JP4971844B2 - Example database creation device, example database creation program, translation device, and translation program - Google Patents

Description

  The present invention relates to an example database creation device and an example database creation program for creating an example database used for translation, and a translation device and a translation program that perform translation using the example database.

  Conventionally, a translation system that translates a first language (for example, Japanese) into a second language (for example, English) using examples (examples such as words, examples of actual use, examples of usage) An input sentence in one language and the most similar example are selected in sentence units and the second language of the parallel translation is edited (see, for example, Non-Patent Document 1), or the input sentence in the first language completely matches. There is one that generates a second language sentence by combining partial language words in a partial example (see, for example, Patent Document 1).

In addition, some translation systems using examples split an input sentence in a first language before translation (see, for example, Non-Patent Document 2), and the second language is a translation source language. There is also one that divides and translates an input sentence in one language (for example, see Patent Documents 2 and 3).
JP 2006-2525290 A JP 2004-110583 A Japanese Patent Laid-Open No. 10-312382 Sumita, Example-based machine translation using DP-matching betaword word sequences, 39th ACL workshop on DDMT, pp. 1-8, 2001. Kim et al., Japanese automatic short sentence segmentation and subject completion for Japanese-English machine translation, IPSJ Journal, Vol. 35, no. 6, pp. 1018-1028, 1994.

  However, in the conventional translation system, when the first language is a long sentence such as a compound sentence, if an attempt is made to use a sentence unit example of the second language, the long sentence of the first language includes a lot of contents. As a result, it is unlikely that all examples with similar contents will be included in the example database used in the translation system. If all examples with similar contents are not included in the example database, It can happen that the accuracy is reduced.

As a result, in the translation system disclosed in Non-Patent Document 1, when the input sentence is a long sentence, there is a problem that variations in the sentence that can be translated are reduced and the sentence cannot be translated into a natural sentence.
For example, when the results of news translation from Japanese to English conducted by a translation professional were investigated, it was found that Japanese news sentences were often translated into multiple English news sentences. In other words, the unit of a natural English news sentence was found to be shorter than the unit of a Japanese news sentence. For this reason, in order to translate long Japanese news into natural English news, it is more appropriate to translate one Japanese sentence into one or more English sentences.

  Further, in the translation system disclosed in Patent Document 1, when an input sentence is a long sentence in Japanese, a sentence is generated by combining a part of expressions in an English sentence. Cannot translate into multiple sentences. Moreover, since this translation system generates the structure of an English sentence itself, there is a possibility that an unnatural English sentence will be generated. The sentence structure is represented by, for example, SVOC. Here, English sentences are created by connecting fragments such as nouns, verbs, and adjectives. A structure will also be generated at this time, but it may not be grammatically correct or natural English. As a result, there is a problem that even the translation system disclosed in Patent Document 1 cannot be translated into a natural sentence.

  Further, in the translation system disclosed in Non-Patent Document 2, when the input sentence is a long Japanese sentence, the Japanese input sentence is divided without considering the unit of the English sentence that is the translation result. Then, it is not always possible to divide the sentence into appropriate English units, and there is a problem that it cannot be translated into a natural sentence unless a Japanese input sentence is divided into appropriate units.

  Furthermore, in the translation system disclosed in Patent Document 2, when the input sentence is a long sentence in Japanese, English is used as the translation source language, and the example is not divided. After all, there is a problem that it cannot be translated into a natural sentence. Further, in the translation system disclosed in Patent Document 3, when an input sentence is a long sentence in Japanese, since a rule that depends on English is used for the division method, a long sentence in Japanese can be divided. After all, there is a problem that it cannot be translated into a natural sentence.

  Therefore, in the present invention, when the above-mentioned problem is solved and one sentence in the first language (for example, Japanese) is translated into a plurality of sentences in the second language (for example, English), An example database creation device and an example database creation program capable of creating an example database in which a part in one sentence and one sentence in English correspond to each other, and a sentence in the first language as a natural language in the second language It is an object of the present invention to provide a translation apparatus and a translation program that can translate a sentence.

  In order to solve the above problem, the example database creation device according to claim 1 is a bilingual database in which translation data obtained by translating a sentence in a first language into a sentence in a second language is stored. One sentence corresponding translation data in which one sentence part of the first language corresponds to one sentence of the second language when one sentence is translated into a plurality of sentences of the second language Is an example database creation device for creating an example database added to the bilingual database, wherein the translation data discrimination means, the quoted expression discrimination means, the quoted expression separation means, the second language multiple sentence discrimination means, and the first A language expression specifying unit, a first language dividing unit, and a divided first language-compatible second language adding unit are provided.

  According to such a configuration, the example database creation device patterns the citation expression frequently appearing in the first language by the citation expression discriminating means to determine whether or not the citation expression representing the citation is included in the sentence of the first language. Discrimination is performed using a quoted expression pattern. Then, the example database creation device separates the quoted expression from the sentence in the first language and discards it when the quoted expression separating unit determines that the quoted expression is included by the quoted expression separating unit. Then, the example database creation device uses the second language multi-sentence discriminating means to make the text in the first language in which the quoted expression separating means is separated into the text, and the text is translated into the plurality of sentences in the second language. Is determined using a sentence break pattern obtained by patterning sentence breaks according to the grammar of the second language.

  Then, the example database creation device, when the first language expression specifying unit determines that the text obtained by separating the cited expression by the second language multiple sentence determination unit is translated into a plurality of sentences in the second language. The bilingual dictionary data in which the words of the first language and the words of the second language are associated with which word of the words contained in the text the words included in the plurality of sentences of the second language correspond to Use to specify.

  In the first language expression specifying means, instead of the bilingual dictionary data, the log likelihood ratio data obtained for the relationship between the first language word and the second language word, the first language word, and the second language You may use at least 1 with the word corresponding | compatible probability calculated | required about the probability corresponding to the word of a language. Further, the first language expression specifying means may use the syntax structure data of the first language sentence, the second language, and the syntax structure data.

  Further, the example database creation device is configured so that the first language dividing unit determines the text according to the correspondence between the words included in the plurality of sentences in the second language specified by the first language expression specifying unit and the words included in the text. Are divided so that one part of the second language corresponds to one sentence of the second language. Then, the example database creation device uses the divided first language-corresponding second language adding means, so that each part of the text divided by the first language dividing means corresponds to one sentence of the second language. Add data to the bilingual database.

  The example database creation device according to claim 2 is the example database creation device according to claim 1, wherein the first language segmentation unit preliminarily becomes a subject or a suggestion of each part of the text. The set word is added.

  According to such a configuration, the example database creation device can add the subject or the proposal to each part of the text by the first language dividing means, and can establish each part of the text as one sentence.

  The translation device according to claim 3 is a translation device for translating an input sentence in the first language into a sentence in the second language, an example database, a quoted expression discriminating means, a quoted expression separating means, The first language clause / parallel phrase discrimination means, the unit translation means, the maximum score translation result selection means, the translation means, and the translation result output means are provided.

  According to this configuration, the translation apparatus includes the example database created by the database creation apparatus according to claim 1, and whether the quotation expression representing the quotation is included in the sentence in the first language by the quotation expression determination unit. Whether or not is determined using a citation expression pattern obtained by patterning citation expressions frequently appearing in the first language. Subsequently, the translation device separates the quoted expression from one sentence of the corresponding first language when the quoted expression separating unit determines that the quoted expression is included by the quoted expression separating unit. Then, the translation device uses the first language clause / parallel phrase discriminating means to determine whether or not the sentence of the first language separated by the quote expression separating means includes a clause or a parallel phrase in the grammar of the first language. Discrimination is performed using clause / parallel phrase patterns obtained by patterning the clauses and parallel phrases of the sentence.

  The translation device includes a unit for translating the sentence of the first language, which is determined by the first language translation unit dividing means to include the clause or the parallel phrase by the first language clause / parallel phrase determining means, into the second language. Divide into translation units. Then, the translation device calculates, for each translation unit, a score indicating the degree of matching between the translation unit divided by the first language translation unit dividing unit and the data included in the example database by the unit translation unit. translate. Furthermore, the translation apparatus selects the maximum score translation result that maximizes the total score of the scores for each translation unit calculated when the translation unit is translated by the unit translation means by the maximum score translation result selection means. . Then, the translation apparatus uses the example database to translate the sentence in the first language, which is determined by the translation unit to include no clause or parallel phrase by the first language clause / parallel phrase determination unit. In the translation device, the translation result output means has separated the maximum score translation result selected by the maximum score translation result selection means or the translation result translated by the translation means and the quote expression by the quote expression separation means. In this case, a translation result obtained by translating the cited expression using bilingual dictionary data in which a word in the first language and a word in the second language are associated is output.

  The translation device according to claim 4 is the translation device according to claim 3, wherein the first language translation unit dividing unit assigns, to the translation unit, a preset word that is a subject or a proposal of the translation unit. It is characterized by adding.

  According to such a configuration, the translation device is included in the example database by adding a subject or a proposal to the translation unit by the first language translation unit dividing unit and establishing the translation unit as one sentence. Can correspond to a second language sentence.

  The translation device according to claim 5 is the translation device according to claim 2 or 3, wherein the unit-by-unit translation means includes example data acquisition means, example data selection means, editing means, and translation candidate output means. It is characterized by providing.

  According to such a configuration, the translation apparatus acquires, as example data, a sentence in the first language that satisfies the predicate included in the translation unit or satisfies the similarity set in advance from the example database by the example data acquisition unit. Subsequently, the translation apparatus calculates a distance representing the degree of approximation of the syntax structure between the example data acquired by the example data acquisition unit and the translation unit by the example data selection unit, and a predetermined number from the smallest distance is used. Select example data for.

  Note that this distance is calculated when the example data and the translation unit are the same, and the minimum edit distance is assumed, and when the example data is edited into the translation unit, the word is deleted at this edit distance. Is added as the deletion cost, the cost when the word is replaced is added as the replacement cost, and the cost when the word is inserted is added as the insertion cost.

In addition, when the translation device edits the example data so that the expression of the example data selected by the example data selection unit and the translation unit is the same by the editing unit, the translation device performs editing in accordance with a preset editing rule. While calculating the cost, the second language expression of the example data was edited using the first language second language correspondence information in which the correspondence between the first language word and the second language word was preset. The later second language sentence is taken as a translation candidate.
This editing cost is calculated when a word included in a second language sentence corresponding to the example data is replaced, deleted, or inserted according to the editing of the example data.

  Then, the translation apparatus outputs a score and a translation candidate for each translation unit from the distance calculated by the example data selection unit and the editing cost calculated by the editing unit by the translation candidate output unit.

  The example database creation program according to claim 6 is a parallel translation database storing translation data obtained by translating a sentence in a first language into a sentence in a second language, wherein one sentence in the first language is the second sentence. When translated into a plurality of sentences in a language, one sentence corresponding translation data in which one sentence part of the first language corresponds to one sentence of the second language is added to the parallel translation database. In order to create the example database, the computer is divided into a translation data discriminating means, a quoted expression discriminating means, a quoted expression separating means, a second language plural sentence discriminating means, a first language expression specifying means, a first language dividing means, The second language adding means corresponding to one language is configured to function.

  According to such a configuration, the example database creation program patterns the citation expression frequently appearing in the first language by the citation expression discriminating means to determine whether or not the citation expression representing the citation is included in the sentence of the first language. Discrimination is performed using a quoted expression pattern. Then, the example database creation program, when the citation representation separation means determines that the citation expression is included by the citation expression determination means, separates the citation expression from the sentence in the first language, and discards it. The second language multi-sentence discriminating means determines whether or not the text separated from the cited expression by the cited expression separating means has been translated into a plurality of sentences in the second language, and patterns of sentence breaks according to the grammar of the second language If the first language expression specifying unit determines that the text is translated into a plurality of sentences in the second language by the first language expression specifying unit, Using bilingual dictionary data in which words in the first language and words in the second language are associated with each other, the words included in the sentences in two languages correspond to which of the words included in the text. Identify. Further, the example database creation program is included in a plurality of translated sentence inclusion sentences in which a word included in a plurality of sentences in the second language specified by the first language expression specifying means and a quoted expression are separated by the first language dividing means. In accordance with the corresponding relationship with each word, each portion of the sentence including the multiple translation sentences separated from the quoted expression and one sentence in the second language correspond to each other. One sentence corresponding translation data in which each part of the text divided by the first language dividing means corresponds to one sentence of the second language is added to the parallel translation database.

  The translation program according to claim 7, a computer having an example database created by the database creation device according to claim 1, for translating an input sentence in the first language into a sentence in the second language, Citation expression discrimination means, citation expression separation means, first language clause / parallel phrase discrimination means, unit-by-unit translation means, maximum score translation result selection means, translation means, and translation result output means The configuration.

  According to such a configuration, the translation program patterns the citation expression frequently appearing in the first language by the citation expression discriminating means to determine whether or not the citation expression representing the citation is included in the sentence of the first language. When the quotation expression pattern is determined and the quotation expression separation means determines that the quotation expression is included, the quotation expression is separated from one sentence of the corresponding first language. The sentence section according to the grammar of the first language is used to determine whether the sentence in the first language from which the quote expression is separated by the quote expression separation means by the first language clause / parallel phrase discrimination means includes a clause or a parallel phrase. And it is discriminated by using a clause / parallel phrase pattern obtained by patterning the parallel phrase. The translation program includes a unit for translating the sentence in the first language, which is determined by the first language translation unit dividing means to include the clause or the parallel phrase in the first language clause / parallel phrase determining means, into the second language. The translation unit for each translation unit calculates a score indicating the degree of matching between the translation unit divided by the first language translation unit division unit and the data included in the example database. translate. Furthermore, the translation program selects the maximum score translation result that maximizes the total sum of the scores for each translation unit calculated when the translation unit is translated by the unit translation means by the maximum score translation result selection means. The sentence in the first language determined not to contain the clause or the parallel phrase by the first language clause / parallel phrase determining means by the translation means is translated using the example database. Then, the translation program outputs the maximum score translation result selected by the maximum score translation result selection means or the translation result translated by the translation means and the quote expression by the quote expression separation means by the translation result output means. In this case, a translation result obtained by translating the cited expression using bilingual dictionary data in which a word in the first language and a word in the second language are associated is output.

  According to the first and sixth aspects of the invention, after removing the citation expression in the first language, one sentence in the first language (for example, Japanese) can be obtained by associating it with the sentence in the second language. When translated into a plurality of sentences in a language (for example, English), it is possible to create an example database in which a part of one sentence in the first language corresponds to one sentence in the second language .

  According to the second aspect of the present invention, a subject or a suggestion is added to each part of the divided text, and each part of the sentence including multiple translations is established as one sentence, so that it is included in the example database. It can correspond to a sentence in the second language.

  According to the third and seventh aspects of the present invention, when a first language including a quoted expression or a plurality of clauses / parallel phrases is input, the first language sentence and Since the example database in which the second language sentence has a one-to-one correspondence can be used appropriately, the sentence in the first language can be translated into a natural sentence in the second language.

  According to the invention described in claim 4, by adding a subject or a proposal to the divided translation unit and establishing the translation unit as one sentence, the sentence of the second language included in the example database Can be matched.

  According to the invention described in claim 5, with respect to the divided translation units, the deletion cost and the replacement are included in the edit distance when the translation unit has the same syntax structure as the example data stored in the example database. When calculating the distance with the cost added and editing by replacing, deleting, or inserting words contained in the second language sentence corresponding to the example data so that the translation unit is the same as the example data The example database can be used appropriately by calculating the editing cost of the above and outputting the score and the translation candidate for each translation unit from the distance and the editing cost.

Next, embodiments of the present invention will be described in detail with reference to the drawings as appropriate.
First, the configuration and operation of the example database creation device will be described, and the configuration and operation of the translation device will be described with specific examples as appropriate.
(Configuration of example database creation device)
FIG. 1 is a block diagram of an example database creation apparatus. As illustrated in FIG. 1, the example database creation device 1 includes a translation database 2 in which translation data obtained by translating a sentence in a first language into a sentence in a second language is stored. When translated into a plurality of sentences in a language, one sentence corresponding translation data translated so that each part of one sentence in the first language corresponds to one sentence in the second language is added to the parallel translation database 2 The example database 4 is created, the translation data discriminating means 3, the quoted expression discriminating means 5, the quoted expression pattern accumulating means 7, the quoted expression separating means 9, the second language plural sentence discriminating means 11, the sentence A delimiter pattern accumulating unit 13, a first language expression specifying unit 15, a bilingual dictionary data accumulating unit 17, a first language dividing unit 19, and a divided first language corresponding second language adding unit 21 are provided.

  In this embodiment, the example database 4 created by the example database creation device 1 is described as the first language is Japanese and the second language is English. However, the bilingual database 2 exists, and the quote expression pattern Any language may be used as long as the storage means 7, the sentence delimiter pattern storage means 13 and the bilingual dictionary data storage means 17 can be prepared.

  Also, the bilingual database 2 is an article correspondence and sentence correspondence for Japanese news articles and English news articles, and here, the one actually used in news programs etc. is adopted. ing.

The translation data discriminating means 3 uses the translation data stored in the parallel translation database 2 as follows:
It is determined whether one sentence in Japanese is translated into a plurality of sentences in English. The discrimination by the translation data discriminating means 3 is simply determined by whether or not two or more periods are included in the corresponding English sentence for one Japanese sentence. In the determination by the translation data determination means 3, if an English sentence includes a period indicating omission, the English sentence is determined as a plurality of sentences.

  The example database creation device 1 processes each sentence of the Japanese sentence determined by the translation data determination means 3 together with the corresponding English sentence.

  The quoted expression discriminating means 5 represents a citation in a sentence (multiple translated sentence inclusion sentence) determined by the translation data discriminating means 3 that one Japanese sentence is translated into a plurality of English sentences. Whether or not a quote expression is included is determined using the quote expression pattern stored in the quote expression pattern storage means 7.

  The quoted expression pattern storage means 7 stores quoted expression patterns obtained by patterning quoted expressions that frequently appear in Japanese, and is constituted by a general recording medium such as a hard disk. The quote expression pattern stored in the quote expression pattern storage means 7 is, for example, “according to ...”, “generally said to ...”, “... according to ... ... ".

The quoted expression separating means 9 is for separating the quoted expression discriminated by the quoted expression discriminating means 5 from a plurality of translated sentence inclusion sentences.
Here, a specific example in the case of separation by the quoted expression separation unit 9 after discrimination by the quoted expression discrimination unit 5 will be described.
The multiple-translation sentence inclusion sentence is "If JR says, the Tokaido / Sanyo Shinkansen diamonds are expected to continue to be disrupted today." The English sentence corresponding to this multiple-translation sentence inclusion sentence is "The Japan Railway". “Company say the Tokyo Sanyo Shinkansen services will be disrupted the last train tonight”.

  In such a case, after the citation expression discriminating means 5 discriminates and the citation expression discriminated by the citation expression separation means 9 is separated, the citation expression becomes “The Japan Railway Company Says”, and the multiple translation sentence inclusion sentence ( Hereinafter, the text is also simply referred to as “the Tokaido Sanyo Shinkansen services will be disrupted the last train tonight.”.

  The second language plural sentence discriminating means 11 stores a sentence delimiter pattern as to whether or not the plural translated sentence inclusion sentence (text) from which the quote expression is separated by the quote expression separating means 9 is composed of a plurality of English sentences. The determination is made using the sentence delimiter pattern stored in the means 13. Then, when the second language plural determination unit 11 determines that the text is not composed of a plurality of English sentences, the text is discarded. That is, no further processing is performed on the text.

  Note that the second language plural determination unit 11 discards the text when it is determined that the text is not composed of a plurality of English sentences. Since the quoted expressions are separated, this result may be output to the parallel translation database 2 to enrich the translation data. That is, at first, the translation data included in the bilingual database 2 includes a Japanese multiple translation sentence inclusion sentence and an English multiple sentence, a Japanese quotation expression and an English quotation expression, and a Japanese one. It is divided into two sentences, one sentence and one sentence in English.

  The sentence delimiter pattern accumulating unit 13 accumulates a sentence delimiter pattern obtained by patterning sentence delimiters according to English grammar, and is configured by a general recording medium such as a hard disk. The sentence delimiter pattern stored in the sentence delimiter pattern accumulating unit 13 is a pattern for identifying a period at the end of the sentence except for a period indicating omission. Examples of periods indicating omission are “Mr.”, “Mt.”, and “Dr.”.

  The first language expression specifying unit 15 determines that the second language plural sentence discriminating unit 11 determines that the plural translated sentence inclusion sentence (text) from which the cited expression is separated is composed of plural English sentences. The bilingual dictionary data accumulated in the bilingual dictionary data accumulating means 17 and the Japanese dictionary to determine which of the words contained in the text the words contained in the plurality of English sentences correspond to. It is specified (corresponds) using the syntax data obtained for the relationship between the sentence structure and the sentence structure in English.

  The syntax data represents a set of clauses and phrases. For example, in the case of a phrase, the Japanese “Meteorological Agency” corresponds to the English “The Meteorological Agency”, and this word is present in the sentence “The Meteorological Agenda has a rain train”. This is reflected as parentheses that delimit this phrase “(The Metalogical Agency) has been a heavy rain advice”.

  In this embodiment, the first language expression specifying means 15 uses the bilingual dictionary data to determine which word included in the plurality of English sentences corresponds to the word included in the text. Log likelihood ratio data obtained for the relationship between the Japanese word and the English word, and the word correspondence probability obtained for the probability that the Japanese word corresponds to the English word. May be specified.

Here, with reference to FIG. 7, the example (association example) specified by the first language expression specifying means 15 will be described.
As shown in FIG. 7, English (from later tonight) is associated with Japanese (from now on), English (until torrow) is associated with Japanese (even tomorrow), and English (Heavy rain) is associated with Japanese (severe rain), English (is forecast) is associated with Japanese (may be feared), and English (The Metalogical Agency) is associated with Japanese (Meteorological Agency) is associated with English (a heavy rain) is associated with Japanese (rain), English (advancery) is associated with Japanese (to be wary), English (Has raised) is associated with Japanese (calling).

  As can be seen from the example in FIG. 7, Japanese (in various places) and (future) are not associated with English words. This is either a part that is not important in terms of content because the translated English is not a direct translation of Japanese, or a part that occurs because it is omitted.

  The bilingual dictionary data accumulating means 17 accumulates bilingual dictionary data which is a verbatim translation of Japanese words and English words, and is constituted by a general recording medium such as a hard disk.

  The first language dividing unit 19 is configured to determine each part of the text according to the correspondence between the words included in the plurality of English sentences specified by the first language expression specifying unit 15 and the words included in the text. The main text is divided so that one sentence in English corresponds, and a subject / proposition adding means 19a is provided.

The subject / proposition adding means 19a adds a Japanese subject or proposal to each part in accordance with one English sentence when the text is divided into each part. The proposal has various meanings such as a proposition to be proved, a topic, a standing, an assertion, and a thesis, but here, the phrase that ends with the particle "ha" is the closest to the beginning of the sentence.
Here, an example in which the text shown in FIG. 7 is divided by the first language dividing unit 19 will be described.

  "(Future) (even tomorrow) (in various places) (it will rain heavily) (may be afraid), (the Meteorological Agency) (future) (in the rain) (to be vigilant) (calling)." "(Future) (even tomorrow) (in various places) (it will rain heavily) (may be afraid)" and "(Japan Meteorological Agency) (future) (in the rain) (to be vigilant) (call The English sentence corresponding to the previous sentence will be "Heavy rain is fore cast from ton until tomtrorow", and the English sentence corresponding to the latter sentence will be "The Metrological agenda". heavy rain advice ".

  “The Tokaido / Sanyo Shinkansen is a typhoon, so the schedule has been greatly disturbed by the fact that eighteen trains have been suspended due to the impact of three stopovers.” “Tokaido / Sanyo Shinkansen is a typhoon, so it was affected by driving for three times,” and “18 trains have been suspended so far.” Divided. And the English sentence corresponding to the first sentence is "Tokaido Sankinshin train services have been recycled toed." typhoon. "and the English sentence corresponding to the next sentence is" 18 trains have bean cancelled so far. "

  In these English sentences, “The Shinkansen bullettrains” corresponds to the subject added by the subject / proposition adding means 19a.

  The divided first language corresponding second language adding means 21 includes a plurality of English sentences corresponding to each part of the text divided by the first language dividing means 19 and one English sentence, and each part of the text. Is added to the parallel translation database 2 as single-text translation data.

  According to this example database creation device 1, when it is determined that one Japanese sentence is translated into a plurality of English sentences in the translation data stored in the bilingual database 2, the Japanese citation After removing the expression, if the sentence in Japanese is translated into multiple sentences in English by associating it with the sentence in English, the part in one sentence in Japanese and the sentence in English are A corresponding example database 4 can be created.

(Operation of example database creation device)
Next, the operation of the example database creation device 1 will be described with reference to the flowchart shown in FIG. 2 (see FIG. 1 as appropriate).
First, the example database creation device 1 uses the translation data discriminating means 3 to discriminate whether one Japanese sentence is translated into a plurality of English sentences in the translation data stored in the parallel translation database 2. (Step S1). The example database creation device 1 ends the operation when there is no translation data in which one Japanese sentence is translated into a plurality of English sentences (step S1, No), and one Japanese sentence is converted to English. If the sentence is translated into a plurality of sentences (step S1, Yes), the quoted expression discriminating means 5 determines whether or not a quoted expression is included in one sentence in Japanese (multiple translated sentences included sentence). It discriminate | determines (step S2).

  Subsequently, the example database creation device 1 determines that the quoted expression discriminating unit 5 determines that the quoted expression is included in the plural translation sentence inclusion sentence (step S2, Yes), and the quoted expression separating unit 9 performs plural translation. The quoted expression is separated from the sentence inclusion sentence (step S3). If it is not determined that a quoted expression is included (No in step S2), step S3 is escaped and the process proceeds to step S4.

  Then, the example database creation device 1 is configured so that the multiple translated sentence inclusion sentence (text) in which the quotation expression is separated by the quotation expression separation means 9 by the second language plural sentence discrimination means 11 is composed of a plurality of English sentences. It is determined whether or not (step S4). When the example database creation device 1 determines that the text is not composed of a plurality of English sentences (No in step S4), the operation ends, and the text is composed of a plurality of English sentences. If it is determined (step S4, Yes), the first language expression specifying means 15 specifies which expression (word) in the body that is a Japanese sentence the word included in the English sentence corresponds to. (Step S5).

  Then, the example database creation device 1 uses the first language dividing unit 19 to include words included in a plurality of English sentences specified by the first language expression specifying unit 15, each part of the text, and one English sentence. The body is divided so that the sentence corresponds (step S6). Then, the example database creation device 1 uses the divided first language-corresponding second language adding unit 21 to generate a plurality of English sentences corresponding to each part of the text and one English sentence and each part of the text. Is added to the bilingual database 2 as single sentence corresponding translation data to be an example database 4 (step S7).

(Configuration of translation device)
Next, a translation apparatus using the example database 4 created by the example database creation apparatus 1 will be described with reference to FIG. FIG. 3 is a block diagram of the translation apparatus.
As shown in FIG. 3, the translation device 31 translates an input sentence in the first language into a sentence in the second language, and includes a quoted expression discriminating means 5, a quoted expression pattern accumulating means 7, and a quoted expression. Separating means 9, first language clause / parallel phrase discriminating means 33, clause / parallel phrase pattern accumulating means 35, first language translation unit dividing means 37, unit-by-unit translation means 39, translation means 41, maximum Score translation result selection means 43, translation result output means 45, and bilingual dictionary data storage means 17 are provided. Note that the same components as those in the example database creation device 1 shown in FIG.

  In this translation apparatus 31, the first language to be input is described as Japanese, and the second language to be translated is English. However, the translation database data storage means 17 can be prepared because the example database 4 exists. Any language can be used.

  In addition, an example of a Japanese sentence (hereinafter also referred to simply as an input sentence) input to the translation device 31 is that, according to the Japan Meteorological Agency, the cyclone passes near Japan, so it will be gathered on the Pacific side of northern Japan. There is a risk that it will snow. "

  The first language clause / parallel phrase discriminating means 33 determines whether or not the Japanese sentence (text) from which the quote expression is separated by the quote expression separating means 9 includes a clause or a parallel phrase. It is determined by using the clause / parallel phrase pattern stored in 35. The first language clause / parallel phrase discriminating means 33 discriminates a clause or parallel phrase when the text matches the clause / parallel phrase pattern. Then, the first language clause / parallel phrase discriminating means 33 discriminates that the text determined to contain the clause or the parallel phrase is contained in the first language translation unit dividing means 37 as containing the clause or the parallel phrase. The missing text is output to the translation means 41.

  In the first language clause / parallel phrase discriminating means 33, the text “quoted expression is removed from the input sentence“ Snow that has gathered around the Pacific side of northern Japan will continue tomorrow because the cyclone passes near Japan. "There is a risk of falling" is entered. By the way, the quoted expression separated from the input sentence by the quoted expression separating means 9 is “It means that it depends on the Japan Meteorological Agency”.

  The clause / parallel phrase pattern storage means 35 stores sentence clauses according to Japanese grammar and clause / parallel phrase patterns obtained by patterning parallel phrases, and is constituted by a recording medium such as a general hard disk. Yes. The clause / parallel phrase pattern stored in the clause / parallel phrase pattern storage unit 35 includes, for example, “verb +” and “verb + to” in the case of a clause (the end of a clause which is the end of a clause). Can be mentioned.

  For example, in the text “It's raining in Kanto, but not in Tohoku.”, The pattern “verb + is” and the part “It ’s falling” match. , It is determined that the end of the clause is included.

In the case of a parallel phrase, “one or more“ nouns + is + numbers + places ”” and one “nouns + is + numbers + places” are listed.
For example, if this parallel phrase pattern is applied to the text “There are two other road damages, one bridge washed away, and five mountain landslides.” It is determined that three parallel phrases, “Two places”, “One washed away bridge”, and “Five places of landslides” are included.

  The first language translation unit dividing means 37 divides the text determined by the first language clause / parallel phrase discriminating means 33 to contain a clause or parallel phrase into various translation units, and translates each unit. And a subject / proposition adding means 37a. The translation unit output from the first language translation unit dividing unit 37 is translated by the unit translation unit 39 for each translation unit, and a score (details will be described later) for each translation unit is calculated. The translation candidate with the maximum score (details will be described later) is selected by being output to the score translation result selection means 43.

  Therefore, in this first language translation unit dividing means 37, all combinations that can be divided according to these clauses or parallel phrases are included in the text containing the clauses or parallel phrases in the first language clause / parallel phrase discriminating means 33. Output translation units for.

  For example, in the first language translation unit dividing means 37, when only a plurality of sections are included in the body, the body is divided so that each section is a translation unit, and three or more sections (section A) , Section B, Section C) are included in the text, Section A and Section B are one translation unit, Section C is one translation unit, Section A is a translation unit, The body is divided so that Section B and Section C are used as one translation unit. Further, the first language translation unit dividing means 37 divides the text so that each parallel phrase is used as a translation unit when no clause is included and only a plurality of parallel phrases are included in the text. When three or more parallel phrases (A phrase, B phrase, C phrase) are included in the text, the A phrase and the B phrase are set as one translation unit, and the C phrase is set as one translation unit. The text is divided so that the A phrase is one translation unit and the B phrase and the C phrase are one translation unit.

  Alternatively, the first language translation unit dividing means 37 does not output all combinations that can be divided according to the clauses or parallel phrases, and includes a plurality of clauses and a plurality of parallel phrases included in the text. Divides the text by using parallel phrases as one translation unit for one section, and each parallel phrase as a translation unit for another section, or the number of parallel phrases for one section is greater than a predetermined number Only when the parallel phrase is used as a translation unit, the text is divided.

Here, examples of translation units are given.
For example, “Since the cyclone passes near Japan, there is a risk of falling snow centering on the Pacific side of northern Japan even in the future.” “Because the cyclone passes near Japan” It is divided into two translation units, "There is a risk of snow falling on the Pacific side of the country."

  The subject / proposition adding means 37a adds a Japanese subject or proposal to the translation unit in accordance with one English sentence when the text is divided into translation units.

  The unit translation means 39 translates each translation unit output from the first language translation unit division means 37 using the example database 4, and provides the score for each translation unit and the translation result for each translation unit. The translation candidate is output to the maximum score translation result selection means 43. A detailed configuration of the unit-by-unit translation means 39 is shown in FIG. As shown in FIG. 4, the unit-by-unit translation means 39 includes an example data acquisition means 39a, an example data selection means 39b, an editing means 39c, and a translation candidate output means 39d.

  The example data obtaining unit 39a stores, in the example database 4 as example data, a Japanese sentence in which predicates included in the translation unit output from the first language translation unit dividing unit 37 match or satisfy a preset similarity. It is obtained from translation data that is stored or translation data corresponding to one sentence. Note that the similarity set in advance is the semantic proximity between one predicate and another predicate. For example, the similarity value can be obtained as the probability that Japanese in one predicate and Japanese in another predicate correspond to the same English in the example database 4. Satisfying a preset similarity indicates that the similarity value is greater than a preset threshold.

  For example, when “because the low pressure passes near Japan” is input to the example data acquisition means 39a as a translation unit, a predicate that matches or is similar to the predicate “passes” included in this translation unit is included. As the example data, “low pressure passes near Okinawa” (the corresponding English sentence is “A low air pressure systems close to Okinawa.”) Is acquired from the example database 4.

  In this embodiment, the example data acquisition unit 39a focuses on a predicate included in a Japanese sentence, but in addition to this predicate, translation data stored in the example database 4 or single sentence correspondence translation In the data, if the predicate included in the English sentence corresponds to a predicate other than the predicate included in the Japanese sentence, only the part that matches the translation unit is acquired for the corresponding part.

  For example, if “because a developing low pressure is traveling south of the Kanto region” is input to this example data acquisition means 39a, “under development” included in this translation unit Only “is developing” is acquired from the example database 4 from the English sentence “A low-pressure system is developing off the Kanto coast” of translation data including matching or similar parts or translation data corresponding to one sentence.

  The example data selection unit 39b calculates the distance between the example data acquired by the example data acquisition unit 39a and the translation unit, and calculates a predetermined number of example data (from the smallest distance to the ranking) from the smallest calculated distance. (Example data positioned at the top of the case) is selected.

  This distance is a guideline indicating how similar the example data and translation units are (representing the degree to which the syntax structure is approximate, and not only the syntax structure but also the contents (words) are similar ), The smaller the distance, the more similar.

  The distance calculated by the example data selection unit 39b is the minimum edit distance when the example data and the translation unit have the same syntax structure (including the case where the same syntax structure is obtained by changing the order of words). Suppose that For example, when the Japanese sentence of the example data is composed of subject + complement 1 + complement 2 + predicate, and the translation unit is composed of subject + complement 1 + complement 2 + predicate, the same syntax structure is assumed. It can be said that even if the translation unit is composed of subject + complement 2 + complement 1 + predicate, the same syntax structure is taken.

  Then, the cost when a Japanese word that has been deleted when the example data (Japanese sentence) is translated into an English sentence is added to the edit distance as a deletion cost. The cost when the semantic attributes are the same or when the meaning is approximated in the thesaurus (thesesaurus: classified and arranged words by semantic similarity, classified vocabulary table) is added as a replacement cost.

  As an example of the deletion cost, when the Japanese sentence of the example data is compared with the English sentence associated with the example data, the deletion cost when omitted is “0”.

  As an example of replacement cost, if a word can be replaced simply, such as words that have similar meanings in the thesaurus, such as “future” and “future”, “dislike” and “bad”, the replacement cost is Lower. On the other hand, if you need to replace words with meanings that are far from each other (for example, opposite words), “future” and “past”, “hate” and “like”, etc. The replacement cost is high.

  Further, as examples of replacement costs, for example, “Saitama Prefecture” and “Kanagawa Prefecture” are the same in terms of place names, so the replacement cost is low.

  The example data (Japanese sentence) selected by the example data selection means 39b is output to the editing means 39c together with the corresponding English sentence, and the calculated distance is passed through the editing means 39c. It is output to the translation candidate output means 39d.

  The editing unit 39c edits the example data so that the example data (Japanese sentence) selected by the example data selecting unit 39b is the same as the translation unit, and the correspondence between the Japanese word and the English word is edited. Using Japanese-English correspondence information with preset relations (pre-set editing rules), replace, delete, or insert words contained in English sentences associated with the example data according to this editing In addition to calculating and outputting the editing cost for the case, the English sentence after replacement, deletion or insertion is output as a translation candidate. Here, a word included in an English sentence is referred to as a replacement cost, a cost when it is replaced, a deletion cost as a deletion cost, and a cost when another word is inserted as an insertion cost. I decided to.

  Note that the Japanese-English correspondence information defines how much reliability (1 or less) is used when replacing Japanese words (Japanese expressions) with English words (English expressions). It is. For example, the Japanese English word correspondence information defines that the Japanese word “Okinawa” corresponds to the English word “Okinawa” with a reliability of 0.9. The Japanese word “Okinawa” is defined as having a reliability of 0.4 and corresponding to the English word “system”.

  Also, the editing cost is calculated based on the reliability defined for Japanese-English words when replacing or deleting words included in the English sentence associated with the example data. In other words, if the reliability defined in correspondence with the Japanese-English word is high, the reliability is low, and if the reliability is low, the reliability is high.

  Further, the editing cost is calculated as follows when a word is inserted into an English sentence associated with the example data. First, the syntax structure of the English sentence associated with the example data is analyzed, and as a result of the analysis, only the position where the modification relationship of the inserted word is correct is set as the insertion position candidate. As a result of the analysis, when there are a plurality of insertion position candidates, the optimal insertion position candidate is determined using statistical information such as a language model recorded in advance.

  Note that the insertion cost for inserting this word increases as the number of words to be inserted increases. Then, the editing cost is obtained as the sum of the deletion cost, the replacement cost, the insertion cost, and the like.

  Here, an example of calculating the editing cost when the example data is edited by the editing means 39c in the same manner as the translation unit, and the English sentence associated with the example data is edited in accordance with this editing. And will be described.

  If the translation unit input to the editing means 39c is “because the low pressure passes near Japan” and the example data is “the low pressure passes near Okinawa”, it is included in the translation unit. To match “Japan”, “Okinawa” included in the example data is replaced with “Japan”. When the English sentence associated with the example data is also edited using the Japanese-English correspondence information in accordance with this editing, “A low air pressure system close to Okinawa.” Is changed to “A low air.” "pressure system passes close to Japan."

  In this case, the editing unit 39c performs editing by replacing “Okinawa” in the example data (Japanese sentence) with “Japan”, and according to this editing, “Okinawa” in the English sentence associated with the example data. Is edited to replace “Japan” with “Japan”. In such editing, the English expression and the editing cost are determined by the reliability of the correspondence between the Japanese word “Okinawa” and the English word (English expression) based on the Japanese / English correspondence information. Will be. In this embodiment, the editing cost is defined as “1-reliability”.

  In the Japanese-English correspondence information, for example, when the Japanese word “Okinawa” is associated with the English expression “Okinawa” and the reliability is 0.9, the editing cost is 1-0. 9 = 0.1. And if the translation unit is “because the low pressure passes near Japan” and the example data is “the low pressure passes near Okinawa”, it matches the “Japan” included in the translation unit. Replace “Okinawa” in the example data with “Japan”. The English sentence associated with this example data is also edited using the Japanese-English correspondence information in accordance with this editing, so that “Okinawa” associated with “Okinawa” is changed to “Japan”. The expression “Japan” is expressed, and the edited English sentence (translation candidate) is “A low air pressure system close to Japan” as described above.

  In the Japanese-English correspondence information, for example, when the Japanese word “Okinawa” is associated with the English expression “system” and the reliability is 0.4, the editing cost is 1− 0.4 = 0.6. And if the translation unit is “because the low pressure passes near Japan” and the example data is “the low pressure passes near Okinawa”, it matches the “Japan” included in the translation unit. Replace “Okinawa” in the example data with “Japan”. The English sentence associated with this example data is also edited using the Japanese-English correspondence information in accordance with this editing, so that “system” associated with “Okinawa” is changed to “Japan”. The expression “Japan” is expressed, and the edited English sentence (translation candidate) is “A low air pressure Japan passes close to Okinawa.”.

  The translation candidate output unit 39d outputs a score and a translation candidate for each translation unit from the distance calculated by the example data selection unit 39b and the editing cost calculated by the editing unit 39c. The score for each translation unit is obtained by adding -1 by adding the editing cost calculated by the editing means 39c to the distance calculated by the example data selecting means 39b.

Here, a series of translation processing examples by the first language translation unit dividing unit 37 and the unit-by-unit translation unit 39 will be described with reference to FIG.
This example of translation processing is the translation device 31 when a Japanese input sentence is translated into English. As shown in FIG. 8, the input sentence is “if the low pressure passes near Japan (according to the Japan Meteorological Agency). Therefore, there is a risk that it will fall down tomorrow, mainly on the Pacific side of northern Japan. " Then, it is assumed that “(according to the Japan Meteorological Agency) (that is)” enclosed in parentheses in this input sentence is a quoted expression and is separated by the quoted expression separating means 9.

  Furthermore, the first language clause / parallel phrase discriminating means 33 uses the first language clause / parallel phrase discriminating means 33 to separate the quoted expression “there is a possibility that snow falls mainly on the Pacific side of northern Japan as the low pressure passes near Japan.・ It is assumed that a parallel phrase is included. In this case, the input sentence divided by the first language translation unit dividing means 37 (to be exact, the translation unit) is “because the low pressure passes near Japan” and “tomorrow, focusing on the Pacific side of northern Japan. There is a risk of falling snow together ”, and for each of these two example data (a Japanese sentence in a similar example) is obtained by the example data acquisition means 39a of the unit-by-unit translation means 39,“ low pressure is Okinawa “Tomorrow is likely to have heavy snowfall in northern Japan”.

  The English sentence associated with the example data (the English sentence of the similar example) is “A low air pressure system passes to Okinawa” is “A low air pressure system passes to Okinawa”. “There is a risk of heavy snowfall in northern Japan” is “Heavy snow will fall in northern Japan tomorrow”.

  Then, the example data selection unit 39b and the editing unit 39c of the unit-by-unit translation unit 39 are "A low air pressure system close to Japan" and "Heavy snowman confused in the world." " Returning to FIG.

  The translation unit 41 translates the text that has not been identified by the first language clause / parallel phrase discrimination unit 33 as containing a clause or parallel phrase, using the example database 4, and outputs it to the translation result output unit 45. Is.

  The maximum score translation result selection unit 43 selects a combination of translation units that has the maximum total score obtained by summing up the scores for the respective translation units output from the unit-by-unit translation unit 39, and selects a translation candidate combination of the corresponding translation unit. The maximum score translation result is output to the translation result output means 45.

Here, an example of a combination of the total score obtained by summing up the scores for each translation unit and the translation candidates for the corresponding translation unit (input sentence is as described above) in the maximum score translation result selection means 43 is shown.
According to the translation candidate with a score of 5 (translation result when not divided into sections), there is a risk that snow will fall on the Pacific side of northern Japan, even if it goes tomorrow because the cyclone passes near Japan. “Heavy snow will fall along the Pacific Ocean in northern Japan from flat ton until torrow.”

  For a translation candidate with a score = 8 (translation result when divided into two sections), “A low air pressure system passes to Japan.” “Tomorrow, there is a risk of falling snow centering on the Pacific side of northern Japan” will be “Heavy snow will fall all the Pacific Ocean in northern Japan tontountil tomorrow.”

  In this case, the translation result (score = 8) when divided into two sections has a higher score than the translation result (score = 5) when not divided into sections. In means 43, “A low air pressure systems close to Japan” and “Heavy snow will fall in the north of Japan” are output as the maximum score translation result.

  The translation result output means 45 translates the citation expression separated by the citation expression separation means 9 by using the translation dictionary data stored in the parallel translation dictionary data storage means 17 and the translation result output from the translation means 41. The translation result or the maximum score translation result selected by the maximum score translation result selection means 43 is output. It should be noted that if the quoted expression is not included in the input sentence input to the translation apparatus 1, the quoted expression is not output from the quoted expression separating unit 9 as a matter of course. It is never output.

  By the way, the translation result obtained by translating the quoted expression is “The Metalogical Agency says” (according to the Japan Meteorological Agency), and “A low air pressure systems close to Japan”, “A low air pressure systems close to Japan.” Will fall along the Pacific Ocean Atta ri s à s e m e s e n e n e s e n e n e n e n e n e n s e m e n e n e n e n e n e n e n e n e n e n e m e n e n e n e n e m e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e m e s e n e e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n e n to e n e n e. , “The Metalogical Ageny Says heavy snow will fall all along the Pacific Ocean in northern Japan tonnage output will be output.”

  According to this translation apparatus 31, when a Japanese sentence including a quoted expression or a plurality of clauses / parallel phrases is input as an input sentence, the sentence is appropriately separated and divided to obtain a Japanese sentence. Since the example database 4 with one-to-one correspondence with English sentences can be used appropriately, Japanese sentences can be translated into natural English sentences.

(Translation device operation)
Next, the operation of the translation apparatus 31 will be described with reference to the flowchart shown in FIG. 5 (see FIG. 3 as appropriate).
First, the translation device 31 inputs a Japanese sentence as an input sentence (step 11), and determines whether or not the input sentence includes a quotation expression by the quotation expression determination means 5 (step S12).

  Subsequently, in the translation device 31, when the quoted expression separating unit 9 determines that the quoted expression discriminating unit 5 includes the quoted expression (step S <b> 12, Yes), Separated from the input sentence (step S13). If it is not determined that the quoted expression is included (step S12, No), step S13 is escaped and the process proceeds to step S14.

  Then, the translation device 31 determines whether or not the input sentence from which the quoted expression is separated includes a clause or a parallel phrase by the first language clause / parallel phrase determination means 33 (step S14). When it is determined that the translation device 31 includes a clause or a parallel phrase (step S14, Yes), the translation unit 31 divides the translation unit into translation units by the first language translation unit division unit 37 (step S15). Then, each divided translation unit is translated (step S16), and the maximum score translation result selection means 43 selects the maximum score translation result that maximizes the total score of the translation units, and outputs it to the translation result output means 45. (Step S17).

  If the translation device 31 does not determine that it contains a clause or a parallel phrase (step S14, No), the translation unit 41 translates the input sentence from which the quoted expression is separated as it is, and outputs the translation result. It outputs to the means 45 (step S18).

  Thereafter, the translation device 31 uses the translation result output means 45 to obtain the maximum score translation result output from the maximum score translation result selection means 43 or the translation result output from the translation means 41 and the quote expression from the quote expression separation means 9. If output, the translation result translated with reference to the bilingual dictionary data storage means 17 is output (step S19).

Here, with reference to the flowchart shown in FIG. 6, the process in the unit translation means 39 of the translation apparatus 31 is demonstrated (refer FIG. 4 suitably).
First, the translation device 31 inputs a translation unit from the first language translation unit dividing unit 37 to the unit-by-unit translation unit 39 (step S21). Subsequently, the translation apparatus 31 acquires, from the example database 4, example data that matches or is similar to the predicate included in the translation unit by the example data acquisition unit 39a of the unit-by-unit translation unit 39 (step S22).

  The translation device 31 calculates the distance between the example data acquired by the example data acquisition unit 39a and the input translation unit by the example data selection unit 39b of the unit-by-unit translation unit 39, and the example data having a small distance is used. Is selected (step S23). Further, the translation device 31 outputs the translation candidate for the selected example data edited by the editing unit 39c of the unit-by-unit translation unit 39 by editing the editing cost for each translation unit and the English sentence associated with the example data. (Step S24).

  Then, the translation device 31 calculates a score for each translation unit from the distance calculated by the example data selection unit 39b and the editing cost calculated by the editing unit 39c by the translation candidate output unit 39d of the unit-by-unit translation unit 39. Then, the score and the translation result for each translation unit are output (step S25).

  As mentioned above, although embodiment of this invention was described, this invention is not limited to the said embodiment. For example, in the present embodiment, the example database creation device 1 and the translation device 31 have been described. However, the processing of each means in these devices is also realized as an example database creation program and a translation program that are described in a computer language in an executable manner. it can.

It is a block diagram of the example database creation apparatus which concerns on embodiment of this invention. It is the flowchart which showed operation | movement of the example database creation apparatus shown in FIG. It is a block diagram of the translation apparatus which concerns on embodiment of this invention. It is a block diagram of the translation means for every unit of the translation apparatus shown in FIG. It is the flowchart which showed the operation | movement of the translation apparatus shown in FIG. It is the flowchart which showed the operation | movement of the unit translation means shown in FIG. It is the figure which showed the example of matching in the example database creation apparatus. It is the figure which showed the example of translation processing in the translation apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Example database creation apparatus 2 Bilingual database 3 Translation data discrimination means 4 Example database 5 Citation expression discrimination means 7 Citation expression pattern storage means 9 Citation expression separation means 11 Second language multiple sentence discrimination means 13 Sentence delimiter pattern accumulation means 15 First language Expression specifying means 17 Bilingual dictionary data storage means 19 First language dividing means 19a, 37a Subject / subject adding means 21 Divided first language compatible second language adding means 31 Translation device 33 First language clause / parallel phrase discriminating means 35 Parallel phrase pattern storage means 37 First language translation unit division means 39 Unit-by-unit translation means 39a Example data acquisition means 39b Example data selection means 39c Editing means 39d Translation candidate output means 41 Translation means 43 Maximum score translation result selection means 45 Translation result Output means

Claims (7)

A bilingual database storing translation data in which a sentence in the first language is translated into a sentence in the second language, and one sentence in the first language is translated into a plurality of sentences in the second language An example database creation device for creating an example database by adding single sentence corresponding translation data in which one sentence part of the first language corresponds to one sentence of the second language to the parallel translation database Because
A quoted expression discriminating means for discriminating whether or not a quoted expression representing a quote is included in the sentence of the first language, using a quoted expression pattern obtained by patterning a quoted expression frequently appearing in the first language;
A quoted expression separating means for separating and discarding the quoted expression from the sentence in the first language when the quoted expression judging means determines that the quoted expression is included;
A sentence in the first language from which the quote expression is separated by the quote expression separating means is set as a text, and whether or not the text is translated into a plurality of sentences in the second language is determined according to the grammar of the second language. A second language multi-sentence discriminating means for discriminating using a sentence delimiter pattern obtained by patterning sentence delimiters;
When it is determined by the second language plural sentence discriminating means that the text from which the cited expression is separated is translated into a plurality of sentences in the second language, words included in the sentences in the second language 1st language expression specification that specifies which of the words included in the text corresponds to using the bilingual dictionary data in which the first language word and the second language word are associated with each other Means,
Each part of the body and one sentence of the second language according to the correspondence relationship between the words included in the plurality of sentences of the second language specified by the first language expression specifying means and the words included in the body First language dividing means for dividing so that and correspond to each other,
Addition of second language corresponding to divided first language for adding one sentence corresponding translation data in which each part of the body divided by the first language dividing means corresponds to one sentence of the second language to the parallel translation database Means,
An example database creation device comprising:   The example database creation device according to claim 1, wherein the first language dividing unit adds a preset word to be a subject or a proposal of each part to each part of the text. A translation device that translates an input sentence in a first language into a sentence in a second language,
An example database created by the database creation device of claim 1;
A quoted expression discriminating means for discriminating whether or not a quoted expression representing a quote is included in the sentence in the first language, using a quoted expression pattern obtained by patterning a quoted expression frequently appearing in the first language;
When it is determined that the quote expression is included by the quote expression determination means, the quote expression separating means for separating the quote expression from one sentence of the corresponding first language;
Whether or not the sentence in the first language from which the quote expression is separated by the quote expression separating means includes a clause or a parallel phrase, the clause / parallel in which the sentence clause and the parallel phrase according to the grammar of the first language are patterned A first language clause / parallel phrase discrimination means for discriminating using a phrase pattern;
The first language translation unit dividing means for dividing the sentence of the first language determined to contain the clause or the parallel phrase by the first language clause / parallel phrase determining means into the translation unit as a unit to be translated into the second language. When,
A unit-by-unit translation unit that calculates and translates, for each translation unit, a score indicating a degree of coincidence between the translation unit divided by the first language translation unit dividing unit and the data included in the example database;
A maximum score translation result selection means for selecting a maximum score translation result in which the total score summed up about the score for each translation unit calculated when the translation unit is translated by the unit translation means;
Translation means for translating a sentence in the first language determined not to contain a clause or parallel phrase by the first language clause / parallel phrase determination means, using the example database;
The maximum score translation result selected by the maximum score translation result selection means, or the translation result translated by the translation means and the quote expression when the quote expression is separated by the quote expression separation means, A translation result output means for outputting a translation result translated using bilingual dictionary data in which a language word and a second language word are associated;
A translation apparatus comprising:   The translation apparatus according to claim 3, wherein the first language translation unit dividing unit adds a preset word which is a subject or a subject of the translation unit to the translation unit. The unit translation means is:
From the example database, example data acquisition means for acquiring, as example data, a sentence in a first language that matches or pre-set the predicate included in the translation unit;
An example data selection unit that calculates a distance representing a degree of approximation between the syntax structure of the example data acquired by the example data acquisition unit and the translation unit, and selects a predetermined number of example data from the smallest distance; ,
When editing the example data so that the example data selected by the example data selection means and the translation unit have the same representation, the editing cost is calculated according to a preset editing rule, and the example data An editing means that uses the second language sentence after editing the data as translation candidates,
Translation candidate output means for outputting a score and a translation candidate for each translation unit from the distance calculated by the example data selection means and the editing cost calculated by the editing means;
The translation apparatus according to claim 3, further comprising: A bilingual database storing translation data in which a sentence in the first language is translated into a sentence in the second language, and one sentence in the first language is translated into a plurality of sentences in the second language In order to create an example database by adding, to the parallel translation database, one sentence corresponding translation data in which one sentence part of the first language corresponds to one sentence of the second language The
A quoted expression discriminating means for discriminating whether or not a quoted expression representing a quote is included in the sentence in the first language, using a quoted expression pattern obtained by patterning a quoted expression frequently appearing in the first language;
A quoted expression separating unit that separates and discards the quoted expression from the sentence in the first language when the quoted expression determining unit determines that the quoted expression is included;
A sentence in the first language from which the quote expression is separated by the quote expression separating means is set as a text, and whether or not the text is translated into a plurality of sentences in the second language is determined according to the grammar of the second language. Second language multi-sentence discriminating means for discriminating using a sentence delimiter pattern obtained by patterning sentence delimiters
When it is determined by the second language plural sentence discriminating means that the text from which the cited expression is separated is translated into a plurality of sentences in the second language, words included in the sentences in the second language 1st language expression specification that specifies which of the words included in the text corresponds to using the bilingual dictionary data in which the first language word and the second language word are associated with each other means,
Each part of the body and one sentence of the second language according to the correspondence relationship between the words included in the plurality of sentences of the second language specified by the first language expression specifying means and the words included in the body A first language dividing means for dividing so as to correspond to
Addition of second language corresponding to divided first language for adding one sentence corresponding translation data in which each part of the body divided by the first language dividing means corresponds to one sentence of the second language to the parallel translation database means,
An example database creation program characterized by being made to function as In order to translate the sentence of the input first language into the sentence of the second language, a computer having an example database created by the database creation device according to claim 1,
A quoted expression discriminating means for discriminating whether or not a quoted expression representing a quote is included in the sentence in the first language, using a quoted expression pattern obtained by patterning a quoted expression frequently appearing in the first language;
A quoted expression separating unit that separates the quoted expression from one sentence of the corresponding first language when the quoted expression determining unit determines that the quoted expression is included;
Whether or not the sentence in the first language from which the quote expression is separated by the quote expression separating means includes a clause or a parallel phrase, the clause / parallel in which the sentence clause and the parallel phrase according to the grammar of the first language are patterned First language clause / parallel phrase discrimination means for discriminating using phrase patterns,
The first language translation unit dividing means for dividing the sentence of the first language determined to contain the clause or the parallel phrase by the first language clause / parallel phrase determining means into the translation unit as a unit to be translated into the second language. When,
A unit-by-unit translation unit that calculates and translates, for each translation unit, a score indicating the degree of matching between the translation unit divided by the first language translation unit dividing unit and the data included in the example database;
Maximum score translation result selection means for selecting a maximum score translation result that maximizes the total score totaled for the scores for the translation units calculated when the translation unit is translated by the unit translation means,
Translation means for translating a sentence in the first language determined not to contain a clause or parallel phrase by the first language clause / parallel phrase determination means, using the example database;
The maximum score translation result selected by the maximum score translation result selection means, or the translation result translated by the translation means and the quote expression when the quote expression is separated by the quote expression separation means, A translation result output means for outputting a translation result translated using bilingual dictionary data in which a language word and a second language word are associated;
Translation program characterized by functioning as

Images