JP4708682B2 - Bilingual word pair learning method, apparatus, and recording medium on which parallel word pair learning program is recorded - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a learning method for extracting bilingual word pairs having a bilingual relationship, and more particularly to a learning method for automatically extracting bilingual word pairs from a natural language sentence having a bilingual relationship.
[0002]
[Prior art]
There is known a method of automatically learning bilingual word pairs from bilingual sentence pairs in a bilingual relationship using statistical information (see Patent Document 1). For example, word pairs with greatly different language structures such as between Japanese and English. In the case of, there was a problem in the accuracy of extracting parallel word pairs. In addition, when such a language structure is greatly different, a learning method (see Non-Patent Document 1) is known that uses the syntax information of one language to improve word pair extraction accuracy.
[0003]
[Patent Document 1]
Japanese Patent Application Laid-Open No. Hei 5-189482 "Computer Operation Method for Translation, Lexical Model Generation Method, Model Generation Method, Translation Computer System, Lexical Model Generation Computer System and Model Generation Computer System"
[Non-Patent Document 1]
Kenji Yamada and Kevin Knight, `` A Syntax-based Statistical Translation Model '', 39th Annual Meeting of the Association for Computation Linguistics (ACL-01), pp. 523-530, 2001
[0004]
[Problems to be solved by the invention]
In the conventional method of learning a pair of translated words in Patent Document 1 and Non-Patent Document 1 described above, when a base language to be associated changes between two languages of a translated sentence, the obtained word pair changes, and a stable word pair cannot be extracted. There's a problem. Further, in the learning method of Non-Patent Document 1, although the extraction accuracy is improved as compared with the method that does not use syntax information, the accuracy is not yet sufficient.
The present invention has been made in view of the above points, and stable bilingual words can be obtained by extracting the same bilingual word pairs from bilingual word pairs obtained by using syntactic information of the bilingual sentences inputted for each language. A bilingual word pair learning method for extracting pairs, and adding the extracted common bilingual word pairs to the input bilingual sentence and repeating the bilingual word pair learning to improve the accuracy of bilingual word pair extraction, It is an object of the present invention to provide an apparatus and a recording medium on which a parallel word pair learning program is recorded.
[0005]
[Means for Solving the Problems]
FIG. 1 and FIG. 4 are schematic flowcharts of a bilingual word pair learning method for explaining the present invention.
The present invention is an apparatus for learning a word pair in a bilingual relationship by inputting a first natural language sentence and a second natural language sentence that is a translation of the first natural language sentence. A natural language sentence is input (step 1), and the word or word string that is the element of the first natural language sentence is associated with the word or word string that is the element of the second natural language sentence, to form a set of bilingual word pairs (Step 2), a word or word string that is an element of the first natural language sentence is associated with a word or word string that is the element of the second natural language sentence, and a set of parallel word pairs is obtained ( Step 3), extracting the same translation pair from the set of these two types of translation word pairs, and the extracted common word pair is already stored in the storage device 9 by the common translation word pair extraction means Judge whether or not all of the pairs match. If it is determined not to store the common word pair not the match to the storage device 9 as a common translation word pair, if it is determined that match all the process ends,
The word-by-language word extracting means converts the common bilingual word pairs stored in the storage device 9 into words or word strings that are elements of the first natural language sentence and words or word strings that are elements of the second natural language sentence. And store in the bilingual corpus and return to step 1 (step 4).
[0006]
Further, the present invention includes a step of associating a word or word string with a syntax tree obtained by providing a syntax analysis step in the process of associating a word or word string in the bilingual word pair acquisition step.
[0007]
In another reference example of the present invention, a word or word string that is an element of a first natural language sentence and a first natural language sentence of a common word pair obtained by the step of extracting the same parallel translation pair. For a word or word string that is an element, a word or word string that is an element of a second natural language sentence, and a second natural language sentence of a common word pair obtained by extracting the same parallel translation pair A second natural language sentence of a common word pair obtained by associating a word or word string as an element, a word or word string as an element of a second natural language sentence, and the step of extracting the same parallel translation pair A first natural language of a common word pair obtained by extracting a word or word string that is an element of a first natural language sentence and the same parallel translation pair with respect to a word or word string that is an element of Comprising the step of associating a word or word string is a component.
In addition, the present invention includes a step of repeating the method of learning a parallel word pair until the common word pair is changed by using the common word pair obtained in the step of extracting the same parallel translation pair.
[0008]
2 and 3 are schematic configuration diagrams of the bilingual word pair learning apparatus of the present invention.
The bilingual word pair learning device includes a bilingual sentence reading unit 1 that reads a first natural language sentence and a second natural language sentence that is a parallel translation thereof, and a word or a word string that is an element of the first natural language sentence. A word or a word string that is an element of the natural language sentence of 2 and a parallel translation word pair that is associated based on the first natural language are stored in the storage device 71, and an element of the second natural language sentence Bilingual word pair acquisition means for associating a word or word string, which is an element of the first natural language sentence, with a certain word or word string, and storing in the storage device 72 bilingual word pairs associated with each other based on the second natural language 6 and the parallel translation word pair which is the same translation word pair in the storage devices 71 and 72 are extracted, and the extracted common word pair is already stored in the storage device 9 by the common translation word pair extraction means. Judges whether or not all of the common bilingual word pairs match If it is determined that they do not match all, the non-matching common word pair is stored in the storage device 9 as a common bilingual word pair, and if it is determined that all match, a common bilingual word pair extraction unit that ends the processing; ,
The common bilingual word pair stored in the storage device 9 by the common bilingual word pair extraction means is a word or word string that is an element of a first natural language sentence and a word or word that is an element of a second natural language sentence. A bilingual word pair acquisition means based on the first natural language, a bilingual word pair acquisition means based on the second natural language, and a common bilingual word pair extraction means 8 which are stored in the bilingual corpus divided into columns. Language-specific word extraction means to be operated is included.
[0009]
The present invention also uses a morphological analysis unit 2 for morphological analysis of a sentence read by the bilingual sentence reading unit 1 and subsequently parses it in order to use a word cut result and a syntax analysis result in the bilingual word pair acquisition unit 6. The parsing means 3 is included.
According to another embodiment of the present invention, in order to use the common translation word pair stored in the storage device 9 by the common translation word pair extraction means 8 in the translation word pair acquisition means 6, the common translation word pair is converted into a language. It includes language-specific word extraction means A.
In addition, another embodiment of the present invention includes means for repeating learning of parallel word pairs until the common parallel word pairs stored in the storage device 9 by the common parallel word pair extracting means 8 do not change.
[0010]
(Function)
As described above, in the parallel word pair learning method of the present invention, the first natural language sentence and the second natural language sentence corresponding to the first natural language sentence are input to the parallel sentence reading means 1, and the parallel word pair acquisition means 6 is input. The word or word string that is the element of the second natural language sentence is associated with the word or word string that is the element of the first natural language sentence, and the bilingual word pair that is associated based on the first natural language is stored In the device 71, the parallel word pair acquisition means 6 associates the word or word string that is the element of the second natural language sentence with the word or word string that is the element of the second natural language sentence, Bilingual word pairs associated with each other in the natural language are stored in the storage device 72, and the common bilingual word pair extraction means 8 extracts common bilingual word pairs that are the same bilingual word pairs in the storage devices 71 and 72. And stored in the storage device 9. Thereby, a stable parallel word pair can be automatically extracted from the first natural language sentence and the second natural language sentence which is a parallel translation thereof.
[0011]
Also, the morpheme analysis unit 2 performs morpheme analysis of the sentence input by the parallel translation reading unit 1, stores the morpheme analysis results in the storage devices 41 and 42, and further performs the syntax analysis by the syntax analysis unit 3 to perform the syntax analysis. The results are stored in the storage devices 51 and 52, and the translated word pair acquisition means 6 uses the morphological analysis results and the syntax analysis results to store the translated word pairs associated with each other based on the first natural language. 71, and the parallel translation word pair associated with the second natural language is stored in the storage device 72. The common translation word pair extraction means 8 uses the same word pair in the storage devices 71 and 72. A common bilingual word pair is extracted and stored in the storage device 9. Thereby, a common parallel translation word pair can be extracted more correctly.
[0012]
In addition, the word-by-language word extraction unit A divides the common bilingual word pair stored in the storage device 9 by the common bilingual word pair extraction unit 8 by language, and in addition to the bilingual sentence input by the bilingual sentence reading unit 1, The parallel word pair acquisition means 6 uses the word or word string that is the element of the first natural language sentence and the word or word string that is the element of the second natural language sentence. Thereby, the common parallel translation word pair extraction means 8 can extract more common parallel translation word pairs more accurately.
[0013]
Further, the means for repeating the learning of the parallel translation word pairs is performed until the common translation word pairs stored in the storage device 9 by the common translation word pair extraction means 8 no longer change. As a result, more common parallel word pairs can be extracted more accurately.
Therefore, by executing all the above methods and using the above means, a parallel word pair is automatically obtained, and a parallel word pair can be learned.
The above description describes the learning method of the parallel word pair, but the same applies to the parallel word pair learning device and the parallel word pair learning program.
[0014]
DETAILED DESCRIPTION OF THE INVENTION
An embodiment of the present invention will be described below with reference to the drawings.
FIG. 3 is a basic block configuration diagram of a bilingual word pair learning device according to an embodiment of the present invention. The bilingual word pair learning device shown in FIG. 1 includes a bilingual sentence reading unit 1, a morpheme analyzing unit 2, a syntax analyzing unit 3, a bilingual word pair acquiring unit 6, a common bilingual word pair extracting unit 8, a language-specific word extracting unit A, The storage devices 41, 42, 51, 52, 71, 72 and 9 are configured.
[0015]
The bilingual sentence reading unit 1 inputs (reads) the first natural language sentence stored in the bilingual corpus and the second natural language sentence that is the parallel translation.
The morpheme analysis unit 2 stores the result of morphological analysis of the input first natural language sentence in the storage device 41 and stores the result of morphological analysis of the input second natural language sentence in the storage device 42.
The syntax analysis unit 3 uses the morphological analysis result stored in the storage device 41 to analyze the syntax of the first natural language sentence, stores the result in the storage device 51, and stores the result in the storage device 42. The second natural language sentence is analyzed by using the morpheme analysis result, and the result is stored in the storage device 52.
[0016]
The parallel word pair acquisition unit 6 includes a parallel word pair acquisition unit 61 based on the first natural language and a parallel word pair acquisition unit 62 based on the second natural language.
The bilingual word pair acquisition unit 61 based on the first natural language, for example, translates the first natural language sentence by associating the result of syntactic analysis of the second natural language sentence with the word string that is the result of morphological analysis. The word pair is extracted and stored in the storage device 71.
The bilingual word pair acquisition unit 62 based on the second natural language, for example, translates the second natural language sentence by associating the syntactic analysis result of the first natural language sentence with the word string that is the result of the morphological analysis. The word pair is extracted and stored in the storage device 72.
The common bilingual word pair extraction unit 8 extracts the same bilingual word pair from the set of bilingual word pairs stored in the storage devices 71 and 72 and stores it in the storage device 9.
The word-specific word extraction unit A divides the common bilingual word pairs stored in the storage device 9 into words of the first natural language and the second natural language, and the bilingual corpus read by the bilingual sentence reading unit 1 Add to the translation.
[0017]
FIG. 4 is a flowchart for repeating learning according to an embodiment of the present invention.
Hereinafter, based on this flowchart, an embodiment in the case where the first natural language is English and the second natural language is Japanese will be described.
In step 101, the first natural language sentence and the second natural language sentence corresponding to the first natural language sentence are input to the parallel sentence reading unit 1. For example, as shown in Fig. 5, the translated text is "The house is somewhere about here", "Look about", and the translated Japanese text is "The house is somewhere around", " Suppose you include "Look around".
[0018]
In step 102, the parallel translation read in step 101 is subjected to morphological analysis by the morphological analysis unit 2, and the results are stored in the storage devices 41 and 42. For example, as shown in FIG. 6, it is assumed that the Japanese sentence is cut into words “that / house / has / where / somewhere / this / around / in / be” and “around / see / turn / turn”. In addition, the morpheme analysis unit 2 assigns parts of speech to each word for syntax analysis. For example, in English, as shown in FIG. 7, it is assumed that “Look” is given a verb, “about” is an adverb, and the part of speech shown in FIG. Also, for example, in Japanese, as shown in FIG. 8, “around” is given a noun, “wa” is a particle, “see” is a verb, “mawasu” is a verb, and the part of speech shown in FIG. Suppose that
In step 103, the syntax analysis unit 3 performs syntax analysis based on the morphological analysis results stored in the storage devices 41 and 42, and stores the results in the storage devices 51 and 52, respectively. For example, it is assumed that the results shown in FIGS. 7 and 9 are obtained for English and the results shown in FIGS. 8 and 10 are obtained for Japanese.
[0019]
In step 104 and step 105, using the morpheme analysis results and the syntax analysis results stored in the storage devices 41, 42, 51, 52, the translated word pairs based on the first natural language are used. The acquisition unit 61 stores the parallel word pairs based on the first natural language in the storage device 71, and the parallel word pair acquisition unit 62 based on the second natural language uses the second natural language as a basis. The translated word pairs are stored in the storage device 72.
For example, in the above example, as shown in Fig. 11, when syntactic analysis results of Japanese sentences are associated based on English sentences that have undergone morphological analysis, "the" and "the", "house" and "house", etc. Are associated with each other, and when the English syntax analysis result is associated based on the Japanese sentence that has been subjected to morphological analysis, it is assumed that "the" and "that", "house" and "house", etc. are associated. . Note that a blank indicates that there is no corresponding item. For example, “ka” in the case of associating Japanese sentence syntactic analysis results based on the English sentence in FIG. 11 indicates that there is no corresponding English word or word string. means. Further, in this example, the word order of the syntax analysis result is exchanged so as to match the word string as much as possible within a range not exceeding the intermediate node unit of the syntax analysis result. Here, the intermediate node of the parsing result indicates a grammatical category on the parsing result such as a verb, a noun, a verb phrase, and a noun phrase. The term “intermediate node unit” refers to a word string that belongs below the intermediate node (that is, the grammatical category) in the parsing result. For example, in the parsing result shown in FIG. 8, the noun phrase unit is lower than the noun phrase. Is a word string that belongs to, and refers to “around” and “to”, and the verb phrase unit refers to “see” and “turn”. The range that does not exceed the intermediate node unit means that it is limited to the range in the word string that is the intermediate node unit. For example, in the result of the parsing shown in FIG. 8, “around” and “to” in the range of the noun phrase unit are interchanged, or “around / to” of the entire noun phrase unit and “see / turn” of the entire verb phrase unit. However, because the particle `` O '' is a word in the range of a noun phrase, and the verb `` mi '' is a word in the range of a verb phrase, the particle `` You can't interchange the verb "see". Therefore, when associating the English sentence “look about” with the syntax analysis result of the Japanese sentence “Per / Oh / Look / Turn”, as shown in FIG. 11 (above), the noun phrase “Oh / O” and the verb phrase “ By changing the word order of “/ Mawasu”, “look” and “look” and “around” and “about” are associated with each other. These parallel translation word pairs are stored in the storage devices 71 and 72.
[0020]
In step 106, the common bilingual word pair extraction unit 8 extracts the same bilingual word pair from the bilingual word pairs stored in the storage device 71 and the bilingual word pairs stored in the storage device 72, and has already been stored in the storage device 9. If all are the same (yes), if they are not the same (no), it will be (no) .If yes is selected, the bilingual word pair learning will end. , No is selected, the process proceeds to the next step 107. In the above example, common bilingual word pairs extracted by the common bilingual word pair extraction unit 8 from the input sentence shown in FIG. 5 are, for example, “that”, “the”, “house”, “ house "etc. are extracted. Since nothing is stored in the storage device 9, no is selected and the process proceeds to the next step 107.
[0021]
In step 107, the parallel translation word pair that does not match the common translation word pair stored in the storage device 9 in step 106 is stored in the storage device 9. In the case of the above example, all the translated word pairs extracted in step 106 shown in FIG.
In step 108, the common bilingual word pairs stored in the storage device 9 are divided for each language and added to the input bilingual translations. That is, the common bilingual word pair is stored in the bilingual corpus. In the case of the above example, “the”, “house”, “is”, “about” are entered in the English sentences that are divided by language, and “that”, “house” ",""Ha" and "around" are added.
[0022]
In the above example, the process proceeds to step 101 again. Therefore, the above example will be further described step by step.
In step 101, both the original translated sentence and the translated word pair added in step 108 are input to the translated sentence reading unit 1.
In step 102, the morphological analysis unit 2 assigns the part of speech of the translated word pair input in step 101, and the result is stored in the storage devices 41 and 42.
In step 103, the morphological analysis result stored in the storage devices 41 and 42 is used for the bilingual word pair input in step 101 in the syntax analysis unit 3, and the result is stored in the storage devices 51 and 52. .
[0023]
In Step 104 and Step 105, the first sentence is stored in the storage devices 41, 42, 51, 52 using the morpheme result and the syntax analysis result of the word or word string of the input sentence and the common word pair. A bilingual word pair acquisition unit 61 based on the first natural language is stored in the storage device 71 by the bilingual word pair acquisition unit 61 based on the natural language, and a bilingual word pair acquisition unit 62 based on the second natural language. As a result, the translated word pairs based on the second natural language are stored in the storage device 72. In the case of the above example, since the common bilingual word pairs “around” and “about” are added to the input sentence (see FIG. 12), when the parsing result corresponds to the word string, “around” and “about” It is considered to be able to respond. For example, if the Japanese sentence “that / house / has / where / somewhere / this / around / n / a” is combined with the English sentence “The house is somewhere about here”, the verb from the parsing result shown in FIG. The phrase “al” and the adverb phrase “this / around / ni” are interchanged. Further, in the adverb phrase, the conjunction “this” and the particle “ni” are interchanged, and the word order is as shown in FIG. As a result, for example, correspondence with English sentences is as shown in FIG. 15, and “this” is associated with “here”. On the other hand, for the same example sentence, when the syntactic analysis result of English is combined with the Japanese sentence, the adverb “about” and the adverb “here” in the adverb phrase are switched from the parsing result shown in FIG. Word order. As a result, for example, correspondence with Japanese sentences is as shown in FIG. 15, and “here” and “this” are associated with each other. These bilingual word pairs shown in FIG. 15 are stored in the storage devices 71 and 72.
[0024]
In step 106, when the common translation word pair extraction unit 8 extracts the same translation word pairs from the translation word pairs stored in the storage device 71 and the translation word pairs stored in the storage device 72, in the above example, A common bilingual word pair of “this” and “here” is newly extracted. This does not match the common bilingual word pair of FIG. 12 currently stored in the storage device 9, so the determination at this step is no and the process proceeds to step 107.
[0025]
In step 107, “this” and “here” that do not match those in FIG. 12 stored in the storage device 9 are added, and the common bilingual word pairs shown in FIG. 16 are stored in the storage device 9.
In step 108, the common bilingual word pairs stored in the storage device 9 are divided for each language and added to the input bilingual translations.
Thereafter, the process further proceeds to step 101. However, when the common bilingual word pair does not change, the determination in step 106 becomes yes, and the bilingual word pair learning ends.
Although the above example has been described based on the configuration diagram shown in FIG. 3, the present invention is not limited to this example, and various modifications and applications can be made within the scope of the claims.
[0026]
The bilingual word pair learning apparatus of the present invention can be composed of a computer having a CPU, a memory, and the like, a user terminal, and a machine-readable recording medium such as a CD-ROM, a magnetic disk device, and a semiconductor memory.
A bilingual word pair learning program recorded on a recording medium or a bilingual word pair learning program transmitted via a communication line is read by a computer to control the operation of the computer. Implement each process.
[0027]
【The invention's effect】
As described above, according to the present invention, a bilingual word in which a first natural language sentence is associated with a second natural language sentence from the first natural language sentence and the second natural language sentence that is a parallel translation thereof. By comparing the pair and the second natural language sentence with the bilingual word pair associated with the first natural language sentence and extracting the common bilingual word pair, the first stable natural language and the first Two natural language bilingual word pairs can be extracted.
Also, by using the result of using the morphological analysis means and the syntax analysis means in the associating step, it is possible to extract the first natural language and second natural language parallel word pairs more accurately.
Further, by adding the bilingual word pairs extracted once to the input bilingual sentence, more bilingual word pairs of the first natural language and the second natural language can be extracted.
Further, by repeating the learning of the above-described bilingual word pairs until the extracted bilingual word pairs no longer change, the bilingual word pairs of the first natural language and the second natural language can be extracted more accurately.
The bilingual word pairs extracted in this way can be used for, for example, construction or expansion of an electronic bilingual dictionary, or can be used as a bilingual dictionary of a machine translation system.
[Brief description of the drawings]
FIG. 1 is a schematic flowchart of a bilingual word pair learning method for explaining the principle of the present invention;
FIG. 2 is a schematic configuration diagram of a bilingual word pair learning apparatus according to the present invention.
FIG. 3 is a basic block configuration diagram of a bilingual word pair learning apparatus according to an embodiment of the present invention.
FIG. 4 is a flowchart for repeating learning according to an embodiment of the present invention.
FIG. 5 is a diagram showing an example of an input parallel translation sentence according to an embodiment of the present invention.
FIG. 6 is a diagram showing an example of a Japanese sentence morphological analysis result (word cut) according to an embodiment of the present invention.
FIG. 7 is a diagram showing an example (part 1) of the English syntax analysis result according to the embodiment of the present invention;
FIG. 8 is a diagram showing an example (part 1) of a Japanese sentence syntactic analysis result according to an embodiment of the present invention;
FIG. 9 is a diagram showing an example (part 2) of the English syntax analysis result according to the embodiment of the present invention;
FIG. 10 is a diagram illustrating an example (part 2) of a Japanese sentence syntactic analysis result according to an embodiment of the present invention;
FIG. 11 is a diagram showing an example (part 1) of the word correspondence result according to the embodiment of the present invention.
FIG. 12 is a diagram showing an example (part 1) of common word correspondence according to an embodiment of the present invention.
FIG. 13 is a diagram showing an example of exchanging the word order of the Japanese sentence syntactic analysis results according to an embodiment of the present invention.
FIG. 14 is a diagram showing an example in which the word order of the English syntax analysis result is switched according to an embodiment of the present invention.
FIG. 15 is a diagram showing an example (part 2) of the word correspondence result according to the embodiment of the present invention.
FIG. 16 is a diagram showing an example (part 2) of common word correspondence that is an embodiment of the present invention.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Bilingual sentence reading part, 2 ... Morphological analysis part, 3 ... Syntax analysis part, 6 ... Parallel word pair acquisition part, 8 ... Common parallel translation word pair extraction part, 9 ... Common bilingual word pair, 71 ... bilingual word pair storage device associated with the first natural language, 72 ... bilingual word pair storage device associated with the second natural language, A ...・ Language extraction unit

Claims (5)

In a bilingual word pair learning method for extracting a word pair having a bilingual relationship with respect to a first natural language sentence stored in a bilingual corpus separately for each natural language sentence and a second natural language sentence that is a bilingual sentence,
First of translation word pair acquisition means based on natural language, the translation of the first natural language sentence stored in corpus element at a word or word string a second stored in the bilingual corpus Generating and storing a bilingual word pair that is a pair of a word or a word string in a bilingual relationship obtained in association with a word or word string that is an element of a natural language sentence;
Bilingual word pairs acquisition means based on a second natural language, said second natural language text which is stored in the bilingual corpus element in which the word or word string a first stored in the bilingual corpus A step 2 of generating and storing a bilingual word pair that is a pair of a word or a word string in a bilingual relationship obtained in association with a word or word string that is an element of a natural language sentence;
The common bilingual word pair extraction means compares the bilingual word pairs stored in the procedure 1 and the procedure 2, extracts common word pairs that are the same bilingual word pairs, and the extracted common word pairs are already common. The bilingual word pair extraction means determines whether or not all the common bilingual word pairs stored in the storage device (9) match, and if it does not match all the common bilingual word pairs, Store in the storage device (9) as a word pair, and if it is determined that all match, step 3 ends the process;
The language-specific word extracting means converts the common bilingual word pairs stored in the storage device (9) into words or word strings that are elements of the first natural language sentence and words that are elements of the second natural language sentence or A method of learning bilingual word pairs, comprising: a step 4 of dividing into word strings and storing in the bilingual corpus and returning to the step 1. In the learning method of the parallel translation word pair of Claim 1,
In the steps 1 and 2, the bilingual word pair acquisition unit based on the first natural language and the bilingual word pair acquisition unit based on the second natural language parse the inputted natural language sentence. A bilingual word pair learning method characterized by associating a syntax tree obtained by analysis with a word or a word string. In a bilingual word pair learning device that extracts a word pair having a bilingual relationship with respect to a first natural language sentence stored in a bilingual corpus separately for each natural language sentence and a second natural language sentence that is a bilingual sentence,
Obtained in association with the first word or word string is an element of the second natural language sentence the word or word string is an element of the natural language text is stored in the bilingual corpus stored in the bilingual corpus A bilingual word pair acquisition means based on a first natural language for generating and storing a bilingual word pair that is a pair of a word or a word string in a translated relation;
Obtained in association with the first word or word string is an element of the natural language text which is stored the element word or word string which is the second natural language text which is stored in the bilingual corpus to the bilingual corpus A bilingual word pair acquisition means based on a second natural language for generating and storing a bilingual word pair that is a pair of words or word strings having a bilingual relationship,
Comparing the bilingual word pair stored by the bilingual word pair acquisition means based on the first natural language and the bilingual word pair stored by the bilingual word pair acquisition means based on the second natural language; same or bilingual word extracting a is common word pair pairs, the extracted common word pairs, already consistent with all common translation word pairs stored in the storage device (9) by a common translation word pair extracting means If not, the common word pair that does not match is stored in the storage device (9) as a common bilingual word pair, and the process ends if it is determined that all match. Common bilingual word pair extraction means;
The common bilingual word pair stored in the storage device (9) by the common bilingual word pair extraction means is a word or word string that is an element of the first natural language sentence and a word that is an element of the second natural language sentence. Alternatively, it is stored in the bilingual corpus divided into word strings, and bilingual word pair acquisition means based on the first natural language and bilingual word pair acquisition means based on the second natural language and common bilingual word pair extraction means And a word-by-language word extraction means for operating the translation word pair learning apparatus. The parallel translation word pair learning device according to claim 3,
The bilingual word pair acquisition means based on the first and second natural languages associates a syntax tree obtained by parsing the input natural language sentence with a word or a word string, Learning device.   A computer-readable recording medium storing a program for causing a computer to execute each process of the bilingual word pair learning device according to claim 3.

Images