JP2000250913A - Example type natural language translation method, production method and device for list of bilingual examples and recording medium recording program of the production method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform the translation by means of a proper list of bilingual examples corresponding to the sentences which are produced by eliminating the pairs of sentences having the low sentence correspondence accuracy out of a sentence correspondence text. SOLUTION: A sentence correspondence text storing the bilingual pairs of sentences of the 1st and 2nd natural languages is inputted (S1), and a list of bilingual examples with which a sentence of the 2nd natural language is equal to the proper translation of a sentence of the 1st natural language is produced from the sentence correspondence text (S2). Then a sentence similar to an input sentence of the 1st natural language is retrieved from the list of the bilingual examples (S3), and the difference is decided between the input sentence and its similar sentence (S4). According to this difference information, a sentence of the 2nd natural language corresponding to the similar sentence is mechanically processed (S5). Then a translation of the input sentence is outputted (S6).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to the first existing device.
Example natural language translation method for automatically translating a second natural language sentence for an input sentence of a first natural language by imitating the relationship between a sentence of a natural language and a sentence of a second natural language, and a method of creating a bilingual example book Further, the present invention relates to an apparatus and a recording medium on which the program is recorded.

[0002]

2. Description of the Related Art First, a sentence correspondence text is prepared from
A bilingual pair of a first natural language sentence and a second natural language sentence similar to the input sentence of the natural language is searched, and a second natural language corresponding to a different part in the first natural language sentence of the searched pair is searched. An example-type natural language translator has been developed which examines a part in a sentence, processes a sentence in a second natural language based on the corresponding part, and generates a translation of the input sentence.

In order to translate various input sentences with such a conventional example-type natural language translation device, a translation pair having a sentence similar to the various input sentences must be searched. First, it is necessary to prepare a large number of translation pairs of the first natural language sentence and the second natural language sentence. Therefore, in order to create a sentence-corresponding text from a text in a first natural language and a text in a second natural language without a sentence correspondence, sentence correspondence has been performed manually or by a machine.

[0004]

However, when sentence matching is performed using texts that are actually used, there is a lack of partial information in a pair or a natural language in which sentence correspondence is incorrectly attached. Sentence corresponding text is included in the sentence corresponding text, so if the sentence corresponding text is used as it is in the example-type natural language translator, if the sentence Problem arises.
In addition, there is a problem in that when manually associating sentences, human cost and time are required.

[0005] The present invention has been made in view of the above,
The purpose is to create an example-type natural language translation method that translates using an appropriate sentence-corresponding bilingual example set created by removing pairs with low sentence-correspondence accuracy from the sentence-corresponding text, and a bilingual example set. It is an object of the present invention to provide a method and apparatus and a recording medium on which the program is recorded.

[0006]

In order to achieve the above object, the present invention according to claim 1 comprises a first natural language sentence and a second natural language sentence.
From the sentence-corresponding text in which the bilingual pair, which is a pair of natural language sentences, is inserted, a bilingual example set, which is a set of bilingual pairs in which the second natural language sentence is an appropriate translation of the first natural language sentence, is created. A similar sentence to the input sentence of the first natural language to be translated is searched from the bilingual example collection, a difference between the input sentence and the similar sentence is extracted, and a second corresponding to the similar sentence is extracted using the difference information. The gist is to mechanically process a sentence in a natural language to generate a translated sentence of the input sentence.

According to the first aspect of the present invention, a sentence of a second natural language is converted from a sentence of a first natural language from a sentence corresponding text containing a bilingual pair of a sentence of a first natural language and a sentence of a second natural language. Creates a bilingual example collection that is an appropriate translation of the sentence, searches the bilingual example collection for a similar sentence to the input sentence of the first natural language, extracts the difference between the input sentence and the similar sentence, and uses the difference information to perform similarity. Since the translation of the input sentence is generated by mechanically processing the second natural language sentence corresponding to the sentence, it is possible to create a bilingual example collection narrowed down to an appropriate bilingual pair. Can generate a translation by imitating an appropriate translation pair.

[0008] The present invention according to claim 2 is based on claim 1.
In the described invention, the gist is that the bilingual example collection is created by combining a bilingual pair, which is extracted by a plurality of different methods, and in which the second natural language sentence is an appropriate translation of the first natural language sentence. I do.

According to the second aspect of the present invention, the bilingual example set is extracted by a plurality of different methods, and is combined with a bilingual pair in which the sentence of the second natural language is an appropriate translation of the sentence of the first natural language. Created.

Further, the present invention according to claim 3 provides the present invention as claimed in claim 2.
In the invention described above, one of the plurality of different methods is to extract a bilingual pair that is often similar to a sentence-corresponding text as a fixed bilingual pair.

According to the third aspect of the present invention, as one of a plurality of different methods, a bilingual pair that is frequently similar to a text corresponding to a sentence is extracted as a fixed bilingual pair.

According to a fourth aspect of the present invention, in the second aspect, as one of the plurality of different methods,
The gist of the present invention is to check the correspondence between words and phrases between a sentence in the first natural language and a sentence in the second natural language, and to extract a translation pair that has word / phrase correspondence in the entire sentence.

According to the present invention, as one of a plurality of different methods, the correspondence between phrases in a sentence in a first natural language and a sentence in a second natural language is checked, and the phrase in the entire sentence is examined. Extract the corresponding translation pairs.

Further, the present invention described in claim 5 is based on claim 1.
In the invention described above, the gist of the invention is that the sentence-corresponding text is composed of a pair of a text in a first natural language and a text in a second natural language without a sentence correspondence, the sentence being correlated by hand and machine.

According to the present invention, the sentence-corresponding text is a pair in which a sentence is manually and manually matched from a text of a first natural language and a text of a second natural language, each of which has no sentence correspondence. Constitute.

Further, according to the present invention, the second natural language sentence is converted into the first natural language sentence from the sentence corresponding text in which a translation pair which is a pair of the first natural language sentence and the second natural language sentence is stored. The gist is to narrow down to bilingual pairs that are appropriate translations of natural language sentences and to create bilingual translation examples.

According to the sixth aspect of the present invention, a sentence in a second natural language is converted from a sentence corresponding text in which a translation pair which is a pair of a sentence in a first natural language and a sentence in a second natural language is stored. (1) A bilingual example book is created by narrowing down to a bilingual pair that is an appropriate translation of a sentence in natural language.

According to a seventh aspect of the present invention, in the invention according to the sixth aspect, the bilingual example collection is extracted by a plurality of different methods, and the second natural language sentence is an appropriate one of the first natural language sentence. The gist is that it is created by combining the translation pairs to be translated.

According to the present invention, a bilingual example set is extracted by a plurality of different methods, and a bilingual pair in which a sentence of the second natural language is an appropriate translation of a sentence of the first natural language is combined. Created.

The present invention according to claim 8 provides the present invention according to claim 7.
In the invention described above, one of the plurality of different methods is to extract a bilingual pair that is often similar to a sentence-corresponding text as a fixed bilingual pair.

According to the present invention, as one of a plurality of different methods, a well-known bilingual pair frequently appearing in a sentence-corresponding text is extracted as a standard bilingual pair.

Further, the present invention described in claim 9 provides the present invention in claim 7.
In the invention described above, as one of the plurality of different methods, a correspondence between words and phrases between a sentence in a first natural language and a sentence in a second natural language is checked, and a translation pair having a word / phrase correspondence in the entire sentence is extracted. The point is to do.

According to the ninth aspect of the present invention, as one of a plurality of different methods, the correspondence between phrases in the first natural language sentence and the second natural language sentence is checked, and the phrase correspondence is determined in the entire sentence. Extract the bilingual pair that has been removed.

According to a tenth aspect of the present invention, in the invention according to the sixth aspect, the sentence-corresponding text is manually or manually converted from a first natural language text and a second natural language text having no sentence correspondence. The gist of the present invention is to form a sentence-matched pair.

According to the tenth aspect of the present invention, the sentence-corresponding text is a pair in which a sentence is manually and manually matched from a text in a first natural language and a text in a second natural language without a sentence correspondence. Constitute.

The present invention according to claim 11 is a sentence-corresponding text input means for reading a sentence-corresponding text containing a pair of a sentence of a first natural language and a sentence of a second natural language; A bilingual pair narrowing-down unit for narrowing down a pair of natural language sentences to be an appropriate translation of the first natural language sentence to create a bilingual example book, and an input sentence inputting a first natural language sentence to be translated Means, a similar sentence search means for searching for a similar sentence to the input sentence from the bilingual example collection,
The difference between the input sentence and the similar sentence is taken, and the first natural language and the second
Using a bilingual dictionary in which the correspondence between natural language terms is described, a phrase correspondence between the difference part of the similar sentence and the second natural language sentence corresponding to the difference part of the input sentence is obtained. And a processing unit for mechanically processing a sentence of the second natural language corresponding to the similar sentence using the difference information to generate a translated sentence of the input sentence, and creating a translated sentence of the input sentence. The gist of the present invention is to have a translation output means.

According to the present invention described in claim 11, the first aspect
A sentence-corresponding text containing a pair of a sentence of a natural language and a sentence of a second natural language is read, and from this sentence-corresponding text, the sentence of the second natural language is narrowed down to a pair that is an appropriate translation of the sentence of the first natural language. A bilingual example set is created, a similar sentence to the input sentence of the first natural language is searched from the bilingual example set, a difference between the input sentence and the similar sentence is obtained, and a second part corresponding to the difference part of the similar sentence is obtained using the bilingual dictionary. A phrase of a part of a natural language sentence is obtained, a translation of a second natural language corresponding to a difference part of the input sentence is obtained, and a sentence of the second natural language corresponding to a similar sentence is mechanically processed using the difference information. In order to generate a translation of the input sentence, it is possible to create a bilingual example set narrowed down to an appropriate bilingual pair, and to use this bilingual example set to generate a translation by imitating a bilingual pair with an appropriate sentence correspondence .

Further, according to a twelfth aspect of the present invention, in accordance with the eleventh aspect of the present invention, the parallel translation narrowing means includes:
An appropriate bilingual pair extracting means for extracting a plurality of different appropriate bilingual pairs from the sentence-corresponding text; and a bilingual example set by combining the bilingual pairs extracted by the appropriate bilingual pair extracting means using a combination condition table and a combination condition identifier. And a means for creating a bilingual example book.

According to the twelfth aspect of the present invention, the translation pair narrowing means extracts a plurality of different suitable translation pairs from the text corresponding to the sentence and uses the extracted translation pairs by using a combination condition table and a combination condition identifier. To create a bilingual example book.

According to a thirteenth aspect of the present invention, in the twelfth aspect of the present invention, the appropriate bilingual pair extraction means performs a morphological analysis to acquire linguistic information of a phrase from the sentence correspondence text. A fixed bilingual translation extracting means for extracting a fixed bilingual pair from the morphologically analyzed sentence-corresponding text; and a phrase correspondence extracting means for extracting a bilingual pair in which the entire sentence has been matched from the morphologically analyzed sentence-corresponding text. The gist is to have.

According to the thirteenth aspect of the present invention, the appropriate bilingual pair extraction means performs a morphological analysis to obtain linguistic information of a phrase from the sentence-corresponding text, and forms a fixed form from the morphologically analyzed sentence-corresponding text. A bilingual pair is extracted, and a bilingual pair in which the whole sentence has a word / phrase correspondence is extracted from the morphologically analyzed sentence corresponding text.

According to a fourteenth aspect of the present invention, there is provided a sentence-corresponding text input means for reading a sentence-corresponding text containing a pair of a sentence of a first natural language and a sentence of a second natural language; The gist of the present invention is to provide a bilingual pair narrowing-down means for narrowing down a sentence in two natural languages to a pair that is an appropriate translation of the first natural language sentence and creating a bilingual example book.

In the present invention according to claim 14, the first aspect
A sentence-corresponding text containing a pair of a sentence of a natural language and a sentence of a second natural language is read, and from this sentence-corresponding text, the sentence of the second natural language is narrowed down to a pair that is an appropriate translation of the sentence of the first natural language. Create a bilingual example book.

Further, according to the present invention as set forth in claim 15, in the invention as set forth in claim 14, the translation pair narrowing means includes:
An appropriate bilingual pair extracting means for extracting a plurality of different appropriate bilingual pairs from the sentence-corresponding text; and a bilingual example set by combining the bilingual pairs extracted by the appropriate bilingual pair extracting means using a combination condition table and a combination condition identifier. And a means for creating a bilingual example book.

According to the present invention, the translation pair narrowing means extracts a plurality of different suitable translation pairs from the text corresponding to the sentence, and compares the extracted translation pairs with a combination condition table and a combination condition identifier. And use it to create a bilingual example book.

According to a sixteenth aspect of the present invention, in the invention according to the fifteenth aspect, the appropriate bilingual pair extraction unit performs a morphological analysis to obtain linguistic information of a phrase from the sentence correspondence text. A fixed bilingual translation extracting means for extracting a fixed bilingual pair from the morphologically analyzed sentence-corresponding text; and a phrase correspondence extracting means for extracting a bilingual pair in which the entire sentence has been matched from the morphologically analyzed sentence-corresponding text. The gist is to have.

According to the sixteenth aspect of the present invention, the appropriate bilingual pair extraction means performs a morphological analysis to acquire linguistic information of a phrase from the sentence-corresponding text, and forms a fixed form from the morphologically analyzed sentence-corresponding text. A bilingual pair is extracted, and a bilingual pair in which the whole sentence has a word / phrase correspondence is extracted from the morphologically analyzed sentence corresponding text.

According to a seventeenth aspect of the present invention, the second natural language sentence is converted from the first natural language sentence and the second natural language sentence pair to the first natural language sentence. A bilingual example set, which is a set of bilingual pairs that are appropriate translations of natural language sentences, is created, and a similar sentence to the first natural language input sentence to be translated is searched from the bilingual example set. A natural language translation program for extracting a difference from a similar sentence and mechanically processing a second natural language sentence corresponding to the similar sentence using the difference information to generate a translated sentence of the input sentence is recorded. The point is to record on a medium.

In the present invention according to claim 17, the first aspect
From the sentence-corresponding text that contains the bilingual pair of the sentence of the natural language and the sentence of the second natural language, a bilingual translation example in which the sentence of the second natural language is an appropriate translation of the sentence of the first natural language is created, A similar sentence to the input sentence of the natural language is searched from the bilingual translation example collection, a difference between the input sentence and the similar sentence is extracted, and a sentence of the second natural language corresponding to the similar sentence is mechanically processed using the difference information. Since the example-type natural language translation program for generating the translated sentence of the input sentence is recorded in the recording medium, it is possible to enhance the circulation by using the recording medium.

Further, in the present invention according to claim 18, in the invention according to claim 17, the bilingual example set includes a bilingual pair in which a sentence in the second natural language is an appropriate translation of a sentence in the first natural language. The gist of the present invention is to record an example-type natural language translation program extracted and created by a plurality of different methods on a recording medium.

According to the eighteenth aspect of the present invention, the bilingual example collection extracts a bilingual pair in which the second natural language sentence is an appropriate translation of the first natural language sentence by a plurality of different methods. Since the example-type natural language translation program to be created is recorded on the recording medium, the distribution of the program can be enhanced by using the recording medium.

According to a nineteenth aspect of the present invention, in the invention according to the eighteenth aspect, as one of the plurality of different methods, a bilingual pair that is frequently similar to a text corresponding to a sentence is extracted as a fixed bilingual pair. The gist is to record an example-type natural language translation program on a recording medium.

According to the nineteenth aspect of the present invention, as one of a plurality of different methods, an example-type natural language translation program for extracting a well-similar bilingual pair frequently appearing in a sentence-corresponding text as a standard bilingual pair. Is recorded on a recording medium, so that the distribution can be improved by using the recording medium.

According to a twentieth aspect of the present invention, in the invention according to the eighteenth aspect, as one of the plurality of different methods, a phrase between a first natural language sentence and a second natural language sentence is used. The gist is to record an example-type natural language translation program for examining the correspondence and extracting a bilingual pair having a phrase correspondence in the entire sentence on a recording medium.

According to the twentieth aspect of the present invention, as one of a plurality of different methods, a sentence in a first natural language and a second
To check the correspondence of phrases between sentences in natural language, and to record on a recording medium an example natural language translation program for extracting a bilingual pair having a phrase correspondence in the entire sentence, using the recording medium, Its distribution can be improved.

Further, in accordance with the present invention as set forth in claim 21, in the invention as set forth in claim 17, the sentence-corresponding text is manually and manually derived from the first natural language text and the second natural language text having no sentence correspondence. The gist of the present invention is to record an example-type natural language translation program composed of pairs matched by a machine with a sentence on a recording medium.

According to the twenty-first aspect of the present invention, the sentence-corresponding text is a pair of a sentence corresponding to a sentence from a first natural language text and a second natural language text to which no sentence correspondence is attached by hand and machine. Since the constituent example natural language translation program is recorded on the recording medium, the distribution of the program can be improved by using the recording medium.

According to a twenty-second aspect of the present invention, a sentence in a second natural language is converted into a first natural language from a sentence corresponding text containing a translation pair, which is a pair of a sentence in a first natural language and a sentence in a second natural language. The gist of the present invention is to record on a recording medium a bilingual example book creation program that creates a bilingual example book by narrowing down to a bilingual pair that is an appropriate translation of the sentence.

In the present invention according to claim 22, the first aspect
From the sentence-corresponding text that contains the bilingual pair, which is a pair of a natural language sentence and a second natural language sentence, narrow down to a bilingual pair in which the second natural language sentence is an appropriate translation of the first natural language sentence. Since the bilingual example book creation program for creating the example book is recorded on the recording medium, the distribution of the example book can be improved by using the recording medium.

According to a twenty-third aspect of the present invention, in the invention according to the twenty-second aspect, the bilingual example collection is obtained by extracting the second natural language sentence by a plurality of different methods from the first natural language sentence. The gist of the present invention is to record a bilingual translation example book creation program created by combining bilingual pairs that are appropriate translations on a recording medium.

According to the twenty-third aspect of the present invention, the bilingual example set is extracted by a plurality of different methods, and a bilingual pair in which the sentence of the second natural language is an appropriate translation of the sentence of the first natural language is combined. Since the bilingual example book creation program created by this method is recorded on a recording medium, the distribution of the bilingual example book collection program can be enhanced using the recording medium.

Further, in the present invention according to claim 24, in the invention according to claim 23, as one of the plurality of different methods, a well-known bilingual pair frequently appearing in a sentence-corresponding text is used as a standard bilingual pair. The gist of the invention is to record the extracted bilingual example book creation program on a recording medium.

According to the present invention described in claim 24, as one of a plurality of different methods, a bilingual example book creation program for extracting a well-similar bilingual pair frequently appearing in a sentence corresponding text as a standard bilingual pair is provided. Since the information is recorded on the recording medium, the distribution property can be improved by using the recording medium.

According to a twenty-fifth aspect of the present invention, in the invention according to the twenty-third aspect, as one of the plurality of different methods, correspondence between a phrase in a first natural language sentence and a second natural language sentence is determined. The gist of the present invention is to record, on a recording medium, a bilingual example book creation program for examining and extracting a bilingual pair having a phrase correspondence in the entire sentence.

According to the twenty-fifth aspect of the present invention, as one of a plurality of different methods, a sentence in a first natural language and a second
By examining the correspondence between words and phrases between sentences in a natural language and recording a bilingual example book creation program that extracts a pair of translations that have word correspondence in the entire sentence on a recording medium, using the recording medium, Distribution can be improved.

According to a twenty-sixth aspect of the present invention, in the twenty-sixth aspect, the sentence-corresponding text is manually and manually converted from a first natural language text and a second natural language text having no sentence correspondence. The gist of the present invention is to record a bilingual translation example book creation program, which is composed of pairs matched by a sentence by a machine, on a recording medium.

According to the twenty-sixth aspect of the present invention, the sentence-corresponding text is a pair in which a sentence is manually and manually matched from a text in a first natural language and a text in a second natural language without a sentence correspondence. Since the constituent bilingual example book creation program is recorded on a recording medium, the distribution of the program can be improved by using the recording medium.

[0058]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a flowchart for explaining the operation principle of an example-type natural language translation apparatus that implements an example-type natural language translation method according to an embodiment of the present invention. The example-type natural language translation method according to the present embodiment includes two methods:
In the first phase, a sentence-corresponding text is input as pre-translation processing (step S1), and the second natural language sentence is converted from the sentence-corresponding text to the first.
A bilingual example collection that is an appropriate translation of a natural language sentence is created (step S2). In the second phase, as a translation process, an input sentence of the first natural language is input, and a similar sentence to the input sentence of the first natural language is searched (step S3).
The difference between the input sentence and the similar sentence is obtained (step S4), and the sentence of the second natural language corresponding to the similar sentence is mechanically processed using the difference information (step S5) to generate a translated sentence of the input sentence. Output (Step S6). If you want to translate another input sentence using a previously created bilingual translation,
Can be skipped and the process can be started from step S3 of phase 2.

Note that steps S1 and S2 in FIG.
Is a flowchart for explaining the operation principle of the bilingual example book creation apparatus that executes the bilingual example book creation method according to the embodiment of the present invention.

Further, in the embodiment, in the sentence-corresponding text, a pair in which the first natural language text having no sentence correspondence and the second natural language text are sentence-associated by hand and machine is used as the sentence-corresponding text. Available as

FIG. 2 is a block diagram showing the configuration of the example-type natural language translator of the embodiment shown in FIG. This apparatus reads a sentence corresponding text containing a pair of a sentence of a first natural language and a sentence of a second natural language.
And a translation pair narrowing unit 200 that narrows down the sentence of the second natural language to an appropriate translation of the first natural language sentence and creates a bilingual example set 300, and an input sentence for inputting the first natural language sentence An input unit 400, a similar sentence search unit 500 that searches the parallel translation example collection 300 for a similar sentence to the input sentence, and a difference between the input sentence and the similar sentence, and the correspondence between the first natural language and the second natural language phrase is determined. Using the described bilingual dictionary 900, a phrase correspondence between the difference portion of the similar sentence and the portion of the second natural language sentence corresponding to the difference portion of the similar sentence, and a difference correspondence unit 600 for obtaining a translation of the second natural language corresponding to the difference portion of the input sentence And a sentence processing unit 700 that mechanically processes a sentence of the second natural language corresponding to the similar sentence based on the difference information, and a translated sentence output unit 800 that generates and outputs a translated sentence of the input sentence.

FIG. 2 is a block diagram showing the configuration of the bilingual example book creating apparatus according to the embodiment shown in FIG. 1. The apparatus comprises a first natural language sentence and a second natural language sentence. A sentence-corresponding text input unit 100 that reads a sentence-corresponding text containing a pair of natural language sentences, and a pair of sentences in which the second natural language sentence is an appropriate translation of the first natural language sentence.
Is provided.

FIG. 3 is a block diagram showing the configuration of the translation pair narrowing unit 200 used in the example-type natural language translator shown in FIG. The bilingual pair narrowing unit 200 extracts the bilingual pair having a plurality of appropriate sentence correspondences from the appropriate bilingual pair extracting unit 201 and the fixed bilingual pair extracting unit 220 from the combination condition table 250 by a method designated by a predetermined combination condition identifier. It has a bilingual example set creating unit 240 that creates the bilingual example set 300 by combining the bilingual pairs obtained by the correspondence extracting unit 230. Appropriate translation pair extraction unit 201
Is a morphological analysis unit 210 that performs a morphological analysis to obtain part-of-speech information of each word in each sentence of the sentence-corresponding text, and a stylized bilingual extraction unit that extracts a stylized bilingual pair from the morphologically analyzed sentence-corresponding text 220 and the second sentence of the first natural language sentence and its pair from the morphologically analyzed sentence corresponding text
It has a phrase correspondence extraction unit 230 that extracts a translation pair having a phrase correspondence as a whole between sentences in a natural language.

FIG. 5 is a block diagram of the fixed-form bilingual extraction unit 22 shown in FIG.
FIG. 3 is a block diagram showing a configuration of a 0. Boilerplate bilingual extraction unit 22
0 uses a substitution target dictionary 224 in which information of a word or part of speech before replacement and information of a symbol after replacement are described, in order to extract a fixed bilingual pair from the morphologically analyzed sentence corresponding text.
A notation replacement unit 221 that replaces a specific phrase in the first natural language and the second natural language with a specific symbol, and a first natural language sentence whose notation is of the same type from the replaced bilingual pair and a bilingual pair consisting of the pair. A first natural language sentence isomorphic bilingual set generation unit 222 for creating a set, and a bilingual sentence composed of similar sentences in the second natural language sentence and a pair thereof from a set of bilingual pairs including the first natural language sentence having the same type It has a fixed form translation pair narrowing unit 223 that creates a subset in pairs and extracts a set of fixed form translation examples.

FIG. 8 is a block diagram of the phrase correspondence extracting unit 23 shown in FIG.
FIG. 3 is a block diagram showing a configuration of a 0. Phrase correspondence extraction unit 23
0 is a phrase correspondence target phrase filter unit 231 that limits a phrase to be investigated for phrase correspondence from the morphologically analyzed sentence correspondence text using the phrase correspondence target table 234, a phrase in the first natural language and a phrase in the second natural language. A phrase correspondence investigation unit 232 that uses the bilingual dictionary 900 describing the correspondence between words and phrases to check a part in which the phrase correspondence is obtained between each sentence of the bilingual pair, and a corresponding phrase in the designated search target phrase Is extracted as a bilingual pair that has phrase correspondence in the entire sentence,
It has a phrase corresponding pass translation pair narrowing unit 233 that creates a phrase correspondence pass translation example book. If there is no description in the phrase correspondence target table 234, the processing is performed for all the phrases in the sentence correspondence text.

Next, the operation of the example-type natural language translating apparatus and the bilingual example book creating apparatus of the present embodiment configured as described above will be described. In the example-type natural language translation device and the bilingual example book creating device of the present embodiment, the sentence-corresponding text has an error in the sentence correspondence in the sentence-corresponding text by extracting the fixed-value parallel translation pair and the appropriate translation. A translation example in which an appropriate translation pair is sent by removing a translation pair with a low translation accuracy that can take a certain translation pair or partial correspondence but cannot be said to correspond to the entire sentence. It is possible to create a collection. Further, in the example-type natural language translation device of the present embodiment, by performing the example-type natural language translation using the bilingual example collection, it is possible to imitate a bilingual pair having a sentence similar to the input sentence and perform appropriate translation. is there. Further, the quality of the translation result can be maintained without the need for manual pre-processing or complicated processing during translation.

[Specific Example] In this embodiment, the first natural language is Japanese and the second natural language is English, but the present invention is not limited to these.

First, the sentence-corresponding text input unit 100 reads a Japanese-English text. FIG. 11 shows an example of the Japanese-English text. A sentence number is attached to the head of the Japanese sentence and the English sentence, and the Japanese sentence number and the English sentence number are described in the Japanese-English sentence correspondence table so as to correspond to each other. The sentence correspondence text includes a case where the sentence correspondence is not one-to-one. For example, the line "4: 4,5" indicates that English sentences 4 and 5 correspond to Japanese sentence 4.

Next, the morphological analysis unit 210 of the bilingual pair narrowing unit 200 performs a morphological analysis on the Japanese sentence and the English sentence of the sentence-corresponding text, and acquires part-of-speech information for each word. FIG. 12 shows a morphological analysis result of the Japanese sentence of FIG. 11, and FIG. 13 shows a morphological analysis result of the English sentence of FIG. In the morphological analysis result, a delimiter “/” and a part of speech are attached after the word. For example, the notation “General / Adverb” of the first word of Japanese sentence 1 in FIG. 12 means that the part of speech of the word “General” is an adverb. Similarly, the notation “Aug./Noun” of the first word of English sentence 1 in FIG. 13 means that the part of speech of the word “Aug.” is a noun.

Next, the fixed phrase translation extraction unit 220 and the phrase correspondence extraction unit 230 use the sentence correspondence text and the morphologically analyzed sentence correspondence text to apply appropriate methods to extract an appropriate translation pair. Boilerplate bilingual extraction unit 220 and phrase correspondence extraction unit 2
If there is a method for extracting an appropriate translation pair other than the processing performed by the processing unit 30, a unit that performs the processing is added to the appropriate translation extraction unit 201, and the combination condition of each extraction unit is described in the combination condition table 250, and It is possible to process.

The fixed form bilingual extraction unit 220 creates a set of bilingual pairs having the same type of Japanese sentence from the sentence corresponding text and the morphologically analyzed corresponding text, and converts a set of similar bilingual pairs having similar English sentences from the set. Extract as a bilingual pair.

The notation replacement unit 221 shown in FIG. 5 replaces specific words and phrases with specific symbols in Japanese and English, respectively, in order to extract typical Japanese and English sentences from the morphologically analyzed sentence corresponding text. If the replacement target dictionary 224 is as shown in FIG. 6, all words in the English sentence whose part of speech is a numeral are replaced with the same symbol “<D>”. The same applies to Japanese. After the replacement, a sentence corresponding text is created using the word and the replaced notation. "225 / number" in English sentence 1 in FIG.
D> / numeral. ”Assuming that“ Aug./noun Nikkei / noun 225 / numeral call / verb options / noun mostly / adverb fell / verb and / connective many / adjective puts / noun firmed / verb Monday / noun morning / noun.
/. "/"Aug./Noun Nikkei / Noun 300 / Numerical call
/ Verb options / Noun mostly / Adverb fell / Verb and
/ Connective many / Adjective puts / Noun firmed / Verb M
onday / noun morning / noun. /. If the result of the morphological analysis of a similar sentence "is morphologically analyzed and the corresponding text, the word whose part of speech is a numeral is replaced with"<D>".
The sentence-corresponding text, which is delimited and converted to a sentence form excluding the part of speech, is a sentence of the same type "Aug. Nikkei <D> call options mos
tly fell and many puts firmed Monday morning. ". As shown in FIG. 6, the designation of the replacement target phrase is not limited to the part of speech, but may be a specific phrase.

Next, the first natural language sentence isomorphic bilingual set creation unit 222 examines the same type of Japanese sentence from the replaced sentence corresponding text. Since the first, third, and fifth sentences in FIG. 11 are all “Trading is low in general.”, A set is created using the translation pairs having the Japanese sentences 1, 3, and 5 shown in FIG.

Next, the fixed-form bilingual pair narrowing unit 223
From the set created by the first natural language sentence isomorphic bilingual set creation unit 222, a subset is created with a bilingual pair having a more general English sentence. A more general way to look up English sentences is
Here, it is based on words that appear commonly in many English sentences. A more general method for examining English sentences is to exclude sentences in which the number of words in the set is extremely different, and if the refinement is insufficient, repeat the refined set in the same way as necessary Alternatively, another method may be used or a combination may be used. Common words (case insensitive) in the English sentence shown in FIG. 14 are shown in bold Gothic

[Outside 1] Therefore, the English sentences 3 and 6 are similar sentences. Using this result and the sentence-corresponding text, as shown in FIG. 7, bilingual pairs of “3: 3” and “5: 6” are extracted as a set of fixed bilingual translation examples. A bilingual pair "1: 1", an unconventional bilingual pair "2: 2",
“4: 4,5” and “6: 7” are removed as unnecessary translation pairs.

The phrase correspondence extraction unit 230 extracts a bilingual pair having overall phrase correspondence between the sentences of the bilingual pair from the sentence corresponding text and the morphologically analyzed sentence corresponding text.

The target phrase filtering unit 2 shown in FIG.
At 31, in the case where the phrase correspondence target table 234 is as shown in FIG. 9, the phrase whose part of speech is a noun in both Japanese and English is limited as a target for phrase correspondence between Japanese and English. If you want to investigate verbs in addition to nouns in both Japanese and English, write "part of speech: verb" in both Japanese and English. If you write "phrase: am" in Japanese and "phrase: am" in English,
You can also search for specific words. A description will be given using FIG. 15 as an example of the phrase correspondence of the bilingual pair in which the correspondence between the Japanese sentence and the English sentence in FIG. 11 is “4: 4,5”. A bold Gothic character is a noun whose word is a noun, and a bold Gothic character surrounded by a square can be associated with the bilingual dictionary of FIG. The Japanese sentence has six nouns, of which five have word correspondence with English, so the word correspondence rate is 5/6／0.
83. In the English sentence, the first sentence has 3 nouns, of which 3 has a phrase correspondence with Japanese, the second sentence has 6 nouns,
There are five nouns with correspondence, and when two are combined, the phrase correspondence rate is (3 + 5) / (3 + 6) ≒ 0.88. Assuming that the threshold value of the Japanese-English phrase correspondence rate is both 0.6, the bilingual pair is extracted using the result and the sentence-corresponding text since the threshold value is equal to or greater than the threshold value in both English and Japanese. In addition, another method can be used for setting the threshold.

FIG. 17 shows an example of a result of a survey corresponding to words and phrases with respect to the translation pair shown in FIG. Since the translation pairs in FIG. 17 having the phrase correspondence ratio of not less than the threshold value 0.6 in both Japanese and English are “4: 4, 5” and “6: 7”, as shown in FIG. A bilingual pair "1: 1" which is extracted as a collection of examples and has an incorrect sentence correspondence, a bilingual pair "2: 2" in which unnecessary phrases are included in an English sentence and the phrase correspondence is small,
Excludes translation pairs "3: 3" and "5: 6" where the phrase correspondence of "low tone / noun" is "muted / verb" and "weak / adjective" and the part of speech of English phrase is not the target of phrase correspondence. Is done.

The bilingual example set creating unit 240 extracts appropriate bilingual pairs by combining bilingual pairs extracted by different methods. The combination method is in the combination condition table 250 and is selected by the combination condition identifier. For example, when the combination condition identifier is set to 1 and the method of the identification symbol 1 corresponding to the combination condition identifier is selected from FIG. 18, either one of a standard translation pair and an appropriate translation pair corresponding to a phrase is included. The translated translation pair and the translation translation example translation pair. This prevents the exclusion of a bilingual pair that is not a standard but has appropriate sentence correspondence when only a standard bilingual pair is extracted. Conversely, if only appropriate translation pairs corresponding to words / phrases are extracted, it is not possible to say that the word / phrase correspondences are not appropriate, but it is possible to prevent a standard translation pair from being excluded. If there is any other suitable bilingual pair extraction method, the extraction method can be added to the appropriate bilingual extraction unit 201 as appropriate, and the result can be added to the combination condition table 250 by adding the combination conditions of the other extraction methods. When the bilingual pair extracted according to the notation of the Japanese-English correspondence table of FIG. 11 is shown, the standard bilingual pair is “3: 3”,
"5: 6", and the translated pair with the corresponding phrase is "4: 4".
5 "and" 6: 7 ", the bilingual example collection of FIG.
“3: 3”, “4: 4,5”, “5: 6”, “6: 7”
4 translations are included.

Next, the translation process will be described. In this description, FIG. 19 shows a case where the bilingual example book 300 created as described above is not used and a case where it is used (a case where translation is performed using a fixed translation pair). It will be described using FIG. For bold Gothic characters, the difference between the input sentence and the similar sentence, the portion of the English sentence corresponding to the difference between the similar sentences, and the portion of the English sentence corresponding to the difference between the similar sentences are replaced with the English translation of the difference between the input sentences Represents a part.

As described above, the prepared bilingual translation example book 3
When the input sentence “Generally the price is low.” Is read from the input sentence input unit 400 and the sentence corresponding text is directly used as the bilingual example collection without using 00, the similar sentence search unit 500 inputs from the bilingual example collection. A Japanese sentence similar to the sentence is searched. Although there are various similar sentence search methods, a sentence having a large number of matching characters is regarded as a similar sentence. However, a similar sentence search method other than this method can be used. Similar sentence search unit 5
The search result of "00" is Japanese sentences 1, 3, and 5 "Trading is generally low." Thereafter, if a plurality of searches are made, the one with the smallest sentence number is selected. Here, a bilingual pair of Japanese sentence 1 is selected. However, it is not limited to this method.

In the difference correspondence unit 600, the “quote”
It can be seen that the similar sentence "Trading" is a difference portion, but the English sentence portion corresponding to the difference portion "Trading" of the similar sentence cannot be found from the bilingual dictionary and the bilingual pair "1: 1" in FIG. Thereafter, when there is no word correspondence of the difference portion between the parallel translation pairs, if the replacement of the difference portion is not performed, the sentence processing section 700 does not perform any processing, and the translation output section 800 outputs “Aug. Nikkei”.
225 call options mostly fell and many puts firmed
Monday morning. "Is output, but this output sentence is inappropriate as a translation of the input sentence.

Therefore, the input sentence input unit 400 uses the bilingual example book 300 created by the present invention as described above to
When the same input sentence “Generally the market is low” is read, the similar sentence search unit 500 searches for Japanese sentences 3 and 5 “General trade is low.” Here, it is assumed that a translation pair having Japanese sentence 3 with the smallest sentence number is used for translation. In the difference correspondence unit 600, “quote” of the input sentence and “business” of the similar sentence are difference portions, and the English sentence corresponding to the difference portion “business” of the similar sentence is the bilingual dictionary and the bilingual pair “3” in FIG. : 3 ”to“ Trading ”. Further, the English corresponding to the difference part “quote” of the input sentence is “market” from the bilingual dictionary of FIG. Therefore, the sentence processing section 700 replaces the input sentence with the difference part of the bilingual pair “3: 3”, and the translated sentence is “Market was generally muted.” Finally, the translation is prepared using the analysis information, and “The mark
et was generally muted. "is output from the translation output unit 800. This output sentence is appropriate for the input sentence.

As another example of the translation processing, FIG. 20 is used for a case where the bilingual translation example book 300 created according to the present invention is not used and a case where it is used (a case where translation is performed using an appropriate bilingual translation pair). Will be explained. For bold Gothic characters, the difference between the input sentence and the similar sentence, the portion of the English sentence corresponding to the difference between the similar sentences, and the portion of the English sentence corresponding to the difference between the similar sentences are replaced with the English translation of the difference between the input sentences Represents a part.

The bilingual example book 30 created by the present invention
If the sentence-corresponding text is directly used as a bilingual example collection without using 0, and the input sentence “Trading is generally strong in pattern-viewing mood” is read from the input sentence input unit 400,
The search result of the similar sentence search unit 500 is Japanese sentence 2 “Trading is generally strong in pattern-view mood.” Since the input sentence and the similar sentence match, the difference correspondence unit 600 and the sentence processing unit 700
Does nothing, and the translation output unit 800 outputs “The tarding
was cautious ahead of the release of USjobs dat
a. ”is output, but this output sentence has a word correspondence only in the“ trading ”part of the input sentence and“ trading ”of the translated sentence, and an inappropriate translation is output for the input sentence.

Therefore, the bilingual translation example book 30 created by the present invention is used.
When the same input sentence “Trading is generally strong in pattern view mood.” Is read from the input sentence input unit 400 using 0, the search result of the similar sentence search unit 500 is Japanese sentence 6 “The market price is generally The pattern view mood is strong. " Next, the difference correspondence unit 600 finds that the “quote” of the input sentence and the “quote” of the similar sentence are difference portions, and the English sentence portion corresponding to the difference portion “quote” of the similar sentence is the Japanese / English of FIG. The bilingual dictionary and the bilingual pair “6: 7” are “market”. In addition, the English corresponding to the difference part “business” of the input sentence is “trading” from the Japanese-English bilingual dictionary of FIG. Therefore, the sentence processing unit 700 replaces the difference part of the bilingual pair “6: 7” with the input sentence to create a translated sentence “The trading was a wait-and-see moo.
d. ”Finally, prepare the translation using the analysis information,
"Trading was a wait-and-see mood."
Output from 00. This output sentence is a translation that is compatible with words and phrases with respect to the input sentence.

As another example of the translation processing, here, the bilingual translation example book 300 created according to the present invention is composed of only a fixed bilingual pair, and is either a fixed bilingual pair or an appropriate bilingual pair corresponding to a phrase. A case of a bilingual pair including one or the other will be described with reference to FIG. For bold Gothic characters, the difference between the input sentence and the similar sentence, the portion of the English sentence corresponding to the difference between the similar sentences, and the portion of the English sentence corresponding to the difference between the similar sentences are replaced with the English translation of the difference between the input sentences Represents a part.

Assuming that the bilingual translation example collection 300 is only a fixed bilingual translation pair (parallel translation pairs “3: 3”, “5: 6”), the input sentences “NTT, KDD” are lowered from the input sentence input unit 400 to NT
T Docomo has hit a new low for the year. Is read, nothing is obtained as a search result of the similar sentence search unit 500, so that the difference handling unit 600 and the sentence processing unit 700 are not performed.
No translated sentence can be obtained from translated sentence output section 800. However, among the bilingual pairs excluded in the standard bilingual pair extraction, there is a bilingual sentence “4: 4, 5” whose sentence correspondence is appropriate and which is similar to the input sentence.

Therefore, it is assumed that the bilingual translation example collection 300 is a fixed translation bilingual pair and a bilingual translation pair having correspondence between words and phrases. From the input sentence input unit 400, the same input sentence “NTT, KDD, DDI is lowered, and NTT DoCoMo is Is read, the search result of the similar sentence search unit 500 is Japanese sentence 4 "PH Morris, IBM, and Apple dropped, and TI updated the year-to-date low." Next, the difference correspondence unit 60
0, the input sentence "NTT, KDD, DDI" and the similar sentence "PH Morris, IBM, Apple" are the difference parts,
Also, the input sentence “NTT DOCOMO” and the similar sentence “TI” are the difference parts. Japanese-English bilingual dictionary and bilingual pair "4:
4, 5 ", the difference part of the similar sentence" PH Morris, IB
The English part of "M, Apple" is "Philip Morris,
"IBM and Apple Computer", and the portion of the English sentence corresponding to "TI" is "Texas Instruments".
According to the Japanese-English bilingual dictionary in FIG. 16, the English part corresponding to the difference part “NTT, KDD, DDI” of the input sentence is “NTT,
"KDD and DDI" and the English language corresponding to "NTT Docomo" is "NTT Docomo". Therefore, the sentence processing unit 7
00, the translated sentence created by replacing the difference part of the translated sentence pair “4: 4,5” with the input sentence is “NTT Docomo hit ay”.
ear's low.Among other decliners were NTT, KDD and D
DI. "Is output from the translated sentence output unit 800. This output sentence is an appropriate translation result for the input sentence.

By recording the processing of the above-described embodiment as a program on a recording medium, it is possible to use the recording medium to enhance its circulation.

[0090]

As described above, according to the present invention,
It is possible to create a bilingual translation example in which a sentence in the second natural language is an appropriate translation of the sentence in the first natural language from a sentence-corresponding text containing a translated pair of a sentence in the first natural language and a sentence in the second natural language. Also, a similar sentence to the input sentence of the first natural language is searched from the bilingual example collection, and a difference between the input sentence and the similar sentence is extracted.
Since the translated sentence of the input sentence is generated by mechanically processing the sentence of the second natural language corresponding to the similar sentence using the difference information, the accuracy of sentence correspondence including a wrong translated sentence pair is low. It is possible to create a bilingual example collection narrowed down to an appropriate bilingual pair even from a sentence-corresponding text, and to use this bilingual example collection in an example-type natural language translation, it is possible to obtain an appropriate translation result.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating the operation principle of an example natural language translation apparatus that implements an example natural language translation method and a bilingual example book creation method according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of an example-type natural language translating apparatus and a bilingual example book creating apparatus of the embodiment shown in FIG. 1;

FIG. 3 is a block diagram showing a configuration of a bilingual-language narrowing-down unit used in the example-type natural language translating apparatus and the bilingual example book creating apparatus shown in FIG. 2;

FIG. 4 is a diagram showing an example of a bilingual translation example book created by the bilingual translation narrowing unit of FIG. 3;

FIG. 5 is a block diagram illustrating a configuration of a fixed form bilingual extraction unit illustrated in FIG. 3;

FIG. 6 is a diagram showing an example of a replacement target dictionary used in the fixed phrase translation extraction unit shown in FIG. 5;

FIG. 7 is a diagram showing an example of a set of fixed bilingual examples extracted by the fixed bilingual extraction unit shown in FIG. 5;

FIG. 8 is a block diagram illustrating a configuration of a phrase correspondence extraction unit illustrated in FIG. 3;

9 is a diagram illustrating an example of a phrase correspondence target table used in the phrase correspondence extraction unit illustrated in FIG. 8;

FIG. 10 is a diagram showing an example of a phrase-matched successful bilingual translation example extracted by the phrase-matching extraction unit shown in FIG. 8;

FIG. 11 is a diagram showing an example of a Japanese-English text.

FIG. 12 is a diagram illustrating an example of morphological analysis of the Japanese sentence illustrated in FIG. 11;

FIG. 13 is a diagram illustrating an example of morphological analysis of the English sentence illustrated in FIG. 11;

FIG. 14 is a diagram showing an example of extraction of a fixed bilingual pair.

FIG. 15 is a diagram showing an example of correspondence between Japanese and English phrases (nouns).

FIG. 16 is a diagram showing an example of a Japanese-English bilingual dictionary.

FIG. 17 is a diagram showing an example of a survey result corresponding to a phrase.

FIG. 18 is a diagram showing an example of a combination condition table.

FIG. 19 is a diagram illustrating a translation example showing the effect of narrowing down appropriate translation pairs.

FIG. 20 is a diagram showing another example of translation showing the effect of narrowing down appropriate translation pairs.

FIG. 21 is a diagram showing still another example of translation showing the effect of narrowing down appropriate translation pairs.

[Explanation of symbols]

 100 text-corresponding text input sentence 200 parallel-translation pair narrowing-down unit 201 appropriate parallel-translation pair extraction unit 210 morphological analysis unit 220 fixed-phrase translation extraction unit 230 word-correspondence extraction unit 240 parallel translation example collection creation unit 250 combination condition table 300 parallel translation example collection 400 input sentence input unit 500 similar sentence search unit 600 difference correspondence unit 700 sentence processing unit 800 translated sentence output unit 900 bilingual dictionary

Claims (26)

[Claims] 1. A sentence corresponding to a sentence corresponding to a translation pair which is a pair of a sentence of a first natural language and a sentence of a second natural language, the sentence of the second natural language is determined as an appropriate translation of the sentence of the first natural language. Creating a bilingual example set, which is a set of bilingual pairs, searching for a similar sentence to the input sentence of the first natural language to be translated from the bilingual example set, extracting a difference between the input sentence and the similar sentence, An example-type natural language translation method, characterized by mechanically processing a sentence of a second natural language corresponding to the similar sentence using the difference information to generate a translated sentence of the input sentence. 2. The bilingual translation example collection is created by extracting a bilingual pair in which a sentence of a second natural language is an appropriate translation of a sentence of a first natural language by a plurality of different methods. Item 4. The example-type natural language translation method according to Item 1. 3. One of the plurality of different methods includes:
3. The example-type natural language translation method according to claim 2, wherein a bilingual pair that is frequently similar to the sentence-corresponding text is extracted as a fixed bilingual pair. 4. One of the plurality of different methods includes:
3. The example-type natural language according to claim 2, wherein a correspondence between words and phrases between a sentence of the first natural language and a sentence of the second natural language is checked, and a bilingual pair having a word correspondence in the entire sentence is extracted. Translation method. 5. The text according to claim 1, wherein the text corresponding to the first natural language and the text of the second natural language to which the text does not correspond have a sentence matched manually and manually. Item 4. The example-type natural language translation method according to Item 1. 6. A sentence corresponding to a sentence of a first natural language sentence and a second natural language sentence, and a second natural language sentence is converted to an appropriate translated sentence of the first natural language sentence. A bilingual example collection creating method, wherein a bilingual example set is created by narrowing down to a bilingual pair. 7. The bilingual example set is created by combining a pair of bilingual sentences extracted by a plurality of different methods, the sentences of the second natural language being appropriate translations of the sentences of the first natural language. 7. The method for creating a bilingual example book according to claim 6. 8. One of the plurality of different methods includes:
8. The bilingual translation example collection creating method according to claim 7, wherein a bilingual pair that is frequently similar to the text corresponding to the sentence is extracted as a fixed bilingual pair. 9. One of the plurality of different methods includes:
8. The bilingual example book creation according to claim 7, wherein a correspondence between words and phrases between a sentence of the first natural language and a sentence of the second natural language is checked, and a bilingual pair having a word / phrase correspondence in the entire sentence is extracted. Method. 10. The text according to claim 1, wherein the text corresponding to the first natural language and the text of the second natural language to which no text corresponding to the text is attached are manually and mechanically paired by a machine. A method for creating a bilingual example book according to item 6. 11. A sentence-corresponding text input means for reading a sentence-corresponding text containing a pair of a sentence of a first natural language and a sentence of a second natural language; A bilingual pair narrowing means for narrowing down to an appropriate translation of a language sentence to create a bilingual example book, an input sentence inputting means for inputting a first natural language sentence to be translated, and a similar sentence for the input sentence A similar sentence search means for searching for the input sentence and the similar sentence from the bilingual example collection,
Using a bilingual dictionary in which the correspondence between natural language terms is described, a phrase correspondence between the difference part of the similar sentence and the second natural language sentence corresponding to the difference part of the input sentence is obtained. A sentence processing means for mechanically processing a sentence of the second natural language corresponding to the similar sentence using difference information; and a translated sentence output means for generating and outputting a translated sentence of the input sentence. An example-type natural language translator characterized by having. 12. The bilingual-parallel pair narrowing-down means includes a proper bilingual-pair extracting means for extracting a plurality of different proper bilingual pairs from the sentence-corresponding text, and a combination condition table combining the bilingual pairs extracted by the proper bilingual-pair extracting means. 13. The example-type natural language translation apparatus according to claim 11, further comprising: a bilingual example collection creating unit configured to create a bilingual example set by combining with the combination condition identifier. 13. The suitable bilingual pair extraction means, comprising: a morphological analysis means for performing morphological analysis to obtain linguistic information of a phrase from the sentence corresponding text; and a fixed bilingual pair from the morphologically analyzed sentence corresponding text. 13. The example-type natural language according to claim 12, comprising: a fixed-form bilingual extraction means for extracting; and a phrase correspondence extracting means for extracting a bilingual pair in which the whole sentence has a phrase correspondence from the morphologically analyzed sentence-corresponding text. Language translator. 14. A sentence-corresponding text input means for reading a sentence-corresponding text containing a pair of a sentence of a first natural language and a sentence of a second natural language; A bilingual translation example narrowing-down device for narrowing down a translation to a pair that becomes an appropriate translation of a language sentence to generate a bilingual translation example collection. 15. The bilingual-parallel pair narrowing-down means includes a proper bilingual-pair extracting means for extracting a plurality of different suitable bilingual pairs from the sentence-corresponding text, and a combination condition table combining the bilingual pairs extracted by the proper bilingual-pair extracting means. 15. The bilingual example book creating apparatus according to claim 14, further comprising: a bilingual example book creating unit that creates a bilingual example book by combining the bilingual example set with the combination condition identifier. 16. The suitable bilingual pair extracting means includes a morphological analysis means for performing morphological analysis to obtain language information of a phrase from the sentence corresponding text, and a fixed bilingual pair from the morphologically analyzed sentence corresponding text. 16. The bilingual example collection according to claim 15, further comprising: a fixed form bilingual extraction means for extracting, and a phrase correspondence extracting means for extracting a bilingual pair in which the whole sentence has a phrase correspondence from the morphologically analyzed sentence corresponding text. Creating device. 17. A sentence corresponding to a sentence in which a translation pair, which is a pair of a sentence of a first natural language and a sentence of a second natural language, is inserted into an appropriate translation of the sentence of the first natural language. A bilingual example set, which is a set of bilingual pairs, is searched from the bilingual example set for a similar sentence to the input sentence of the first natural language to be translated, and a difference between the input sentence and the similar sentence is extracted; A recording medium storing an example-type natural language translation program, wherein a sentence of a second natural language corresponding to the similar sentence is mechanically processed using the difference information to generate a translated sentence of the input sentence. 18. The bilingual example set is created by extracting a bilingual pair in which a sentence of a second natural language is an appropriate translation of a sentence of a first natural language by a plurality of different methods. Item 18. A recording medium on which the example-type natural language translation program according to Item 17 is recorded. 19. A method according to claim 1, wherein, as one of the plurality of different methods, a bilingual pair that is frequently similar to the sentence-corresponding text is extracted as a fixed bilingual pair.
8. A recording medium on which the example-type natural language translation program described in 8 is recorded. 20. As one of the plurality of different methods, a correspondence between words and phrases between a sentence in a first natural language and a sentence in a second natural language is checked, and a translation pair having a word / phrase correspondence in the entire sentence is extracted. 19. A recording medium on which the example-type natural language translation program according to claim 18 is recorded. 21. The sentence-corresponding text is composed of a pair in which a sentence is associated with a text of a first natural language and a text of a second natural language that have no sentence correspondence by hand and machine. Item 18. A recording medium on which the example-type natural language translation program according to Item 17 is recorded. 22. A sentence corresponding to a sentence corresponding to a first natural language sentence and a second natural language sentence, and a second natural language sentence is converted to an appropriate translated sentence of the first natural language sentence. A recording medium storing a bilingual example book creation program, wherein a bilingual example set is created by narrowing down to a bilingual pair. 23. The bilingual example set is created by combining a pair of bilingual sentences extracted by a plurality of different methods so that the second natural language sentence is an appropriate translation of the first natural language sentence. 23. A recording medium storing the bilingual example book creation program according to claim 22. 24. As one of the plurality of different methods, a bilingual pair that is frequently similar to a sentence-corresponding text is extracted as a fixed bilingual pair.
3. A recording medium on which the bilingual example book creation program described in 3 is recorded. 25. As one of the plurality of different methods, a correspondence between words and phrases between a sentence in a first natural language and a sentence in a second natural language is checked, and a translation pair having a word / phrase correspondence in the entire sentence is extracted. 24. A recording medium on which the bilingual example book creation program according to claim 23 is recorded. 26. The sentence-corresponding text is composed of a pair of a sentence of a first natural language text and a text of a second natural language, to which no sentence correspondence is attached, which is manually and manually matched by a machine. A recording medium recording the bilingual example book creation program according to item 22.