) whose similarity S ₁ (x, y) is greater than or equal to a predetermined threshold value.

  More preferably, the predetermined threshold value is a function of the number of common translation words of the third language included in the entry pair (x, y).

  More preferably, the function of the number of common translations is a monotonically decreasing function with respect to the number of common translations.

  The function of the number of common translations may be set to 0 when the number of common translations is 4 or more.

  Preferably, the bilingual dictionary creation method includes a third dictionary having a direction from the third language to the first language for a speaker in the first language and a third dictionary for a speaker in the second language. Preparing a fourth dictionary having directionality from the language to the second language in an electronically readable form, a third dictionary entry corresponding to the third language heading, and a fourth dictionary The third language example sentence that satisfies the predetermined condition and its translation are extracted from both the dictionary entries, and the parallel sentence pair created in the step of creating the parallel sentence pair and the step of creating the parallel sentence pair Of these, the second step of extracting the functional word pairs from the chunks aligned with each other in the alignment step and the step of taking the alignment, and storing them in an electronically readable form, and electronically readable The first and second steps stored in various formats Further comprising the step of merging up in stored content words paired with a function word pairs.

  When the computer program according to the second aspect of the present invention is executed by a computer, the computer program controls the computer to perform all the steps of any of the above-described bilingual dictionary creation methods.

  A bilingual dictionary creation device according to a third aspect of the present invention is a bilingual dictionary creation device for automatically creating a bilingual dictionary between a first language and a second language, using the third language as an intermediary. A first dictionary with directionality from the first language to the third language for speakers of the first language and a direction from second language to the third language for speakers of the second language A dictionary preparing means for preparing the second dictionary having the characteristics in an electronically readable form, each entry relating to the content word of the first dictionary, and each entry relating to the content word of the second dictionary Entry pair extraction means for extracting an entry pair whose similarity value determined as a predetermined function for a set of translated words in the third language included in each entry is equal to or greater than a predetermined threshold value Corresponds to the entry pair extracted by the entry pair extraction means That the content words pairs and a content word pair storage means for storing in electronically readable form.

[Dictionary properties]
In the prior art, the nature of the dictionary is not considered much. In practice, however, the nature of the dictionary is very important in the automatic creation of a bilingual dictionary. Typical characteristics of the dictionary include the native language assumed by the dictionary and the directionality of the dictionary.

  In this specification, “dictionary directionality” indicates a directionality meaning which language word is drawn from which language entry. In this specification, this directionality is represented by “⇒”. For example, the direction of the English-Japanese dictionary is English → Japanese, and the direction of the Japanese-English dictionary is “Japanese → English”. The Korean-English dictionary is “Korean → English”, and the English-Korean dictionary is “English → Korean”. In the following description, abbreviations are used for languages in order to simplify the description. For example, “J” for Japanese, “E” for English, “K” for Korean, etc. Therefore, the Japanese-English dictionary is expressed as “J → E”, and the English-Japanese dictionary is expressed as “E → J”.

The native language assumed by the dictionary refers to the user's native language assumed by the dictionary. In general, both English-Japanese and Japanese-English dictionaries created in Japan are assumed to be users who speak Japanese as their mother tongue and English as a foreign language (hereinafter referred to as “Japanese speakers”). On the other hand, for example, if there is an English-Japanese dictionary created in the United Kingdom, an assumed user is called a speaker whose native language is English ("English speaker". The same applies to other languages hereinafter. ). In the present specification, an abbreviation representing a native language assumed by the dictionary is given in small letters before the directionality of the dictionary. Therefore, the English-Japanese dictionary for Japanese speakers _"J E⇒J", Japanese-English dictionary for Japanese speakers _"J J⇒E", Korean-English dictionary for the Korean speaker is _{"K K} ⇒E ”, the English-Korean dictionary for Korean speakers is expressed as“ _K E⇒K ”. In this specification, a dictionary from language X to language Y, which is generally created for speakers of language Z, is represented as “ _{Z X} => Y”.

The directionality of the dictionary is not considered in the prior art. However, directionality is actually a very important concept in creating a bilingual dictionary. For example, Japanese-English dictionary _(J J⇒E) that Japanese speaker to be used is used when a Japanese speaker or speak in English mainly or write a sentence in English. In such a situation, of course, a Japanese speaker knows the meaning of the Japanese word he is trying to translate into English. Therefore, it is not necessary to write a detailed explanation about the word in the dictionary. As an exception, in the case of a word in Japanese that does not have a concept corresponding to English, an explanatory description for expressing the concept in Japanese in English may be provided. In addition, when trying to look up English corresponding to Japanese with a Japanese-English dictionary, information on the part of speech (POS) of the translation is also secondary and is not so important.

On the other hand, English-Japanese dictionary _(J E⇒J) for Japanese speakers, as seen from the point of view of the Japanese speaker is used to know, such as English the meaning and usage of the word. For this reason, not only the translated words but also grammatical information such as descriptive information such as wording and usage, example sentences, and parts of speech are often attached to each entry.

  For example, as shown in FIG. 4, for the words “tango” and “tango” that correspond to each other in Japanese and English, an entry 160 of “tango” in the Japanese-English dictionary and an entry 162 of “tango” in the English-Japanese dictionary. The contents are very different. The entry 160 lists only the translated words very simply, whereas the entry 162 is detailed. Not only the translated words but also explanations about the words, grammatical information such as plurals and parts of speech, and information such as example sentences are described. When creating a bilingual dictionary between languages, the case where only translated words are listed as in the entry 160 is less complicated in machine processing, and a highly accurate bilingual dictionary can be created efficiently. On the other hand, an entry 162 containing many explanatory texts related to translated words includes unnecessary information for creating a bilingual dictionary, so that necessary machine processing is complicated and errors are increased.

  Therefore, in creating a bilingual dictionary for machine translation, it is necessary to consider the directionality of such a dictionary and the language of the target speaker.

The inventor of the present invention conducted an experiment to create a bilingual dictionary of Japanese and Korean (hereinafter referred to as “Japanese-Korean bilingual dictionary”) using English as an intermediate language in consideration of the directionality. As a result, it is creating a Japanese-Korean translation dictionary in combination dictionary that _J J⇒E and _K K⇒E, it was possible to obtain much better results in comparison with the case of using the other combinations. Since the directionality of the dictionary is considered to have universality, the creation method considering the directionality of the dictionary in this way is considered to be effective not only in a Japanese-Korean bilingual dictionary but also in any combination of languages. Hereinafter, an embodiment of such a bilingual dictionary automatic creation apparatus will be described.

[Constitution]
FIG. 1 is a block diagram of an automatic dictionary creation system 20 according to an embodiment of the present invention. Referring to FIG. 1, a dictionary automatic creation system 20 includes a dictionary storage device 38 that stores a large number of dictionaries between two languages of various combinations, and a dictionary attribute storage device 36 that stores attribute information related to the directionality of these dictionaries. The dictionary language combination information (information 30 designating the first language X and information 32 designating the second language Y) given by the user is received, and the dictionary is referred to the dictionary attribute storage device 36. An optimal combination of dictionaries is selected from the dictionaries stored in the storage device 38 according to the first language X and the second language Y, and the electronic parallel translation dictionary 40 between the first language X and the second language Y is automatically selected. And an automatic dictionary creation device 34 for creating it.

  FIG. 2 shows the contents of the dictionary storage device 38. As shown in FIG. 2, the dictionary storage device 38 includes, for each of various languages, a dictionary group 80, 82, 84,... With another language for a user whose native language is the language. For example, the dictionary group 80 is a dictionary group for Japanese speakers, and includes a Japanese-English dictionary 90, an English-Japanese dictionary 92,. Similarly, the dictionary group 82 is a dictionary group for Korean speakers, and includes a Korean-English dictionary 100, an English-Korean dictionary 102,. The dictionary group 84 is a dictionary group for English speakers, and includes a Japanese-English dictionary 110, an English-Japanese dictionary 112,.

  Both the dictionary 90 and the dictionary 110 are Japanese-English dictionaries, with the difference that the dictionary 90 is for Japanese speakers and the dictionary 110 is for English speakers. Similarly, the dictionary 92 and the dictionary 112 are both English-Japanese dictionaries, but the dictionary 92 is for Japanese speakers and the dictionary 112 is for English speakers.

  FIG. 3 shows a configuration of the dictionary attribute storage device 36 shown in FIG. Referring to FIG. 3, the dictionary attribute storage device 36 stores a plurality of dictionary attribute information 140, 142, 144, 146,. The attribute information follows the format of the attribute information 130 shown in FIG. The attribute information 130 includes a dictionary file name, a path name to the dictionary file (URL if the dictionary file is on the network; hereinafter simply referred to as “path name”), and the language in which the dictionary is the native language. Native language type information indicating whether the dictionary is for a speaker, the type of the first language constituting the entry, the type of the second language constituting the translated word, and the number of entries included in the dictionary are included.

For example, in the case of a dictionary _J J⇒E, native types are Japanese (J), the first language type is Japanese (J), the second language type is the English (E). In the case of a dictionary _J E⇒J, native types are Japanese (J), the first language type is English (E), the second language type is a Japanese (J).

  Referring again to FIG. 1, the dictionary automatic creation device 34 includes an input device 50 for receiving input of information 30 specifying the first language X and information 32 specifying the second language Y from the user, A content word dictionary selection unit 52 for selecting a dictionary combination for extracting an entry related to a content word from the bilingual dictionary according to the language combination received by the device 50 with reference to the dictionary attribute storage device 36. A content word translation extraction processing unit 54 for extracting a translation (content word pair) related to the content word with reference to the two dictionaries selected by the content word dictionary selection unit 52 in the dictionary storage device 38; A content word pair storage unit 56 for electronically storing the content word pairs extracted by the word translation extraction processing unit 54.

  The dictionary automatic creation device 34 further selects a dictionary combination for extracting an entry relating to a functional word from the bilingual dictionary in accordance with the language combination received by the input device 50 in the same manner as the content word dictionary selection unit 52. A function word dictionary selection unit 60 for selecting with reference to the storage device 36 and a function word with reference to two dictionaries selected by the function word dictionary selection unit 60 among the dictionaries in the dictionary storage device 38 A functional word translation extraction processing unit 62 for extracting bilingual translations (functional word pairs), and a functional word pair storage unit 64 for electronically storing the functional word pairs extracted by the functional word translation extraction processing unit 62; including.

  The dictionary automatic creation device 34 further merges the content word pair stored in the content word pair storage unit 56 and the function word pair stored in the function word pair storage unit 64 to change from the first language to the second language. A merge processing unit 70 for creating the bilingual dictionary 40 is included. Here, “merge” refers to a language that uses a set of content word pairs and a set of function word pairs in one set, and in some order, for example, Ai-Oh order in Japanese, English, etc. If so, it means that content word pairs and function word pairs are arranged in alphabetical order. The language that becomes the reference of the order may be either the first language or the second language, or two may be created by merging them based on each.

Assuming that the first language X and the second language Y are designated, the content word dictionary selection unit 52 refers to the dictionary attribute storage device 2 so that _{X X} → Z and _{Y Y} → Z. Has the ability to select two dictionaries. The language Z is an arbitrary language, and any language can be used as long as the corresponding dictionary exists in the dictionary storage device 38. That is, the content word dictionary selection unit 52 is for a speaker whose first language X is a native language, for a speaker whose first language X is a third language Z, and for a speaker whose second language Y is a native language. And a dictionary from the second language Y to the third language Z is selected.

On the other hand the function word dictionary selecting section 60, first language X, assuming that the second language Y is designated, by referring to the dictionary attribute storage device 36, such that _X Z⇒X, _Y Z⇒Y Select two dictionaries. That is, the functional language dictionary selection unit 60 is for a speaker whose first language X is a native language, a dictionary from the third language Z to the first language X, and a speaker whose native language is the second language Y. And a dictionary from the third language Z to the second language Y is selected.

FIG. 5 is a more detailed block diagram of the content word translation extraction processing unit 54 shown in FIG. Referring to FIG. 5, content word translation extraction processing unit 54 extracts words for extracting all combinations of two dictionary 170 entries and dictionary 172 entries selected by content word dictionary selecting unit 52. Of the word pairs extracted by the pair extraction unit 180 and the word pair extraction unit 180, both entries based on the contents of the translation in the entry extracted from the dictionary 170 and the translation in the entry extracted from the dictionary 172 And a similarity calculation unit 182 for calculating the similarity S ₁ between them by the following equation.

ただしｘ、ｙはそれぞれ言語Ｘ、Ｙのエントリ内に含まれる、言語Ｚの訳語の集合を表し、記号｜・｜は集合の要素数を表す。

Here, x and y represent a set of translated words of language Z included in the entries of languages X and Y, respectively, and symbols | · | represent the number of elements of the set.

  The content word translation extraction processing unit 54 further determines a threshold of similarity determined in advance by experiments so that a content word pair can be appropriately selected or rejected according to the number of common translation words included between two entries. The threshold value calculated by the similarity calculation unit 182 for each of the word pairs extracted by the threshold value table storage unit 186 and the word pair extraction unit 180 is stored in the threshold value table storage unit. A content word pair selection unit 184 for determining whether to select or reject the word pair according to whether or not the threshold value stored in 186 is equal to or greater than, and outputting the selected word pair to the content word pair storage unit 56 Including.

FIG. 6 shows an example of a word pair extraction result by the word pair extraction unit 180 when a bilingual dictionary of Korean and Japanese is created with English as an intermediate language. In this case, _K K⇒E and _J J⇒E are respectively selected as the dictionary 170 and 172. With reference to FIG. 6, by referring to the _K K⇒E, for both the word 190 and words 192 Korean, for example it is assumed that two English translation (check and Cheque) are extracted . On the other hand, the same two English translations (check and check) are extracted for the Japanese word “check”, and three translations (check, prevent, prevent from) are extracted for “prevent”. It is assumed that two translations (examine, check) are extracted for “check”. Similarly, it is assumed that the translated word check is extracted for “verification” and four translated words (leave, deposit, check, entrust) are extracted for “deposit”.

  In this case, the similarity is calculated according to the expression (1) between the Korean words 190 and 192 and the Japanese words, and is classified according to the number of common words in the translated words in each entry. Is shown in FIG.

  For example, two translated words (check, check) are common between Korean 190 and Japanese “cheque”. In addition, since both have two translations, similarity = 2 × 2 / (2 + 2) = 1.000. The relationship between Korean 192 and Japanese “cheque” is similar.

  On the other hand, there is one common translation between Korean 190 and Japanese “match”. Further, since there are two translations of Korean 190 and one translation of “verify”, similarity = 2 × 1 / (2 + 1) = 0.667. Similarities in other rows in FIG. 7 are calculated in the same manner.

  Referring to FIG. 7, if the English translation for Korean K1 and the English translation for Japanese J1 completely match, it is very likely that words K1 and J1 correspond to each other. However, as shown in FIG. 7, as the number of translated words to be shared decreases, the possibility that the words K1 and J1 correspond to each other decreases. In the example shown in FIG. 7, for example, it is preferable not to use one belonging to the group (3).

  Therefore, the similarity threshold for determining whether or not to adopt as a content word pair is changed according to the number of common translation words. Specifically, the threshold value was determined so that the content word pair could be adopted with an accuracy of about 90 percent by various experiments. FIG. 8 shows the number of common translation words when a bilingual dictionary of Japanese and Korean is created using English as an intermediary, and the threshold value τ of similarity that varies depending on the number of common translation words. As shown in FIG. 8, the threshold τ is a monotonically decreasing function of the number of common translation words. The threshold τ is set to 0 when the number of common translation words is 4 or more. In other words, content word pairs having a common translation word count of 4 or more are used unconditionally.

  Using the threshold values stored in the threshold table storage unit 186 shown in FIG. 8, an experiment was conducted between Korean and Japanese to extract content word pairs using English as an intermediary. The number of combinations of Korean and Japanese entries is 157,618. For this combination, 28,479 entries were used in the used Korean dictionary (50,826 entries), and 17,687 entries were used in the Japanese dictionary (28,310 entries). Of these, the number of content word pairs extracted as entries in the Korean-Japanese bilingual dictionary with 90% accuracy was 25,703.

  On the other hand, it is more preferable that not only content word pairs but also function word pairs can be extracted. The functional word translation extraction processing unit 62 shown in FIG. 1 is for that purpose. The detailed structure is shown in FIG.

Referring to FIG. 9, a functional translation extraction processing unit 62, the two dictionaries 210 selected by the function word dictionary selecting section 60 as described above _(X Z⇒X) and 212 _(Y Z⇒Y) An example sentence extracting unit 220 for examining all combinations of the same word entries in the language Z and extracting all the example sentences or phrases (hereinafter simply referred to as “example sentences”) of the language Z and their translations; Among the example sentences of the language Z in the entry of the same language Z extracted by the example sentence extracting unit 220, the one extracted from the dictionary 210 and the one extracted from the dictionary 212 are identical or similar. And a matching sentence extracting unit 222 for extracting the sentence together with the translated sentences described in the two dictionaries for the example sentence. Here, out of a pair of example sentences of language Z described in two dictionaries, one that completely matches each other or a similar one having seven or more matching words is extracted and its translation is extracted. .

  Note that example sentences in language Z that completely match each other may be extracted unconditionally. Further, the threshold of the number of matching words when selecting as similar sentences is preferably 7 or more, and may be 8 or more.

  The function word translation extraction processing unit 62 further refers to the contents of the content word pair storage unit 56 and aligns the two translated sentences extracted by the matching sentence extraction unit 222 in units of chunks centering on the content words. By removing the content words stored in the content word pair storage unit 56 from the chunks of the two languages X and Y associated with each other by the alignment by the alignment processing unit 224 and the alignment processing unit 224, A function word pair selection unit 226 for generating a pair of only function words remaining in the chunk. The function word pair selection unit 226 adopts only those chunks whose corresponding language Z chunks completely match each other, and do not adopt those that are different from each other. The function word pair selection unit 226 writes the adopted function word pair in the function word pair storage unit 64.

  Processing performed by the function word pair selection unit 226 will be described with reference to FIG. In the following, in order to facilitate understanding, a case will be described in which functional word pairs are extracted between Korean and Japanese using English as an intermediary. FIG. 10 shows example sentence pairs 240 and 242.

  The example sentence pair 240 indicates a case where the English sentence of the example sentence 250 extracted from the English-Japanese dictionary and the example sentence 252 extracted from the English-Korean dictionary completely match. The symbol “=” described at the beginning of the sentence of the example sentence pair 240 indicates that the English sentences are completely identical to each other, and “10” indicates the number of words included in the example sentence.

  In the case of the example sentence pair 240, the Japanese chunk “I am” 270 and the Korean chunk 280 are associated with each other by alignment. Further, the Japanese chunk “Jiyuga” 272 and the Korean chunk 282 are associated with each other. If the chunks are associated with each other in this way, the function words associated with each other remain if the content word is subtracted from each chunk. Such function words may be adopted as function word pairs.

  In the case of the example sentence pair 242, the English sentence of the example sentence 260 from the English-Japanese dictionary does not completely match the English sentence of the example sentence 262 from the English-Korean dictionary. The symbol at the beginning of the sentence indicates that both English sentences are similar but not exactly the same. The subsequent number “8” indicates the number of matching words in the English example sentence. In this example, the chunk 290 extracted from the English-Japanese dictionary is different from the chunk 292 extracted from the English-Korean dictionary. There is also a difference in prepositions before “class”.

  In this case, when there is a difference in the corresponding English part, the function word pair selection unit 226 does not extract the function word pair from the chunk corresponding to the part. A function word pair is simply extracted from a chunk corresponding to a part where both are completely matched.

[Operation]
With reference to FIGS. 1 to 10, the dictionary automatic creation system 20 whose configuration has been described above operates as follows. First, referring to FIG. 1, the user inputs information 30 specifying the first language X and information 32 specifying the second language Y to the input device 50. The input device 50 provides the given information to the content word dictionary selecting unit 52 and the function word dictionary selecting unit 60.

The content word dictionary selection unit 52 refers to the dictionary attribute storage device 36 and determines a language Z in which a dictionary of _XX → Z and _YY → Z exists. In determining the language Z, there is a method of selecting the one having the largest total number of entries in the two dictionaries based on the number of entries included in the dictionary.

On the other hand, the function word dictionary selection unit 60 refers to the dictionary attribute storage device 36 and determines a language W in which a dictionary of _X W => _X and _Y W => _Y exists. Language W may be the same as language Z described above, but may be different. In order to simplify the description, language W = language Z is assumed below. Even in this way, generality is not lost.

  The content word dictionary selection unit 52 gives information about the selected dictionary, that is, the file name and path name, to the content word translation extraction processing unit 54. Upon receiving the file name and path name, the content word translation extraction processing unit 54 accesses the two dictionaries in the dictionary storage device 38 (or on the network), creates a content word pair, and stores the content word pair storage unit. 56.

More specifically, the content word translation extraction processing unit 54 operates as follows. Referring to FIG. 5, word pair extraction section 180 extracts all word pairs from designated dictionaries 170 and 172 and provides them to similarity calculation section 182. The similarity calculation unit 182, all word pairs (x, y) with respect to, a set of translation language Z in the entry x extracted from the dictionary _X X⇒Z Z (x), from the dictionary _Y Y⇒Z The similarity S ₁ (x, y) is calculated according to the above-described equation (1) with the translated word Z (y) of the language Z in the extracted entry y, and is given to the content word pair selection unit 184.

The content word pair selection unit 184 refers to the threshold value table storage unit 186 and obtains a threshold value corresponding to the number of elements in the language Z word set Z (x) ∩Z (y) common to both. The similarity S ₁ given from the similarity calculation unit 182 is compared with this threshold value. If the similarity S ₁ (x, y) is greater than the threshold value, the content word pair selection unit 184 adopts the word pair x, y as the content word pair and writes it into the content word pair storage unit 56. If the similarity S ₁ (x, y) is less than or equal to the threshold value, the content word pair selection unit 184 rejects the word pair.

  By repeating the above processing for all the word pairs extracted by the word pair extraction unit 180, a set of content word pairs of language X and language Y is accumulated in the content word pair storage unit 56.

  When the accumulation of the content word pairs in the content word pair storage unit 56 is completed, the function word pair extraction is performed by the function word translation extraction processing unit 62 shown in FIG. More specifically, the function word translation extraction processing unit 62 operates as follows.

Referring to FIG. 9, example sentence extraction unit 220 converts word dictionary Z to word z for two dictionaries 210 ( _X Z => _X ) and 212 ( _Y Z => _Y ) selected by function word dictionary selection unit 60. The process of extracting example sentences from the two corresponding entries is performed for all the words z, and the extracted example sentences are given to the matching sentence extraction unit 222. The matching sentence extraction unit 222 determines whether or not there is an example sentence in the language Z that is exactly the same as or similar to each other among the example sentences obtained from the two entries corresponding to the word z, and uses the example sentence as an alignment processing part. 224.

  The alignment processing unit 224 stores the example sentence pairs of the language X and the language Y, which are translations of the same example sentence in the language Z among the example sentences given from the matching sentence extraction unit 222, in the content word pair storage unit 56. Align using content word pairs. The alignment processing unit 224 gives the aligned example sentence pair to the function word pair selection unit 226. The function word pair selection unit 226 removes the content words of each language by referring to the content word pair storage unit 56 from the chunks in which the corresponding language Z portions completely match among the aligned example sentences, The remaining function words are paired and given to the function word pair storage unit 64.

  By repeating such processing for all the example sentences extracted by the example sentence extraction unit 220, the function word pairs of the language X and the language Y are accumulated in the function word pair storage unit 64.

  Referring to FIG. 1, merge processing unit 70 combines a large number of content word pairs stored in content word pair storage unit 56 and a large number of function word pairs stored in function word pair storage unit 64. The electronic bilingual dictionary 40 between the language X and the language Y is created by merging into one dictionary format file.

  As described above, according to the dictionary automatic creation system 20 according to the present embodiment, by specifying two languages X and Y first, a specific word suitable for creating a content word pair and a function word pair is created. Each directional dictionary is selected, and a content word pair and a function word pair are created. Furthermore, the electronic bilingual dictionary 40 is created by merging them. The dictionary is selected so that as many entry pairs as possible can be obtained corresponding to each processing, so that the number of entries in the electronic bilingual dictionary 40 finally obtained is large, which is useful when used for machine translation. .

  In the above embodiment, an electronic bilingual dictionary is created using both content word pairs and function word pairs. However, the present invention is not limited to such an embodiment. For example, a dictionary of only content word pairs may be created. For example, in the case of languages belonging to an agglutinative language such as Japanese and Korean, function word pairs can be extracted with a relatively high accuracy by the method of the above-described embodiment, but the accuracy may be lowered depending on the combination of languages. It can happen. In such a case, a selection may be made to stop the extraction of function word pairs.

  Moreover, in the said embodiment, the language Z to mediate is determined by the dictionary automatic creation system 20. As a criterion for this decision, in addition to selecting the one with the larger number of entries (the sum of the entries) as described above, prioritizing by language, considering the type of language, era or year when the dictionary was created May be considered. In addition, the dictionary automatic creation system 20 does not determine the language Z to mediate, but the user may explicitly specify it. Also in this case, when a plurality of dictionary combinations are possible, it is preferable to select a dictionary combination that maximizes the sum of the number of entries.

  Furthermore, the mediation language when creating the content word pair and the mediation language when creating the function word pair may be made to coincide with each other.

  Furthermore, in the above-described embodiment, all available dictionaries are stored in the dictionary storage device 38 in advance, but the present invention is not limited to such an embodiment. For example, there may be a method in which a dictionary is located in a remote place and can be accessed via a network, or a person stored in some storage medium is set in a reading device by a human according to the designation from the dictionary automatic creation system 20. Is possible.

[Realization by computer]
The dictionary automatic creation system 20 according to this embodiment can be realized by computer hardware, a program executed by the computer hardware, and data stored in the computer hardware. FIG. 11 shows the external appearance of the computer system 330, and FIG. 12 shows the internal configuration of the computer system 330.

  Referring to FIG. 11, a computer system 330 includes a computer 340 having an FD (flexible disk) drive 352 and a CD-ROM (compact disk read only memory) drive 350, a keyboard 346, a mouse 348, and a monitor 342. including.

  Referring to FIG. 12, in addition to FD drive 352 and CD-ROM drive 350, computer 340 includes CPU (Central Processing Unit) 356 and bus 366 connected to CPU 356, FD drive 352, and CD-ROM drive 350. And a read only memory (ROM) 358 for storing a boot-up program and the like, and a random access memory (RAM) 360 connected to the bus 366 for storing a program command, a system program, work data, and the like. Computer system 330 further includes a printer 344.

  Although not shown here, the computer 340 may further include a network adapter board that provides a connection to a local area network (LAN).

  A computer program for causing the computer system 330 to operate as the dictionary automatic creation system 20 according to this embodiment is stored in the CD-ROM 362 or FD 364 inserted in the CD-ROM drive 350 or FD drive 352, and further Transferred to the hard disk 354. Alternatively, the program may be transmitted to the computer 340 through a network (not shown) and stored in the hard disk 354. The program is loaded into the RAM 360 when executed. The program may be loaded directly into the RAM 360 from the CD-ROM 362, from the FD 364, or via a network.

  This program includes a plurality of instructions for causing the computer 340 to operate as the dictionary automatic creation system 20 according to this embodiment. Some of the basic functions necessary to perform this method are provided by operating system (OS) or third party programs running on the computer 340 or various toolkit modules installed on the computer 340. Therefore, this program does not necessarily include all functions necessary for realizing the operation as the dictionary automatic creation system 20 of this embodiment. This program includes only instructions that realize the operation as the dictionary automatic creation system 20 by calling an appropriate function or “tool” in a controlled manner so as to obtain a desired result. Just go out. The operation of computer system 330 is well known and will not be repeated here.

  The embodiment disclosed herein is merely an example, and the present invention is not limited to the above-described embodiment. The scope of the present invention is indicated by each claim in the claims after taking into account the description of the detailed description of the invention, and all modifications within the meaning and scope equivalent to the wording described therein are intended. Including.

1 is a block diagram of an automatic dictionary creation system 20 according to an embodiment of the present invention. 3 is a diagram schematically showing an example of the configuration of a dictionary storage device 38. FIG. 3 is a diagram schematically illustrating an example of a configuration of a dictionary attribute storage device 36. FIG. It is a figure for demonstrating the directionality of a dictionary. 4 is a block diagram of a content word translation extraction processing unit 54. FIG. It is a figure for demonstrating the similarity of the words between Korean and Japanese. It is a figure for demonstrating the similarity of the words between Korean and Japanese. It is a figure which shows the threshold value which changes with the number of common translation between Korean and Japanese by a table form. 6 is a block diagram of a function word translation extraction processing unit 62. FIG. It is a figure for demonstrating the principle of the function word pair extraction by the function word translation extraction process part 62. FIG. 1 is an external view of a computer system that realizes an automatic dictionary creation system 20 according to an embodiment of the present invention. It is a block diagram of the computer system shown in FIG.

Explanation of symbols

  20 dictionary automatic creation system, 30 information specifying the first language, 32 information specifying the second language, 34 dictionary automatic creation device, 36 dictionary attribute storage device, 38 dictionary storage device, 40 electronic bilingual dictionary, 50 input device, 52 content word dictionary selection unit, 54 content word translation extraction processing unit, 56 content word pair storage unit, 60 function word dictionary selection unit, 62 function word translation extraction processing unit, 64 function word pair storage unit, 70 merge processing unit , 180 word pair extraction unit, 182 similarity calculation unit, 184 content word pair selection unit, 186 threshold value table storage unit, 220 example sentence extraction unit, 222 matching sentence extraction unit, 224 alignment processing unit, 226 function word pair selection unit

Claims (6)

A bilingual dictionary creation method for automatically creating a bilingual dictionary between a first language and a second language by using a third language as an intermediary,
A first dictionary with directionality from the first language to the third language for the first language speaker, and the second language to the second language for the second language speaker; Preparing an electronically readable second dictionary having directionality to three languages;
Of entry pairs consisting of entries related to content words in the first dictionary and entries related to content words in the second dictionary, defined as a predetermined function for a set of translated words in the third language included in each entry Extracting an entry pair whose similarity value is greater than or equal to a predetermined threshold;
A bilingual dictionary creation method including a first step of storing the content word pairs corresponding to the entry pairs extracted in the extracting step in an electronically readable form. The step of preparing comprises
Providing a plurality of electronically readable bilingual dictionaries, each having a predetermined direction, for a predetermined language speaker;
Receiving a designation for the first language and the second language;
A dictionary having a direction from the first language to another language designated in the step of receiving the designation from the plurality of bilingual dictionaries, and the other language from the second language designated in the step of receiving the designation. The bilingual dictionary creation method according to claim 1, further comprising a step of selecting a dictionary pair including a dictionary having directionality to a language. The step of preparing comprises
Providing a plurality of electronically readable bilingual dictionaries, each having a predetermined direction, for a predetermined language speaker;
Receiving designations for the first language and the second language, and the third language;
From the plurality of bilingual dictionaries, a dictionary having a direction from the first language to the third language designated in the step of receiving the designation, and from the second language designated in the step of receiving the designation The bilingual dictionary creation method of Claim 1 including the step of selecting the dictionary pair which consists of a dictionary with the directionality to the said 3rd language. A third dictionary with direction from the third language to the first language for speakers of the first language and a third dictionary from the third language for speakers of the second language; Preparing a fourth dictionary with directionality in two languages in an electronically readable form;
Extracting the third language example sentence that satisfies a predetermined condition and its translation from both the third dictionary entry and the fourth dictionary entry corresponding to the same third language heading Creating a bilingual sentence pair;
Of the bilingual sentence pairs created in the step of creating the bilingual sentence pair, taking the alignment between the translated sentences;
A second step of extracting function word pairs from the chunks aligned with each other in the step of taking the alignment and storing them in an electronically readable form;
4. The method according to claim 1, further comprising: merging the content word pairs accumulated in the first and second steps and the function word pairs accumulated in the electronically readable form. How to create a bilingual dictionary as described in Crab. A computer program that, when executed by a computer, controls the computer to execute all the steps of the bilingual dictionary creation method according to any one of claims 1 to 4. A bilingual dictionary creation device for automatically creating a bilingual dictionary between a first language and a second language, using the third language as an intermediary,
A first dictionary with directionality from the first language to the third language for the first language speaker, and the second language to the second language for the second language speaker; Dictionary preparing means for preparing a second dictionary having directionality to three languages in an electronically readable form;
Of entry pairs consisting of entries related to content words in the first dictionary and entries related to content words in the second dictionary, defined as a predetermined function for a set of translated words in the third language included in each entry An entry pair extraction means for extracting an entry pair whose similarity value is equal to or greater than a predetermined threshold value;
A bilingual dictionary creation device comprising: content word pair storage means for storing content word pairs corresponding to the entry pairs extracted by the entry pair extraction means in an electronically readable form.

Images